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_tunnel *l2tp_tunnel(struct sock *sk)
 119 {
 120         return sk->sk_user_data;
 121 }
 122
 123 static inline struct l2tp_net *l2tp_pernet(struct net *net)
 124 {
 125         BUG_ON(!net);
 126
 127         return net_generic(net, l2tp_net_id);
 128 }
 129
 130 /* Tunnel reference counts. Incremented per session that is added to
 131  * the tunnel.
 132  */
 133 static inline void l2tp_tunnel_inc_refcount_1(struct l2tp_tunnel *tunnel)
 134 {
 135         atomic_inc(&tunnel->ref_count);
 136 }
 137
 138 static inline void l2tp_tunnel_dec_refcount_1(struct l2tp_tunnel *tunnel)
 139 {
 140         if (atomic_dec_and_test(&tunnel->ref_count))
 141                 l2tp_tunnel_free(tunnel);
 142 }
 143 #ifdef L2TP_REFCNT_DEBUG
 144 #define l2tp_tunnel_inc_refcount(_t)                                    \
 145 do {                                                                    \
 146         pr_debug("l2tp_tunnel_inc_refcount: %s:%d %s: cnt=%d\n",        \
 147                  __func__, __LINE__, (_t)->name,                        \
 148                  atomic_read(&_t->ref_count));                          \
 149         l2tp_tunnel_inc_refcount_1(_t);                                 \
 150 } while (0)
 151 #define l2tp_tunnel_dec_refcount(_t)
 152 do {                                                                    \
 153         pr_debug("l2tp_tunnel_dec_refcount: %s:%d %s: cnt=%d\n",        \
 154                  __func__, __LINE__, (_t)->name,                        \
 155                  atomic_read(&_t->ref_count));                          \
 156         l2tp_tunnel_dec_refcount_1(_t);                                 \
 157 } while (0)
 158 #else
 159 #define l2tp_tunnel_inc_refcount(t) l2tp_tunnel_inc_refcount_1(t)
 160 #define l2tp_tunnel_dec_refcount(t) l2tp_tunnel_dec_refcount_1(t)
 161 #endif
 162
 163 /* Session hash global list for L2TPv3.
 164  * The session_id SHOULD be random according to RFC3931, but several
 165  * L2TP implementations use incrementing session_ids.  So we do a real
 166  * hash on the session_id, rather than a simple bitmask.
 167  */
 168 static inline struct hlist_head *
 169 l2tp_session_id_hash_2(struct l2tp_net *pn, u32 session_id)
 170 {
 171         return &pn->l2tp_session_hlist[hash_32(session_id, L2TP_HASH_BITS_2)];
 172
 173 }
 174
 175 /* Lookup the tunnel socket, possibly involving the fs code if the socket is
 176  * owned by userspace.  A struct sock returned from this function must be
 177  * released using l2tp_tunnel_sock_put once you're done with it.
 178  */
 179 struct sock *l2tp_tunnel_sock_lookup(struct l2tp_tunnel *tunnel)
 180 {
 181         int err = 0;
 182         struct socket *sock = NULL;
 183         struct sock *sk = NULL;
 184
 185         if (!tunnel)
 186                 goto out;
 187
 188         if (tunnel->fd >= 0) {
 189                 /* Socket is owned by userspace, who might be in the process
 190                  * of closing it.  Look the socket up using the fd to ensure
 191                  * consistency.
 192                  */
 193                 sock = sockfd_lookup(tunnel->fd, &err);
 194                 if (sock)
 195                         sk = sock->sk;
 196         } else {
 197                 /* Socket is owned by kernelspace */
 198                 sk = tunnel->sock;
 199                 sock_hold(sk);
 200         }
 201
 202 out:
 203         return sk;
 204 }
 205 EXPORT_SYMBOL_GPL(l2tp_tunnel_sock_lookup);
 206
 207 /* Drop a reference to a tunnel socket obtained via. l2tp_tunnel_sock_put */
 208 void l2tp_tunnel_sock_put(struct sock *sk)
 209 {
 210         struct l2tp_tunnel *tunnel = l2tp_sock_to_tunnel(sk);
 211         if (tunnel) {
 212                 if (tunnel->fd >= 0) {
 213                         /* Socket is owned by userspace */
 214                         sockfd_put(sk->sk_socket);
 215                 }
 216                 sock_put(sk);
 217         }
 218         sock_put(sk);
 219 }
 220 EXPORT_SYMBOL_GPL(l2tp_tunnel_sock_put);
 221
 222 /* Lookup a session by id in the global session list
 223  */
 224 static struct l2tp_session *l2tp_session_find_2(struct net *net, u32 session_id)
 225 {
 226         struct l2tp_net *pn = l2tp_pernet(net);
 227         struct hlist_head *session_list =
 228                 l2tp_session_id_hash_2(pn, session_id);
 229         struct l2tp_session *session;
 230
 231         rcu_read_lock_bh();
 232         hlist_for_each_entry_rcu(session, session_list, global_hlist) {
 233                 if (session->session_id == session_id) {
 234                         rcu_read_unlock_bh();
 235                         return session;
 236                 }
 237         }
 238         rcu_read_unlock_bh();
 239
 240         return NULL;
 241 }
 242
 243 /* Session hash list.
 244  * The session_id SHOULD be random according to RFC2661, but several
 245  * L2TP implementations (Cisco and Microsoft) use incrementing
 246  * session_ids.  So we do a real hash on the session_id, rather than a
 247  * simple bitmask.
 248  */
 249 static inline struct hlist_head *
 250 l2tp_session_id_hash(struct l2tp_tunnel *tunnel, u32 session_id)
 251 {
 252         return &tunnel->session_hlist[hash_32(session_id, L2TP_HASH_BITS)];
 253 }
 254
 255 /* Lookup a session by id
 256  */
 257 struct l2tp_session *l2tp_session_find(struct net *net, struct l2tp_tunnel *tunnel, u32 session_id)
 258 {
 259         struct hlist_head *session_list;
 260         struct l2tp_session *session;
 261
 262         /* In L2TPv3, session_ids are unique over all tunnels and we
 263          * sometimes need to look them up before we know the
 264          * tunnel.
 265          */
 266         if (tunnel == NULL)
 267                 return l2tp_session_find_2(net, session_id);
 268
 269         session_list = l2tp_session_id_hash(tunnel, session_id);
 270         read_lock_bh(&tunnel->hlist_lock);
 271         hlist_for_each_entry(session, session_list, hlist) {
 272                 if (session->session_id == session_id) {
 273                         read_unlock_bh(&tunnel->hlist_lock);
 274                         return session;
 275                 }
 276         }
 277         read_unlock_bh(&tunnel->hlist_lock);
 278
 279         return NULL;
 280 }
 281 EXPORT_SYMBOL_GPL(l2tp_session_find);
 282
 283 struct l2tp_session *l2tp_session_find_nth(struct l2tp_tunnel *tunnel, int nth)
 284 {
 285         int hash;
 286         struct l2tp_session *session;
 287         int count = 0;
 288
 289         read_lock_bh(&tunnel->hlist_lock);
 290         for (hash = 0; hash < L2TP_HASH_SIZE; hash++) {
 291                 hlist_for_each_entry(session, &tunnel->session_hlist[hash], hlist) {
 292                         if (++count > nth) {
 293                                 read_unlock_bh(&tunnel->hlist_lock);
 294                                 return session;
 295                         }
 296                 }
 297         }
 298
 299         read_unlock_bh(&tunnel->hlist_lock);
 300
 301         return NULL;
 302 }
 303 EXPORT_SYMBOL_GPL(l2tp_session_find_nth);
 304
 305 /* Lookup a session by interface name.
 306  * This is very inefficient but is only used by management interfaces.
 307  */
 308 struct l2tp_session *l2tp_session_find_by_ifname(struct net *net, char *ifname)
 309 {
 310         struct l2tp_net *pn = l2tp_pernet(net);
 311         int hash;
 312         struct l2tp_session *session;
 313
 314         rcu_read_lock_bh();
 315         for (hash = 0; hash < L2TP_HASH_SIZE_2; hash++) {
 316                 hlist_for_each_entry_rcu(session, &pn->l2tp_session_hlist[hash], global_hlist) {
 317                         if (!strcmp(session->ifname, ifname)) {
 318                                 rcu_read_unlock_bh();
 319                                 return session;
 320                         }
 321                 }
 322         }
 323
 324         rcu_read_unlock_bh();
 325
 326         return NULL;
 327 }
 328 EXPORT_SYMBOL_GPL(l2tp_session_find_by_ifname);
 329
 330 /* Lookup a tunnel by id
 331  */
 332 struct l2tp_tunnel *l2tp_tunnel_find(struct net *net, u32 tunnel_id)
 333 {
 334         struct l2tp_tunnel *tunnel;
 335         struct l2tp_net *pn = l2tp_pernet(net);
 336
 337         rcu_read_lock_bh();
 338         list_for_each_entry_rcu(tunnel, &pn->l2tp_tunnel_list, list) {
 339                 if (tunnel->tunnel_id == tunnel_id) {
 340                         rcu_read_unlock_bh();
 341                         return tunnel;
 342                 }
 343         }
 344         rcu_read_unlock_bh();
 345
 346         return NULL;
 347 }
 348 EXPORT_SYMBOL_GPL(l2tp_tunnel_find);
 349
 350 struct l2tp_tunnel *l2tp_tunnel_find_nth(struct net *net, int nth)
 351 {
 352         struct l2tp_net *pn = l2tp_pernet(net);
 353         struct l2tp_tunnel *tunnel;
 354         int count = 0;
 355
 356         rcu_read_lock_bh();
 357         list_for_each_entry_rcu(tunnel, &pn->l2tp_tunnel_list, list) {
 358                 if (++count > nth) {
 359                         rcu_read_unlock_bh();
 360                         return tunnel;
 361                 }
 362         }
 363
 364         rcu_read_unlock_bh();
 365
 366         return NULL;
 367 }
 368 EXPORT_SYMBOL_GPL(l2tp_tunnel_find_nth);
 369
 370 /*****************************************************************************
 371  * Receive data handling
 372  *****************************************************************************/
 373
 374 /* Queue a skb in order. We come here only if the skb has an L2TP sequence
 375  * number.
