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_add_entry {
  26         struct list_head        list;
  27         struct mptcp_addr_info  addr;
  28         struct timer_list       add_timer;
  29         struct mptcp_sock       *sock;
  30         u8                      retrans_times;
  31 };
  32
  33 struct pm_nl_pernet {
  34         /* protects pernet updates */
  35         spinlock_t              lock;
  36         struct list_head        local_addr_list;
  37         unsigned int            addrs;
  38         unsigned int            stale_loss_cnt;
  39         unsigned int            add_addr_signal_max;
  40         unsigned int            add_addr_accept_max;
  41         unsigned int            local_addr_max;
  42         unsigned int            subflows_max;
  43         unsigned int            next_id;
  44         DECLARE_BITMAP(id_bitmap, MPTCP_PM_MAX_ADDR_ID + 1);
  45 };
  46
  47 #define MPTCP_PM_ADDR_MAX       8
  48 #define ADD_ADDR_RETRANS_MAX    3
  49
  50 static struct pm_nl_pernet *pm_nl_get_pernet(const struct net *net)
  51 {
  52         return net_generic(net, pm_nl_pernet_id);
  53 }
  54
  55 static struct pm_nl_pernet *
  56 pm_nl_get_pernet_from_msk(const struct mptcp_sock *msk)
  57 {
  58         return pm_nl_get_pernet(sock_net((struct sock *)msk));
  59 }
  60
  61 bool mptcp_addresses_equal(const struct mptcp_addr_info *a,
  62                            const 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 void local_address(const struct sock_common *skc,
  90                           struct mptcp_addr_info *addr)
  91 {
  92         addr->family = skc->skc_family;
  93         addr->port = htons(skc->skc_num);
  94         if (addr->family == AF_INET)
  95                 addr->addr.s_addr = skc->skc_rcv_saddr;
  96 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
  97         else if (addr->family == AF_INET6)
  98                 addr->addr6 = skc->skc_v6_rcv_saddr;
  99 #endif
 100 }
 101
 102 static void remote_address(const struct sock_common *skc,
 103                            struct mptcp_addr_info *addr)
 104 {
 105         addr->family = skc->skc_family;
 106         addr->port = skc->skc_dport;
 107         if (addr->family == AF_INET)
 108                 addr->addr.s_addr = skc->skc_daddr;
 109 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
 110         else if (addr->family == AF_INET6)
 111                 addr->addr6 = skc->skc_v6_daddr;
 112 #endif
 113 }
 114
 115 static bool lookup_subflow_by_saddr(const struct list_head *list,
 116                                     const struct mptcp_addr_info *saddr)
 117 {
 118         struct mptcp_subflow_context *subflow;
 119         struct mptcp_addr_info cur;
 120         struct sock_common *skc;
 121
 122         list_for_each_entry(subflow, list, node) {
 123                 skc = (struct sock_common *)mptcp_subflow_tcp_sock(subflow);
 124
 125                 local_address(skc, &cur);
 126                 if (mptcp_addresses_equal(&cur, saddr, saddr->port))
 127                         return true;
 128         }
 129
 130         return false;
 131 }
 132
 133 static bool lookup_subflow_by_daddr(const struct list_head *list,
 134                                     const struct mptcp_addr_info *daddr)
 135 {
 136         struct mptcp_subflow_context *subflow;
 137         struct mptcp_addr_info cur;
 138         struct sock_common *skc;
 139
 140         list_for_each_entry(subflow, list, node) {
 141                 skc = (struct sock_common *)mptcp_subflow_tcp_sock(subflow);
 142
 143                 remote_address(skc, &cur);
 144                 if (mptcp_addresses_equal(&cur, daddr, daddr->port))
 145                         return true;
 146         }
 147
 148         return false;
 149 }
 150
 151 static struct mptcp_pm_addr_entry *
 152 select_local_address(const struct pm_nl_pernet *pernet,
 153                      const struct mptcp_sock *msk)
 154 {
 155         const struct sock *sk = (const struct sock *)msk;
 156         struct mptcp_pm_addr_entry *entry, *ret = NULL;
 157
 158         msk_owned_by_me(msk);
 159
 160         rcu_read_lock();
 161         list_for_each_entry_rcu(entry, &pernet->local_addr_list, list) {
 162                 if (!(entry->flags & MPTCP_PM_ADDR_FLAG_SUBFLOW))
 163                         continue;
 164
 165                 if (!test_bit(entry->addr.id, msk->pm.id_avail_bitmap))
 166                         continue;
 167
 168                 if (entry->addr.family != sk->sk_family) {
 169 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
 170                         if ((entry->addr.family == AF_INET &&
 171                              !ipv6_addr_v4mapped(&sk->sk_v6_daddr)) ||
 172                             (sk->sk_family == AF_INET &&
 173                              !ipv6_addr_v4mapped(&entry->addr.addr6)))
 174 #endif
 175                                 continue;
 176                 }
 177
 178                 ret = entry;
 179                 break;
 180         }
 181         rcu_read_unlock();
 182         return ret;
 183 }
 184
 185 static struct mptcp_pm_addr_entry *
 186 select_signal_address(struct pm_nl_pernet *pernet, const struct mptcp_sock *msk)
 187 {
 188         struct mptcp_pm_addr_entry *entry, *ret = NULL;
 189
 190         rcu_read_lock();
 191         /* do not keep any additional per socket state, just signal
 192          * the address list in order.
 193          * Note: removal from the local address list during the msk life-cycle
 194          * can lead to additional addresses not being announced.
 195          */
 196         list_for_each_entry_rcu(entry, &pernet->local_addr_list, list) {
 197                 if (!test_bit(entry->addr.id, msk->pm.id_avail_bitmap))
 198                         continue;
 199
 200                 if (!(entry->flags & MPTCP_PM_ADDR_FLAG_SIGNAL))
 201                         continue;
 202
 203                 ret = entry;
 204                 break;
 205         }
 206         rcu_read_unlock();
 207         return ret;
 208 }
 209
 210 unsigned int mptcp_pm_get_add_addr_signal_max(const struct mptcp_sock *msk)
 211 {
 212         const struct pm_nl_pernet *pernet = pm_nl_get_pernet_from_msk(msk);
 213
 214         return READ_ONCE(pernet->add_addr_signal_max);
 215 }
 216 EXPORT_SYMBOL_GPL(mptcp_pm_get_add_addr_signal_max);
 217
 218 unsigned int mptcp_pm_get_add_addr_accept_max(const struct mptcp_sock *msk)
 219 {
 220         struct pm_nl_pernet *pernet = pm_nl_get_pernet_from_msk(msk);
 221
 222         return READ_ONCE(pernet->add_addr_accept_max);
 223 }
 224 EXPORT_SYMBOL_GPL(mptcp_pm_get_add_addr_accept_max);
 225
 226 unsigned int mptcp_pm_get_subflows_max(const struct mptcp_sock *msk)
 227 {
 228         struct pm_nl_pernet *pernet = pm_nl_get_pernet_from_msk(msk);
 229
 230         return READ_ONCE(pernet->subflows_max);
 231 }
 232 EXPORT_SYMBOL_GPL(mptcp_pm_get_subflows_max);
 233
 234 unsigned int mptcp_pm_get_local_addr_max(const struct mptcp_sock *msk)
 235 {
 236         struct pm_nl_pernet *pernet = pm_nl_get_pernet_from_msk(msk);
 237
 238         return READ_ONCE(pernet->local_addr_max);
 239 }
 240 EXPORT_SYMBOL_GPL(mptcp_pm_get_local_addr_max);
 241
 242 bool mptcp_pm_nl_check_work_pending(struct mptcp_sock *msk)
 243 {
 244         struct pm_nl_pernet *pernet = pm_nl_get_pernet_from_msk(msk);
 245
 246         if (msk->pm.subflows == mptcp_pm_get_subflows_max(msk) ||
 247             (find_next_and_bit(pernet->id_bitmap, msk->pm.id_avail_bitmap,
 248                                MPTCP_PM_MAX_ADDR_ID + 1, 0) == MPTCP_PM_MAX_ADDR_ID + 1)) {
 249                 WRITE_ONCE(msk->pm.work_pending, false);
 250                 return false;
 251         }
 252         return true;
 253 }
 254
 255 struct mptcp_pm_add_entry *
 256 mptcp_lookup_anno_list_by_saddr(const struct mptcp_sock *msk,
 257                                 const struct mptcp_addr_info *addr)
 258 {
 259         struct mptcp_pm_add_entry *entry;
 260
 261         lockdep_assert_held(&msk->pm.lock);
 262
 263         list_for_each_entry(entry, &msk->pm.anno_list, list) {
 264                 if (mptcp_addresses_equal(&entry->addr, addr, true))
 265                         return entry;
 266         }
 267
 268         return NULL;
 269 }
 270
 271 bool mptcp_pm_sport_in_anno_list(struct mptcp_sock *msk, const struct sock *sk)
 272 {
 273         struct mptcp_pm_add_entry *entry;
 274         struct mptcp_addr_info saddr;
 275         bool ret = false;
 276
 277         local_address((struct sock_common *)sk, &saddr);
 278
 279         spin_lock_bh(&msk->pm.lock);
 280         list_for_each_entry(entry, &msk->pm.anno_list, list) {
 281                 if (mptcp_addresses_equal(&entry->addr, &saddr, true)) {
 282                         ret = true;
 283                         goto out;
 284                 }
 285         }
 286
 287 out:
 288         spin_unlock_bh(&msk->pm.lock);
 289         return ret;
 290 }
 291
 292 static void mptcp_pm_add_timer(struct timer_list *timer)
 293 {
 294         struct mptcp_pm_add_entry *entry = from_timer(entry, timer, add_timer);
 295         struct mptcp_sock *msk = entry->sock;
 296         struct sock *sk = (struct sock *)msk;
 297
 298         pr_debug("msk=%p", msk);
 299
 300         if (!msk)
 301                 return;
 302
 303         if (inet_sk_state_load(sk) == TCP_CLOSE)
 304                 return;
 305
 306         if (!entry->addr.id)
 307                 return;
 308
 309         if (mptcp_pm_should_add_signal_addr(msk)) {
 310                 sk_reset_timer(sk, timer, jiffies + TCP_RTO_MAX / 8);
 311                 goto out;
 312         }
 313
 314         spin_lock_bh(&msk->pm.lock);
 315
 316         if (!mptcp_pm_should_add_signal_addr(msk)) {
 317                 pr_debug("retransmit ADD_ADDR id=%d", entry->addr.id);
 318                 mptcp_pm_announce_addr(msk, &entry->addr, false);
 319                 mptcp_pm_add_addr_send_ack(msk);
 320                 entry->retrans_times++;
 321         }
 322
 323         if (entry->retrans_times < ADD_ADDR_RETRANS_MAX)
 324                 sk_reset_timer(sk, timer,
 325                                jiffies + mptcp_get_add_addr_timeout(sock_net(sk)));
 326
 327         spin_unlock_bh(&msk->pm.lock);
 328
 329         if (entry->retrans_times == ADD_ADDR_RETRANS_MAX)
 330                 mptcp_pm_subflow_established(msk);
 331
 332 out:
 333         __sock_put(sk);
 334 }
 335
 336 struct mptcp_pm_add_entry *
 337 mptcp_pm_del_add_timer(struct mptcp_sock *msk,
 338                        const struct mptcp_addr_info *addr, bool check_id)
 339 {
 340         struct mptcp_pm_add_entry *entry;
 341         struct sock *sk = (struct sock *)msk;
 342
 343         spin_lock_bh(&msk->pm.lock);
 344         entry = mptcp_lookup_anno_list_by_saddr(msk, addr);
 345         if (entry && (!check_id || entry->addr.id == addr->id))
 346                 entry->retrans_times = ADD_ADDR_RETRANS_MAX;
 347         spin_unlock_bh(&msk->pm.lock);
 348
 349         if (entry && (!check_id || entry->addr.id == addr->id))
 350                 sk_stop_timer_sync(sk, &entry->add_timer);
 351
 352         return entry;
 353 }
 354
 355 bool mptcp_pm_alloc_anno_list(struct mptcp_sock *msk,
 356                               const struct mptcp_pm_addr_entry *entry)
 357 {
 358         struct mptcp_pm_add_entry *add_entry = NULL;
 359         struct sock *sk = (struct sock *)msk;
 360         struct net *net = sock_net(sk);
 361
 362         lockdep_assert_held(&msk->pm.lock);
 363
 364         add_entry = mptcp_lookup_anno_list_by_saddr(msk, &entry->addr);
 365
 366         if (add_entry) {
 367                 if (mptcp_pm_is_kernel(msk))
 368                         return false;
 369
 370                 sk_reset_timer(sk, &add_entry->add_timer,
 371                                jiffies + mptcp_get_add_addr_timeout(net));
 372                 return true;
 373         }
 374
 375         add_entry = kmalloc(sizeof(*add_entry), GFP_ATOMIC);
 376         if (!add_entry)
 377                 return false;
 378
 379         list_add(&add_entry->list, &msk->pm.anno_list);
 380
 381         add_entry->addr = entry->addr;
 382         add_entry->sock = msk;
 383         add_entry->retrans_times = 0;
 384
 385         timer_setup(&add_entry->add_timer, mptcp_pm_add_timer, 0);
 386         sk_reset_timer(sk, &add_entry->add_timer,
 387                        jiffies + mptcp_get_add_addr_timeout(net));
 388
 389         return true;
 390 }
 391
 392 void mptcp_pm_free_anno_list(struct mptcp_sock *msk)
 393 {
 394         struct mptcp_pm_add_entry *entry, *tmp;
 395         struct sock *sk = (struct sock *)msk;
 396         LIST_HEAD(free_list);
 397
 398         pr_debug("msk=%p", msk);
 399
 400         spin_lock_bh(&msk->pm.lock);
 401         list_splice_init(&msk->pm.anno_list, &free_list);
 402         spin_unlock_bh(&msk->pm.lock);
 403
 404         list_for_each_entry_safe(entry, tmp, &free_list, list) {
 405                 sk_stop_timer_sync(sk, &entry->add_timer);
 406                 kfree(entry);
 407         }
 408 }
 409
 410 /* Fill all the remote addresses into the array addrs[],
 411  * and return the array size.
