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         u8                      flags;
  29         int                     ifindex;
  30         struct socket           *lsk;
  31 };
  32
  33 struct mptcp_pm_add_entry {
  34         struct list_head        list;
  35         struct mptcp_addr_info  addr;
  36         struct timer_list       add_timer;
  37         struct mptcp_sock       *sock;
  38         u8                      retrans_times;
  39 };
  40
  41 #define MAX_ADDR_ID             255
  42 #define BITMAP_SZ DIV_ROUND_UP(MAX_ADDR_ID + 1, BITS_PER_LONG)
  43
  44 struct pm_nl_pernet {
  45         /* protects pernet updates */
  46         spinlock_t              lock;
  47         struct list_head        local_addr_list;
  48         unsigned int            addrs;
  49         unsigned int            stale_loss_cnt;
  50         unsigned int            add_addr_signal_max;
  51         unsigned int            add_addr_accept_max;
  52         unsigned int            local_addr_max;
  53         unsigned int            subflows_max;
  54         unsigned int            next_id;
  55         unsigned long           id_bitmap[BITMAP_SZ];
  56 };
  57
  58 #define MPTCP_PM_ADDR_MAX       8
  59 #define ADD_ADDR_RETRANS_MAX    3
  60
  61 static bool addresses_equal(const struct mptcp_addr_info *a,
  62                             struct mptcp_addr_info *b, bool use_port)
  63 {
  64         bool addr_equals = false;
  65
  66         if (a->family == b->family) {
  67                 if (a->family == AF_INET)
  68                         addr_equals = a->addr.s_addr == b->addr.s_addr;
  69 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
  70                 else
  71                         addr_equals = !ipv6_addr_cmp(&a->addr6, &b->addr6);
  72         } else if (a->family == AF_INET) {
  73                 if (ipv6_addr_v4mapped(&b->addr6))
  74                         addr_equals = a->addr.s_addr == b->addr6.s6_addr32[3];
  75         } else if (b->family == AF_INET) {
  76                 if (ipv6_addr_v4mapped(&a->addr6))
  77                         addr_equals = a->addr6.s6_addr32[3] == b->addr.s_addr;
  78 #endif
  79         }
  80
  81         if (!addr_equals)
  82                 return false;
  83         if (!use_port)
  84                 return true;
  85
  86         return a->port == b->port;
  87 }
  88
  89 static bool address_zero(const struct mptcp_addr_info *addr)
  90 {
  91         struct mptcp_addr_info zero;
  92
  93         memset(&zero, 0, sizeof(zero));
  94         zero.family = addr->family;
  95
  96         return addresses_equal(addr, &zero, true);
  97 }
  98
  99 static void local_address(const struct sock_common *skc,
 100                           struct mptcp_addr_info *addr)
 101 {
 102         addr->family = skc->skc_family;
 103         addr->port = htons(skc->skc_num);
 104         if (addr->family == AF_INET)
 105                 addr->addr.s_addr = skc->skc_rcv_saddr;
 106 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
 107         else if (addr->family == AF_INET6)
 108                 addr->addr6 = skc->skc_v6_rcv_saddr;
 109 #endif
 110 }
 111
 112 static void remote_address(const struct sock_common *skc,
 113                            struct mptcp_addr_info *addr)
 114 {
 115         addr->family = skc->skc_family;
 116         addr->port = skc->skc_dport;
 117         if (addr->family == AF_INET)
 118                 addr->addr.s_addr = skc->skc_daddr;
 119 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
 120         else if (addr->family == AF_INET6)
 121                 addr->addr6 = skc->skc_v6_daddr;
 122 #endif
 123 }
 124
 125 static bool lookup_subflow_by_saddr(const struct list_head *list,
 126                                     struct mptcp_addr_info *saddr)
 127 {
 128         struct mptcp_subflow_context *subflow;
 129         struct mptcp_addr_info cur;
 130         struct sock_common *skc;
 131
 132         list_for_each_entry(subflow, list, node) {
 133                 skc = (struct sock_common *)mptcp_subflow_tcp_sock(subflow);
 134
 135                 local_address(skc, &cur);
 136                 if (addresses_equal(&cur, saddr, saddr->port))
 137                         return true;
 138         }
 139
 140         return false;
 141 }
 142
 143 static bool lookup_subflow_by_daddr(const struct list_head *list,
 144                                     struct mptcp_addr_info *daddr)
 145 {
 146         struct mptcp_subflow_context *subflow;
 147         struct mptcp_addr_info cur;
 148         struct sock_common *skc;
 149
 150         list_for_each_entry(subflow, list, node) {
 151                 skc = (struct sock_common *)mptcp_subflow_tcp_sock(subflow);
 152
 153                 remote_address(skc, &cur);
 154                 if (addresses_equal(&cur, daddr, daddr->port))
 155                         return true;
 156         }
 157
 158         return false;
 159 }
 160
 161 static struct mptcp_pm_addr_entry *
 162 select_local_address(const struct pm_nl_pernet *pernet,
 163                      struct mptcp_sock *msk)
 164 {
 165         struct mptcp_pm_addr_entry *entry, *ret = NULL;
 166         struct sock *sk = (struct sock *)msk;
 167
 168         msk_owned_by_me(msk);
 169
 170         rcu_read_lock();
 171         __mptcp_flush_join_list(msk);
 172         list_for_each_entry_rcu(entry, &pernet->local_addr_list, list) {
 173                 if (!(entry->flags & MPTCP_PM_ADDR_FLAG_SUBFLOW))
 174                         continue;
 175
 176                 if (entry->addr.family != sk->sk_family) {
 177 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
 178                         if ((entry->addr.family == AF_INET &&
 179                              !ipv6_addr_v4mapped(&sk->sk_v6_daddr)) ||
 180                             (sk->sk_family == AF_INET &&
 181                              !ipv6_addr_v4mapped(&entry->addr.addr6)))
 182 #endif
 183                                 continue;
 184                 }
 185
 186                 /* avoid any address already in use by subflows and
 187                  * pending join
 188                  */
 189                 if (!lookup_subflow_by_saddr(&msk->conn_list, &entry->addr)) {
 190                         ret = entry;
 191                         break;
 192                 }
 193         }
 194         rcu_read_unlock();
 195         return ret;
 196 }
 197
 198 static struct mptcp_pm_addr_entry *
 199 select_signal_address(struct pm_nl_pernet *pernet, unsigned int pos)
 200 {
 201         struct mptcp_pm_addr_entry *entry, *ret = NULL;
 202         int i = 0;
 203
 204         rcu_read_lock();
 205         /* do not keep any additional per socket state, just signal
 206          * the address list in order.
 207          * Note: removal from the local address list during the msk life-cycle
 208          * can lead to additional addresses not being announced.
 209          */
 210         list_for_each_entry_rcu(entry, &pernet->local_addr_list, list) {
 211                 if (!(entry->flags & MPTCP_PM_ADDR_FLAG_SIGNAL))
 212                         continue;
 213                 if (i++ == pos) {
 214                         ret = entry;
 215                         break;
 216                 }
 217         }
 218         rcu_read_unlock();
 219         return ret;
 220 }
 221
 222 unsigned int mptcp_pm_get_add_addr_signal_max(struct mptcp_sock *msk)
 223 {
 224         struct pm_nl_pernet *pernet;
 225
 226         pernet = net_generic(sock_net((struct sock *)msk), pm_nl_pernet_id);
 227         return READ_ONCE(pernet->add_addr_signal_max);
 228 }
 229 EXPORT_SYMBOL_GPL(mptcp_pm_get_add_addr_signal_max);
 230
 231 unsigned int mptcp_pm_get_add_addr_accept_max(struct mptcp_sock *msk)
 232 {
 233         struct pm_nl_pernet *pernet;
 234
 235         pernet = net_generic(sock_net((struct sock *)msk), pm_nl_pernet_id);
 236         return READ_ONCE(pernet->add_addr_accept_max);
 237 }
 238 EXPORT_SYMBOL_GPL(mptcp_pm_get_add_addr_accept_max);
 239
 240 unsigned int mptcp_pm_get_subflows_max(struct mptcp_sock *msk)
 241 {
 242         struct pm_nl_pernet *pernet;
 243
 244         pernet = net_generic(sock_net((struct sock *)msk), pm_nl_pernet_id);
 245         return READ_ONCE(pernet->subflows_max);
 246 }
 247 EXPORT_SYMBOL_GPL(mptcp_pm_get_subflows_max);
 248
 249 unsigned int mptcp_pm_get_local_addr_max(struct mptcp_sock *msk)
 250 {
 251         struct pm_nl_pernet *pernet;
 252
 253         pernet = net_generic(sock_net((struct sock *)msk), pm_nl_pernet_id);
 254         return READ_ONCE(pernet->local_addr_max);
 255 }
 256 EXPORT_SYMBOL_GPL(mptcp_pm_get_local_addr_max);
 257
 258 static void check_work_pending(struct mptcp_sock *msk)
 259 {
 260         if (msk->pm.add_addr_signaled == mptcp_pm_get_add_addr_signal_max(msk) &&
 261             (msk->pm.local_addr_used == mptcp_pm_get_local_addr_max(msk) ||
 262              msk->pm.subflows == mptcp_pm_get_subflows_max(msk)))
 263                 WRITE_ONCE(msk->pm.work_pending, false);
 264 }
 265
 266 struct mptcp_pm_add_entry *
 267 mptcp_lookup_anno_list_by_saddr(struct mptcp_sock *msk,
 268                                 struct mptcp_addr_info *addr)
 269 {
 270         struct mptcp_pm_add_entry *entry;
 271
 272         lockdep_assert_held(&msk->pm.lock);
 273
 274         list_for_each_entry(entry, &msk->pm.anno_list, list) {
 275                 if (addresses_equal(&entry->addr, addr, true))
 276                         return entry;
 277         }
 278
 279         return NULL;
 280 }
 281
 282 bool mptcp_pm_sport_in_anno_list(struct mptcp_sock *msk, const struct sock *sk)
 283 {
 284         struct mptcp_pm_add_entry *entry;
 285         struct mptcp_addr_info saddr;
 286         bool ret = false;
 287
 288         local_address((struct sock_common *)sk, &saddr);
 289
 290         spin_lock_bh(&msk->pm.lock);
 291         list_for_each_entry(entry, &msk->pm.anno_list, list) {
