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 #include "trace.h"
  65
  66 #define CREATE_TRACE_POINTS
  67 #include "trace.h"
  68
  69 #define L2TP_DRV_VERSION        "V2.0"
  70
  71 /* L2TP header constants */
  72 #define L2TP_HDRFLAG_T     0x8000
  73 #define L2TP_HDRFLAG_L     0x4000
  74 #define L2TP_HDRFLAG_S     0x0800
  75 #define L2TP_HDRFLAG_O     0x0200
  76 #define L2TP_HDRFLAG_P     0x0100
  77
  78 #define L2TP_HDR_VER_MASK  0x000F
  79 #define L2TP_HDR_VER_2     0x0002
  80 #define L2TP_HDR_VER_3     0x0003
  81
  82 /* L2TPv3 default L2-specific sublayer */
  83 #define L2TP_SLFLAG_S      0x40000000
  84 #define L2TP_SL_SEQ_MASK   0x00ffffff
  85
  86 #define L2TP_HDR_SIZE_MAX               14
  87
  88 /* Default trace flags */
  89 #define L2TP_DEFAULT_DEBUG_FLAGS        0
  90
  91 /* Private data stored for received packets in the skb.
  92  */
  93 struct l2tp_skb_cb {
  94         u32                     ns;
  95         u16                     has_seq;
  96         u16                     length;
  97         unsigned long           expires;
  98 };
  99
 100 #define L2TP_SKB_CB(skb)        ((struct l2tp_skb_cb *)&(skb)->cb[sizeof(struct inet_skb_parm)])
 101
 102 static struct workqueue_struct *l2tp_wq;
 103
 104 /* per-net private data for this module */
 105 static unsigned int l2tp_net_id;
 106 struct l2tp_net {
 107         struct list_head l2tp_tunnel_list;
 108         /* Lock for write access to l2tp_tunnel_list */
 109         spinlock_t l2tp_tunnel_list_lock;
 110         struct hlist_head l2tp_session_hlist[L2TP_HASH_SIZE_2];
 111         /* Lock for write access to l2tp_session_hlist */
 112         spinlock_t l2tp_session_hlist_lock;
 113 };
 114
 115 #if IS_ENABLED(CONFIG_IPV6)
 116 static bool l2tp_sk_is_v6(struct sock *sk)
 117 {
 118         return sk->sk_family == PF_INET6 &&
 119                !ipv6_addr_v4mapped(&sk->sk_v6_daddr);
 120 }
 121 #endif
 122
 123 static inline struct l2tp_net *l2tp_pernet(const struct net *net)
 124 {
 125         return net_generic(net, l2tp_net_id);
 126 }
 127
 128 /* Session hash global list for L2TPv3.
 129  * The session_id SHOULD be random according to RFC3931, but several
 130  * L2TP implementations use incrementing session_ids.  So we do a real
 131  * hash on the session_id, rather than a simple bitmask.
 132  */
 133 static inline struct hlist_head *
 134 l2tp_session_id_hash_2(struct l2tp_net *pn, u32 session_id)
 135 {
 136         return &pn->l2tp_session_hlist[hash_32(session_id, L2TP_HASH_BITS_2)];
 137 }
 138
 139 /* Session hash list.
 140  * The session_id SHOULD be random according to RFC2661, but several
 141  * L2TP implementations (Cisco and Microsoft) use incrementing
 142  * session_ids.  So we do a real hash on the session_id, rather than a
 143  * simple bitmask.
 144  */
 145 static inline struct hlist_head *
 146 l2tp_session_id_hash(struct l2tp_tunnel *tunnel, u32 session_id)
 147 {
 148         return &tunnel->session_hlist[hash_32(session_id, L2TP_HASH_BITS)];
 149 }
 150
 151 static void l2tp_tunnel_free(struct l2tp_tunnel *tunnel)
 152 {
 153         trace_free_tunnel(tunnel);
 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         trace_free_session(session);
 161         if (session->tunnel)
 162                 l2tp_tunnel_dec_refcount(session->tunnel);
 163         kfree(session);
 164 }
 165
 166 struct l2tp_tunnel *l2tp_sk_to_tunnel(struct sock *sk)
 167 {
 168         struct l2tp_tunnel *tunnel = sk->sk_user_data;
 169
 170         if (tunnel)
 171                 if (WARN_ON(tunnel->magic != L2TP_TUNNEL_MAGIC))
 172                         return NULL;
 173
 174         return tunnel;
 175 }
 176 EXPORT_SYMBOL_GPL(l2tp_sk_to_tunnel);
 177
 178 void l2tp_tunnel_inc_refcount(struct l2tp_tunnel *tunnel)
 179 {
 180         refcount_inc(&tunnel->ref_count);
 181 }
 182 EXPORT_SYMBOL_GPL(l2tp_tunnel_inc_refcount);
 183
 184 void l2tp_tunnel_dec_refcount(struct l2tp_tunnel *tunnel)
 185 {
 186         if (refcount_dec_and_test(&tunnel->ref_count))
 187                 l2tp_tunnel_free(tunnel);
 188 }
 189 EXPORT_SYMBOL_GPL(l2tp_tunnel_dec_refcount);
 190
 191 void l2tp_session_inc_refcount(struct l2tp_session *session)
 192 {
 193         refcount_inc(&session->ref_count);
 194 }
 195 EXPORT_SYMBOL_GPL(l2tp_session_inc_refcount);
 196
 197 void l2tp_session_dec_refcount(struct l2tp_session *session)
 198 {
 199         if (refcount_dec_and_test(&session->ref_count))
 200                 l2tp_session_free(session);
 201 }
 202 EXPORT_SYMBOL_GPL(l2tp_session_dec_refcount);
 203
 204 /* Lookup a tunnel. A new reference is held on the returned tunnel. */
 205 struct l2tp_tunnel *l2tp_tunnel_get(const struct net *net, u32 tunnel_id)
 206 {
 207         const struct l2tp_net *pn = l2tp_pernet(net);
 208         struct l2tp_tunnel *tunnel;
 209
 210         rcu_read_lock_bh();
 211         list_for_each_entry_rcu(tunnel, &pn->l2tp_tunnel_list, list) {
 212                 if (tunnel->tunnel_id == tunnel_id &&
 213                     refcount_inc_not_zero(&tunnel->ref_count)) {
 214                         rcu_read_unlock_bh();
 215
 216                         return tunnel;
 217                 }
 218         }
 219         rcu_read_unlock_bh();
 220
 221         return NULL;
 222 }
 223 EXPORT_SYMBOL_GPL(l2tp_tunnel_get);
 224
 225 struct l2tp_tunnel *l2tp_tunnel_get_nth(const struct net *net, int nth)
 226 {
 227         const struct l2tp_net *pn = l2tp_pernet(net);
 228         struct l2tp_tunnel *tunnel;
 229         int count = 0;
 230
 231         rcu_read_lock_bh();
 232         list_for_each_entry_rcu(tunnel, &pn->l2tp_tunnel_list, list) {
 233                 if (++count > nth &&
 234                     refcount_inc_not_zero(&tunnel->ref_count)) {
 235                         rcu_read_unlock_bh();
 236                         return tunnel;
 237                 }
 238         }
 239         rcu_read_unlock_bh();
 240
 241         return NULL;
 242 }
 243 EXPORT_SYMBOL_GPL(l2tp_tunnel_get_nth);
 244
 245 struct l2tp_session *l2tp_tunnel_get_session(struct l2tp_tunnel *tunnel,
 246                                              u32 session_id)
 247 {
 248         struct hlist_head *session_list;
 249         struct l2tp_session *session;
 250
 251         session_list = l2tp_session_id_hash(tunnel, session_id);
 252
 253         read_lock_bh(&tunnel->hlist_lock);
 254         hlist_for_each_entry(session, session_list, hlist)
 255                 if (session->session_id == session_id) {
 256                         l2tp_session_inc_refcount(session);
 257                         read_unlock_bh(&tunnel->hlist_lock);
 258
 259                         return session;
 260                 }
 261         read_unlock_bh(&tunnel->hlist_lock);
 262
 263         return NULL;
 264 }
 265 EXPORT_SYMBOL_GPL(l2tp_tunnel_get_session);
 266
 267 struct l2tp_session *l2tp_session_get(const struct net *net, u32 session_id)
 268 {
 269         struct hlist_head *session_list;
 270         struct l2tp_session *session;
 271
 272         session_list = l2tp_session_id_hash_2(l2tp_pernet(net), session_id);
 273
 274         rcu_read_lock_bh();
 275         hlist_for_each_entry_rcu(session, session_list, global_hlist)
 276                 if (session->session_id == session_id) {
 277                         l2tp_session_inc_refcount(session);
 278                         rcu_read_unlock_bh();
 279
 280                         return session;
 281                 }
 282         rcu_read_unlock_bh();
 283
 284         return NULL;
 285 }
 286 EXPORT_SYMBOL_GPL(l2tp_session_get);
 287
 288 struct l2tp_session *l2tp_session_get_nth(struct l2tp_tunnel *tunnel, int nth)
 289 {
 290         int hash;
 291         struct l2tp_session *session;
 292         int count = 0;
 293
 294         read_lock_bh(&tunnel->hlist_lock);
 295         for (hash = 0; hash < L2TP_HASH_SIZE; hash++) {
 296                 hlist_for_each_entry(session, &tunnel->session_hlist[hash], hlist) {
 297                         if (++count > nth) {
 298                                 l2tp_session_inc_refcount(session);
 299                                 read_unlock_bh(&tunnel->hlist_lock);
 300                                 return session;
 301                         }
 302                 }
 303         }
 304
 305         read_unlock_bh(&tunnel->hlist_lock);
 306
 307         return NULL;
 308 }
 309 EXPORT_SYMBOL_GPL(l2tp_session_get_nth);
 310
 311 /* Lookup a session by interface name.
