Pileus Git - ~andy/linux/blob - net/l2tp/l2tp_core.c

   1 /*
   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  * This program is free software; you can redistribute it and/or modify
  17  * it under the terms of the GNU General Public License version 2 as
  18  * published by the Free Software Foundation.
  19  */
  20
  21 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
  22
  23 #include <linux/module.h>
  24 #include <linux/string.h>
  25 #include <linux/list.h>
  26 #include <linux/rculist.h>
  27 #include <linux/uaccess.h>
  28
  29 #include <linux/kernel.h>
  30 #include <linux/spinlock.h>
  31 #include <linux/kthread.h>
  32 #include <linux/sched.h>
  33 #include <linux/slab.h>
  34 #include <linux/errno.h>
  35 #include <linux/jiffies.h>
  36
  37 #include <linux/netdevice.h>
  38 #include <linux/net.h>
  39 #include <linux/inetdevice.h>
  40 #include <linux/skbuff.h>
  41 #include <linux/init.h>
  42 #include <linux/in.h>
  43 #include <linux/ip.h>
  44 #include <linux/udp.h>
  45 #include <linux/l2tp.h>
  46 #include <linux/hash.h>
  47 #include <linux/sort.h>
  48 #include <linux/file.h>
  49 #include <linux/nsproxy.h>
  50 #include <net/net_namespace.h>
  51 #include <net/netns/generic.h>
  52 #include <net/dst.h>
  53 #include <net/ip.h>
  54 #include <net/udp.h>
  55 #include <net/inet_common.h>
  56 #include <net/xfrm.h>
  57 #include <net/protocol.h>
  58 #include <net/inet6_connection_sock.h>
  59 #include <net/inet_ecn.h>
  60 #include <net/ip6_route.h>
  61 #include <net/ip6_checksum.h>
  62
  63 #include <asm/byteorder.h>
  64 #include <linux/atomic.h>
  65
  66 #include "l2tp_core.h"
  67
  68 #define L2TP_DRV_VERSION        "V2.0"
  69
  70 /* L2TP header constants */
  71 #define L2TP_HDRFLAG_T     0x8000
  72 #define L2TP_HDRFLAG_L     0x4000
  73 #define L2TP_HDRFLAG_S     0x0800
  74 #define L2TP_HDRFLAG_O     0x0200
  75 #define L2TP_HDRFLAG_P     0x0100
  76
  77 #define L2TP_HDR_VER_MASK  0x000F
  78 #define L2TP_HDR_VER_2     0x0002
  79 #define L2TP_HDR_VER_3     0x0003
  80
  81 /* L2TPv3 default L2-specific sublayer */
  82 #define L2TP_SLFLAG_S      0x40000000
  83 #define L2TP_SL_SEQ_MASK   0x00ffffff
  84
  85 #define L2TP_HDR_SIZE_SEQ               10
  86 #define L2TP_HDR_SIZE_NOSEQ             6
  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 atomic_t l2tp_tunnel_count;
 103 static atomic_t l2tp_session_count;
 104 static struct workqueue_struct *l2tp_wq;
 105
 106 /* per-net private data for this module */
 107 static unsigned int l2tp_net_id;
 108 struct l2tp_net {
 109         struct list_head l2tp_tunnel_list;
 110         spinlock_t l2tp_tunnel_list_lock;
 111         struct hlist_head l2tp_session_hlist[L2TP_HASH_SIZE_2];
 112         spinlock_t l2tp_session_hlist_lock;
 113 };
 114
 115 static void l2tp_session_set_header_len(struct l2tp_session *session, int version);
 116 static void l2tp_tunnel_free(struct l2tp_tunnel *tunnel);
 117
 118 static inline struct l2tp_net *l2tp_pernet(struct net *net)
 119 {
 120         BUG_ON(!net);
 121
 122         return net_generic(net, l2tp_net_id);
 123 }
 124
 125 /* Tunnel reference counts. Incremented per session that is added to
 126  * the tunnel.
 127  */
 128 static inline void l2tp_tunnel_inc_refcount_1(struct l2tp_tunnel *tunnel)
 129 {
 130         atomic_inc(&tunnel->ref_count);
 131 }
 132
 133 static inline void l2tp_tunnel_dec_refcount_1(struct l2tp_tunnel *tunnel)
 134 {
 135         if (atomic_dec_and_test(&tunnel->ref_count))
 136                 l2tp_tunnel_free(tunnel);
 137 }
 138 #ifdef L2TP_REFCNT_DEBUG
 139 #define l2tp_tunnel_inc_refcount(_t)                                    \
 140 do {                                                                    \
 141         pr_debug("l2tp_tunnel_inc_refcount: %s:%d %s: cnt=%d\n",        \
 142                  __func__, __LINE__, (_t)->name,                        \
 143                  atomic_read(&_t->ref_count));                          \
 144         l2tp_tunnel_inc_refcount_1(_t);                                 \
 145 } while (0)
 146 #define l2tp_tunnel_dec_refcount(_t)
 147 do {                                                                    \
 148         pr_debug("l2tp_tunnel_dec_refcount: %s:%d %s: cnt=%d\n",        \
 149                  __func__, __LINE__, (_t)->name,                        \
 150                  atomic_read(&_t->ref_count));                          \
 151         l2tp_tunnel_dec_refcount_1(_t);                                 \
 152 } while (0)
 153 #else
 154 #define l2tp_tunnel_inc_refcount(t) l2tp_tunnel_inc_refcount_1(t)
 155 #define l2tp_tunnel_dec_refcount(t) l2tp_tunnel_dec_refcount_1(t)
 156 #endif
 157
 158 /* Session hash global list for L2TPv3.
 159  * The session_id SHOULD be random according to RFC3931, but several
 160  * L2TP implementations use incrementing session_ids.  So we do a real
 161  * hash on the session_id, rather than a simple bitmask.
 162  */
 163 static inline struct hlist_head *
 164 l2tp_session_id_hash_2(struct l2tp_net *pn, u32 session_id)
 165 {
 166         return &pn->l2tp_session_hlist[hash_32(session_id, L2TP_HASH_BITS_2)];
 167
 168 }
 169
 170 /* Lookup the tunnel socket, possibly involving the fs code if the socket is
 171  * owned by userspace.  A struct sock returned from this function must be
 172  * released using l2tp_tunnel_sock_put once you're done with it.
 173  */
 174 struct sock *l2tp_tunnel_sock_lookup(struct l2tp_tunnel *tunnel)
 175 {
 176         int err = 0;
 177         struct socket *sock = NULL;
 178         struct sock *sk = NULL;
 179
 180         if (!tunnel)
 181                 goto out;
 182
 183         if (tunnel->fd >= 0) {
 184                 /* Socket is owned by userspace, who might be in the process
 185                  * of closing it.  Look the socket up using the fd to ensure
 186                  * consistency.
 187                  */
 188                 sock = sockfd_lookup(tunnel->fd, &err);
 189                 if (sock)
 190                         sk = sock->sk;
 191         } else {
 192                 /* Socket is owned by kernelspace */
 193                 sk = tunnel->sock;
 194                 sock_hold(sk);
 195         }
 196
 197 out:
 198         return sk;
 199 }
 200 EXPORT_SYMBOL_GPL(l2tp_tunnel_sock_lookup);
 201
 202 /* Drop a reference to a tunnel socket obtained via. l2tp_tunnel_sock_put */
 203 void l2tp_tunnel_sock_put(struct sock *sk)
 204 {
 205         struct l2tp_tunnel *tunnel = l2tp_sock_to_tunnel(sk);
 206         if (tunnel) {
 207                 if (tunnel->fd >= 0) {
 208                         /* Socket is owned by userspace */
 209                         sockfd_put(sk->sk_socket);
 210                 }
 211                 sock_put(sk);
 212         }
 213         sock_put(sk);
 214 }
 215 EXPORT_SYMBOL_GPL(l2tp_tunnel_sock_put);
 216
 217 /* Lookup a session by id in the global session list
 218  */
 219 static struct l2tp_session *l2tp_session_find_2(struct net *net, u32 session_id)
 220 {
 221         struct l2tp_net *pn = l2tp_pernet(net);
 222         struct hlist_head *session_list =
 223                 l2tp_session_id_hash_2(pn, session_id);
 224         struct l2tp_session *session;
 225
 226         rcu_read_lock_bh();
 227         hlist_for_each_entry_rcu(session, session_list, global_hlist) {
 228                 if (session->session_id == session_id) {
 229                         rcu_read_unlock_bh();
 230                         return session;
 231                 }
 232         }
 233         rcu_read_unlock_bh();
 234
 235         return NULL;
 236 }
 237
 238 /* Session hash list.
 239  * The session_id SHOULD be random according to RFC2661, but several
 240  * L2TP implementations (Cisco and Microsoft) use incrementing
 241  * session_ids.  So we do a real hash on the session_id, rather than a
 242  * simple bitmask.
 243  */
 244 static inline struct hlist_head *
 245 l2tp_session_id_hash(struct l2tp_tunnel *tunnel, u32 session_id)
 246 {
 247         return &tunnel->session_hlist[hash_32(session_id, L2TP_HASH_BITS)];
 248 }
 249
 250 /* Lookup a session by id
 251  */
 252 struct l2tp_session *l2tp_session_find(struct net *net, struct l2tp_tunnel *tunnel, u32 session_id)
 253 {
 254         struct hlist_head *session_list;
 255         struct l2tp_session *session;
 256
 257         /* In L2TPv3, session_ids are unique over all tunnels and we
 258          * sometimes need to look them up before we know the
 259          * tunnel.