 376  */
 377 static void l2tp_recv_queue_skb(struct l2tp_session *session, struct sk_buff *skb)
 378 {
 379         struct sk_buff *skbp;
 380         struct sk_buff *tmp;
 381         u32 ns = L2TP_SKB_CB(skb)->ns;
 382
 383         spin_lock_bh(&session->reorder_q.lock);
 384         skb_queue_walk_safe(&session->reorder_q, skbp, tmp) {
 385                 if (L2TP_SKB_CB(skbp)->ns > ns) {
 386                         __skb_queue_before(&session->reorder_q, skbp, skb);
 387                         l2tp_dbg(session, L2TP_MSG_SEQ,
 388                                  "%s: pkt %hu, inserted before %hu, reorder_q len=%d\n",
 389                                  session->name, ns, L2TP_SKB_CB(skbp)->ns,
 390                                  skb_queue_len(&session->reorder_q));
 391                         atomic_long_inc(&session->stats.rx_oos_packets);
 392                         goto out;
 393                 }
 394         }
 395
 396         __skb_queue_tail(&session->reorder_q, skb);
 397
 398 out:
 399         spin_unlock_bh(&session->reorder_q.lock);
 400 }
 401
 402 /* Dequeue a single skb.
 403  */
 404 static void l2tp_recv_dequeue_skb(struct l2tp_session *session, struct sk_buff *skb)
 405 {
 406         struct l2tp_tunnel *tunnel = session->tunnel;
 407         int length = L2TP_SKB_CB(skb)->length;
 408
 409         /* We're about to requeue the skb, so return resources
 410          * to its current owner (a socket receive buffer).
 411          */
 412         skb_orphan(skb);
 413
 414         atomic_long_inc(&tunnel->stats.rx_packets);
 415         atomic_long_add(length, &tunnel->stats.rx_bytes);
 416         atomic_long_inc(&session->stats.rx_packets);
 417         atomic_long_add(length, &session->stats.rx_bytes);
 418
 419         if (L2TP_SKB_CB(skb)->has_seq) {
 420                 /* Bump our Nr */
 421                 session->nr++;
 422                 if (tunnel->version == L2TP_HDR_VER_2)
 423                         session->nr &= 0xffff;
 424                 else
 425                         session->nr &= 0xffffff;
 426
 427                 l2tp_dbg(session, L2TP_MSG_SEQ, "%s: updated nr to %hu\n",
 428                          session->name, session->nr);
 429         }
 430
 431         /* call private receive handler */
 432         if (session->recv_skb != NULL)
 433                 (*session->recv_skb)(session, skb, L2TP_SKB_CB(skb)->length);
 434         else
 435                 kfree_skb(skb);
 436
 437         if (session->deref)
 438                 (*session->deref)(session);
 439 }
 440
 441 /* Dequeue skbs from the session's reorder_q, subject to packet order.
 442  * Skbs that have been in the queue for too long are simply discarded.
 443  */
 444 static void l2tp_recv_dequeue(struct l2tp_session *session)
 445 {
 446         struct sk_buff *skb;
 447         struct sk_buff *tmp;
 448
 449         /* If the pkt at the head of the queue has the nr that we
 450          * expect to send up next, dequeue it and any other
 451          * in-sequence packets behind it.
 452          */
 453 start:
 454         spin_lock_bh(&session->reorder_q.lock);
 455         skb_queue_walk_safe(&session->reorder_q, skb, tmp) {
 456                 if (time_after(jiffies, L2TP_SKB_CB(skb)->expires)) {
 457                         atomic_long_inc(&session->stats.rx_seq_discards);
 458                         atomic_long_inc(&session->stats.rx_errors);
 459                         l2tp_dbg(session, L2TP_MSG_SEQ,
 460                                  "%s: oos pkt %u len %d discarded (too old), waiting for %u, reorder_q_len=%d\n",
 461                                  session->name, L2TP_SKB_CB(skb)->ns,
 462                                  L2TP_SKB_CB(skb)->length, session->nr,
 463                                  skb_queue_len(&session->reorder_q));
 464                         session->reorder_skip = 1;
 465                         __skb_unlink(skb, &session->reorder_q);
 466                         kfree_skb(skb);
 467                         if (session->deref)
 468                                 (*session->deref)(session);
 469                         continue;
 470                 }
 471
 472                 if (L2TP_SKB_CB(skb)->has_seq) {
 473                         if (session->reorder_skip) {
 474                                 l2tp_dbg(session, L2TP_MSG_SEQ,
 475                                          "%s: advancing nr to next pkt: %u -> %u",
 476                                          session->name, session->nr,
 477                                          L2TP_SKB_CB(skb)->ns);
 478                                 session->reorder_skip = 0;
 479                                 session->nr = L2TP_SKB_CB(skb)->ns;
 480                         }
 481                         if (L2TP_SKB_CB(skb)->ns != session->nr) {
 482                                 l2tp_dbg(session, L2TP_MSG_SEQ,
 483                                          "%s: holding oos pkt %u len %d, waiting for %u, reorder_q_len=%d\n",
 484                                          session->name, L2TP_SKB_CB(skb)->ns,
 485                                          L2TP_SKB_CB(skb)->length, session->nr,
 486                                          skb_queue_len(&session->reorder_q));
 487                                 goto out;
 488                         }
 489                 }
 490                 __skb_unlink(skb, &session->reorder_q);
 491
 492                 /* Process the skb. We release the queue lock while we
 493                  * do so to let other contexts process the queue.
 494                  */
 495                 spin_unlock_bh(&session->reorder_q.lock);
 496                 l2tp_recv_dequeue_skb(session, skb);
 497                 goto start;
 498         }
 499
 500 out:
 501         spin_unlock_bh(&session->reorder_q.lock);
 502 }
 503
 504 static inline int l2tp_verify_udp_checksum(struct sock *sk,
 505                                            struct sk_buff *skb)
 506 {
 507         struct udphdr *uh = udp_hdr(skb);
 508         u16 ulen = ntohs(uh->len);
 509         __wsum psum;
 510
 511         if (sk->sk_no_check || skb_csum_unnecessary(skb))
 512                 return 0;
 513
 514 #if IS_ENABLED(CONFIG_IPV6)
 515         if (sk->sk_family == PF_INET6 && !l2tp_tunnel(sk)->v4mapped) {
 516                 if (!uh->check) {
 517                         LIMIT_NETDEBUG(KERN_INFO "L2TP: IPv6: checksum is 0\n");
 518                         return 1;
 519                 }
 520                 if ((skb->ip_summed == CHECKSUM_COMPLETE) &&
 521                     !csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
 522                                      &ipv6_hdr(skb)->daddr, ulen,
 523                                      IPPROTO_UDP, skb->csum)) {
 524                         skb->ip_summed = CHECKSUM_UNNECESSARY;
 525                         return 0;
 526                 }
 527                 skb->csum = ~csum_unfold(csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
 528                                                          &ipv6_hdr(skb)->daddr,
 529                                                          skb->len, IPPROTO_UDP,
 530                                                          0));
 531         } else
 532 #endif
 533         {
 534                 struct inet_sock *inet;
 535                 if (!uh->check)
 536                         return 0;
 537                 inet = inet_sk(sk);
 538                 psum = csum_tcpudp_nofold(inet->inet_saddr, inet->inet_daddr,
 539                                           ulen, IPPROTO_UDP, 0);
 540
 541                 if ((skb->ip_summed == CHECKSUM_COMPLETE) &&
 542                     !csum_fold(csum_add(psum, skb->csum)))
 543                         return 0;
 544                 skb->csum = psum;
 545         }
 546
 547         return __skb_checksum_complete(skb);
 548 }
 549
 550 /* Do receive processing of L2TP data frames. We handle both L2TPv2
 551  * and L2TPv3 data frames here.
 552  *
 553  * L2TPv2 Data Message Header
 554  *
 555  *  0                   1                   2                   3
 556  *  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
 557  * +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
 558  * |T|L|x|x|S|x|O|P|x|x|x|x|  Ver  |          Length (opt)         |
 559  * +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
 560  * |           Tunnel ID           |           Session ID          |
 561  * +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
 562  * |             Ns (opt)          |             Nr (opt)          |
 563  * +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
 564  * |      Offset Size (opt)        |    Offset pad... (opt)
 565  * +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
 566  *
 567  * Data frames are marked by T=0. All other fields are the same as
 568  * those in L2TP control frames.
 569  *
 570  * L2TPv3 Data Message Header
 571  *
 572  * +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
 573  * |                      L2TP Session Header                      |
 574  * +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
 575  * |                      L2-Specific Sublayer                     |
 576  * +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
 577  * |                        Tunnel Payload                      ...
 578  * +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
 579  *
 580  * L2TPv3 Session Header Over IP
 581  *
 582  *  0                   1                   2                   3
 583  *  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
 584  * +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
 585  * |                           Session ID                          |
 586  * +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
 587  * |               Cookie (optional, maximum 64 bits)...
