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 struct pm_nl_pernet {
  42         /* protects pernet updates */
  43         spinlock_t              lock;
  44         struct list_head        local_addr_list;
  45         unsigned int            addrs;
  46         unsigned int            stale_loss_cnt;
  47         unsigned int            add_addr_signal_max;
  48         unsigned int            add_addr_accept_max;
  49         unsigned int            local_addr_max;
  50         unsigned int            subflows_max;
  51         unsigned int            next_id;
  52         DECLARE_BITMAP(id_bitmap, MPTCP_PM_MAX_ADDR_ID + 1);
  53 };
  54
  55 #define MPTCP_PM_ADDR_MAX       8
  56 #define ADD_ADDR_RETRANS_MAX    3
  57
  58 static bool addresses_equal(const struct mptcp_addr_info *a,
  59                             const struct mptcp_addr_info *b, bool use_port)
  60 {
  61         bool addr_equals = false;
  62
  63         if (a->family == b->family) {
  64                 if (a->family == AF_INET)
  65                         addr_equals = a->addr.s_addr == b->addr.s_addr;
  66 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
  67                 else
  68                         addr_equals = !ipv6_addr_cmp(&a->addr6, &b->addr6);
  69         } else if (a->family == AF_INET) {
  70                 if (ipv6_addr_v4mapped(&b->addr6))
  71                         addr_equals = a->addr.s_addr == b->addr6.s6_addr32[3];
  72         } else if (b->family == AF_INET) {
  73                 if (ipv6_addr_v4mapped(&a->addr6))
  74                         addr_equals = a->addr6.s6_addr32[3] == b->addr.s_addr;
  75 #endif
  76         }
  77
  78         if (!addr_equals)
  79                 return false;
  80         if (!use_port)
  81                 return true;
  82
  83         return a->port == b->port;
  84 }
  85
  86 static bool address_zero(const struct mptcp_addr_info *addr)
  87 {
  88         struct mptcp_addr_info zero;
  89
  90         memset(&zero, 0, sizeof(zero));
  91         zero.family = addr->family;
  92
  93         return addresses_equal(addr, &zero, true);
  94 }
  95
  96 static void local_address(const struct sock_common *skc,
  97                           struct mptcp_addr_info *addr)
  98 {
  99         addr->family = skc->skc_family;
 100         addr->port = htons(skc->skc_num);
 101         if (addr->family == AF_INET)
 102                 addr->addr.s_addr = skc->skc_rcv_saddr;
 103 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
 104         else if (addr->family == AF_INET6)
 105                 addr->addr6 = skc->skc_v6_rcv_saddr;
 106 #endif
 107 }
 108
 109 static void remote_address(const struct sock_common *skc,
 110                            struct mptcp_addr_info *addr)
 111 {
 112         addr->family = skc->skc_family;
 113         addr->port = skc->skc_dport;
 114         if (addr->family == AF_INET)
 115                 addr->addr.s_addr = skc->skc_daddr;
 116 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
 117         else if (addr->family == AF_INET6)
 118                 addr->addr6 = skc->skc_v6_daddr;
 119 #endif
 120 }
 121
 122 static bool lookup_subflow_by_saddr(const struct list_head *list,
 123                                     struct mptcp_addr_info *saddr)
 124 {
 125         struct mptcp_subflow_context *subflow;
 126         struct mptcp_addr_info cur;
 127         struct sock_common *skc;
 128
 129         list_for_each_entry(subflow, list, node) {
 130                 skc = (struct sock_common *)mptcp_subflow_tcp_sock(subflow);
 131
 132                 local_address(skc, &cur);
 133                 if (addresses_equal(&cur, saddr, saddr->port))
 134                         return true;
 135         }
 136
 137         return false;
 138 }
 139
 140 static bool lookup_subflow_by_daddr(const struct list_head *list,
 141                                     struct mptcp_addr_info *daddr)
 142 {
 143         struct mptcp_subflow_context *subflow;
 144         struct mptcp_addr_info cur;
 145         struct sock_common *skc;
 146
 147         list_for_each_entry(subflow, list, node) {
 148                 skc = (struct sock_common *)mptcp_subflow_tcp_sock(subflow);
 149
 150                 remote_address(skc, &cur);
 151                 if (addresses_equal(&cur, daddr, daddr->port))
 152                         return true;
 153         }
 154
 155         return false;
 156 }
 157
 158 static struct mptcp_pm_addr_entry *
 159 select_local_address(const struct pm_nl_pernet *pernet,
 160                      struct mptcp_sock *msk)
 161 {
 162         struct mptcp_pm_addr_entry *entry, *ret = NULL;
 163         struct sock *sk = (struct sock *)msk;
 164
 165         msk_owned_by_me(msk);
 166
 167         rcu_read_lock();
 168         list_for_each_entry_rcu(entry, &pernet->local_addr_list, list) {
 169                 if (!(entry->flags & MPTCP_PM_ADDR_FLAG_SUBFLOW))
 170                         continue;
 171
 172                 if (!test_bit(entry->addr.id, msk->pm.id_avail_bitmap))
 173                         continue;
 174
 175                 if (entry->addr.family != sk->sk_family) {
 176 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
 177                         if ((entry->addr.family == AF_INET &&
 178                              !ipv6_addr_v4mapped(&sk->sk_v6_daddr)) ||
 179                             (sk->sk_family == AF_INET &&
 180                              !ipv6_addr_v4mapped(&entry->addr.addr6)))
 181 #endif
 182                                 continue;
 183                 }
 184
 185                 ret = entry;
 186                 break;
 187         }
 188         rcu_read_unlock();
 189         return ret;
 190 }
 191
 192 static struct mptcp_pm_addr_entry *
 193 select_signal_address(struct pm_nl_pernet *pernet, struct mptcp_sock *msk)
 194 {
 195         struct mptcp_pm_addr_entry *entry, *ret = NULL;
 196
 197         rcu_read_lock();
 198         /* do not keep any additional per socket state, just signal
 199          * the address list in order.
 200          * Note: removal from the local address list during the msk life-cycle
 201          * can lead to additional addresses not being announced.
 202          */
 203         list_for_each_entry_rcu(entry, &pernet->local_addr_list, list) {
 204                 if (!test_bit(entry->addr.id, msk->pm.id_avail_bitmap))
 205                         continue;
 206
 207                 if (!(entry->flags & MPTCP_PM_ADDR_FLAG_SIGNAL))
 208                         continue;
 209
 210                 ret = entry;
 211                 break;
 212         }
 213         rcu_read_unlock();
 214         return ret;
 215 }
 216
 217 unsigned int mptcp_pm_get_add_addr_signal_max(struct mptcp_sock *msk)
 218 {
 219         struct pm_nl_pernet *pernet;
 220
 221         pernet = net_generic(sock_net((struct sock *)msk), pm_nl_pernet_id);
 222         return READ_ONCE(pernet->add_addr_signal_max);
 223 }
 224 EXPORT_SYMBOL_GPL(mptcp_pm_get_add_addr_signal_max);
 225
 226 unsigned int mptcp_pm_get_add_addr_accept_max(struct mptcp_sock *msk)
 227 {
 228         struct pm_nl_pernet *pernet;
 229
 230         pernet = net_generic(sock_net((struct sock *)msk), pm_nl_pernet_id);
 231         return READ_ONCE(pernet->add_addr_accept_max);
 232 }
 233 EXPORT_SYMBOL_GPL(mptcp_pm_get_add_addr_accept_max);
 234
 235 unsigned int mptcp_pm_get_subflows_max(struct mptcp_sock *msk)
 236 {
 237         struct pm_nl_pernet *pernet;
 238
 239         pernet = net_generic(sock_net((struct sock *)msk), pm_nl_pernet_id);
 240         return READ_ONCE(pernet->subflows_max);
 241 }
 242 EXPORT_SYMBOL_GPL(mptcp_pm_get_subflows_max);
 243
 244 unsigned int mptcp_pm_get_local_addr_max(struct mptcp_sock *msk)
 245 {
 246         struct pm_nl_pernet *pernet;
 247
 248         pernet = net_generic(sock_net((struct sock *)msk), pm_nl_pernet_id);
 249         return READ_ONCE(pernet->local_addr_max);
 250 }
 251 EXPORT_SYMBOL_GPL(mptcp_pm_get_local_addr_max);
 252
 253 bool mptcp_pm_nl_check_work_pending(struct mptcp_sock *msk)
 254 {
 255         struct pm_nl_pernet *pernet = net_generic(sock_net((struct sock *)msk), pm_nl_pernet_id);
 256
 257         if (msk->pm.subflows == mptcp_pm_get_subflows_max(msk) ||
 258             (find_next_and_bit(pernet->id_bitmap, msk->pm.id_avail_bitmap,
 259                                MPTCP_PM_MAX_ADDR_ID + 1, 0) == MPTCP_PM_MAX_ADDR_ID + 1)) {
 260                 WRITE_ONCE(msk->pm.work_pending, false);
 261                 return false;
 262         }
 263         return true;
 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         bool deny_id0 = READ_ONCE(msk->pm.remote_deny_join_id0);
 433         struct sock *sk = (struct sock *)msk, *ssk;
 434         struct mptcp_subflow_context *subflow;
 435         struct mptcp_addr_info remote = { 0 };
 436         unsigned int subflows_max;
 437         int i = 0;
 438
 439         subflows_max = mptcp_pm_get_subflows_max(msk);
 440         remote_address((struct sock_common *)sk, &remote);
 441