 260          */
 261         if (tunnel == NULL)
 262                 return l2tp_session_find_2(net, session_id);
 263
 264         session_list = l2tp_session_id_hash(tunnel, session_id);
 265         read_lock_bh(&tunnel->hlist_lock);
 266         hlist_for_each_entry(session, session_list, hlist) {
 267                 if (session->session_id == session_id) {
 268                         read_unlock_bh(&tunnel->hlist_lock);
 269                         return session;
 270                 }
 271         }
 272         read_unlock_bh(&tunnel->hlist_lock);
 273
 274         return NULL;
 275 }
 276 EXPORT_SYMBOL_GPL(l2tp_session_find);
 277
 278 struct l2tp_session *l2tp_session_find_nth(struct l2tp_tunnel *tunnel, int nth)
 279 {
 280         int hash;
 281         struct l2tp_session *session;
 282         int count = 0;
 283
 284         read_lock_bh(&tunnel->hlist_lock);
 285         for (hash = 0; hash < L2TP_HASH_SIZE; hash++) {
 286                 hlist_for_each_entry(session, &tunnel->session_hlist[hash], hlist) {
 287                         if (++count > nth) {
 288                                 read_unlock_bh(&tunnel->hlist_lock);
 289                                 return session;
 290                         }
 291                 }
 292         }
 293
 294         read_unlock_bh(&tunnel->hlist_lock);
 295
 296         return NULL;
 297 }
 298 EXPORT_SYMBOL_GPL(l2tp_session_find_nth);
 299
 300 /* Lookup a session by interface name.
 301  * This is very inefficient but is only used by management interfaces.
 302  */
 303 struct l2tp_session *l2tp_session_find_by_ifname(struct net *net, char *ifname)
 304 {
 305         struct l2tp_net *pn = l2tp_pernet(net);
 306         int hash;
 307         struct l2tp_session *session;
 308
 309         rcu_read_lock_bh();
 310         for (hash = 0; hash < L2TP_HASH_SIZE_2; hash++) {
 311                 hlist_for_each_entry_rcu(session, &pn->l2tp_session_hlist[hash], global_hlist) {
 312                         if (!strcmp(session->ifname, ifname)) {
 313                                 rcu_read_unlock_bh();
 314                                 return session;
 315                         }
 316                 }
 317         }
 318
 319         rcu_read_unlock_bh();
 320
 321         return NULL;
 322 }
 323 EXPORT_SYMBOL_GPL(l2tp_session_find_by_ifname);
 324
 325 /* Lookup a tunnel by id
 326  */
 327 struct l2tp_tunnel *l2tp_tunnel_find(struct net *net, u32 tunnel_id)
 328 {
 329         struct l2tp_tunnel *tunnel;
 330         struct l2tp_net *pn = l2tp_pernet(net);
 331
 332         rcu_read_lock_bh();
 333         list_for_each_entry_rcu(tunnel, &pn->l2tp_tunnel_list, list) {
 334                 if (tunnel->tunnel_id == tunnel_id) {
 335                         rcu_read_unlock_bh();
 336                         return tunnel;
 337                 }
 338         }
 339         rcu_read_unlock_bh();
 340
 341         return NULL;
 342 }
 343 EXPORT_SYMBOL_GPL(l2tp_tunnel_find);
 344
 345 struct l2tp_tunnel *l2tp_tunnel_find_nth(struct net *net, int nth)
 346 {
 347         struct l2tp_net *pn = l2tp_pernet(net);
 348         struct l2tp_tunnel *tunnel;
 349         int count = 0;
 350
 351         rcu_read_lock_bh();
 352         list_for_each_entry_rcu(tunnel, &pn->l2tp_tunnel_list, list) {
 353                 if (++count > nth) {
 354                         rcu_read_unlock_bh();
 355                         return tunnel;
 356                 }
 357         }
 358
 359         rcu_read_unlock_bh();
 360
 361         return NULL;
 362 }
 363 EXPORT_SYMBOL_GPL(l2tp_tunnel_find_nth);
 364
 365 /*****************************************************************************
 366  * Receive data handling
 367  *****************************************************************************/
 368
 369 /* Queue a skb in order. We come here only if the skb has an L2TP sequence
 370  * number.
 371  */
 372 static void l2tp_recv_queue_skb(struct l2tp_session *session, struct sk_buff *skb)
 373 {
 374         struct sk_buff *skbp;
 375         struct sk_buff *tmp;
 376         u32 ns = L2TP_SKB_CB(skb)->ns;
 377
 378         spin_lock_bh(&session->reorder_q.lock);
 379         skb_queue_walk_safe(&session->reorder_q, skbp, tmp) {
 380                 if (L2TP_SKB_CB(skbp)->ns > ns) {
 381                         __skb_queue_before(&session->reorder_q, skbp, skb);
 382                         l2tp_dbg(session, L2TP_MSG_SEQ,
 383                                  "%s: pkt %hu, inserted before %hu, reorder_q len=%d\n",
 384                                  session->name, ns, L2TP_SKB_CB(skbp)->ns,
 385                                  skb_queue_len(&session->reorder_q));
 386                         atomic_long_inc(&session->stats.rx_oos_packets);
 387                         goto out;
 388                 }
 389         }
 390
 391         __skb_queue_tail(&session->reorder_q, skb);
 392
 393 out:
 394         spin_unlock_bh(&session->reorder_q.lock);
 395 }
 396
 397 /* Dequeue a single skb.
 398  */
 399 static void l2tp_recv_dequeue_skb(struct l2tp_session *session, struct sk_buff *skb)
 400 {
 401         struct l2tp_tunnel *tunnel = session->tunnel;
 402         int length = L2TP_SKB_CB(skb)->length;
 403
 404         /* We're about to requeue the skb, so return resources
 405          * to its current owner (a socket receive buffer).
 406          */
 407         skb_orphan(skb);
 408
 409         atomic_long_inc(&tunnel->stats.rx_packets);
 410         atomic_long_add(length, &tunnel->stats.rx_bytes);
 411         atomic_long_inc(&session->stats.rx_packets);
 412         atomic_long_add(length, &session->stats.rx_bytes);
 413
 414         if (L2TP_SKB_CB(skb)->has_seq) {
 415                 /* Bump our Nr */
 416                 session->nr++;
 417                 if (tunnel->version == L2TP_HDR_VER_2)
 418                         session->nr &= 0xffff;
 419                 else
 420                         session->nr &= 0xffffff;
 421
 422                 l2tp_dbg(session, L2TP_MSG_SEQ, "%s: updated nr to %hu\n",
 423                          session->name, session->nr);
 424         }
 425
 426         /* call private receive handler */
 427         if (session->recv_skb != NULL)
 428                 (*session->recv_skb)(session, skb, L2TP_SKB_CB(skb)->length);
 429         else
 430                 kfree_skb(skb);
 431
 432         if (session->deref)
 433                 (*session->deref)(session);
 434 }
 435
 436 /* Dequeue skbs from the session's reorder_q, subject to packet order.
 437  * Skbs that have been in the queue for too long are simply discarded.
 438  */
 439 static void l2tp_recv_dequeue(struct l2tp_session *session)
 440 {
 441         struct sk_buff *skb;
 442         struct sk_buff *tmp;
 443
 444         /* If the pkt at the head of the queue has the nr that we
 445          * expect to send up next, dequeue it and any other
 446          * in-sequence packets behind it.
 447          */
 448 start:
 449         spin_lock_bh(&session->reorder_q.lock);
 450         skb_queue_walk_safe(&session->reorder_q, skb, tmp) {
 451                 if (time_after(jiffies, L2TP_SKB_CB(skb)->expires)) {
 452                         atomic_long_inc(&session->stats.rx_seq_discards);
 453                         atomic_long_inc(&session->stats.rx_errors);
 454                         l2tp_dbg(session, L2TP_MSG_SEQ,
 455                                  "%s: oos pkt %u len %d discarded (too old), waiting for %u, reorder_q_len=%d\n",
 456                                  session->name, L2TP_SKB_CB(skb)->ns,
 457                                  L2TP_SKB_CB(skb)->length, session->nr,
 458                                  skb_queue_len(&session->reorder_q));
 459                         session->reorder_skip = 1;
 460                         __skb_unlink(skb, &session->reorder_q);
 461                         kfree_skb(skb);
 462                         if (session->deref)
 463                                 (*session->deref)(session);
 464                         continue;
 465                 }
 466
 467                 if (L2TP_SKB_CB(skb)->has_seq) {
 468                         if (session->reorder_skip) {
 469                                 l2tp_dbg(session, L2TP_MSG_SEQ,
 470                                          "%s: advancing nr to next pkt: %u -> %u",
 471                                          session->name, session->nr,
 472                                          L2TP_SKB_CB(skb)->ns);
 473                                 session->reorder_skip = 0;
 474                                 session->nr = L2TP_SKB_CB(skb)->ns;
 475                         }
 476                         if (L2TP_SKB_CB(skb)->ns != session->nr) {
 477                                 l2tp_dbg(session, L2TP_MSG_SEQ,
 478                                          "%s: holding oos pkt %u len %d, waiting for %u, reorder_q_len=%d\n",
 479                                          session->name, L2TP_SKB_CB(skb)->ns,
 480                                          L2TP_SKB_CB(skb)->length, session->nr,
 481                                          skb_queue_len(&session->reorder_q));
 482                                 goto out;
 483                         }
 484                 }
 485                 __skb_unlink(skb, &session->reorder_q);
 486
 487                 /* Process the skb. We release the queue lock while we
 488                  * do so to let other contexts process the queue.