 588  * +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
 589  *                                                                 |
 590  * +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
 591  *
 592  * L2TPv3 L2-Specific Sublayer Format
 593  *
 594  *  0                   1                   2                   3
 595  *  0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1
 596  * +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
 597  * |x|S|x|x|x|x|x|x|              Sequence Number                  |
 598  * +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
 599  *
 600  * Cookie value, sublayer format and offset (pad) are negotiated with
 601  * the peer when the session is set up. Unlike L2TPv2, we do not need
 602  * to parse the packet header to determine if optional fields are
 603  * present.
 604  *
 605  * Caller must already have parsed the frame and determined that it is
 606  * a data (not control) frame before coming here. Fields up to the
 607  * session-id have already been parsed and ptr points to the data
 608  * after the session-id.
 609  */
 610 void l2tp_recv_common(struct l2tp_session *session, struct sk_buff *skb,
 611                       unsigned char *ptr, unsigned char *optr, u16 hdrflags,
 612                       int length, int (*payload_hook)(struct sk_buff *skb))
 613 {
 614         struct l2tp_tunnel *tunnel = session->tunnel;
 615         int offset;
 616         u32 ns, nr;
 617
 618         /* The ref count is increased since we now hold a pointer to
 619          * the session. Take care to decrement the refcnt when exiting
 620          * this function from now on...
 621          */
 622         l2tp_session_inc_refcount(session);
 623         if (session->ref)
 624                 (*session->ref)(session);
 625
 626         /* Parse and check optional cookie */
 627         if (session->peer_cookie_len > 0) {
 628                 if (memcmp(ptr, &session->peer_cookie[0], session->peer_cookie_len)) {
 629                         l2tp_info(tunnel, L2TP_MSG_DATA,
 630                                   "%s: cookie mismatch (%u/%u). Discarding.\n",
 631                                   tunnel->name, tunnel->tunnel_id,
 632                                   session->session_id);
 633                         atomic_long_inc(&session->stats.rx_cookie_discards);
 634                         goto discard;
 635                 }
 636                 ptr += session->peer_cookie_len;
 637         }
 638
 639         /* Handle the optional sequence numbers. Sequence numbers are
 640          * in different places for L2TPv2 and L2TPv3.
 641          *
 642          * If we are the LAC, enable/disable sequence numbers under
 643          * the control of the LNS.  If no sequence numbers present but
 644          * we were expecting them, discard frame.
 645          */
 646         ns = nr = 0;
 647         L2TP_SKB_CB(skb)->has_seq = 0;
 648         if (tunnel->version == L2TP_HDR_VER_2) {
 649                 if (hdrflags & L2TP_HDRFLAG_S) {
 650                         ns = ntohs(*(__be16 *) ptr);
 651                         ptr += 2;
 652                         nr = ntohs(*(__be16 *) ptr);
 653                         ptr += 2;
 654
 655                         /* Store L2TP info in the skb */
 656                         L2TP_SKB_CB(skb)->ns = ns;
 657                         L2TP_SKB_CB(skb)->has_seq = 1;
 658
 659                         l2tp_dbg(session, L2TP_MSG_SEQ,
 660                                  "%s: recv data ns=%u, nr=%u, session nr=%u\n",
 661                                  session->name, ns, nr, session->nr);
 662                 }
 663         } else if (session->l2specific_type == L2TP_L2SPECTYPE_DEFAULT) {
 664                 u32 l2h = ntohl(*(__be32 *) ptr);
 665
 666                 if (l2h & 0x40000000) {
 667                         ns = l2h & 0x00ffffff;
 668
 669                         /* Store L2TP info in the skb */
 670                         L2TP_SKB_CB(skb)->ns = ns;
 671                         L2TP_SKB_CB(skb)->has_seq = 1;
 672
 673                         l2tp_dbg(session, L2TP_MSG_SEQ,
 674                                  "%s: recv data ns=%u, session nr=%u\n",
 675                                  session->name, ns, session->nr);
 676                 }
 677         }
 678
 679         /* Advance past L2-specific header, if present */
 680         ptr += session->l2specific_len;
 681
 682         if (L2TP_SKB_CB(skb)->has_seq) {
 683                 /* Received a packet with sequence numbers. If we're the LNS,
 684                  * check if we sre sending sequence numbers and if not,
 685                  * configure it so.
 686                  */
 687                 if ((!session->lns_mode) && (!session->send_seq)) {
 688                         l2tp_info(session, L2TP_MSG_SEQ,
 689                                   "%s: requested to enable seq numbers by LNS\n",
 690                                   session->name);
 691                         session->send_seq = -1;
 692                         l2tp_session_set_header_len(session, tunnel->version);
 693                 }
 694         } else {
 695                 /* No sequence numbers.
 696                  * If user has configured mandatory sequence numbers, discard.
 697                  */
 698                 if (session->recv_seq) {
 699                         l2tp_warn(session, L2TP_MSG_SEQ,
 700                                   "%s: recv data has no seq numbers when required. Discarding.\n",
 701                                   session->name);
 702                         atomic_long_inc(&session->stats.rx_seq_discards);
 703                         goto discard;
 704                 }
 705
 706                 /* If we're the LAC and we're sending sequence numbers, the
 707                  * LNS has requested that we no longer send sequence numbers.
 708                  * If we're the LNS and we're sending sequence numbers, the
 709                  * LAC is broken. Discard the frame.
 710                  */
 711                 if ((!session->lns_mode) && (session->send_seq)) {
 712                         l2tp_info(session, L2TP_MSG_SEQ,
 713                                   "%s: requested to disable seq numbers by LNS\n",
 714                                   session->name);
 715                         session->send_seq = 0;
 716                         l2tp_session_set_header_len(session, tunnel->version);
 717                 } else if (session->send_seq) {
 718                         l2tp_warn(session, L2TP_MSG_SEQ,
 719                                   "%s: recv data has no seq numbers when required. Discarding.\n",
 720                                   session->name);
 721                         atomic_long_inc(&session->stats.rx_seq_discards);
 722                         goto discard;
 723                 }
 724         }
 725
 726         /* Session data offset is handled differently for L2TPv2 and
 727          * L2TPv3. For L2TPv2, there is an optional 16-bit value in
 728          * the header. For L2TPv3, the offset is negotiated using AVPs
 729          * in the session setup control protocol.
 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         } else
 738                 ptr += session->offset;
 739
 740         offset = ptr - optr;
 741         if (!pskb_may_pull(skb, offset))
 742                 goto discard;
 743
 744         __skb_pull(skb, offset);
 745
 746         /* If caller wants to process the payload before we queue the
 747          * packet, do so now.
 748          */
 749         if (payload_hook)
 750                 if ((*payload_hook)(skb))
 751                         goto discard;
 752
 753         /* Prepare skb for adding to the session's reorder_q.  Hold
 754          * packets for max reorder_timeout or 1 second if not
 755          * reordering.
 756          */
 757         L2TP_SKB_CB(skb)->length = length;
 758         L2TP_SKB_CB(skb)->expires = jiffies +
 759                 (session->reorder_timeout ? session->reorder_timeout : HZ);
 760
 761         /* Add packet to the session's receive queue. Reordering is done here, if
 762          * enabled. Saved L2TP protocol info is stored in skb->sb[].
 763          */
 764         if (L2TP_SKB_CB(skb)->has_seq) {
 765                 if (session->reorder_timeout != 0) {
 766                         /* Packet reordering enabled. Add skb to session's
 767                          * reorder queue, in order of ns.
 768                          */
 769                         l2tp_recv_queue_skb(session, skb);
 770                 } else {
 771                         /* Packet reordering disabled. Discard out-of-sequence
 772                          * packets
 773                          */
 774                         if (L2TP_SKB_CB(skb)->ns != session->nr) {
 775                                 atomic_long_inc(&session->stats.rx_seq_discards);
 776                                 l2tp_dbg(session, L2TP_MSG_SEQ,
 777                                          "%s: oos pkt %u len %d discarded, waiting for %u, reorder_q_len=%d\n",
 778                                          session->name, L2TP_SKB_CB(skb)->ns,
 779                                          L2TP_SKB_CB(skb)->length, session->nr,
 780                                          skb_queue_len(&session->reorder_q));
 781                                 goto discard;
 782                         }
 783                         skb_queue_tail(&session->reorder_q, skb);
 784                 }
 785         } else {
 786                 /* No sequence numbers. Add the skb to the tail of the
 787                  * reorder queue. This ensures that it will be
 788                  * delivered after all previous sequenced skbs.
 789                  */
 790                 skb_queue_tail(&session->reorder_q, skb);
 791         }
 792
 793         /* Try to dequeue as many skbs from reorder_q as we can. */
 794         l2tp_recv_dequeue(session);
 795
 796         l2tp_session_dec_refcount(session);
 797
 798         return;
 799
 800 discard:
 801         atomic_long_inc(&session->stats.rx_errors);
 802         kfree_skb(skb);
 803
 804         if (session->deref)
 805                 (*session->deref)(session);
 806
 807         l2tp_session_dec_refcount(session);
 808 }
 809 EXPORT_SYMBOL(l2tp_recv_common);
 810
 811 /* Drop skbs from the session's reorder_q
 812  */
 813 int l2tp_session_queue_purge(struct l2tp_session *session)
 814 {
 815         struct sk_buff *skb = NULL;
 816         BUG_ON(!session);
 817         BUG_ON(session->magic != L2TP_SESSION_MAGIC);
 818         while ((skb = skb_dequeue(&session->reorder_q))) {
 819                 atomic_long_inc(&session->stats.rx_errors);
 820                 kfree_skb(skb);
 821                 if (session->deref)
 822                         (*session->deref)(session);
 823         }
 824         return 0;
 825 }
 826 EXPORT_SYMBOL_GPL(l2tp_session_queue_purge);
 827
 828 /* Internal UDP receive frame. Do the real work of receiving an L2TP data frame
 829  * here. The skb is not on a list when we get here.