 412  */
 413 static unsigned int fill_remote_addresses_vec(struct mptcp_sock *msk,
 414                                               struct mptcp_addr_info *local,
 415                                               bool fullmesh,
 416                                               struct mptcp_addr_info *addrs)
 417 {
 418         bool deny_id0 = READ_ONCE(msk->pm.remote_deny_join_id0);
 419         struct sock *sk = (struct sock *)msk, *ssk;
 420         struct mptcp_subflow_context *subflow;
 421         struct mptcp_addr_info remote = { 0 };
 422         unsigned int subflows_max;
 423         int i = 0;
 424
 425         subflows_max = mptcp_pm_get_subflows_max(msk);
 426         remote_address((struct sock_common *)sk, &remote);
 427
 428         /* Non-fullmesh endpoint, fill in the single entry
 429          * corresponding to the primary MPC subflow remote address
 430          */
 431         if (!fullmesh) {
 432                 if (deny_id0)
 433                         return 0;
 434
 435                 msk->pm.subflows++;
 436                 addrs[i++] = remote;
 437         } else {
 438                 DECLARE_BITMAP(unavail_id, MPTCP_PM_MAX_ADDR_ID + 1);
 439
 440                 /* Forbid creation of new subflows matching existing
 441                  * ones, possibly already created by incoming ADD_ADDR
 442                  */
 443                 bitmap_zero(unavail_id, MPTCP_PM_MAX_ADDR_ID + 1);
 444                 mptcp_for_each_subflow(msk, subflow)
 445                         if (READ_ONCE(subflow->local_id) == local->id)
 446                                 __set_bit(subflow->remote_id, unavail_id);
 447
 448                 mptcp_for_each_subflow(msk, subflow) {
 449                         ssk = mptcp_subflow_tcp_sock(subflow);
 450                         remote_address((struct sock_common *)ssk, &addrs[i]);
 451                         addrs[i].id = READ_ONCE(subflow->remote_id);
 452                         if (deny_id0 && !addrs[i].id)
 453                                 continue;
 454
 455                         if (msk->pm.subflows < subflows_max) {
 456                                 /* forbid creating multiple address towards
 457                                  * this id
 458                                  */
 459                                 __set_bit(addrs[i].id, unavail_id);
 460                                 msk->pm.subflows++;
 461                                 i++;
 462                         }
 463                 }
 464         }
 465
 466         return i;
 467 }
 468
 469 static void __mptcp_pm_send_ack(struct mptcp_sock *msk, struct mptcp_subflow_context *subflow,
 470                                 bool prio, bool backup)
 471 {
 472         struct sock *ssk = mptcp_subflow_tcp_sock(subflow);
 473         bool slow;
 474
 475         pr_debug("send ack for %s",
 476                  prio ? "mp_prio" : (mptcp_pm_should_add_signal(msk) ? "add_addr" : "rm_addr"));
 477
 478         slow = lock_sock_fast(ssk);
 479         if (prio) {
 480                 if (subflow->backup != backup)
 481                         msk->last_snd = NULL;
 482
 483                 subflow->send_mp_prio = 1;
 484                 subflow->backup = backup;
 485                 subflow->request_bkup = backup;
 486         }
 487
 488         __mptcp_subflow_send_ack(ssk);
 489         unlock_sock_fast(ssk, slow);
 490 }
 491
 492 static void mptcp_pm_send_ack(struct mptcp_sock *msk, struct mptcp_subflow_context *subflow,
 493                               bool prio, bool backup)
 494 {
 495         spin_unlock_bh(&msk->pm.lock);
 496         __mptcp_pm_send_ack(msk, subflow, prio, backup);
 497         spin_lock_bh(&msk->pm.lock);
 498 }
 499
 500 static struct mptcp_pm_addr_entry *
 501 __lookup_addr_by_id(struct pm_nl_pernet *pernet, unsigned int id)
 502 {
 503         struct mptcp_pm_addr_entry *entry;
 504
 505         list_for_each_entry(entry, &pernet->local_addr_list, list) {
 506                 if (entry->addr.id == id)
 507                         return entry;
 508         }
 509         return NULL;
 510 }
 511
 512 static struct mptcp_pm_addr_entry *
 513 __lookup_addr(struct pm_nl_pernet *pernet, const struct mptcp_addr_info *info,
 514               bool lookup_by_id)
 515 {
 516         struct mptcp_pm_addr_entry *entry;
 517
 518         list_for_each_entry(entry, &pernet->local_addr_list, list) {
 519                 if ((!lookup_by_id &&
 520                      mptcp_addresses_equal(&entry->addr, info, entry->addr.port)) ||
 521                     (lookup_by_id && entry->addr.id == info->id))
 522                         return entry;
 523         }
 524         return NULL;
 525 }
 526
 527 static void mptcp_pm_create_subflow_or_signal_addr(struct mptcp_sock *msk)
 528 {
 529         struct sock *sk = (struct sock *)msk;
 530         struct mptcp_pm_addr_entry *local;
 531         unsigned int add_addr_signal_max;
 532         unsigned int local_addr_max;
 533         struct pm_nl_pernet *pernet;
 534         unsigned int subflows_max;
 535
 536         pernet = pm_nl_get_pernet(sock_net(sk));
 537
 538         add_addr_signal_max = mptcp_pm_get_add_addr_signal_max(msk);
 539         local_addr_max = mptcp_pm_get_local_addr_max(msk);
 540         subflows_max = mptcp_pm_get_subflows_max(msk);
 541
 542         /* do lazy endpoint usage accounting for the MPC subflows */
 543         if (unlikely(!(msk->pm.status & BIT(MPTCP_PM_MPC_ENDPOINT_ACCOUNTED))) && msk->first) {
 544                 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(msk->first);
 545                 struct mptcp_pm_addr_entry *entry;
 546                 struct mptcp_addr_info mpc_addr;
 547                 bool backup = false;
 548
 549                 local_address((struct sock_common *)msk->first, &mpc_addr);
 550                 rcu_read_lock();
 551                 entry = __lookup_addr(pernet, &mpc_addr, false);
 552                 if (entry) {
 553                         __clear_bit(entry->addr.id, msk->pm.id_avail_bitmap);
 554                         msk->mpc_endpoint_id = entry->addr.id;
 555                         backup = !!(entry->flags & MPTCP_PM_ADDR_FLAG_BACKUP);
 556                 }
 557                 rcu_read_unlock();
 558
 559                 if (backup)
 560                         mptcp_pm_send_ack(msk, subflow, true, backup);
 561
 562                 msk->pm.status |= BIT(MPTCP_PM_MPC_ENDPOINT_ACCOUNTED);
 563         }
 564
 565         pr_debug("local %d:%d signal %d:%d subflows %d:%d\n",
 566                  msk->pm.local_addr_used, local_addr_max,
 567                  msk->pm.add_addr_signaled, add_addr_signal_max,
 568                  msk->pm.subflows, subflows_max);
 569
 570         /* check first for announce */
 571         if (msk->pm.add_addr_signaled < add_addr_signal_max) {
 572                 local = select_signal_address(pernet, msk);
 573
 574                 /* due to racing events on both ends we can reach here while
 575                  * previous add address is still running: if we invoke now
 576                  * mptcp_pm_announce_addr(), that will fail and the
 577                  * corresponding id will be marked as used.
 578                  * Instead let the PM machinery reschedule us when the
 579                  * current address announce will be completed.
 580                  */
 581                 if (msk->pm.addr_signal & BIT(MPTCP_ADD_ADDR_SIGNAL))
 582                         return;
 583
 584                 if (local) {
 585                         if (mptcp_pm_alloc_anno_list(msk, local)) {
 586                                 __clear_bit(local->addr.id, msk->pm.id_avail_bitmap);
 587                                 msk->pm.add_addr_signaled++;
 588                                 mptcp_pm_announce_addr(msk, &local->addr, false);
 589                                 mptcp_pm_nl_addr_send_ack(msk);
 590                         }
 591                 }
 592         }
 593
 594         /* check if should create a new subflow */
 595         while (msk->pm.local_addr_used < local_addr_max &&
 596                msk->pm.subflows < subflows_max) {
 597                 struct mptcp_addr_info addrs[MPTCP_PM_ADDR_MAX];
 598                 bool fullmesh;
 599                 int i, nr;
 600
 601                 local = select_local_address(pernet, msk);
 602                 if (!local)
 603                         break;
 604
 605                 fullmesh = !!(local->flags & MPTCP_PM_ADDR_FLAG_FULLMESH);
 606
 607                 msk->pm.local_addr_used++;
 608                 nr = fill_remote_addresses_vec(msk, &local->addr, fullmesh, addrs);
 609                 if (nr)
 610                         __clear_bit(local->addr.id, msk->pm.id_avail_bitmap);
 611                 spin_unlock_bh(&msk->pm.lock);
 612                 for (i = 0; i < nr; i++)
 613                         __mptcp_subflow_connect(sk, &local->addr, &addrs[i]);
 614                 spin_lock_bh(&msk->pm.lock);
 615         }
 616         mptcp_pm_nl_check_work_pending(msk);
 617 }
 618
 619 static void mptcp_pm_nl_fully_established(struct mptcp_sock *msk)
 620 {
 621         mptcp_pm_create_subflow_or_signal_addr(msk);
 622 }
 623
 624 static void mptcp_pm_nl_subflow_established(struct mptcp_sock *msk)
 625 {
 626         mptcp_pm_create_subflow_or_signal_addr(msk);
 627 }
 628
 629 /* Fill all the local addresses into the array addrs[],
 630  * and return the array size.