 489                  */
 490                 spin_unlock_bh(&session->reorder_q.lock);
 491                 l2tp_recv_dequeue_skb(session, skb);
 492                 goto start;
 493         }
 494
 495 out:
 496         spin_unlock_bh(&session->reorder_q.lock);
 497 }
 498
 499 static inline int l2tp_verify_udp_checksum(struct sock *sk,
 500                                            struct sk_buff *skb)
 501 {
 502         struct udphdr *uh = udp_hdr(skb);
 503         u16 ulen = ntohs(uh->len);
 504         __wsum psum;
 505
 506         if (sk->sk_no_check || skb_csum_unnecessary(skb))
 507                 return 0;
 508
 509 #if IS_ENABLED(CONFIG_IPV6)
 510         if (sk->sk_family == PF_INET6) {
 511                 if (!uh->check) {
 512                         LIMIT_NETDEBUG(KERN_INFO "L2TP: IPv6: checksum is 0\n");
 513                         return 1;
 514                 }
 515                 if ((skb->ip_summed == CHECKSUM_COMPLETE) &&
 516                     !csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
 517                                      &ipv6_hdr(skb)->daddr, ulen,
 518                                      IPPROTO_UDP, skb->csum)) {
 519                         skb->ip_summed = CHECKSUM_UNNECESSARY;
 520                         return 0;
 521                 }
 522                 skb->csum = ~csum_unfold(csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
 523                                                          &ipv6_hdr(skb)->daddr,
 524                                                          skb->len, IPPROTO_UDP,
 525                                                          0));
 526         } else
 527 #endif
 528         {
 529                 struct inet_sock *inet;
 530                 if (!uh->check)
 531                         return 0;
 532                 inet = inet_sk(sk);
 533                 psum = csum_tcpudp_nofold(inet->inet_saddr, inet->inet_daddr,
 534                                           ulen, IPPROTO_UDP, 0);
 535
 536                 if ((skb->ip_summed == CHECKSUM_COMPLETE) &&
 537                     !csum_fold(csum_add(psum, skb->csum)))
 538                         return 0;
 539                 skb->csum = psum;
 540         }
 541
 542         return __skb_checksum_complete(skb);
 543 }
 544
 545 /* Do receive processing of L2TP data frames. We handle both L2TPv2
 546  * and L2TPv3 data frames here.
 547  *
 548  * L2TPv2 Data Message Header
 549  *
 550  *  0                   1                   2                   3
 551  *  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
 552  * +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
 553  * |T|L|x|x|S|x|O|P|x|x|x|x|  Ver  |          Length (opt)         |
 554  * +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
 555  * |           Tunnel ID           |           Session ID          |
 556  * +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
 557  * |             Ns (opt)          |             Nr (opt)          |
 558  * +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
 559  * |      Offset Size (opt)        |    Offset pad... (opt)
 560  * +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
 561  *
 562  * Data frames are marked by T=0. All other fields are the same as
 563  * those in L2TP control frames.
 564  *
 565  * L2TPv3 Data Message Header
 566  *
 567  * +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
 568  * |                      L2TP Session Header                      |
 569  * +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
 570  * |                      L2-Specific Sublayer                     |
 571  * +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
 572  * |                        Tunnel Payload                      ...
 573  * +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
 574  *
 575  * L2TPv3 Session Header Over IP
 576  *
 577  *  0                   1                   2                   3
 578  *  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
 579  * +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
 580  * |                           Session ID                          |
 581  * +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
 582  * |               Cookie (optional, maximum 64 bits)...
 583  * +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
 584  *                                                                 |
 585  * +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
 586  *
 587  * L2TPv3 L2-Specific Sublayer Format
 588  *
 589  *  0                   1                   2                   3
 590  *  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
 591  * +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
 592  * |x|S|x|x|x|x|x|x|              Sequence Number                  |
 593  * +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
 594  *
 595  * Cookie value, sublayer format and offset (pad) are negotiated with
 596  * the peer when the session is set up. Unlike L2TPv2, we do not need
 597  * to parse the packet header to determine if optional fields are
 598  * present.
 599  *
 600  * Caller must already have parsed the frame and determined that it is
 601  * a data (not control) frame before coming here. Fields up to the
 602  * session-id have already been parsed and ptr points to the data
 603  * after the session-id.
 604  */
 605 void l2tp_recv_common(struct l2tp_session *session, struct sk_buff *skb,
 606                       unsigned char *ptr, unsigned char *optr, u16 hdrflags,
 607                       int length, int (*payload_hook)(struct sk_buff *skb))
 608 {
 609         struct l2tp_tunnel *tunnel = session->tunnel;
 610         int offset;
 611         u32 ns, nr;
 612
 613         /* The ref count is increased since we now hold a pointer to
 614          * the session. Take care to decrement the refcnt when exiting
 615          * this function from now on...
 616          */
 617         l2tp_session_inc_refcount(session);
 618         if (session->ref)
 619                 (*session->ref)(session);
 620
 621         /* Parse and check optional cookie */
 622         if (session->peer_cookie_len > 0) {
 623                 if (memcmp(ptr, &session->peer_cookie[0], session->peer_cookie_len)) {
 624                         l2tp_info(tunnel, L2TP_MSG_DATA,
 625                                   "%s: cookie mismatch (%u/%u). Discarding.\n",
 626                                   tunnel->name, tunnel->tunnel_id,
 627                                   session->session_id);
 628                         atomic_long_inc(&session->stats.rx_cookie_discards);
 629                         goto discard;
 630                 }
 631                 ptr += session->peer_cookie_len;
 632         }
 633
 634         /* Handle the optional sequence numbers. Sequence numbers are
 635          * in different places for L2TPv2 and L2TPv3.
 636          *
 637          * If we are the LAC, enable/disable sequence numbers under
 638          * the control of the LNS.  If no sequence numbers present but
 639          * we were expecting them, discard frame.
 640          */
 641         ns = nr = 0;
 642         L2TP_SKB_CB(skb)->has_seq = 0;
 643         if (tunnel->version == L2TP_HDR_VER_2) {
 644                 if (hdrflags & L2TP_HDRFLAG_S) {
 645                         ns = ntohs(*(__be16 *) ptr);
 646                         ptr += 2;
 647                         nr = ntohs(*(__be16 *) ptr);
 648                         ptr += 2;
 649
 650                         /* Store L2TP info in the skb */
 651                         L2TP_SKB_CB(skb)->ns = ns;
 652                         L2TP_SKB_CB(skb)->has_seq = 1;
 653
 654                         l2tp_dbg(session, L2TP_MSG_SEQ,
 655                                  "%s: recv data ns=%u, nr=%u, session nr=%u\n",
 656                                  session->name, ns, nr, session->nr);
 657                 }
 658         } else if (session->l2specific_type == L2TP_L2SPECTYPE_DEFAULT) {
 659                 u32 l2h = ntohl(*(__be32 *) ptr);
 660
 661                 if (l2h & 0x40000000) {
 662                         ns = l2h & 0x00ffffff;
 663
 664                         /* Store L2TP info in the skb */
 665                         L2TP_SKB_CB(skb)->ns = ns;
 666                         L2TP_SKB_CB(skb)->has_seq = 1;
 667
 668                         l2tp_dbg(session, L2TP_MSG_SEQ,
 669                                  "%s: recv data ns=%u, session nr=%u\n",
 670                                  session->name, ns, session->nr);
 671                 }
 672         }
 673
 674         /* Advance past L2-specific header, if present */
 675         ptr += session->l2specific_len;
 676
 677         if (L2TP_SKB_CB(skb)->has_seq) {
 678                 /* Received a packet with sequence numbers. If we're the LNS,
 679                  * check if we sre sending sequence numbers and if not,
 680                  * configure it so.
 681                  */
 682                 if ((!session->lns_mode) && (!session->send_seq)) {
 683                         l2tp_info(session, L2TP_MSG_SEQ,
 684                                   "%s: requested to enable seq numbers by LNS\n",
 685                                   session->name);
 686                         session->send_seq = -1;
 687                         l2tp_session_set_header_len(session, tunnel->version);
 688                 }
 689         } else {
 690                 /* No sequence numbers.
 691                  * If user has configured mandatory sequence numbers, discard.
 692                  */
 693                 if (session->recv_seq) {
 694                         l2tp_warn(session, L2TP_MSG_SEQ,
 695                                   "%s: recv data has no seq numbers when required. Discarding.\n",
 696                                   session->name);
 697                         atomic_long_inc(&session->stats.rx_seq_discards);
 698                         goto discard;
 699                 }
 700
 701                 /* If we're the LAC and we're sending sequence numbers, the
 702                  * LNS has requested that we no longer send sequence numbers.
 703                  * If we're the LNS and we're sending sequence numbers, the
 704                  * LAC is broken. Discard the frame.
 705                  */
 706                 if ((!session->lns_mode) && (session->send_seq)) {
 707                         l2tp_info(session, L2TP_MSG_SEQ,
 708                                   "%s: requested to disable seq numbers by LNS\n",
 709                                   session->name);
 710                         session->send_seq = 0;
 711                         l2tp_session_set_header_len(session, tunnel->version);
 712                 } else if (session->send_seq) {
 713                         l2tp_warn(session, L2TP_MSG_SEQ,
 714                                   "%s: recv data has no seq numbers when required. Discarding.\n",
 715                                   session->name);
 716                         atomic_long_inc(&session->stats.rx_seq_discards);
 717                         goto discard;
 718                 }
 719         }
 720
 721         /* Session data offset is handled differently for L2TPv2 and
 722          * L2TPv3. For L2TPv2, there is an optional 16-bit value in
 723          * the header. For L2TPv3, the offset is negotiated using AVPs
 724          * in the session setup control protocol.
 725          */
 726         if (tunnel->version == L2TP_HDR_VER_2) {
 727                 /* If offset bit set, skip it. */
 728                 if (hdrflags & L2TP_HDRFLAG_O) {
 729                         offset = ntohs(*(__be16 *)ptr);
 730                         ptr += 2 + offset;
 731                 }
 732         } else
 733                 ptr += session->offset;
 734
 735         offset = ptr - optr;
 736         if (!pskb_may_pull(skb, offset))
 737                 goto discard;
 738
 739         __skb_pull(skb, offset);
 740
 741         /* If caller wants to process the payload before we queue the
 742          * packet, do so now.
 743          */
 744         if (payload_hook)
 745                 if ((*payload_hook)(skb))
 746                         goto discard;
 747
 748         /* Prepare skb for adding to the session's reorder_q.  Hold
 749          * packets for max reorder_timeout or 1 second if not
 750          * reordering.
 751          */
 752         L2TP_SKB_CB(skb)->length = length;
 753         L2TP_SKB_CB(skb)->expires = jiffies +
 754                 (session->reorder_timeout ? session->reorder_timeout : HZ);
 755
 756         /* Add packet to the session's receive queue. Reordering is done here, if
 757          * enabled. Saved L2TP protocol info is stored in skb->sb[].