 830  * Returns 0 if the packet was a data packet and was successfully passed on.
 831  * Returns 1 if the packet was not a good data packet and could not be
 832  * forwarded.  All such packets are passed up to userspace to deal with.
 833  */
 834 static int l2tp_udp_recv_core(struct l2tp_tunnel *tunnel, struct sk_buff *skb,
 835                               int (*payload_hook)(struct sk_buff *skb))
 836 {
 837         struct l2tp_session *session = NULL;
 838         unsigned char *ptr, *optr;
 839         u16 hdrflags;
 840         u32 tunnel_id, session_id;
 841         u16 version;
 842         int length;
 843
 844         if (tunnel->sock && l2tp_verify_udp_checksum(tunnel->sock, skb))
 845                 goto discard_bad_csum;
 846
 847         /* UDP always verifies the packet length. */
 848         __skb_pull(skb, sizeof(struct udphdr));
 849
 850         /* Short packet? */
 851         if (!pskb_may_pull(skb, L2TP_HDR_SIZE_SEQ)) {
 852                 l2tp_info(tunnel, L2TP_MSG_DATA,
 853                           "%s: recv short packet (len=%d)\n",
 854                           tunnel->name, skb->len);
 855                 goto error;
 856         }
 857
 858         /* Trace packet contents, if enabled */
 859         if (tunnel->debug & L2TP_MSG_DATA) {
 860                 length = min(32u, skb->len);
 861                 if (!pskb_may_pull(skb, length))
 862                         goto error;
 863
 864                 pr_debug("%s: recv\n", tunnel->name);
 865                 print_hex_dump_bytes("", DUMP_PREFIX_OFFSET, skb->data, length);
 866         }
 867
 868         /* Point to L2TP header */
 869         optr = ptr = skb->data;
 870
 871         /* Get L2TP header flags */
 872         hdrflags = ntohs(*(__be16 *) ptr);
 873
 874         /* Check protocol version */
 875         version = hdrflags & L2TP_HDR_VER_MASK;
 876         if (version != tunnel->version) {
 877                 l2tp_info(tunnel, L2TP_MSG_DATA,
 878                           "%s: recv protocol version mismatch: got %d expected %d\n",
 879                           tunnel->name, version, tunnel->version);
 880                 goto error;
 881         }
 882
 883         /* Get length of L2TP packet */
 884         length = skb->len;
 885
 886         /* If type is control packet, it is handled by userspace. */
 887         if (hdrflags & L2TP_HDRFLAG_T) {
 888                 l2tp_dbg(tunnel, L2TP_MSG_DATA,
 889                          "%s: recv control packet, len=%d\n",
 890                          tunnel->name, length);
 891                 goto error;
 892         }
 893
 894         /* Skip flags */
 895         ptr += 2;
 896
 897         if (tunnel->version == L2TP_HDR_VER_2) {
 898                 /* If length is present, skip it */
 899                 if (hdrflags & L2TP_HDRFLAG_L)
 900                         ptr += 2;
 901
 902                 /* Extract tunnel and session ID */
 903                 tunnel_id = ntohs(*(__be16 *) ptr);
 904                 ptr += 2;
 905                 session_id = ntohs(*(__be16 *) ptr);
 906                 ptr += 2;
 907         } else {
 908                 ptr += 2;       /* skip reserved bits */
 909                 tunnel_id = tunnel->tunnel_id;
 910                 session_id = ntohl(*(__be32 *) ptr);
 911                 ptr += 4;
 912         }
 913
 914         /* Find the session context */
 915         session = l2tp_session_find(tunnel->l2tp_net, tunnel, session_id);
 916         if (!session || !session->recv_skb) {
 917                 /* Not found? Pass to userspace to deal with */
 918                 l2tp_info(tunnel, L2TP_MSG_DATA,
 919                           "%s: no session found (%u/%u). Passing up.\n",
 920                           tunnel->name, tunnel_id, session_id);
 921                 goto error;
 922         }
 923
 924         l2tp_recv_common(session, skb, ptr, optr, hdrflags, length, payload_hook);
 925
 926         return 0;
 927
 928 discard_bad_csum:
 929         LIMIT_NETDEBUG("%s: UDP: bad checksum\n", tunnel->name);
 930         UDP_INC_STATS_USER(tunnel->l2tp_net, UDP_MIB_INERRORS, 0);
 931         atomic_long_inc(&tunnel->stats.rx_errors);
 932         kfree_skb(skb);
 933
 934         return 0;
 935
 936 error:
 937         /* Put UDP header back */
 938         __skb_push(skb, sizeof(struct udphdr));
 939
 940         return 1;
 941 }
 942
 943 /* UDP encapsulation receive handler. See net/ipv4/udp.c.
 944  * Return codes:
 945  * 0 : success.
 946  * <0: error
 947  * >0: skb should be passed up to userspace as UDP.
 948  */
 949 int l2tp_udp_encap_recv(struct sock *sk, struct sk_buff *skb)
 950 {
 951         struct l2tp_tunnel *tunnel;
 952
 953         tunnel = l2tp_sock_to_tunnel(sk);
 954         if (tunnel == NULL)
 955                 goto pass_up;
 956
 957         l2tp_dbg(tunnel, L2TP_MSG_DATA, "%s: received %d bytes\n",
 958                  tunnel->name, skb->len);
 959
 960         if (l2tp_udp_recv_core(tunnel, skb, tunnel->recv_payload_hook))
 961                 goto pass_up_put;
 962
 963         sock_put(sk);
 964         return 0;
 965
 966 pass_up_put:
 967         sock_put(sk);
 968 pass_up:
 969         return 1;
 970 }
 971 EXPORT_SYMBOL_GPL(l2tp_udp_encap_recv);
 972
 973 /************************************************************************
 974  * Transmit handling
 975  ***********************************************************************/
 976
 977 /* Build an L2TP header for the session into the buffer provided.
 978  */
 979 static int l2tp_build_l2tpv2_header(struct l2tp_session *session, void *buf)
 980 {
 981         struct l2tp_tunnel *tunnel = session->tunnel;
 982         __be16 *bufp = buf;
 983         __be16 *optr = buf;
 984         u16 flags = L2TP_HDR_VER_2;
 985         u32 tunnel_id = tunnel->peer_tunnel_id;
 986         u32 session_id = session->peer_session_id;
 987
 988         if (session->send_seq)
 989                 flags |= L2TP_HDRFLAG_S;
 990
 991         /* Setup L2TP header. */
 992         *bufp++ = htons(flags);
 993         *bufp++ = htons(tunnel_id);
 994         *bufp++ = htons(session_id);
 995         if (session->send_seq) {
 996                 *bufp++ = htons(session->ns);
 997                 *bufp++ = 0;
 998                 session->ns++;
 999                 session->ns &= 0xffff;
1000                 l2tp_dbg(session, L2TP_MSG_SEQ, "%s: updated ns to %u\n",
1001                          session->name, session->ns);
1002         }
1003
1004         return bufp - optr;
1005 }
1006
1007 static int l2tp_build_l2tpv3_header(struct l2tp_session *session, void *buf)
1008 {
1009         struct l2tp_tunnel *tunnel = session->tunnel;
1010         char *bufp = buf;
1011         char *optr = bufp;
1012
1013         /* Setup L2TP header. The header differs slightly for UDP and
1014          * IP encapsulations. For UDP, there is 4 bytes of flags.
1015          */
1016         if (tunnel->encap == L2TP_ENCAPTYPE_UDP) {
1017                 u16 flags = L2TP_HDR_VER_3;
1018                 *((__be16 *) bufp) = htons(flags);
1019                 bufp += 2;
1020                 *((__be16 *) bufp) = 0;
1021                 bufp += 2;
1022         }
1023
1024         *((__be32 *) bufp) = htonl(session->peer_session_id);
1025         bufp += 4;
1026         if (session->cookie_len) {
1027                 memcpy(bufp, &session->cookie[0], session->cookie_len);
1028                 bufp += session->cookie_len;
1029         }
1030         if (session->l2specific_len) {
1031                 if (session->l2specific_type == L2TP_L2SPECTYPE_DEFAULT) {
1032                         u32 l2h = 0;
1033                         if (session->send_seq) {
1034                                 l2h = 0x40000000 | session->ns;
1035                                 session->ns++;
1036                                 session->ns &= 0xffffff;
1037                                 l2tp_dbg(session, L2TP_MSG_SEQ,
1038                                          "%s: updated ns to %u\n",
1039                                          session->name, session->ns);
1040                         }
1041
1042                         *((__be32 *) bufp) = htonl(l2h);
1043                 }
1044                 bufp += session->l2specific_len;
1045         }
1046         if (session->offset)
1047                 bufp += session->offset;
1048
1049         return bufp - optr;
1050 }
1051
1052 static int l2tp_xmit_core(struct l2tp_session *session, struct sk_buff *skb,
1053                           struct flowi *fl, size_t data_len)
1054 {
1055         struct l2tp_tunnel *tunnel = session->tunnel;
1056         unsigned int len = skb->len;
1057         int error;
1058
1059         /* Debug */
1060         if (session->send_seq)
1061                 l2tp_dbg(session, L2TP_MSG_DATA, "%s: send %Zd bytes, ns=%u\n",
1062                          session->name, data_len, session->ns - 1);
1063         else
1064                 l2tp_dbg(session, L2TP_MSG_DATA, "%s: send %Zd bytes\n",
1065                          session->name, data_len);
1066
1067         if (session->debug & L2TP_MSG_DATA) {
1068                 int uhlen = (tunnel->encap == L2TP_ENCAPTYPE_UDP) ? sizeof(struct udphdr) : 0;
1069                 unsigned char *datap = skb->data + uhlen;
1070
1071                 pr_debug("%s: xmit\n", session->name);
1072                 print_hex_dump_bytes("", DUMP_PREFIX_OFFSET,
1073                                      datap, min_t(size_t, 32, len - uhlen));
1074         }
1075
1076         /* Queue the packet to IP for output */
1077         skb->local_df = 1;
1078 #if IS_ENABLED(CONFIG_IPV6)
1079         if (skb->sk->sk_family == PF_INET6 && !tunnel->v4mapped)
1080                 error = inet6_csk_xmit(skb, NULL);
1081         else
1082 #endif
1083                 error = ip_queue_xmit(skb, fl);
1084
1085         /* Update stats */
1086         if (error >= 0) {
1087                 atomic_long_inc(&tunnel->stats.tx_packets);
1088                 atomic_long_add(len, &tunnel->stats.tx_bytes);
1089                 atomic_long_inc(&session->stats.tx_packets);
1090                 atomic_long_add(len, &session->stats.tx_bytes);
1091         } else {
1092                 atomic_long_inc(&tunnel->stats.tx_errors);
1093                 atomic_long_inc(&session->stats.tx_errors);
1094         }
1095
1096         return 0;
1097 }
1098
1099 /* Automatically called when the skb is freed.
