net/mptcp/pm_netlink.c

   1 // SPDX-License-Identifier: GPL-2.0
   2 /* Multipath TCP
   3  *
   4  * Copyright (c) 2020, Red Hat, Inc.
   5  */
   6
   7 #define pr_fmt(fmt) "MPTCP: " fmt
   8
   9 #include <linux/inet.h>
  10 #include <linux/kernel.h>
  11 #include <net/tcp.h>
  12 #include <net/netns/generic.h>
  13 #include <net/mptcp.h>
  14 #include <net/genetlink.h>
  15 #include <uapi/linux/mptcp.h>
  16
  17 #include "protocol.h"
  18 #include "mib.h"
  19
  20 /* forward declaration */
  21 static struct genl_family mptcp_genl_family;
  22
  23 static int pm_nl_pernet_id;
  24
  25 struct mptcp_pm_addr_entry {
  26         struct list_head        list;
  27         struct mptcp_addr_info  addr;
  28         struct rcu_head         rcu;
  29 };
  30
  31 struct mptcp_pm_add_entry {
  32         struct list_head        list;
  33         struct mptcp_addr_info  addr;
  34         struct timer_list       add_timer;
  35         struct mptcp_sock       *sock;
  36         u8                      retrans_times;
  37 };
  38
  39 struct pm_nl_pernet {
  40         /* protects pernet updates */
  41         spinlock_t              lock;
  42         struct list_head        local_addr_list;
  43         unsigned int            addrs;
  44         unsigned int            add_addr_signal_max;
  45         unsigned int            add_addr_accept_max;
  46         unsigned int            local_addr_max;
  47         unsigned int            subflows_max;
  48         unsigned int            next_id;
  49 };
  50
  51 #define MPTCP_PM_ADDR_MAX       8
  52 #define ADD_ADDR_RETRANS_MAX    3
  53
  54 static bool addresses_equal(const struct mptcp_addr_info *a,
  55                             struct mptcp_addr_info *b, bool use_port)
  56 {
  57         bool addr_equals = false;
  58
  59         if (a->family != b->family)
  60                 return false;
  61
  62         if (a->family == AF_INET)
  63                 addr_equals = a->addr.s_addr == b->addr.s_addr;
  64 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
  65         else
  66                 addr_equals = !ipv6_addr_cmp(&a->addr6, &b->addr6);
  67 #endif
  68
  69         if (!addr_equals)
  70                 return false;
  71         if (!use_port)
  72                 return true;
  73
  74         return a->port == b->port;
  75 }
  76
  77 static bool address_zero(const struct mptcp_addr_info *addr)
  78 {
  79         struct mptcp_addr_info zero;
  80
  81         memset(&zero, 0, sizeof(zero));
  82         zero.family = addr->family;
  83
  84         return addresses_equal(addr, &zero, false);
  85 }
  86
  87 static void local_address(const struct sock_common *skc,
  88                           struct mptcp_addr_info *addr)
  89 {
  90         addr->port = 0;
  91         addr->family = skc->skc_family;
  92         if (addr->family == AF_INET)
  93                 addr->addr.s_addr = skc->skc_rcv_saddr;
  94 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
  95         else if (addr->family == AF_INET6)
  96                 addr->addr6 = skc->skc_v6_rcv_saddr;
  97 #endif
  98 }
  99
 100 static void remote_address(const struct sock_common *skc,
 101                            struct mptcp_addr_info *addr)
 102 {
 103         addr->family = skc->skc_family;
 104         addr->port = skc->skc_dport;
 105         if (addr->family == AF_INET)
 106                 addr->addr.s_addr = skc->skc_daddr;
 107 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
 108         else if (addr->family == AF_INET6)
 109                 addr->addr6 = skc->skc_v6_daddr;
 110 #endif
 111 }
 112
 113 static bool lookup_subflow_by_saddr(const struct list_head *list,
 114                                     struct mptcp_addr_info *saddr)
 115 {
 116         struct mptcp_subflow_context *subflow;
 117         struct mptcp_addr_info cur;
 118         struct sock_common *skc;
 119
 120         list_for_each_entry(subflow, list, node) {
 121                 skc = (struct sock_common *)mptcp_subflow_tcp_sock(subflow);
 122
 123                 local_address(skc, &cur);
 124                 if (addresses_equal(&cur, saddr, false))
 125                         return true;
 126         }
 127
 128         return false;
 129 }
 130
 131 static struct mptcp_pm_addr_entry *
 132 select_local_address(const struct pm_nl_pernet *pernet,
 133                      struct mptcp_sock *msk)
 134 {
 135         struct mptcp_pm_addr_entry *entry, *ret = NULL;
 136
 137         rcu_read_lock();
 138         spin_lock_bh(&msk->join_list_lock);
 139         list_for_each_entry_rcu(entry, &pernet->local_addr_list, list) {
 140                 if (!(entry->addr.flags & MPTCP_PM_ADDR_FLAG_SUBFLOW))
 141                         continue;
 142
 143                 /* avoid any address already in use by subflows and
 144                  * pending join
 145                  */
 146                 if (entry->addr.family == ((struct sock *)msk)->sk_family &&
 147                     !lookup_subflow_by_saddr(&msk->conn_list, &entry->addr) &&
 148                     !lookup_subflow_by_saddr(&msk->join_list, &entry->addr)) {
 149                         ret = entry;
 150                         break;
 151                 }
 152         }
 153         spin_unlock_bh(&msk->join_list_lock);
 154         rcu_read_unlock();
 155         return ret;
 156 }
 157
 158 static struct mptcp_pm_addr_entry *
 159 select_signal_address(struct pm_nl_pernet *pernet, unsigned int pos)
 160 {
 161         struct mptcp_pm_addr_entry *entry, *ret = NULL;
 162         int i = 0;
 163
 164         rcu_read_lock();
 165         /* do not keep any additional per socket state, just signal
 166          * the address list in order.
 167          * Note: removal from the local address list during the msk life-cycle
 168          * can lead to additional addresses not being announced.