 758          */
 759         if (L2TP_SKB_CB(skb)->has_seq) {
 760                 if (session->reorder_timeout != 0) {
 761                         /* Packet reordering enabled. Add skb to session's
 762                          * reorder queue, in order of ns.
 763                          */
 764                         l2tp_recv_queue_skb(session, skb);
 765                 } else {
 766                         /* Packet reordering disabled. Discard out-of-sequence
 767                          * packets
 768                          */
 769                         if (L2TP_SKB_CB(skb)->ns != session->nr) {
 770                                 atomic_long_inc(&session->stats.rx_seq_discards);
 771                                 l2tp_dbg(session, L2TP_MSG_SEQ,
 772                                          "%s: oos pkt %u len %d discarded, waiting for %u, reorder_q_len=%d\n",
 773                                          session->name, L2TP_SKB_CB(skb)->ns,
 774                                          L2TP_SKB_CB(skb)->length, session->nr,
 775                                          skb_queue_len(&session->reorder_q));
 776                                 goto discard;
 777                         }
 778                         skb_queue_tail(&session->reorder_q, skb);
 779                 }
 780         } else {
 781                 /* No sequence numbers. Add the skb to the tail of the
 782                  * reorder queue. This ensures that it will be
 783                  * delivered after all previous sequenced skbs.
 784                  */
 785                 skb_queue_tail(&session->reorder_q, skb);
 786         }
 787
 788         /* Try to dequeue as many skbs from reorder_q as we can. */
 789         l2tp_recv_dequeue(session);
 790
 791         l2tp_session_dec_refcount(session);
 792
 793         return;
 794
 795 discard:
 796         atomic_long_inc(&session->stats.rx_errors);
 797         kfree_skb(skb);
 798
 799         if (session->deref)
 800                 (*session->deref)(session);
 801
 802         l2tp_session_dec_refcount(session);
 803 }
 804 EXPORT_SYMBOL(l2tp_recv_common);
 805
 806 /* Drop skbs from the session's reorder_q
 807  */
 808 int l2tp_session_queue_purge(struct l2tp_session *session)
 809 {
 810         struct sk_buff *skb = NULL;
 811         BUG_ON(!session);
 812         BUG_ON(session->magic != L2TP_SESSION_MAGIC);
 813         while ((skb = skb_dequeue(&session->reorder_q))) {
 814                 atomic_long_inc(&session->stats.rx_errors);
 815                 kfree_skb(skb);
 816                 if (session->deref)
 817                         (*session->deref)(session);
 818         }
 819         return 0;
 820 }
 821 EXPORT_SYMBOL_GPL(l2tp_session_queue_purge);
 822
 823 /* Internal UDP receive frame. Do the real work of receiving an L2TP data frame
 824  * here. The skb is not on a list when we get here.
 825  * Returns 0 if the packet was a data packet and was successfully passed on.
 826  * Returns 1 if the packet was not a good data packet and could not be
 827  * forwarded.  All such packets are passed up to userspace to deal with.
 828  */
 829 static int l2tp_udp_recv_core(struct l2tp_tunnel *tunnel, struct sk_buff *skb,
 830                               int (*payload_hook)(struct sk_buff *skb))
 831 {
 832         struct l2tp_session *session = NULL;
 833         unsigned char *ptr, *optr;
 834         u16 hdrflags;
 835         u32 tunnel_id, session_id;
 836         u16 version;
 837         int length;
 838
 839         if (tunnel->sock && l2tp_verify_udp_checksum(tunnel->sock, skb))
 840                 goto discard_bad_csum;
 841
 842         /* UDP always verifies the packet length. */
 843         __skb_pull(skb, sizeof(struct udphdr));
 844
 845         /* Short packet? */
 846         if (!pskb_may_pull(skb, L2TP_HDR_SIZE_SEQ)) {
 847                 l2tp_info(tunnel, L2TP_MSG_DATA,
 848                           "%s: recv short packet (len=%d)\n",
 849                           tunnel->name, skb->len);
 850                 goto error;
 851         }
 852
 853         /* Trace packet contents, if enabled */
 854         if (tunnel->debug & L2TP_MSG_DATA) {
 855                 length = min(32u, skb->len);
 856                 if (!pskb_may_pull(skb, length))
 857                         goto error;
 858
 859                 pr_debug("%s: recv\n", tunnel->name);
 860                 print_hex_dump_bytes("", DUMP_PREFIX_OFFSET, skb->data, length);
 861         }
 862
 863         /* Point to L2TP header */
 864         optr = ptr = skb->data;
 865
 866         /* Get L2TP header flags */
 867         hdrflags = ntohs(*(__be16 *) ptr);
 868
 869         /* Check protocol version */
 870         version = hdrflags & L2TP_HDR_VER_MASK;
 871         if (version != tunnel->version) {
 872                 l2tp_info(tunnel, L2TP_MSG_DATA,
 873                           "%s: recv protocol version mismatch: got %d expected %d\n",
 874                           tunnel->name, version, tunnel->version);
 875                 goto error;
 876         }
 877
 878         /* Get length of L2TP packet */
 879         length = skb->len;
 880
 881         /* If type is control packet, it is handled by userspace. */
 882         if (hdrflags & L2TP_HDRFLAG_T) {
 883                 l2tp_dbg(tunnel, L2TP_MSG_DATA,
 884                          "%s: recv control packet, len=%d\n",
 885                          tunnel->name, length);
 886                 goto error;
 887         }
 888
 889         /* Skip flags */
 890         ptr += 2;
 891
 892         if (tunnel->version == L2TP_HDR_VER_2) {
 893                 /* If length is present, skip it */
 894                 if (hdrflags & L2TP_HDRFLAG_L)
 895                         ptr += 2;
 896
 897                 /* Extract tunnel and session ID */
 898                 tunnel_id = ntohs(*(__be16 *) ptr);
 899                 ptr += 2;
 900                 session_id = ntohs(*(__be16 *) ptr);
 901                 ptr += 2;
 902         } else {
 903                 ptr += 2;       /* skip reserved bits */
 904                 tunnel_id = tunnel->tunnel_id;
 905                 session_id = ntohl(*(__be32 *) ptr);
 906                 ptr += 4;
 907         }
 908
 909         /* Find the session context */
 910         session = l2tp_session_find(tunnel->l2tp_net, tunnel, session_id);
 911         if (!session || !session->recv_skb) {
 912                 /* Not found? Pass to userspace to deal with */
 913                 l2tp_info(tunnel, L2TP_MSG_DATA,
 914                           "%s: no session found (%u/%u). Passing up.\n",
 915                           tunnel->name, tunnel_id, session_id);
 916                 goto error;
 917         }
 918
 919         l2tp_recv_common(session, skb, ptr, optr, hdrflags, length, payload_hook);
 920
 921         return 0;
 922
 923 discard_bad_csum:
 924         LIMIT_NETDEBUG("%s: UDP: bad checksum\n", tunnel->name);
 925         UDP_INC_STATS_USER(tunnel->l2tp_net, UDP_MIB_INERRORS, 0);
 926         atomic_long_inc(&tunnel->stats.rx_errors);
 927         kfree_skb(skb);
 928
 929         return 0;
 930
 931 error:
 932         /* Put UDP header back */
 933         __skb_push(skb, sizeof(struct udphdr));
 934
 935         return 1;
 936 }
 937
 938 /* UDP encapsulation receive handler. See net/ipv4/udp.c.
 939  * Return codes:
 940  * 0 : success.
 941  * <0: error
 942  * >0: skb should be passed up to userspace as UDP.
 943  */
 944 int l2tp_udp_encap_recv(struct sock *sk, struct sk_buff *skb)
 945 {
 946         struct l2tp_tunnel *tunnel;
 947
 948         tunnel = l2tp_sock_to_tunnel(sk);
 949         if (tunnel == NULL)
 950                 goto pass_up;
 951
 952         l2tp_dbg(tunnel, L2TP_MSG_DATA, "%s: received %d bytes\n",
 953                  tunnel->name, skb->len);
 954
 955         if (l2tp_udp_recv_core(tunnel, skb, tunnel->recv_payload_hook))
 956                 goto pass_up_put;
 957
 958         sock_put(sk);
 959         return 0;
 960
 961 pass_up_put:
 962         sock_put(sk);
 963 pass_up:
 964         return 1;
 965 }
 966 EXPORT_SYMBOL_GPL(l2tp_udp_encap_recv);
 967
 968 /************************************************************************
 969  * Transmit handling
 970  ***********************************************************************/
 971
 972 /* Build an L2TP header for the session into the buffer provided.
 973  */
 974 static int l2tp_build_l2tpv2_header(struct l2tp_session *session, void *buf)
 975 {
 976         struct l2tp_tunnel *tunnel = session->tunnel;
 977         __be16 *bufp = buf;
 978         __be16 *optr = buf;
 979         u16 flags = L2TP_HDR_VER_2;
 980         u32 tunnel_id = tunnel->peer_tunnel_id;
 981         u32 session_id = session->peer_session_id;
 982
 983         if (session->send_seq)
 984                 flags |= L2TP_HDRFLAG_S;
 985
 986         /* Setup L2TP header. */
 987         *bufp++ = htons(flags);
 988         *bufp++ = htons(tunnel_id);
 989         *bufp++ = htons(session_id);
 990         if (session->send_seq) {
 991                 *bufp++ = htons(session->ns);
 992                 *bufp++ = 0;
 993                 session->ns++;
 994                 session->ns &= 0xffff;
 995                 l2tp_dbg(session, L2TP_MSG_SEQ, "%s: updated ns to %u\n",
 996                          session->name, session->ns);
 997         }
 998
 999         return bufp - optr;
1000 }
1001
1002 static int l2tp_build_l2tpv3_header(struct l2tp_session *session, void *buf)
1003 {
1004         struct l2tp_tunnel *tunnel = session->tunnel;
1005         char *bufp = buf;
1006         char *optr = bufp;
1007
1008         /* Setup L2TP header. The header differs slightly for UDP and
1009          * IP encapsulations. For UDP, there is 4 bytes of flags.