 631  */
 632 static unsigned int fill_local_addresses_vec(struct mptcp_sock *msk,
 633                                              struct mptcp_addr_info *addrs)
 634 {
 635         struct sock *sk = (struct sock *)msk;
 636         struct mptcp_pm_addr_entry *entry;
 637         struct mptcp_addr_info local;
 638         struct pm_nl_pernet *pernet;
 639         unsigned int subflows_max;
 640         int i = 0;
 641
 642         pernet = pm_nl_get_pernet_from_msk(msk);
 643         subflows_max = mptcp_pm_get_subflows_max(msk);
 644
 645         rcu_read_lock();
 646         list_for_each_entry_rcu(entry, &pernet->local_addr_list, list) {
 647                 if (!(entry->flags & MPTCP_PM_ADDR_FLAG_FULLMESH))
 648                         continue;
 649
 650                 if (entry->addr.family != sk->sk_family) {
 651 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
 652                         if ((entry->addr.family == AF_INET &&
 653                              !ipv6_addr_v4mapped(&sk->sk_v6_daddr)) ||
 654                             (sk->sk_family == AF_INET &&
 655                              !ipv6_addr_v4mapped(&entry->addr.addr6)))
 656 #endif
 657                                 continue;
 658                 }
 659
 660                 if (msk->pm.subflows < subflows_max) {
 661                         msk->pm.subflows++;
 662                         addrs[i++] = entry->addr;
 663                 }
 664         }
 665         rcu_read_unlock();
 666
 667         /* If the array is empty, fill in the single
 668          * 'IPADDRANY' local address
 669          */
 670         if (!i) {
 671                 memset(&local, 0, sizeof(local));
 672                 local.family = msk->pm.remote.family;
 673
 674                 msk->pm.subflows++;
 675                 addrs[i++] = local;
 676         }
 677
 678         return i;
 679 }
 680
 681 static void mptcp_pm_nl_add_addr_received(struct mptcp_sock *msk)
 682 {
 683         struct mptcp_addr_info addrs[MPTCP_PM_ADDR_MAX];
 684         struct sock *sk = (struct sock *)msk;
 685         unsigned int add_addr_accept_max;
 686         struct mptcp_addr_info remote;
 687         unsigned int subflows_max;
 688         int i, nr;
 689
 690         add_addr_accept_max = mptcp_pm_get_add_addr_accept_max(msk);
 691         subflows_max = mptcp_pm_get_subflows_max(msk);
 692
 693         pr_debug("accepted %d:%d remote family %d",
 694                  msk->pm.add_addr_accepted, add_addr_accept_max,
 695                  msk->pm.remote.family);
 696
 697         remote = msk->pm.remote;
 698         mptcp_pm_announce_addr(msk, &remote, true);
 699         mptcp_pm_nl_addr_send_ack(msk);
 700
 701         if (lookup_subflow_by_daddr(&msk->conn_list, &remote))
 702                 return;
 703
 704         /* pick id 0 port, if none is provided the remote address */
 705         if (!remote.port)
 706                 remote.port = sk->sk_dport;
 707
 708         /* connect to the specified remote address, using whatever
 709          * local address the routing configuration will pick.
 710          */
 711         nr = fill_local_addresses_vec(msk, addrs);
 712
 713         msk->pm.add_addr_accepted++;
 714         if (msk->pm.add_addr_accepted >= add_addr_accept_max ||
 715             msk->pm.subflows >= subflows_max)
 716                 WRITE_ONCE(msk->pm.accept_addr, false);
 717
 718         spin_unlock_bh(&msk->pm.lock);
 719         for (i = 0; i < nr; i++)
 720                 __mptcp_subflow_connect(sk, &addrs[i], &remote);
 721         spin_lock_bh(&msk->pm.lock);
 722 }
 723
 724 void mptcp_pm_nl_addr_send_ack(struct mptcp_sock *msk)
 725 {
 726         struct mptcp_subflow_context *subflow;
 727
 728         msk_owned_by_me(msk);
 729         lockdep_assert_held(&msk->pm.lock);
 730
 731         if (!mptcp_pm_should_add_signal(msk) &&
 732             !mptcp_pm_should_rm_signal(msk))
 733                 return;
 734
 735         subflow = list_first_entry_or_null(&msk->conn_list, typeof(*subflow), node);
 736         if (subflow)
 737                 mptcp_pm_send_ack(msk, subflow, false, false);
 738 }
 739
 740 int mptcp_pm_nl_mp_prio_send_ack(struct mptcp_sock *msk,
 741                                  struct mptcp_addr_info *addr,
 742                                  struct mptcp_addr_info *rem,
 743                                  u8 bkup)
 744 {
 745         struct mptcp_subflow_context *subflow;
 746
 747         pr_debug("bkup=%d", bkup);
 748
 749         mptcp_for_each_subflow(msk, subflow) {
 750                 struct sock *ssk = mptcp_subflow_tcp_sock(subflow);
 751                 struct mptcp_addr_info local, remote;
 752
 753                 local_address((struct sock_common *)ssk, &local);
 754                 if (!mptcp_addresses_equal(&local, addr, addr->port))
 755                         continue;
 756
 757                 if (rem && rem->family != AF_UNSPEC) {
 758                         remote_address((struct sock_common *)ssk, &remote);
 759                         if (!mptcp_addresses_equal(&remote, rem, rem->port))
 760                                 continue;
 761                 }
 762
 763                 __mptcp_pm_send_ack(msk, subflow, true, bkup);
 764                 return 0;
 765         }
 766
 767         return -EINVAL;
 768 }
 769
 770 static bool mptcp_local_id_match(const struct mptcp_sock *msk, u8 local_id, u8 id)
 771 {
 772         return local_id == id || (!local_id && msk->mpc_endpoint_id == id);
 773 }
 774
 775 static void mptcp_pm_nl_rm_addr_or_subflow(struct mptcp_sock *msk,
 776                                            const struct mptcp_rm_list *rm_list,
 777                                            enum linux_mptcp_mib_field rm_type)
 778 {
 779         struct mptcp_subflow_context *subflow, *tmp;
 780         struct sock *sk = (struct sock *)msk;
 781         u8 i;
 782
 783         pr_debug("%s rm_list_nr %d",
 784                  rm_type == MPTCP_MIB_RMADDR ? "address" : "subflow", rm_list->nr);
 785
 786         msk_owned_by_me(msk);
 787
 788         if (sk->sk_state == TCP_LISTEN)
 789                 return;
 790
 791         if (!rm_list->nr)
 792                 return;
 793
 794         if (list_empty(&msk->conn_list))
 795                 return;
 796
 797         for (i = 0; i < rm_list->nr; i++) {
 798                 u8 rm_id = rm_list->ids[i];
 799                 bool removed = false;
 800
 801                 mptcp_for_each_subflow_safe(msk, subflow, tmp) {
 802                         struct sock *ssk = mptcp_subflow_tcp_sock(subflow);
 803                         u8 remote_id = READ_ONCE(subflow->remote_id);
 804                         int how = RCV_SHUTDOWN | SEND_SHUTDOWN;
 805                         u8 id = subflow_get_local_id(subflow);
 806
 807                         if (rm_type == MPTCP_MIB_RMADDR && remote_id != rm_id)
 808                                 continue;
 809                         if (rm_type == MPTCP_MIB_RMSUBFLOW && !mptcp_local_id_match(msk, id, rm_id))
 810                                 continue;
 811
 812                         pr_debug(" -> %s rm_list_ids[%d]=%u local_id=%u remote_id=%u mpc_id=%u",
 813                                  rm_type == MPTCP_MIB_RMADDR ? "address" : "subflow",
 814                                  i, rm_id, id, remote_id, msk->mpc_endpoint_id);
 815                         spin_unlock_bh(&msk->pm.lock);
 816                         mptcp_subflow_shutdown(sk, ssk, how);
 817
 818                         /* the following takes care of updating the subflows counter */
 819                         mptcp_close_ssk(sk, ssk, subflow);
 820                         spin_lock_bh(&msk->pm.lock);
 821
 822                         removed = true;
 823                         __MPTCP_INC_STATS(sock_net(sk), rm_type);
 824                 }
 825                 if (rm_type == MPTCP_MIB_RMSUBFLOW)
 826                         __set_bit(rm_id ? rm_id : msk->mpc_endpoint_id, msk->pm.id_avail_bitmap);
 827                 if (!removed)
 828                         continue;
 829
 830                 if (!mptcp_pm_is_kernel(msk))
 831                         continue;
 832
 833                 if (rm_type == MPTCP_MIB_RMADDR) {
 834                         msk->pm.add_addr_accepted--;
 835                         WRITE_ONCE(msk->pm.accept_addr, true);
 836                 } else if (rm_type == MPTCP_MIB_RMSUBFLOW) {
 837                         msk->pm.local_addr_used--;
 838                 }
 839         }
 840 }
 841
 842 static void mptcp_pm_nl_rm_addr_received(struct mptcp_sock *msk)
 843 {
 844         mptcp_pm_nl_rm_addr_or_subflow(msk, &msk->pm.rm_list_rx, MPTCP_MIB_RMADDR);
 845 }
 846
 847 void mptcp_pm_nl_rm_subflow_received(struct mptcp_sock *msk,
 848                                      const struct mptcp_rm_list *rm_list)
 849 {
 850         mptcp_pm_nl_rm_addr_or_subflow(msk, rm_list, MPTCP_MIB_RMSUBFLOW);
 851 }
 852
 853 void mptcp_pm_nl_work(struct mptcp_sock *msk)
 854 {
 855         struct mptcp_pm_data *pm = &msk->pm;
 856
 857         msk_owned_by_me(msk);
 858
 859         if (!(pm->status & MPTCP_PM_WORK_MASK))
 860                 return;
 861
 862         spin_lock_bh(&msk->pm.lock);
 863
 864         pr_debug("msk=%p status=%x", msk, pm->status);
 865         if (pm->status & BIT(MPTCP_PM_ADD_ADDR_RECEIVED)) {
 866                 pm->status &= ~BIT(MPTCP_PM_ADD_ADDR_RECEIVED);
 867                 mptcp_pm_nl_add_addr_received(msk);
 868         }
 869         if (pm->status & BIT(MPTCP_PM_ADD_ADDR_SEND_ACK)) {
 870                 pm->status &= ~BIT(MPTCP_PM_ADD_ADDR_SEND_ACK);
 871                 mptcp_pm_nl_addr_send_ack(msk);
 872         }
 873         if (pm->status & BIT(MPTCP_PM_RM_ADDR_RECEIVED)) {
 874                 pm->status &= ~BIT(MPTCP_PM_RM_ADDR_RECEIVED);
 875                 mptcp_pm_nl_rm_addr_received(msk);
 876         }
 877         if (pm->status & BIT(MPTCP_PM_ESTABLISHED)) {
 878                 pm->status &= ~BIT(MPTCP_PM_ESTABLISHED);
 879                 mptcp_pm_nl_fully_established(msk);
 880         }
 881         if (pm->status & BIT(MPTCP_PM_SUBFLOW_ESTABLISHED)) {
 882                 pm->status &= ~BIT(MPTCP_PM_SUBFLOW_ESTABLISHED);
 883                 mptcp_pm_nl_subflow_established(msk);
 884         }
 885
 886         spin_unlock_bh(&msk->pm.lock);