 292                 if (addresses_equal(&entry->addr, &saddr, true)) {
 293                         ret = true;
 294                         goto out;
 295                 }
 296         }
 297
 298 out:
 299         spin_unlock_bh(&msk->pm.lock);
 300         return ret;
 301 }
 302
 303 static void mptcp_pm_add_timer(struct timer_list *timer)
 304 {
 305         struct mptcp_pm_add_entry *entry = from_timer(entry, timer, add_timer);
 306         struct mptcp_sock *msk = entry->sock;
 307         struct sock *sk = (struct sock *)msk;
 308
 309         pr_debug("msk=%p", msk);
 310
 311         if (!msk)
 312                 return;
 313
 314         if (inet_sk_state_load(sk) == TCP_CLOSE)
 315                 return;
 316
 317         if (!entry->addr.id)
 318                 return;
 319
 320         if (mptcp_pm_should_add_signal_addr(msk)) {
 321                 sk_reset_timer(sk, timer, jiffies + TCP_RTO_MAX / 8);
 322                 goto out;
 323         }
 324
 325         spin_lock_bh(&msk->pm.lock);
 326
 327         if (!mptcp_pm_should_add_signal_addr(msk)) {
 328                 pr_debug("retransmit ADD_ADDR id=%d", entry->addr.id);
 329                 mptcp_pm_announce_addr(msk, &entry->addr, false);
 330                 mptcp_pm_add_addr_send_ack(msk);
 331                 entry->retrans_times++;
 332         }
 333
 334         if (entry->retrans_times < ADD_ADDR_RETRANS_MAX)
 335                 sk_reset_timer(sk, timer,
 336                                jiffies + mptcp_get_add_addr_timeout(sock_net(sk)));
 337
 338         spin_unlock_bh(&msk->pm.lock);
 339
 340         if (entry->retrans_times == ADD_ADDR_RETRANS_MAX)
 341                 mptcp_pm_subflow_established(msk);
 342
 343 out:
 344         __sock_put(sk);
 345 }
 346
 347 struct mptcp_pm_add_entry *
 348 mptcp_pm_del_add_timer(struct mptcp_sock *msk,
 349                        struct mptcp_addr_info *addr, bool check_id)
 350 {
 351         struct mptcp_pm_add_entry *entry;
 352         struct sock *sk = (struct sock *)msk;
 353
 354         spin_lock_bh(&msk->pm.lock);
 355         entry = mptcp_lookup_anno_list_by_saddr(msk, addr);
 356         if (entry && (!check_id || entry->addr.id == addr->id))
 357                 entry->retrans_times = ADD_ADDR_RETRANS_MAX;
 358         spin_unlock_bh(&msk->pm.lock);
 359
 360         if (entry && (!check_id || entry->addr.id == addr->id))
 361                 sk_stop_timer_sync(sk, &entry->add_timer);
 362
 363         return entry;
 364 }
 365
 366 static bool mptcp_pm_alloc_anno_list(struct mptcp_sock *msk,
 367                                      struct mptcp_pm_addr_entry *entry)
 368 {
 369         struct mptcp_pm_add_entry *add_entry = NULL;
 370         struct sock *sk = (struct sock *)msk;
 371         struct net *net = sock_net(sk);
 372
 373         lockdep_assert_held(&msk->pm.lock);
 374
 375         if (mptcp_lookup_anno_list_by_saddr(msk, &entry->addr))
 376                 return false;
 377
 378         add_entry = kmalloc(sizeof(*add_entry), GFP_ATOMIC);
 379         if (!add_entry)
 380                 return false;
 381
 382         list_add(&add_entry->list, &msk->pm.anno_list);
 383
 384         add_entry->addr = entry->addr;
 385         add_entry->sock = msk;
 386         add_entry->retrans_times = 0;
 387
 388         timer_setup(&add_entry->add_timer, mptcp_pm_add_timer, 0);
 389         sk_reset_timer(sk, &add_entry->add_timer,
 390                        jiffies + mptcp_get_add_addr_timeout(net));
 391
 392         return true;
 393 }
 394
 395 void mptcp_pm_free_anno_list(struct mptcp_sock *msk)
 396 {
 397         struct mptcp_pm_add_entry *entry, *tmp;
 398         struct sock *sk = (struct sock *)msk;
 399         LIST_HEAD(free_list);
 400
 401         pr_debug("msk=%p", msk);
 402
 403         spin_lock_bh(&msk->pm.lock);
 404         list_splice_init(&msk->pm.anno_list, &free_list);
 405         spin_unlock_bh(&msk->pm.lock);
 406
 407         list_for_each_entry_safe(entry, tmp, &free_list, list) {
 408                 sk_stop_timer_sync(sk, &entry->add_timer);
 409                 kfree(entry);
 410         }
 411 }
 412
 413 static bool lookup_address_in_vec(struct mptcp_addr_info *addrs, unsigned int nr,
 414                                   struct mptcp_addr_info *addr)
 415 {
 416         int i;
 417
 418         for (i = 0; i < nr; i++) {
 419                 if (addresses_equal(&addrs[i], addr, addr->port))
 420                         return true;
 421         }
 422
 423         return false;
 424 }
 425
 426 /* Fill all the remote addresses into the array addrs[],
 427  * and return the array size.
 428  */
 429 static unsigned int fill_remote_addresses_vec(struct mptcp_sock *msk, bool fullmesh,
 430                                               struct mptcp_addr_info *addrs)
 431 {
 432         struct sock *sk = (struct sock *)msk, *ssk;
 433         struct mptcp_subflow_context *subflow;
 434         struct mptcp_addr_info remote = { 0 };
 435         unsigned int subflows_max;
 436         int i = 0;
 437
 438         subflows_max = mptcp_pm_get_subflows_max(msk);
 439
 440         /* Non-fullmesh endpoint, fill in the single entry
 441          * corresponding to the primary MPC subflow remote address
 442          */
 443         if (!fullmesh) {
 444                 remote_address((struct sock_common *)sk, &remote);
 445                 msk->pm.subflows++;
 446                 addrs[i++] = remote;
 447         } else {
 448                 mptcp_for_each_subflow(msk, subflow) {
 449                         ssk = mptcp_subflow_tcp_sock(subflow);
 450                         remote_address((struct sock_common *)ssk, &remote);
 451                         if (!lookup_address_in_vec(addrs, i, &remote) &&
 452                             msk->pm.subflows < subflows_max) {
 453                                 msk->pm.subflows++;
 454                                 addrs[i++] = remote;
 455                         }
 456                 }
 457         }
 458
 459         return i;
 460 }
 461
 462 static struct mptcp_pm_addr_entry *
 463 __lookup_addr(struct pm_nl_pernet *pernet, struct mptcp_addr_info *info)
 464 {
 465         struct mptcp_pm_addr_entry *entry;
 466
 467         list_for_each_entry(entry, &pernet->local_addr_list, list) {
 468                 if (addresses_equal(&entry->addr, info, true))
 469                         return entry;
 470         }
 471         return NULL;
 472 }
 473
 474 static void mptcp_pm_create_subflow_or_signal_addr(struct mptcp_sock *msk)
 475 {
 476         struct sock *sk = (struct sock *)msk;
 477         struct mptcp_pm_addr_entry *local;
 478         unsigned int add_addr_signal_max;
 479         unsigned int local_addr_max;
 480         struct pm_nl_pernet *pernet;
 481         unsigned int subflows_max;
 482
 483         pernet = net_generic(sock_net(sk), pm_nl_pernet_id);
 484
 485         add_addr_signal_max = mptcp_pm_get_add_addr_signal_max(msk);
 486         local_addr_max = mptcp_pm_get_local_addr_max(msk);
 487         subflows_max = mptcp_pm_get_subflows_max(msk);
 488
 489         pr_debug("local %d:%d signal %d:%d subflows %d:%d\n",
 490                  msk->pm.local_addr_used, local_addr_max,
 491                  msk->pm.add_addr_signaled, add_addr_signal_max,
 492                  msk->pm.subflows, subflows_max);
 493
 494         /* check first for announce */
 495         if (msk->pm.add_addr_signaled < add_addr_signal_max) {
 496                 local = select_signal_address(pernet,
 497                                               msk->pm.add_addr_signaled);
 498
 499                 if (local) {
 500                         if (mptcp_pm_alloc_anno_list(msk, local)) {
 501                                 msk->pm.add_addr_signaled++;
 502                                 mptcp_pm_announce_addr(msk, &local->addr, false);
 503                                 mptcp_pm_nl_addr_send_ack(msk);
 504                         }
 505                 } else {
 506                         /* pick failed, avoid fourther attempts later */
 507                         msk->pm.local_addr_used = add_addr_signal_max;
 508                 }
 509
 510                 check_work_pending(msk);
 511         }
 512
 513         /* check if should create a new subflow */
 514         if (msk->pm.local_addr_used < local_addr_max &&
 515             msk->pm.subflows < subflows_max &&
 516             !READ_ONCE(msk->pm.remote_deny_join_id0)) {
 517                 local = select_local_address(pernet, msk);
 518                 if (local) {
 519                         bool fullmesh = !!(local->flags & MPTCP_PM_ADDR_FLAG_FULLMESH);
 520                         struct mptcp_addr_info addrs[MPTCP_PM_ADDR_MAX];
 521                         int i, nr;
 522
 523                         msk->pm.local_addr_used++;
 524                         check_work_pending(msk);
 525                         nr = fill_remote_addresses_vec(msk, fullmesh, addrs);
 526                         spin_unlock_bh(&msk->pm.lock);
 527                         for (i = 0; i < nr; i++)
 528                                 __mptcp_subflow_connect(sk, &local->addr, &addrs[i]);
 529                         spin_lock_bh(&msk->pm.lock);
 530                         return;
 531                 }
 532
 533                 /* lookup failed, avoid fourther attempts later */
 534                 msk->pm.local_addr_used = local_addr_max;
 535                 check_work_pending(msk);
 536         }
 537 }
 538
 539 static void mptcp_pm_nl_fully_established(struct mptcp_sock *msk)
 540 {
 541         mptcp_pm_create_subflow_or_signal_addr(msk);
 542 }
 543
 544 static void mptcp_pm_nl_subflow_established(struct mptcp_sock *msk)
 545 {
 546         mptcp_pm_create_subflow_or_signal_addr(msk);
 547 }
 548
 549 /* Fill all the local addresses into the array addrs[],
 550  * and return the array size.