 442         /* Non-fullmesh endpoint, fill in the single entry
 443          * corresponding to the primary MPC subflow remote address
 444          */
 445         if (!fullmesh) {
 446                 if (deny_id0)
 447                         return 0;
 448
 449                 msk->pm.subflows++;
 450                 addrs[i++] = remote;
 451         } else {
 452                 mptcp_for_each_subflow(msk, subflow) {
 453                         ssk = mptcp_subflow_tcp_sock(subflow);
 454                         remote_address((struct sock_common *)ssk, &addrs[i]);
 455                         if (deny_id0 && addresses_equal(&addrs[i], &remote, false))
 456                                 continue;
 457
 458                         if (!lookup_address_in_vec(addrs, i, &addrs[i]) &&
 459                             msk->pm.subflows < subflows_max) {
 460                                 msk->pm.subflows++;
 461                                 i++;
 462                         }
 463                 }
 464         }
 465
 466         return i;
 467 }
 468
 469 static struct mptcp_pm_addr_entry *
 470 __lookup_addr_by_id(struct pm_nl_pernet *pernet, unsigned int id)
 471 {
 472         struct mptcp_pm_addr_entry *entry;
 473
 474         list_for_each_entry(entry, &pernet->local_addr_list, list) {
 475                 if (entry->addr.id == id)
 476                         return entry;
 477         }
 478         return NULL;
 479 }
 480
 481 static struct mptcp_pm_addr_entry *
 482 __lookup_addr(struct pm_nl_pernet *pernet, const struct mptcp_addr_info *info,
 483               bool lookup_by_id)
 484 {
 485         struct mptcp_pm_addr_entry *entry;
 486
 487         list_for_each_entry(entry, &pernet->local_addr_list, list) {
 488                 if ((!lookup_by_id && addresses_equal(&entry->addr, info, true)) ||
 489                     (lookup_by_id && entry->addr.id == info->id))
 490                         return entry;
 491         }
 492         return NULL;
 493 }
 494
 495 static int
 496 lookup_id_by_addr(struct pm_nl_pernet *pernet, const struct mptcp_addr_info *addr)
 497 {
 498         struct mptcp_pm_addr_entry *entry;
 499         int ret = -1;
 500
 501         rcu_read_lock();
 502         list_for_each_entry(entry, &pernet->local_addr_list, list) {
 503                 if (addresses_equal(&entry->addr, addr, entry->addr.port)) {
 504                         ret = entry->addr.id;
 505                         break;
 506                 }
 507         }
 508         rcu_read_unlock();
 509         return ret;
 510 }
 511
 512 static void mptcp_pm_create_subflow_or_signal_addr(struct mptcp_sock *msk)
 513 {
 514         struct sock *sk = (struct sock *)msk;
 515         struct mptcp_pm_addr_entry *local;
 516         unsigned int add_addr_signal_max;
 517         unsigned int local_addr_max;
 518         struct pm_nl_pernet *pernet;
 519         unsigned int subflows_max;
 520
 521         pernet = net_generic(sock_net(sk), pm_nl_pernet_id);
 522
 523         add_addr_signal_max = mptcp_pm_get_add_addr_signal_max(msk);
 524         local_addr_max = mptcp_pm_get_local_addr_max(msk);
 525         subflows_max = mptcp_pm_get_subflows_max(msk);
 526
 527         /* do lazy endpoint usage accounting for the MPC subflows */
 528         if (unlikely(!(msk->pm.status & BIT(MPTCP_PM_MPC_ENDPOINT_ACCOUNTED))) && msk->first) {
 529                 struct mptcp_addr_info mpc_addr;
 530                 int mpc_id;
 531
 532                 local_address((struct sock_common *)msk->first, &mpc_addr);
 533                 mpc_id = lookup_id_by_addr(pernet, &mpc_addr);
 534                 if (mpc_id >= 0)
 535                         __clear_bit(mpc_id, msk->pm.id_avail_bitmap);
 536
 537                 msk->pm.status |= BIT(MPTCP_PM_MPC_ENDPOINT_ACCOUNTED);
 538         }
 539
 540         pr_debug("local %d:%d signal %d:%d subflows %d:%d\n",
 541                  msk->pm.local_addr_used, local_addr_max,
 542                  msk->pm.add_addr_signaled, add_addr_signal_max,
 543                  msk->pm.subflows, subflows_max);
 544
 545         /* check first for announce */
 546         if (msk->pm.add_addr_signaled < add_addr_signal_max) {
 547                 local = select_signal_address(pernet, msk);
 548
 549                 if (local) {
 550                         if (mptcp_pm_alloc_anno_list(msk, local)) {
 551                                 __clear_bit(local->addr.id, msk->pm.id_avail_bitmap);
 552                                 msk->pm.add_addr_signaled++;
 553                                 mptcp_pm_announce_addr(msk, &local->addr, false);
 554                                 mptcp_pm_nl_addr_send_ack(msk);
 555                         }
 556                 }
 557         }
 558
 559         /* check if should create a new subflow */
 560         while (msk->pm.local_addr_used < local_addr_max &&
 561                msk->pm.subflows < subflows_max) {
 562                 struct mptcp_addr_info addrs[MPTCP_PM_ADDR_MAX];
 563                 bool fullmesh;
 564                 int i, nr;
 565
 566                 local = select_local_address(pernet, msk);
 567                 if (!local)
 568                         break;
 569
 570                 fullmesh = !!(local->flags & MPTCP_PM_ADDR_FLAG_FULLMESH);
 571
 572                 msk->pm.local_addr_used++;
 573                 nr = fill_remote_addresses_vec(msk, fullmesh, addrs);
 574                 if (nr)
 575                         __clear_bit(local->addr.id, msk->pm.id_avail_bitmap);
 576                 spin_unlock_bh(&msk->pm.lock);
 577                 for (i = 0; i < nr; i++)
 578                         __mptcp_subflow_connect(sk, &local->addr, &addrs[i]);
 579                 spin_lock_bh(&msk->pm.lock);
 580         }
 581         mptcp_pm_nl_check_work_pending(msk);
 582 }
 583
 584 static void mptcp_pm_nl_fully_established(struct mptcp_sock *msk)
 585 {
 586         mptcp_pm_create_subflow_or_signal_addr(msk);
 587 }
 588
 589 static void mptcp_pm_nl_subflow_established(struct mptcp_sock *msk)
 590 {
 591         mptcp_pm_create_subflow_or_signal_addr(msk);
 592 }
 593
 594 /* Fill all the local addresses into the array addrs[],
 595  * and return the array size.
 596  */
 597 static unsigned int fill_local_addresses_vec(struct mptcp_sock *msk,
 598                                              struct mptcp_addr_info *addrs)
 599 {
 600         struct sock *sk = (struct sock *)msk;
 601         struct mptcp_pm_addr_entry *entry;
 602         struct mptcp_addr_info local;
 603         struct pm_nl_pernet *pernet;
 604         unsigned int subflows_max;
 605         int i = 0;
 606
 607         pernet = net_generic(sock_net(sk), pm_nl_pernet_id);
 608         subflows_max = mptcp_pm_get_subflows_max(msk);
 609
 610         rcu_read_lock();
 611         list_for_each_entry_rcu(entry, &pernet->local_addr_list, list) {
 612                 if (!(entry->flags & MPTCP_PM_ADDR_FLAG_FULLMESH))
 613                         continue;
 614
 615                 if (entry->addr.family != sk->sk_family) {
 616 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
 617                         if ((entry->addr.family == AF_INET &&
 618                              !ipv6_addr_v4mapped(&sk->sk_v6_daddr)) ||
 619                             (sk->sk_family == AF_INET &&
 620                              !ipv6_addr_v4mapped(&entry->addr.addr6)))
 621 #endif
 622                                 continue;
 623                 }
 624
 625                 if (msk->pm.subflows < subflows_max) {
 626                         msk->pm.subflows++;
 627                         addrs[i++] = entry->addr;
 628                 }
 629         }
 630         rcu_read_unlock();
 631
 632         /* If the array is empty, fill in the single
 633          * 'IPADDRANY' local address
 634          */
 635         if (!i) {
 636                 memset(&local, 0, sizeof(local));
 637                 local.family = msk->pm.remote.family;
 638
 639                 msk->pm.subflows++;
 640                 addrs[i++] = local;
 641         }
 642
 643         return i;
 644 }
 645
 646 static void mptcp_pm_nl_add_addr_received(struct mptcp_sock *msk)
 647 {
 648         struct mptcp_addr_info addrs[MPTCP_PM_ADDR_MAX];
 649         struct sock *sk = (struct sock *)msk;
 650         unsigned int add_addr_accept_max;
 651         struct mptcp_addr_info remote;
 652         unsigned int subflows_max;
 653         int i, nr;
 654
 655         add_addr_accept_max = mptcp_pm_get_add_addr_accept_max(msk);
 656         subflows_max = mptcp_pm_get_subflows_max(msk);
 657
 658         pr_debug("accepted %d:%d remote family %d",
 659                  msk->pm.add_addr_accepted, add_addr_accept_max,
 660                  msk->pm.remote.family);
 661
 662         if (lookup_subflow_by_daddr(&msk->conn_list, &msk->pm.remote))
 663                 goto add_addr_echo;
 664
 665         /* connect to the specified remote address, using whatever
 666          * local address the routing configuration will pick.