 312  * This is very inefficient but is only used by management interfaces.
 313  */
 314 struct l2tp_session *l2tp_session_get_by_ifname(const struct net *net,
 315                                                 const char *ifname)
 316 {
 317         struct l2tp_net *pn = l2tp_pernet(net);
 318         int hash;
 319         struct l2tp_session *session;
 320
 321         rcu_read_lock_bh();
 322         for (hash = 0; hash < L2TP_HASH_SIZE_2; hash++) {
 323                 hlist_for_each_entry_rcu(session, &pn->l2tp_session_hlist[hash], global_hlist) {
 324                         if (!strcmp(session->ifname, ifname)) {
 325                                 l2tp_session_inc_refcount(session);
 326                                 rcu_read_unlock_bh();
 327
 328                                 return session;
 329                         }
 330                 }
 331         }
 332
 333         rcu_read_unlock_bh();
 334
 335         return NULL;
 336 }
 337 EXPORT_SYMBOL_GPL(l2tp_session_get_by_ifname);
 338
 339 int l2tp_session_register(struct l2tp_session *session,
 340                           struct l2tp_tunnel *tunnel)
 341 {
 342         struct l2tp_session *session_walk;
 343         struct hlist_head *g_head;
 344         struct hlist_head *head;
 345         struct l2tp_net *pn;
 346         int err;
 347
 348         head = l2tp_session_id_hash(tunnel, session->session_id);
 349
 350         write_lock_bh(&tunnel->hlist_lock);
 351         if (!tunnel->acpt_newsess) {
 352                 err = -ENODEV;
 353                 goto err_tlock;
 354         }
 355
 356         hlist_for_each_entry(session_walk, head, hlist)
 357                 if (session_walk->session_id == session->session_id) {
 358                         err = -EEXIST;
 359                         goto err_tlock;
 360                 }
 361
 362         if (tunnel->version == L2TP_HDR_VER_3) {
 363                 pn = l2tp_pernet(tunnel->l2tp_net);
 364                 g_head = l2tp_session_id_hash_2(pn, session->session_id);
 365
 366                 spin_lock_bh(&pn->l2tp_session_hlist_lock);
 367
 368                 /* IP encap expects session IDs to be globally unique, while
 369                  * UDP encap doesn't.
 370                  */
 371                 hlist_for_each_entry(session_walk, g_head, global_hlist)
 372                         if (session_walk->session_id == session->session_id &&
 373                             (session_walk->tunnel->encap == L2TP_ENCAPTYPE_IP ||
 374                              tunnel->encap == L2TP_ENCAPTYPE_IP)) {
 375                                 err = -EEXIST;
 376                                 goto err_tlock_pnlock;
 377                         }
 378
 379                 l2tp_tunnel_inc_refcount(tunnel);
 380                 hlist_add_head_rcu(&session->global_hlist, g_head);
 381
 382                 spin_unlock_bh(&pn->l2tp_session_hlist_lock);
 383         } else {
 384                 l2tp_tunnel_inc_refcount(tunnel);
 385         }
 386
 387         hlist_add_head(&session->hlist, head);
 388         write_unlock_bh(&tunnel->hlist_lock);
 389
 390         trace_register_session(session);
 391
 392         return 0;
 393
 394 err_tlock_pnlock:
 395         spin_unlock_bh(&pn->l2tp_session_hlist_lock);
 396 err_tlock:
 397         write_unlock_bh(&tunnel->hlist_lock);
 398
 399         return err;
 400 }
 401 EXPORT_SYMBOL_GPL(l2tp_session_register);
 402
 403 /*****************************************************************************
 404  * Receive data handling
 405  *****************************************************************************/
 406
 407 /* Queue a skb in order. We come here only if the skb has an L2TP sequence
 408  * number.
 409  */
 410 static void l2tp_recv_queue_skb(struct l2tp_session *session, struct sk_buff *skb)
 411 {
 412         struct sk_buff *skbp;
 413         struct sk_buff *tmp;
 414         u32 ns = L2TP_SKB_CB(skb)->ns;
 415
 416         spin_lock_bh(&session->reorder_q.lock);
 417         skb_queue_walk_safe(&session->reorder_q, skbp, tmp) {
 418                 if (L2TP_SKB_CB(skbp)->ns > ns) {
 419                         __skb_queue_before(&session->reorder_q, skbp, skb);
 420                         atomic_long_inc(&session->stats.rx_oos_packets);
 421                         goto out;
 422                 }
 423         }
 424
 425         __skb_queue_tail(&session->reorder_q, skb);
 426
 427 out:
 428         spin_unlock_bh(&session->reorder_q.lock);
 429 }
 430
 431 /* Dequeue a single skb.
 432  */
 433 static void l2tp_recv_dequeue_skb(struct l2tp_session *session, struct sk_buff *skb)
 434 {
 435         struct l2tp_tunnel *tunnel = session->tunnel;
 436         int length = L2TP_SKB_CB(skb)->length;
 437
 438         /* We're about to requeue the skb, so return resources
 439          * to its current owner (a socket receive buffer).
 440          */
 441         skb_orphan(skb);
 442
 443         atomic_long_inc(&tunnel->stats.rx_packets);
 444         atomic_long_add(length, &tunnel->stats.rx_bytes);
 445         atomic_long_inc(&session->stats.rx_packets);
 446         atomic_long_add(length, &session->stats.rx_bytes);
 447
 448         if (L2TP_SKB_CB(skb)->has_seq) {
 449                 /* Bump our Nr */
 450                 session->nr++;
 451                 session->nr &= session->nr_max;
 452                 trace_session_seqnum_update(session);
 453         }
 454
 455         /* call private receive handler */
 456         if (session->recv_skb)
 457                 (*session->recv_skb)(session, skb, L2TP_SKB_CB(skb)->length);
 458         else
 459                 kfree_skb(skb);
 460 }
 461
 462 /* Dequeue skbs from the session's reorder_q, subject to packet order.
 463  * Skbs that have been in the queue for too long are simply discarded.
 464  */
 465 static void l2tp_recv_dequeue(struct l2tp_session *session)
 466 {
 467         struct sk_buff *skb;
 468         struct sk_buff *tmp;
 469
 470         /* If the pkt at the head of the queue has the nr that we
 471          * expect to send up next, dequeue it and any other
 472          * in-sequence packets behind it.