 551  */
 552 static unsigned int fill_local_addresses_vec(struct mptcp_sock *msk,
 553                                              struct mptcp_addr_info *addrs)
 554 {
 555         struct sock *sk = (struct sock *)msk;
 556         struct mptcp_pm_addr_entry *entry;
 557         struct mptcp_addr_info local;
 558         struct pm_nl_pernet *pernet;
 559         unsigned int subflows_max;
 560         int i = 0;
 561
 562         pernet = net_generic(sock_net(sk), pm_nl_pernet_id);
 563         subflows_max = mptcp_pm_get_subflows_max(msk);
 564
 565         rcu_read_lock();
 566         __mptcp_flush_join_list(msk);
 567         list_for_each_entry_rcu(entry, &pernet->local_addr_list, list) {
 568                 if (!(entry->flags & MPTCP_PM_ADDR_FLAG_FULLMESH))
 569                         continue;
 570
 571                 if (entry->addr.family != sk->sk_family) {
 572 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
 573                         if ((entry->addr.family == AF_INET &&
 574                              !ipv6_addr_v4mapped(&sk->sk_v6_daddr)) ||
 575                             (sk->sk_family == AF_INET &&
 576                              !ipv6_addr_v4mapped(&entry->addr.addr6)))
 577 #endif
 578                                 continue;
 579                 }
 580
 581                 if (msk->pm.subflows < subflows_max) {
 582                         msk->pm.subflows++;
 583                         addrs[i++] = entry->addr;
 584                 }
 585         }
 586         rcu_read_unlock();
 587
 588         /* If the array is empty, fill in the single
 589          * 'IPADDRANY' local address
 590          */
 591         if (!i) {
 592                 memset(&local, 0, sizeof(local));
 593                 local.family = msk->pm.remote.family;
 594
 595                 msk->pm.subflows++;
 596                 addrs[i++] = local;
 597         }
 598
 599         return i;
 600 }
 601
 602 static void mptcp_pm_nl_add_addr_received(struct mptcp_sock *msk)
 603 {
 604         struct mptcp_addr_info addrs[MPTCP_PM_ADDR_MAX];
 605         struct sock *sk = (struct sock *)msk;
 606         unsigned int add_addr_accept_max;
 607         struct mptcp_addr_info remote;
 608         unsigned int subflows_max;
 609         bool reset_port = false;
 610         int i, nr;
 611
 612         add_addr_accept_max = mptcp_pm_get_add_addr_accept_max(msk);
 613         subflows_max = mptcp_pm_get_subflows_max(msk);
 614
 615         pr_debug("accepted %d:%d remote family %d",
 616                  msk->pm.add_addr_accepted, add_addr_accept_max,
 617                  msk->pm.remote.family);
 618
 619         remote = msk->pm.remote;
 620         if (lookup_subflow_by_daddr(&msk->conn_list, &remote))
 621                 goto add_addr_echo;
 622
 623         /* pick id 0 port, if none is provided the remote address */
 624         if (!remote.port) {
 625                 reset_port = true;
 626                 remote.port = sk->sk_dport;
 627         }
 628
 629         /* connect to the specified remote address, using whatever
 630          * local address the routing configuration will pick.
 631          */
 632         nr = fill_local_addresses_vec(msk, addrs);
 633
 634         msk->pm.add_addr_accepted++;
 635         if (msk->pm.add_addr_accepted >= add_addr_accept_max ||
 636             msk->pm.subflows >= subflows_max)
 637                 WRITE_ONCE(msk->pm.accept_addr, false);
 638
 639         spin_unlock_bh(&msk->pm.lock);
 640         for (i = 0; i < nr; i++)
 641                 __mptcp_subflow_connect(sk, &addrs[i], &remote);
 642         spin_lock_bh(&msk->pm.lock);
 643
 644         /* be sure to echo exactly the received address */
 645         if (reset_port)
 646                 remote.port = 0;
 647
 648 add_addr_echo:
 649         mptcp_pm_announce_addr(msk, &remote, true);
 650         mptcp_pm_nl_addr_send_ack(msk);
 651 }
 652
 653 void mptcp_pm_nl_addr_send_ack(struct mptcp_sock *msk)
 654 {
 655         struct mptcp_subflow_context *subflow;
 656
 657         msk_owned_by_me(msk);
 658         lockdep_assert_held(&msk->pm.lock);
 659
 660         if (!mptcp_pm_should_add_signal(msk) &&
 661             !mptcp_pm_should_rm_signal(msk))
 662                 return;
 663
 664         __mptcp_flush_join_list(msk);
 665         subflow = list_first_entry_or_null(&msk->conn_list, typeof(*subflow), node);
 666         if (subflow) {
 667                 struct sock *ssk = mptcp_subflow_tcp_sock(subflow);
 668
 669                 spin_unlock_bh(&msk->pm.lock);
 670                 pr_debug("send ack for %s",
 671                          mptcp_pm_should_add_signal(msk) ? "add_addr" : "rm_addr");
 672
 673                 mptcp_subflow_send_ack(ssk);
 674                 spin_lock_bh(&msk->pm.lock);
 675         }
 676 }
 677
 678 int mptcp_pm_nl_mp_prio_send_ack(struct mptcp_sock *msk,
 679                                  struct mptcp_addr_info *addr,
 680                                  u8 bkup)
 681 {
 682         struct mptcp_subflow_context *subflow;
 683
 684         pr_debug("bkup=%d", bkup);
 685
 686         mptcp_for_each_subflow(msk, subflow) {
 687                 struct sock *ssk = mptcp_subflow_tcp_sock(subflow);
 688                 struct sock *sk = (struct sock *)msk;
 689                 struct mptcp_addr_info local;
 690
 691                 local_address((struct sock_common *)ssk, &local);
 692                 if (!addresses_equal(&local, addr, addr->port))
 693                         continue;
 694
 695                 if (subflow->backup != bkup)
 696                         msk->last_snd = NULL;
 697                 subflow->backup = bkup;
 698                 subflow->send_mp_prio = 1;
 699                 subflow->request_bkup = bkup;
 700                 __MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_MPPRIOTX);
 701
 702                 spin_unlock_bh(&msk->pm.lock);
 703                 pr_debug("send ack for mp_prio");
 704                 mptcp_subflow_send_ack(ssk);
 705                 spin_lock_bh(&msk->pm.lock);
 706
 707                 return 0;
 708         }
 709
 710         return -EINVAL;
 711 }
 712
 713 static void mptcp_pm_nl_rm_addr_or_subflow(struct mptcp_sock *msk,
 714                                            const struct mptcp_rm_list *rm_list,
 715                                            enum linux_mptcp_mib_field rm_type)
 716 {
 717         struct mptcp_subflow_context *subflow, *tmp;
 718         struct sock *sk = (struct sock *)msk;
 719         u8 i;
 720
 721         pr_debug("%s rm_list_nr %d",
 722                  rm_type == MPTCP_MIB_RMADDR ? "address" : "subflow", rm_list->nr);
 723
 724         msk_owned_by_me(msk);
 725
 726         if (sk->sk_state == TCP_LISTEN)
 727                 return;
 728
 729         if (!rm_list->nr)
 730                 return;
 731
 732         if (list_empty(&msk->conn_list))
 733                 return;
 734
 735         for (i = 0; i < rm_list->nr; i++) {
 736                 bool removed = false;
 737
 738                 list_for_each_entry_safe(subflow, tmp, &msk->conn_list, node) {
 739                         struct sock *ssk = mptcp_subflow_tcp_sock(subflow);
 740                         int how = RCV_SHUTDOWN | SEND_SHUTDOWN;
 741                         u8 id = subflow->local_id;
 742
 743                         if (rm_type == MPTCP_MIB_RMADDR)
 744                                 id = subflow->remote_id;
 745
 746                         if (rm_list->ids[i] != id)
 747                                 continue;
 748
 749                         pr_debug(" -> %s rm_list_ids[%d]=%u local_id=%u remote_id=%u",
 750                                  rm_type == MPTCP_MIB_RMADDR ? "address" : "subflow",
 751                                  i, rm_list->ids[i], subflow->local_id, subflow->remote_id);
 752                         spin_unlock_bh(&msk->pm.lock);
 753                         mptcp_subflow_shutdown(sk, ssk, how);
 754                         mptcp_close_ssk(sk, ssk, subflow);
 755                         spin_lock_bh(&msk->pm.lock);
 756
 757                         removed = true;
 758                         msk->pm.subflows--;
 759                         __MPTCP_INC_STATS(sock_net(sk), rm_type);
 760                 }
 761                 if (!removed)
 762                         continue;
 763
 764                 if (rm_type == MPTCP_MIB_RMADDR) {
 765                         msk->pm.add_addr_accepted--;
 766                         WRITE_ONCE(msk->pm.accept_addr, true);
 767                 } else if (rm_type == MPTCP_MIB_RMSUBFLOW) {
 768                         msk->pm.local_addr_used--;
 769                 }
 770         }
 771 }
 772
 773 static void mptcp_pm_nl_rm_addr_received(struct mptcp_sock *msk)
 774 {
 775         mptcp_pm_nl_rm_addr_or_subflow(msk, &msk->pm.rm_list_rx, MPTCP_MIB_RMADDR);
 776 }
 777
 778 void mptcp_pm_nl_rm_subflow_received(struct mptcp_sock *msk,
 779                                      const struct mptcp_rm_list *rm_list)
 780 {
 781         mptcp_pm_nl_rm_addr_or_subflow(msk, rm_list, MPTCP_MIB_RMSUBFLOW);
 782 }
 783
 784 void mptcp_pm_nl_work(struct mptcp_sock *msk)
 785 {
 786         struct mptcp_pm_data *pm = &msk->pm;
 787
 788         msk_owned_by_me(msk);
 789
 790         spin_lock_bh(&msk->pm.lock);
 791
 792         pr_debug("msk=%p status=%x", msk, pm->status);
 793         if (pm->status & BIT(MPTCP_PM_ADD_ADDR_RECEIVED)) {
 794                 pm->status &= ~BIT(MPTCP_PM_ADD_ADDR_RECEIVED);
 795                 mptcp_pm_nl_add_addr_received(msk);
 796         }
 797         if (pm->status & BIT(MPTCP_PM_ADD_ADDR_SEND_ACK)) {
 798                 pm->status &= ~BIT(MPTCP_PM_ADD_ADDR_SEND_ACK);
 799                 mptcp_pm_nl_addr_send_ack(msk);
 800         }
 801         if (pm->status & BIT(MPTCP_PM_RM_ADDR_RECEIVED)) {
 802                 pm->status &= ~BIT(MPTCP_PM_RM_ADDR_RECEIVED);
 803                 mptcp_pm_nl_rm_addr_received(msk);
 804         }
 805         if (pm->status & BIT(MPTCP_PM_ESTABLISHED)) {
 806                 pm->status &= ~BIT(MPTCP_PM_ESTABLISHED);
 807                 mptcp_pm_nl_fully_established(msk);
 808         }
 809         if (pm->status & BIT(MPTCP_PM_SUBFLOW_ESTABLISHED)) {
 810                 pm->status &= ~BIT(MPTCP_PM_SUBFLOW_ESTABLISHED);
 811                 mptcp_pm_nl_subflow_established(msk);
 812         }
 813
 814         spin_unlock_bh(&msk->pm.lock);
 815 }
 816
 817 static bool address_use_port(struct mptcp_pm_addr_entry *entry)
 818 {
 819         return (entry->flags &
 820                 (MPTCP_PM_ADDR_FLAG_SIGNAL | MPTCP_PM_ADDR_FLAG_SUBFLOW)) ==
 821                 MPTCP_PM_ADDR_FLAG_SIGNAL;
 822 }
 823
 824 static int mptcp_pm_nl_append_new_local_addr(struct pm_nl_pernet *pernet,