 887 }
 888
 889 static bool address_use_port(struct mptcp_pm_addr_entry *entry)
 890 {
 891         return (entry->flags &
 892                 (MPTCP_PM_ADDR_FLAG_SIGNAL | MPTCP_PM_ADDR_FLAG_SUBFLOW)) ==
 893                 MPTCP_PM_ADDR_FLAG_SIGNAL;
 894 }
 895
 896 /* caller must ensure the RCU grace period is already elapsed */
 897 static void __mptcp_pm_release_addr_entry(struct mptcp_pm_addr_entry *entry)
 898 {
 899         if (entry->lsk)
 900                 sock_release(entry->lsk);
 901         kfree(entry);
 902 }
 903
 904 static int mptcp_pm_nl_append_new_local_addr(struct pm_nl_pernet *pernet,
 905                                              struct mptcp_pm_addr_entry *entry,
 906                                              bool needs_id)
 907 {
 908         struct mptcp_pm_addr_entry *cur, *del_entry = NULL;
 909         unsigned int addr_max;
 910         int ret = -EINVAL;
 911
 912         spin_lock_bh(&pernet->lock);
 913         /* to keep the code simple, don't do IDR-like allocation for address ID,
 914          * just bail when we exceed limits
 915          */
 916         if (pernet->next_id == MPTCP_PM_MAX_ADDR_ID)
 917                 pernet->next_id = 1;
 918         if (pernet->addrs >= MPTCP_PM_ADDR_MAX)
 919                 goto out;
 920         if (test_bit(entry->addr.id, pernet->id_bitmap))
 921                 goto out;
 922
 923         /* do not insert duplicate address, differentiate on port only
 924          * singled addresses
 925          */
 926         if (!address_use_port(entry))
 927                 entry->addr.port = 0;
 928         list_for_each_entry(cur, &pernet->local_addr_list, list) {
 929                 if (mptcp_addresses_equal(&cur->addr, &entry->addr,
 930                                           cur->addr.port || entry->addr.port)) {
 931                         /* allow replacing the exiting endpoint only if such
 932                          * endpoint is an implicit one and the user-space
 933                          * did not provide an endpoint id
 934                          */
 935                         if (!(cur->flags & MPTCP_PM_ADDR_FLAG_IMPLICIT))
 936                                 goto out;
 937                         if (entry->addr.id)
 938                                 goto out;
 939
 940                         pernet->addrs--;
 941                         entry->addr.id = cur->addr.id;
 942                         list_del_rcu(&cur->list);
 943                         del_entry = cur;
 944                         break;
 945                 }
 946         }
 947
 948         if (!entry->addr.id && needs_id) {
 949 find_next:
 950                 entry->addr.id = find_next_zero_bit(pernet->id_bitmap,
 951                                                     MPTCP_PM_MAX_ADDR_ID + 1,
 952                                                     pernet->next_id);
 953                 if (!entry->addr.id && pernet->next_id != 1) {
 954                         pernet->next_id = 1;
 955                         goto find_next;
 956                 }
 957         }
 958
 959         if (!entry->addr.id && needs_id)
 960                 goto out;
 961
 962         __set_bit(entry->addr.id, pernet->id_bitmap);
 963         if (entry->addr.id > pernet->next_id)
 964                 pernet->next_id = entry->addr.id;
 965
 966         if (entry->flags & MPTCP_PM_ADDR_FLAG_SIGNAL) {
 967                 addr_max = pernet->add_addr_signal_max;
 968                 WRITE_ONCE(pernet->add_addr_signal_max, addr_max + 1);
 969         }
 970         if (entry->flags & MPTCP_PM_ADDR_FLAG_SUBFLOW) {
 971                 addr_max = pernet->local_addr_max;
 972                 WRITE_ONCE(pernet->local_addr_max, addr_max + 1);
 973         }
 974
 975         pernet->addrs++;
 976         if (!entry->addr.port)
 977                 list_add_tail_rcu(&entry->list, &pernet->local_addr_list);
 978         else
 979                 list_add_rcu(&entry->list, &pernet->local_addr_list);
 980         ret = entry->addr.id;
 981
 982 out:
 983         spin_unlock_bh(&pernet->lock);
 984
 985         /* just replaced an existing entry, free it */
 986         if (del_entry) {
 987                 synchronize_rcu();
 988                 __mptcp_pm_release_addr_entry(del_entry);
 989         }
 990         return ret;
 991 }
 992
 993 static struct lock_class_key mptcp_slock_keys[2];
 994 static struct lock_class_key mptcp_keys[2];
 995
 996 static int mptcp_pm_nl_create_listen_socket(struct sock *sk,
 997                                             struct mptcp_pm_addr_entry *entry)
 998 {
 999         bool is_ipv6 = sk->sk_family == AF_INET6;
1000         int addrlen = sizeof(struct sockaddr_in);
1001         struct sockaddr_storage addr;
1002         struct socket *ssock;
1003         struct sock *newsk;
1004         int backlog = 1024;
1005         int err;
1006
1007         err = sock_create_kern(sock_net(sk), entry->addr.family,
1008                                SOCK_STREAM, IPPROTO_MPTCP, &entry->lsk);
1009         if (err)
1010                 return err;
1011
1012         newsk = entry->lsk->sk;
1013         if (!newsk)
1014                 return -EINVAL;
1015
1016         /* The subflow socket lock is acquired in a nested to the msk one
1017          * in several places, even by the TCP stack, and this msk is a kernel
1018          * socket: lockdep complains. Instead of propagating the _nested
1019          * modifiers in several places, re-init the lock class for the msk
1020          * socket to an mptcp specific one.
1021          */
1022         sock_lock_init_class_and_name(newsk,
1023                                       is_ipv6 ? "mlock-AF_INET6" : "mlock-AF_INET",
1024                                       &mptcp_slock_keys[is_ipv6],
1025                                       is_ipv6 ? "msk_lock-AF_INET6" : "msk_lock-AF_INET",
1026                                       &mptcp_keys[is_ipv6]);
1027
1028         lock_sock(newsk);
1029         ssock = __mptcp_nmpc_socket(mptcp_sk(newsk));
1030         release_sock(newsk);
1031         if (!ssock)
1032                 return -EINVAL;
1033
1034         mptcp_info2sockaddr(&entry->addr, &addr, entry->addr.family);
1035 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
1036         if (entry->addr.family == AF_INET6)
1037                 addrlen = sizeof(struct sockaddr_in6);
1038 #endif
1039         err = kernel_bind(ssock, (struct sockaddr *)&addr, addrlen);
1040         if (err) {
1041                 pr_warn("kernel_bind error, err=%d", err);
1042                 return err;
1043         }
1044
1045         inet_sk_state_store(newsk, TCP_LISTEN);
1046         err = kernel_listen(ssock, backlog);
1047         if (err) {
1048                 pr_warn("kernel_listen error, err=%d", err);
1049                 return err;
1050         }
1051
1052         return 0;
1053 }
1054
1055 int mptcp_pm_nl_get_local_id(struct mptcp_sock *msk, struct sock_common *skc)
1056 {
1057         struct mptcp_pm_addr_entry *entry;
1058         struct mptcp_addr_info skc_local;
1059         struct mptcp_addr_info msk_local;
1060         struct pm_nl_pernet *pernet;
1061         int ret = -1;
1062
1063         if (WARN_ON_ONCE(!msk))
1064                 return -1;
1065
1066         /* The 0 ID mapping is defined by the first subflow, copied into the msk
1067          * addr
1068          */
1069         local_address((struct sock_common *)msk, &msk_local);
1070         local_address((struct sock_common *)skc, &skc_local);
1071         if (mptcp_addresses_equal(&msk_local, &skc_local, false))
1072                 return 0;
1073
1074         if (mptcp_pm_is_userspace(msk))
1075                 return mptcp_userspace_pm_get_local_id(msk, &skc_local);
1076
1077         pernet = pm_nl_get_pernet_from_msk(msk);
1078
1079         rcu_read_lock();
1080         list_for_each_entry_rcu(entry, &pernet->local_addr_list, list) {
1081                 if (mptcp_addresses_equal(&entry->addr, &skc_local, entry->addr.port)) {
1082                         ret = entry->addr.id;
1083                         break;
1084                 }
1085         }
1086         rcu_read_unlock();
1087         if (ret >= 0)
1088                 return ret;
1089
1090         /* address not found, add to local list */
1091         entry = kmalloc(sizeof(*entry), GFP_ATOMIC);
1092         if (!entry)
1093                 return -ENOMEM;
1094
1095         entry->addr = skc_local;
1096         entry->addr.id = 0;
1097         entry->addr.port = 0;
1098         entry->ifindex = 0;
1099         entry->flags = MPTCP_PM_ADDR_FLAG_IMPLICIT;
1100         entry->lsk = NULL;
1101         ret = mptcp_pm_nl_append_new_local_addr(pernet, entry, true);
1102         if (ret < 0)
1103                 kfree(entry);
1104
1105         return ret;
1106 }
1107
1108 #define MPTCP_PM_CMD_GRP_OFFSET       0
1109 #define MPTCP_PM_EV_GRP_OFFSET        1
1110
1111 static const struct genl_multicast_group mptcp_pm_mcgrps[] = {
1112         [MPTCP_PM_CMD_GRP_OFFSET]       = { .name = MPTCP_PM_CMD_GRP_NAME, },
1113         [MPTCP_PM_EV_GRP_OFFSET]        = { .name = MPTCP_PM_EV_GRP_NAME,
1114                                             .flags = GENL_UNS_ADMIN_PERM,
1115                                           },
1116 };
1117
1118 static const struct nla_policy
1119 mptcp_pm_addr_policy[MPTCP_PM_ADDR_ATTR_MAX + 1] = {
1120         [MPTCP_PM_ADDR_ATTR_FAMILY]     = { .type       = NLA_U16,      },
1121         [MPTCP_PM_ADDR_ATTR_ID]         = { .type       = NLA_U8,       },
1122         [MPTCP_PM_ADDR_ATTR_ADDR4]      = { .type       = NLA_U32,      },
1123         [MPTCP_PM_ADDR_ATTR_ADDR6]      =
1124                 NLA_POLICY_EXACT_LEN(sizeof(struct in6_addr)),
1125         [MPTCP_PM_ADDR_ATTR_PORT]       = { .type       = NLA_U16       },
1126         [MPTCP_PM_ADDR_ATTR_FLAGS]      = { .type       = NLA_U32       },
1127         [MPTCP_PM_ADDR_ATTR_IF_IDX]     = { .type       = NLA_S32       },
1128 };
1129
1130 static const struct nla_policy mptcp_pm_policy[MPTCP_PM_ATTR_MAX + 1] = {
1131         [MPTCP_PM_ATTR_ADDR]            =
1132                                         NLA_POLICY_NESTED(mptcp_pm_addr_policy),
1133         [MPTCP_PM_ATTR_RCV_ADD_ADDRS]   = { .type       = NLA_U32,      },
1134         [MPTCP_PM_ATTR_SUBFLOWS]        = { .type       = NLA_U32,      },
