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_warn_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_warn_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_warn_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 verifed 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_warn_ratelimited("%s: recv short packet (len=%d)\n",
 813                                     tunnel->name, skb->len);
 814                 goto error;
 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_warn_ratelimited("%s: recv protocol version mismatch: got %d expected %d\n",
 828                                     tunnel->name, version, tunnel->version);
 829                 goto error;
 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 error;
 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_warn_ratelimited("%s: no session found (%u/%u). Passing up.\n",
 867                                     tunnel->name, tunnel_id, session_id);
 868                 goto error;
 869         }
 870
 871         if (tunnel->version == L2TP_HDR_VER_3 &&
 872             l2tp_v3_ensure_opt_in_linear(session, skb, &ptr, &optr))
 873                 goto error;
 874
 875         l2tp_recv_common(session, skb, ptr, optr, hdrflags, length);
 876         l2tp_session_dec_refcount(session);
 877
 878         return 0;
 879
 880 error:
 881         /* Put UDP header back */
 882         __skb_push(skb, sizeof(struct udphdr));
 883
 884         return 1;
 885 }
 886
 887 /* UDP encapsulation receive handler. See net/ipv4/udp.c.
 888  * Return codes:
 889  * 0 : success.
 890  * <0: error
 891  * >0: skb should be passed up to userspace as UDP.
 892  */
 893 int l2tp_udp_encap_recv(struct sock *sk, struct sk_buff *skb)
 894 {
 895         struct l2tp_tunnel *tunnel;
 896
 897         /* Note that this is called from the encap_rcv hook inside an
 898          * RCU-protected region, but without the socket being locked.
 899          * Hence we use rcu_dereference_sk_user_data to access the
 900          * tunnel data structure rather the usual l2tp_sk_to_tunnel
 901          * accessor function.
 902          */
 903         tunnel = rcu_dereference_sk_user_data(sk);
 904         if (!tunnel)
 905                 goto pass_up;
 906         if (WARN_ON(tunnel->magic != L2TP_TUNNEL_MAGIC))
 907                 goto pass_up;
 908
 909         if (l2tp_udp_recv_core(tunnel, skb))
 910                 goto pass_up;
 911
 912         return 0;
 913
 914 pass_up:
 915         return 1;
 916 }
 917 EXPORT_SYMBOL_GPL(l2tp_udp_encap_recv);
 918
 919 /************************************************************************
 920  * Transmit handling
 921  ***********************************************************************/
 922
 923 /* Build an L2TP header for the session into the buffer provided.
 924  */
 925 static int l2tp_build_l2tpv2_header(struct l2tp_session *session, void *buf)
 926 {
 927         struct l2tp_tunnel *tunnel = session->tunnel;
 928         __be16 *bufp = buf;
 929         __be16 *optr = buf;
 930         u16 flags = L2TP_HDR_VER_2;
 931         u32 tunnel_id = tunnel->peer_tunnel_id;
 932         u32 session_id = session->peer_session_id;
 933
 934         if (session->send_seq)
 935                 flags |= L2TP_HDRFLAG_S;
 936
 937         /* Setup L2TP header. */
 938         *bufp++ = htons(flags);
 939         *bufp++ = htons(tunnel_id);
 940         *bufp++ = htons(session_id);
 941         if (session->send_seq) {
 942                 *bufp++ = htons(session->ns);
 943                 *bufp++ = 0;
 944                 session->ns++;
 945                 session->ns &= 0xffff;
 946                 trace_session_seqnum_update(session);
 947         }
 948
 949         return bufp - optr;
 950 }
 951
 952 static int l2tp_build_l2tpv3_header(struct l2tp_session *session, void *buf)
 953 {
 954         struct l2tp_tunnel *tunnel = session->tunnel;
 955         char *bufp = buf;
 956         char *optr = bufp;
 957
 958         /* Setup L2TP header. The header differs slightly for UDP and
 959          * IP encapsulations. For UDP, there is 4 bytes of flags.