1010          */
1011         if (tunnel->encap == L2TP_ENCAPTYPE_UDP) {
1012                 u16 flags = L2TP_HDR_VER_3;
1013                 *((__be16 *) bufp) = htons(flags);
1014                 bufp += 2;
1015                 *((__be16 *) bufp) = 0;
1016                 bufp += 2;
1017         }
1018
1019         *((__be32 *) bufp) = htonl(session->peer_session_id);
1020         bufp += 4;
1021         if (session->cookie_len) {
1022                 memcpy(bufp, &session->cookie[0], session->cookie_len);
1023                 bufp += session->cookie_len;
1024         }
1025         if (session->l2specific_len) {
1026                 if (session->l2specific_type == L2TP_L2SPECTYPE_DEFAULT) {
1027                         u32 l2h = 0;
1028                         if (session->send_seq) {
1029                                 l2h = 0x40000000 | session->ns;
1030                                 session->ns++;
1031                                 session->ns &= 0xffffff;
1032                                 l2tp_dbg(session, L2TP_MSG_SEQ,
1033                                          "%s: updated ns to %u\n",
1034                                          session->name, session->ns);
1035                         }
1036
1037                         *((__be32 *) bufp) = htonl(l2h);
1038                 }
1039                 bufp += session->l2specific_len;
1040         }
1041         if (session->offset)
1042                 bufp += session->offset;
1043
1044         return bufp - optr;
1045 }
1046
1047 static int l2tp_xmit_core(struct l2tp_session *session, struct sk_buff *skb,
1048                           struct flowi *fl, size_t data_len)
1049 {
1050         struct l2tp_tunnel *tunnel = session->tunnel;
1051         unsigned int len = skb->len;
1052         int error;
1053
1054         /* Debug */
1055         if (session->send_seq)
1056                 l2tp_dbg(session, L2TP_MSG_DATA, "%s: send %Zd bytes, ns=%u\n",
1057                          session->name, data_len, session->ns - 1);
1058         else
1059                 l2tp_dbg(session, L2TP_MSG_DATA, "%s: send %Zd bytes\n",
1060                          session->name, data_len);
1061
1062         if (session->debug & L2TP_MSG_DATA) {
1063                 int uhlen = (tunnel->encap == L2TP_ENCAPTYPE_UDP) ? sizeof(struct udphdr) : 0;
1064                 unsigned char *datap = skb->data + uhlen;
1065
1066                 pr_debug("%s: xmit\n", session->name);
1067                 print_hex_dump_bytes("", DUMP_PREFIX_OFFSET,
1068                                      datap, min_t(size_t, 32, len - uhlen));
1069         }
1070
1071         /* Queue the packet to IP for output */
1072         skb->local_df = 1;
1073 #if IS_ENABLED(CONFIG_IPV6)
1074         if (skb->sk->sk_family == PF_INET6)
1075                 error = inet6_csk_xmit(skb, NULL);
1076         else
1077 #endif
1078                 error = ip_queue_xmit(skb, fl);
1079
1080         /* Update stats */
1081         if (error >= 0) {
1082                 atomic_long_inc(&tunnel->stats.tx_packets);
1083                 atomic_long_add(len, &tunnel->stats.tx_bytes);
1084                 atomic_long_inc(&session->stats.tx_packets);
1085                 atomic_long_add(len, &session->stats.tx_bytes);
1086         } else {
1087                 atomic_long_inc(&tunnel->stats.tx_errors);
1088                 atomic_long_inc(&session->stats.tx_errors);
1089         }
1090
1091         return 0;
1092 }
1093
1094 /* Automatically called when the skb is freed.
1095  */
1096 static void l2tp_sock_wfree(struct sk_buff *skb)
1097 {
1098         sock_put(skb->sk);
1099 }
1100
1101 /* For data skbs that we transmit, we associate with the tunnel socket
1102  * but don't do accounting.
1103  */
1104 static inline void l2tp_skb_set_owner_w(struct sk_buff *skb, struct sock *sk)
1105 {
1106         sock_hold(sk);
1107         skb->sk = sk;
1108         skb->destructor = l2tp_sock_wfree;
1109 }
1110
1111 #if IS_ENABLED(CONFIG_IPV6)
1112 static void l2tp_xmit_ipv6_csum(struct sock *sk, struct sk_buff *skb,
1113                                 int udp_len)
1114 {
1115         struct ipv6_pinfo *np = inet6_sk(sk);
1116         struct udphdr *uh = udp_hdr(skb);
1117
1118         if (!skb_dst(skb) || !skb_dst(skb)->dev ||
1119             !(skb_dst(skb)->dev->features & NETIF_F_IPV6_CSUM)) {
1120                 __wsum csum = skb_checksum(skb, 0, udp_len, 0);
1121                 skb->ip_summed = CHECKSUM_UNNECESSARY;
1122                 uh->check = csum_ipv6_magic(&np->saddr, &np->daddr, udp_len,
1123                                             IPPROTO_UDP, csum);
1124                 if (uh->check == 0)
1125                         uh->check = CSUM_MANGLED_0;
1126         } else {
1127                 skb->ip_summed = CHECKSUM_PARTIAL;
1128                 skb->csum_start = skb_transport_header(skb) - skb->head;
1129                 skb->csum_offset = offsetof(struct udphdr, check);
1130                 uh->check = ~csum_ipv6_magic(&np->saddr, &np->daddr,
1131                                              udp_len, IPPROTO_UDP, 0);
1132         }
1133 }
1134 #endif
1135
1136 /* If caller requires the skb to have a ppp header, the header must be
1137  * inserted in the skb data before calling this function.
1138  */
1139 int l2tp_xmit_skb(struct l2tp_session *session, struct sk_buff *skb, int hdr_len)
1140 {
1141         int data_len = skb->len;
1142         struct l2tp_tunnel *tunnel = session->tunnel;
1143         struct sock *sk = tunnel->sock;
1144         struct flowi *fl;
1145         struct udphdr *uh;
1146         struct inet_sock *inet;
1147         __wsum csum;
1148         int headroom;
1149         int uhlen = (tunnel->encap == L2TP_ENCAPTYPE_UDP) ? sizeof(struct udphdr) : 0;
1150         int udp_len;
1151         int ret = NET_XMIT_SUCCESS;
1152
1153         /* Check that there's enough headroom in the skb to insert IP,
1154          * UDP and L2TP headers. If not enough, expand it to
1155          * make room. Adjust truesize.
1156          */
1157         headroom = NET_SKB_PAD + sizeof(struct iphdr) +
1158                 uhlen + hdr_len;
1159         if (skb_cow_head(skb, headroom)) {
1160                 kfree_skb(skb);
1161                 return NET_XMIT_DROP;
1162         }
1163
1164         skb_orphan(skb);
1165         /* Setup L2TP header */
1166         session->build_header(session, __skb_push(skb, hdr_len));
1167
1168         /* Reset skb netfilter state */
1169         memset(&(IPCB(skb)->opt), 0, sizeof(IPCB(skb)->opt));
1170         IPCB(skb)->flags &= ~(IPSKB_XFRM_TUNNEL_SIZE | IPSKB_XFRM_TRANSFORMED |
1171                               IPSKB_REROUTED);
1172         nf_reset(skb);
1173
1174         bh_lock_sock(sk);
1175         if (sock_owned_by_user(sk)) {
1176                 kfree_skb(skb);
1177                 ret = NET_XMIT_DROP;
1178                 goto out_unlock;
1179         }
1180
1181         /* Get routing info from the tunnel socket */
1182         skb_dst_drop(skb);
1183         skb_dst_set(skb, dst_clone(__sk_dst_check(sk, 0)));
1184
1185         inet = inet_sk(sk);
1186         fl = &inet->cork.fl;
1187         switch (tunnel->encap) {
1188         case L2TP_ENCAPTYPE_UDP:
1189                 /* Setup UDP header */
1190                 __skb_push(skb, sizeof(*uh));
1191                 skb_reset_transport_header(skb);
1192                 uh = udp_hdr(skb);
1193                 uh->source = inet->inet_sport;
1194                 uh->dest = inet->inet_dport;
1195                 udp_len = uhlen + hdr_len + data_len;
1196                 uh->len = htons(udp_len);
1197                 uh->check = 0;
1198
1199                 /* Calculate UDP checksum if configured to do so */
1200 #if IS_ENABLED(CONFIG_IPV6)
1201                 if (sk->sk_family == PF_INET6)
1202                         l2tp_xmit_ipv6_csum(sk, skb, udp_len);
1203                 else
1204 #endif
1205                 if (sk->sk_no_check == UDP_CSUM_NOXMIT)
1206                         skb->ip_summed = CHECKSUM_NONE;
1207                 else if ((skb_dst(skb) && skb_dst(skb)->dev) &&
1208                          (!(skb_dst(skb)->dev->features & NETIF_F_V4_CSUM))) {
1209                         skb->ip_summed = CHECKSUM_COMPLETE;
1210                         csum = skb_checksum(skb, 0, udp_len, 0);
1211                         uh->check = csum_tcpudp_magic(inet->inet_saddr,
1212                                                       inet->inet_daddr,
1213                                                       udp_len, IPPROTO_UDP, csum);
1214                         if (uh->check == 0)
1215                                 uh->check = CSUM_MANGLED_0;
1216                 } else {
1217                         skb->ip_summed = CHECKSUM_PARTIAL;
1218                         skb->csum_start = skb_transport_header(skb) - skb->head;
1219                         skb->csum_offset = offsetof(struct udphdr, check);
1220                         uh->check = ~csum_tcpudp_magic(inet->inet_saddr,
1221                                                        inet->inet_daddr,
1222                                                        udp_len, IPPROTO_UDP, 0);
1223                 }
1224                 break;
1225
1226         case L2TP_ENCAPTYPE_IP:
1227                 break;
1228         }
1229
1230         l2tp_skb_set_owner_w(skb, sk);
1231
1232         l2tp_xmit_core(session, skb, fl, data_len);
1233 out_unlock:
1234         bh_unlock_sock(sk);
1235
1236         return ret;
1237 }
1238 EXPORT_SYMBOL_GPL(l2tp_xmit_skb);
1239
1240 /*****************************************************************************
1241  * Tinnel and session create/destroy.