 169          */
 170         list_for_each_entry_rcu(entry, &pernet->local_addr_list, list) {
 171                 if (!(entry->addr.flags & MPTCP_PM_ADDR_FLAG_SIGNAL))
 172                         continue;
 173                 if (i++ == pos) {
 174                         ret = entry;
 175                         break;
 176                 }
 177         }
 178         rcu_read_unlock();
 179         return ret;
 180 }
 181
 182 static void check_work_pending(struct mptcp_sock *msk)
 183 {
 184         if (msk->pm.add_addr_signaled == msk->pm.add_addr_signal_max &&
 185             (msk->pm.local_addr_used == msk->pm.local_addr_max ||
 186              msk->pm.subflows == msk->pm.subflows_max))
 187                 WRITE_ONCE(msk->pm.work_pending, false);
 188 }
 189
 190 static struct mptcp_pm_add_entry *
 191 lookup_anno_list_by_saddr(struct mptcp_sock *msk,
 192                           struct mptcp_addr_info *addr)
 193 {
 194         struct mptcp_pm_add_entry *entry;
 195
 196         list_for_each_entry(entry, &msk->pm.anno_list, list) {
 197                 if (addresses_equal(&entry->addr, addr, false))
 198                         return entry;
 199         }
 200
 201         return NULL;
 202 }
 203
 204 static void mptcp_pm_add_timer(struct timer_list *timer)
 205 {
 206         struct mptcp_pm_add_entry *entry = from_timer(entry, timer, add_timer);
 207         struct mptcp_sock *msk = entry->sock;
 208         struct sock *sk = (struct sock *)msk;
 209
 210         pr_debug("msk=%p", msk);
 211
 212         if (!msk)
 213                 return;
 214
 215         if (inet_sk_state_load(sk) == TCP_CLOSE)
 216                 return;
 217
 218         if (!entry->addr.id)
 219                 return;
 220
 221         if (mptcp_pm_should_add_signal(msk)) {
 222                 sk_reset_timer(sk, timer, jiffies + TCP_RTO_MAX / 8);
 223                 goto out;
 224         }
 225
 226         spin_lock_bh(&msk->pm.lock);
 227
 228         if (!mptcp_pm_should_add_signal(msk)) {
 229                 pr_debug("retransmit ADD_ADDR id=%d", entry->addr.id);
 230                 mptcp_pm_announce_addr(msk, &entry->addr, false);
 231                 mptcp_pm_add_addr_send_ack(msk);
 232                 entry->retrans_times++;
 233         }
 234
 235         if (entry->retrans_times < ADD_ADDR_RETRANS_MAX)
 236                 sk_reset_timer(sk, timer,
 237                                jiffies + mptcp_get_add_addr_timeout(sock_net(sk)));
 238
 239         spin_unlock_bh(&msk->pm.lock);
 240
 241 out:
 242         __sock_put(sk);
 243 }
 244
 245 struct mptcp_pm_add_entry *
 246 mptcp_pm_del_add_timer(struct mptcp_sock *msk,
 247                        struct mptcp_addr_info *addr)
 248 {
 249         struct mptcp_pm_add_entry *entry;
 250         struct sock *sk = (struct sock *)msk;
 251
 252         spin_lock_bh(&msk->pm.lock);
 253         entry = lookup_anno_list_by_saddr(msk, addr);
 254         if (entry)
 255                 entry->retrans_times = ADD_ADDR_RETRANS_MAX;
 256         spin_unlock_bh(&msk->pm.lock);
 257
 258         if (entry)
 259                 sk_stop_timer_sync(sk, &entry->add_timer);
 260
 261         return entry;
 262 }
 263
 264 static bool mptcp_pm_alloc_anno_list(struct mptcp_sock *msk,
 265                                      struct mptcp_pm_addr_entry *entry)
 266 {
 267         struct mptcp_pm_add_entry *add_entry = NULL;
 268         struct sock *sk = (struct sock *)msk;
 269         struct net *net = sock_net(sk);
 270
 271         if (lookup_anno_list_by_saddr(msk, &entry->addr))
 272                 return false;
 273
 274         add_entry = kmalloc(sizeof(*add_entry), GFP_ATOMIC);
 275         if (!add_entry)
 276                 return false;
 277
 278         list_add(&add_entry->list, &msk->pm.anno_list);
 279
 280         add_entry->addr = entry->addr;
 281         add_entry->sock = msk;
 282         add_entry->retrans_times = 0;
 283
 284         timer_setup(&add_entry->add_timer, mptcp_pm_add_timer, 0);
 285         sk_reset_timer(sk, &add_entry->add_timer,
 286                        jiffies + mptcp_get_add_addr_timeout(net));
 287
 288         return true;
 289 }
 290
 291 void mptcp_pm_free_anno_list(struct mptcp_sock *msk)
 292 {
 293         struct mptcp_pm_add_entry *entry, *tmp;
 294         struct sock *sk = (struct sock *)msk;
 295         LIST_HEAD(free_list);
 296
 297         pr_debug("msk=%p", msk);
 298
 299         spin_lock_bh(&msk->pm.lock);
 300         list_splice_init(&msk->pm.anno_list, &free_list);
 301         spin_unlock_bh(&msk->pm.lock);
 302
 303         list_for_each_entry_safe(entry, tmp, &free_list, list) {
 304                 sk_stop_timer_sync(sk, &entry->add_timer);
 305                 kfree(entry);
 306         }
 307 }
 308
 309 static void mptcp_pm_create_subflow_or_signal_addr(struct mptcp_sock *msk)
 310 {
 311         struct mptcp_addr_info remote = { 0 };
 312         struct sock *sk = (struct sock *)msk;
 313         struct mptcp_pm_addr_entry *local;
 314         struct pm_nl_pernet *pernet;
 315
 316         pernet = net_generic(sock_net((struct sock *)msk), pm_nl_pernet_id);
 317
 318         pr_debug("local %d:%d signal %d:%d subflows %d:%d\n",
 319                  msk->pm.local_addr_used, msk->pm.local_addr_max,
 320                  msk->pm.add_addr_signaled, msk->pm.add_addr_signal_max,