 667          */
 668         remote = msk->pm.remote;
 669         if (!remote.port)
 670                 remote.port = sk->sk_dport;
 671         nr = fill_local_addresses_vec(msk, addrs);
 672
 673         msk->pm.add_addr_accepted++;
 674         if (msk->pm.add_addr_accepted >= add_addr_accept_max ||
 675             msk->pm.subflows >= subflows_max)
 676                 WRITE_ONCE(msk->pm.accept_addr, false);
 677
 678         spin_unlock_bh(&msk->pm.lock);
 679         for (i = 0; i < nr; i++)
 680                 __mptcp_subflow_connect(sk, &addrs[i], &remote);
 681         spin_lock_bh(&msk->pm.lock);
 682
 683 add_addr_echo:
 684         mptcp_pm_announce_addr(msk, &msk->pm.remote, true);
 685         mptcp_pm_nl_addr_send_ack(msk);
 686 }
 687
 688 void mptcp_pm_nl_addr_send_ack(struct mptcp_sock *msk)
 689 {
 690         struct mptcp_subflow_context *subflow;
 691
 692         msk_owned_by_me(msk);
 693         lockdep_assert_held(&msk->pm.lock);
 694
 695         if (!mptcp_pm_should_add_signal(msk) &&
 696             !mptcp_pm_should_rm_signal(msk))
 697                 return;
 698
 699         subflow = list_first_entry_or_null(&msk->conn_list, typeof(*subflow), node);
 700         if (subflow) {
 701                 struct sock *ssk = mptcp_subflow_tcp_sock(subflow);
 702
 703                 spin_unlock_bh(&msk->pm.lock);
 704                 pr_debug("send ack for %s",
 705                          mptcp_pm_should_add_signal(msk) ? "add_addr" : "rm_addr");
 706
 707                 mptcp_subflow_send_ack(ssk);
 708                 spin_lock_bh(&msk->pm.lock);
 709         }
 710 }
 711
 712 static int mptcp_pm_nl_mp_prio_send_ack(struct mptcp_sock *msk,
 713                                         struct mptcp_addr_info *addr,
 714                                         u8 bkup)
 715 {
 716         struct mptcp_subflow_context *subflow;
 717
 718         pr_debug("bkup=%d", bkup);
 719
 720         mptcp_for_each_subflow(msk, subflow) {
 721                 struct sock *ssk = mptcp_subflow_tcp_sock(subflow);
 722                 struct sock *sk = (struct sock *)msk;
 723                 struct mptcp_addr_info local;
 724
 725                 local_address((struct sock_common *)ssk, &local);
 726                 if (!addresses_equal(&local, addr, addr->port))
 727                         continue;
 728
 729                 subflow->backup = bkup;
 730                 subflow->send_mp_prio = 1;
 731                 subflow->request_bkup = bkup;
 732                 __MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_MPPRIOTX);
 733
 734                 spin_unlock_bh(&msk->pm.lock);
 735                 pr_debug("send ack for mp_prio");
 736                 mptcp_subflow_send_ack(ssk);
 737                 spin_lock_bh(&msk->pm.lock);
 738
 739                 return 0;
 740         }
 741
 742         return -EINVAL;
 743 }
 744
 745 static void mptcp_pm_nl_rm_addr_or_subflow(struct mptcp_sock *msk,
 746                                            const struct mptcp_rm_list *rm_list,
 747                                            enum linux_mptcp_mib_field rm_type)
 748 {
 749         struct mptcp_subflow_context *subflow, *tmp;
 750         struct sock *sk = (struct sock *)msk;
 751         u8 i;
 752
 753         pr_debug("%s rm_list_nr %d",
 754                  rm_type == MPTCP_MIB_RMADDR ? "address" : "subflow", rm_list->nr);
 755
 756         msk_owned_by_me(msk);
 757
 758         if (sk->sk_state == TCP_LISTEN)
 759                 return;
 760
 761         if (!rm_list->nr)
 762                 return;
 763
 764         if (list_empty(&msk->conn_list))
 765                 return;
 766
 767         for (i = 0; i < rm_list->nr; i++) {
 768                 bool removed = false;
 769
 770                 list_for_each_entry_safe(subflow, tmp, &msk->conn_list, node) {
 771                         struct sock *ssk = mptcp_subflow_tcp_sock(subflow);
 772                         int how = RCV_SHUTDOWN | SEND_SHUTDOWN;
 773                         u8 id = subflow->local_id;
 774
 775                         if (rm_type == MPTCP_MIB_RMADDR)
 776                                 id = subflow->remote_id;
 777
 778                         if (rm_list->ids[i] != id)
 779                                 continue;
 780
 781                         pr_debug(" -> %s rm_list_ids[%d]=%u local_id=%u remote_id=%u",
 782                                  rm_type == MPTCP_MIB_RMADDR ? "address" : "subflow",
 783                                  i, rm_list->ids[i], subflow->local_id, subflow->remote_id);
 784                         spin_unlock_bh(&msk->pm.lock);
 785                         mptcp_subflow_shutdown(sk, ssk, how);
 786
 787                         /* the following takes care of updating the subflows counter */
 788                         mptcp_close_ssk(sk, ssk, subflow);
 789                         spin_lock_bh(&msk->pm.lock);
 790
 791                         removed = true;
 792                         __MPTCP_INC_STATS(sock_net(sk), rm_type);
 793                 }
 794                 __set_bit(rm_list->ids[i], msk->pm.id_avail_bitmap);
 795                 if (!removed)
 796                         continue;
 797
 798                 if (rm_type == MPTCP_MIB_RMADDR) {
 799                         msk->pm.add_addr_accepted--;
 800                         WRITE_ONCE(msk->pm.accept_addr, true);
 801                 } else if (rm_type == MPTCP_MIB_RMSUBFLOW) {
 802                         msk->pm.local_addr_used--;
 803                 }
 804         }
 805 }
 806
 807 static void mptcp_pm_nl_rm_addr_received(struct mptcp_sock *msk)
 808 {
 809         mptcp_pm_nl_rm_addr_or_subflow(msk, &msk->pm.rm_list_rx, MPTCP_MIB_RMADDR);
 810 }
 811
 812 void mptcp_pm_nl_rm_subflow_received(struct mptcp_sock *msk,
 813                                      const struct mptcp_rm_list *rm_list)
 814 {
 815         mptcp_pm_nl_rm_addr_or_subflow(msk, rm_list, MPTCP_MIB_RMSUBFLOW);
 816 }
 817
 818 void mptcp_pm_nl_work(struct mptcp_sock *msk)
 819 {
 820         struct mptcp_pm_data *pm = &msk->pm;
 821
 822         msk_owned_by_me(msk);
 823
 824         if (!(pm->status & MPTCP_PM_WORK_MASK))
 825                 return;
 826
 827         spin_lock_bh(&msk->pm.lock);
 828
 829         pr_debug("msk=%p status=%x", msk, pm->status);
 830         if (pm->status & BIT(MPTCP_PM_ADD_ADDR_RECEIVED)) {
 831                 pm->status &= ~BIT(MPTCP_PM_ADD_ADDR_RECEIVED);
 832                 mptcp_pm_nl_add_addr_received(msk);
 833         }
 834         if (pm->status & BIT(MPTCP_PM_ADD_ADDR_SEND_ACK)) {
 835                 pm->status &= ~BIT(MPTCP_PM_ADD_ADDR_SEND_ACK);
 836                 mptcp_pm_nl_addr_send_ack(msk);
 837         }
 838         if (pm->status & BIT(MPTCP_PM_RM_ADDR_RECEIVED)) {
 839                 pm->status &= ~BIT(MPTCP_PM_RM_ADDR_RECEIVED);
 840                 mptcp_pm_nl_rm_addr_received(msk);
 841         }
 842         if (pm->status & BIT(MPTCP_PM_ESTABLISHED)) {
 843                 pm->status &= ~BIT(MPTCP_PM_ESTABLISHED);
 844                 mptcp_pm_nl_fully_established(msk);
 845         }
 846         if (pm->status & BIT(MPTCP_PM_SUBFLOW_ESTABLISHED)) {
 847                 pm->status &= ~BIT(MPTCP_PM_SUBFLOW_ESTABLISHED);
 848                 mptcp_pm_nl_subflow_established(msk);
 849         }
 850
 851         spin_unlock_bh(&msk->pm.lock);
 852 }
 853
 854 static bool address_use_port(struct mptcp_pm_addr_entry *entry)
 855 {
 856         return (entry->flags &
 857                 (MPTCP_PM_ADDR_FLAG_SIGNAL | MPTCP_PM_ADDR_FLAG_SUBFLOW)) ==
 858                 MPTCP_PM_ADDR_FLAG_SIGNAL;
 859 }
 860
 861 static int mptcp_pm_nl_append_new_local_addr(struct pm_nl_pernet *pernet,
 862                                              struct mptcp_pm_addr_entry *entry)
 863 {
 864         struct mptcp_pm_addr_entry *cur;
 865         unsigned int addr_max;
 866         int ret = -EINVAL;
 867
 868         spin_lock_bh(&pernet->lock);
 869         /* to keep the code simple, don't do IDR-like allocation for address ID,
 870          * just bail when we exceed limits
 871          */
 872         if (pernet->next_id == MPTCP_PM_MAX_ADDR_ID)
 873                 pernet->next_id = 1;
 874         if (pernet->addrs >= MPTCP_PM_ADDR_MAX)
 875                 goto out;
 876         if (test_bit(entry->addr.id, pernet->id_bitmap))
 877                 goto out;
 878
 879         /* do not insert duplicate address, differentiate on port only
 880          * singled addresses
 881          */
 882         list_for_each_entry(cur, &pernet->local_addr_list, list) {
 883                 if (addresses_equal(&cur->addr, &entry->addr,
 884                                     address_use_port(entry) &&
 885                                     address_use_port(cur)))
 886                         goto out;
 887         }
 888
 889         if (!entry->addr.id) {
 890 find_next:
 891                 entry->addr.id = find_next_zero_bit(pernet->id_bitmap,
 892                                                     MPTCP_PM_MAX_ADDR_ID + 1,
 893                                                     pernet->next_id);
 894                 if (!entry->addr.id && pernet->next_id != 1) {
 895                         pernet->next_id = 1;
 896                         goto find_next;
 897                 }
 898         }
 899
 900         if (!entry->addr.id)
 901                 goto out;
 902
 903         __set_bit(entry->addr.id, pernet->id_bitmap);
 904         if (entry->addr.id > pernet->next_id)
 905                 pernet->next_id = entry->addr.id;
 906
 907         if (entry->flags & MPTCP_PM_ADDR_FLAG_SIGNAL) {