 960          */
 961         if (tunnel->encap == L2TP_ENCAPTYPE_UDP) {
 962                 u16 flags = L2TP_HDR_VER_3;
 963                 *((__be16 *)bufp) = htons(flags);
 964                 bufp += 2;
 965                 *((__be16 *)bufp) = 0;
 966                 bufp += 2;
 967         }
 968
 969         *((__be32 *)bufp) = htonl(session->peer_session_id);
 970         bufp += 4;
 971         if (session->cookie_len) {
 972                 memcpy(bufp, &session->cookie[0], session->cookie_len);
 973                 bufp += session->cookie_len;
 974         }
 975         if (session->l2specific_type == L2TP_L2SPECTYPE_DEFAULT) {
 976                 u32 l2h = 0;
 977
 978                 if (session->send_seq) {
 979                         l2h = 0x40000000 | session->ns;
 980                         session->ns++;
 981                         session->ns &= 0xffffff;
 982                         trace_session_seqnum_update(session);
 983                 }
 984
 985                 *((__be32 *)bufp) = htonl(l2h);
 986                 bufp += 4;
 987         }
 988
 989         return bufp - optr;
 990 }
 991
 992 /* Queue the packet to IP for output: tunnel socket lock must be held */
 993 static int l2tp_xmit_queue(struct l2tp_tunnel *tunnel, struct sk_buff *skb, struct flowi *fl)
 994 {
 995         int err;
 996
 997         skb->ignore_df = 1;
 998         skb_dst_drop(skb);
 999 #if IS_ENABLED(CONFIG_IPV6)
1000         if (l2tp_sk_is_v6(tunnel->sock))
1001                 err = inet6_csk_xmit(tunnel->sock, skb, NULL);
1002         else
1003 #endif
1004                 err = ip_queue_xmit(tunnel->sock, skb, fl);
1005
1006         return err >= 0 ? NET_XMIT_SUCCESS : NET_XMIT_DROP;
1007 }
1008
1009 static int l2tp_xmit_core(struct l2tp_session *session, struct sk_buff *skb, unsigned int *len)
1010 {
1011         struct l2tp_tunnel *tunnel = session->tunnel;
1012         unsigned int data_len = skb->len;
1013         struct sock *sk = tunnel->sock;
1014         int headroom, uhlen, udp_len;
1015         int ret = NET_XMIT_SUCCESS;
1016         struct inet_sock *inet;
1017         struct udphdr *uh;
1018
1019         /* Check that there's enough headroom in the skb to insert IP,
1020          * UDP and L2TP headers. If not enough, expand it to
1021          * make room. Adjust truesize.
1022          */
1023         uhlen = (tunnel->encap == L2TP_ENCAPTYPE_UDP) ? sizeof(*uh) : 0;
1024         headroom = NET_SKB_PAD + sizeof(struct iphdr) + uhlen + session->hdr_len;
1025         if (skb_cow_head(skb, headroom)) {
1026                 kfree_skb(skb);
1027                 return NET_XMIT_DROP;
1028         }
1029
1030         /* Setup L2TP header */
1031         if (tunnel->version == L2TP_HDR_VER_2)
1032                 l2tp_build_l2tpv2_header(session, __skb_push(skb, session->hdr_len));
1033         else
1034                 l2tp_build_l2tpv3_header(session, __skb_push(skb, session->hdr_len));
1035
1036         /* Reset skb netfilter state */
1037         memset(&(IPCB(skb)->opt), 0, sizeof(IPCB(skb)->opt));
1038         IPCB(skb)->flags &= ~(IPSKB_XFRM_TUNNEL_SIZE | IPSKB_XFRM_TRANSFORMED | IPSKB_REROUTED);
1039         nf_reset_ct(skb);
1040
1041         bh_lock_sock(sk);
1042         if (sock_owned_by_user(sk)) {
1043                 kfree_skb(skb);
1044                 ret = NET_XMIT_DROP;
1045                 goto out_unlock;
1046         }
1047
1048         /* The user-space may change the connection status for the user-space
1049          * provided socket at run time: we must check it under the socket lock
1050          */
1051         if (tunnel->fd >= 0 && sk->sk_state != TCP_ESTABLISHED) {
1052                 kfree_skb(skb);
1053                 ret = NET_XMIT_DROP;
1054                 goto out_unlock;
1055         }
1056
1057         /* Report transmitted length before we add encap header, which keeps
1058          * statistics consistent for both UDP and IP encap tx/rx paths.