 321                  msk->pm.subflows, msk->pm.subflows_max);
 322
 323         /* check first for announce */
 324         if (msk->pm.add_addr_signaled < msk->pm.add_addr_signal_max) {
 325                 local = select_signal_address(pernet,
 326                                               msk->pm.add_addr_signaled);
 327
 328                 if (local) {
 329                         if (mptcp_pm_alloc_anno_list(msk, local)) {
 330                                 msk->pm.add_addr_signaled++;
 331                                 mptcp_pm_announce_addr(msk, &local->addr, false);
 332                                 mptcp_pm_nl_add_addr_send_ack(msk);
 333                         }
 334                 } else {
 335                         /* pick failed, avoid fourther attempts later */
 336                         msk->pm.local_addr_used = msk->pm.add_addr_signal_max;
 337                 }
 338
 339                 check_work_pending(msk);
 340         }
 341
 342         /* check if should create a new subflow */
 343         if (msk->pm.local_addr_used < msk->pm.local_addr_max &&
 344             msk->pm.subflows < msk->pm.subflows_max) {
 345                 remote_address((struct sock_common *)sk, &remote);
 346
 347                 local = select_local_address(pernet, msk);
 348                 if (local) {
 349                         msk->pm.local_addr_used++;
 350                         msk->pm.subflows++;
 351                         check_work_pending(msk);
 352                         spin_unlock_bh(&msk->pm.lock);
 353                         __mptcp_subflow_connect(sk, &local->addr, &remote);
 354                         spin_lock_bh(&msk->pm.lock);
 355                         return;
 356                 }
 357
 358                 /* lookup failed, avoid fourther attempts later */
 359                 msk->pm.local_addr_used = msk->pm.local_addr_max;
 360                 check_work_pending(msk);
 361         }
 362 }
 363
 364 void mptcp_pm_nl_fully_established(struct mptcp_sock *msk)
 365 {
 366         mptcp_pm_create_subflow_or_signal_addr(msk);
 367 }
 368
 369 void mptcp_pm_nl_subflow_established(struct mptcp_sock *msk)
 370 {
 371         mptcp_pm_create_subflow_or_signal_addr(msk);
 372 }
 373
 374 void mptcp_pm_nl_add_addr_received(struct mptcp_sock *msk)
 375 {
 376         struct sock *sk = (struct sock *)msk;
 377         struct mptcp_addr_info remote;
 378         struct mptcp_addr_info local;
 379
 380         pr_debug("accepted %d:%d remote family %d",
 381                  msk->pm.add_addr_accepted, msk->pm.add_addr_accept_max,
 382                  msk->pm.remote.family);
 383         msk->pm.add_addr_accepted++;
 384         msk->pm.subflows++;
 385         if (msk->pm.add_addr_accepted >= msk->pm.add_addr_accept_max ||
 386             msk->pm.subflows >= msk->pm.subflows_max)
 387                 WRITE_ONCE(msk->pm.accept_addr, false);
 388
 389         /* connect to the specified remote address, using whatever
 390          * local address the routing configuration will pick.
 391          */
 392         remote = msk->pm.remote;
 393         if (!remote.port)
 394                 remote.port = sk->sk_dport;
 395         memset(&local, 0, sizeof(local));
 396         local.family = remote.family;
 397
 398         spin_unlock_bh(&msk->pm.lock);
 399         __mptcp_subflow_connect((struct sock *)msk, &local, &remote);
 400         spin_lock_bh(&msk->pm.lock);
 401
 402         mptcp_pm_announce_addr(msk, &remote, true);
 403         mptcp_pm_nl_add_addr_send_ack(msk);
 404 }
 405
 406 void mptcp_pm_nl_add_addr_send_ack(struct mptcp_sock *msk)
 407 {
 408         struct mptcp_subflow_context *subflow;
 409
 410         if (!mptcp_pm_should_add_signal_ipv6(msk) &&
 411             !mptcp_pm_should_add_signal_port(msk))
 412                 return;
 413
 414         __mptcp_flush_join_list(msk);
 415         subflow = list_first_entry_or_null(&msk->conn_list, typeof(*subflow), node);
 416         if (subflow) {
 417                 struct sock *ssk = mptcp_subflow_tcp_sock(subflow);
 418                 u8 add_addr;
 419
 420                 spin_unlock_bh(&msk->pm.lock);
 421                 if (mptcp_pm_should_add_signal_ipv6(msk))
 422                         pr_debug("send ack for add_addr6");
 423                 if (mptcp_pm_should_add_signal_port(msk))
 424                         pr_debug("send ack for add_addr_port");
 425
 426                 lock_sock(ssk);
 427                 tcp_send_ack(ssk);
 428                 release_sock(ssk);
 429                 spin_lock_bh(&msk->pm.lock);
 430
 431                 add_addr = READ_ONCE(msk->pm.add_addr_signal);
 432                 if (mptcp_pm_should_add_signal_ipv6(msk))
 433                         add_addr &= ~BIT(MPTCP_ADD_ADDR_IPV6);
 434                 if (mptcp_pm_should_add_signal_port(msk))
 435                         add_addr &= ~BIT(MPTCP_ADD_ADDR_PORT);
 436                 WRITE_ONCE(msk->pm.add_addr_signal, add_addr);
 437         }
 438 }
 439
 440 void mptcp_pm_nl_rm_addr_received(struct mptcp_sock *msk)
 441 {
 442         struct mptcp_subflow_context *subflow, *tmp;
 443         struct sock *sk = (struct sock *)msk;
 444
 445         pr_debug("address rm_id %d", msk->pm.rm_id);
 446
 447         if (!msk->pm.rm_id)
 448                 return;
 449