1242  *****************************************************************************/
1243
1244 /* Tunnel socket destruct hook.
1245  * The tunnel context is deleted only when all session sockets have been
1246  * closed.
1247  */
1248 static void l2tp_tunnel_destruct(struct sock *sk)
1249 {
1250         struct l2tp_tunnel *tunnel;
1251         struct l2tp_net *pn;
1252
1253         tunnel = sk->sk_user_data;
1254         if (tunnel == NULL)
1255                 goto end;
1256
1257         l2tp_info(tunnel, L2TP_MSG_CONTROL, "%s: closing...\n", tunnel->name);
1258
1259
1260         /* Disable udp encapsulation */
1261         switch (tunnel->encap) {
1262         case L2TP_ENCAPTYPE_UDP:
1263                 /* No longer an encapsulation socket. See net/ipv4/udp.c */
1264                 (udp_sk(sk))->encap_type = 0;
1265                 (udp_sk(sk))->encap_rcv = NULL;
1266                 (udp_sk(sk))->encap_destroy = NULL;
1267                 break;
1268         case L2TP_ENCAPTYPE_IP:
1269                 break;
1270         }
1271
1272         /* Remove hooks into tunnel socket */
1273         sk->sk_destruct = tunnel->old_sk_destruct;
1274         sk->sk_user_data = NULL;
1275         tunnel->sock = NULL;
1276
1277         /* Remove the tunnel struct from the tunnel list */
1278         pn = l2tp_pernet(tunnel->l2tp_net);
1279         spin_lock_bh(&pn->l2tp_tunnel_list_lock);
1280         list_del_rcu(&tunnel->list);
1281         spin_unlock_bh(&pn->l2tp_tunnel_list_lock);
1282         atomic_dec(&l2tp_tunnel_count);
1283
1284         l2tp_tunnel_closeall(tunnel);
1285         l2tp_tunnel_dec_refcount(tunnel);
1286
1287         /* Call the original destructor */
1288         if (sk->sk_destruct)
1289                 (*sk->sk_destruct)(sk);
1290 end:
1291         return;
1292 }
1293
1294 /* When the tunnel is closed, all the attached sessions need to go too.
1295  */
1296 void l2tp_tunnel_closeall(struct l2tp_tunnel *tunnel)
1297 {
1298         int hash;
1299         struct hlist_node *walk;
1300         struct hlist_node *tmp;
1301         struct l2tp_session *session;
1302
1303         BUG_ON(tunnel == NULL);
1304
1305         l2tp_info(tunnel, L2TP_MSG_CONTROL, "%s: closing all sessions...\n",
1306                   tunnel->name);
1307
1308         write_lock_bh(&tunnel->hlist_lock);
1309         for (hash = 0; hash < L2TP_HASH_SIZE; hash++) {
1310 again:
1311                 hlist_for_each_safe(walk, tmp, &tunnel->session_hlist[hash]) {
1312                         session = hlist_entry(walk, struct l2tp_session, hlist);
1313
1314                         l2tp_info(session, L2TP_MSG_CONTROL,
1315                                   "%s: closing session\n", session->name);
1316
1317                         hlist_del_init(&session->hlist);
1318
1319                         if (session->ref != NULL)
1320                                 (*session->ref)(session);
1321
1322                         write_unlock_bh(&tunnel->hlist_lock);
1323
1324                         __l2tp_session_unhash(session);
1325                         l2tp_session_queue_purge(session);
1326
1327                         if (session->session_close != NULL)
1328                                 (*session->session_close)(session);
1329
1330                         if (session->deref != NULL)
1331                                 (*session->deref)(session);
1332
1333                         l2tp_session_dec_refcount(session);
1334
1335                         write_lock_bh(&tunnel->hlist_lock);
1336
1337                         /* Now restart from the beginning of this hash
1338                          * chain.  We always remove a session from the
1339                          * list so we are guaranteed to make forward
1340                          * progress.
1341                          */
1342                         goto again;
1343                 }
1344         }
1345         write_unlock_bh(&tunnel->hlist_lock);
1346 }
1347 EXPORT_SYMBOL_GPL(l2tp_tunnel_closeall);
1348
1349 /* Tunnel socket destroy hook for UDP encapsulation */
1350 static void l2tp_udp_encap_destroy(struct sock *sk)
1351 {
1352         struct l2tp_tunnel *tunnel = l2tp_sock_to_tunnel(sk);
1353         if (tunnel) {
1354                 l2tp_tunnel_closeall(tunnel);
1355                 sock_put(sk);
1356         }
1357 }
1358
1359 /* Really kill the tunnel.
1360  * Come here only when all sessions have been cleared from the tunnel.
1361  */
1362 static void l2tp_tunnel_free(struct l2tp_tunnel *tunnel)
1363 {
1364         BUG_ON(atomic_read(&tunnel->ref_count) != 0);
1365         BUG_ON(tunnel->sock != NULL);
1366         l2tp_info(tunnel, L2TP_MSG_CONTROL, "%s: free...\n", tunnel->name);
1367         kfree_rcu(tunnel, rcu);
1368 }
1369
1370 /* Workqueue tunnel deletion function */
1371 static void l2tp_tunnel_del_work(struct work_struct *work)
1372 {
1373         struct l2tp_tunnel *tunnel = NULL;
1374         struct socket *sock = NULL;
1375         struct sock *sk = NULL;
1376
1377         tunnel = container_of(work, struct l2tp_tunnel, del_work);
1378         sk = l2tp_tunnel_sock_lookup(tunnel);
1379         if (!sk)
1380                 return;
1381
1382         sock = sk->sk_socket;
1383
1384         /* If the tunnel socket was created by userspace, then go through the
1385          * inet layer to shut the socket down, and let userspace close it.
1386          * Otherwise, if we created the socket directly within the kernel, use
1387          * the sk API to release it here.
1388          * In either case the tunnel resources are freed in the socket
1389          * destructor when the tunnel socket goes away.
1390          */
1391         if (tunnel->fd >= 0) {
1392                 if (sock)
1393                         inet_shutdown(sock, 2);
1394         } else {
1395                 if (sock)
1396                         kernel_sock_shutdown(sock, SHUT_RDWR);
1397                 sk_release_kernel(sk);
1398         }
1399
1400         l2tp_tunnel_sock_put(sk);
1401 }
1402
1403 /* Create a socket for the tunnel, if one isn't set up by
1404  * userspace. This is used for static tunnels where there is no
1405  * managing L2TP daemon.
1406  *
1407  * Since we don't want these sockets to keep a namespace alive by
1408  * themselves, we drop the socket's namespace refcount after creation.
1409  * These sockets are freed when the namespace exits using the pernet
1410  * exit hook.
1411  */
1412 static int l2tp_tunnel_sock_create(struct net *net,
1413                                 u32 tunnel_id,
1414                                 u32 peer_tunnel_id,
1415                                 struct l2tp_tunnel_cfg *cfg,
1416                                 struct socket **sockp)
1417 {
1418         int err = -EINVAL;
1419         struct socket *sock = NULL;
1420         struct sockaddr_in udp_addr = {0};
1421         struct sockaddr_l2tpip ip_addr = {0};
1422 #if IS_ENABLED(CONFIG_IPV6)
1423         struct sockaddr_in6 udp6_addr = {0};
1424         struct sockaddr_l2tpip6 ip6_addr = {0};
1425 #endif
1426
1427         switch (cfg->encap) {
1428         case L2TP_ENCAPTYPE_UDP:
1429 #if IS_ENABLED(CONFIG_IPV6)
1430                 if (cfg->local_ip6 && cfg->peer_ip6) {
1431                         err = sock_create_kern(AF_INET6, SOCK_DGRAM, 0, &sock);
1432                         if (err < 0)
1433                                 goto out;
1434
1435                         sk_change_net(sock->sk, net);
1436
1437                         udp6_addr.sin6_family = AF_INET6;
1438                         memcpy(&udp6_addr.sin6_addr, cfg->local_ip6,
1439                                sizeof(udp6_addr.sin6_addr));
1440                         udp6_addr.sin6_port = htons(cfg->local_udp_port);
1441                         err = kernel_bind(sock, (struct sockaddr *) &udp6_addr,
1442                                           sizeof(udp6_addr));
1443                         if (err < 0)
1444                                 goto out;
1445
1446                         udp6_addr.sin6_family = AF_INET6;
1447                         memcpy(&udp6_addr.sin6_addr, cfg->peer_ip6,
1448                                sizeof(udp6_addr.sin6_addr));
1449                         udp6_addr.sin6_port = htons(cfg->peer_udp_port);
1450                         err = kernel_connect(sock,
1451                                              (struct sockaddr *) &udp6_addr,
1452                                              sizeof(udp6_addr), 0);
1453                         if (err < 0)
1454                                 goto out;
1455                 } else
1456 #endif
1457                 {
1458                         err = sock_create_kern(AF_INET, SOCK_DGRAM, 0, &sock);
1459                         if (err < 0)
1460                                 goto out;
1461
1462                         sk_change_net(sock->sk, net);
1463
1464                         udp_addr.sin_family = AF_INET;
1465                         udp_addr.sin_addr = cfg->local_ip;
1466                         udp_addr.sin_port = htons(cfg->local_udp_port);
1467                         err = kernel_bind(sock, (struct sockaddr *) &udp_addr,
1468                                           sizeof(udp_addr));
1469                         if (err < 0)
1470                                 goto out;
1471
1472                         udp_addr.sin_family = AF_INET;
1473                         udp_addr.sin_addr = cfg->peer_ip;
1474                         udp_addr.sin_port = htons(cfg->peer_udp_port);
1475                         err = kernel_connect(sock,
1476                                              (struct sockaddr *) &udp_addr,