1100  */
1101 static void l2tp_sock_wfree(struct sk_buff *skb)
1102 {
1103         sock_put(skb->sk);
1104 }
1105
1106 /* For data skbs that we transmit, we associate with the tunnel socket
1107  * but don't do accounting.
1108  */
1109 static inline void l2tp_skb_set_owner_w(struct sk_buff *skb, struct sock *sk)
1110 {
1111         sock_hold(sk);
1112         skb->sk = sk;
1113         skb->destructor = l2tp_sock_wfree;
1114 }
1115
1116 #if IS_ENABLED(CONFIG_IPV6)
1117 static void l2tp_xmit_ipv6_csum(struct sock *sk, struct sk_buff *skb,
1118                                 int udp_len)
1119 {
1120         struct ipv6_pinfo *np = inet6_sk(sk);
1121         struct udphdr *uh = udp_hdr(skb);
1122
1123         if (!skb_dst(skb) || !skb_dst(skb)->dev ||
1124             !(skb_dst(skb)->dev->features & NETIF_F_IPV6_CSUM)) {
1125                 __wsum csum = skb_checksum(skb, 0, udp_len, 0);
1126                 skb->ip_summed = CHECKSUM_UNNECESSARY;
1127                 uh->check = csum_ipv6_magic(&np->saddr, &np->daddr, udp_len,
1128                                             IPPROTO_UDP, csum);
1129                 if (uh->check == 0)
1130                         uh->check = CSUM_MANGLED_0;
1131         } else {
1132                 skb->ip_summed = CHECKSUM_PARTIAL;
1133                 skb->csum_start = skb_transport_header(skb) - skb->head;
1134                 skb->csum_offset = offsetof(struct udphdr, check);
1135                 uh->check = ~csum_ipv6_magic(&np->saddr, &np->daddr,
1136                                              udp_len, IPPROTO_UDP, 0);
1137         }
1138 }
1139 #endif
1140
1141 /* If caller requires the skb to have a ppp header, the header must be
1142  * inserted in the skb data before calling this function.
1143  */
1144 int l2tp_xmit_skb(struct l2tp_session *session, struct sk_buff *skb, int hdr_len)
1145 {
1146         int data_len = skb->len;
1147         struct l2tp_tunnel *tunnel = session->tunnel;
1148         struct sock *sk = tunnel->sock;
1149         struct flowi *fl;
1150         struct udphdr *uh;
1151         struct inet_sock *inet;
1152         __wsum csum;
1153         int headroom;
1154         int uhlen = (tunnel->encap == L2TP_ENCAPTYPE_UDP) ? sizeof(struct udphdr) : 0;
1155         int udp_len;
1156         int ret = NET_XMIT_SUCCESS;
1157
1158         /* Check that there's enough headroom in the skb to insert IP,
1159          * UDP and L2TP headers. If not enough, expand it to
1160          * make room. Adjust truesize.
1161          */
1162         headroom = NET_SKB_PAD + sizeof(struct iphdr) +
1163                 uhlen + hdr_len;
1164         if (skb_cow_head(skb, headroom)) {
1165                 kfree_skb(skb);
1166                 return NET_XMIT_DROP;
1167         }
1168
1169         skb_orphan(skb);
1170         /* Setup L2TP header */
1171         session->build_header(session, __skb_push(skb, hdr_len));
1172
1173         /* Reset skb netfilter state */
1174         memset(&(IPCB(skb)->opt), 0, sizeof(IPCB(skb)->opt));
1175         IPCB(skb)->flags &= ~(IPSKB_XFRM_TUNNEL_SIZE | IPSKB_XFRM_TRANSFORMED |
1176                               IPSKB_REROUTED);
1177         nf_reset(skb);
1178
1179         bh_lock_sock(sk);
1180         if (sock_owned_by_user(sk)) {
1181                 kfree_skb(skb);
1182                 ret = NET_XMIT_DROP;
1183                 goto out_unlock;
1184         }
1185
1186         /* Get routing info from the tunnel socket */
1187         skb_dst_drop(skb);
1188         skb_dst_set(skb, dst_clone(__sk_dst_check(sk, 0)));
1189
1190         inet = inet_sk(sk);
1191         fl = &inet->cork.fl;
1192         switch (tunnel->encap) {
1193         case L2TP_ENCAPTYPE_UDP:
1194                 /* Setup UDP header */
1195                 __skb_push(skb, sizeof(*uh));
1196                 skb_reset_transport_header(skb);
1197                 uh = udp_hdr(skb);
1198                 uh->source = inet->inet_sport;
1199                 uh->dest = inet->inet_dport;
1200                 udp_len = uhlen + hdr_len + data_len;
1201                 uh->len = htons(udp_len);
1202                 uh->check = 0;
1203
1204                 /* Calculate UDP checksum if configured to do so */
1205 #if IS_ENABLED(CONFIG_IPV6)
1206                 if (sk->sk_family == PF_INET6 && !tunnel->v4mapped)
1207                         l2tp_xmit_ipv6_csum(sk, skb, udp_len);
1208                 else
1209 #endif
1210                 if (sk->sk_no_check == UDP_CSUM_NOXMIT)
1211                         skb->ip_summed = CHECKSUM_NONE;
1212                 else if ((skb_dst(skb) && skb_dst(skb)->dev) &&
1213                          (!(skb_dst(skb)->dev->features & NETIF_F_V4_CSUM))) {
1214                         skb->ip_summed = CHECKSUM_COMPLETE;
1215                         csum = skb_checksum(skb, 0, udp_len, 0);
1216                         uh->check = csum_tcpudp_magic(inet->inet_saddr,
1217                                                       inet->inet_daddr,
1218                                                       udp_len, IPPROTO_UDP, csum);
1219                         if (uh->check == 0)
1220                                 uh->check = CSUM_MANGLED_0;
1221                 } else {
1222                         skb->ip_summed = CHECKSUM_PARTIAL;
1223                         skb->csum_start = skb_transport_header(skb) - skb->head;
1224                         skb->csum_offset = offsetof(struct udphdr, check);
1225                         uh->check = ~csum_tcpudp_magic(inet->inet_saddr,
1226                                                        inet->inet_daddr,
1227                                                        udp_len, IPPROTO_UDP, 0);
1228                 }
1229                 break;
1230
1231         case L2TP_ENCAPTYPE_IP:
1232                 break;
1233         }
1234
1235         l2tp_skb_set_owner_w(skb, sk);
1236
1237         l2tp_xmit_core(session, skb, fl, data_len);
1238 out_unlock:
1239         bh_unlock_sock(sk);
1240
1241         return ret;
1242 }
1243 EXPORT_SYMBOL_GPL(l2tp_xmit_skb);
1244
1245 /*****************************************************************************
1246  * Tinnel and session create/destroy.
1247  *****************************************************************************/
1248
1249 /* Tunnel socket destruct hook.
1250  * The tunnel context is deleted only when all session sockets have been
1251  * closed.
1252  */
1253 static void l2tp_tunnel_destruct(struct sock *sk)
1254 {
1255         struct l2tp_tunnel *tunnel = l2tp_tunnel(sk);
1256         struct l2tp_net *pn;
1257
1258         if (tunnel == NULL)
1259                 goto end;
1260
1261         l2tp_info(tunnel, L2TP_MSG_CONTROL, "%s: closing...\n", tunnel->name);
1262
1263
1264         /* Disable udp encapsulation */
1265         switch (tunnel->encap) {
1266         case L2TP_ENCAPTYPE_UDP:
1267                 /* No longer an encapsulation socket. See net/ipv4/udp.c */
1268                 (udp_sk(sk))->encap_type = 0;
1269                 (udp_sk(sk))->encap_rcv = NULL;
1270                 (udp_sk(sk))->encap_destroy = NULL;
1271                 break;
1272         case L2TP_ENCAPTYPE_IP:
1273                 break;
1274         }
1275
1276         /* Remove hooks into tunnel socket */
1277         sk->sk_destruct = tunnel->old_sk_destruct;
1278         sk->sk_user_data = NULL;
1279         tunnel->sock = NULL;
1280
1281         /* Remove the tunnel struct from the tunnel list */
1282         pn = l2tp_pernet(tunnel->l2tp_net);
1283         spin_lock_bh(&pn->l2tp_tunnel_list_lock);
1284         list_del_rcu(&tunnel->list);
1285         spin_unlock_bh(&pn->l2tp_tunnel_list_lock);
1286         atomic_dec(&l2tp_tunnel_count);
1287
1288         l2tp_tunnel_closeall(tunnel);
1289         l2tp_tunnel_dec_refcount(tunnel);
1290
1291         /* Call the original destructor */
1292         if (sk->sk_destruct)
1293                 (*sk->sk_destruct)(sk);
1294 end:
1295         return;
1296 }
1297
1298 /* When the tunnel is closed, all the attached sessions need to go too.