 908                 addr_max = pernet->add_addr_signal_max;
 909                 WRITE_ONCE(pernet->add_addr_signal_max, addr_max + 1);
 910         }
 911         if (entry->flags & MPTCP_PM_ADDR_FLAG_SUBFLOW) {
 912                 addr_max = pernet->local_addr_max;
 913                 WRITE_ONCE(pernet->local_addr_max, addr_max + 1);
 914         }
 915
 916         pernet->addrs++;
 917         list_add_tail_rcu(&entry->list, &pernet->local_addr_list);
 918         ret = entry->addr.id;
 919
 920 out:
 921         spin_unlock_bh(&pernet->lock);
 922         return ret;
 923 }
 924
 925 static int mptcp_pm_nl_create_listen_socket(struct sock *sk,
 926                                             struct mptcp_pm_addr_entry *entry)
 927 {
 928         int addrlen = sizeof(struct sockaddr_in);
 929         struct sockaddr_storage addr;
 930         struct mptcp_sock *msk;
 931         struct socket *ssock;
 932         int backlog = 1024;
 933         int err;
 934
 935         err = sock_create_kern(sock_net(sk), entry->addr.family,
 936                                SOCK_STREAM, IPPROTO_MPTCP, &entry->lsk);
 937         if (err)
 938                 return err;
 939
 940         msk = mptcp_sk(entry->lsk->sk);
 941         if (!msk) {
 942                 err = -EINVAL;
 943                 goto out;
 944         }
 945
 946         ssock = __mptcp_nmpc_socket(msk);
 947         if (!ssock) {
 948                 err = -EINVAL;
 949                 goto out;
 950         }
 951
 952         mptcp_info2sockaddr(&entry->addr, &addr, entry->addr.family);
 953 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
 954         if (entry->addr.family == AF_INET6)
 955                 addrlen = sizeof(struct sockaddr_in6);
 956 #endif
 957         err = kernel_bind(ssock, (struct sockaddr *)&addr, addrlen);
 958         if (err) {
 959                 pr_warn("kernel_bind error, err=%d", err);
 960                 goto out;
 961         }
 962
 963         err = kernel_listen(ssock, backlog);
 964         if (err) {
 965                 pr_warn("kernel_listen error, err=%d", err);
 966                 goto out;
 967         }
 968
 969         return 0;
 970
 971 out:
 972         sock_release(entry->lsk);
 973         return err;
 974 }
 975
 976 int mptcp_pm_nl_get_local_id(struct mptcp_sock *msk, struct sock_common *skc)
 977 {
 978         struct mptcp_pm_addr_entry *entry;
 979         struct mptcp_addr_info skc_local;
 980         struct mptcp_addr_info msk_local;
 981         struct pm_nl_pernet *pernet;
 982         int ret = -1;
 983
 984         if (WARN_ON_ONCE(!msk))
 985                 return -1;
 986
 987         /* The 0 ID mapping is defined by the first subflow, copied into the msk
 988          * addr
 989          */
 990         local_address((struct sock_common *)msk, &msk_local);
 991         local_address((struct sock_common *)skc, &skc_local);
 992         if (addresses_equal(&msk_local, &skc_local, false))
 993                 return 0;
 994
 995         if (address_zero(&skc_local))
 996                 return 0;
 997
 998         pernet = net_generic(sock_net((struct sock *)msk), pm_nl_pernet_id);
 999
1000         rcu_read_lock();
1001         list_for_each_entry_rcu(entry, &pernet->local_addr_list, list) {
1002                 if (addresses_equal(&entry->addr, &skc_local, entry->addr.port)) {
1003                         ret = entry->addr.id;
1004                         break;
1005                 }
1006         }
1007         rcu_read_unlock();
1008         if (ret >= 0)
1009                 return ret;
1010
1011         /* address not found, add to local list */
1012         entry = kmalloc(sizeof(*entry), GFP_ATOMIC);
1013         if (!entry)
1014                 return -ENOMEM;
1015
1016         entry->addr = skc_local;
1017         entry->addr.id = 0;
1018         entry->addr.port = 0;
1019         entry->ifindex = 0;
1020         entry->flags = 0;
1021         entry->lsk = NULL;
1022         ret = mptcp_pm_nl_append_new_local_addr(pernet, entry);
1023         if (ret < 0)
1024                 kfree(entry);
1025
1026         return ret;
1027 }
1028
1029 void mptcp_pm_nl_data_init(struct mptcp_sock *msk)
1030 {
1031         struct mptcp_pm_data *pm = &msk->pm;
1032         bool subflows;
1033
1034         subflows = !!mptcp_pm_get_subflows_max(msk);
1035         WRITE_ONCE(pm->work_pending, (!!mptcp_pm_get_local_addr_max(msk) && subflows) ||
1036                    !!mptcp_pm_get_add_addr_signal_max(msk));
1037         WRITE_ONCE(pm->accept_addr, !!mptcp_pm_get_add_addr_accept_max(msk) && subflows);
1038         WRITE_ONCE(pm->accept_subflow, subflows);
1039 }
1040
1041 #define MPTCP_PM_CMD_GRP_OFFSET       0
1042 #define MPTCP_PM_EV_GRP_OFFSET        1
1043
1044 static const struct genl_multicast_group mptcp_pm_mcgrps[] = {
1045         [MPTCP_PM_CMD_GRP_OFFSET]       = { .name = MPTCP_PM_CMD_GRP_NAME, },
1046         [MPTCP_PM_EV_GRP_OFFSET]        = { .name = MPTCP_PM_EV_GRP_NAME,
1047                                             .flags = GENL_UNS_ADMIN_PERM,
1048                                           },
1049 };
1050
1051 static const struct nla_policy
1052 mptcp_pm_addr_policy[MPTCP_PM_ADDR_ATTR_MAX + 1] = {
1053         [MPTCP_PM_ADDR_ATTR_FAMILY]     = { .type       = NLA_U16,      },
1054         [MPTCP_PM_ADDR_ATTR_ID]         = { .type       = NLA_U8,       },
1055         [MPTCP_PM_ADDR_ATTR_ADDR4]      = { .type       = NLA_U32,      },
1056         [MPTCP_PM_ADDR_ATTR_ADDR6]      =
1057                 NLA_POLICY_EXACT_LEN(sizeof(struct in6_addr)),
1058         [MPTCP_PM_ADDR_ATTR_PORT]       = { .type       = NLA_U16       },
1059         [MPTCP_PM_ADDR_ATTR_FLAGS]      = { .type       = NLA_U32       },
1060         [MPTCP_PM_ADDR_ATTR_IF_IDX]     = { .type       = NLA_S32       },
1061 };
1062
1063 static const struct nla_policy mptcp_pm_policy[MPTCP_PM_ATTR_MAX + 1] = {
1064         [MPTCP_PM_ATTR_ADDR]            =
1065                                         NLA_POLICY_NESTED(mptcp_pm_addr_policy),
1066         [MPTCP_PM_ATTR_RCV_ADD_ADDRS]   = { .type       = NLA_U32,      },
1067         [MPTCP_PM_ATTR_SUBFLOWS]        = { .type       = NLA_U32,      },
1068 };
1069
1070 void mptcp_pm_nl_subflow_chk_stale(const struct mptcp_sock *msk, struct sock *ssk)
1071 {
1072         struct mptcp_subflow_context *iter, *subflow = mptcp_subflow_ctx(ssk);
1073         struct sock *sk = (struct sock *)msk;
1074         unsigned int active_max_loss_cnt;
1075         struct net *net = sock_net(sk);
1076         unsigned int stale_loss_cnt;
1077         bool slow;
1078
1079         stale_loss_cnt = mptcp_stale_loss_cnt(net);
1080         if (subflow->stale || !stale_loss_cnt || subflow->stale_count <= stale_loss_cnt)
1081                 return;
1082
1083         /* look for another available subflow not in loss state */
1084         active_max_loss_cnt = max_t(int, stale_loss_cnt - 1, 1);
1085         mptcp_for_each_subflow(msk, iter) {
1086                 if (iter != subflow && mptcp_subflow_active(iter) &&
1087                     iter->stale_count < active_max_loss_cnt) {
1088                         /* we have some alternatives, try to mark this subflow as idle ...*/
1089                         slow = lock_sock_fast(ssk);
1090                         if (!tcp_rtx_and_write_queues_empty(ssk)) {
1091                                 subflow->stale = 1;
1092                                 __mptcp_retransmit_pending_data(sk);
1093                                 MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_SUBFLOWSTALE);
1094                         }
1095                         unlock_sock_fast(ssk, slow);
1096
1097                         /* always try to push the pending data regarless of re-injections:
1098                          * we can possibly use backup subflows now, and subflow selection
1099                          * is cheap under the msk socket lock
1100                          */
1101                         __mptcp_push_pending(sk, 0);
1102                         return;
1103                 }
1104         }
1105 }
1106
1107 static int mptcp_pm_family_to_addr(int family)
1108 {
1109 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
1110         if (family == AF_INET6)
1111                 return MPTCP_PM_ADDR_ATTR_ADDR6;
1112 #endif
1113         return MPTCP_PM_ADDR_ATTR_ADDR4;
1114 }
1115
1116 static int mptcp_pm_parse_addr(struct nlattr *attr, struct genl_info *info,
1117                                bool require_family,
1118                                struct mptcp_pm_addr_entry *entry)
1119 {
1120         struct nlattr *tb[MPTCP_PM_ADDR_ATTR_MAX + 1];
1121         int err, addr_addr;
1122
1123         if (!attr) {
1124                 GENL_SET_ERR_MSG(info, "missing address info");
1125                 return -EINVAL;
1126         }
1127
1128         /* no validation needed - was already done via nested policy */
1129         err = nla_parse_nested_deprecated(tb, MPTCP_PM_ADDR_ATTR_MAX, attr,
1130                                           mptcp_pm_addr_policy, info->extack);
1131         if (err)
1132                 return err;
1133
1134         memset(entry, 0, sizeof(*entry));
1135         if (!tb[MPTCP_PM_ADDR_ATTR_FAMILY]) {
1136                 if (!require_family)
1137                         goto skip_family;
1138
1139                 NL_SET_ERR_MSG_ATTR(info->extack, attr,
1140                                     "missing family");
1141                 return -EINVAL;
1142         }
1143
1144         entry->addr.family = nla_get_u16(tb[MPTCP_PM_ADDR_ATTR_FAMILY]);
1145         if (entry->addr.family != AF_INET