1059          */
1060         *len = skb->len;
1061
1062         inet = inet_sk(sk);
1063         switch (tunnel->encap) {
1064         case L2TP_ENCAPTYPE_UDP:
1065                 /* Setup UDP header */
1066                 __skb_push(skb, sizeof(*uh));
1067                 skb_reset_transport_header(skb);
1068                 uh = udp_hdr(skb);
1069                 uh->source = inet->inet_sport;
1070                 uh->dest = inet->inet_dport;
1071                 udp_len = uhlen + session->hdr_len + data_len;
1072                 uh->len = htons(udp_len);
1073
1074                 /* Calculate UDP checksum if configured to do so */
1075 #if IS_ENABLED(CONFIG_IPV6)
1076                 if (l2tp_sk_is_v6(sk))
1077                         udp6_set_csum(udp_get_no_check6_tx(sk),
1078                                       skb, &inet6_sk(sk)->saddr,
1079                                       &sk->sk_v6_daddr, udp_len);
1080                 else
1081 #endif
1082                         udp_set_csum(sk->sk_no_check_tx, skb, inet->inet_saddr,
1083                                      inet->inet_daddr, udp_len);
1084                 break;
1085
1086         case L2TP_ENCAPTYPE_IP:
1087                 break;
1088         }
1089
1090         ret = l2tp_xmit_queue(tunnel, skb, &inet->cork.fl);
1091
1092 out_unlock:
1093         bh_unlock_sock(sk);
1094
1095         return ret;
1096 }
1097
1098 /* If caller requires the skb to have a ppp header, the header must be
1099  * inserted in the skb data before calling this function.
1100  */
1101 int l2tp_xmit_skb(struct l2tp_session *session, struct sk_buff *skb)
1102 {
1103         unsigned int len = 0;
1104         int ret;
1105
1106         ret = l2tp_xmit_core(session, skb, &len);
1107         if (ret == NET_XMIT_SUCCESS) {
1108                 atomic_long_inc(&session->tunnel->stats.tx_packets);
1109                 atomic_long_add(len, &session->tunnel->stats.tx_bytes);
1110                 atomic_long_inc(&session->stats.tx_packets);
1111                 atomic_long_add(len, &session->stats.tx_bytes);
1112         } else {
1113                 atomic_long_inc(&session->tunnel->stats.tx_errors);
1114                 atomic_long_inc(&session->stats.tx_errors);
1115         }
1116         return ret;
1117 }
1118 EXPORT_SYMBOL_GPL(l2tp_xmit_skb);
1119
1120 /*****************************************************************************
1121  * Tinnel and session create/destroy.
1122  *****************************************************************************/
1123
1124 /* Tunnel socket destruct hook.
1125  * The tunnel context is deleted only when all session sockets have been
1126  * closed.
1127  */
1128 static void l2tp_tunnel_destruct(struct sock *sk)
1129 {
1130         struct l2tp_tunnel *tunnel = l2tp_sk_to_tunnel(sk);
1131
1132         if (!tunnel)
1133                 goto end;
1134
1135         /* Disable udp encapsulation */
1136         switch (tunnel->encap) {
1137         case L2TP_ENCAPTYPE_UDP:
1138                 /* No longer an encapsulation socket. See net/ipv4/udp.c */
1139                 (udp_sk(sk))->encap_type = 0;
1140                 (udp_sk(sk))->encap_rcv = NULL;
1141                 (udp_sk(sk))->encap_destroy = NULL;
1142                 break;
1143         case L2TP_ENCAPTYPE_IP:
1144                 break;
1145         }
1146
1147         /* Remove hooks into tunnel socket */
1148         sk->sk_destruct = tunnel->old_sk_destruct;
1149         sk->sk_user_data = NULL;
1150
1151         /* Call the original destructor */
1152         if (sk->sk_destruct)
1153                 (*sk->sk_destruct)(sk);
1154
1155         kfree_rcu(tunnel, rcu);
1156 end:
1157         return;
1158 }
1159
1160 /* Remove an l2tp session from l2tp_core's hash lists. */
1161 static void l2tp_session_unhash(struct l2tp_session *session)
1162 {
1163         struct l2tp_tunnel *tunnel = session->tunnel;
1164
1165         /* Remove the session from core hashes */
1166         if (tunnel) {
1167                 /* Remove from the per-tunnel hash */
1168                 write_lock_bh(&tunnel->hlist_lock);
1169                 hlist_del_init(&session->hlist);
1170                 write_unlock_bh(&tunnel->hlist_lock);
1171
1172                 /* For L2TPv3 we have a per-net hash: remove from there, too */
1173                 if (tunnel->version != L2TP_HDR_VER_2) {
1174                         struct l2tp_net *pn = l2tp_pernet(tunnel->l2tp_net);
1175
1176                         spin_lock_bh(&pn->l2tp_session_hlist_lock);
1177                         hlist_del_init_rcu(&session->global_hlist);
1178                         spin_unlock_bh(&pn->l2tp_session_hlist_lock);
1179                         synchronize_rcu();
1180                 }
1181         }
1182 }
1183
1184 /* When the tunnel is closed, all the attached sessions need to go too.