 450         if (list_empty(&msk->conn_list))
 451                 return;
 452
 453         list_for_each_entry_safe(subflow, tmp, &msk->conn_list, node) {
 454                 struct sock *ssk = mptcp_subflow_tcp_sock(subflow);
 455                 int how = RCV_SHUTDOWN | SEND_SHUTDOWN;
 456
 457                 if (msk->pm.rm_id != subflow->remote_id)
 458                         continue;
 459
 460                 spin_unlock_bh(&msk->pm.lock);
 461                 mptcp_subflow_shutdown(sk, ssk, how);
 462                 __mptcp_close_ssk(sk, ssk, subflow);
 463                 spin_lock_bh(&msk->pm.lock);
 464
 465                 msk->pm.add_addr_accepted--;
 466                 msk->pm.subflows--;
 467                 WRITE_ONCE(msk->pm.accept_addr, true);
 468
 469                 __MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_RMADDR);
 470
 471                 break;
 472         }
 473 }
 474
 475 void mptcp_pm_nl_rm_subflow_received(struct mptcp_sock *msk, u8 rm_id)
 476 {
 477         struct mptcp_subflow_context *subflow, *tmp;
 478         struct sock *sk = (struct sock *)msk;
 479
 480         pr_debug("subflow rm_id %d", rm_id);
 481
 482         if (!rm_id)
 483                 return;
 484
 485         if (list_empty(&msk->conn_list))
 486                 return;
 487
 488         list_for_each_entry_safe(subflow, tmp, &msk->conn_list, node) {
 489                 struct sock *ssk = mptcp_subflow_tcp_sock(subflow);
 490                 int how = RCV_SHUTDOWN | SEND_SHUTDOWN;
 491
 492                 if (rm_id != subflow->local_id)
 493                         continue;
 494
 495                 spin_unlock_bh(&msk->pm.lock);
 496                 mptcp_subflow_shutdown(sk, ssk, how);
 497                 __mptcp_close_ssk(sk, ssk, subflow);
 498                 spin_lock_bh(&msk->pm.lock);
 499
 500                 msk->pm.local_addr_used--;
 501                 msk->pm.subflows--;
 502
 503                 __MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_RMSUBFLOW);
 504
 505                 break;
 506         }
 507 }
 508
 509 static bool address_use_port(struct mptcp_pm_addr_entry *entry)
 510 {
 511         return (entry->addr.flags &
 512                 (MPTCP_PM_ADDR_FLAG_SIGNAL | MPTCP_PM_ADDR_FLAG_SUBFLOW)) ==
 513                 MPTCP_PM_ADDR_FLAG_SIGNAL;
 514 }
 515
 516 static int mptcp_pm_nl_append_new_local_addr(struct pm_nl_pernet *pernet,
 517                                              struct mptcp_pm_addr_entry *entry)
 518 {
 519         struct mptcp_pm_addr_entry *cur;
 520         int ret = -EINVAL;
 521
 522         spin_lock_bh(&pernet->lock);
 523         /* to keep the code simple, don't do IDR-like allocation for address ID,
 524          * just bail when we exceed limits
 525          */
 526         if (pernet->next_id > 255)
 527                 goto out;
 528         if (pernet->addrs >= MPTCP_PM_ADDR_MAX)
 529                 goto out;
 530
 531         /* do not insert duplicate address, differentiate on port only
 532          * singled addresses
 533          */
 534         list_for_each_entry(cur, &pernet->local_addr_list, list) {
 535                 if (addresses_equal(&cur->addr, &entry->addr,
 536                                     address_use_port(entry) &&
 537                                     address_use_port(cur)))
 538                         goto out;
 539         }
 540
 541         if (entry->addr.flags & MPTCP_PM_ADDR_FLAG_SIGNAL)
 542                 pernet->add_addr_signal_max++;
 543         if (entry->addr.flags & MPTCP_PM_ADDR_FLAG_SUBFLOW)
 544                 pernet->local_addr_max++;
 545
 546         entry->addr.id = pernet->next_id++;
 547         pernet->addrs++;
 548         list_add_tail_rcu(&entry->list, &pernet->local_addr_list);
 549         ret = entry->addr.id;
 550
 551 out:
 552         spin_unlock_bh(&pernet->lock);
 553         return ret;
 554 }
 555
 556 int mptcp_pm_nl_get_local_id(struct mptcp_sock *msk, struct sock_common *skc)
 557 {
 558         struct mptcp_pm_addr_entry *entry;
 559         struct mptcp_addr_info skc_local;
 560         struct mptcp_addr_info msk_local;
 561         struct pm_nl_pernet *pernet;
 562         int ret = -1;
 563
 564         if (WARN_ON_ONCE(!msk))
 565                 return -1;
 566
 567         /* The 0 ID mapping is defined by the first subflow, copied into the msk
 568          * addr
 569          */
 570         local_address((struct sock_common *)msk, &msk_local);
 571         local_address((struct sock_common *)skc, &skc_local);
 572         if (addresses_equal(&msk_local, &skc_local, false))
 573                 return 0;
 574
 575         if (address_zero(&skc_local))
 576                 return 0;
 577
 578         pernet = net_generic(sock_net((struct sock *)msk), pm_nl_pernet_id);
 579
 580         rcu_read_lock();
 581         list_for_each_entry_rcu(entry, &pernet->local_addr_list, list) {
 582                 if (addresses_equal(&entry->addr, &skc_local, false)) {
 583                         ret = entry->addr.id;
 584                         break;
 585                 }
 586         }
 587         rcu_read_unlock();
 588         if (ret >= 0)
 589                 return ret;
 590
 591         /* address not found, add to local list */
 592         entry = kmalloc(sizeof(*entry), GFP_ATOMIC);
 593         if (!entry)