1146 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
1147             && entry->addr.family != AF_INET6
1148 #endif
1149             ) {
1150                 NL_SET_ERR_MSG_ATTR(info->extack, attr,
1151                                     "unknown address family");
1152                 return -EINVAL;
1153         }
1154         addr_addr = mptcp_pm_family_to_addr(entry->addr.family);
1155         if (!tb[addr_addr]) {
1156                 NL_SET_ERR_MSG_ATTR(info->extack, attr,
1157                                     "missing address data");
1158                 return -EINVAL;
1159         }
1160
1161 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
1162         if (entry->addr.family == AF_INET6)
1163                 entry->addr.addr6 = nla_get_in6_addr(tb[addr_addr]);
1164         else
1165 #endif
1166                 entry->addr.addr.s_addr = nla_get_in_addr(tb[addr_addr]);
1167
1168 skip_family:
1169         if (tb[MPTCP_PM_ADDR_ATTR_IF_IDX]) {
1170                 u32 val = nla_get_s32(tb[MPTCP_PM_ADDR_ATTR_IF_IDX]);
1171
1172                 entry->ifindex = val;
1173         }
1174
1175         if (tb[MPTCP_PM_ADDR_ATTR_ID])
1176                 entry->addr.id = nla_get_u8(tb[MPTCP_PM_ADDR_ATTR_ID]);
1177
1178         if (tb[MPTCP_PM_ADDR_ATTR_FLAGS])
1179                 entry->flags = nla_get_u32(tb[MPTCP_PM_ADDR_ATTR_FLAGS]);
1180
1181         if (tb[MPTCP_PM_ADDR_ATTR_PORT]) {
1182                 if (!(entry->flags & MPTCP_PM_ADDR_FLAG_SIGNAL)) {
1183                         NL_SET_ERR_MSG_ATTR(info->extack, attr,
1184                                             "flags must have signal when using port");
1185                         return -EINVAL;
1186                 }
1187                 entry->addr.port = htons(nla_get_u16(tb[MPTCP_PM_ADDR_ATTR_PORT]));
1188         }
1189
1190         return 0;
1191 }
1192
1193 static struct pm_nl_pernet *genl_info_pm_nl(struct genl_info *info)
1194 {
1195         return net_generic(genl_info_net(info), pm_nl_pernet_id);
1196 }
1197
1198 static int mptcp_nl_add_subflow_or_signal_addr(struct net *net)
1199 {
1200         struct mptcp_sock *msk;
1201         long s_slot = 0, s_num = 0;
1202
1203         while ((msk = mptcp_token_iter_next(net, &s_slot, &s_num)) != NULL) {
1204                 struct sock *sk = (struct sock *)msk;
1205
1206                 if (!READ_ONCE(msk->fully_established))
1207                         goto next;
1208
1209                 lock_sock(sk);
1210                 spin_lock_bh(&msk->pm.lock);
1211                 mptcp_pm_create_subflow_or_signal_addr(msk);
1212                 spin_unlock_bh(&msk->pm.lock);
1213                 release_sock(sk);
1214
1215 next:
1216                 sock_put(sk);
1217                 cond_resched();
1218         }
1219
1220         return 0;
1221 }
1222
1223 static int mptcp_nl_cmd_add_addr(struct sk_buff *skb, struct genl_info *info)
1224 {
1225         struct nlattr *attr = info->attrs[MPTCP_PM_ATTR_ADDR];
1226         struct pm_nl_pernet *pernet = genl_info_pm_nl(info);
1227         struct mptcp_pm_addr_entry addr, *entry;
1228         int ret;
1229
1230         ret = mptcp_pm_parse_addr(attr, info, true, &addr);
1231         if (ret < 0)
1232                 return ret;
1233
1234         entry = kmalloc(sizeof(*entry), GFP_KERNEL);
1235         if (!entry) {
1236                 GENL_SET_ERR_MSG(info, "can't allocate addr");
1237                 return -ENOMEM;
1238         }
1239
1240         *entry = addr;
1241         if (entry->addr.port) {
1242                 ret = mptcp_pm_nl_create_listen_socket(skb->sk, entry);
1243                 if (ret) {
1244                         GENL_SET_ERR_MSG(info, "create listen socket error");
1245                         kfree(entry);
1246                         return ret;
1247                 }
1248         }
1249         ret = mptcp_pm_nl_append_new_local_addr(pernet, entry);
1250         if (ret < 0) {
1251                 GENL_SET_ERR_MSG(info, "too many addresses or duplicate one");
1252                 if (entry->lsk)
1253                         sock_release(entry->lsk);
1254                 kfree(entry);
1255                 return ret;
1256         }
1257
1258         mptcp_nl_add_subflow_or_signal_addr(sock_net(skb->sk));
1259
1260         return 0;
1261 }
1262
1263 int mptcp_pm_get_flags_and_ifindex_by_id(struct net *net, unsigned int id,
1264                                          u8 *flags, int *ifindex)
1265 {
1266         struct mptcp_pm_addr_entry *entry;
1267
1268         *flags = 0;
1269         *ifindex = 0;
1270
1271         if (id) {
1272                 rcu_read_lock();
1273                 entry = __lookup_addr_by_id(net_generic(net, pm_nl_pernet_id), id);
1274                 if (entry) {
1275                         *flags = entry->flags;
1276                         *ifindex = entry->ifindex;
1277                 }
1278                 rcu_read_unlock();
1279         }
1280
1281         return 0;
1282 }
1283
1284 static bool remove_anno_list_by_saddr(struct mptcp_sock *msk,
1285                                       struct mptcp_addr_info *addr)
1286 {
1287         struct mptcp_pm_add_entry *entry;
1288
1289         entry = mptcp_pm_del_add_timer(msk, addr, false);
1290         if (entry) {
1291                 list_del(&entry->list);
1292                 kfree(entry);
1293                 return true;
1294         }
1295
1296         return false;
1297 }
1298
1299 static bool mptcp_pm_remove_anno_addr(struct mptcp_sock *msk,
1300                                       struct mptcp_addr_info *addr,
1301                                       bool force)
1302 {
1303         struct mptcp_rm_list list = { .nr = 0 };
1304         bool ret;
1305
1306         list.ids[list.nr++] = addr->id;
1307
1308         ret = remove_anno_list_by_saddr(msk, addr);
1309         if (ret || force) {
1310                 spin_lock_bh(&msk->pm.lock);
1311                 mptcp_pm_remove_addr(msk, &list);
1312                 spin_unlock_bh(&msk->pm.lock);
1313         }
1314         return ret;
1315 }
1316
1317 static int mptcp_nl_remove_subflow_and_signal_addr(struct net *net,
1318                                                    struct mptcp_addr_info *addr)
1319 {
1320         struct mptcp_sock *msk;
1321         long s_slot = 0, s_num = 0;
1322         struct mptcp_rm_list list = { .nr = 0 };
1323
1324         pr_debug("remove_id=%d", addr->id);
1325
1326         list.ids[list.nr++] = addr->id;
1327
1328         while ((msk = mptcp_token_iter_next(net, &s_slot, &s_num)) != NULL) {
1329                 struct sock *sk = (struct sock *)msk;
1330                 bool remove_subflow;
1331
1332                 if (list_empty(&msk->conn_list)) {
1333                         mptcp_pm_remove_anno_addr(msk, addr, false);
1334                         goto next;
1335                 }
1336
1337                 lock_sock(sk);
1338                 remove_subflow = lookup_subflow_by_saddr(&msk->conn_list, addr);
1339                 mptcp_pm_remove_anno_addr(msk, addr, remove_subflow);
1340                 if (remove_subflow)
1341                         mptcp_pm_remove_subflow(msk, &list);
1342                 release_sock(sk);
1343
1344 next:
1345                 sock_put(sk);
1346                 cond_resched();
1347         }
1348
1349         return 0;
1350 }
1351
1352 /* caller must ensure the RCU grace period is already elapsed */
1353 static void __mptcp_pm_release_addr_entry(struct mptcp_pm_addr_entry *entry)
1354 {
1355         if (entry->lsk)
1356                 sock_release(entry->lsk);
1357         kfree(entry);
1358 }
1359
1360 static int mptcp_nl_remove_id_zero_address(struct net *net,
1361                                            struct mptcp_addr_info *addr)
1362 {
1363         struct mptcp_rm_list list = { .nr = 0 };
1364         long s_slot = 0, s_num = 0;
1365         struct mptcp_sock *msk;
1366
1367         list.ids[list.nr++] = 0;
1368
1369         while ((msk = mptcp_token_iter_next(net, &s_slot, &s_num)) != NULL) {
1370                 struct sock *sk = (struct sock *)msk;
1371                 struct mptcp_addr_info msk_local;
1372
1373                 if (list_empty(&msk->conn_list))
1374                         goto next;
1375
1376                 local_address((struct sock_common *)msk, &msk_local);
1377                 if (!addresses_equal(&msk_local, addr, addr->port))
1378                         goto next;
1379
1380                 lock_sock(sk);
1381                 spin_lock_bh(&msk->pm.lock);
1382                 mptcp_pm_remove_addr(msk, &list);
1383                 mptcp_pm_nl_rm_subflow_received(msk, &list);
1384                 spin_unlock_bh(&msk->pm.lock);
1385                 release_sock(sk);
1386
1387 next:
1388                 sock_put(sk);
1389                 cond_resched();
1390         }
1391
1392         return 0;
1393 }
1394
1395 static int mptcp_nl_cmd_del_addr(struct sk_buff *skb, struct genl_info *info)
1396 {
1397         struct nlattr *attr = info->attrs[MPTCP_PM_ATTR_ADDR];
1398         struct pm_nl_pernet *pernet = genl_info_pm_nl(info);
1399         struct mptcp_pm_addr_entry addr, *entry;
1400         unsigned int addr_max;
1401         int ret;
1402
1403         ret = mptcp_pm_parse_addr(attr, info, false, &addr);
1404         if (ret < 0)
1405                 return ret;
1406
1407         /* the zero id address is special: the first address used by the msk
1408          * always gets such an id, so different subflows can have different zero
1409          * id addresses. Additionally zero id is not accounted for in id_bitmap.
1410          * Let's use an 'mptcp_rm_list' instead of the common remove code.