1185  */
1186 static void l2tp_tunnel_closeall(struct l2tp_tunnel *tunnel)
1187 {
1188         int hash;
1189         struct hlist_node *walk;
1190         struct hlist_node *tmp;
1191         struct l2tp_session *session;
1192
1193         write_lock_bh(&tunnel->hlist_lock);
1194         tunnel->acpt_newsess = false;
1195         for (hash = 0; hash < L2TP_HASH_SIZE; hash++) {
1196 again:
1197                 hlist_for_each_safe(walk, tmp, &tunnel->session_hlist[hash]) {
1198                         session = hlist_entry(walk, struct l2tp_session, hlist);
1199                         hlist_del_init(&session->hlist);
1200
1201                         write_unlock_bh(&tunnel->hlist_lock);
1202                         l2tp_session_delete(session);
1203                         write_lock_bh(&tunnel->hlist_lock);
1204
1205                         /* Now restart from the beginning of this hash
1206                          * chain.  We always remove a session from the
1207                          * list so we are guaranteed to make forward
1208                          * progress.
1209                          */
1210                         goto again;
1211                 }
1212         }
1213         write_unlock_bh(&tunnel->hlist_lock);
1214 }
1215
1216 /* Tunnel socket destroy hook for UDP encapsulation */
1217 static void l2tp_udp_encap_destroy(struct sock *sk)
1218 {
1219         struct l2tp_tunnel *tunnel = l2tp_sk_to_tunnel(sk);
1220
1221         if (tunnel)
1222                 l2tp_tunnel_delete(tunnel);
1223 }
1224
1225 /* Workqueue tunnel deletion function */
1226 static void l2tp_tunnel_del_work(struct work_struct *work)
1227 {
1228         struct l2tp_tunnel *tunnel = container_of(work, struct l2tp_tunnel,
1229                                                   del_work);
1230         struct sock *sk = tunnel->sock;
1231         struct socket *sock = sk->sk_socket;
1232         struct l2tp_net *pn;
1233
1234         l2tp_tunnel_closeall(tunnel);
1235
1236         /* If the tunnel socket was created within the kernel, use
1237          * the sk API to release it here.
1238          */
1239         if (tunnel->fd < 0) {
1240                 if (sock) {
1241                         kernel_sock_shutdown(sock, SHUT_RDWR);
1242                         sock_release(sock);
1243                 }
1244         }
1245
1246         /* Remove the tunnel struct from the tunnel list */
1247         pn = l2tp_pernet(tunnel->l2tp_net);
1248         spin_lock_bh(&pn->l2tp_tunnel_list_lock);
1249         list_del_rcu(&tunnel->list);
1250         spin_unlock_bh(&pn->l2tp_tunnel_list_lock);
1251
1252         /* drop initial ref */
1253         l2tp_tunnel_dec_refcount(tunnel);
1254
1255         /* drop workqueue ref */
1256         l2tp_tunnel_dec_refcount(tunnel);
1257 }
1258
1259 /* Create a socket for the tunnel, if one isn't set up by
1260  * userspace. This is used for static tunnels where there is no
1261  * managing L2TP daemon.
1262  *
1263  * Since we don't want these sockets to keep a namespace alive by
1264  * themselves, we drop the socket's namespace refcount after creation.
1265  * These sockets are freed when the namespace exits using the pernet
1266  * exit hook.