 594                 return -ENOMEM;
 595
 596         entry->addr = skc_local;
 597         entry->addr.ifindex = 0;
 598         entry->addr.flags = 0;
 599         ret = mptcp_pm_nl_append_new_local_addr(pernet, entry);
 600         if (ret < 0)
 601                 kfree(entry);
 602
 603         return ret;
 604 }
 605
 606 void mptcp_pm_nl_data_init(struct mptcp_sock *msk)
 607 {
 608         struct mptcp_pm_data *pm = &msk->pm;
 609         struct pm_nl_pernet *pernet;
 610         bool subflows;
 611
 612         pernet = net_generic(sock_net((struct sock *)msk), pm_nl_pernet_id);
 613
 614         pm->add_addr_signal_max = READ_ONCE(pernet->add_addr_signal_max);
 615         pm->add_addr_accept_max = READ_ONCE(pernet->add_addr_accept_max);
 616         pm->local_addr_max = READ_ONCE(pernet->local_addr_max);
 617         pm->subflows_max = READ_ONCE(pernet->subflows_max);
 618         subflows = !!pm->subflows_max;
 619         WRITE_ONCE(pm->work_pending, (!!pm->local_addr_max && subflows) ||
 620                    !!pm->add_addr_signal_max);
 621         WRITE_ONCE(pm->accept_addr, !!pm->add_addr_accept_max && subflows);
 622         WRITE_ONCE(pm->accept_subflow, subflows);
 623 }
 624
 625 #define MPTCP_PM_CMD_GRP_OFFSET 0
 626
 627 static const struct genl_multicast_group mptcp_pm_mcgrps[] = {
 628         [MPTCP_PM_CMD_GRP_OFFSET]       = { .name = MPTCP_PM_CMD_GRP_NAME, },
 629 };
 630
 631 static const struct nla_policy
 632 mptcp_pm_addr_policy[MPTCP_PM_ADDR_ATTR_MAX + 1] = {
 633         [MPTCP_PM_ADDR_ATTR_FAMILY]     = { .type       = NLA_U16,      },
 634         [MPTCP_PM_ADDR_ATTR_ID]         = { .type       = NLA_U8,       },
 635         [MPTCP_PM_ADDR_ATTR_ADDR4]      = { .type       = NLA_U32,      },
 636         [MPTCP_PM_ADDR_ATTR_ADDR6]      =
 637                 NLA_POLICY_EXACT_LEN(sizeof(struct in6_addr)),
 638         [MPTCP_PM_ADDR_ATTR_PORT]       = { .type       = NLA_U16       },
 639         [MPTCP_PM_ADDR_ATTR_FLAGS]      = { .type       = NLA_U32       },
 640         [MPTCP_PM_ADDR_ATTR_IF_IDX]     = { .type       = NLA_S32       },
 641 };
 642
 643 static const struct nla_policy mptcp_pm_policy[MPTCP_PM_ATTR_MAX + 1] = {
 644         [MPTCP_PM_ATTR_ADDR]            =
 645                                         NLA_POLICY_NESTED(mptcp_pm_addr_policy),
 646         [MPTCP_PM_ATTR_RCV_ADD_ADDRS]   = { .type       = NLA_U32,      },
 647         [MPTCP_PM_ATTR_SUBFLOWS]        = { .type       = NLA_U32,      },
 648 };
 649
 650 static int mptcp_pm_family_to_addr(int family)
 651 {
 652 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
 653         if (family == AF_INET6)
 654                 return MPTCP_PM_ADDR_ATTR_ADDR6;
 655 #endif
 656         return MPTCP_PM_ADDR_ATTR_ADDR4;
 657 }
 658
 659 static int mptcp_pm_parse_addr(struct nlattr *attr, struct genl_info *info,
 660                                bool require_family,
 661                                struct mptcp_pm_addr_entry *entry)
 662 {
 663         struct nlattr *tb[MPTCP_PM_ADDR_ATTR_MAX + 1];
 664         int err, addr_addr;
 665
 666         if (!attr) {
 667                 GENL_SET_ERR_MSG(info, "missing address info");
 668                 return -EINVAL;
 669         }
 670
 671         /* no validation needed - was already done via nested policy */
 672         err = nla_parse_nested_deprecated(tb, MPTCP_PM_ADDR_ATTR_MAX, attr,
 673                                           mptcp_pm_addr_policy, info->extack);
 674         if (err)
 675                 return err;
 676
 677         memset(entry, 0, sizeof(*entry));
 678         if (!tb[MPTCP_PM_ADDR_ATTR_FAMILY]) {
 679                 if (!require_family)
 680                         goto skip_family;
 681
 682                 NL_SET_ERR_MSG_ATTR(info->extack, attr,
 683                                     "missing family");
 684                 return -EINVAL;
 685         }
 686
 687         entry->addr.family = nla_get_u16(tb[MPTCP_PM_ADDR_ATTR_FAMILY]);
 688         if (entry->addr.family != AF_INET
 689 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
 690             && entry->addr.family != AF_INET6
 691 #endif
 692             ) {
 693                 NL_SET_ERR_MSG_ATTR(info->extack, attr,
 694                                     "unknown address family");
 695                 return -EINVAL;
 696         }
 697         addr_addr = mptcp_pm_family_to_addr(entry->addr.family);
 698         if (!tb[addr_addr]) {
 699                 NL_SET_ERR_MSG_ATTR(info->extack, attr,
 700                                     "missing address data");
 701                 return -EINVAL;
 702         }
 703
 704 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
 705         if (entry->addr.family == AF_INET6)
 706                 entry->addr.addr6 = nla_get_in6_addr(tb[addr_addr]);
 707         else
 708 #endif
 709                 entry->addr.addr.s_addr = nla_get_in_addr(tb[addr_addr]);
 710
 711 skip_family:
 712         if (tb[MPTCP_PM_ADDR_ATTR_IF_IDX]) {
 713                 u32 val = nla_get_s32(tb[MPTCP_PM_ADDR_ATTR_IF_IDX]);
 714
 715                 entry->addr.ifindex = val;
 716         }
 717
 718         if (tb[MPTCP_PM_ADDR_ATTR_ID])
 719                 entry->addr.id = nla_get_u8(tb[MPTCP_PM_ADDR_ATTR_ID]);
 720
 721         if (tb[MPTCP_PM_ADDR_ATTR_FLAGS])