1411          */
1412         if (addr.addr.id == 0)
1413                 return mptcp_nl_remove_id_zero_address(sock_net(skb->sk), &addr.addr);
1414
1415         spin_lock_bh(&pernet->lock);
1416         entry = __lookup_addr_by_id(pernet, addr.addr.id);
1417         if (!entry) {
1418                 GENL_SET_ERR_MSG(info, "address not found");
1419                 spin_unlock_bh(&pernet->lock);
1420                 return -EINVAL;
1421         }
1422         if (entry->flags & MPTCP_PM_ADDR_FLAG_SIGNAL) {
1423                 addr_max = pernet->add_addr_signal_max;
1424                 WRITE_ONCE(pernet->add_addr_signal_max, addr_max - 1);
1425         }
1426         if (entry->flags & MPTCP_PM_ADDR_FLAG_SUBFLOW) {
1427                 addr_max = pernet->local_addr_max;
1428                 WRITE_ONCE(pernet->local_addr_max, addr_max - 1);
1429         }
1430
1431         pernet->addrs--;
1432         list_del_rcu(&entry->list);
1433         __clear_bit(entry->addr.id, pernet->id_bitmap);
1434         spin_unlock_bh(&pernet->lock);
1435
1436         mptcp_nl_remove_subflow_and_signal_addr(sock_net(skb->sk), &entry->addr);
1437         synchronize_rcu();
1438         __mptcp_pm_release_addr_entry(entry);
1439
1440         return ret;
1441 }
1442
1443 static void mptcp_pm_remove_addrs_and_subflows(struct mptcp_sock *msk,
1444                                                struct list_head *rm_list)
1445 {
1446         struct mptcp_rm_list alist = { .nr = 0 }, slist = { .nr = 0 };
1447         struct mptcp_pm_addr_entry *entry;
1448
1449         list_for_each_entry(entry, rm_list, list) {
1450                 if (lookup_subflow_by_saddr(&msk->conn_list, &entry->addr) &&
1451                     alist.nr < MPTCP_RM_IDS_MAX &&
1452                     slist.nr < MPTCP_RM_IDS_MAX) {
1453                         alist.ids[alist.nr++] = entry->addr.id;
1454                         slist.ids[slist.nr++] = entry->addr.id;
1455                 } else if (remove_anno_list_by_saddr(msk, &entry->addr) &&
1456                          alist.nr < MPTCP_RM_IDS_MAX) {
1457                         alist.ids[alist.nr++] = entry->addr.id;
1458                 }
1459         }
1460
1461         if (alist.nr) {
1462                 spin_lock_bh(&msk->pm.lock);
1463                 mptcp_pm_remove_addr(msk, &alist);
1464                 spin_unlock_bh(&msk->pm.lock);
1465         }
1466         if (slist.nr)
1467                 mptcp_pm_remove_subflow(msk, &slist);
1468 }
1469
1470 static void mptcp_nl_remove_addrs_list(struct net *net,
1471                                        struct list_head *rm_list)
1472 {
1473         long s_slot = 0, s_num = 0;
1474         struct mptcp_sock *msk;
1475
1476         if (list_empty(rm_list))
1477                 return;
1478
1479         while ((msk = mptcp_token_iter_next(net, &s_slot, &s_num)) != NULL) {
1480                 struct sock *sk = (struct sock *)msk;
1481
1482                 lock_sock(sk);
1483                 mptcp_pm_remove_addrs_and_subflows(msk, rm_list);
1484                 release_sock(sk);
1485
1486                 sock_put(sk);
1487                 cond_resched();
1488         }
1489 }
1490
1491 /* caller must ensure the RCU grace period is already elapsed */
1492 static void __flush_addrs(struct list_head *list)
1493 {
1494         while (!list_empty(list)) {
1495                 struct mptcp_pm_addr_entry *cur;
1496
1497                 cur = list_entry(list->next,
1498                                  struct mptcp_pm_addr_entry, list);
1499                 list_del_rcu(&cur->list);
1500                 __mptcp_pm_release_addr_entry(cur);
1501         }
1502 }
1503
1504 static void __reset_counters(struct pm_nl_pernet *pernet)
1505 {
1506         WRITE_ONCE(pernet->add_addr_signal_max, 0);
1507         WRITE_ONCE(pernet->add_addr_accept_max, 0);
1508         WRITE_ONCE(pernet->local_addr_max, 0);
1509         pernet->addrs = 0;
1510 }
1511
1512 static int mptcp_nl_cmd_flush_addrs(struct sk_buff *skb, struct genl_info *info)
1513 {
1514         struct pm_nl_pernet *pernet = genl_info_pm_nl(info);
1515         LIST_HEAD(free_list);
1516
1517         spin_lock_bh(&pernet->lock);
1518         list_splice_init(&pernet->local_addr_list, &free_list);
1519         __reset_counters(pernet);
1520         pernet->next_id = 1;
1521         bitmap_zero(pernet->id_bitmap, MPTCP_PM_MAX_ADDR_ID + 1);
1522         spin_unlock_bh(&pernet->lock);
1523         mptcp_nl_remove_addrs_list(sock_net(skb->sk), &free_list);
1524         synchronize_rcu();
1525         __flush_addrs(&free_list);
1526         return 0;
1527 }
1528
1529 static int mptcp_nl_fill_addr(struct sk_buff *skb,
1530                               struct mptcp_pm_addr_entry *entry)
1531 {
1532         struct mptcp_addr_info *addr = &entry->addr;
1533         struct nlattr *attr;
1534
1535         attr = nla_nest_start(skb, MPTCP_PM_ATTR_ADDR);
1536         if (!attr)
1537                 return -EMSGSIZE;
1538
1539         if (nla_put_u16(skb, MPTCP_PM_ADDR_ATTR_FAMILY, addr->family))
1540                 goto nla_put_failure;
1541         if (nla_put_u16(skb, MPTCP_PM_ADDR_ATTR_PORT, ntohs(addr->port)))
1542                 goto nla_put_failure;
1543         if (nla_put_u8(skb, MPTCP_PM_ADDR_ATTR_ID, addr->id))
1544                 goto nla_put_failure;
1545         if (nla_put_u32(skb, MPTCP_PM_ADDR_ATTR_FLAGS, entry->flags))
1546                 goto nla_put_failure;
1547         if (entry->ifindex &&
1548             nla_put_s32(skb, MPTCP_PM_ADDR_ATTR_IF_IDX, entry->ifindex))
1549                 goto nla_put_failure;
1550
1551         if (addr->family == AF_INET &&
1552             nla_put_in_addr(skb, MPTCP_PM_ADDR_ATTR_ADDR4,
1553                             addr->addr.s_addr))
1554                 goto nla_put_failure;
1555 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
1556         else if (addr->family == AF_INET6 &&
1557                  nla_put_in6_addr(skb, MPTCP_PM_ADDR_ATTR_ADDR6, &addr->addr6))
1558                 goto nla_put_failure;
1559 #endif
1560         nla_nest_end(skb, attr);
1561         return 0;
1562
1563 nla_put_failure:
1564         nla_nest_cancel(skb, attr);
1565         return -EMSGSIZE;
1566 }
1567
1568 static int mptcp_nl_cmd_get_addr(struct sk_buff *skb, struct genl_info *info)
1569 {
1570         struct nlattr *attr = info->attrs[MPTCP_PM_ATTR_ADDR];
1571         struct pm_nl_pernet *pernet = genl_info_pm_nl(info);
1572         struct mptcp_pm_addr_entry addr, *entry;
1573         struct sk_buff *msg;
1574         void *reply;
1575         int ret;
1576
1577         ret = mptcp_pm_parse_addr(attr, info, false, &addr);
1578         if (ret < 0)
1579                 return ret;
1580
1581         msg = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
1582         if (!msg)
1583                 return -ENOMEM;
1584
1585         reply = genlmsg_put_reply(msg, info, &mptcp_genl_family, 0,
1586                                   info->genlhdr->cmd);
1587         if (!reply) {
1588                 GENL_SET_ERR_MSG(info, "not enough space in Netlink message");
1589                 ret = -EMSGSIZE;
1590                 goto fail;
1591         }
1592
1593         spin_lock_bh(&pernet->lock);
1594         entry = __lookup_addr_by_id(pernet, addr.addr.id);
1595         if (!entry) {
1596                 GENL_SET_ERR_MSG(info, "address not found");
1597                 ret = -EINVAL;
1598                 goto unlock_fail;
1599         }
1600
1601         ret = mptcp_nl_fill_addr(msg, entry);
1602         if (ret)
1603                 goto unlock_fail;
1604
1605         genlmsg_end(msg, reply);
1606         ret = genlmsg_reply(msg, info);
1607         spin_unlock_bh(&pernet->lock);
1608         return ret;
1609
1610 unlock_fail:
1611         spin_unlock_bh(&pernet->lock);
1612
1613 fail:
1614         nlmsg_free(msg);
1615         return ret;
1616 }
1617
1618 static int mptcp_nl_cmd_dump_addrs(struct sk_buff *msg,
1619                                    struct netlink_callback *cb)
1620 {
1621         struct net *net = sock_net(msg->sk);
1622         struct mptcp_pm_addr_entry *entry;
1623         struct pm_nl_pernet *pernet;
1624         int id = cb->args[0];
1625         void *hdr;
1626         int i;
1627
1628         pernet = net_generic(net, pm_nl_pernet_id);
1629
1630         spin_lock_bh(&pernet->lock);
1631         for (i = id; i < MPTCP_PM_MAX_ADDR_ID + 1; i++) {
1632                 if (test_bit(i, pernet->id_bitmap)) {
1633                         entry = __lookup_addr_by_id(pernet, i);
1634                         if (!entry)
1635                                 break;
1636
1637                         if (entry->addr.id <= id)
1638                                 continue;
1639
1640                         hdr = genlmsg_put(msg, NETLINK_CB(cb->skb).portid,
1641                                           cb->nlh->nlmsg_seq, &mptcp_genl_family,
1642                                           NLM_F_MULTI, MPTCP_PM_CMD_GET_ADDR);
1643                         if (!hdr)
1644                                 break;
1645
1646                         if (mptcp_nl_fill_addr(msg, entry) < 0) {
1647                                 genlmsg_cancel(msg, hdr);
1648                                 break;