 825                                              struct mptcp_pm_addr_entry *entry)
 826 {
 827         struct mptcp_pm_addr_entry *cur;
 828         unsigned int addr_max;
 829         int ret = -EINVAL;
 830
 831         spin_lock_bh(&pernet->lock);
 832         /* to keep the code simple, don't do IDR-like allocation for address ID,
 833          * just bail when we exceed limits
 834          */
 835         if (pernet->next_id == MAX_ADDR_ID)
 836                 pernet->next_id = 1;
 837         if (pernet->addrs >= MPTCP_PM_ADDR_MAX)
 838                 goto out;
 839         if (test_bit(entry->addr.id, pernet->id_bitmap))
 840                 goto out;
 841
 842         /* do not insert duplicate address, differentiate on port only
 843          * singled addresses
 844          */
 845         list_for_each_entry(cur, &pernet->local_addr_list, list) {
 846                 if (addresses_equal(&cur->addr, &entry->addr,
 847                                     address_use_port(entry) &&
 848                                     address_use_port(cur)))
 849                         goto out;
 850         }
 851
 852         if (!entry->addr.id) {
 853 find_next:
 854                 entry->addr.id = find_next_zero_bit(pernet->id_bitmap,
 855                                                     MAX_ADDR_ID + 1,
 856                                                     pernet->next_id);
 857                 if ((!entry->addr.id || entry->addr.id > MAX_ADDR_ID) &&
 858                     pernet->next_id != 1) {
 859                         pernet->next_id = 1;
 860                         goto find_next;
 861                 }
 862         }
 863
 864         if (!entry->addr.id || entry->addr.id > MAX_ADDR_ID)
 865                 goto out;
 866
 867         __set_bit(entry->addr.id, pernet->id_bitmap);
 868         if (entry->addr.id > pernet->next_id)
 869                 pernet->next_id = entry->addr.id;
 870
 871         if (entry->flags & MPTCP_PM_ADDR_FLAG_SIGNAL) {
 872                 addr_max = pernet->add_addr_signal_max;
 873                 WRITE_ONCE(pernet->add_addr_signal_max, addr_max + 1);
 874         }
 875         if (entry->flags & MPTCP_PM_ADDR_FLAG_SUBFLOW) {
 876                 addr_max = pernet->local_addr_max;
 877                 WRITE_ONCE(pernet->local_addr_max, addr_max + 1);
 878         }
 879
 880         pernet->addrs++;
 881         list_add_tail_rcu(&entry->list, &pernet->local_addr_list);
 882         ret = entry->addr.id;
 883
 884 out:
 885         spin_unlock_bh(&pernet->lock);
 886         return ret;
 887 }
 888
 889 static int mptcp_pm_nl_create_listen_socket(struct sock *sk,
 890                                             struct mptcp_pm_addr_entry *entry)
 891 {
 892         int addrlen = sizeof(struct sockaddr_in);
 893         struct sockaddr_storage addr;
 894         struct mptcp_sock *msk;
 895         struct socket *ssock;
 896         int backlog = 1024;
 897         int err;
 898
 899         err = sock_create_kern(sock_net(sk), entry->addr.family,
 900                                SOCK_STREAM, IPPROTO_MPTCP, &entry->lsk);
 901         if (err)
 902                 return err;
 903
 904         msk = mptcp_sk(entry->lsk->sk);
 905         if (!msk) {
 906                 err = -EINVAL;
 907                 goto out;
 908         }
 909
 910         ssock = __mptcp_nmpc_socket(msk);
 911         if (!ssock) {
 912                 err = -EINVAL;
 913                 goto out;
 914         }
 915
 916         mptcp_info2sockaddr(&entry->addr, &addr, entry->addr.family);
 917 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
 918         if (entry->addr.family == AF_INET6)
 919                 addrlen = sizeof(struct sockaddr_in6);
 920 #endif
 921         err = kernel_bind(ssock, (struct sockaddr *)&addr, addrlen);
 922         if (err) {
 923                 pr_warn("kernel_bind error, err=%d", err);
 924                 goto out;
 925         }
 926
 927         err = kernel_listen(ssock, backlog);
 928         if (err) {
 929                 pr_warn("kernel_listen error, err=%d", err);
 930                 goto out;
 931         }
 932
 933         return 0;
 934
 935 out:
 936         sock_release(entry->lsk);
 937         return err;
 938 }
 939
 940 int mptcp_pm_nl_get_local_id(struct mptcp_sock *msk, struct sock_common *skc)
 941 {
 942         struct mptcp_pm_addr_entry *entry;
 943         struct mptcp_addr_info skc_local;
 944         struct mptcp_addr_info msk_local;
 945         struct pm_nl_pernet *pernet;
 946         int ret = -1;
 947
 948         if (WARN_ON_ONCE(!msk))
 949                 return -1;
 950
 951         /* The 0 ID mapping is defined by the first subflow, copied into the msk
 952          * addr
 953          */
 954         local_address((struct sock_common *)msk, &msk_local);
 955         local_address((struct sock_common *)skc, &skc_local);
 956         if (addresses_equal(&msk_local, &skc_local, false))
 957                 return 0;
 958
 959         if (address_zero(&skc_local))
 960                 return 0;
 961
 962         pernet = net_generic(sock_net((struct sock *)msk), pm_nl_pernet_id);
 963
 964         rcu_read_lock();
 965         list_for_each_entry_rcu(entry, &pernet->local_addr_list, list) {
 966                 if (addresses_equal(&entry->addr, &skc_local, entry->addr.port)) {
 967                         ret = entry->addr.id;
 968                         break;
 969                 }
 970         }
 971         rcu_read_unlock();
 972         if (ret >= 0)
 973                 return ret;
 974
 975         /* address not found, add to local list */
 976         entry = kmalloc(sizeof(*entry), GFP_ATOMIC);
 977         if (!entry)
 978                 return -ENOMEM;
 979
 980         entry->addr = skc_local;
 981         entry->addr.id = 0;
 982         entry->addr.port = 0;
 983         entry->ifindex = 0;
 984         entry->flags = 0;
 985         entry->lsk = NULL;
 986         ret = mptcp_pm_nl_append_new_local_addr(pernet, entry);
 987         if (ret < 0)
 988                 kfree(entry);
 989
 990         return ret;
 991 }
 992
 993 void mptcp_pm_nl_data_init(struct mptcp_sock *msk)
 994 {
 995         struct mptcp_pm_data *pm = &msk->pm;
 996         bool subflows;
 997
 998         subflows = !!mptcp_pm_get_subflows_max(msk);
 999         WRITE_ONCE(pm->work_pending, (!!mptcp_pm_get_local_addr_max(msk) && subflows) ||
1000                    !!mptcp_pm_get_add_addr_signal_max(msk));
1001         WRITE_ONCE(pm->accept_addr, !!mptcp_pm_get_add_addr_accept_max(msk) && subflows);
1002         WRITE_ONCE(pm->accept_subflow, subflows);
1003 }
1004
1005 #define MPTCP_PM_CMD_GRP_OFFSET       0
1006 #define MPTCP_PM_EV_GRP_OFFSET        1
1007
1008 static const struct genl_multicast_group mptcp_pm_mcgrps[] = {
1009         [MPTCP_PM_CMD_GRP_OFFSET]       = { .name = MPTCP_PM_CMD_GRP_NAME, },
1010         [MPTCP_PM_EV_GRP_OFFSET]        = { .name = MPTCP_PM_EV_GRP_NAME,
1011                                             .flags = GENL_UNS_ADMIN_PERM,
1012                                           },
1013 };
1014
1015 static const struct nla_policy
1016 mptcp_pm_addr_policy[MPTCP_PM_ADDR_ATTR_MAX + 1] = {
1017         [MPTCP_PM_ADDR_ATTR_FAMILY]     = { .type       = NLA_U16,      },
1018         [MPTCP_PM_ADDR_ATTR_ID]         = { .type       = NLA_U8,       },
1019         [MPTCP_PM_ADDR_ATTR_ADDR4]      = { .type       = NLA_U32,      },
1020         [MPTCP_PM_ADDR_ATTR_ADDR6]      =
1021                 NLA_POLICY_EXACT_LEN(sizeof(struct in6_addr)),
1022         [MPTCP_PM_ADDR_ATTR_PORT]       = { .type       = NLA_U16       },
1023         [MPTCP_PM_ADDR_ATTR_FLAGS]      = { .type       = NLA_U32       },
1024         [MPTCP_PM_ADDR_ATTR_IF_IDX]     = { .type       = NLA_S32       },
1025 };
1026
1027 static const struct nla_policy mptcp_pm_policy[MPTCP_PM_ATTR_MAX + 1] = {
1028         [MPTCP_PM_ATTR_ADDR]            =
1029                                         NLA_POLICY_NESTED(mptcp_pm_addr_policy),
1030         [MPTCP_PM_ATTR_RCV_ADD_ADDRS]   = { .type       = NLA_U32,      },
1031         [MPTCP_PM_ATTR_SUBFLOWS]        = { .type       = NLA_U32,      },
1032 };
1033
1034 void mptcp_pm_nl_subflow_chk_stale(const struct mptcp_sock *msk, struct sock *ssk)
1035 {
1036         struct mptcp_subflow_context *iter, *subflow = mptcp_subflow_ctx(ssk);
1037         struct sock *sk = (struct sock *)msk;
1038         unsigned int active_max_loss_cnt;
1039         struct net *net = sock_net(sk);
1040         unsigned int stale_loss_cnt;
1041         bool slow;
1042
1043         stale_loss_cnt = mptcp_stale_loss_cnt(net);
1044         if (subflow->stale || !stale_loss_cnt || subflow->stale_count <= stale_loss_cnt)
1045                 return;
1046
1047         /* look for another available subflow not in loss state */
1048         active_max_loss_cnt = max_t(int, stale_loss_cnt - 1, 1);
1049         mptcp_for_each_subflow(msk, iter) {
1050                 if (iter != subflow && mptcp_subflow_active(iter) &&
1051                     iter->stale_count < active_max_loss_cnt) {
1052                         /* we have some alternatives, try to mark this subflow as idle ...*/
1053                         slow = lock_sock_fast(ssk);
1054                         if (!tcp_rtx_and_write_queues_empty(ssk)) {
1055                                 subflow->stale = 1;
1056                                 __mptcp_retransmit_pending_data(sk);
1057                                 MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_SUBFLOWSTALE);
1058                         }
1059                         unlock_sock_fast(ssk, slow);
1060
1061                         /* always try to push the pending data regarless of re-injections:
1062                          * we can possibly use backup subflows now, and subflow selection
1063                          * is cheap under the msk socket lock
1064                          */
1065                         __mptcp_push_pending(sk, 0);
1066                         return;
1067                 }
1068         }
1069 }
1070
1071 static int mptcp_pm_family_to_addr(int family)
1072 {
1073 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
1074         if (family == AF_INET6)
1075                 return MPTCP_PM_ADDR_ATTR_ADDR6;
1076 #endif
1077         return MPTCP_PM_ADDR_ATTR_ADDR4;
1078 }
1079
1080 static int mptcp_pm_parse_addr(struct nlattr *attr, struct genl_info *info,
1081                                bool require_family,
1082                                struct mptcp_pm_addr_entry *entry)
1083 {
1084         struct nlattr *tb[MPTCP_PM_ADDR_ATTR_MAX + 1];
1085         int err, addr_addr;
1086
1087         if (!attr) {
1088                 GENL_SET_ERR_MSG(info, "missing address info");
1089                 return -EINVAL;