1135         [MPTCP_PM_ATTR_TOKEN]           = { .type       = NLA_U32,      },
1136         [MPTCP_PM_ATTR_LOC_ID]          = { .type       = NLA_U8,       },
1137         [MPTCP_PM_ATTR_ADDR_REMOTE]     =
1138                                         NLA_POLICY_NESTED(mptcp_pm_addr_policy),
1139 };
1140
1141 void mptcp_pm_nl_subflow_chk_stale(const struct mptcp_sock *msk, struct sock *ssk)
1142 {
1143         struct mptcp_subflow_context *iter, *subflow = mptcp_subflow_ctx(ssk);
1144         struct sock *sk = (struct sock *)msk;
1145         unsigned int active_max_loss_cnt;
1146         struct net *net = sock_net(sk);
1147         unsigned int stale_loss_cnt;
1148         bool slow;
1149
1150         stale_loss_cnt = mptcp_stale_loss_cnt(net);
1151         if (subflow->stale || !stale_loss_cnt || subflow->stale_count <= stale_loss_cnt)
1152                 return;
1153
1154         /* look for another available subflow not in loss state */
1155         active_max_loss_cnt = max_t(int, stale_loss_cnt - 1, 1);
1156         mptcp_for_each_subflow(msk, iter) {
1157                 if (iter != subflow && mptcp_subflow_active(iter) &&
1158                     iter->stale_count < active_max_loss_cnt) {
1159                         /* we have some alternatives, try to mark this subflow as idle ...*/
1160                         slow = lock_sock_fast(ssk);
1161                         if (!tcp_rtx_and_write_queues_empty(ssk)) {
1162                                 subflow->stale = 1;
1163                                 __mptcp_retransmit_pending_data(sk);
1164                                 MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_SUBFLOWSTALE);
1165                         }
1166                         unlock_sock_fast(ssk, slow);
1167
1168                         /* always try to push the pending data regardless of re-injections:
1169                          * we can possibly use backup subflows now, and subflow selection
1170                          * is cheap under the msk socket lock
1171                          */
1172                         __mptcp_push_pending(sk, 0);
1173                         return;
1174                 }
1175         }
1176 }
1177
1178 static int mptcp_pm_family_to_addr(int family)
1179 {
1180 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
1181         if (family == AF_INET6)
1182                 return MPTCP_PM_ADDR_ATTR_ADDR6;
1183 #endif
1184         return MPTCP_PM_ADDR_ATTR_ADDR4;
1185 }
1186
1187 static int mptcp_pm_parse_pm_addr_attr(struct nlattr *tb[],
1188                                        const struct nlattr *attr,
1189                                        struct genl_info *info,
1190                                        struct mptcp_addr_info *addr,
1191                                        bool require_family)
1192 {
1193         int err, addr_addr;
1194
1195         if (!attr) {
1196                 GENL_SET_ERR_MSG(info, "missing address info");
1197                 return -EINVAL;
1198         }
1199
1200         /* no validation needed - was already done via nested policy */
1201         err = nla_parse_nested_deprecated(tb, MPTCP_PM_ADDR_ATTR_MAX, attr,
1202                                           mptcp_pm_addr_policy, info->extack);
1203         if (err)
1204                 return err;
1205
1206         if (tb[MPTCP_PM_ADDR_ATTR_ID])
1207                 addr->id = nla_get_u8(tb[MPTCP_PM_ADDR_ATTR_ID]);
1208
1209         if (!tb[MPTCP_PM_ADDR_ATTR_FAMILY]) {
1210                 if (!require_family)
1211                         return err;
1212
1213                 NL_SET_ERR_MSG_ATTR(info->extack, attr,
1214                                     "missing family");
1215                 return -EINVAL;
1216         }
1217
1218         addr->family = nla_get_u16(tb[MPTCP_PM_ADDR_ATTR_FAMILY]);
1219         if (addr->family != AF_INET
1220 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
1221             && addr->family != AF_INET6
1222 #endif
1223             ) {
1224                 NL_SET_ERR_MSG_ATTR(info->extack, attr,
1225                                     "unknown address family");
1226                 return -EINVAL;
1227         }
1228         addr_addr = mptcp_pm_family_to_addr(addr->family);
1229         if (!tb[addr_addr]) {
1230                 NL_SET_ERR_MSG_ATTR(info->extack, attr,
1231                                     "missing address data");
1232                 return -EINVAL;
1233         }
1234
1235 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
1236         if (addr->family == AF_INET6)
1237                 addr->addr6 = nla_get_in6_addr(tb[addr_addr]);
1238         else
1239 #endif
1240                 addr->addr.s_addr = nla_get_in_addr(tb[addr_addr]);
1241
1242         if (tb[MPTCP_PM_ADDR_ATTR_PORT])
1243                 addr->port = htons(nla_get_u16(tb[MPTCP_PM_ADDR_ATTR_PORT]));
1244
1245         return err;
1246 }
1247
1248 int mptcp_pm_parse_addr(struct nlattr *attr, struct genl_info *info,
1249                         struct mptcp_addr_info *addr)
1250 {
1251         struct nlattr *tb[MPTCP_PM_ADDR_ATTR_MAX + 1];
1252
1253         memset(addr, 0, sizeof(*addr));
1254
1255         return mptcp_pm_parse_pm_addr_attr(tb, attr, info, addr, true);
1256 }
1257
1258 int mptcp_pm_parse_entry(struct nlattr *attr, struct genl_info *info,
1259                          bool require_family,
1260                          struct mptcp_pm_addr_entry *entry)
1261 {
1262         struct nlattr *tb[MPTCP_PM_ADDR_ATTR_MAX + 1];
1263         int err;
1264
1265         memset(entry, 0, sizeof(*entry));
1266
1267         err = mptcp_pm_parse_pm_addr_attr(tb, attr, info, &entry->addr, require_family);
1268         if (err)
1269                 return err;
1270
1271         if (tb[MPTCP_PM_ADDR_ATTR_IF_IDX]) {
1272                 u32 val = nla_get_s32(tb[MPTCP_PM_ADDR_ATTR_IF_IDX]);
1273
1274                 entry->ifindex = val;
1275         }
1276
1277         if (tb[MPTCP_PM_ADDR_ATTR_FLAGS])
1278                 entry->flags = nla_get_u32(tb[MPTCP_PM_ADDR_ATTR_FLAGS]);
1279
1280         if (tb[MPTCP_PM_ADDR_ATTR_PORT])
1281                 entry->addr.port = htons(nla_get_u16(tb[MPTCP_PM_ADDR_ATTR_PORT]));
1282
1283         return 0;
1284 }
1285
1286 static struct pm_nl_pernet *genl_info_pm_nl(struct genl_info *info)
1287 {
1288         return pm_nl_get_pernet(genl_info_net(info));
1289 }
1290
1291 static int mptcp_nl_add_subflow_or_signal_addr(struct net *net)
1292 {
1293         struct mptcp_sock *msk;
1294         long s_slot = 0, s_num = 0;
1295
1296         while ((msk = mptcp_token_iter_next(net, &s_slot, &s_num)) != NULL) {
1297                 struct sock *sk = (struct sock *)msk;
1298
1299                 if (!READ_ONCE(msk->fully_established) ||
1300                     mptcp_pm_is_userspace(msk))
1301                         goto next;
1302
1303                 lock_sock(sk);
1304                 spin_lock_bh(&msk->pm.lock);
1305                 mptcp_pm_create_subflow_or_signal_addr(msk);
1306                 spin_unlock_bh(&msk->pm.lock);
1307                 release_sock(sk);
1308
1309 next:
1310                 sock_put(sk);
1311                 cond_resched();
1312         }
1313
1314         return 0;
1315 }
1316
1317 static bool mptcp_pm_has_addr_attr_id(const struct nlattr *attr,
1318                                       struct genl_info *info)
1319 {
1320         struct nlattr *tb[MPTCP_PM_ADDR_ATTR_MAX + 1];
1321
1322         if (!nla_parse_nested_deprecated(tb, MPTCP_PM_ADDR_ATTR_MAX, attr,
1323                                          mptcp_pm_addr_policy, info->extack) &&
1324             tb[MPTCP_PM_ADDR_ATTR_ID])
1325                 return true;
1326         return false;
1327 }
1328
1329 static int mptcp_nl_cmd_add_addr(struct sk_buff *skb, struct genl_info *info)
1330 {
1331         struct nlattr *attr = info->attrs[MPTCP_PM_ATTR_ADDR];
1332         struct pm_nl_pernet *pernet = genl_info_pm_nl(info);
1333         struct mptcp_pm_addr_entry addr, *entry;
1334         int ret;
1335
1336         ret = mptcp_pm_parse_entry(attr, info, true, &addr);
1337         if (ret < 0)
1338                 return ret;
1339
1340         if (addr.addr.port && !(addr.flags & MPTCP_PM_ADDR_FLAG_SIGNAL)) {
1341                 GENL_SET_ERR_MSG(info, "flags must have signal when using port");
1342                 return -EINVAL;
1343         }
1344
1345         if (addr.flags & MPTCP_PM_ADDR_FLAG_SIGNAL &&
1346             addr.flags & MPTCP_PM_ADDR_FLAG_FULLMESH) {
1347                 GENL_SET_ERR_MSG(info, "flags mustn't have both signal and fullmesh");
1348                 return -EINVAL;
1349         }
1350
1351         if (addr.flags & MPTCP_PM_ADDR_FLAG_IMPLICIT) {
1352                 GENL_SET_ERR_MSG(info, "can't create IMPLICIT endpoint");
1353                 return -EINVAL;
1354         }
1355
1356         entry = kzalloc(sizeof(*entry), GFP_KERNEL_ACCOUNT);
1357         if (!entry) {
1358                 GENL_SET_ERR_MSG(info, "can't allocate addr");
1359                 return -ENOMEM;
1360         }
1361
1362         *entry = addr;
1363         if (entry->addr.port) {
1364                 ret = mptcp_pm_nl_create_listen_socket(skb->sk, entry);
1365                 if (ret) {
1366                         GENL_SET_ERR_MSG(info, "create listen socket error");
1367                         goto out_free;
1368                 }
1369         }
1370         ret = mptcp_pm_nl_append_new_local_addr(pernet, entry,
1371                                                 !mptcp_pm_has_addr_attr_id(attr, info));
1372         if (ret < 0) {
1373                 GENL_SET_ERR_MSG(info, "too many addresses or duplicate one");
1374                 goto out_free;
1375         }
1376
1377         mptcp_nl_add_subflow_or_signal_addr(sock_net(skb->sk));
1378         return 0;
1379
1380 out_free:
1381         __mptcp_pm_release_addr_entry(entry);
1382         return ret;
1383 }
1384
1385 int mptcp_pm_get_flags_and_ifindex_by_id(struct mptcp_sock *msk, unsigned int id,
1386                                          u8 *flags, int *ifindex)
1387 {
1388         struct mptcp_pm_addr_entry *entry;
1389         struct sock *sk = (struct sock *)msk;
1390         struct net *net = sock_net(sk);
1391
1392         *flags = 0;
1393         *ifindex = 0;
1394
1395         if (id) {
1396                 if (mptcp_pm_is_userspace(msk))