 473          */
 474 start:
 475         spin_lock_bh(&session->reorder_q.lock);
 476         skb_queue_walk_safe(&session->reorder_q, skb, tmp) {
 477                 struct l2tp_skb_cb *cb = L2TP_SKB_CB(skb);
 478
 479                 /* If the packet has been pending on the queue for too long, discard it */
 480                 if (time_after(jiffies, cb->expires)) {
 481                         atomic_long_inc(&session->stats.rx_seq_discards);
 482                         atomic_long_inc(&session->stats.rx_errors);
 483                         trace_session_pkt_expired(session, cb->ns);
 484                         session->reorder_skip = 1;
 485                         __skb_unlink(skb, &session->reorder_q);
 486                         kfree_skb(skb);
 487                         continue;
 488                 }
 489
 490                 if (cb->has_seq) {
 491                         if (session->reorder_skip) {
 492                                 session->reorder_skip = 0;
 493                                 session->nr = cb->ns;
 494                                 trace_session_seqnum_reset(session);
 495                         }
 496                         if (cb->ns != session->nr)
 497                                 goto out;
 498                 }
 499                 __skb_unlink(skb, &session->reorder_q);
 500
 501                 /* Process the skb. We release the queue lock while we
 502                  * do so to let other contexts process the queue.
 503                  */
 504                 spin_unlock_bh(&session->reorder_q.lock);
 505                 l2tp_recv_dequeue_skb(session, skb);
 506                 goto start;
 507         }
 508
 509 out:
 510         spin_unlock_bh(&session->reorder_q.lock);
 511 }
 512
 513 static int l2tp_seq_check_rx_window(struct l2tp_session *session, u32 nr)
 514 {
 515         u32 nws;
 516
 517         if (nr >= session->nr)
 518                 nws = nr - session->nr;
 519         else
 520                 nws = (session->nr_max + 1) - (session->nr - nr);
 521
 522         return nws < session->nr_window_size;
 523 }
 524
 525 /* If packet has sequence numbers, queue it if acceptable. Returns 0 if
 526  * acceptable, else non-zero.
 527  */
 528 static int l2tp_recv_data_seq(struct l2tp_session *session, struct sk_buff *skb)
 529 {
 530         struct l2tp_skb_cb *cb = L2TP_SKB_CB(skb);
 531
 532         if (!l2tp_seq_check_rx_window(session, cb->ns)) {
 533                 /* Packet sequence number is outside allowed window.
 534                  * Discard it.
 535                  */
 536                 trace_session_pkt_outside_rx_window(session, cb->ns);
 537                 goto discard;
 538         }
 539
 540         if (session->reorder_timeout != 0) {
 541                 /* Packet reordering enabled. Add skb to session's
 542                  * reorder queue, in order of ns.
 543                  */
 544                 l2tp_recv_queue_skb(session, skb);
 545                 goto out;
 546         }
 547
 548         /* Packet reordering disabled. Discard out-of-sequence packets, while
 549          * tracking the number if in-sequence packets after the first OOS packet
 550          * is seen. After nr_oos_count_max in-sequence packets, reset the
 551          * sequence number to re-enable packet reception.
 552          */
 553         if (cb->ns == session->nr) {
 554                 skb_queue_tail(&session->reorder_q, skb);
 555         } else {
 556                 u32 nr_oos = cb->ns;
 557                 u32 nr_next = (session->nr_oos + 1) & session->nr_max;
 558
 559                 if (nr_oos == nr_next)
 560                         session->nr_oos_count++;
 561                 else
 562                         session->nr_oos_count = 0;
 563
 564                 session->nr_oos = nr_oos;
 565                 if (session->nr_oos_count > session->nr_oos_count_max) {
 566                         session->reorder_skip = 1;
 567                 }
 568                 if (!session->reorder_skip) {
 569                         atomic_long_inc(&session->stats.rx_seq_discards);
 570                         trace_session_pkt_oos(session, cb->ns);
 571                         goto discard;
 572                 }
 573                 skb_queue_tail(&session->reorder_q, skb);
 574         }
 575
 576 out:
 577         return 0;
 578
 579 discard:
 580         return 1;
 581 }
 582
 583 /* Do receive processing of L2TP data frames. We handle both L2TPv2
 584  * and L2TPv3 data frames here.
 585  *
 586  * L2TPv2 Data Message Header
 587  *
 588  *  0                   1                   2                   3
 589  *  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
 590  * +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
 591  * |T|L|x|x|S|x|O|P|x|x|x|x|  Ver  |          Length (opt)         |
 592  * +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
 593  * |           Tunnel ID           |           Session ID          |
 594  * +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
 595  * |             Ns (opt)          |             Nr (opt)          |
 596  * +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
 597  * |      Offset Size (opt)        |    Offset pad... (opt)
 598  * +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
 599  *
 600  * Data frames are marked by T=0. All other fields are the same as
 601  * those in L2TP control frames.
 602  *
 603  * L2TPv3 Data Message Header
 604  *
 605  * +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
 606  * |                      L2TP Session Header                      |
 607  * +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
 608  * |                      L2-Specific Sublayer                     |
 609  * +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
 610  * |                        Tunnel Payload                      ...
 611  * +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
 612  *
 613  * L2TPv3 Session Header Over IP
 614  *
 615  *  0                   1                   2                   3
 616  *  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
 617  * +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
 618  * |                           Session ID                          |
 619  * +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
 620  * |               Cookie (optional, maximum 64 bits)...
 621  * +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
 622  *                                                                 |
 623  * +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
 624  *
 625  * L2TPv3 L2-Specific Sublayer Format
 626  *
 627  *  0                   1                   2                   3
 628  *  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
 629  * +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
 630  * |x|S|x|x|x|x|x|x|              Sequence Number                  |
 631  * +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
 632  *
 633  * Cookie value and sublayer format are negotiated with the peer when
 634  * the session is set up. Unlike L2TPv2, we do not need to parse the
 635  * packet header to determine if optional fields are present.
 636  *
 637  * Caller must already have parsed the frame and determined that it is
 638  * a data (not control) frame before coming here. Fields up to the
 639  * session-id have already been parsed and ptr points to the data
 640  * after the session-id.
 641  */
 642 void l2tp_recv_common(struct l2tp_session *session, struct sk_buff *skb,
 643                       unsigned char *ptr, unsigned char *optr, u16 hdrflags,
 644                       int length)
 645 {
 646         struct l2tp_tunnel *tunnel = session->tunnel;
 647         int offset;
 648
 649         /* Parse and check optional cookie */
 650         if (session->peer_cookie_len > 0) {
 651                 if (memcmp(ptr, &session->peer_cookie[0], session->peer_cookie_len)) {
 652                         pr_debug_ratelimited("%s: cookie mismatch (%u/%u). Discarding.\n",
 653                                              tunnel->name, tunnel->tunnel_id,
 654                                              session->session_id);
 655                         atomic_long_inc(&session->stats.rx_cookie_discards);
 656                         goto discard;
 657                 }
 658                 ptr += session->peer_cookie_len;
 659         }
 660
 661         /* Handle the optional sequence numbers. Sequence numbers are
 662          * in different places for L2TPv2 and L2TPv3.
 663          *
 664          * If we are the LAC, enable/disable sequence numbers under
 665          * the control of the LNS.  If no sequence numbers present but
 666          * we were expecting them, discard frame.
 667          */
 668         L2TP_SKB_CB(skb)->has_seq = 0;
 669         if (tunnel->version == L2TP_HDR_VER_2) {
 670                 if (hdrflags & L2TP_HDRFLAG_S) {
 671                         /* Store L2TP info in the skb */
 672                         L2TP_SKB_CB(skb)->ns = ntohs(*(__be16 *)ptr);
 673                         L2TP_SKB_CB(skb)->has_seq = 1;
 674                         ptr += 2;
 675                         /* Skip past nr in the header */
 676                         ptr += 2;
 677
 678                 }
 679         } else if (session->l2specific_type == L2TP_L2SPECTYPE_DEFAULT) {
 680                 u32 l2h = ntohl(*(__be32 *)ptr);
 681
 682                 if (l2h & 0x40000000) {
 683                         /* Store L2TP info in the skb */
 684                         L2TP_SKB_CB(skb)->ns = l2h & 0x00ffffff;
 685                         L2TP_SKB_CB(skb)->has_seq = 1;
 686                 }
 687                 ptr += 4;
 688         }
 689
 690         if (L2TP_SKB_CB(skb)->has_seq) {
 691                 /* Received a packet with sequence numbers. If we're the LAC,
 692                  * check if we sre sending sequence numbers and if not,
 693                  * configure it so.