1090         }
1091
1092         /* no validation needed - was already done via nested policy */
1093         err = nla_parse_nested_deprecated(tb, MPTCP_PM_ADDR_ATTR_MAX, attr,
1094                                           mptcp_pm_addr_policy, info->extack);
1095         if (err)
1096                 return err;
1097
1098         memset(entry, 0, sizeof(*entry));
1099         if (!tb[MPTCP_PM_ADDR_ATTR_FAMILY]) {
1100                 if (!require_family)
1101                         goto skip_family;
1102
1103                 NL_SET_ERR_MSG_ATTR(info->extack, attr,
1104                                     "missing family");
1105                 return -EINVAL;
1106         }
1107
1108         entry->addr.family = nla_get_u16(tb[MPTCP_PM_ADDR_ATTR_FAMILY]);
1109         if (entry->addr.family != AF_INET
1110 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
1111             && entry->addr.family != AF_INET6
1112 #endif
1113             ) {
1114                 NL_SET_ERR_MSG_ATTR(info->extack, attr,
1115                                     "unknown address family");
1116                 return -EINVAL;
1117         }
1118         addr_addr = mptcp_pm_family_to_addr(entry->addr.family);
1119         if (!tb[addr_addr]) {
1120                 NL_SET_ERR_MSG_ATTR(info->extack, attr,
1121                                     "missing address data");
1122                 return -EINVAL;
1123         }
1124
1125 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
1126         if (entry->addr.family == AF_INET6)
1127                 entry->addr.addr6 = nla_get_in6_addr(tb[addr_addr]);
1128         else
1129 #endif
1130                 entry->addr.addr.s_addr = nla_get_in_addr(tb[addr_addr]);
1131
1132 skip_family:
1133         if (tb[MPTCP_PM_ADDR_ATTR_IF_IDX]) {
1134                 u32 val = nla_get_s32(tb[MPTCP_PM_ADDR_ATTR_IF_IDX]);
1135
1136                 entry->ifindex = val;
1137         }
1138
1139         if (tb[MPTCP_PM_ADDR_ATTR_ID])
1140                 entry->addr.id = nla_get_u8(tb[MPTCP_PM_ADDR_ATTR_ID]);
1141
1142         if (tb[MPTCP_PM_ADDR_ATTR_FLAGS])
1143                 entry->flags = nla_get_u32(tb[MPTCP_PM_ADDR_ATTR_FLAGS]);
1144
1145         if (tb[MPTCP_PM_ADDR_ATTR_PORT]) {
1146                 if (!(entry->flags & MPTCP_PM_ADDR_FLAG_SIGNAL)) {
1147                         NL_SET_ERR_MSG_ATTR(info->extack, attr,
1148                                             "flags must have signal when using port");
1149                         return -EINVAL;
1150                 }
1151                 entry->addr.port = htons(nla_get_u16(tb[MPTCP_PM_ADDR_ATTR_PORT]));
1152         }
1153
1154         return 0;
1155 }
1156
1157 static struct pm_nl_pernet *genl_info_pm_nl(struct genl_info *info)
1158 {
1159         return net_generic(genl_info_net(info), pm_nl_pernet_id);
1160 }
1161
1162 static int mptcp_nl_add_subflow_or_signal_addr(struct net *net)
1163 {
1164         struct mptcp_sock *msk;
1165         long s_slot = 0, s_num = 0;
1166
1167         while ((msk = mptcp_token_iter_next(net, &s_slot, &s_num)) != NULL) {
1168                 struct sock *sk = (struct sock *)msk;
1169
1170                 if (!READ_ONCE(msk->fully_established))
1171                         goto next;
1172
1173                 lock_sock(sk);
1174                 spin_lock_bh(&msk->pm.lock);
1175                 mptcp_pm_create_subflow_or_signal_addr(msk);
1176                 spin_unlock_bh(&msk->pm.lock);
1177                 release_sock(sk);
1178
1179 next:
1180                 sock_put(sk);
1181                 cond_resched();
1182         }
1183
1184         return 0;
1185 }
1186
1187 static int mptcp_nl_cmd_add_addr(struct sk_buff *skb, struct genl_info *info)
1188 {
1189         struct nlattr *attr = info->attrs[MPTCP_PM_ATTR_ADDR];
1190         struct pm_nl_pernet *pernet = genl_info_pm_nl(info);
1191         struct mptcp_pm_addr_entry addr, *entry;
1192         int ret;
1193
1194         ret = mptcp_pm_parse_addr(attr, info, true, &addr);
1195         if (ret < 0)
1196                 return ret;
1197
1198         entry = kmalloc(sizeof(*entry), GFP_KERNEL);
1199         if (!entry) {
1200                 GENL_SET_ERR_MSG(info, "can't allocate addr");
1201                 return -ENOMEM;
1202         }
1203
1204         *entry = addr;
1205         if (entry->addr.port) {
1206                 ret = mptcp_pm_nl_create_listen_socket(skb->sk, entry);
1207                 if (ret) {
1208                         GENL_SET_ERR_MSG(info, "create listen socket error");
1209                         kfree(entry);
1210                         return ret;
1211                 }
1212         }
1213         ret = mptcp_pm_nl_append_new_local_addr(pernet, entry);
1214         if (ret < 0) {
1215                 GENL_SET_ERR_MSG(info, "too many addresses or duplicate one");
1216                 if (entry->lsk)
1217                         sock_release(entry->lsk);
1218                 kfree(entry);
1219                 return ret;
1220         }
1221
1222         mptcp_nl_add_subflow_or_signal_addr(sock_net(skb->sk));
1223
1224         return 0;
1225 }
1226
1227 static struct mptcp_pm_addr_entry *
1228 __lookup_addr_by_id(struct pm_nl_pernet *pernet, unsigned int id)
1229 {
1230         struct mptcp_pm_addr_entry *entry;
1231
1232         list_for_each_entry(entry, &pernet->local_addr_list, list) {
1233                 if (entry->addr.id == id)
1234                         return entry;
1235         }
1236         return NULL;
1237 }
1238
1239 int mptcp_pm_get_flags_and_ifindex_by_id(struct net *net, unsigned int id,
1240                                          u8 *flags, int *ifindex)
1241 {
1242         struct mptcp_pm_addr_entry *entry;
1243
1244         *flags = 0;
1245         *ifindex = 0;
1246
1247         if (id) {
1248                 rcu_read_lock();
1249                 entry = __lookup_addr_by_id(net_generic(net, pm_nl_pernet_id), id);
1250                 if (entry) {
1251                         *flags = entry->flags;
1252                         *ifindex = entry->ifindex;
1253                 }
1254                 rcu_read_unlock();
1255         }
1256
1257         return 0;
1258 }
1259
1260 static bool remove_anno_list_by_saddr(struct mptcp_sock *msk,
1261                                       struct mptcp_addr_info *addr)
1262 {
1263         struct mptcp_pm_add_entry *entry;
1264
1265         entry = mptcp_pm_del_add_timer(msk, addr, false);
1266         if (entry) {
1267                 list_del(&entry->list);
1268                 kfree(entry);
1269                 return true;
1270         }
1271
1272         return false;
1273 }
1274
1275 static bool mptcp_pm_remove_anno_addr(struct mptcp_sock *msk,
1276                                       struct mptcp_addr_info *addr,
1277                                       bool force)
1278 {
1279         struct mptcp_rm_list list = { .nr = 0 };
1280         bool ret;
1281
1282         list.ids[list.nr++] = addr->id;
1283
1284         ret = remove_anno_list_by_saddr(msk, addr);
1285         if (ret || force) {
1286                 spin_lock_bh(&msk->pm.lock);
1287                 mptcp_pm_remove_addr(msk, &list);
1288                 spin_unlock_bh(&msk->pm.lock);
1289         }
1290         return ret;
1291 }
1292
1293 static int mptcp_nl_remove_subflow_and_signal_addr(struct net *net,
1294                                                    struct mptcp_addr_info *addr)
1295 {
1296         struct mptcp_sock *msk;
1297         long s_slot = 0, s_num = 0;
1298         struct mptcp_rm_list list = { .nr = 0 };
1299
1300         pr_debug("remove_id=%d", addr->id);
1301
1302         list.ids[list.nr++] = addr->id;
1303
1304         while ((msk = mptcp_token_iter_next(net, &s_slot, &s_num)) != NULL) {
1305                 struct sock *sk = (struct sock *)msk;
1306                 bool remove_subflow;
1307
1308                 if (list_empty(&msk->conn_list)) {
1309                         mptcp_pm_remove_anno_addr(msk, addr, false);
1310                         goto next;
1311                 }
1312
1313                 lock_sock(sk);
1314                 remove_subflow = lookup_subflow_by_saddr(&msk->conn_list, addr);
1315                 mptcp_pm_remove_anno_addr(msk, addr, remove_subflow);
1316                 if (remove_subflow)
1317                         mptcp_pm_remove_subflow(msk, &list);
1318                 release_sock(sk);
1319
1320 next:
1321                 sock_put(sk);
1322                 cond_resched();
1323         }
1324
1325         return 0;
1326 }
1327
1328 /* caller must ensure the RCU grace period is already elapsed */
1329 static void __mptcp_pm_release_addr_entry(struct mptcp_pm_addr_entry *entry)
1330 {
1331         if (entry->lsk)
1332                 sock_release(entry->lsk);
1333         kfree(entry);
1334 }
1335
1336 static int mptcp_nl_remove_id_zero_address(struct net *net,
1337                                            struct mptcp_addr_info *addr)
1338 {
1339         struct mptcp_rm_list list = { .nr = 0 };
1340         long s_slot = 0, s_num = 0;
1341         struct mptcp_sock *msk;
1342
1343         list.ids[list.nr++] = 0;
1344
1345         while ((msk = mptcp_token_iter_next(net, &s_slot, &s_num)) != NULL) {
1346                 struct sock *sk = (struct sock *)msk;
1347                 struct mptcp_addr_info msk_local;
1348
1349                 if (list_empty(&msk->conn_list))
1350                         goto next;
1351
1352                 local_address((struct sock_common *)msk, &msk_local);
1353                 if (!addresses_equal(&msk_local, addr, addr->port))
1354                         goto next;
1355
1356                 lock_sock(sk);
1357                 spin_lock_bh(&msk->pm.lock);
1358                 mptcp_pm_remove_addr(msk, &list);
1359                 mptcp_pm_nl_rm_subflow_received(msk, &list);
1360                 spin_unlock_bh(&msk->pm.lock);
1361                 release_sock(sk);
1362
1363 next:
1364                 sock_put(sk);
1365                 cond_resched();
1366         }
1367
1368         return 0;
1369 }
1370
1371 static int mptcp_nl_cmd_del_addr(struct sk_buff *skb, struct genl_info *info)
1372 {
1373         struct nlattr *attr = info->attrs[MPTCP_PM_ATTR_ADDR];
1374         struct pm_nl_pernet *pernet = genl_info_pm_nl(info);
1375         struct mptcp_pm_addr_entry addr, *entry;
1376         unsigned int addr_max;
1377         int ret;
1378
1379         ret = mptcp_pm_parse_addr(attr, info, false, &addr);
1380         if (ret < 0)
1381                 return ret;
1382
1383         /* the zero id address is special: the first address used by the msk
1384          * always gets such an id, so different subflows can have different zero
1385          * id addresses. Additionally zero id is not accounted for in id_bitmap.