1649                         }
1650
1651                         id = entry->addr.id;
1652                         genlmsg_end(msg, hdr);
1653                 }
1654         }
1655         spin_unlock_bh(&pernet->lock);
1656
1657         cb->args[0] = id;
1658         return msg->len;
1659 }
1660
1661 static int parse_limit(struct genl_info *info, int id, unsigned int *limit)
1662 {
1663         struct nlattr *attr = info->attrs[id];
1664
1665         if (!attr)
1666                 return 0;
1667
1668         *limit = nla_get_u32(attr);
1669         if (*limit > MPTCP_PM_ADDR_MAX) {
1670                 GENL_SET_ERR_MSG(info, "limit greater than maximum");
1671                 return -EINVAL;
1672         }
1673         return 0;
1674 }
1675
1676 static int
1677 mptcp_nl_cmd_set_limits(struct sk_buff *skb, struct genl_info *info)
1678 {
1679         struct pm_nl_pernet *pernet = genl_info_pm_nl(info);
1680         unsigned int rcv_addrs, subflows;
1681         int ret;
1682
1683         spin_lock_bh(&pernet->lock);
1684         rcv_addrs = pernet->add_addr_accept_max;
1685         ret = parse_limit(info, MPTCP_PM_ATTR_RCV_ADD_ADDRS, &rcv_addrs);
1686         if (ret)
1687                 goto unlock;
1688
1689         subflows = pernet->subflows_max;
1690         ret = parse_limit(info, MPTCP_PM_ATTR_SUBFLOWS, &subflows);
1691         if (ret)
1692                 goto unlock;
1693
1694         WRITE_ONCE(pernet->add_addr_accept_max, rcv_addrs);
1695         WRITE_ONCE(pernet->subflows_max, subflows);
1696
1697 unlock:
1698         spin_unlock_bh(&pernet->lock);
1699         return ret;
1700 }
1701
1702 static int
1703 mptcp_nl_cmd_get_limits(struct sk_buff *skb, struct genl_info *info)
1704 {
1705         struct pm_nl_pernet *pernet = genl_info_pm_nl(info);
1706         struct sk_buff *msg;
1707         void *reply;
1708
1709         msg = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
1710         if (!msg)
1711                 return -ENOMEM;
1712
1713         reply = genlmsg_put_reply(msg, info, &mptcp_genl_family, 0,
1714                                   MPTCP_PM_CMD_GET_LIMITS);
1715         if (!reply)
1716                 goto fail;
1717
1718         if (nla_put_u32(msg, MPTCP_PM_ATTR_RCV_ADD_ADDRS,
1719                         READ_ONCE(pernet->add_addr_accept_max)))
1720                 goto fail;
1721
1722         if (nla_put_u32(msg, MPTCP_PM_ATTR_SUBFLOWS,
1723                         READ_ONCE(pernet->subflows_max)))
1724                 goto fail;
1725
1726         genlmsg_end(msg, reply);
1727         return genlmsg_reply(msg, info);
1728
1729 fail:
1730         GENL_SET_ERR_MSG(info, "not enough space in Netlink message");
1731         nlmsg_free(msg);
1732         return -EMSGSIZE;
1733 }
1734
1735 static int mptcp_nl_addr_backup(struct net *net,
1736                                 struct mptcp_addr_info *addr,
1737                                 u8 bkup)
1738 {
1739         long s_slot = 0, s_num = 0;
1740         struct mptcp_sock *msk;
1741         int ret = -EINVAL;
1742
1743         while ((msk = mptcp_token_iter_next(net, &s_slot, &s_num)) != NULL) {
1744                 struct sock *sk = (struct sock *)msk;
1745
1746                 if (list_empty(&msk->conn_list))
1747                         goto next;
1748
1749                 lock_sock(sk);
1750                 spin_lock_bh(&msk->pm.lock);
1751                 ret = mptcp_pm_nl_mp_prio_send_ack(msk, addr, bkup);
1752                 spin_unlock_bh(&msk->pm.lock);
1753                 release_sock(sk);
1754
1755 next:
1756                 sock_put(sk);
1757                 cond_resched();
1758         }
1759
1760         return ret;
1761 }
1762
1763 static int mptcp_nl_cmd_set_flags(struct sk_buff *skb, struct genl_info *info)
1764 {
1765         struct mptcp_pm_addr_entry addr = { .addr = { .family = AF_UNSPEC }, }, *entry;
1766         struct nlattr *attr = info->attrs[MPTCP_PM_ATTR_ADDR];
1767         struct pm_nl_pernet *pernet = genl_info_pm_nl(info);
1768         struct net *net = sock_net(skb->sk);
1769         u8 bkup = 0, lookup_by_id = 0;
1770         int ret;
1771
1772         ret = mptcp_pm_parse_addr(attr, info, false, &addr);
1773         if (ret < 0)
1774                 return ret;
1775
1776         if (addr.flags & MPTCP_PM_ADDR_FLAG_BACKUP)
1777                 bkup = 1;
1778         if (addr.addr.family == AF_UNSPEC) {
1779                 lookup_by_id = 1;
1780                 if (!addr.addr.id)
1781                         return -EOPNOTSUPP;
1782         }
1783
1784         spin_lock_bh(&pernet->lock);
1785         entry = __lookup_addr(pernet, &addr.addr, lookup_by_id);
1786         if (!entry) {
1787                 spin_unlock_bh(&pernet->lock);
1788                 return -EINVAL;
1789         }
1790
1791         if (bkup)
1792                 entry->flags |= MPTCP_PM_ADDR_FLAG_BACKUP;
1793         else
1794                 entry->flags &= ~MPTCP_PM_ADDR_FLAG_BACKUP;
1795         addr = *entry;
1796         spin_unlock_bh(&pernet->lock);
1797
1798         mptcp_nl_addr_backup(net, &addr.addr, bkup);
1799         return 0;
1800 }
1801
1802 static void mptcp_nl_mcast_send(struct net *net, struct sk_buff *nlskb, gfp_t gfp)
1803 {
1804         genlmsg_multicast_netns(&mptcp_genl_family, net,
1805                                 nlskb, 0, MPTCP_PM_EV_GRP_OFFSET, gfp);
1806 }
1807
1808 static int mptcp_event_add_subflow(struct sk_buff *skb, const struct sock *ssk)
1809 {
1810         const struct inet_sock *issk = inet_sk(ssk);
1811         const struct mptcp_subflow_context *sf;
1812
1813         if (nla_put_u16(skb, MPTCP_ATTR_FAMILY, ssk->sk_family))
1814                 return -EMSGSIZE;
1815
1816         switch (ssk->sk_family) {
1817         case AF_INET:
1818                 if (nla_put_in_addr(skb, MPTCP_ATTR_SADDR4, issk->inet_saddr))
1819                         return -EMSGSIZE;
1820                 if (nla_put_in_addr(skb, MPTCP_ATTR_DADDR4, issk->inet_daddr))
1821                         return -EMSGSIZE;
1822                 break;
1823 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
1824         case AF_INET6: {
1825                 const struct ipv6_pinfo *np = inet6_sk(ssk);
1826
1827                 if (nla_put_in6_addr(skb, MPTCP_ATTR_SADDR6, &np->saddr))
1828                         return -EMSGSIZE;
1829                 if (nla_put_in6_addr(skb, MPTCP_ATTR_DADDR6, &ssk->sk_v6_daddr))
1830                         return -EMSGSIZE;
1831                 break;
1832         }
1833 #endif
1834         default:
1835                 WARN_ON_ONCE(1);
1836                 return -EMSGSIZE;
1837         }
1838
1839         if (nla_put_be16(skb, MPTCP_ATTR_SPORT, issk->inet_sport))
1840                 return -EMSGSIZE;
1841         if (nla_put_be16(skb, MPTCP_ATTR_DPORT, issk->inet_dport))
1842                 return -EMSGSIZE;
1843
1844         sf = mptcp_subflow_ctx(ssk);
1845         if (WARN_ON_ONCE(!sf))
1846                 return -EINVAL;
1847
1848         if (nla_put_u8(skb, MPTCP_ATTR_LOC_ID, sf->local_id))
1849                 return -EMSGSIZE;
1850
1851         if (nla_put_u8(skb, MPTCP_ATTR_REM_ID, sf->remote_id))
1852                 return -EMSGSIZE;
1853
1854         return 0;
1855 }
1856
1857 static int mptcp_event_put_token_and_ssk(struct sk_buff *skb,
1858                                          const struct mptcp_sock *msk,
1859                                          const struct sock *ssk)
1860 {
1861         const struct sock *sk = (const struct sock *)msk;
1862         const struct mptcp_subflow_context *sf;
1863         u8 sk_err;
1864
1865         if (nla_put_u32(skb, MPTCP_ATTR_TOKEN, msk->token))
1866                 return -EMSGSIZE;
1867
1868         if (mptcp_event_add_subflow(skb, ssk))
1869                 return -EMSGSIZE;
1870
1871         sf = mptcp_subflow_ctx(ssk);
1872         if (WARN_ON_ONCE(!sf))
1873                 return -EINVAL;
1874
1875         if (nla_put_u8(skb, MPTCP_ATTR_BACKUP, sf->backup))
1876                 return -EMSGSIZE;
1877
1878         if (ssk->sk_bound_dev_if &&
1879             nla_put_s32(skb, MPTCP_ATTR_IF_IDX, ssk->sk_bound_dev_if))
1880                 return -EMSGSIZE;
1881
1882         sk_err = ssk->sk_err;
1883         if (sk_err && sk->sk_state == TCP_ESTABLISHED &&
1884             nla_put_u8(skb, MPTCP_ATTR_ERROR, sk_err))
1885                 return -EMSGSIZE;
1886
1887         return 0;
1888 }
1889
1890 static int mptcp_event_sub_established(struct sk_buff *skb,
1891                                        const struct mptcp_sock *msk,
1892                                        const struct sock *ssk)
1893 {
1894         return mptcp_event_put_token_and_ssk(skb, msk, ssk);
1895 }
1896
1897 static int mptcp_event_sub_closed(struct sk_buff *skb,
1898                                   const struct mptcp_sock *msk,
1899                                   const struct sock *ssk)
1900 {
1901         const struct mptcp_subflow_context *sf;
1902
1903         if (mptcp_event_put_token_and_ssk(skb, msk, ssk))
1904                 return -EMSGSIZE;
1905
1906         sf = mptcp_subflow_ctx(ssk);
1907         if (!sf->reset_seen)
1908                 return 0;
1909
1910         if (nla_put_u32(skb, MPTCP_ATTR_RESET_REASON, sf->reset_reason))