1299  */
1300 void l2tp_tunnel_closeall(struct l2tp_tunnel *tunnel)
1301 {
1302         int hash;
1303         struct hlist_node *walk;
1304         struct hlist_node *tmp;
1305         struct l2tp_session *session;
1306
1307         BUG_ON(tunnel == NULL);
1308
1309         l2tp_info(tunnel, L2TP_MSG_CONTROL, "%s: closing all sessions...\n",
1310                   tunnel->name);
1311
1312         write_lock_bh(&tunnel->hlist_lock);
1313         for (hash = 0; hash < L2TP_HASH_SIZE; hash++) {
1314 again:
1315                 hlist_for_each_safe(walk, tmp, &tunnel->session_hlist[hash]) {
1316                         session = hlist_entry(walk, struct l2tp_session, hlist);
1317
1318                         l2tp_info(session, L2TP_MSG_CONTROL,
1319                                   "%s: closing session\n", session->name);
1320
1321                         hlist_del_init(&session->hlist);
1322
1323                         if (session->ref != NULL)
1324                                 (*session->ref)(session);
1325
1326                         write_unlock_bh(&tunnel->hlist_lock);
1327
1328                         __l2tp_session_unhash(session);
1329                         l2tp_session_queue_purge(session);
1330
1331                         if (session->session_close != NULL)
1332                                 (*session->session_close)(session);
1333
1334                         if (session->deref != NULL)
1335                                 (*session->deref)(session);
1336
1337                         l2tp_session_dec_refcount(session);
1338
1339                         write_lock_bh(&tunnel->hlist_lock);
1340
1341                         /* Now restart from the beginning of this hash
1342                          * chain.  We always remove a session from the
1343                          * list so we are guaranteed to make forward
1344                          * progress.
1345                          */
1346                         goto again;
1347                 }
1348         }
1349         write_unlock_bh(&tunnel->hlist_lock);
1350 }
1351 EXPORT_SYMBOL_GPL(l2tp_tunnel_closeall);
1352
1353 /* Tunnel socket destroy hook for UDP encapsulation */
1354 static void l2tp_udp_encap_destroy(struct sock *sk)
1355 {
1356         struct l2tp_tunnel *tunnel = l2tp_sock_to_tunnel(sk);
1357         if (tunnel) {
1358                 l2tp_tunnel_closeall(tunnel);
1359                 sock_put(sk);
1360         }
1361 }
1362
1363 /* Really kill the tunnel.
1364  * Come here only when all sessions have been cleared from the tunnel.
1365  */
1366 static void l2tp_tunnel_free(struct l2tp_tunnel *tunnel)
1367 {
1368         BUG_ON(atomic_read(&tunnel->ref_count) != 0);
1369         BUG_ON(tunnel->sock != NULL);
1370         l2tp_info(tunnel, L2TP_MSG_CONTROL, "%s: free...\n", tunnel->name);
1371         kfree_rcu(tunnel, rcu);
1372 }
1373
1374 /* Workqueue tunnel deletion function */
1375 static void l2tp_tunnel_del_work(struct work_struct *work)
1376 {
1377         struct l2tp_tunnel *tunnel = NULL;
1378         struct socket *sock = NULL;
1379         struct sock *sk = NULL;
1380
1381         tunnel = container_of(work, struct l2tp_tunnel, del_work);
1382         sk = l2tp_tunnel_sock_lookup(tunnel);
1383         if (!sk)
1384                 return;
1385
1386         sock = sk->sk_socket;
1387
1388         /* If the tunnel socket was created by userspace, then go through the
1389          * inet layer to shut the socket down, and let userspace close it.
1390          * Otherwise, if we created the socket directly within the kernel, use
1391          * the sk API to release it here.
1392          * In either case the tunnel resources are freed in the socket
1393          * destructor when the tunnel socket goes away.
1394          */
1395         if (tunnel->fd >= 0) {
1396                 if (sock)
1397                         inet_shutdown(sock, 2);
1398         } else {
1399                 if (sock)
1400                         kernel_sock_shutdown(sock, SHUT_RDWR);
1401                 sk_release_kernel(sk);
1402         }
1403
1404         l2tp_tunnel_sock_put(sk);
1405 }
1406
1407 /* Create a socket for the tunnel, if one isn't set up by
1408  * userspace. This is used for static tunnels where there is no
1409  * managing L2TP daemon.
1410  *
1411  * Since we don't want these sockets to keep a namespace alive by
1412  * themselves, we drop the socket's namespace refcount after creation.
1413  * These sockets are freed when the namespace exits using the pernet
1414  * exit hook.
1415  */
1416 static int l2tp_tunnel_sock_create(struct net *net,
1417                                 u32 tunnel_id,
1418                                 u32 peer_tunnel_id,
1419                                 struct l2tp_tunnel_cfg *cfg,
1420                                 struct socket **sockp)
1421 {
1422         int err = -EINVAL;
1423         struct socket *sock = NULL;
1424         struct sockaddr_in udp_addr = {0};
1425         struct sockaddr_l2tpip ip_addr = {0};
1426 #if IS_ENABLED(CONFIG_IPV6)
1427         struct sockaddr_in6 udp6_addr = {0};
1428         struct sockaddr_l2tpip6 ip6_addr = {0};
1429 #endif
1430
1431         switch (cfg->encap) {
1432         case L2TP_ENCAPTYPE_UDP:
1433 #if IS_ENABLED(CONFIG_IPV6)
1434                 if (cfg->local_ip6 && cfg->peer_ip6) {
1435                         err = sock_create_kern(AF_INET6, SOCK_DGRAM, 0, &sock);
1436                         if (err < 0)
1437                                 goto out;
1438
1439                         sk_change_net(sock->sk, net);
1440
1441                         udp6_addr.sin6_family = AF_INET6;
1442                         memcpy(&udp6_addr.sin6_addr, cfg->local_ip6,
1443                                sizeof(udp6_addr.sin6_addr));
1444                         udp6_addr.sin6_port = htons(cfg->local_udp_port);
1445                         err = kernel_bind(sock, (struct sockaddr *) &udp6_addr,
1446                                           sizeof(udp6_addr));
1447                         if (err < 0)
1448                                 goto out;
1449
1450                         udp6_addr.sin6_family = AF_INET6;
1451                         memcpy(&udp6_addr.sin6_addr, cfg->peer_ip6,
1452                                sizeof(udp6_addr.sin6_addr));
1453                         udp6_addr.sin6_port = htons(cfg->peer_udp_port);
1454                         err = kernel_connect(sock,
1455                                              (struct sockaddr *) &udp6_addr,
1456                                              sizeof(udp6_addr), 0);
1457                         if (err < 0)
1458                                 goto out;
1459                 } else
1460 #endif
1461                 {
1462                         err = sock_create_kern(AF_INET, SOCK_DGRAM, 0, &sock);
1463                         if (err < 0)
1464                                 goto out;
1465
1466                         sk_change_net(sock->sk, net);
1467
1468                         udp_addr.sin_family = AF_INET;
1469                         udp_addr.sin_addr = cfg->local_ip;
1470                         udp_addr.sin_port = htons(cfg->local_udp_port);
1471                         err = kernel_bind(sock, (struct sockaddr *) &udp_addr,
1472                                           sizeof(udp_addr));
1473                         if (err < 0)
1474                                 goto out;
1475
1476                         udp_addr.sin_family = AF_INET;
1477                         udp_addr.sin_addr = cfg->peer_ip;
1478                         udp_addr.sin_port = htons(cfg->peer_udp_port);
1479                         err = kernel_connect(sock,
1480                                              (struct sockaddr *) &udp_addr,
1481                                              sizeof(udp_addr), 0);
1482                         if (err < 0)
1483                                 goto out;
1484                 }
1485
1486                 if (!cfg->use_udp_checksums)
1487                         sock->sk->sk_no_check = UDP_CSUM_NOXMIT;
1488
1489                 break;
1490
1491         case L2TP_ENCAPTYPE_IP:
1492 #if IS_ENABLED(CONFIG_IPV6)
1493                 if (cfg->local_ip6 && cfg->peer_ip6) {
1494                         err = sock_create_kern(AF_INET6, SOCK_DGRAM,
1495                                           IPPROTO_L2TP, &sock);
1496                         if (err < 0)
1497                                 goto out;
1498
1499                         sk_change_net(sock->sk, net);
1500
1501                         ip6_addr.l2tp_family = AF_INET6;
1502                         memcpy(&ip6_addr.l2tp_addr, cfg->local_ip6,
1503                                sizeof(ip6_addr.l2tp_addr));
1504                         ip6_addr.l2tp_conn_id = tunnel_id;
1505                         err = kernel_bind(sock, (struct sockaddr *) &ip6_addr,
1506                                           sizeof(ip6_addr));
1507                         if (err < 0)
1508                                 goto out;
1509
1510                         ip6_addr.l2tp_family = AF_INET6;
1511                         memcpy(&ip6_addr.l2tp_addr, cfg->peer_ip6,
1512                                sizeof(ip6_addr.l2tp_addr));
1513                         ip6_addr.l2tp_conn_id = peer_tunnel_id;
1514                         err = kernel_connect(sock,
1515                                              (struct sockaddr *) &ip6_addr,
1516                                              sizeof(ip6_addr), 0);
1517                         if (err < 0)
1518                                 goto out;
1519                 } else
1520 #endif
1521                 {
1522                         err = sock_create_kern(AF_INET, SOCK_DGRAM,
1523                                           IPPROTO_L2TP, &sock);
1524                         if (err < 0)
1525                                 goto out;
1526
1527                         sk_change_net(sock->sk, net);
1528
1529                         ip_addr.l2tp_family = AF_INET;
1530                         ip_addr.l2tp_addr = cfg->local_ip;
1531                         ip_addr.l2tp_conn_id = tunnel_id;
1532                         err = kernel_bind(sock, (struct sockaddr *) &ip_addr,
1533                                           sizeof(ip_addr));
1534                         if (err < 0)
1535                                 goto out;
1536
1537                         ip_addr.l2tp_family = AF_INET;
1538                         ip_addr.l2tp_addr = cfg->peer_ip;
1539                         ip_addr.l2tp_conn_id = peer_tunnel_id;
1540                         err = kernel_connect(sock, (struct sockaddr *) &ip_addr,
1541                                              sizeof(ip_addr), 0);
1542                         if (err < 0)
1543                                 goto out;
1544                 }
1545                 break;
1546
1547         default:
1548                 goto out;
1549         }
1550
1551 out:
1552         *sockp = sock;
1553         if ((err < 0) && sock) {
1554                 kernel_sock_shutdown(sock, SHUT_RDWR);
1555                 sk_release_kernel(sock->sk);
1556                 *sockp = NULL;
1557         }
1558
1559         return err;
1560 }
1561
1562 static struct lock_class_key l2tp_socket_class;
1563
1564 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)
1565 {
1566         struct l2tp_tunnel *tunnel = NULL;
1567         int err;
1568         struct socket *sock = NULL;
1569         struct sock *sk = NULL;
1570         struct l2tp_net *pn;
1571         enum l2tp_encap_type encap = L2TP_ENCAPTYPE_UDP;
1572
1573         /* Get the tunnel socket from the fd, which was opened by
1574          * the userspace L2TP daemon. If not specified, create a
1575          * kernel socket.