1267  */
1268 static int l2tp_tunnel_sock_create(struct net *net,
1269                                    u32 tunnel_id,
1270                                    u32 peer_tunnel_id,
1271                                    struct l2tp_tunnel_cfg *cfg,
1272                                    struct socket **sockp)
1273 {
1274         int err = -EINVAL;
1275         struct socket *sock = NULL;
1276         struct udp_port_cfg udp_conf;
1277
1278         switch (cfg->encap) {
1279         case L2TP_ENCAPTYPE_UDP:
1280                 memset(&udp_conf, 0, sizeof(udp_conf));
1281
1282 #if IS_ENABLED(CONFIG_IPV6)
1283                 if (cfg->local_ip6 && cfg->peer_ip6) {
1284                         udp_conf.family = AF_INET6;
1285                         memcpy(&udp_conf.local_ip6, cfg->local_ip6,
1286                                sizeof(udp_conf.local_ip6));
1287                         memcpy(&udp_conf.peer_ip6, cfg->peer_ip6,
1288                                sizeof(udp_conf.peer_ip6));
1289                         udp_conf.use_udp6_tx_checksums =
1290                           !cfg->udp6_zero_tx_checksums;
1291                         udp_conf.use_udp6_rx_checksums =
1292                           !cfg->udp6_zero_rx_checksums;
1293                 } else
1294 #endif
1295                 {
1296                         udp_conf.family = AF_INET;
1297                         udp_conf.local_ip = cfg->local_ip;
1298                         udp_conf.peer_ip = cfg->peer_ip;
1299                         udp_conf.use_udp_checksums = cfg->use_udp_checksums;
1300                 }
1301
1302                 udp_conf.local_udp_port = htons(cfg->local_udp_port);
1303                 udp_conf.peer_udp_port = htons(cfg->peer_udp_port);
1304
1305                 err = udp_sock_create(net, &udp_conf, &sock);
1306                 if (err < 0)
1307                         goto out;
1308
1309                 break;
1310
1311         case L2TP_ENCAPTYPE_IP:
1312 #if IS_ENABLED(CONFIG_IPV6)
1313                 if (cfg->local_ip6 && cfg->peer_ip6) {
1314                         struct sockaddr_l2tpip6 ip6_addr = {0};
1315
1316                         err = sock_create_kern(net, AF_INET6, SOCK_DGRAM,
1317                                                IPPROTO_L2TP, &sock);
1318                         if (err < 0)
1319                                 goto out;
1320
1321                         ip6_addr.l2tp_family = AF_INET6;
1322                         memcpy(&ip6_addr.l2tp_addr, cfg->local_ip6,
1323                                sizeof(ip6_addr.l2tp_addr));
1324                         ip6_addr.l2tp_conn_id = tunnel_id;
1325                         err = kernel_bind(sock, (struct sockaddr *)&ip6_addr,
1326                                           sizeof(ip6_addr));
1327                         if (err < 0)
1328                                 goto out;
1329
1330                         ip6_addr.l2tp_family = AF_INET6;
1331                         memcpy(&ip6_addr.l2tp_addr, cfg->peer_ip6,
1332                                sizeof(ip6_addr.l2tp_addr));
1333                         ip6_addr.l2tp_conn_id = peer_tunnel_id;
1334                         err = kernel_connect(sock,
1335                                              (struct sockaddr *)&ip6_addr,
1336                                              sizeof(ip6_addr), 0);
1337                         if (err < 0)
1338                                 goto out;
1339                 } else
1340 #endif
1341                 {
1342                         struct sockaddr_l2tpip ip_addr = {0};
1343
1344                         err = sock_create_kern(net, AF_INET, SOCK_DGRAM,
1345                                                IPPROTO_L2TP, &sock);
1346                         if (err < 0)
1347                                 goto out;
1348
1349                         ip_addr.l2tp_family = AF_INET;
1350                         ip_addr.l2tp_addr = cfg->local_ip;
1351                         ip_addr.l2tp_conn_id = tunnel_id;
1352                         err = kernel_bind(sock, (struct sockaddr *)&ip_addr,
1353                                           sizeof(ip_addr));
1354                         if (err < 0)
1355                                 goto out;
1356