 722                 entry->addr.flags = nla_get_u32(tb[MPTCP_PM_ADDR_ATTR_FLAGS]);
 723
 724         return 0;
 725 }
 726
 727 static struct pm_nl_pernet *genl_info_pm_nl(struct genl_info *info)
 728 {
 729         return net_generic(genl_info_net(info), pm_nl_pernet_id);
 730 }
 731
 732 static int mptcp_nl_cmd_add_addr(struct sk_buff *skb, struct genl_info *info)
 733 {
 734         struct nlattr *attr = info->attrs[MPTCP_PM_ATTR_ADDR];
 735         struct pm_nl_pernet *pernet = genl_info_pm_nl(info);
 736         struct mptcp_pm_addr_entry addr, *entry;
 737         int ret;
 738
 739         ret = mptcp_pm_parse_addr(attr, info, true, &addr);
 740         if (ret < 0)
 741                 return ret;
 742
 743         entry = kmalloc(sizeof(*entry), GFP_KERNEL);
 744         if (!entry) {
 745                 GENL_SET_ERR_MSG(info, "can't allocate addr");
 746                 return -ENOMEM;
 747         }
 748
 749         *entry = addr;
 750         ret = mptcp_pm_nl_append_new_local_addr(pernet, entry);
 751         if (ret < 0) {
 752                 GENL_SET_ERR_MSG(info, "too many addresses or duplicate one");
 753                 kfree(entry);
 754                 return ret;
 755         }
 756
 757         return 0;
 758 }
 759
 760 static struct mptcp_pm_addr_entry *
 761 __lookup_addr_by_id(struct pm_nl_pernet *pernet, unsigned int id)
 762 {
 763         struct mptcp_pm_addr_entry *entry;
 764
 765         list_for_each_entry(entry, &pernet->local_addr_list, list) {
 766                 if (entry->addr.id == id)
 767                         return entry;
 768         }
 769         return NULL;
 770 }
 771
 772 static bool remove_anno_list_by_saddr(struct mptcp_sock *msk,
 773                                       struct mptcp_addr_info *addr)
 774 {
 775         struct mptcp_pm_add_entry *entry;
 776
 777         entry = mptcp_pm_del_add_timer(msk, addr);
 778         if (entry) {
 779                 list_del(&entry->list);
 780                 kfree(entry);
 781                 return true;
 782         }
 783
 784         return false;
 785 }
 786
 787 static bool mptcp_pm_remove_anno_addr(struct mptcp_sock *msk,
 788                                       struct mptcp_addr_info *addr,
 789                                       bool force)
 790 {
 791         bool ret;
 792
 793         ret = remove_anno_list_by_saddr(msk, addr);
 794         if (ret || force) {
 795                 spin_lock_bh(&msk->pm.lock);
 796                 mptcp_pm_remove_addr(msk, addr->id);
 797                 spin_unlock_bh(&msk->pm.lock);
 798         }
 799         return ret;
 800 }
 801
 802 static int mptcp_nl_remove_subflow_and_signal_addr(struct net *net,
 803                                                    struct mptcp_addr_info *addr)
 804 {
 805         struct mptcp_sock *msk;
 806         long s_slot = 0, s_num = 0;
 807
 808         pr_debug("remove_id=%d", addr->id);
 809
 810         while ((msk = mptcp_token_iter_next(net, &s_slot, &s_num)) != NULL) {
 811                 struct sock *sk = (struct sock *)msk;
 812                 bool remove_subflow;
 813
 814                 if (list_empty(&msk->conn_list)) {
 815                         mptcp_pm_remove_anno_addr(msk, addr, false);
 816                         goto next;
 817                 }
 818
 819                 lock_sock(sk);
 820                 remove_subflow = lookup_subflow_by_saddr(&msk->conn_list, addr);
 821                 mptcp_pm_remove_anno_addr(msk, addr, remove_subflow);
 822                 if (remove_subflow)
 823                         mptcp_pm_remove_subflow(msk, addr->id);
 824                 release_sock(sk);
 825
 826 next:
 827                 sock_put(sk);
 828                 cond_resched();
 829         }
 830
 831         return 0;
 832 }
 833
 834 static int mptcp_nl_cmd_del_addr(struct sk_buff *skb, struct genl_info *info)
 835 {
 836         struct nlattr *attr = info->attrs[MPTCP_PM_ATTR_ADDR];
 837         struct pm_nl_pernet *pernet = genl_info_pm_nl(info);
 838         struct mptcp_pm_addr_entry addr, *entry;
 839         int ret;
 840
 841         ret = mptcp_pm_parse_addr(attr, info, false, &addr);
 842         if (ret < 0)
 843                 return ret;
 844
 845         spin_lock_bh(&pernet->lock);
 846         entry = __lookup_addr_by_id(pernet, addr.addr.id);
 847         if (!entry) {
 848                 GENL_SET_ERR_MSG(info, "address not found");
 849                 spin_unlock_bh(&pernet->lock);
 850                 return -EINVAL;
 851         }
 852         if (entry->addr.flags & MPTCP_PM_ADDR_FLAG_SIGNAL)
 853                 pernet->add_addr_signal_max--;
 854         if (entry->addr.flags & MPTCP_PM_ADDR_FLAG_SUBFLOW)
 855                 pernet->local_addr_max--;
 856
 857         pernet->addrs--;
 858         list_del_rcu(&entry->list);
 859         spin_unlock_bh(&pernet->lock);
 860
 861         mptcp_nl_remove_subflow_and_signal_addr(sock_net(skb->sk), &entry->addr);
 862         kfree_rcu(entry, rcu);
 863
 864         return ret;
 865 }
 866
 867 static void __flush_addrs(struct pm_nl_pernet *pernet)
 868 {
 869         while (!list_empty(&pernet->local_addr_list)) {
 870                 struct mptcp_pm_addr_entry *cur;
 871
 872                 cur = list_entry(pernet->local_addr_list.next,
 873                                  struct mptcp_pm_addr_entry, list);