1386          * Let's use an 'mptcp_rm_list' instead of the common remove code.
1387          */
1388         if (addr.addr.id == 0)
1389                 return mptcp_nl_remove_id_zero_address(sock_net(skb->sk), &addr.addr);
1390
1391         spin_lock_bh(&pernet->lock);
1392         entry = __lookup_addr_by_id(pernet, addr.addr.id);
1393         if (!entry) {
1394                 GENL_SET_ERR_MSG(info, "address not found");
1395                 spin_unlock_bh(&pernet->lock);
1396                 return -EINVAL;
1397         }
1398         if (entry->flags & MPTCP_PM_ADDR_FLAG_SIGNAL) {
1399                 addr_max = pernet->add_addr_signal_max;
1400                 WRITE_ONCE(pernet->add_addr_signal_max, addr_max - 1);
1401         }
1402         if (entry->flags & MPTCP_PM_ADDR_FLAG_SUBFLOW) {
1403                 addr_max = pernet->local_addr_max;
1404                 WRITE_ONCE(pernet->local_addr_max, addr_max - 1);
1405         }
1406
1407         pernet->addrs--;
1408         list_del_rcu(&entry->list);
1409         __clear_bit(entry->addr.id, pernet->id_bitmap);
1410         spin_unlock_bh(&pernet->lock);
1411
1412         mptcp_nl_remove_subflow_and_signal_addr(sock_net(skb->sk), &entry->addr);
1413         synchronize_rcu();
1414         __mptcp_pm_release_addr_entry(entry);
1415
1416         return ret;
1417 }
1418
1419 static void mptcp_pm_remove_addrs_and_subflows(struct mptcp_sock *msk,
1420                                                struct list_head *rm_list)
1421 {
1422         struct mptcp_rm_list alist = { .nr = 0 }, slist = { .nr = 0 };
1423         struct mptcp_pm_addr_entry *entry;
1424
1425         list_for_each_entry(entry, rm_list, list) {
1426                 if (lookup_subflow_by_saddr(&msk->conn_list, &entry->addr) &&
1427                     alist.nr < MPTCP_RM_IDS_MAX &&
1428                     slist.nr < MPTCP_RM_IDS_MAX) {
1429                         alist.ids[alist.nr++] = entry->addr.id;
1430                         slist.ids[slist.nr++] = entry->addr.id;
1431                 } else if (remove_anno_list_by_saddr(msk, &entry->addr) &&
1432                          alist.nr < MPTCP_RM_IDS_MAX) {
1433                         alist.ids[alist.nr++] = entry->addr.id;
1434                 }
1435         }
1436
1437         if (alist.nr) {
1438                 spin_lock_bh(&msk->pm.lock);
1439                 mptcp_pm_remove_addr(msk, &alist);
1440                 spin_unlock_bh(&msk->pm.lock);
1441         }
1442         if (slist.nr)
1443                 mptcp_pm_remove_subflow(msk, &slist);
1444 }
1445
1446 static void mptcp_nl_remove_addrs_list(struct net *net,
1447                                        struct list_head *rm_list)
1448 {
1449         long s_slot = 0, s_num = 0;
1450         struct mptcp_sock *msk;
1451
1452         if (list_empty(rm_list))
1453                 return;
1454
1455         while ((msk = mptcp_token_iter_next(net, &s_slot, &s_num)) != NULL) {
1456                 struct sock *sk = (struct sock *)msk;
1457
1458                 lock_sock(sk);
1459                 mptcp_pm_remove_addrs_and_subflows(msk, rm_list);
1460                 release_sock(sk);
1461
1462                 sock_put(sk);
1463                 cond_resched();
1464         }
1465 }
1466
1467 /* caller must ensure the RCU grace period is already elapsed */
1468 static void __flush_addrs(struct list_head *list)
1469 {
1470         while (!list_empty(list)) {
1471                 struct mptcp_pm_addr_entry *cur;
1472
1473                 cur = list_entry(list->next,
1474                                  struct mptcp_pm_addr_entry, list);
1475                 list_del_rcu(&cur->list);
1476                 __mptcp_pm_release_addr_entry(cur);
1477         }
1478 }
1479
1480 static void __reset_counters(struct pm_nl_pernet *pernet)
1481 {
1482         WRITE_ONCE(pernet->add_addr_signal_max, 0);
1483         WRITE_ONCE(pernet->add_addr_accept_max, 0);
1484         WRITE_ONCE(pernet->local_addr_max, 0);
1485         pernet->addrs = 0;
1486 }
1487
1488 static int mptcp_nl_cmd_flush_addrs(struct sk_buff *skb, struct genl_info *info)
1489 {
1490         struct pm_nl_pernet *pernet = genl_info_pm_nl(info);
1491         LIST_HEAD(free_list);
1492
1493         spin_lock_bh(&pernet->lock);
1494         list_splice_init(&pernet->local_addr_list, &free_list);
1495         __reset_counters(pernet);
1496         pernet->next_id = 1;
1497         bitmap_zero(pernet->id_bitmap, MAX_ADDR_ID + 1);
1498         spin_unlock_bh(&pernet->lock);
1499         mptcp_nl_remove_addrs_list(sock_net(skb->sk), &free_list);
1500         synchronize_rcu();
1501         __flush_addrs(&free_list);
1502         return 0;
1503 }
1504
1505 static int mptcp_nl_fill_addr(struct sk_buff *skb,
1506                               struct mptcp_pm_addr_entry *entry)
1507 {
1508         struct mptcp_addr_info *addr = &entry->addr;
1509         struct nlattr *attr;
1510
1511         attr = nla_nest_start(skb, MPTCP_PM_ATTR_ADDR);
1512         if (!attr)
1513                 return -EMSGSIZE;
1514
1515         if (nla_put_u16(skb, MPTCP_PM_ADDR_ATTR_FAMILY, addr->family))
1516                 goto nla_put_failure;
1517         if (nla_put_u16(skb, MPTCP_PM_ADDR_ATTR_PORT, ntohs(addr->port)))
1518                 goto nla_put_failure;
1519         if (nla_put_u8(skb, MPTCP_PM_ADDR_ATTR_ID, addr->id))
1520                 goto nla_put_failure;
1521         if (nla_put_u32(skb, MPTCP_PM_ADDR_ATTR_FLAGS, entry->flags))
1522                 goto nla_put_failure;
1523         if (entry->ifindex &&
1524             nla_put_s32(skb, MPTCP_PM_ADDR_ATTR_IF_IDX, entry->ifindex))
1525                 goto nla_put_failure;
1526
1527         if (addr->family == AF_INET &&
1528             nla_put_in_addr(skb, MPTCP_PM_ADDR_ATTR_ADDR4,
1529                             addr->addr.s_addr))
1530                 goto nla_put_failure;
1531 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
1532         else if (addr->family == AF_INET6 &&
1533                  nla_put_in6_addr(skb, MPTCP_PM_ADDR_ATTR_ADDR6, &addr->addr6))
1534                 goto nla_put_failure;
1535 #endif
1536         nla_nest_end(skb, attr);
1537         return 0;
1538
1539 nla_put_failure:
1540         nla_nest_cancel(skb, attr);
1541         return -EMSGSIZE;
1542 }
1543
1544 static int mptcp_nl_cmd_get_addr(struct sk_buff *skb, struct genl_info *info)
1545 {
1546         struct nlattr *attr = info->attrs[MPTCP_PM_ATTR_ADDR];
1547         struct pm_nl_pernet *pernet = genl_info_pm_nl(info);
1548         struct mptcp_pm_addr_entry addr, *entry;
1549         struct sk_buff *msg;
1550         void *reply;
1551         int ret;
1552
1553         ret = mptcp_pm_parse_addr(attr, info, false, &addr);
1554         if (ret < 0)
1555                 return ret;
1556
1557         msg = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
1558         if (!msg)
1559                 return -ENOMEM;
1560
1561         reply = genlmsg_put_reply(msg, info, &mptcp_genl_family, 0,
1562                                   info->genlhdr->cmd);
1563         if (!reply) {
1564                 GENL_SET_ERR_MSG(info, "not enough space in Netlink message");
1565                 ret = -EMSGSIZE;
1566                 goto fail;
1567         }
1568
1569         spin_lock_bh(&pernet->lock);
1570         entry = __lookup_addr_by_id(pernet, addr.addr.id);
1571         if (!entry) {
1572                 GENL_SET_ERR_MSG(info, "address not found");
1573                 ret = -EINVAL;
1574                 goto unlock_fail;
1575         }
1576
1577         ret = mptcp_nl_fill_addr(msg, entry);
1578         if (ret)
1579                 goto unlock_fail;
1580
1581         genlmsg_end(msg, reply);
1582         ret = genlmsg_reply(msg, info);
1583         spin_unlock_bh(&pernet->lock);
1584         return ret;
1585
1586 unlock_fail:
1587         spin_unlock_bh(&pernet->lock);
1588
1589 fail:
1590         nlmsg_free(msg);
1591         return ret;
1592 }
1593
1594 static int mptcp_nl_cmd_dump_addrs(struct sk_buff *msg,
1595                                    struct netlink_callback *cb)
1596 {
1597         struct net *net = sock_net(msg->sk);
1598         struct mptcp_pm_addr_entry *entry;
1599         struct pm_nl_pernet *pernet;
1600         int id = cb->args[0];
1601         void *hdr;
1602         int i;
1603
1604         pernet = net_generic(net, pm_nl_pernet_id);
1605
1606         spin_lock_bh(&pernet->lock);
1607         for (i = id; i < MAX_ADDR_ID + 1; i++) {
1608                 if (test_bit(i, pernet->id_bitmap)) {
1609                         entry = __lookup_addr_by_id(pernet, i);
1610                         if (!entry)
1611                                 break;
1612
1613                         if (entry->addr.id <= id)
1614                                 continue;
1615
1616                         hdr = genlmsg_put(msg, NETLINK_CB(cb->skb).portid,
1617                                           cb->nlh->nlmsg_seq, &mptcp_genl_family,
1618                                           NLM_F_MULTI, MPTCP_PM_CMD_GET_ADDR);
1619                         if (!hdr)
1620                                 break;
1621