1357                         ip_addr.l2tp_family = AF_INET;
1358                         ip_addr.l2tp_addr = cfg->peer_ip;
1359                         ip_addr.l2tp_conn_id = peer_tunnel_id;
1360                         err = kernel_connect(sock, (struct sockaddr *)&ip_addr,
1361                                              sizeof(ip_addr), 0);
1362                         if (err < 0)
1363                                 goto out;
1364                 }
1365                 break;
1366
1367         default:
1368                 goto out;
1369         }
1370
1371 out:
1372         *sockp = sock;
1373         if (err < 0 && sock) {
1374                 kernel_sock_shutdown(sock, SHUT_RDWR);
1375                 sock_release(sock);
1376                 *sockp = NULL;
1377         }
1378
1379         return err;
1380 }
1381
1382 static struct lock_class_key l2tp_socket_class;
1383
1384 int l2tp_tunnel_create(int fd, int version, u32 tunnel_id, u32 peer_tunnel_id,
1385                        struct l2tp_tunnel_cfg *cfg, struct l2tp_tunnel **tunnelp)
1386 {
1387         struct l2tp_tunnel *tunnel = NULL;
1388         int err;
1389         enum l2tp_encap_type encap = L2TP_ENCAPTYPE_UDP;
1390
1391         if (cfg)
1392                 encap = cfg->encap;
1393
1394         tunnel = kzalloc(sizeof(*tunnel), GFP_KERNEL);
1395         if (!tunnel) {
1396                 err = -ENOMEM;
1397                 goto err;
1398         }
1399
1400         tunnel->version = version;
1401         tunnel->tunnel_id = tunnel_id;
1402         tunnel->peer_tunnel_id = peer_tunnel_id;
1403
1404         tunnel->magic = L2TP_TUNNEL_MAGIC;
1405         sprintf(&tunnel->name[0], "tunl %u", tunnel_id);
1406         rwlock_init(&tunnel->hlist_lock);
1407         tunnel->acpt_newsess = true;
1408
1409         tunnel->encap = encap;
1410
1411         refcount_set(&tunnel->ref_count, 1);
1412         tunnel->fd = fd;
1413
1414         /* Init delete workqueue struct */
1415         INIT_WORK(&tunnel->del_work, l2tp_tunnel_del_work);
1416
1417         INIT_LIST_HEAD(&tunnel->list);
1418
1419         err = 0;
1420 err:
1421         if (tunnelp)
1422                 *tunnelp = tunnel;
1423
1424         return err;
1425 }
1426 EXPORT_SYMBOL_GPL(l2tp_tunnel_create);
1427
1428 static int l2tp_validate_socket(const struct sock *sk, const struct net *net,
1429                                 enum l2tp_encap_type encap)
1430 {
1431         if (!net_eq(sock_net(sk), net))
1432                 return -EINVAL;
1433
1434         if (sk->sk_type != SOCK_DGRAM)
1435                 return -EPROTONOSUPPORT;
1436
1437         if (sk->sk_family != PF_INET && sk->sk_family != PF_INET6)
1438                 return -EPROTONOSUPPORT;
1439
1440         if ((encap == L2TP_ENCAPTYPE_UDP && sk->sk_protocol != IPPROTO_UDP) ||
1441             (encap == L2TP_ENCAPTYPE_IP && sk->sk_protocol != IPPROTO_L2TP))
1442                 return -EPROTONOSUPPORT;
1443
1444         if (sk->sk_user_data)
1445                 return -EBUSY;
1446
1447         return 0;
1448 }
1449
1450 int l2tp_tunnel_register(struct l2tp_tunnel *tunnel, struct net *net,
1451                          struct l2tp_tunnel_cfg *cfg)
1452 {
1453         struct l2tp_tunnel *tunnel_walk;
1454         struct l2tp_net *pn;
1455         struct socket *sock;
1456         struct sock *sk;
1457         int ret;
1458
1459         if (tunnel->fd < 0) {
1460                 ret = l2tp_tunnel_sock_create(net, tunnel->tunnel_id,
1461                                               tunnel->peer_tunnel_id, cfg,
1462                                               &sock);
1463                 if (ret < 0)
1464                         goto err;
1465         } else {
1466                 sock = sockfd_lookup(tunnel->fd, &ret);
1467                 if (!sock)
1468                         goto err;
1469
1470                 ret = l2tp_validate_socket(sock->sk, net, tunnel->encap);
1471                 if (ret < 0)
1472                         goto err_sock;
1473         }
1474
1475         tunnel->l2tp_net = net;
1476         pn = l2tp_pernet(net);
1477
1478         spin_lock_bh(&pn->l2tp_tunnel_list_lock);
1479         list_for_each_entry(tunnel_walk, &pn->l2tp_tunnel_list, list) {
1480                 if (tunnel_walk->tunnel_id == tunnel->tunnel_id) {