1397                         return mptcp_userspace_pm_get_flags_and_ifindex_by_id(msk,
1398                                                                               id,
1399                                                                               flags,
1400                                                                               ifindex);
1401
1402                 rcu_read_lock();
1403                 entry = __lookup_addr_by_id(pm_nl_get_pernet(net), id);
1404                 if (entry) {
1405                         *flags = entry->flags;
1406                         *ifindex = entry->ifindex;
1407                 }
1408                 rcu_read_unlock();
1409         }
1410
1411         return 0;
1412 }
1413
1414 static bool remove_anno_list_by_saddr(struct mptcp_sock *msk,
1415                                       const struct mptcp_addr_info *addr)
1416 {
1417         struct mptcp_pm_add_entry *entry;
1418
1419         entry = mptcp_pm_del_add_timer(msk, addr, false);
1420         if (entry) {
1421                 list_del(&entry->list);
1422                 kfree(entry);
1423                 return true;
1424         }
1425
1426         return false;
1427 }
1428
1429 static bool mptcp_pm_remove_anno_addr(struct mptcp_sock *msk,
1430                                       const struct mptcp_addr_info *addr,
1431                                       bool force)
1432 {
1433         struct mptcp_rm_list list = { .nr = 0 };
1434         bool ret;
1435
1436         list.ids[list.nr++] = addr->id;
1437
1438         ret = remove_anno_list_by_saddr(msk, addr);
1439         if (ret || force) {
1440                 spin_lock_bh(&msk->pm.lock);
1441                 mptcp_pm_remove_addr(msk, &list);
1442                 spin_unlock_bh(&msk->pm.lock);
1443         }
1444         return ret;
1445 }
1446
1447 static int mptcp_nl_remove_subflow_and_signal_addr(struct net *net,
1448                                                    const struct mptcp_pm_addr_entry *entry)
1449 {
1450         const struct mptcp_addr_info *addr = &entry->addr;
1451         struct mptcp_rm_list list = { .nr = 0 };
1452         long s_slot = 0, s_num = 0;
1453         struct mptcp_sock *msk;
1454
1455         pr_debug("remove_id=%d", addr->id);
1456
1457         list.ids[list.nr++] = addr->id;
1458
1459         while ((msk = mptcp_token_iter_next(net, &s_slot, &s_num)) != NULL) {
1460                 struct sock *sk = (struct sock *)msk;
1461                 bool remove_subflow;
1462
1463                 if (mptcp_pm_is_userspace(msk))
1464                         goto next;
1465
1466                 if (list_empty(&msk->conn_list)) {
1467                         mptcp_pm_remove_anno_addr(msk, addr, false);
1468                         goto next;
1469                 }
1470
1471                 lock_sock(sk);
1472                 remove_subflow = lookup_subflow_by_saddr(&msk->conn_list, addr);
1473                 mptcp_pm_remove_anno_addr(msk, addr, remove_subflow &&
1474                                           !(entry->flags & MPTCP_PM_ADDR_FLAG_IMPLICIT));
1475                 if (remove_subflow)
1476                         mptcp_pm_remove_subflow(msk, &list);
1477                 release_sock(sk);
1478
1479 next:
1480                 sock_put(sk);
1481                 cond_resched();
1482         }
1483
1484         return 0;
1485 }
1486
1487 static int mptcp_nl_remove_id_zero_address(struct net *net,
1488                                            struct mptcp_addr_info *addr)
1489 {
1490         struct mptcp_rm_list list = { .nr = 0 };
1491         long s_slot = 0, s_num = 0;
1492         struct mptcp_sock *msk;
1493
1494         list.ids[list.nr++] = 0;
1495
1496         while ((msk = mptcp_token_iter_next(net, &s_slot, &s_num)) != NULL) {
1497                 struct sock *sk = (struct sock *)msk;
1498                 struct mptcp_addr_info msk_local;
1499
1500                 if (list_empty(&msk->conn_list) || mptcp_pm_is_userspace(msk))
1501                         goto next;
1502
1503                 local_address((struct sock_common *)msk, &msk_local);
1504                 if (!mptcp_addresses_equal(&msk_local, addr, addr->port))
1505                         goto next;
1506
1507                 lock_sock(sk);
1508                 spin_lock_bh(&msk->pm.lock);
1509                 mptcp_pm_remove_addr(msk, &list);
1510                 mptcp_pm_nl_rm_subflow_received(msk, &list);
1511                 spin_unlock_bh(&msk->pm.lock);
1512                 release_sock(sk);
1513
1514 next:
1515                 sock_put(sk);
1516                 cond_resched();
1517         }
1518
1519         return 0;
1520 }
1521
1522 static int mptcp_nl_cmd_del_addr(struct sk_buff *skb, struct genl_info *info)
1523 {
1524         struct nlattr *attr = info->attrs[MPTCP_PM_ATTR_ADDR];
1525         struct pm_nl_pernet *pernet = genl_info_pm_nl(info);
1526         struct mptcp_pm_addr_entry addr, *entry;
1527         unsigned int addr_max;
1528         int ret;
1529
1530         ret = mptcp_pm_parse_entry(attr, info, false, &addr);
1531         if (ret < 0)
1532                 return ret;
1533
1534         /* the zero id address is special: the first address used by the msk
1535          * always gets such an id, so different subflows can have different zero
1536          * id addresses. Additionally zero id is not accounted for in id_bitmap.
1537          * Let's use an 'mptcp_rm_list' instead of the common remove code.
1538          */
1539         if (addr.addr.id == 0)
1540                 return mptcp_nl_remove_id_zero_address(sock_net(skb->sk), &addr.addr);
1541
1542         spin_lock_bh(&pernet->lock);
1543         entry = __lookup_addr_by_id(pernet, addr.addr.id);
1544         if (!entry) {
1545                 GENL_SET_ERR_MSG(info, "address not found");
1546                 spin_unlock_bh(&pernet->lock);
1547                 return -EINVAL;
1548         }
1549         if (entry->flags & MPTCP_PM_ADDR_FLAG_SIGNAL) {
1550                 addr_max = pernet->add_addr_signal_max;
1551                 WRITE_ONCE(pernet->add_addr_signal_max, addr_max - 1);
1552         }
1553         if (entry->flags & MPTCP_PM_ADDR_FLAG_SUBFLOW) {
1554                 addr_max = pernet->local_addr_max;
1555                 WRITE_ONCE(pernet->local_addr_max, addr_max - 1);
1556         }
1557
1558         pernet->addrs--;
1559         list_del_rcu(&entry->list);
1560         __clear_bit(entry->addr.id, pernet->id_bitmap);
1561         spin_unlock_bh(&pernet->lock);
1562
1563         mptcp_nl_remove_subflow_and_signal_addr(sock_net(skb->sk), entry);
1564         synchronize_rcu();
1565         __mptcp_pm_release_addr_entry(entry);
1566
1567         return ret;
1568 }
1569
1570 void mptcp_pm_remove_addrs(struct mptcp_sock *msk, struct list_head *rm_list)
1571 {
1572         struct mptcp_rm_list alist = { .nr = 0 };
1573         struct mptcp_pm_addr_entry *entry;
1574
1575         list_for_each_entry(entry, rm_list, list) {
1576                 if ((remove_anno_list_by_saddr(msk, &entry->addr) ||
1577                      lookup_subflow_by_saddr(&msk->conn_list, &entry->addr)) &&
1578                     alist.nr < MPTCP_RM_IDS_MAX)
1579                         alist.ids[alist.nr++] = entry->addr.id;
1580         }
1581
1582         if (alist.nr) {
1583                 spin_lock_bh(&msk->pm.lock);
1584                 mptcp_pm_remove_addr(msk, &alist);
1585                 spin_unlock_bh(&msk->pm.lock);
1586         }
1587 }
1588
1589 void mptcp_pm_remove_addrs_and_subflows(struct mptcp_sock *msk,
1590                                         struct list_head *rm_list)
1591 {
1592         struct mptcp_rm_list alist = { .nr = 0 }, slist = { .nr = 0 };
1593         struct mptcp_pm_addr_entry *entry;
1594
1595         list_for_each_entry(entry, rm_list, list) {
1596                 if (lookup_subflow_by_saddr(&msk->conn_list, &entry->addr) &&
1597                     slist.nr < MPTCP_RM_IDS_MAX)
1598                         slist.ids[slist.nr++] = entry->addr.id;
1599
1600                 if (remove_anno_list_by_saddr(msk, &entry->addr) &&
1601                     alist.nr < MPTCP_RM_IDS_MAX)
1602                         alist.ids[alist.nr++] = entry->addr.id;
1603         }
1604
1605         if (alist.nr) {
1606                 spin_lock_bh(&msk->pm.lock);
1607                 mptcp_pm_remove_addr(msk, &alist);
1608                 spin_unlock_bh(&msk->pm.lock);
1609         }
1610         if (slist.nr)
1611                 mptcp_pm_remove_subflow(msk, &slist);
1612 }
1613
1614 static void mptcp_nl_remove_addrs_list(struct net *net,
1615                                        struct list_head *rm_list)
1616 {
1617         long s_slot = 0, s_num = 0;
1618         struct mptcp_sock *msk;
1619
1620         if (list_empty(rm_list))
1621                 return;
1622
1623         while ((msk = mptcp_token_iter_next(net, &s_slot, &s_num)) != NULL) {
1624                 struct sock *sk = (struct sock *)msk;
1625
1626                 if (!mptcp_pm_is_userspace(msk)) {
1627                         lock_sock(sk);
1628                         mptcp_pm_remove_addrs_and_subflows(msk, rm_list);
1629                         release_sock(sk);
1630                 }
1631
1632                 sock_put(sk);
1633                 cond_resched();
1634         }
1635 }
1636
1637 /* caller must ensure the RCU grace period is already elapsed */
1638 static void __flush_addrs(struct list_head *list)
1639 {
1640         while (!list_empty(list)) {
1641                 struct mptcp_pm_addr_entry *cur;
1642
1643                 cur = list_entry(list->next,
1644                                  struct mptcp_pm_addr_entry, list);
1645                 list_del_rcu(&cur->list);
1646                 __mptcp_pm_release_addr_entry(cur);
1647         }
1648 }
1649
1650 static void __reset_counters(struct pm_nl_pernet *pernet)
1651 {
1652         WRITE_ONCE(pernet->add_addr_signal_max, 0);
1653         WRITE_ONCE(pernet->add_addr_accept_max, 0);
1654         WRITE_ONCE(pernet->local_addr_max, 0);
1655         pernet->addrs = 0;
1656 }
1657
1658 static int mptcp_nl_cmd_flush_addrs(struct sk_buff *skb, struct genl_info *info)
1659 {
1660         struct pm_nl_pernet *pernet = genl_info_pm_nl(info);
1661         LIST_HEAD(free_list);
1662
1663         spin_lock_bh(&pernet->lock);