1622                         if (mptcp_nl_fill_addr(msg, entry) < 0) {
1623                                 genlmsg_cancel(msg, hdr);
1624                                 break;
1625                         }
1626
1627                         id = entry->addr.id;
1628                         genlmsg_end(msg, hdr);
1629                 }
1630         }
1631         spin_unlock_bh(&pernet->lock);
1632
1633         cb->args[0] = id;
1634         return msg->len;
1635 }
1636
1637 static int parse_limit(struct genl_info *info, int id, unsigned int *limit)
1638 {
1639         struct nlattr *attr = info->attrs[id];
1640
1641         if (!attr)
1642                 return 0;
1643
1644         *limit = nla_get_u32(attr);
1645         if (*limit > MPTCP_PM_ADDR_MAX) {
1646                 GENL_SET_ERR_MSG(info, "limit greater than maximum");
1647                 return -EINVAL;
1648         }
1649         return 0;
1650 }
1651
1652 static int
1653 mptcp_nl_cmd_set_limits(struct sk_buff *skb, struct genl_info *info)
1654 {
1655         struct pm_nl_pernet *pernet = genl_info_pm_nl(info);
1656         unsigned int rcv_addrs, subflows;
1657         int ret;
1658
1659         spin_lock_bh(&pernet->lock);
1660         rcv_addrs = pernet->add_addr_accept_max;
1661         ret = parse_limit(info, MPTCP_PM_ATTR_RCV_ADD_ADDRS, &rcv_addrs);
1662         if (ret)
1663                 goto unlock;
1664
1665         subflows = pernet->subflows_max;
1666         ret = parse_limit(info, MPTCP_PM_ATTR_SUBFLOWS, &subflows);
1667         if (ret)
1668                 goto unlock;
1669
1670         WRITE_ONCE(pernet->add_addr_accept_max, rcv_addrs);
1671         WRITE_ONCE(pernet->subflows_max, subflows);
1672
1673 unlock:
1674         spin_unlock_bh(&pernet->lock);
1675         return ret;
1676 }
1677
1678 static int
1679 mptcp_nl_cmd_get_limits(struct sk_buff *skb, struct genl_info *info)
1680 {
1681         struct pm_nl_pernet *pernet = genl_info_pm_nl(info);
1682         struct sk_buff *msg;
1683         void *reply;
1684
1685         msg = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
1686         if (!msg)
1687                 return -ENOMEM;
1688
1689         reply = genlmsg_put_reply(msg, info, &mptcp_genl_family, 0,
1690                                   MPTCP_PM_CMD_GET_LIMITS);
1691         if (!reply)
1692                 goto fail;
1693
1694         if (nla_put_u32(msg, MPTCP_PM_ATTR_RCV_ADD_ADDRS,
1695                         READ_ONCE(pernet->add_addr_accept_max)))
1696                 goto fail;
1697
1698         if (nla_put_u32(msg, MPTCP_PM_ATTR_SUBFLOWS,
1699                         READ_ONCE(pernet->subflows_max)))
1700                 goto fail;
1701
1702         genlmsg_end(msg, reply);
1703         return genlmsg_reply(msg, info);
1704
1705 fail:
1706         GENL_SET_ERR_MSG(info, "not enough space in Netlink message");
1707         nlmsg_free(msg);
1708         return -EMSGSIZE;
1709 }
1710
1711 static int mptcp_nl_addr_backup(struct net *net,
1712                                 struct mptcp_addr_info *addr,
1713                                 u8 bkup)
1714 {
1715         long s_slot = 0, s_num = 0;
1716         struct mptcp_sock *msk;
1717         int ret = -EINVAL;
1718
1719         while ((msk = mptcp_token_iter_next(net, &s_slot, &s_num)) != NULL) {
1720                 struct sock *sk = (struct sock *)msk;
1721
1722                 if (list_empty(&msk->conn_list))
1723                         goto next;
1724
1725                 lock_sock(sk);
1726                 spin_lock_bh(&msk->pm.lock);
1727                 ret = mptcp_pm_nl_mp_prio_send_ack(msk, addr, bkup);
1728                 spin_unlock_bh(&msk->pm.lock);
1729                 release_sock(sk);
1730
1731 next:
1732                 sock_put(sk);
1733                 cond_resched();
1734         }
1735
1736         return ret;
1737 }
1738
1739 static int mptcp_nl_cmd_set_flags(struct sk_buff *skb, struct genl_info *info)
1740 {
1741         struct nlattr *attr = info->attrs[MPTCP_PM_ATTR_ADDR];
1742         struct pm_nl_pernet *pernet = genl_info_pm_nl(info);
1743         struct mptcp_pm_addr_entry addr, *entry;
1744         struct net *net = sock_net(skb->sk);
1745         u8 bkup = 0;
1746         int ret;
1747
1748         ret = mptcp_pm_parse_addr(attr, info, true, &addr);
1749         if (ret < 0)
1750                 return ret;
1751
1752         if (addr.flags & MPTCP_PM_ADDR_FLAG_BACKUP)
1753                 bkup = 1;
1754
1755         spin_lock_bh(&pernet->lock);
1756         entry = __lookup_addr(pernet, &addr.addr);
1757         if (!entry) {
1758                 spin_unlock_bh(&pernet->lock);
1759                 return -EINVAL;
1760         }
1761
1762         if (bkup)
1763                 entry->flags |= MPTCP_PM_ADDR_FLAG_BACKUP;
1764         else
1765                 entry->flags &= ~MPTCP_PM_ADDR_FLAG_BACKUP;
1766         addr = *entry;
1767         spin_unlock_bh(&pernet->lock);
1768
1769         mptcp_nl_addr_backup(net, &addr.addr, bkup);
1770         return 0;
1771 }
1772
1773 static void mptcp_nl_mcast_send(struct net *net, struct sk_buff *nlskb, gfp_t gfp)
1774 {
1775         genlmsg_multicast_netns(&mptcp_genl_family, net,
1776                                 nlskb, 0, MPTCP_PM_EV_GRP_OFFSET, gfp);
1777 }
1778
1779 static int mptcp_event_add_subflow(struct sk_buff *skb, const struct sock *ssk)
1780 {
1781         const struct inet_sock *issk = inet_sk(ssk);
1782         const struct mptcp_subflow_context *sf;
1783
1784         if (nla_put_u16(skb, MPTCP_ATTR_FAMILY, ssk->sk_family))
1785                 return -EMSGSIZE;
1786
1787         switch (ssk->sk_family) {
1788         case AF_INET:
1789                 if (nla_put_in_addr(skb, MPTCP_ATTR_SADDR4, issk->inet_saddr))
1790                         return -EMSGSIZE;
1791                 if (nla_put_in_addr(skb, MPTCP_ATTR_DADDR4, issk->inet_daddr))
1792                         return -EMSGSIZE;
1793                 break;
1794 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
1795         case AF_INET6: {
1796                 const struct ipv6_pinfo *np = inet6_sk(ssk);
1797
1798                 if (nla_put_in6_addr(skb, MPTCP_ATTR_SADDR6, &np->saddr))
1799                         return -EMSGSIZE;
1800                 if (nla_put_in6_addr(skb, MPTCP_ATTR_DADDR6, &ssk->sk_v6_daddr))
1801                         return -EMSGSIZE;
1802                 break;
1803         }
1804 #endif
1805         default:
1806                 WARN_ON_ONCE(1);
1807                 return -EMSGSIZE;
1808         }
1809
1810         if (nla_put_be16(skb, MPTCP_ATTR_SPORT, issk->inet_sport))
1811                 return -EMSGSIZE;
1812         if (nla_put_be16(skb, MPTCP_ATTR_DPORT, issk->inet_dport))
1813                 return -EMSGSIZE;
1814
1815         sf = mptcp_subflow_ctx(ssk);
1816         if (WARN_ON_ONCE(!sf))
1817                 return -EINVAL;
1818
1819         if (nla_put_u8(skb, MPTCP_ATTR_LOC_ID, sf->local_id))
1820                 return -EMSGSIZE;
1821
1822         if (nla_put_u8(skb, MPTCP_ATTR_REM_ID, sf->remote_id))
1823                 return -EMSGSIZE;
1824
1825         return 0;
1826 }
1827
1828 static int mptcp_event_put_token_and_ssk(struct sk_buff *skb,
1829                                          const struct mptcp_sock *msk,
1830                                          const struct sock *ssk)
1831 {
1832         const struct sock *sk = (const struct sock *)msk;
1833         const struct mptcp_subflow_context *sf;
1834         u8 sk_err;
1835
1836         if (nla_put_u32(skb, MPTCP_ATTR_TOKEN, msk->token))
1837                 return -EMSGSIZE;
1838
1839         if (mptcp_event_add_subflow(skb, ssk))
1840                 return -EMSGSIZE;
1841
1842         sf = mptcp_subflow_ctx(ssk);
1843         if (WARN_ON_ONCE(!sf))
1844                 return -EINVAL;
1845
1846         if (nla_put_u8(skb, MPTCP_ATTR_BACKUP, sf->backup))
1847                 return -EMSGSIZE;
1848
1849         if (ssk->sk_bound_dev_if &&
1850             nla_put_s32(skb, MPTCP_ATTR_IF_IDX, ssk->sk_bound_dev_if))
1851                 return -EMSGSIZE;
1852
1853         sk_err = ssk->sk_err;
1854         if (sk_err && sk->sk_state == TCP_ESTABLISHED &&
1855             nla_put_u8(skb, MPTCP_ATTR_ERROR, sk_err))
1856                 return -EMSGSIZE;
1857
1858         return 0;
1859 }
1860
1861 static int mptcp_event_sub_established(struct sk_buff *skb,
1862                                        const struct mptcp_sock *msk,
1863                                        const struct sock *ssk)
1864 {
1865         return mptcp_event_put_token_and_ssk(skb, msk, ssk);
1866 }
1867
1868 static int mptcp_event_sub_closed(struct sk_buff *skb,
1869                                   const struct mptcp_sock *msk,
1870                                   const struct sock *ssk)
1871 {
1872         const struct mptcp_subflow_context *sf;
1873
1874         if (mptcp_event_put_token_and_ssk(skb, msk, ssk))
1875                 return -EMSGSIZE;
1876
1877         sf = mptcp_subflow_ctx(ssk);
1878         if (!sf->reset_seen)
1879                 return 0;
1880