 694                  */
 695                 if (!session->lns_mode && !session->send_seq) {
 696                         trace_session_seqnum_lns_enable(session);
 697                         session->send_seq = 1;
 698                         l2tp_session_set_header_len(session, tunnel->version);
 699                 }
 700         } else {
 701                 /* No sequence numbers.
 702                  * If user has configured mandatory sequence numbers, discard.
 703                  */
 704                 if (session->recv_seq) {
 705                         pr_debug_ratelimited("%s: recv data has no seq numbers when required. Discarding.\n",
 706                                              session->name);
 707                         atomic_long_inc(&session->stats.rx_seq_discards);
 708                         goto discard;
 709                 }
 710
 711                 /* If we're the LAC and we're sending sequence numbers, the
 712                  * LNS has requested that we no longer send sequence numbers.
 713                  * If we're the LNS and we're sending sequence numbers, the
 714                  * LAC is broken. Discard the frame.
 715                  */
 716                 if (!session->lns_mode && session->send_seq) {
 717                         trace_session_seqnum_lns_disable(session);
 718                         session->send_seq = 0;
 719                         l2tp_session_set_header_len(session, tunnel->version);
 720                 } else if (session->send_seq) {
 721                         pr_debug_ratelimited("%s: recv data has no seq numbers when required. Discarding.\n",
 722                                              session->name);
 723                         atomic_long_inc(&session->stats.rx_seq_discards);
 724                         goto discard;
 725                 }
 726         }
 727
 728         /* Session data offset is defined only for L2TPv2 and is
 729          * indicated by an optional 16-bit value in the header.
 730          */
 731         if (tunnel->version == L2TP_HDR_VER_2) {
 732                 /* If offset bit set, skip it. */
 733                 if (hdrflags & L2TP_HDRFLAG_O) {
 734                         offset = ntohs(*(__be16 *)ptr);
 735                         ptr += 2 + offset;
 736                 }
 737         }
 738
 739         offset = ptr - optr;
 740         if (!pskb_may_pull(skb, offset))
 741                 goto discard;
 742
 743         __skb_pull(skb, offset);
 744
 745         /* Prepare skb for adding to the session's reorder_q.  Hold
 746          * packets for max reorder_timeout or 1 second if not
 747          * reordering.
 748          */
 749         L2TP_SKB_CB(skb)->length = length;
 750         L2TP_SKB_CB(skb)->expires = jiffies +
 751                 (session->reorder_timeout ? session->reorder_timeout : HZ);
 752
 753         /* Add packet to the session's receive queue. Reordering is done here, if
 754          * enabled. Saved L2TP protocol info is stored in skb->sb[].
 755          */
 756         if (L2TP_SKB_CB(skb)->has_seq) {
 757                 if (l2tp_recv_data_seq(session, skb))
 758                         goto discard;
 759         } else {
 760                 /* No sequence numbers. Add the skb to the tail of the
 761                  * reorder queue. This ensures that it will be
 762                  * delivered after all previous sequenced skbs.
 763                  */
 764                 skb_queue_tail(&session->reorder_q, skb);
 765         }
 766
 767         /* Try to dequeue as many skbs from reorder_q as we can. */
 768         l2tp_recv_dequeue(session);
 769
 770         return;
 771
 772 discard:
 773         atomic_long_inc(&session->stats.rx_errors);
 774         kfree_skb(skb);
 775 }
 776 EXPORT_SYMBOL_GPL(l2tp_recv_common);
 777
 778 /* Drop skbs from the session's reorder_q
 779  */
 780 static void l2tp_session_queue_purge(struct l2tp_session *session)
 781 {
 782         struct sk_buff *skb = NULL;
 783
 784         while ((skb = skb_dequeue(&session->reorder_q))) {
 785                 atomic_long_inc(&session->stats.rx_errors);
 786                 kfree_skb(skb);
 787         }
 788 }
 789
 790 /* Internal UDP receive frame. Do the real work of receiving an L2TP data frame
 791  * here. The skb is not on a list when we get here.
 792  * Returns 0 if the packet was a data packet and was successfully passed on.
 793  * Returns 1 if the packet was not a good data packet and could not be
 794  * forwarded.  All such packets are passed up to userspace to deal with.
 795  */
 796 static int l2tp_udp_recv_core(struct l2tp_tunnel *tunnel, struct sk_buff *skb)
 797 {
 798         struct l2tp_session *session = NULL;
 799         unsigned char *ptr, *optr;
 800         u16 hdrflags;
 801         u32 tunnel_id, session_id;
 802         u16 version;
 803         int length;
 804
 805         /* UDP has verified checksum */
 806
 807         /* UDP always verifies the packet length. */
 808         __skb_pull(skb, sizeof(struct udphdr));
 809
 810         /* Short packet? */
 811         if (!pskb_may_pull(skb, L2TP_HDR_SIZE_MAX)) {
 812                 pr_debug_ratelimited("%s: recv short packet (len=%d)\n",
 813                                      tunnel->name, skb->len);
 814                 goto invalid;
 815         }
 816
 817         /* Point to L2TP header */
 818         optr = skb->data;
 819         ptr = skb->data;
 820
 821         /* Get L2TP header flags */
 822         hdrflags = ntohs(*(__be16 *)ptr);
 823
 824         /* Check protocol version */
 825         version = hdrflags & L2TP_HDR_VER_MASK;
 826         if (version != tunnel->version) {
 827                 pr_debug_ratelimited("%s: recv protocol version mismatch: got %d expected %d\n",
 828                                      tunnel->name, version, tunnel->version);
 829                 goto invalid;
 830         }
 831
 832         /* Get length of L2TP packet */
 833         length = skb->len;
 834
 835         /* If type is control packet, it is handled by userspace. */
 836         if (hdrflags & L2TP_HDRFLAG_T)
 837                 goto pass;
 838
 839         /* Skip flags */
 840         ptr += 2;
 841
 842         if (tunnel->version == L2TP_HDR_VER_2) {
 843                 /* If length is present, skip it */
 844                 if (hdrflags & L2TP_HDRFLAG_L)
 845                         ptr += 2;
 846
 847                 /* Extract tunnel and session ID */
 848                 tunnel_id = ntohs(*(__be16 *)ptr);
 849                 ptr += 2;
 850                 session_id = ntohs(*(__be16 *)ptr);
 851                 ptr += 2;
 852         } else {
 853                 ptr += 2;       /* skip reserved bits */
 854                 tunnel_id = tunnel->tunnel_id;
 855                 session_id = ntohl(*(__be32 *)ptr);
 856                 ptr += 4;
 857         }
 858
 859         /* Find the session context */
 860         session = l2tp_tunnel_get_session(tunnel, session_id);
 861         if (!session || !session->recv_skb) {
 862                 if (session)
 863                         l2tp_session_dec_refcount(session);
 864
 865                 /* Not found? Pass to userspace to deal with */
 866                 pr_debug_ratelimited("%s: no session found (%u/%u). Passing up.\n",
 867                                      tunnel->name, tunnel_id, session_id);
 868                 goto pass;
 869         }
 870
 871         if (tunnel->version == L2TP_HDR_VER_3 &&
 872             l2tp_v3_ensure_opt_in_linear(session, skb, &ptr, &optr))
 873                 goto invalid;
 874
 875         l2tp_recv_common(session, skb, ptr, optr, hdrflags, length);
 876         l2tp_session_dec_refcount(session);
 877
 878         return 0;
 879
 880 invalid:
 881         atomic_long_inc(&tunnel->stats.rx_invalid);
 882
 883 pass:
 884         /* Put UDP header back */
 885         __skb_push(skb, sizeof(struct udphdr));
 886
 887         return 1;
 888 }
 889
 890 /* UDP encapsulation receive handler. See net/ipv4/udp.c.
 891  * Return codes:
 892  * 0 : success.
 893  * <0: error
 894  * >0: skb should be passed up to userspace as UDP.
 895  */
 896 int l2tp_udp_encap_recv(struct sock *sk, struct sk_buff *skb)
 897 {
 898         struct l2tp_tunnel *tunnel;
 899
 900         /* Note that this is called from the encap_rcv hook inside an
 901          * RCU-protected region, but without the socket being locked.
 902          * Hence we use rcu_dereference_sk_user_data to access the
 903          * tunnel data structure rather the usual l2tp_sk_to_tunnel
 904          * accessor function.