1576          */
1577         if (fd < 0) {
1578                 err = l2tp_tunnel_sock_create(net, tunnel_id, peer_tunnel_id,
1579                                 cfg, &sock);
1580                 if (err < 0)
1581                         goto err;
1582         } else {
1583                 sock = sockfd_lookup(fd, &err);
1584                 if (!sock) {
1585                         pr_err("tunl %u: sockfd_lookup(fd=%d) returned %d\n",
1586                                tunnel_id, fd, err);
1587                         err = -EBADF;
1588                         goto err;
1589                 }
1590
1591                 /* Reject namespace mismatches */
1592                 if (!net_eq(sock_net(sock->sk), net)) {
1593                         pr_err("tunl %u: netns mismatch\n", tunnel_id);
1594                         err = -EINVAL;
1595                         goto err;
1596                 }
1597         }
1598
1599         sk = sock->sk;
1600
1601         if (cfg != NULL)
1602                 encap = cfg->encap;
1603
1604         /* Quick sanity checks */
1605         switch (encap) {
1606         case L2TP_ENCAPTYPE_UDP:
1607                 err = -EPROTONOSUPPORT;
1608                 if (sk->sk_protocol != IPPROTO_UDP) {
1609                         pr_err("tunl %hu: fd %d wrong protocol, got %d, expected %d\n",
1610                                tunnel_id, fd, sk->sk_protocol, IPPROTO_UDP);
1611                         goto err;
1612                 }
1613                 break;
1614         case L2TP_ENCAPTYPE_IP:
1615                 err = -EPROTONOSUPPORT;
1616                 if (sk->sk_protocol != IPPROTO_L2TP) {
1617                         pr_err("tunl %hu: fd %d wrong protocol, got %d, expected %d\n",
1618                                tunnel_id, fd, sk->sk_protocol, IPPROTO_L2TP);
1619                         goto err;
1620                 }
1621                 break;
1622         }
1623
1624         /* Check if this socket has already been prepped */
1625         tunnel = l2tp_tunnel(sk);
1626         if (tunnel != NULL) {
1627                 /* This socket has already been prepped */
1628                 err = -EBUSY;
1629                 goto err;
1630         }
1631
1632         tunnel = kzalloc(sizeof(struct l2tp_tunnel), GFP_KERNEL);
1633         if (tunnel == NULL) {
1634                 err = -ENOMEM;
1635                 goto err;
1636         }
1637
1638         tunnel->version = version;
1639         tunnel->tunnel_id = tunnel_id;
1640         tunnel->peer_tunnel_id = peer_tunnel_id;
1641         tunnel->debug = L2TP_DEFAULT_DEBUG_FLAGS;
1642
1643         tunnel->magic = L2TP_TUNNEL_MAGIC;
1644         sprintf(&tunnel->name[0], "tunl %u", tunnel_id);
1645         rwlock_init(&tunnel->hlist_lock);
1646
1647         /* The net we belong to */
1648         tunnel->l2tp_net = net;
1649         pn = l2tp_pernet(net);
1650
1651         if (cfg != NULL)
1652                 tunnel->debug = cfg->debug;
1653
1654 #if IS_ENABLED(CONFIG_IPV6)
1655         if (sk->sk_family == PF_INET6) {
1656                 struct ipv6_pinfo *np = inet6_sk(sk);
1657
1658                 if (ipv6_addr_v4mapped(&np->saddr) &&
1659                     ipv6_addr_v4mapped(&np->daddr)) {
1660                         struct inet_sock *inet = inet_sk(sk);
1661
1662                         tunnel->v4mapped = true;
1663                         inet->inet_saddr = np->saddr.s6_addr32[3];
1664                         inet->inet_rcv_saddr = np->rcv_saddr.s6_addr32[3];
1665                         inet->inet_daddr = np->daddr.s6_addr32[3];
1666                 } else {
1667                         tunnel->v4mapped = false;
1668                 }
1669         }
1670 #endif
1671
1672         /* Mark socket as an encapsulation socket. See net/ipv4/udp.c */
1673         tunnel->encap = encap;
1674         if (encap == L2TP_ENCAPTYPE_UDP) {
1675                 /* Mark socket as an encapsulation socket. See net/ipv4/udp.c */
1676                 udp_sk(sk)->encap_type = UDP_ENCAP_L2TPINUDP;
1677                 udp_sk(sk)->encap_rcv = l2tp_udp_encap_recv;
1678                 udp_sk(sk)->encap_destroy = l2tp_udp_encap_destroy;
1679 #if IS_ENABLED(CONFIG_IPV6)
1680                 if (sk->sk_family == PF_INET6 && !tunnel->v4mapped)
1681                         udpv6_encap_enable();
1682                 else
1683 #endif
1684                 udp_encap_enable();
1685         }
1686
1687         sk->sk_user_data = tunnel;
1688
1689         /* Hook on the tunnel socket destructor so that we can cleanup
1690          * if the tunnel socket goes away.
1691          */
1692         tunnel->old_sk_destruct = sk->sk_destruct;
1693         sk->sk_destruct = &l2tp_tunnel_destruct;
1694         tunnel->sock = sk;
1695         tunnel->fd = fd;
1696         lockdep_set_class_and_name(&sk->sk_lock.slock, &l2tp_socket_class, "l2tp_sock");
1697
1698         sk->sk_allocation = GFP_ATOMIC;
1699
1700         /* Init delete workqueue struct */
1701         INIT_WORK(&tunnel->del_work, l2tp_tunnel_del_work);
1702
1703         /* Add tunnel to our list */
1704         INIT_LIST_HEAD(&tunnel->list);
1705         atomic_inc(&l2tp_tunnel_count);
1706
1707         /* Bump the reference count. The tunnel context is deleted
1708          * only when this drops to zero. Must be done before list insertion
1709          */
1710         l2tp_tunnel_inc_refcount(tunnel);
1711         spin_lock_bh(&pn->l2tp_tunnel_list_lock);
1712         list_add_rcu(&tunnel->list, &pn->l2tp_tunnel_list);
1713         spin_unlock_bh(&pn->l2tp_tunnel_list_lock);
1714
1715         err = 0;
1716 err:
1717         if (tunnelp)
1718                 *tunnelp = tunnel;
1719
1720         /* If tunnel's socket was created by the kernel, it doesn't
1721          *  have a file.
1722          */
1723         if (sock && sock->file)
1724                 sockfd_put(sock);
1725
1726         return err;
1727 }
1728 EXPORT_SYMBOL_GPL(l2tp_tunnel_create);
1729
1730 /* This function is used by the netlink TUNNEL_DELETE command.
1731  */
1732 int l2tp_tunnel_delete(struct l2tp_tunnel *tunnel)
1733 {
1734         l2tp_tunnel_closeall(tunnel);
1735         return (false == queue_work(l2tp_wq, &tunnel->del_work));
1736 }
1737 EXPORT_SYMBOL_GPL(l2tp_tunnel_delete);
1738
1739 /* Really kill the session.