1664         list_splice_init(&pernet->local_addr_list, &free_list);
1665         __reset_counters(pernet);
1666         pernet->next_id = 1;
1667         bitmap_zero(pernet->id_bitmap, MPTCP_PM_MAX_ADDR_ID + 1);
1668         spin_unlock_bh(&pernet->lock);
1669         mptcp_nl_remove_addrs_list(sock_net(skb->sk), &free_list);
1670         synchronize_rcu();
1671         __flush_addrs(&free_list);
1672         return 0;
1673 }
1674
1675 static int mptcp_nl_fill_addr(struct sk_buff *skb,
1676                               struct mptcp_pm_addr_entry *entry)
1677 {
1678         struct mptcp_addr_info *addr = &entry->addr;
1679         struct nlattr *attr;
1680
1681         attr = nla_nest_start(skb, MPTCP_PM_ATTR_ADDR);
1682         if (!attr)
1683                 return -EMSGSIZE;
1684
1685         if (nla_put_u16(skb, MPTCP_PM_ADDR_ATTR_FAMILY, addr->family))
1686                 goto nla_put_failure;
1687         if (nla_put_u16(skb, MPTCP_PM_ADDR_ATTR_PORT, ntohs(addr->port)))
1688                 goto nla_put_failure;
1689         if (nla_put_u8(skb, MPTCP_PM_ADDR_ATTR_ID, addr->id))
1690                 goto nla_put_failure;
1691         if (nla_put_u32(skb, MPTCP_PM_ADDR_ATTR_FLAGS, entry->flags))
1692                 goto nla_put_failure;
1693         if (entry->ifindex &&
1694             nla_put_s32(skb, MPTCP_PM_ADDR_ATTR_IF_IDX, entry->ifindex))
1695                 goto nla_put_failure;
1696
1697         if (addr->family == AF_INET &&
1698             nla_put_in_addr(skb, MPTCP_PM_ADDR_ATTR_ADDR4,
1699                             addr->addr.s_addr))
1700                 goto nla_put_failure;
1701 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
1702         else if (addr->family == AF_INET6 &&
1703                  nla_put_in6_addr(skb, MPTCP_PM_ADDR_ATTR_ADDR6, &addr->addr6))
1704                 goto nla_put_failure;
1705 #endif
1706         nla_nest_end(skb, attr);
1707         return 0;
1708
1709 nla_put_failure:
1710         nla_nest_cancel(skb, attr);
1711         return -EMSGSIZE;
1712 }
1713
1714 static int mptcp_nl_cmd_get_addr(struct sk_buff *skb, struct genl_info *info)
1715 {
1716         struct nlattr *attr = info->attrs[MPTCP_PM_ATTR_ADDR];
1717         struct pm_nl_pernet *pernet = genl_info_pm_nl(info);
1718         struct mptcp_pm_addr_entry addr, *entry;
1719         struct sk_buff *msg;
1720         void *reply;
1721         int ret;
1722
1723         ret = mptcp_pm_parse_entry(attr, info, false, &addr);
1724         if (ret < 0)
1725                 return ret;
1726
1727         msg = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
1728         if (!msg)
1729                 return -ENOMEM;
1730
1731         reply = genlmsg_put_reply(msg, info, &mptcp_genl_family, 0,
1732                                   info->genlhdr->cmd);
1733         if (!reply) {
1734                 GENL_SET_ERR_MSG(info, "not enough space in Netlink message");
1735                 ret = -EMSGSIZE;
1736                 goto fail;
1737         }
1738
1739         spin_lock_bh(&pernet->lock);
1740         entry = __lookup_addr_by_id(pernet, addr.addr.id);
1741         if (!entry) {
1742                 GENL_SET_ERR_MSG(info, "address not found");
1743                 ret = -EINVAL;
1744                 goto unlock_fail;
1745         }
1746
1747         ret = mptcp_nl_fill_addr(msg, entry);
1748         if (ret)
1749                 goto unlock_fail;
1750
1751         genlmsg_end(msg, reply);
1752         ret = genlmsg_reply(msg, info);
1753         spin_unlock_bh(&pernet->lock);
1754         return ret;
1755
1756 unlock_fail:
1757         spin_unlock_bh(&pernet->lock);
1758
1759 fail:
1760         nlmsg_free(msg);
1761         return ret;
1762 }
1763
1764 static int mptcp_nl_cmd_dump_addrs(struct sk_buff *msg,
1765                                    struct netlink_callback *cb)
1766 {
1767         struct net *net = sock_net(msg->sk);
1768         struct mptcp_pm_addr_entry *entry;
1769         struct pm_nl_pernet *pernet;
1770         int id = cb->args[0];
1771         void *hdr;
1772         int i;
1773
1774         pernet = pm_nl_get_pernet(net);
1775
1776         spin_lock_bh(&pernet->lock);
1777         for (i = id; i < MPTCP_PM_MAX_ADDR_ID + 1; i++) {
1778                 if (test_bit(i, pernet->id_bitmap)) {
1779                         entry = __lookup_addr_by_id(pernet, i);
1780                         if (!entry)
1781                                 break;
1782
1783                         if (entry->addr.id <= id)
1784                                 continue;
1785
1786                         hdr = genlmsg_put(msg, NETLINK_CB(cb->skb).portid,
1787                                           cb->nlh->nlmsg_seq, &mptcp_genl_family,
1788                                           NLM_F_MULTI, MPTCP_PM_CMD_GET_ADDR);
1789                         if (!hdr)
1790                                 break;
1791
1792                         if (mptcp_nl_fill_addr(msg, entry) < 0) {
1793                                 genlmsg_cancel(msg, hdr);
1794                                 break;
1795                         }
1796
1797                         id = entry->addr.id;
1798                         genlmsg_end(msg, hdr);
1799                 }
1800         }
1801         spin_unlock_bh(&pernet->lock);
1802
1803         cb->args[0] = id;
1804         return msg->len;
1805 }
1806
1807 static int parse_limit(struct genl_info *info, int id, unsigned int *limit)
1808 {
1809         struct nlattr *attr = info->attrs[id];
1810
1811         if (!attr)
1812                 return 0;
1813
1814         *limit = nla_get_u32(attr);
1815         if (*limit > MPTCP_PM_ADDR_MAX) {
1816                 GENL_SET_ERR_MSG(info, "limit greater than maximum");
1817                 return -EINVAL;
1818         }
1819         return 0;
1820 }
1821
1822 static int
1823 mptcp_nl_cmd_set_limits(struct sk_buff *skb, struct genl_info *info)
1824 {
1825         struct pm_nl_pernet *pernet = genl_info_pm_nl(info);
1826         unsigned int rcv_addrs, subflows;
1827         int ret;
1828
1829         spin_lock_bh(&pernet->lock);
1830         rcv_addrs = pernet->add_addr_accept_max;
1831         ret = parse_limit(info, MPTCP_PM_ATTR_RCV_ADD_ADDRS, &rcv_addrs);
1832         if (ret)
1833                 goto unlock;
1834
1835         subflows = pernet->subflows_max;
1836         ret = parse_limit(info, MPTCP_PM_ATTR_SUBFLOWS, &subflows);
1837         if (ret)
1838                 goto unlock;
1839
1840         WRITE_ONCE(pernet->add_addr_accept_max, rcv_addrs);
1841         WRITE_ONCE(pernet->subflows_max, subflows);
1842
1843 unlock:
1844         spin_unlock_bh(&pernet->lock);
1845         return ret;
1846 }
1847
1848 static int
1849 mptcp_nl_cmd_get_limits(struct sk_buff *skb, struct genl_info *info)
1850 {
1851         struct pm_nl_pernet *pernet = genl_info_pm_nl(info);
1852         struct sk_buff *msg;
1853         void *reply;
1854
1855         msg = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
1856         if (!msg)
1857                 return -ENOMEM;
1858
1859         reply = genlmsg_put_reply(msg, info, &mptcp_genl_family, 0,
1860                                   MPTCP_PM_CMD_GET_LIMITS);
1861         if (!reply)
1862                 goto fail;
1863
1864         if (nla_put_u32(msg, MPTCP_PM_ATTR_RCV_ADD_ADDRS,
1865                         READ_ONCE(pernet->add_addr_accept_max)))
1866                 goto fail;
1867
1868         if (nla_put_u32(msg, MPTCP_PM_ATTR_SUBFLOWS,
1869                         READ_ONCE(pernet->subflows_max)))
1870                 goto fail;
1871
1872         genlmsg_end(msg, reply);
1873         return genlmsg_reply(msg, info);
1874
1875 fail:
1876         GENL_SET_ERR_MSG(info, "not enough space in Netlink message");
1877         nlmsg_free(msg);
1878         return -EMSGSIZE;
1879 }
1880
1881 static void mptcp_pm_nl_fullmesh(struct mptcp_sock *msk,
1882                                  struct mptcp_addr_info *addr)
1883 {
1884         struct mptcp_rm_list list = { .nr = 0 };
1885
1886         list.ids[list.nr++] = addr->id;
1887
1888         spin_lock_bh(&msk->pm.lock);
1889         mptcp_pm_nl_rm_subflow_received(msk, &list);
1890         mptcp_pm_create_subflow_or_signal_addr(msk);
1891         spin_unlock_bh(&msk->pm.lock);
1892 }
1893
1894 static int mptcp_nl_set_flags(struct net *net,
1895                               struct mptcp_addr_info *addr,
1896                               u8 bkup, u8 changed)
1897 {
1898         long s_slot = 0, s_num = 0;
1899         struct mptcp_sock *msk;
1900         int ret = -EINVAL;
1901
1902         while ((msk = mptcp_token_iter_next(net, &s_slot, &s_num)) != NULL) {
1903                 struct sock *sk = (struct sock *)msk;
1904
1905                 if (list_empty(&msk->conn_list) || mptcp_pm_is_userspace(msk))
1906                         goto next;
1907
1908                 lock_sock(sk);
1909                 if (changed & MPTCP_PM_ADDR_FLAG_BACKUP)
1910                         ret = mptcp_pm_nl_mp_prio_send_ack(msk, addr, NULL, bkup);
1911                 if (changed & MPTCP_PM_ADDR_FLAG_FULLMESH)
1912                         mptcp_pm_nl_fullmesh(msk, addr);
1913                 release_sock(sk);
1914
1915 next:
1916                 sock_put(sk);
1917                 cond_resched();
1918         }
1919
1920         return ret;
1921 }
1922
1923 static int mptcp_nl_cmd_set_flags(struct sk_buff *skb, struct genl_info *info)
1924 {
1925         struct mptcp_pm_addr_entry addr = { .addr = { .family = AF_UNSPEC }, }, *entry;
1926         struct mptcp_pm_addr_entry remote = { .addr = { .family = AF_UNSPEC }, };
1927         struct nlattr *attr_rem = info->attrs[MPTCP_PM_ATTR_ADDR_REMOTE];
1928         struct nlattr *token = info->attrs[MPTCP_PM_ATTR_TOKEN];
1929         struct nlattr *attr = info->attrs[MPTCP_PM_ATTR_ADDR];
1930         struct pm_nl_pernet *pernet = genl_info_pm_nl(info);
1931         u8 changed, mask = MPTCP_PM_ADDR_FLAG_BACKUP |
1932                            MPTCP_PM_ADDR_FLAG_FULLMESH;
1933         struct net *net = sock_net(skb->sk);
1934         u8 bkup = 0, lookup_by_id = 0;
1935         int ret;
1936
1937         ret = mptcp_pm_parse_entry(attr, info, false, &addr);
1938         if (ret < 0)
1939                 return ret;
1940
1941         if (attr_rem) {
1942                 ret = mptcp_pm_parse_entry(attr_rem, info, false, &remote);
1943                 if (ret < 0)