1881         if (nla_put_u32(skb, MPTCP_ATTR_RESET_REASON, sf->reset_reason))
1882                 return -EMSGSIZE;
1883
1884         if (nla_put_u32(skb, MPTCP_ATTR_RESET_FLAGS, sf->reset_transient))
1885                 return -EMSGSIZE;
1886
1887         return 0;
1888 }
1889
1890 static int mptcp_event_created(struct sk_buff *skb,
1891                                const struct mptcp_sock *msk,
1892                                const struct sock *ssk)
1893 {
1894         int err = nla_put_u32(skb, MPTCP_ATTR_TOKEN, msk->token);
1895
1896         if (err)
1897                 return err;
1898
1899         return mptcp_event_add_subflow(skb, ssk);
1900 }
1901
1902 void mptcp_event_addr_removed(const struct mptcp_sock *msk, uint8_t id)
1903 {
1904         struct net *net = sock_net((const struct sock *)msk);
1905         struct nlmsghdr *nlh;
1906         struct sk_buff *skb;
1907
1908         if (!genl_has_listeners(&mptcp_genl_family, net, MPTCP_PM_EV_GRP_OFFSET))
1909                 return;
1910
1911         skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_ATOMIC);
1912         if (!skb)
1913                 return;
1914
1915         nlh = genlmsg_put(skb, 0, 0, &mptcp_genl_family, 0, MPTCP_EVENT_REMOVED);
1916         if (!nlh)
1917                 goto nla_put_failure;
1918
1919         if (nla_put_u32(skb, MPTCP_ATTR_TOKEN, msk->token))
1920                 goto nla_put_failure;
1921
1922         if (nla_put_u8(skb, MPTCP_ATTR_REM_ID, id))
1923                 goto nla_put_failure;
1924
1925         genlmsg_end(skb, nlh);
1926         mptcp_nl_mcast_send(net, skb, GFP_ATOMIC);
1927         return;
1928
1929 nla_put_failure:
1930         kfree_skb(skb);
1931 }
1932
1933 void mptcp_event_addr_announced(const struct mptcp_sock *msk,
1934                                 const struct mptcp_addr_info *info)
1935 {
1936         struct net *net = sock_net((const struct sock *)msk);
1937         struct nlmsghdr *nlh;
1938         struct sk_buff *skb;
1939
1940         if (!genl_has_listeners(&mptcp_genl_family, net, MPTCP_PM_EV_GRP_OFFSET))
1941                 return;
1942
1943         skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_ATOMIC);
1944         if (!skb)
1945                 return;
1946
1947         nlh = genlmsg_put(skb, 0, 0, &mptcp_genl_family, 0,
1948                           MPTCP_EVENT_ANNOUNCED);
1949         if (!nlh)
1950                 goto nla_put_failure;
1951
1952         if (nla_put_u32(skb, MPTCP_ATTR_TOKEN, msk->token))
1953                 goto nla_put_failure;
1954
1955         if (nla_put_u8(skb, MPTCP_ATTR_REM_ID, info->id))
1956                 goto nla_put_failure;
1957
1958         if (nla_put_be16(skb, MPTCP_ATTR_DPORT, info->port))
1959                 goto nla_put_failure;
1960
1961         switch (info->family) {
1962         case AF_INET:
1963                 if (nla_put_in_addr(skb, MPTCP_ATTR_DADDR4, info->addr.s_addr))
1964                         goto nla_put_failure;
1965                 break;
1966 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
1967         case AF_INET6:
1968                 if (nla_put_in6_addr(skb, MPTCP_ATTR_DADDR6, &info->addr6))
1969                         goto nla_put_failure;
1970                 break;
1971 #endif
1972         default:
1973                 WARN_ON_ONCE(1);
1974                 goto nla_put_failure;
1975         }
1976
1977         genlmsg_end(skb, nlh);
1978         mptcp_nl_mcast_send(net, skb, GFP_ATOMIC);
1979         return;
1980
1981 nla_put_failure:
1982         kfree_skb(skb);
1983 }
1984
1985 void mptcp_event(enum mptcp_event_type type, const struct mptcp_sock *msk,
1986                  const struct sock *ssk, gfp_t gfp)
1987 {
1988         struct net *net = sock_net((const struct sock *)msk);
1989         struct nlmsghdr *nlh;
1990         struct sk_buff *skb;
1991
1992         if (!genl_has_listeners(&mptcp_genl_family, net, MPTCP_PM_EV_GRP_OFFSET))
1993                 return;
1994
1995         skb = nlmsg_new(NLMSG_DEFAULT_SIZE, gfp);
1996         if (!skb)
1997                 return;
1998
1999         nlh = genlmsg_put(skb, 0, 0, &mptcp_genl_family, 0, type);
2000         if (!nlh)
2001                 goto nla_put_failure;
2002
2003         switch (type) {
2004         case MPTCP_EVENT_UNSPEC:
2005                 WARN_ON_ONCE(1);
2006                 break;
2007         case MPTCP_EVENT_CREATED:
2008         case MPTCP_EVENT_ESTABLISHED:
2009                 if (mptcp_event_created(skb, msk, ssk) < 0)
2010                         goto nla_put_failure;
2011                 break;
2012         case MPTCP_EVENT_CLOSED:
2013                 if (nla_put_u32(skb, MPTCP_ATTR_TOKEN, msk->token) < 0)
2014                         goto nla_put_failure;
2015                 break;
2016         case MPTCP_EVENT_ANNOUNCED:
2017         case MPTCP_EVENT_REMOVED:
2018                 /* call mptcp_event_addr_announced()/removed instead */
2019                 WARN_ON_ONCE(1);
2020                 break;
2021         case MPTCP_EVENT_SUB_ESTABLISHED:
2022         case MPTCP_EVENT_SUB_PRIORITY:
2023                 if (mptcp_event_sub_established(skb, msk, ssk) < 0)
2024                         goto nla_put_failure;
2025                 break;
2026         case MPTCP_EVENT_SUB_CLOSED:
2027                 if (mptcp_event_sub_closed(skb, msk, ssk) < 0)
2028                         goto nla_put_failure;
2029                 break;
2030         }
2031
2032         genlmsg_end(skb, nlh);
2033         mptcp_nl_mcast_send(net, skb, gfp);
2034         return;
2035
2036 nla_put_failure:
2037         kfree_skb(skb);
2038 }
2039
2040 static const struct genl_small_ops mptcp_pm_ops[] = {
2041         {
2042                 .cmd    = MPTCP_PM_CMD_ADD_ADDR,
2043                 .doit   = mptcp_nl_cmd_add_addr,
2044                 .flags  = GENL_ADMIN_PERM,
2045         },
2046         {
2047                 .cmd    = MPTCP_PM_CMD_DEL_ADDR,
2048                 .doit   = mptcp_nl_cmd_del_addr,
2049                 .flags  = GENL_ADMIN_PERM,
2050         },
2051         {
2052                 .cmd    = MPTCP_PM_CMD_FLUSH_ADDRS,
2053                 .doit   = mptcp_nl_cmd_flush_addrs,
2054                 .flags  = GENL_ADMIN_PERM,
2055         },
2056         {
2057                 .cmd    = MPTCP_PM_CMD_GET_ADDR,
2058                 .doit   = mptcp_nl_cmd_get_addr,
2059                 .dumpit   = mptcp_nl_cmd_dump_addrs,
2060         },
2061         {
2062                 .cmd    = MPTCP_PM_CMD_SET_LIMITS,
2063                 .doit   = mptcp_nl_cmd_set_limits,
2064                 .flags  = GENL_ADMIN_PERM,
2065         },
2066         {
2067                 .cmd    = MPTCP_PM_CMD_GET_LIMITS,
2068                 .doit   = mptcp_nl_cmd_get_limits,
2069         },
2070         {
2071                 .cmd    = MPTCP_PM_CMD_SET_FLAGS,
2072                 .doit   = mptcp_nl_cmd_set_flags,
2073                 .flags  = GENL_ADMIN_PERM,
2074         },
2075 };
2076
2077 static struct genl_family mptcp_genl_family __ro_after_init = {
2078         .name           = MPTCP_PM_NAME,
2079         .version        = MPTCP_PM_VER,
2080         .maxattr        = MPTCP_PM_ATTR_MAX,
2081         .policy         = mptcp_pm_policy,
2082         .netnsok        = true,
2083         .module         = THIS_MODULE,
2084         .small_ops      = mptcp_pm_ops,
2085         .n_small_ops    = ARRAY_SIZE(mptcp_pm_ops),
2086         .mcgrps         = mptcp_pm_mcgrps,
2087         .n_mcgrps       = ARRAY_SIZE(mptcp_pm_mcgrps),
2088 };
2089
2090 static int __net_init pm_nl_init_net(struct net *net)
2091 {
2092         struct pm_nl_pernet *pernet = net_generic(net, pm_nl_pernet_id);
2093
2094         INIT_LIST_HEAD_RCU(&pernet->local_addr_list);
2095         pernet->next_id = 1;
2096         pernet->stale_loss_cnt = 4;
2097         spin_lock_init(&pernet->lock);
2098
2099         /* No need to initialize other pernet fields, the struct is zeroed at
2100          * allocation time.
2101          */
2102
2103         return 0;
2104 }
2105
2106 static void __net_exit pm_nl_exit_net(struct list_head *net_list)
2107 {
2108         struct net *net;
2109
2110         list_for_each_entry(net, net_list, exit_list) {
2111                 struct pm_nl_pernet *pernet = net_generic(net, pm_nl_pernet_id);
2112
2113                 /* net is removed from namespace list, can't race with
2114                  * other modifiers, also netns core already waited for a
2115                  * RCU grace period.
2116                  */
2117                 __flush_addrs(&pernet->local_addr_list);
2118         }
2119 }
2120
2121 static struct pernet_operations mptcp_pm_pernet_ops = {
2122         .init = pm_nl_init_net,
2123         .exit_batch = pm_nl_exit_net,
2124         .id = &pm_nl_pernet_id,
2125         .size = sizeof(struct pm_nl_pernet),
2126 };
2127
2128 void __init mptcp_pm_nl_init(void)
2129 {
2130         if (register_pernet_subsys(&mptcp_pm_pernet_ops) < 0)
2131                 panic("Failed to register MPTCP PM pernet subsystem.\n");
2132
2133         if (genl_register_family(&mptcp_genl_family))
2134                 panic("Failed to register MPTCP PM netlink family\n");
2135 }