1481                         spin_unlock_bh(&pn->l2tp_tunnel_list_lock);
1482
1483                         ret = -EEXIST;
1484                         goto err_sock;
1485                 }
1486         }
1487         list_add_rcu(&tunnel->list, &pn->l2tp_tunnel_list);
1488         spin_unlock_bh(&pn->l2tp_tunnel_list_lock);
1489
1490         sk = sock->sk;
1491         sock_hold(sk);
1492         tunnel->sock = sk;
1493
1494         if (tunnel->encap == L2TP_ENCAPTYPE_UDP) {
1495                 struct udp_tunnel_sock_cfg udp_cfg = {
1496                         .sk_user_data = tunnel,
1497                         .encap_type = UDP_ENCAP_L2TPINUDP,
1498                         .encap_rcv = l2tp_udp_encap_recv,
1499                         .encap_destroy = l2tp_udp_encap_destroy,
1500                 };
1501
1502                 setup_udp_tunnel_sock(net, sock, &udp_cfg);
1503         } else {
1504                 sk->sk_user_data = tunnel;
1505         }
1506
1507         tunnel->old_sk_destruct = sk->sk_destruct;
1508         sk->sk_destruct = &l2tp_tunnel_destruct;
1509         lockdep_set_class_and_name(&sk->sk_lock.slock, &l2tp_socket_class,
1510                                    "l2tp_sock");
1511         sk->sk_allocation = GFP_ATOMIC;
1512
1513         trace_register_tunnel(tunnel);
1514
1515         if (tunnel->fd >= 0)
1516                 sockfd_put(sock);
1517
1518         return 0;
1519
1520 err_sock:
1521         if (tunnel->fd < 0)
1522                 sock_release(sock);
1523         else
1524                 sockfd_put(sock);
1525 err:
1526         return ret;
1527 }
1528 EXPORT_SYMBOL_GPL(l2tp_tunnel_register);
1529
1530 /* This function is used by the netlink TUNNEL_DELETE command.
1531  */
1532 void l2tp_tunnel_delete(struct l2tp_tunnel *tunnel)
1533 {
1534         if (!test_and_set_bit(0, &tunnel->dead)) {
1535                 trace_delete_tunnel(tunnel);
1536                 l2tp_tunnel_inc_refcount(tunnel);
1537                 queue_work(l2tp_wq, &tunnel->del_work);
1538         }
1539 }
1540 EXPORT_SYMBOL_GPL(l2tp_tunnel_delete);
1541
1542 void l2tp_session_delete(struct l2tp_session *session)
1543 {
1544         if (test_and_set_bit(0, &session->dead))
1545                 return;
1546
1547         trace_delete_session(session);
1548         l2tp_session_unhash(session);
1549         l2tp_session_queue_purge(session);
1550         if (session->session_close)
1551                 (*session->session_close)(session);
1552
1553         l2tp_session_dec_refcount(session);
1554 }
1555 EXPORT_SYMBOL_GPL(l2tp_session_delete);
1556
1557 /* We come here whenever a session's send_seq, cookie_len or
1558  * l2specific_type parameters are set.
1559  */
1560 void l2tp_session_set_header_len(struct l2tp_session *session, int version)
1561 {
1562         if (version == L2TP_HDR_VER_2) {
1563                 session->hdr_len = 6;
1564                 if (session->send_seq)
1565                         session->hdr_len += 4;
1566         } else {
1567                 session->hdr_len = 4 + session->cookie_len;
1568                 session->hdr_len += l2tp_get_l2specific_len(session);
1569                 if (session->tunnel->encap == L2TP_ENCAPTYPE_UDP)
1570                         session->hdr_len += 4;
1571         }
1572 }
1573 EXPORT_SYMBOL_GPL(l2tp_session_set_header_len);
1574
1575 struct l2tp_session *l2tp_session_create(int priv_size, struct l2tp_tunnel *tunnel, u32 session_id,
1576                                          u32 peer_session_id, struct l2tp_session_cfg *cfg)
1577 {
1578         struct l2tp_session *session;
1579
1580         session = kzalloc(sizeof(*session) + priv_size, GFP_KERNEL);
1581         if (session) {
1582                 session->magic = L2TP_SESSION_MAGIC;
1583                 session->tunnel = tunnel;
1584
1585                 session->session_id = session_id;
1586                 session->peer_session_id = peer_session_id;
1587                 session->nr = 0;
1588                 if (tunnel->version == L2TP_HDR_VER_2)
1589                         session->nr_max = 0xffff;
1590                 else
1591                         session->nr_max = 0xffffff;
1592                 session->nr_window_size = session->nr_max / 2;