1477                                              sizeof(udp_addr), 0);
1478                         if (err < 0)
1479                                 goto out;
1480                 }
1481
1482                 if (!cfg->use_udp_checksums)
1483                         sock->sk->sk_no_check = UDP_CSUM_NOXMIT;
1484
1485                 break;
1486
1487         case L2TP_ENCAPTYPE_IP:
1488 #if IS_ENABLED(CONFIG_IPV6)
1489                 if (cfg->local_ip6 && cfg->peer_ip6) {
1490                         err = sock_create_kern(AF_INET6, SOCK_DGRAM,
1491                                           IPPROTO_L2TP, &sock);
1492                         if (err < 0)
1493                                 goto out;
1494
1495                         sk_change_net(sock->sk, net);
1496
1497                         ip6_addr.l2tp_family = AF_INET6;
1498                         memcpy(&ip6_addr.l2tp_addr, cfg->local_ip6,
1499                                sizeof(ip6_addr.l2tp_addr));
1500                         ip6_addr.l2tp_conn_id = tunnel_id;
1501                         err = kernel_bind(sock, (struct sockaddr *) &ip6_addr,
1502                                           sizeof(ip6_addr));
1503                         if (err < 0)
1504                                 goto out;
1505
1506                         ip6_addr.l2tp_family = AF_INET6;
1507                         memcpy(&ip6_addr.l2tp_addr, cfg->peer_ip6,
1508                                sizeof(ip6_addr.l2tp_addr));
1509                         ip6_addr.l2tp_conn_id = peer_tunnel_id;
1510                         err = kernel_connect(sock,
1511                                              (struct sockaddr *) &ip6_addr,
1512                                              sizeof(ip6_addr), 0);
1513                         if (err < 0)
1514                                 goto out;
1515                 } else
1516 #endif
1517                 {
1518                         err = sock_create_kern(AF_INET, SOCK_DGRAM,
1519                                           IPPROTO_L2TP, &sock);
1520                         if (err < 0)
1521                                 goto out;
1522
1523                         sk_change_net(sock->sk, net);
1524
1525                         ip_addr.l2tp_family = AF_INET;
1526                         ip_addr.l2tp_addr = cfg->local_ip;
1527                         ip_addr.l2tp_conn_id = tunnel_id;
1528                         err = kernel_bind(sock, (struct sockaddr *) &ip_addr,
1529                                           sizeof(ip_addr));
1530                         if (err < 0)
1531                                 goto out;
1532
1533                         ip_addr.l2tp_family = AF_INET;
1534                         ip_addr.l2tp_addr = cfg->peer_ip;
1535                         ip_addr.l2tp_conn_id = peer_tunnel_id;
1536                         err = kernel_connect(sock, (struct sockaddr *) &ip_addr,
1537                                              sizeof(ip_addr), 0);
1538                         if (err < 0)
1539                                 goto out;
1540                 }
1541                 break;
1542
1543         default:
1544                 goto out;
1545         }
1546
1547 out:
1548         *sockp = sock;
1549         if ((err < 0) && sock) {
1550                 kernel_sock_shutdown(sock, SHUT_RDWR);
1551                 sk_release_kernel(sock->sk);
1552                 *sockp = NULL;
1553         }
1554
1555         return err;
1556 }
1557
1558 static struct lock_class_key l2tp_socket_class;
1559
1560 int l2tp_tunnel_create(struct net *net, int fd, int version, u32 tunnel_id, u32 peer_tunnel_id, struct l2tp_tunnel_cfg *cfg, struct l2tp_tunnel **tunnelp)
1561 {
1562         struct l2tp_tunnel *tunnel = NULL;
1563         int err;
1564         struct socket *sock = NULL;
1565         struct sock *sk = NULL;
1566         struct l2tp_net *pn;
1567         enum l2tp_encap_type encap = L2TP_ENCAPTYPE_UDP;
1568
1569         /* Get the tunnel socket from the fd, which was opened by
1570          * the userspace L2TP daemon. If not specified, create a
1571          * kernel socket.
1572          */
1573         if (fd < 0) {
1574                 err = l2tp_tunnel_sock_create(net, tunnel_id, peer_tunnel_id,
1575                                 cfg, &sock);
1576                 if (err < 0)
1577                         goto err;
1578         } else {
1579                 sock = sockfd_lookup(fd, &err);
1580                 if (!sock) {
1581                         pr_err("tunl %u: sockfd_lookup(fd=%d) returned %d\n",
1582                                tunnel_id, fd, err);
1583                         err = -EBADF;
1584                         goto err;
1585                 }
1586
1587                 /* Reject namespace mismatches */
1588                 if (!net_eq(sock_net(sock->sk), net)) {
1589                         pr_err("tunl %u: netns mismatch\n", tunnel_id);
1590                         err = -EINVAL;
1591                         goto err;
1592                 }
1593         }
1594
1595         sk = sock->sk;
1596
1597         if (cfg != NULL)
1598                 encap = cfg->encap;
1599
1600         /* Quick sanity checks */
1601         switch (encap) {
1602         case L2TP_ENCAPTYPE_UDP:
1603                 err = -EPROTONOSUPPORT;
1604                 if (sk->sk_protocol != IPPROTO_UDP) {
1605                         pr_err("tunl %hu: fd %d wrong protocol, got %d, expected %d\n",
1606                                tunnel_id, fd, sk->sk_protocol, IPPROTO_UDP);
1607                         goto err;
1608                 }
1609                 break;
1610         case L2TP_ENCAPTYPE_IP:
1611                 err = -EPROTONOSUPPORT;
1612                 if (sk->sk_protocol != IPPROTO_L2TP) {
1613                         pr_err("tunl %hu: fd %d wrong protocol, got %d, expected %d\n",
1614                                tunnel_id, fd, sk->sk_protocol, IPPROTO_L2TP);
1615                         goto err;
1616                 }
1617                 break;
1618         }
1619
1620         /* Check if this socket has already been prepped */
1621         tunnel = (struct l2tp_tunnel *)sk->sk_user_data;
1622         if (tunnel != NULL) {
1623                 /* This socket has already been prepped */
1624                 err = -EBUSY;
1625                 goto err;
1626         }
1627
1628         tunnel = kzalloc(sizeof(struct l2tp_tunnel), GFP_KERNEL);
1629         if (tunnel == NULL) {
1630                 err = -ENOMEM;
1631                 goto err;
1632         }
1633
1634         tunnel->version = version;
1635         tunnel->tunnel_id = tunnel_id;
1636         tunnel->peer_tunnel_id = peer_tunnel_id;
1637         tunnel->debug = L2TP_DEFAULT_DEBUG_FLAGS;
1638
1639         tunnel->magic = L2TP_TUNNEL_MAGIC;
1640         sprintf(&tunnel->name[0], "tunl %u", tunnel_id);
1641         rwlock_init(&tunnel->hlist_lock);
1642
1643         /* The net we belong to */
1644         tunnel->l2tp_net = net;
1645         pn = l2tp_pernet(net);
1646
1647         if (cfg != NULL)
1648                 tunnel->debug = cfg->debug;
1649
1650         /* Mark socket as an encapsulation socket. See net/ipv4/udp.c */
1651         tunnel->encap = encap;
1652         if (encap == L2TP_ENCAPTYPE_UDP) {
1653                 /* Mark socket as an encapsulation socket. See net/ipv4/udp.c */
1654                 udp_sk(sk)->encap_type = UDP_ENCAP_L2TPINUDP;
1655                 udp_sk(sk)->encap_rcv = l2tp_udp_encap_recv;
1656                 udp_sk(sk)->encap_destroy = l2tp_udp_encap_destroy;
1657 #if IS_ENABLED(CONFIG_IPV6)
1658                 if (sk->sk_family == PF_INET6)
1659                         udpv6_encap_enable();
1660                 else
1661 #endif
1662                 udp_encap_enable();
1663         }
1664
1665         sk->sk_user_data = tunnel;
1666
1667         /* Hook on the tunnel socket destructor so that we can cleanup
1668          * if the tunnel socket goes away.
1669          */
1670         tunnel->old_sk_destruct = sk->sk_destruct;
1671         sk->sk_destruct = &l2tp_tunnel_destruct;
1672         tunnel->sock = sk;
1673         tunnel->fd = fd;
1674         lockdep_set_class_and_name(&sk->sk_lock.slock, &l2tp_socket_class, "l2tp_sock");
1675
1676         sk->sk_allocation = GFP_ATOMIC;
1677
1678         /* Init delete workqueue struct */
1679         INIT_WORK(&tunnel->del_work, l2tp_tunnel_del_work);
1680
1681         /* Add tunnel to our list */
1682         INIT_LIST_HEAD(&tunnel->list);
1683         atomic_inc(&l2tp_tunnel_count);
1684
1685         /* Bump the reference count. The tunnel context is deleted
1686          * only when this drops to zero. Must be done before list insertion
1687          */
1688         l2tp_tunnel_inc_refcount(tunnel);
1689         spin_lock_bh(&pn->l2tp_tunnel_list_lock);
1690         list_add_rcu(&tunnel->list, &pn->l2tp_tunnel_list);
1691         spin_unlock_bh(&pn->l2tp_tunnel_list_lock);
1692
1693         err = 0;
1694 err:
1695         if (tunnelp)
1696                 *tunnelp = tunnel;
1697
1698         /* If tunnel's socket was created by the kernel, it doesn't
1699          *  have a file.
1700          */
1701         if (sock && sock->file)
1702                 sockfd_put(sock);
1703
1704         return err;
1705 }
1706 EXPORT_SYMBOL_GPL(l2tp_tunnel_create);
1707
1708 /* This function is used by the netlink TUNNEL_DELETE command.
1709  */
1710 int l2tp_tunnel_delete(struct l2tp_tunnel *tunnel)
1711 {
1712         l2tp_tunnel_closeall(tunnel);
1713         return (false == queue_work(l2tp_wq, &tunnel->del_work));
1714 }
1715 EXPORT_SYMBOL_GPL(l2tp_tunnel_delete);
1716
1717 /* Really kill the session.