 874                 list_del_rcu(&cur->list);
 875                 kfree_rcu(cur, rcu);
 876         }
 877 }
 878
 879 static void __reset_counters(struct pm_nl_pernet *pernet)
 880 {
 881         pernet->add_addr_signal_max = 0;
 882         pernet->add_addr_accept_max = 0;
 883         pernet->local_addr_max = 0;
 884         pernet->addrs = 0;
 885 }
 886
 887 static int mptcp_nl_cmd_flush_addrs(struct sk_buff *skb, struct genl_info *info)
 888 {
 889         struct pm_nl_pernet *pernet = genl_info_pm_nl(info);
 890
 891         spin_lock_bh(&pernet->lock);
 892         __flush_addrs(pernet);
 893         __reset_counters(pernet);
 894         spin_unlock_bh(&pernet->lock);
 895         return 0;
 896 }
 897
 898 static int mptcp_nl_fill_addr(struct sk_buff *skb,
 899                               struct mptcp_pm_addr_entry *entry)
 900 {
 901         struct mptcp_addr_info *addr = &entry->addr;
 902         struct nlattr *attr;
 903
 904         attr = nla_nest_start(skb, MPTCP_PM_ATTR_ADDR);
 905         if (!attr)
 906                 return -EMSGSIZE;
 907
 908         if (nla_put_u16(skb, MPTCP_PM_ADDR_ATTR_FAMILY, addr->family))
 909                 goto nla_put_failure;
 910         if (nla_put_u8(skb, MPTCP_PM_ADDR_ATTR_ID, addr->id))
 911                 goto nla_put_failure;
 912         if (nla_put_u32(skb, MPTCP_PM_ADDR_ATTR_FLAGS, entry->addr.flags))
 913                 goto nla_put_failure;
 914         if (entry->addr.ifindex &&
 915             nla_put_s32(skb, MPTCP_PM_ADDR_ATTR_IF_IDX, entry->addr.ifindex))
 916                 goto nla_put_failure;
 917
 918         if (addr->family == AF_INET &&
 919             nla_put_in_addr(skb, MPTCP_PM_ADDR_ATTR_ADDR4,
 920                             addr->addr.s_addr))
 921                 goto nla_put_failure;
 922 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
 923         else if (addr->family == AF_INET6 &&
 924                  nla_put_in6_addr(skb, MPTCP_PM_ADDR_ATTR_ADDR6, &addr->addr6))
 925                 goto nla_put_failure;
 926 #endif
 927         nla_nest_end(skb, attr);
 928         return 0;
 929
 930 nla_put_failure:
 931         nla_nest_cancel(skb, attr);
 932         return -EMSGSIZE;
 933 }
 934
 935 static int mptcp_nl_cmd_get_addr(struct sk_buff *skb, struct genl_info *info)
 936 {
 937         struct nlattr *attr = info->attrs[MPTCP_PM_ATTR_ADDR];
 938         struct pm_nl_pernet *pernet = genl_info_pm_nl(info);
 939         struct mptcp_pm_addr_entry addr, *entry;
 940         struct sk_buff *msg;
 941         void *reply;
 942         int ret;
 943
 944         ret = mptcp_pm_parse_addr(attr, info, false, &addr);
 945         if (ret < 0)
 946                 return ret;
 947
 948         msg = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
 949         if (!msg)
 950                 return -ENOMEM;
 951
 952         reply = genlmsg_put_reply(msg, info, &mptcp_genl_family, 0,
 953                                   info->genlhdr->cmd);
 954         if (!reply) {
 955                 GENL_SET_ERR_MSG(info, "not enough space in Netlink message");
 956                 ret = -EMSGSIZE;
 957                 goto fail;
 958         }
 959
 960         spin_lock_bh(&pernet->lock);
 961         entry = __lookup_addr_by_id(pernet, addr.addr.id);
 962         if (!entry) {
 963                 GENL_SET_ERR_MSG(info, "address not found");
 964                 ret = -EINVAL;
 965                 goto unlock_fail;
 966         }
 967
 968         ret = mptcp_nl_fill_addr(msg, entry);
 969         if (ret)
 970                 goto unlock_fail;
 971
 972         genlmsg_end(msg, reply);
 973         ret = genlmsg_reply(msg, info);
 974         spin_unlock_bh(&pernet->lock);
 975         return ret;
 976
 977 unlock_fail:
 978         spin_unlock_bh(&pernet->lock);
 979
 980 fail:
 981         nlmsg_free(msg);
 982         return ret;
 983 }
 984
 985 static int mptcp_nl_cmd_dump_addrs(struct sk_buff *msg,
 986                                    struct netlink_callback *cb)
 987 {
 988         struct net *net = sock_net(msg->sk);
 989         struct mptcp_pm_addr_entry *entry;
 990         struct pm_nl_pernet *pernet;
 991         int id = cb->args[0];
 992         void *hdr;
 993
 994         pernet = net_generic(net, pm_nl_pernet_id);
 995
 996         spin_lock_bh(&pernet->lock);
 997         list_for_each_entry(entry, &pernet->local_addr_list, list) {
 998                 if (entry->addr.id <= id)
 999                         continue;
1000
1001                 hdr = genlmsg_put(msg, NETLINK_CB(cb->skb).portid,
1002                                   cb->nlh->nlmsg_seq, &mptcp_genl_family,
1003                                   NLM_F_MULTI, MPTCP_PM_CMD_GET_ADDR);
1004                 if (!hdr)
1005                         break;
1006
1007                 if (mptcp_nl_fill_addr(msg, entry) < 0) {
1008                         genlmsg_cancel(msg, hdr);
1009                         break;
1010                 }
1011
1012                 id = entry->addr.id;
1013                 genlmsg_end(msg, hdr);
1014         }
1015         spin_unlock_bh(&pernet->lock);
1016
1017         cb->args[0] = id;
1018         return msg->len;
1019 }
1020
1021 static int parse_limit(struct genl_info *info, int id, unsigned int *limit)
1022 {
1023         struct nlattr *attr = info->attrs[id];
1024
1025         if (!attr)