 905          */
 906         tunnel = rcu_dereference_sk_user_data(sk);
 907         if (!tunnel)
 908                 goto pass_up;
 909         if (WARN_ON(tunnel->magic != L2TP_TUNNEL_MAGIC))
 910                 goto pass_up;
 911
 912         if (l2tp_udp_recv_core(tunnel, skb))
 913                 goto pass_up;
 914
 915         return 0;
 916
 917 pass_up:
 918         return 1;
 919 }
 920 EXPORT_SYMBOL_GPL(l2tp_udp_encap_recv);
 921
 922 /************************************************************************
 923  * Transmit handling
 924  ***********************************************************************/
 925
 926 /* Build an L2TP header for the session into the buffer provided.
 927  */
 928 static int l2tp_build_l2tpv2_header(struct l2tp_session *session, void *buf)
 929 {
 930         struct l2tp_tunnel *tunnel = session->tunnel;
 931         __be16 *bufp = buf;
 932         __be16 *optr = buf;
 933         u16 flags = L2TP_HDR_VER_2;
 934         u32 tunnel_id = tunnel->peer_tunnel_id;
 935         u32 session_id = session->peer_session_id;
 936
 937         if (session->send_seq)
 938                 flags |= L2TP_HDRFLAG_S;
 939
 940         /* Setup L2TP header. */
 941         *bufp++ = htons(flags);
 942         *bufp++ = htons(tunnel_id);
 943         *bufp++ = htons(session_id);
 944         if (session->send_seq) {
 945                 *bufp++ = htons(session->ns);
 946                 *bufp++ = 0;
 947                 session->ns++;
 948                 session->ns &= 0xffff;
 949                 trace_session_seqnum_update(session);
 950         }
 951
 952         return bufp - optr;
 953 }
 954
 955 static int l2tp_build_l2tpv3_header(struct l2tp_session *session, void *buf)
 956 {
 957         struct l2tp_tunnel *tunnel = session->tunnel;
 958         char *bufp = buf;
 959         char *optr = bufp;
 960
 961         /* Setup L2TP header. The header differs slightly for UDP and
 962          * IP encapsulations. For UDP, there is 4 bytes of flags.
 963          */
 964         if (tunnel->encap == L2TP_ENCAPTYPE_UDP) {
 965                 u16 flags = L2TP_HDR_VER_3;
 966                 *((__be16 *)bufp) = htons(flags);
 967                 bufp += 2;
 968                 *((__be16 *)bufp) = 0;
 969                 bufp += 2;
 970         }
 971
 972         *((__be32 *)bufp) = htonl(session->peer_session_id);
 973         bufp += 4;
 974         if (session->cookie_len) {
 975                 memcpy(bufp, &session->cookie[0], session->cookie_len);
 976                 bufp += session->cookie_len;
 977         }
 978         if (session->l2specific_type == L2TP_L2SPECTYPE_DEFAULT) {
 979                 u32 l2h = 0;
 980
 981                 if (session->send_seq) {
 982                         l2h = 0x40000000 | session->ns;
 983                         session->ns++;
 984                         session->ns &= 0xffffff;
 985                         trace_session_seqnum_update(session);
 986                 }
 987
 988                 *((__be32 *)bufp) = htonl(l2h);
 989                 bufp += 4;
 990         }
 991
 992         return bufp - optr;
 993 }
 994
 995 /* Queue the packet to IP for output: tunnel socket lock must be held */
 996 static int l2tp_xmit_queue(struct l2tp_tunnel *tunnel, struct sk_buff *skb, struct flowi *fl)
 997 {
 998         int err;
 999
1000         skb->ignore_df = 1;
1001         skb_dst_drop(skb);
1002 #if IS_ENABLED(CONFIG_IPV6)
1003         if (l2tp_sk_is_v6(tunnel->sock))
1004                 err = inet6_csk_xmit(tunnel->sock, skb, NULL);
1005         else
1006 #endif
1007                 err = ip_queue_xmit(tunnel->sock, skb, fl);
1008
1009         return err >= 0 ? NET_XMIT_SUCCESS : NET_XMIT_DROP;
1010 }
1011
1012 static int l2tp_xmit_core(struct l2tp_session *session, struct sk_buff *skb, unsigned int *len)
1013 {
1014         struct l2tp_tunnel *tunnel = session->tunnel;
1015         unsigned int data_len = skb->len;
1016         struct sock *sk = tunnel->sock;
1017         int headroom, uhlen, udp_len;
1018         int ret = NET_XMIT_SUCCESS;
1019         struct inet_sock *inet;
1020         struct udphdr *uh;
1021
1022         /* Check that there's enough headroom in the skb to insert IP,
1023          * UDP and L2TP headers. If not enough, expand it to
1024          * make room. Adjust truesize.
1025          */
1026         uhlen = (tunnel->encap == L2TP_ENCAPTYPE_UDP) ? sizeof(*uh) : 0;
1027         headroom = NET_SKB_PAD + sizeof(struct iphdr) + uhlen + session->hdr_len;
1028         if (skb_cow_head(skb, headroom)) {
1029                 kfree_skb(skb);
1030                 return NET_XMIT_DROP;
1031         }
1032
1033         /* Setup L2TP header */
1034         if (tunnel->version == L2TP_HDR_VER_2)
1035                 l2tp_build_l2tpv2_header(session, __skb_push(skb, session->hdr_len));
1036         else
1037                 l2tp_build_l2tpv3_header(session, __skb_push(skb, session->hdr_len));
1038
1039         /* Reset skb netfilter state */
1040         memset(&(IPCB(skb)->opt), 0, sizeof(IPCB(skb)->opt));
1041         IPCB(skb)->flags &= ~(IPSKB_XFRM_TUNNEL_SIZE | IPSKB_XFRM_TRANSFORMED | IPSKB_REROUTED);
1042         nf_reset_ct(skb);
1043
1044         bh_lock_sock(sk);
1045         if (sock_owned_by_user(sk)) {
1046                 kfree_skb(skb);
1047                 ret = NET_XMIT_DROP;
1048                 goto out_unlock;
1049         }
1050
1051         /* The user-space may change the connection status for the user-space
1052          * provided socket at run time: we must check it under the socket lock
1053          */
1054         if (tunnel->fd >= 0 && sk->sk_state != TCP_ESTABLISHED) {
1055                 kfree_skb(skb);
1056                 ret = NET_XMIT_DROP;
1057                 goto out_unlock;
1058         }
1059
1060         /* Report transmitted length before we add encap header, which keeps
1061          * statistics consistent for both UDP and IP encap tx/rx paths.
1062          */
1063         *len = skb->len;
1064
1065         inet = inet_sk(sk);
1066         switch (tunnel->encap) {
1067         case L2TP_ENCAPTYPE_UDP:
1068                 /* Setup UDP header */
1069                 __skb_push(skb, sizeof(*uh));
1070                 skb_reset_transport_header(skb);
1071                 uh = udp_hdr(skb);
1072                 uh->source = inet->inet_sport;
1073                 uh->dest = inet->inet_dport;
1074                 udp_len = uhlen + session->hdr_len + data_len;
1075                 uh->len = htons(udp_len);
1076
1077                 /* Calculate UDP checksum if configured to do so */
1078 #if IS_ENABLED(CONFIG_IPV6)
1079                 if (l2tp_sk_is_v6(sk))
1080                         udp6_set_csum(udp_get_no_check6_tx(sk),
1081                                       skb, &inet6_sk(sk)->saddr,
1082                                       &sk->sk_v6_daddr, udp_len);
1083                 else
1084 #endif
1085                         udp_set_csum(sk->sk_no_check_tx, skb, inet->inet_saddr,
1086                                      inet->inet_daddr, udp_len);
1087                 break;
1088
1089         case L2TP_ENCAPTYPE_IP:
1090                 break;
1091         }
1092
1093         ret = l2tp_xmit_queue(tunnel, skb, &inet->cork.fl);
1094
1095 out_unlock:
1096         bh_unlock_sock(sk);
1097
1098         return ret;
1099 }
1100
1101 /* If caller requires the skb to have a ppp header, the header must be
1102  * inserted in the skb data before calling this function.