1911                 return -EMSGSIZE;
1912
1913         if (nla_put_u32(skb, MPTCP_ATTR_RESET_FLAGS, sf->reset_transient))
1914                 return -EMSGSIZE;
1915
1916         return 0;
1917 }
1918
1919 static int mptcp_event_created(struct sk_buff *skb,
1920                                const struct mptcp_sock *msk,
1921                                const struct sock *ssk)
1922 {
1923         int err = nla_put_u32(skb, MPTCP_ATTR_TOKEN, msk->token);
1924
1925         if (err)
1926                 return err;
1927
1928         return mptcp_event_add_subflow(skb, ssk);
1929 }
1930
1931 void mptcp_event_addr_removed(const struct mptcp_sock *msk, uint8_t id)
1932 {
1933         struct net *net = sock_net((const struct sock *)msk);
1934         struct nlmsghdr *nlh;
1935         struct sk_buff *skb;
1936
1937         if (!genl_has_listeners(&mptcp_genl_family, net, MPTCP_PM_EV_GRP_OFFSET))
1938                 return;
1939
1940         skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_ATOMIC);
1941         if (!skb)
1942                 return;
1943
1944         nlh = genlmsg_put(skb, 0, 0, &mptcp_genl_family, 0, MPTCP_EVENT_REMOVED);
1945         if (!nlh)
1946                 goto nla_put_failure;
1947
1948         if (nla_put_u32(skb, MPTCP_ATTR_TOKEN, msk->token))
1949                 goto nla_put_failure;
1950
1951         if (nla_put_u8(skb, MPTCP_ATTR_REM_ID, id))
1952                 goto nla_put_failure;
1953
1954         genlmsg_end(skb, nlh);
1955         mptcp_nl_mcast_send(net, skb, GFP_ATOMIC);
1956         return;
1957
1958 nla_put_failure:
1959         kfree_skb(skb);
1960 }
1961
1962 void mptcp_event_addr_announced(const struct mptcp_sock *msk,
1963                                 const struct mptcp_addr_info *info)
1964 {
1965         struct net *net = sock_net((const struct sock *)msk);
1966         struct nlmsghdr *nlh;
1967         struct sk_buff *skb;
1968
1969         if (!genl_has_listeners(&mptcp_genl_family, net, MPTCP_PM_EV_GRP_OFFSET))
1970                 return;
1971
1972         skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_ATOMIC);
1973         if (!skb)
1974                 return;
1975
1976         nlh = genlmsg_put(skb, 0, 0, &mptcp_genl_family, 0,
1977                           MPTCP_EVENT_ANNOUNCED);
1978         if (!nlh)
1979                 goto nla_put_failure;
1980
1981         if (nla_put_u32(skb, MPTCP_ATTR_TOKEN, msk->token))
1982                 goto nla_put_failure;
1983
1984         if (nla_put_u8(skb, MPTCP_ATTR_REM_ID, info->id))
1985                 goto nla_put_failure;
1986
1987         if (nla_put_be16(skb, MPTCP_ATTR_DPORT, info->port))
1988                 goto nla_put_failure;
1989
1990         switch (info->family) {
1991         case AF_INET:
1992                 if (nla_put_in_addr(skb, MPTCP_ATTR_DADDR4, info->addr.s_addr))
1993                         goto nla_put_failure;
1994                 break;
1995 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
1996         case AF_INET6:
1997                 if (nla_put_in6_addr(skb, MPTCP_ATTR_DADDR6, &info->addr6))
1998                         goto nla_put_failure;
1999                 break;
2000 #endif
2001         default:
2002                 WARN_ON_ONCE(1);
2003                 goto nla_put_failure;
2004         }
2005
2006         genlmsg_end(skb, nlh);
2007         mptcp_nl_mcast_send(net, skb, GFP_ATOMIC);
2008         return;
2009
2010 nla_put_failure:
2011         kfree_skb(skb);
2012 }
2013
2014 void mptcp_event(enum mptcp_event_type type, const struct mptcp_sock *msk,
2015                  const struct sock *ssk, gfp_t gfp)
2016 {
2017         struct net *net = sock_net((const struct sock *)msk);
2018         struct nlmsghdr *nlh;
2019         struct sk_buff *skb;
2020
2021         if (!genl_has_listeners(&mptcp_genl_family, net, MPTCP_PM_EV_GRP_OFFSET))
2022                 return;
2023
2024         skb = nlmsg_new(NLMSG_DEFAULT_SIZE, gfp);
2025         if (!skb)
2026                 return;
2027
2028         nlh = genlmsg_put(skb, 0, 0, &mptcp_genl_family, 0, type);
2029         if (!nlh)
2030                 goto nla_put_failure;
2031
2032         switch (type) {
2033         case MPTCP_EVENT_UNSPEC:
2034                 WARN_ON_ONCE(1);
2035                 break;
2036         case MPTCP_EVENT_CREATED:
2037         case MPTCP_EVENT_ESTABLISHED:
2038                 if (mptcp_event_created(skb, msk, ssk) < 0)
2039                         goto nla_put_failure;
2040                 break;
2041         case MPTCP_EVENT_CLOSED:
2042                 if (nla_put_u32(skb, MPTCP_ATTR_TOKEN, msk->token) < 0)
2043                         goto nla_put_failure;
2044                 break;
2045         case MPTCP_EVENT_ANNOUNCED:
2046         case MPTCP_EVENT_REMOVED:
2047                 /* call mptcp_event_addr_announced()/removed instead */
2048                 WARN_ON_ONCE(1);
2049                 break;
2050         case MPTCP_EVENT_SUB_ESTABLISHED:
2051         case MPTCP_EVENT_SUB_PRIORITY:
2052                 if (mptcp_event_sub_established(skb, msk, ssk) < 0)
2053                         goto nla_put_failure;
2054                 break;
2055         case MPTCP_EVENT_SUB_CLOSED:
2056                 if (mptcp_event_sub_closed(skb, msk, ssk) < 0)
2057                         goto nla_put_failure;
2058                 break;
2059         }
2060
2061         genlmsg_end(skb, nlh);
2062         mptcp_nl_mcast_send(net, skb, gfp);
2063         return;
2064
2065 nla_put_failure:
2066         kfree_skb(skb);
2067 }
2068
2069 static const struct genl_small_ops mptcp_pm_ops[] = {
2070         {
2071                 .cmd    = MPTCP_PM_CMD_ADD_ADDR,
2072                 .doit   = mptcp_nl_cmd_add_addr,
2073                 .flags  = GENL_ADMIN_PERM,
2074         },
2075         {
2076                 .cmd    = MPTCP_PM_CMD_DEL_ADDR,
2077                 .doit   = mptcp_nl_cmd_del_addr,
2078                 .flags  = GENL_ADMIN_PERM,
2079         },
2080         {
2081                 .cmd    = MPTCP_PM_CMD_FLUSH_ADDRS,
2082                 .doit   = mptcp_nl_cmd_flush_addrs,
2083                 .flags  = GENL_ADMIN_PERM,
2084         },
2085         {
2086                 .cmd    = MPTCP_PM_CMD_GET_ADDR,
2087                 .doit   = mptcp_nl_cmd_get_addr,
2088                 .dumpit   = mptcp_nl_cmd_dump_addrs,
2089         },
2090         {
2091                 .cmd    = MPTCP_PM_CMD_SET_LIMITS,
2092                 .doit   = mptcp_nl_cmd_set_limits,
2093                 .flags  = GENL_ADMIN_PERM,
2094         },
2095         {
2096                 .cmd    = MPTCP_PM_CMD_GET_LIMITS,
2097                 .doit   = mptcp_nl_cmd_get_limits,
2098         },
2099         {
2100                 .cmd    = MPTCP_PM_CMD_SET_FLAGS,
2101                 .doit   = mptcp_nl_cmd_set_flags,
2102                 .flags  = GENL_ADMIN_PERM,
2103         },
2104 };
2105
2106 static struct genl_family mptcp_genl_family __ro_after_init = {
2107         .name           = MPTCP_PM_NAME,
2108         .version        = MPTCP_PM_VER,
2109         .maxattr        = MPTCP_PM_ATTR_MAX,
2110         .policy         = mptcp_pm_policy,
2111         .netnsok        = true,
2112         .module         = THIS_MODULE,
2113         .small_ops      = mptcp_pm_ops,
2114         .n_small_ops    = ARRAY_SIZE(mptcp_pm_ops),
2115         .mcgrps         = mptcp_pm_mcgrps,
2116         .n_mcgrps       = ARRAY_SIZE(mptcp_pm_mcgrps),
2117 };
2118
2119 static int __net_init pm_nl_init_net(struct net *net)
2120 {
2121         struct pm_nl_pernet *pernet = net_generic(net, pm_nl_pernet_id);
2122
2123         INIT_LIST_HEAD_RCU(&pernet->local_addr_list);
2124
2125         /* Cit. 2 subflows ought to be enough for anybody. */
2126         pernet->subflows_max = 2;
2127         pernet->next_id = 1;
2128         pernet->stale_loss_cnt = 4;
2129         spin_lock_init(&pernet->lock);
2130
2131         /* No need to initialize other pernet fields, the struct is zeroed at
2132          * allocation time.
2133          */
2134
2135         return 0;
2136 }
2137
2138 static void __net_exit pm_nl_exit_net(struct list_head *net_list)
2139 {
2140         struct net *net;
2141
2142         list_for_each_entry(net, net_list, exit_list) {
2143                 struct pm_nl_pernet *pernet = net_generic(net, pm_nl_pernet_id);
2144
2145                 /* net is removed from namespace list, can't race with
2146                  * other modifiers, also netns core already waited for a
2147                  * RCU grace period.
2148                  */
2149                 __flush_addrs(&pernet->local_addr_list);
2150         }
2151 }
2152
2153 static struct pernet_operations mptcp_pm_pernet_ops = {
2154         .init = pm_nl_init_net,
2155         .exit_batch = pm_nl_exit_net,
2156         .id = &pm_nl_pernet_id,
2157         .size = sizeof(struct pm_nl_pernet),
2158 };
2159
2160 void __init mptcp_pm_nl_init(void)
2161 {
2162         if (register_pernet_subsys(&mptcp_pm_pernet_ops) < 0)
2163                 panic("Failed to register MPTCP PM pernet subsystem.\n");
2164
2165         if (genl_register_family(&mptcp_genl_family))
2166                 panic("Failed to register MPTCP PM netlink family\n");
2167 }