1026                 return 0;
1027
1028         *limit = nla_get_u32(attr);
1029         if (*limit > MPTCP_PM_ADDR_MAX) {
1030                 GENL_SET_ERR_MSG(info, "limit greater than maximum");
1031                 return -EINVAL;
1032         }
1033         return 0;
1034 }
1035
1036 static int
1037 mptcp_nl_cmd_set_limits(struct sk_buff *skb, struct genl_info *info)
1038 {
1039         struct pm_nl_pernet *pernet = genl_info_pm_nl(info);
1040         unsigned int rcv_addrs, subflows;
1041         int ret;
1042
1043         spin_lock_bh(&pernet->lock);
1044         rcv_addrs = pernet->add_addr_accept_max;
1045         ret = parse_limit(info, MPTCP_PM_ATTR_RCV_ADD_ADDRS, &rcv_addrs);
1046         if (ret)
1047                 goto unlock;
1048
1049         subflows = pernet->subflows_max;
1050         ret = parse_limit(info, MPTCP_PM_ATTR_SUBFLOWS, &subflows);
1051         if (ret)
1052                 goto unlock;
1053
1054         WRITE_ONCE(pernet->add_addr_accept_max, rcv_addrs);
1055         WRITE_ONCE(pernet->subflows_max, subflows);
1056
1057 unlock:
1058         spin_unlock_bh(&pernet->lock);
1059         return ret;
1060 }
1061
1062 static int
1063 mptcp_nl_cmd_get_limits(struct sk_buff *skb, struct genl_info *info)
1064 {
1065         struct pm_nl_pernet *pernet = genl_info_pm_nl(info);
1066         struct sk_buff *msg;
1067         void *reply;
1068
1069         msg = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
1070         if (!msg)
1071                 return -ENOMEM;
1072
1073         reply = genlmsg_put_reply(msg, info, &mptcp_genl_family, 0,
1074                                   MPTCP_PM_CMD_GET_LIMITS);
1075         if (!reply)
1076                 goto fail;
1077
1078         if (nla_put_u32(msg, MPTCP_PM_ATTR_RCV_ADD_ADDRS,
1079                         READ_ONCE(pernet->add_addr_accept_max)))
1080                 goto fail;
1081
1082         if (nla_put_u32(msg, MPTCP_PM_ATTR_SUBFLOWS,
1083                         READ_ONCE(pernet->subflows_max)))
1084                 goto fail;
1085
1086         genlmsg_end(msg, reply);
1087         return genlmsg_reply(msg, info);
1088
1089 fail:
1090         GENL_SET_ERR_MSG(info, "not enough space in Netlink message");
1091         nlmsg_free(msg);
1092         return -EMSGSIZE;
1093 }
1094
1095 static const struct genl_small_ops mptcp_pm_ops[] = {
1096         {
1097                 .cmd    = MPTCP_PM_CMD_ADD_ADDR,
1098                 .doit   = mptcp_nl_cmd_add_addr,
1099                 .flags  = GENL_ADMIN_PERM,
1100         },
1101         {
1102                 .cmd    = MPTCP_PM_CMD_DEL_ADDR,
1103                 .doit   = mptcp_nl_cmd_del_addr,
1104                 .flags  = GENL_ADMIN_PERM,
1105         },
1106         {
1107                 .cmd    = MPTCP_PM_CMD_FLUSH_ADDRS,
1108                 .doit   = mptcp_nl_cmd_flush_addrs,
1109                 .flags  = GENL_ADMIN_PERM,
1110         },
1111         {
1112                 .cmd    = MPTCP_PM_CMD_GET_ADDR,
1113                 .doit   = mptcp_nl_cmd_get_addr,
1114                 .dumpit   = mptcp_nl_cmd_dump_addrs,
1115         },
1116         {
1117                 .cmd    = MPTCP_PM_CMD_SET_LIMITS,
1118                 .doit   = mptcp_nl_cmd_set_limits,
1119                 .flags  = GENL_ADMIN_PERM,
1120         },
1121         {
1122                 .cmd    = MPTCP_PM_CMD_GET_LIMITS,
1123                 .doit   = mptcp_nl_cmd_get_limits,
1124         },
1125 };
1126
1127 static struct genl_family mptcp_genl_family __ro_after_init = {
1128         .name           = MPTCP_PM_NAME,
1129         .version        = MPTCP_PM_VER,
1130         .maxattr        = MPTCP_PM_ATTR_MAX,
1131         .policy         = mptcp_pm_policy,
1132         .netnsok        = true,
1133         .module         = THIS_MODULE,
1134         .small_ops      = mptcp_pm_ops,
1135         .n_small_ops    = ARRAY_SIZE(mptcp_pm_ops),
1136         .mcgrps         = mptcp_pm_mcgrps,
1137         .n_mcgrps       = ARRAY_SIZE(mptcp_pm_mcgrps),
1138 };
1139
1140 static int __net_init pm_nl_init_net(struct net *net)
1141 {
1142         struct pm_nl_pernet *pernet = net_generic(net, pm_nl_pernet_id);
1143
1144         INIT_LIST_HEAD_RCU(&pernet->local_addr_list);
1145         __reset_counters(pernet);
1146         pernet->next_id = 1;
1147         spin_lock_init(&pernet->lock);
1148         return 0;
1149 }
1150
1151 static void __net_exit pm_nl_exit_net(struct list_head *net_list)
1152 {
1153         struct net *net;
1154
1155         list_for_each_entry(net, net_list, exit_list) {
1156                 /* net is removed from namespace list, can't race with
1157                  * other modifiers
1158                  */
1159                 __flush_addrs(net_generic(net, pm_nl_pernet_id));
1160         }
1161 }
1162
1163 static struct pernet_operations mptcp_pm_pernet_ops = {
1164         .init = pm_nl_init_net,
1165         .exit_batch = pm_nl_exit_net,
1166         .id = &pm_nl_pernet_id,
1167         .size = sizeof(struct pm_nl_pernet),
1168 };
1169
1170 void __init mptcp_pm_nl_init(void)
1171 {
1172         if (register_pernet_subsys(&mptcp_pm_pernet_ops) < 0)
1173                 panic("Failed to register MPTCP PM pernet subsystem.\n");
1174
1175         if (genl_register_family(&mptcp_genl_family))
1176                 panic("Failed to register MPTCP PM netlink family\n");
1177 }