1740  */
1741 void l2tp_session_free(struct l2tp_session *session)
1742 {
1743         struct l2tp_tunnel *tunnel = session->tunnel;
1744
1745         BUG_ON(atomic_read(&session->ref_count) != 0);
1746
1747         if (tunnel) {
1748                 BUG_ON(tunnel->magic != L2TP_TUNNEL_MAGIC);
1749                 if (session->session_id != 0)
1750                         atomic_dec(&l2tp_session_count);
1751                 sock_put(tunnel->sock);
1752                 session->tunnel = NULL;
1753                 l2tp_tunnel_dec_refcount(tunnel);
1754         }
1755
1756         kfree(session);
1757
1758         return;
1759 }
1760 EXPORT_SYMBOL_GPL(l2tp_session_free);
1761
1762 /* Remove an l2tp session from l2tp_core's hash lists.
1763  * Provides a tidyup interface for pseudowire code which can't just route all
1764  * shutdown via. l2tp_session_delete and a pseudowire-specific session_close
1765  * callback.
1766  */
1767 void __l2tp_session_unhash(struct l2tp_session *session)
1768 {
1769         struct l2tp_tunnel *tunnel = session->tunnel;
1770
1771         /* Remove the session from core hashes */
1772         if (tunnel) {
1773                 /* Remove from the per-tunnel hash */
1774                 write_lock_bh(&tunnel->hlist_lock);
1775                 hlist_del_init(&session->hlist);
1776                 write_unlock_bh(&tunnel->hlist_lock);
1777
1778                 /* For L2TPv3 we have a per-net hash: remove from there, too */
1779                 if (tunnel->version != L2TP_HDR_VER_2) {
1780                         struct l2tp_net *pn = l2tp_pernet(tunnel->l2tp_net);
1781                         spin_lock_bh(&pn->l2tp_session_hlist_lock);
1782                         hlist_del_init_rcu(&session->global_hlist);
1783                         spin_unlock_bh(&pn->l2tp_session_hlist_lock);
1784                         synchronize_rcu();
1785                 }
1786         }
1787 }
1788 EXPORT_SYMBOL_GPL(__l2tp_session_unhash);
1789
1790 /* This function is used by the netlink SESSION_DELETE command and by
1791    pseudowire modules.
1792  */
1793 int l2tp_session_delete(struct l2tp_session *session)
1794 {
1795         if (session->ref)
1796                 (*session->ref)(session);
1797         __l2tp_session_unhash(session);
1798         l2tp_session_queue_purge(session);
1799         if (session->session_close != NULL)
1800                 (*session->session_close)(session);
1801         if (session->deref)
1802                 (*session->deref)(session);
1803         l2tp_session_dec_refcount(session);
1804         return 0;
1805 }
1806 EXPORT_SYMBOL_GPL(l2tp_session_delete);
1807
1808 /* We come here whenever a session's send_seq, cookie_len or
1809  * l2specific_len parameters are set.
1810  */
1811 static void l2tp_session_set_header_len(struct l2tp_session *session, int version)
1812 {
1813         if (version == L2TP_HDR_VER_2) {
1814                 session->hdr_len = 6;
1815                 if (session->send_seq)
1816                         session->hdr_len += 4;
1817         } else {
1818                 session->hdr_len = 4 + session->cookie_len + session->l2specific_len + session->offset;
1819                 if (session->tunnel->encap == L2TP_ENCAPTYPE_UDP)
1820                         session->hdr_len += 4;
1821         }
1822
1823 }
1824
1825 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)
1826 {
1827         struct l2tp_session *session;
1828
1829         session = kzalloc(sizeof(struct l2tp_session) + priv_size, GFP_KERNEL);
1830         if (session != NULL) {
1831                 session->magic = L2TP_SESSION_MAGIC;
1832                 session->tunnel = tunnel;
1833
1834                 session->session_id = session_id;
1835                 session->peer_session_id = peer_session_id;
1836                 session->nr = 0;
1837
1838                 sprintf(&session->name[0], "sess %u/%u",
1839                         tunnel->tunnel_id, session->session_id);
1840
1841                 skb_queue_head_init(&session->reorder_q);
1842
1843                 INIT_HLIST_NODE(&session->hlist);
1844                 INIT_HLIST_NODE(&session->global_hlist);
1845
1846                 /* Inherit debug options from tunnel */
1847                 session->debug = tunnel->debug;
1848
1849                 if (cfg) {
1850                         session->pwtype = cfg->pw_type;
1851                         session->debug = cfg->debug;
1852                         session->mtu = cfg->mtu;
1853                         session->mru = cfg->mru;
1854                         session->send_seq = cfg->send_seq;
1855                         session->recv_seq = cfg->recv_seq;
1856                         session->lns_mode = cfg->lns_mode;
1857                         session->reorder_timeout = cfg->reorder_timeout;
1858                         session->offset = cfg->offset;
1859                         session->l2specific_type = cfg->l2specific_type;
1860                         session->l2specific_len = cfg->l2specific_len;
1861                         session->cookie_len = cfg->cookie_len;
1862                         memcpy(&session->cookie[0], &cfg->cookie[0], cfg->cookie_len);
1863                         session->peer_cookie_len = cfg->peer_cookie_len;
1864                         memcpy(&session->peer_cookie[0], &cfg->peer_cookie[0], cfg->peer_cookie_len);
1865                 }
1866
1867                 if (tunnel->version == L2TP_HDR_VER_2)
1868                         session->build_header = l2tp_build_l2tpv2_header;
1869                 else
1870                         session->build_header = l2tp_build_l2tpv3_header;
1871
1872                 l2tp_session_set_header_len(session, tunnel->version);
1873
1874                 /* Bump the reference count. The session context is deleted
1875                  * only when this drops to zero.
1876                  */
1877                 l2tp_session_inc_refcount(session);
1878                 l2tp_tunnel_inc_refcount(tunnel);
1879
1880                 /* Ensure tunnel socket isn't deleted */
1881                 sock_hold(tunnel->sock);
1882
1883                 /* Add session to the tunnel's hash list */
1884                 write_lock_bh(&tunnel->hlist_lock);
1885                 hlist_add_head(&session->hlist,
1886                                l2tp_session_id_hash(tunnel, session_id));
1887                 write_unlock_bh(&tunnel->hlist_lock);
1888
1889                 /* And to the global session list if L2TPv3 */
1890                 if (tunnel->version != L2TP_HDR_VER_2) {
1891                         struct l2tp_net *pn = l2tp_pernet(tunnel->l2tp_net);
1892
1893                         spin_lock_bh(&pn->l2tp_session_hlist_lock);
1894                         hlist_add_head_rcu(&session->global_hlist,
1895                                            l2tp_session_id_hash_2(pn, session_id));
1896                         spin_unlock_bh(&pn->l2tp_session_hlist_lock);
1897                 }
1898
1899                 /* Ignore management session in session count value */
1900                 if (session->session_id != 0)
1901                         atomic_inc(&l2tp_session_count);
1902         }
1903
1904         return session;
1905 }
1906 EXPORT_SYMBOL_GPL(l2tp_session_create);
1907
1908 /*****************************************************************************
1909  * Init and cleanup
1910  *****************************************************************************/
1911
1912 static __net_init int l2tp_init_net(struct net *net)
1913 {
1914         struct l2tp_net *pn = net_generic(net, l2tp_net_id);
1915         int hash;
1916
1917         INIT_LIST_HEAD(&pn->l2tp_tunnel_list);
1918         spin_lock_init(&pn->l2tp_tunnel_list_lock);
1919
1920         for (hash = 0; hash < L2TP_HASH_SIZE_2; hash++)
1921                 INIT_HLIST_HEAD(&pn->l2tp_session_hlist[hash]);
1922
1923         spin_lock_init(&pn->l2tp_session_hlist_lock);
1924
1925         return 0;
1926 }
1927
1928 static __net_exit void l2tp_exit_net(struct net *net)
1929 {
1930         struct l2tp_net *pn = l2tp_pernet(net);
1931         struct l2tp_tunnel *tunnel = NULL;
1932
1933         rcu_read_lock_bh();
1934         list_for_each_entry_rcu(tunnel, &pn->l2tp_tunnel_list, list) {
1935                 (void)l2tp_tunnel_delete(tunnel);
1936         }
1937         rcu_read_unlock_bh();
1938 }
1939
1940 static struct pernet_operations l2tp_net_ops = {
1941         .init = l2tp_init_net,
1942         .exit = l2tp_exit_net,
1943         .id   = &l2tp_net_id,
1944         .size = sizeof(struct l2tp_net),
1945 };
1946
1947 static int __init l2tp_init(void)
1948 {
1949         int rc = 0;
1950
1951         rc = register_pernet_device(&l2tp_net_ops);
1952         if (rc)
1953                 goto out;
1954
1955         l2tp_wq = alloc_workqueue("l2tp", WQ_NON_REENTRANT | WQ_UNBOUND, 0);
1956         if (!l2tp_wq) {
1957                 pr_err("alloc_workqueue failed\n");
1958                 rc = -ENOMEM;
1959                 goto out;
1960         }
1961
1962         pr_info("L2TP core driver, %s\n", L2TP_DRV_VERSION);
1963
1964 out:
1965         return rc;
1966 }
1967
1968 static void __exit l2tp_exit(void)
1969 {
1970         unregister_pernet_device(&l2tp_net_ops);
1971         if (l2tp_wq) {
1972                 destroy_workqueue(l2tp_wq);
1973                 l2tp_wq = NULL;
1974         }
1975 }
1976
1977 module_init(l2tp_init);
1978 module_exit(l2tp_exit);
1979
1980 MODULE_AUTHOR("James Chapman <jchapman@katalix.com>");
1981 MODULE_DESCRIPTION("L2TP core");
1982 MODULE_LICENSE("GPL");
1983 MODULE_VERSION(L2TP_DRV_VERSION);
1984