1103  */
1104 int l2tp_xmit_skb(struct l2tp_session *session, struct sk_buff *skb)
1105 {
1106         unsigned int len = 0;
1107         int ret;
1108
1109         ret = l2tp_xmit_core(session, skb, &len);
1110         if (ret == NET_XMIT_SUCCESS) {
1111                 atomic_long_inc(&session->tunnel->stats.tx_packets);
1112                 atomic_long_add(len, &session->tunnel->stats.tx_bytes);
1113                 atomic_long_inc(&session->stats.tx_packets);
1114                 atomic_long_add(len, &session->stats.tx_bytes);
1115         } else {
1116                 atomic_long_inc(&session->tunnel->stats.tx_errors);
1117                 atomic_long_inc(&session->stats.tx_errors);
1118         }
1119         return ret;
1120 }
1121 EXPORT_SYMBOL_GPL(l2tp_xmit_skb);
1122
1123 /*****************************************************************************
1124  * Tinnel and session create/destroy.
1125  *****************************************************************************/
1126
1127 /* Tunnel socket destruct hook.
1128  * The tunnel context is deleted only when all session sockets have been
1129  * closed.
1130  */
1131 static void l2tp_tunnel_destruct(struct sock *sk)
1132 {
1133         struct l2tp_tunnel *tunnel = l2tp_sk_to_tunnel(sk);
1134
1135         if (!tunnel)
1136                 goto end;
1137
1138         /* Disable udp encapsulation */
1139         switch (tunnel->encap) {
1140         case L2TP_ENCAPTYPE_UDP:
1141                 /* No longer an encapsulation socket. See net/ipv4/udp.c */
1142                 (udp_sk(sk))->encap_type = 0;
1143                 (udp_sk(sk))->encap_rcv = NULL;
1144                 (udp_sk(sk))->encap_destroy = NULL;
1145                 break;
1146         case L2TP_ENCAPTYPE_IP:
1147                 break;
1148         }
1149
1150         /* Remove hooks into tunnel socket */
1151         sk->sk_destruct = tunnel->old_sk_destruct;
1152         sk->sk_user_data = NULL;
1153
1154         /* Call the original destructor */
1155         if (sk->sk_destruct)
1156                 (*sk->sk_destruct)(sk);
1157
1158         kfree_rcu(tunnel, rcu);
1159 end:
1160         return;
1161 }
1162
1163 /* Remove an l2tp session from l2tp_core's hash lists. */
1164 static void l2tp_session_unhash(struct l2tp_session *session)
1165 {
1166         struct l2tp_tunnel *tunnel = session->tunnel;
1167
1168         /* Remove the session from core hashes */
1169         if (tunnel) {
1170                 /* Remove from the per-tunnel hash */
1171                 write_lock_bh(&tunnel->hlist_lock);
1172                 hlist_del_init(&session->hlist);
1173                 write_unlock_bh(&tunnel->hlist_lock);
1174
1175                 /* For L2TPv3 we have a per-net hash: remove from there, too */
1176                 if (tunnel->version != L2TP_HDR_VER_2) {
1177                         struct l2tp_net *pn = l2tp_pernet(tunnel->l2tp_net);
1178
1179                         spin_lock_bh(&pn->l2tp_session_hlist_lock);
1180                         hlist_del_init_rcu(&session->global_hlist);
1181                         spin_unlock_bh(&pn->l2tp_session_hlist_lock);
1182                         synchronize_rcu();
1183                 }
1184         }
1185 }
1186
1187 /* When the tunnel is closed, all the attached sessions need to go too.
1188  */
1189 static void l2tp_tunnel_closeall(struct l2tp_tunnel *tunnel)
1190 {
1191         int hash;
1192         struct hlist_node *walk;
1193         struct hlist_node *tmp;
1194         struct l2tp_session *session;
1195
1196         write_lock_bh(&tunnel->hlist_lock);
1197         tunnel->acpt_newsess = false;
1198         for (hash = 0; hash < L2TP_HASH_SIZE; hash++) {
1199 again:
1200                 hlist_for_each_safe(walk, tmp, &tunnel->session_hlist[hash]) {
1201                         session = hlist_entry(walk, struct l2tp_session, hlist);
1202                         hlist_del_init(&session->hlist);
1203
1204                         write_unlock_bh(&tunnel->hlist_lock);
1205                         l2tp_session_delete(session);
1206                         write_lock_bh(&tunnel->hlist_lock);
1207
1208                         /* Now restart from the beginning of this hash
1209                          * chain.  We always remove a session from the
1210                          * list so we are guaranteed to make forward
1211                          * progress.
1212                          */
1213                         goto again;
1214                 }
1215         }
1216         write_unlock_bh(&tunnel->hlist_lock);
1217 }
1218
1219 /* Tunnel socket destroy hook for UDP encapsulation */
1220 static void l2tp_udp_encap_destroy(struct sock *sk)
1221 {
1222         struct l2tp_tunnel *tunnel = l2tp_sk_to_tunnel(sk);
1223
1224         if (tunnel)
1225                 l2tp_tunnel_delete(tunnel);
1226 }
1227
1228 /* Workqueue tunnel deletion function */
1229 static void l2tp_tunnel_del_work(struct work_struct *work)
1230 {
1231         struct l2tp_tunnel *tunnel = container_of(work, struct l2tp_tunnel,
1232                                                   del_work);
1233         struct sock *sk = tunnel->sock;
1234         struct socket *sock = sk->sk_socket;
1235         struct l2tp_net *pn;
1236
1237         l2tp_tunnel_closeall(tunnel);
1238
1239         /* If the tunnel socket was created within the kernel, use
1240          * the sk API to release it here.
1241          */
1242         if (tunnel->fd < 0) {
1243                 if (sock) {
1244                         kernel_sock_shutdown(sock, SHUT_RDWR);
1245                         sock_release(sock);
1246                 }
1247         }
1248
1249         /* Remove the tunnel struct from the tunnel list */
1250         pn = l2tp_pernet(tunnel->l2tp_net);
1251         spin_lock_bh(&pn->l2tp_tunnel_list_lock);
1252         list_del_rcu(&tunnel->list);
1253         spin_unlock_bh(&pn->l2tp_tunnel_list_lock);
1254
1255         /* drop initial ref */
1256         l2tp_tunnel_dec_refcount(tunnel);
1257
1258         /* drop workqueue ref */
1259         l2tp_tunnel_dec_refcount(tunnel);
1260 }
1261
1262 /* Create a socket for the tunnel, if one isn't set up by
1263  * userspace. This is used for static tunnels where there is no
1264  * managing L2TP daemon.
1265  *
1266  * Since we don't want these sockets to keep a namespace alive by
1267  * themselves, we drop the socket's namespace refcount after creation.
1268  * These sockets are freed when the namespace exits using the pernet
1269  * exit hook.