1718  */
1719 void l2tp_session_free(struct l2tp_session *session)
1720 {
1721         struct l2tp_tunnel *tunnel = session->tunnel;
1722
1723         BUG_ON(atomic_read(&session->ref_count) != 0);
1724
1725         if (tunnel) {
1726                 BUG_ON(tunnel->magic != L2TP_TUNNEL_MAGIC);
1727                 if (session->session_id != 0)
1728                         atomic_dec(&l2tp_session_count);
1729                 sock_put(tunnel->sock);
1730                 session->tunnel = NULL;
1731                 l2tp_tunnel_dec_refcount(tunnel);
1732         }
1733
1734         kfree(session);
1735
1736         return;
1737 }
1738 EXPORT_SYMBOL_GPL(l2tp_session_free);
1739
1740 /* Remove an l2tp session from l2tp_core's hash lists.
1741  * Provides a tidyup interface for pseudowire code which can't just route all
1742  * shutdown via. l2tp_session_delete and a pseudowire-specific session_close
1743  * callback.
1744  */
1745 void __l2tp_session_unhash(struct l2tp_session *session)
1746 {
1747         struct l2tp_tunnel *tunnel = session->tunnel;
1748
1749         /* Remove the session from core hashes */
1750         if (tunnel) {
1751                 /* Remove from the per-tunnel hash */
1752                 write_lock_bh(&tunnel->hlist_lock);
1753                 hlist_del_init(&session->hlist);
1754                 write_unlock_bh(&tunnel->hlist_lock);
1755
1756                 /* For L2TPv3 we have a per-net hash: remove from there, too */
1757                 if (tunnel->version != L2TP_HDR_VER_2) {
1758                         struct l2tp_net *pn = l2tp_pernet(tunnel->l2tp_net);
1759                         spin_lock_bh(&pn->l2tp_session_hlist_lock);
1760                         hlist_del_init_rcu(&session->global_hlist);
1761                         spin_unlock_bh(&pn->l2tp_session_hlist_lock);
1762                         synchronize_rcu();
1763                 }
1764         }
1765 }
1766 EXPORT_SYMBOL_GPL(__l2tp_session_unhash);
1767
1768 /* This function is used by the netlink SESSION_DELETE command and by
1769    pseudowire modules.
1770  */
1771 int l2tp_session_delete(struct l2tp_session *session)
1772 {
1773         if (session->ref)
1774                 (*session->ref)(session);
1775         __l2tp_session_unhash(session);
1776         l2tp_session_queue_purge(session);
1777         if (session->session_close != NULL)
1778                 (*session->session_close)(session);
1779         if (session->deref)
1780                 (*session->ref)(session);
1781         l2tp_session_dec_refcount(session);
1782         return 0;
1783 }
1784 EXPORT_SYMBOL_GPL(l2tp_session_delete);
1785
1786 /* We come here whenever a session's send_seq, cookie_len or
1787  * l2specific_len parameters are set.
1788  */
1789 static void l2tp_session_set_header_len(struct l2tp_session *session, int version)
1790 {
1791         if (version == L2TP_HDR_VER_2) {
1792                 session->hdr_len = 6;
1793                 if (session->send_seq)
1794                         session->hdr_len += 4;
1795         } else {
1796                 session->hdr_len = 4 + session->cookie_len + session->l2specific_len + session->offset;
1797                 if (session->tunnel->encap == L2TP_ENCAPTYPE_UDP)
1798                         session->hdr_len += 4;
1799         }
1800
1801 }
1802
1803 struct l2tp_session *l2tp_session_create(int priv_size, struct l2tp_tunnel *tunnel, u32 session_id, u32 peer_session_id, struct l2tp_session_cfg *cfg)
1804 {
1805         struct l2tp_session *session;
1806
1807         session = kzalloc(sizeof(struct l2tp_session) + priv_size, GFP_KERNEL);
1808         if (session != NULL) {
1809                 session->magic = L2TP_SESSION_MAGIC;
1810                 session->tunnel = tunnel;
1811
1812                 session->session_id = session_id;
1813                 session->peer_session_id = peer_session_id;
1814                 session->nr = 0;
1815
1816                 sprintf(&session->name[0], "sess %u/%u",
1817                         tunnel->tunnel_id, session->session_id);
1818
1819                 skb_queue_head_init(&session->reorder_q);
1820
1821                 INIT_HLIST_NODE(&session->hlist);
1822                 INIT_HLIST_NODE(&session->global_hlist);
1823
1824                 /* Inherit debug options from tunnel */
1825                 session->debug = tunnel->debug;
1826
1827                 if (cfg) {
1828                         session->pwtype = cfg->pw_type;
1829                         session->debug = cfg->debug;
1830                         session->mtu = cfg->mtu;
1831                         session->mru = cfg->mru;
1832                         session->send_seq = cfg->send_seq;
1833                         session->recv_seq = cfg->recv_seq;
1834                         session->lns_mode = cfg->lns_mode;
1835                         session->reorder_timeout = cfg->reorder_timeout;
1836                         session->offset = cfg->offset;
1837                         session->l2specific_type = cfg->l2specific_type;
1838                         session->l2specific_len = cfg->l2specific_len;
1839                         session->cookie_len = cfg->cookie_len;
1840                         memcpy(&session->cookie[0], &cfg->cookie[0], cfg->cookie_len);
1841                         session->peer_cookie_len = cfg->peer_cookie_len;
1842                         memcpy(&session->peer_cookie[0], &cfg->peer_cookie[0], cfg->peer_cookie_len);
1843                 }
1844
1845                 if (tunnel->version == L2TP_HDR_VER_2)
1846                         session->build_header = l2tp_build_l2tpv2_header;
1847                 else
1848                         session->build_header = l2tp_build_l2tpv3_header;
1849
1850                 l2tp_session_set_header_len(session, tunnel->version);
1851
1852                 /* Bump the reference count. The session context is deleted
1853                  * only when this drops to zero.
1854                  */
1855                 l2tp_session_inc_refcount(session);
1856                 l2tp_tunnel_inc_refcount(tunnel);
1857
1858                 /* Ensure tunnel socket isn't deleted */
1859                 sock_hold(tunnel->sock);
1860
1861                 /* Add session to the tunnel's hash list */
1862                 write_lock_bh(&tunnel->hlist_lock);
1863                 hlist_add_head(&session->hlist,
1864                                l2tp_session_id_hash(tunnel, session_id));
1865                 write_unlock_bh(&tunnel->hlist_lock);
1866
1867                 /* And to the global session list if L2TPv3 */
1868                 if (tunnel->version != L2TP_HDR_VER_2) {
1869                         struct l2tp_net *pn = l2tp_pernet(tunnel->l2tp_net);
1870
1871                         spin_lock_bh(&pn->l2tp_session_hlist_lock);
1872                         hlist_add_head_rcu(&session->global_hlist,
1873                                            l2tp_session_id_hash_2(pn, session_id));
1874                         spin_unlock_bh(&pn->l2tp_session_hlist_lock);
1875                 }
1876
1877                 /* Ignore management session in session count value */
1878                 if (session->session_id != 0)
1879                         atomic_inc(&l2tp_session_count);
1880         }
1881
1882         return session;
1883 }
1884 EXPORT_SYMBOL_GPL(l2tp_session_create);
1885
1886 /*****************************************************************************
1887  * Init and cleanup
1888  *****************************************************************************/
1889
1890 static __net_init int l2tp_init_net(struct net *net)
1891 {
1892         struct l2tp_net *pn = net_generic(net, l2tp_net_id);
1893         int hash;
1894
1895         INIT_LIST_HEAD(&pn->l2tp_tunnel_list);
1896         spin_lock_init(&pn->l2tp_tunnel_list_lock);
1897
1898         for (hash = 0; hash < L2TP_HASH_SIZE_2; hash++)
1899                 INIT_HLIST_HEAD(&pn->l2tp_session_hlist[hash]);
1900
1901         spin_lock_init(&pn->l2tp_session_hlist_lock);
1902
1903         return 0;
1904 }
1905
1906 static __net_exit void l2tp_exit_net(struct net *net)
1907 {
1908         struct l2tp_net *pn = l2tp_pernet(net);
1909         struct l2tp_tunnel *tunnel = NULL;
1910
1911         rcu_read_lock_bh();
1912         list_for_each_entry_rcu(tunnel, &pn->l2tp_tunnel_list, list) {
1913                 (void)l2tp_tunnel_delete(tunnel);
1914         }
1915         rcu_read_unlock_bh();
1916 }
1917
1918 static struct pernet_operations l2tp_net_ops = {
1919         .init = l2tp_init_net,
1920         .exit = l2tp_exit_net,
1921         .id   = &l2tp_net_id,
1922         .size = sizeof(struct l2tp_net),
1923 };
1924
1925 static int __init l2tp_init(void)
1926 {
1927         int rc = 0;
1928
1929         rc = register_pernet_device(&l2tp_net_ops);
1930         if (rc)
1931                 goto out;
1932
1933         l2tp_wq = alloc_workqueue("l2tp", WQ_NON_REENTRANT | WQ_UNBOUND, 0);
1934         if (!l2tp_wq) {
1935                 pr_err("alloc_workqueue failed\n");
1936                 rc = -ENOMEM;
1937                 goto out;
1938         }
1939
1940         pr_info("L2TP core driver, %s\n", L2TP_DRV_VERSION);
1941
1942 out:
1943         return rc;
1944 }
1945
1946 static void __exit l2tp_exit(void)
1947 {
1948         unregister_pernet_device(&l2tp_net_ops);
1949         if (l2tp_wq) {
1950                 destroy_workqueue(l2tp_wq);
1951                 l2tp_wq = NULL;
1952         }
1953 }
1954
1955 module_init(l2tp_init);
1956 module_exit(l2tp_exit);
1957
1958 MODULE_AUTHOR("James Chapman <jchapman@katalix.com>");
1959 MODULE_DESCRIPTION("L2TP core");
1960 MODULE_LICENSE("GPL");
1961 MODULE_VERSION(L2TP_DRV_VERSION);
1962