1593                 session->nr_oos_count_max = 4;
1594
1595                 /* Use NR of first received packet */
1596                 session->reorder_skip = 1;
1597
1598                 sprintf(&session->name[0], "sess %u/%u",
1599                         tunnel->tunnel_id, session->session_id);
1600
1601                 skb_queue_head_init(&session->reorder_q);
1602
1603                 INIT_HLIST_NODE(&session->hlist);
1604                 INIT_HLIST_NODE(&session->global_hlist);
1605
1606                 if (cfg) {
1607                         session->pwtype = cfg->pw_type;
1608                         session->send_seq = cfg->send_seq;
1609                         session->recv_seq = cfg->recv_seq;
1610                         session->lns_mode = cfg->lns_mode;
1611                         session->reorder_timeout = cfg->reorder_timeout;
1612                         session->l2specific_type = cfg->l2specific_type;
1613                         session->cookie_len = cfg->cookie_len;
1614                         memcpy(&session->cookie[0], &cfg->cookie[0], cfg->cookie_len);
1615                         session->peer_cookie_len = cfg->peer_cookie_len;
1616                         memcpy(&session->peer_cookie[0], &cfg->peer_cookie[0], cfg->peer_cookie_len);
1617                 }
1618
1619                 l2tp_session_set_header_len(session, tunnel->version);
1620
1621                 refcount_set(&session->ref_count, 1);
1622
1623                 return session;
1624         }
1625
1626         return ERR_PTR(-ENOMEM);
1627 }
1628 EXPORT_SYMBOL_GPL(l2tp_session_create);
1629
1630 /*****************************************************************************
1631  * Init and cleanup
1632  *****************************************************************************/
1633
1634 static __net_init int l2tp_init_net(struct net *net)
1635 {
1636         struct l2tp_net *pn = net_generic(net, l2tp_net_id);
1637         int hash;
1638
1639         INIT_LIST_HEAD(&pn->l2tp_tunnel_list);
1640         spin_lock_init(&pn->l2tp_tunnel_list_lock);
1641
1642         for (hash = 0; hash < L2TP_HASH_SIZE_2; hash++)
1643                 INIT_HLIST_HEAD(&pn->l2tp_session_hlist[hash]);
1644
1645         spin_lock_init(&pn->l2tp_session_hlist_lock);
1646
1647         return 0;
1648 }
1649
1650 static __net_exit void l2tp_exit_net(struct net *net)
1651 {
1652         struct l2tp_net *pn = l2tp_pernet(net);
1653         struct l2tp_tunnel *tunnel = NULL;
1654         int hash;
1655
1656         rcu_read_lock_bh();
1657         list_for_each_entry_rcu(tunnel, &pn->l2tp_tunnel_list, list) {
1658                 l2tp_tunnel_delete(tunnel);
1659         }
1660         rcu_read_unlock_bh();
1661
1662         if (l2tp_wq)
1663                 flush_workqueue(l2tp_wq);
1664         rcu_barrier();
1665
1666         for (hash = 0; hash < L2TP_HASH_SIZE_2; hash++)
1667                 WARN_ON_ONCE(!hlist_empty(&pn->l2tp_session_hlist[hash]));
1668 }
1669
1670 static struct pernet_operations l2tp_net_ops = {
1671         .init = l2tp_init_net,
1672         .exit = l2tp_exit_net,
1673         .id   = &l2tp_net_id,
1674         .size = sizeof(struct l2tp_net),
1675 };
1676
1677 static int __init l2tp_init(void)
1678 {
1679         int rc = 0;
1680
1681         rc = register_pernet_device(&l2tp_net_ops);
1682         if (rc)
1683                 goto out;
1684
1685         l2tp_wq = alloc_workqueue("l2tp", WQ_UNBOUND, 0);
1686         if (!l2tp_wq) {
1687                 pr_err("alloc_workqueue failed\n");
1688                 unregister_pernet_device(&l2tp_net_ops);
1689                 rc = -ENOMEM;
1690                 goto out;
1691         }
1692
1693         pr_info("L2TP core driver, %s\n", L2TP_DRV_VERSION);
1694
1695 out:
1696         return rc;
1697 }
1698
1699 static void __exit l2tp_exit(void)
1700 {
1701         unregister_pernet_device(&l2tp_net_ops);
1702         if (l2tp_wq) {
1703                 destroy_workqueue(l2tp_wq);
1704                 l2tp_wq = NULL;
1705         }
1706 }
1707
1708 module_init(l2tp_init);
1709 module_exit(l2tp_exit);
1710
1711 MODULE_AUTHOR("James Chapman <jchapman@katalix.com>");
1712 MODULE_DESCRIPTION("L2TP core");
1713 MODULE_LICENSE("GPL");
1714 MODULE_VERSION(L2TP_DRV_VERSION);