1270  */
1271 static int l2tp_tunnel_sock_create(struct net *net,
1272                                    u32 tunnel_id,
1273                                    u32 peer_tunnel_id,
1274                                    struct l2tp_tunnel_cfg *cfg,
1275                                    struct socket **sockp)
1276 {
1277         int err = -EINVAL;
1278         struct socket *sock = NULL;
1279         struct udp_port_cfg udp_conf;
1280
1281         switch (cfg->encap) {
1282         case L2TP_ENCAPTYPE_UDP:
1283                 memset(&udp_conf, 0, sizeof(udp_conf));
1284
1285 #if IS_ENABLED(CONFIG_IPV6)
1286                 if (cfg->local_ip6 && cfg->peer_ip6) {
1287                         udp_conf.family = AF_INET6;
1288                         memcpy(&udp_conf.local_ip6, cfg->local_ip6,
1289                                sizeof(udp_conf.local_ip6));
1290                         memcpy(&udp_conf.peer_ip6, cfg->peer_ip6,
1291                                sizeof(udp_conf.peer_ip6));
1292                         udp_conf.use_udp6_tx_checksums =
1293                           !cfg->udp6_zero_tx_checksums;
1294                         udp_conf.use_udp6_rx_checksums =
1295                           !cfg->udp6_zero_rx_checksums;
1296                 } else
1297 #endif
1298                 {
1299                         udp_conf.family = AF_INET;
1300                         udp_conf.local_ip = cfg->local_ip;
1301                         udp_conf.peer_ip = cfg->peer_ip;
1302                         udp_conf.use_udp_checksums = cfg->use_udp_checksums;
1303                 }
1304
1305                 udp_conf.local_udp_port = htons(cfg->local_udp_port);
1306                 udp_conf.peer_udp_port = htons(cfg->peer_udp_port);
1307
1308                 err = udp_sock_create(net, &udp_conf, &sock);
1309                 if (err < 0)
1310                         goto out;
1311
1312                 break;
1313
1314         case L2TP_ENCAPTYPE_IP:
1315 #if IS_ENABLED(CONFIG_IPV6)
1316                 if (cfg->local_ip6 && cfg->peer_ip6) {
1317                         struct sockaddr_l2tpip6 ip6_addr = {0};
1318
1319                         err = sock_create_kern(net, AF_INET6, SOCK_DGRAM,
1320                                                IPPROTO_L2TP, &sock);
1321                         if (err < 0)
1322                                 goto out;
1323
1324                         ip6_addr.l2tp_family = AF_INET6;
1325                         memcpy(&ip6_addr.l2tp_addr, cfg->local_ip6,
1326                                sizeof(ip6_addr.l2tp_addr));
1327                         ip6_addr.l2tp_conn_id = tunnel_id;
1328                         err = kernel_bind(sock, (struct sockaddr *)&ip6_addr,
1329                                           sizeof(ip6_addr));
1330                         if (err < 0)
1331                                 goto out;
1332
1333                         ip6_addr.l2tp_family = AF_INET6;
1334                         memcpy(&ip6_addr.l2tp_addr, cfg->peer_ip6,
1335                                sizeof(ip6_addr.l2tp_addr));
1336                         ip6_addr.l2tp_conn_id = peer_tunnel_id;
1337                         err = kernel_connect(sock,
1338                                              (struct sockaddr *)&ip6_addr,
1339                                              sizeof(ip6_addr), 0);
1340                         if (err < 0)
1341                                 goto out;
1342                 } else
1343 #endif
1344                 {
1345                         struct sockaddr_l2tpip ip_addr = {0};
1346
1347                         err = sock_create_kern(net, AF_INET, SOCK_DGRAM,
1348                                                IPPROTO_L2TP, &sock);
1349                         if (err < 0)
1350                                 goto out;
1351
1352                         ip_addr.l2tp_family = AF_INET;
1353                         ip_addr.l2tp_addr = cfg->local_ip;
1354                         ip_addr.l2tp_conn_id = tunnel_id;
1355                         err = kernel_bind(sock, (struct sockaddr *)&ip_addr,
1356                                           sizeof(ip_addr));
1357                         if (err < 0)
1358                                 goto out;
1359
1360                         ip_addr.l2tp_family = AF_INET;
1361                         ip_addr.l2tp_addr = cfg->peer_ip;
1362                         ip_addr.l2tp_conn_id = peer_tunnel_id;
1363                         err = kernel_connect(sock, (struct sockaddr *)&ip_addr,
1364                                              sizeof(ip_addr), 0);
1365                         if (err < 0)
1366                                 goto out;
1367                 }
1368                 break;
1369
1370         default:
1371                 goto out;
1372         }
1373
1374 out:
1375         *sockp = sock;
1376         if (err < 0 && sock) {
1377                 kernel_sock_shutdown(sock, SHUT_RDWR);
1378                 sock_release(sock);
1379                 *sockp = NULL;
1380         }
1381
1382         return err;
1383 }
1384
1385 static struct lock_class_key l2tp_socket_class;
1386
1387 int l2tp_tunnel_create(int fd, int version, u32 tunnel_id, u32 peer_tunnel_id,
1388                        struct l2tp_tunnel_cfg *cfg, struct l2tp_tunnel **tunnelp)
1389 {
1390         struct l2tp_tunnel *tunnel = NULL;
1391         int err;
1392         enum l2tp_encap_type encap = L2TP_ENCAPTYPE_UDP;
1393
1394         if (cfg)
1395                 encap = cfg->encap;
1396
1397         tunnel = kzalloc(sizeof(*tunnel), GFP_KERNEL);
1398         if (!tunnel) {
1399                 err = -ENOMEM;
1400                 goto err;
1401         }
1402
1403         tunnel->version = version;
1404         tunnel->tunnel_id = tunnel_id;
1405         tunnel->peer_tunnel_id = peer_tunnel_id;
1406
1407         tunnel->magic = L2TP_TUNNEL_MAGIC;
1408         sprintf(&tunnel->name[0], "tunl %u", tunnel_id);
1409         rwlock_init(&tunnel->hlist_lock);
1410         tunnel->acpt_newsess = true;
1411
1412         tunnel->encap = encap;
1413
1414         refcount_set(&tunnel->ref_count, 1);
1415         tunnel->fd = fd;
1416
1417         /* Init delete workqueue struct */
1418         INIT_WORK(&tunnel->del_work, l2tp_tunnel_del_work);
1419
1420         INIT_LIST_HEAD(&tunnel->list);
1421
1422         err = 0;
1423 err:
1424         if (tunnelp)
1425                 *tunnelp = tunnel;
1426
1427         return err;
1428 }
1429 EXPORT_SYMBOL_GPL(l2tp_tunnel_create);
1430
1431 static int l2tp_validate_socket(const struct sock *sk, const struct net *net,
1432                                 enum l2tp_encap_type encap)
1433 {
1434         if (!net_eq(sock_net(sk), net))
1435                 return -EINVAL;
1436
1437         if (sk->sk_type != SOCK_DGRAM)
1438                 return -EPROTONOSUPPORT;
1439
1440         if (sk->sk_family != PF_INET && sk->sk_family != PF_INET6)
1441                 return -EPROTONOSUPPORT;
1442
1443         if ((encap == L2TP_ENCAPTYPE_UDP && sk->sk_protocol != IPPROTO_UDP) ||
1444             (encap == L2TP_ENCAPTYPE_IP && sk->sk_protocol != IPPROTO_L2TP))
1445                 return -EPROTONOSUPPORT;
1446
1447         if (sk->sk_user_data)
1448                 return -EBUSY;
1449
1450         return 0;
1451 }
1452
1453 int l2tp_tunnel_register(struct l2tp_tunnel *tunnel, struct net *net,
1454                          struct l2tp_tunnel_cfg *cfg)
1455 {
1456         struct l2tp_tunnel *tunnel_walk;
1457         struct l2tp_net *pn;
1458         struct socket *sock;
1459         struct sock *sk;
1460         int ret;
1461
1462         if (tunnel->fd < 0) {
1463                 ret = l2tp_tunnel_sock_create(net, tunnel->tunnel_id,
1464                                               tunnel->peer_tunnel_id, cfg,
1465                                               &sock);
1466                 if (ret < 0)
1467                         goto err;
1468         } else {
1469                 sock = sockfd_lookup(tunnel->fd, &ret);
1470                 if (!sock)
1471                         goto err;
1472
1473                 ret = l2tp_validate_socket(sock->sk, net, tunnel->encap);
1474                 if (ret < 0)
1475                         goto err_sock;
1476         }
1477
1478         tunnel->l2tp_net = net;
1479         pn = l2tp_pernet(net);
1480
1481         sk = sock->sk;
1482         sock_hold(sk);
1483         tunnel->sock = sk;
1484
1485         spin_lock_bh(&pn->l2tp_tunnel_list_lock);
1486         list_for_each_entry(tunnel_walk, &pn->l2tp_tunnel_list, list) {
1487                 if (tunnel_walk->tunnel_id == tunnel->tunnel_id) {
1488                         spin_unlock_bh(&pn->l2tp_tunnel_list_lock);
1489                         sock_put(sk);
1490                         ret = -EEXIST;
1491                         goto err_sock;
1492                 }
1493         }
1494         list_add_rcu(&tunnel->list, &pn->l2tp_tunnel_list);
1495         spin_unlock_bh(&pn->l2tp_tunnel_list_lock);
1496
1497         if (tunnel->encap == L2TP_ENCAPTYPE_UDP) {
1498                 struct udp_tunnel_sock_cfg udp_cfg = {
1499                         .sk_user_data = tunnel,
1500                         .encap_type = UDP_ENCAP_L2TPINUDP,
1501                         .encap_rcv = l2tp_udp_encap_recv,
1502                         .encap_destroy = l2tp_udp_encap_destroy,
1503                 };
1504
1505                 setup_udp_tunnel_sock(net, sock, &udp_cfg);
1506         } else {
1507                 sk->sk_user_data = tunnel;
1508         }
1509
1510         tunnel->old_sk_destruct = sk->sk_destruct;
1511         sk->sk_destruct = &l2tp_tunnel_destruct;
1512         lockdep_set_class_and_name(&sk->sk_lock.slock, &l2tp_socket_class,
1513                                    "l2tp_sock");
1514         sk->sk_allocation = GFP_ATOMIC;
1515
1516         trace_register_tunnel(tunnel);
1517
1518         if (tunnel->fd >= 0)
1519                 sockfd_put(sock);
1520
1521         return 0;
1522
1523 err_sock:
1524         if (tunnel->fd < 0)
1525                 sock_release(sock);
1526         else
1527                 sockfd_put(sock);
1528 err:
1529         return ret;
1530 }
1531 EXPORT_SYMBOL_GPL(l2tp_tunnel_register);
1532
1533 /* This function is used by the netlink TUNNEL_DELETE command.
1534  */
1535 void l2tp_tunnel_delete(struct l2tp_tunnel *tunnel)
1536 {
1537         if (!test_and_set_bit(0, &tunnel->dead)) {
1538                 trace_delete_tunnel(tunnel);
1539                 l2tp_tunnel_inc_refcount(tunnel);
1540                 queue_work(l2tp_wq, &tunnel->del_work);
1541         }
1542 }
1543 EXPORT_SYMBOL_GPL(l2tp_tunnel_delete);
1544
1545 void l2tp_session_delete(struct l2tp_session *session)
1546 {
1547         if (test_and_set_bit(0, &session->dead))
1548                 return;
1549
1550         trace_delete_session(session);
1551         l2tp_session_unhash(session);
1552         l2tp_session_queue_purge(session);
1553         if (session->session_close)
1554                 (*session->session_close)(session);
1555
1556         l2tp_session_dec_refcount(session);
1557 }
1558 EXPORT_SYMBOL_GPL(l2tp_session_delete);
1559
1560 /* We come here whenever a session's send_seq, cookie_len or
1561  * l2specific_type parameters are set.
1562  */
1563 void l2tp_session_set_header_len(struct l2tp_session *session, int version)
1564 {
1565         if (version == L2TP_HDR_VER_2) {
1566                 session->hdr_len = 6;
1567                 if (session->send_seq)
1568                         session->hdr_len += 4;
1569         } else {
1570                 session->hdr_len = 4 + session->cookie_len;
1571                 session->hdr_len += l2tp_get_l2specific_len(session);
1572                 if (session->tunnel->encap == L2TP_ENCAPTYPE_UDP)
1573                         session->hdr_len += 4;
1574         }
1575 }
1576 EXPORT_SYMBOL_GPL(l2tp_session_set_header_len);
1577
1578 struct l2tp_session *l2tp_session_create(int priv_size, struct l2tp_tunnel *tunnel, u32 session_id,
1579                                          u32 peer_session_id, struct l2tp_session_cfg *cfg)
1580 {
1581         struct l2tp_session *session;
1582
1583         session = kzalloc(sizeof(*session) + priv_size, GFP_KERNEL);
1584         if (session) {
1585                 session->magic = L2TP_SESSION_MAGIC;
1586                 session->tunnel = tunnel;
1587
1588                 session->session_id = session_id;
1589                 session->peer_session_id = peer_session_id;
1590                 session->nr = 0;
1591                 if (tunnel->version == L2TP_HDR_VER_2)
1592                         session->nr_max = 0xffff;
1593                 else
1594                         session->nr_max = 0xffffff;
1595                 session->nr_window_size = session->nr_max / 2;
1596                 session->nr_oos_count_max = 4;
1597
1598                 /* Use NR of first received packet */
1599                 session->reorder_skip = 1;
1600
1601                 sprintf(&session->name[0], "sess %u/%u",
1602                         tunnel->tunnel_id, session->session_id);
1603
1604                 skb_queue_head_init(&session->reorder_q);
1605
1606                 INIT_HLIST_NODE(&session->hlist);
1607                 INIT_HLIST_NODE(&session->global_hlist);
1608
1609                 if (cfg) {
1610                         session->pwtype = cfg->pw_type;
1611                         session->send_seq = cfg->send_seq;
1612                         session->recv_seq = cfg->recv_seq;
1613                         session->lns_mode = cfg->lns_mode;
1614                         session->reorder_timeout = cfg->reorder_timeout;
1615                         session->l2specific_type = cfg->l2specific_type;
1616                         session->cookie_len = cfg->cookie_len;
1617                         memcpy(&session->cookie[0], &cfg->cookie[0], cfg->cookie_len);
1618                         session->peer_cookie_len = cfg->peer_cookie_len;
1619                         memcpy(&session->peer_cookie[0], &cfg->peer_cookie[0], cfg->peer_cookie_len);
1620                 }
1621
1622                 l2tp_session_set_header_len(session, tunnel->version);
1623
1624                 refcount_set(&session->ref_count, 1);
1625
1626                 return session;
1627         }
1628
1629         return ERR_PTR(-ENOMEM);
1630 }
1631 EXPORT_SYMBOL_GPL(l2tp_session_create);
1632
1633 /*****************************************************************************
1634  * Init and cleanup
1635  *****************************************************************************/
1636
1637 static __net_init int l2tp_init_net(struct net *net)
1638 {
1639         struct l2tp_net *pn = net_generic(net, l2tp_net_id);
1640         int hash;
1641
1642         INIT_LIST_HEAD(&pn->l2tp_tunnel_list);
1643         spin_lock_init(&pn->l2tp_tunnel_list_lock);
1644
1645         for (hash = 0; hash < L2TP_HASH_SIZE_2; hash++)
1646                 INIT_HLIST_HEAD(&pn->l2tp_session_hlist[hash]);
1647
1648         spin_lock_init(&pn->l2tp_session_hlist_lock);
1649
1650         return 0;
1651 }
1652
1653 static __net_exit void l2tp_exit_net(struct net *net)
1654 {
1655         struct l2tp_net *pn = l2tp_pernet(net);
1656         struct l2tp_tunnel *tunnel = NULL;
1657         int hash;
1658
1659         rcu_read_lock_bh();
1660         list_for_each_entry_rcu(tunnel, &pn->l2tp_tunnel_list, list) {
1661                 l2tp_tunnel_delete(tunnel);
1662         }
1663         rcu_read_unlock_bh();
1664
1665         if (l2tp_wq)
1666                 flush_workqueue(l2tp_wq);
1667         rcu_barrier();
1668
1669         for (hash = 0; hash < L2TP_HASH_SIZE_2; hash++)
1670                 WARN_ON_ONCE(!hlist_empty(&pn->l2tp_session_hlist[hash]));
1671 }
1672
1673 static struct pernet_operations l2tp_net_ops = {
1674         .init = l2tp_init_net,
1675         .exit = l2tp_exit_net,
1676         .id   = &l2tp_net_id,
1677         .size = sizeof(struct l2tp_net),
1678 };
1679
1680 static int __init l2tp_init(void)
1681 {
1682         int rc = 0;
1683
1684         rc = register_pernet_device(&l2tp_net_ops);
1685         if (rc)
1686                 goto out;
1687
1688         l2tp_wq = alloc_workqueue("l2tp", WQ_UNBOUND, 0);
1689         if (!l2tp_wq) {
1690                 pr_err("alloc_workqueue failed\n");
1691                 unregister_pernet_device(&l2tp_net_ops);
1692                 rc = -ENOMEM;
1693                 goto out;
1694         }
1695
1696         pr_info("L2TP core driver, %s\n", L2TP_DRV_VERSION);
1697
1698 out:
1699         return rc;
1700 }
1701
1702 static void __exit l2tp_exit(void)
1703 {
1704         unregister_pernet_device(&l2tp_net_ops);
1705         if (l2tp_wq) {
1706                 destroy_workqueue(l2tp_wq);
1707                 l2tp_wq = NULL;
1708         }
1709 }
1710
1711 module_init(l2tp_init);
1712 module_exit(l2tp_exit);
1713
1714 MODULE_AUTHOR("James Chapman <jchapman@katalix.com>");
1715 MODULE_DESCRIPTION("L2TP core");
1716 MODULE_LICENSE("GPL");
1717 MODULE_VERSION(L2TP_DRV_VERSION);