1944                         return ret;
1945         }
1946
1947         if (addr.flags & MPTCP_PM_ADDR_FLAG_BACKUP)
1948                 bkup = 1;
1949         if (addr.addr.family == AF_UNSPEC) {
1950                 lookup_by_id = 1;
1951                 if (!addr.addr.id)
1952                         return -EOPNOTSUPP;
1953         }
1954
1955         if (token)
1956                 return mptcp_userspace_pm_set_flags(sock_net(skb->sk),
1957                                                     token, &addr, &remote, bkup);
1958
1959         spin_lock_bh(&pernet->lock);
1960         entry = __lookup_addr(pernet, &addr.addr, lookup_by_id);
1961         if (!entry) {
1962                 spin_unlock_bh(&pernet->lock);
1963                 return -EINVAL;
1964         }
1965         if ((addr.flags & MPTCP_PM_ADDR_FLAG_FULLMESH) &&
1966             (entry->flags & MPTCP_PM_ADDR_FLAG_SIGNAL)) {
1967                 spin_unlock_bh(&pernet->lock);
1968                 return -EINVAL;
1969         }
1970
1971         changed = (addr.flags ^ entry->flags) & mask;
1972         entry->flags = (entry->flags & ~mask) | (addr.flags & mask);
1973         addr = *entry;
1974         spin_unlock_bh(&pernet->lock);
1975
1976         mptcp_nl_set_flags(net, &addr.addr, bkup, changed);
1977         return 0;
1978 }
1979
1980 static void mptcp_nl_mcast_send(struct net *net, struct sk_buff *nlskb, gfp_t gfp)
1981 {
1982         genlmsg_multicast_netns(&mptcp_genl_family, net,
1983                                 nlskb, 0, MPTCP_PM_EV_GRP_OFFSET, gfp);
1984 }
1985
1986 bool mptcp_userspace_pm_active(const struct mptcp_sock *msk)
1987 {
1988         return genl_has_listeners(&mptcp_genl_family,
1989                                   sock_net((const struct sock *)msk),
1990                                   MPTCP_PM_EV_GRP_OFFSET);
1991 }
1992
1993 static int mptcp_event_add_subflow(struct sk_buff *skb, const struct sock *ssk)
1994 {
1995         const struct inet_sock *issk = inet_sk(ssk);
1996         const struct mptcp_subflow_context *sf;
1997
1998         if (nla_put_u16(skb, MPTCP_ATTR_FAMILY, ssk->sk_family))
1999                 return -EMSGSIZE;
2000
2001         switch (ssk->sk_family) {
2002         case AF_INET:
2003                 if (nla_put_in_addr(skb, MPTCP_ATTR_SADDR4, issk->inet_saddr))
2004                         return -EMSGSIZE;
2005                 if (nla_put_in_addr(skb, MPTCP_ATTR_DADDR4, issk->inet_daddr))
2006                         return -EMSGSIZE;
2007                 break;
2008 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
2009         case AF_INET6: {
2010                 const struct ipv6_pinfo *np = inet6_sk(ssk);
2011
2012                 if (nla_put_in6_addr(skb, MPTCP_ATTR_SADDR6, &np->saddr))
2013                         return -EMSGSIZE;
2014                 if (nla_put_in6_addr(skb, MPTCP_ATTR_DADDR6, &ssk->sk_v6_daddr))
2015                         return -EMSGSIZE;
2016                 break;
2017         }
2018 #endif
2019         default:
2020                 WARN_ON_ONCE(1);
2021                 return -EMSGSIZE;
2022         }
2023
2024         if (nla_put_be16(skb, MPTCP_ATTR_SPORT, issk->inet_sport))
2025                 return -EMSGSIZE;
2026         if (nla_put_be16(skb, MPTCP_ATTR_DPORT, issk->inet_dport))
2027                 return -EMSGSIZE;
2028
2029         sf = mptcp_subflow_ctx(ssk);
2030         if (WARN_ON_ONCE(!sf))
2031                 return -EINVAL;
2032
2033         if (nla_put_u8(skb, MPTCP_ATTR_LOC_ID, subflow_get_local_id(sf)))
2034                 return -EMSGSIZE;
2035
2036         if (nla_put_u8(skb, MPTCP_ATTR_REM_ID, sf->remote_id))
2037                 return -EMSGSIZE;
2038
2039         return 0;
2040 }
2041
2042 static int mptcp_event_put_token_and_ssk(struct sk_buff *skb,
2043                                          const struct mptcp_sock *msk,
2044                                          const struct sock *ssk)
2045 {
2046         const struct sock *sk = (const struct sock *)msk;
2047         const struct mptcp_subflow_context *sf;
2048         u8 sk_err;
2049
2050         if (nla_put_u32(skb, MPTCP_ATTR_TOKEN, msk->token))
2051                 return -EMSGSIZE;
2052
2053         if (mptcp_event_add_subflow(skb, ssk))
2054                 return -EMSGSIZE;
2055
2056         sf = mptcp_subflow_ctx(ssk);
2057         if (WARN_ON_ONCE(!sf))
2058                 return -EINVAL;
2059
2060         if (nla_put_u8(skb, MPTCP_ATTR_BACKUP, sf->backup))
2061                 return -EMSGSIZE;
2062
2063         if (ssk->sk_bound_dev_if &&
2064             nla_put_s32(skb, MPTCP_ATTR_IF_IDX, ssk->sk_bound_dev_if))
2065                 return -EMSGSIZE;
2066
2067         sk_err = READ_ONCE(ssk->sk_err);
2068         if (sk_err && sk->sk_state == TCP_ESTABLISHED &&
2069             nla_put_u8(skb, MPTCP_ATTR_ERROR, sk_err))
2070                 return -EMSGSIZE;
2071
2072         return 0;
2073 }
2074
2075 static int mptcp_event_sub_established(struct sk_buff *skb,
2076                                        const struct mptcp_sock *msk,
2077                                        const struct sock *ssk)
2078 {
2079         return mptcp_event_put_token_and_ssk(skb, msk, ssk);
2080 }
2081
2082 static int mptcp_event_sub_closed(struct sk_buff *skb,
2083                                   const struct mptcp_sock *msk,
2084                                   const struct sock *ssk)
2085 {
2086         const struct mptcp_subflow_context *sf;
2087
2088         if (mptcp_event_put_token_and_ssk(skb, msk, ssk))
2089                 return -EMSGSIZE;
2090
2091         sf = mptcp_subflow_ctx(ssk);
2092         if (!sf->reset_seen)
2093                 return 0;
2094
2095         if (nla_put_u32(skb, MPTCP_ATTR_RESET_REASON, sf->reset_reason))
2096                 return -EMSGSIZE;
2097
2098         if (nla_put_u32(skb, MPTCP_ATTR_RESET_FLAGS, sf->reset_transient))
2099                 return -EMSGSIZE;
2100
2101         return 0;
2102 }
2103
2104 static int mptcp_event_created(struct sk_buff *skb,
2105                                const struct mptcp_sock *msk,
2106                                const struct sock *ssk)
2107 {
2108         int err = nla_put_u32(skb, MPTCP_ATTR_TOKEN, msk->token);
2109
2110         if (err)
2111                 return err;
2112
2113         if (nla_put_u8(skb, MPTCP_ATTR_SERVER_SIDE, READ_ONCE(msk->pm.server_side)))
2114                 return -EMSGSIZE;
2115
2116         return mptcp_event_add_subflow(skb, ssk);
2117 }
2118
2119 void mptcp_event_addr_removed(const struct mptcp_sock *msk, uint8_t id)
2120 {
2121         struct net *net = sock_net((const struct sock *)msk);
2122         struct nlmsghdr *nlh;
2123         struct sk_buff *skb;
2124
2125         if (!genl_has_listeners(&mptcp_genl_family, net, MPTCP_PM_EV_GRP_OFFSET))
2126                 return;
2127
2128         skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_ATOMIC);
2129         if (!skb)
2130                 return;
2131
2132         nlh = genlmsg_put(skb, 0, 0, &mptcp_genl_family, 0, MPTCP_EVENT_REMOVED);
2133         if (!nlh)
2134                 goto nla_put_failure;
2135
2136         if (nla_put_u32(skb, MPTCP_ATTR_TOKEN, msk->token))
2137                 goto nla_put_failure;
2138
2139         if (nla_put_u8(skb, MPTCP_ATTR_REM_ID, id))
2140                 goto nla_put_failure;
2141
2142         genlmsg_end(skb, nlh);
2143         mptcp_nl_mcast_send(net, skb, GFP_ATOMIC);
2144         return;
2145
2146 nla_put_failure:
2147         kfree_skb(skb);
2148 }
2149
2150 void mptcp_event_addr_announced(const struct sock *ssk,
2151                                 const struct mptcp_addr_info *info)
2152 {
2153         struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk);
2154         struct mptcp_sock *msk = mptcp_sk(subflow->conn);
2155         struct net *net = sock_net(ssk);
2156         struct nlmsghdr *nlh;
2157         struct sk_buff *skb;
2158
2159         if (!genl_has_listeners(&mptcp_genl_family, net, MPTCP_PM_EV_GRP_OFFSET))
2160                 return;
2161
2162         skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_ATOMIC);
2163         if (!skb)
2164                 return;
2165
2166         nlh = genlmsg_put(skb, 0, 0, &mptcp_genl_family, 0,
2167                           MPTCP_EVENT_ANNOUNCED);
2168         if (!nlh)
2169                 goto nla_put_failure;
2170
2171         if (nla_put_u32(skb, MPTCP_ATTR_TOKEN, msk->token))
2172                 goto nla_put_failure;
2173
2174         if (nla_put_u8(skb, MPTCP_ATTR_REM_ID, info->id))
2175                 goto nla_put_failure;
2176
2177         if (nla_put_be16(skb, MPTCP_ATTR_DPORT,
2178                          info->port == 0 ?
2179                          inet_sk(ssk)->inet_dport :
2180                          info->port))
2181                 goto nla_put_failure;
2182
2183         switch (info->family) {
2184         case AF_INET:
2185                 if (nla_put_in_addr(skb, MPTCP_ATTR_DADDR4, info->addr.s_addr))
2186                         goto nla_put_failure;
2187                 break;
2188 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
2189         case AF_INET6:
2190                 if (nla_put_in6_addr(skb, MPTCP_ATTR_DADDR6, &info->addr6))
2191                         goto nla_put_failure;
2192                 break;
2193 #endif
2194         default:
2195                 WARN_ON_ONCE(1);
2196                 goto nla_put_failure;
2197         }
2198
2199         genlmsg_end(skb, nlh);
2200         mptcp_nl_mcast_send(net, skb, GFP_ATOMIC);
2201         return;
2202
2203 nla_put_failure:
2204         kfree_skb(skb);
2205 }
2206
2207 void mptcp_event(enum mptcp_event_type type, const struct mptcp_sock *msk,
2208                  const struct sock *ssk, gfp_t gfp)
2209 {
2210         struct net *net = sock_net((const struct sock *)msk);
2211         struct nlmsghdr *nlh;
2212         struct sk_buff *skb;
2213
2214         if (!genl_has_listeners(&mptcp_genl_family, net, MPTCP_PM_EV_GRP_OFFSET))
2215                 return;
2216
2217         skb = nlmsg_new(NLMSG_DEFAULT_SIZE, gfp);
2218         if (!skb)
2219                 return;
2220
2221         nlh = genlmsg_put(skb, 0, 0, &mptcp_genl_family, 0, type);
2222         if (!nlh)
2223                 goto nla_put_failure;
2224
2225         switch (type) {
2226         case MPTCP_EVENT_UNSPEC:
2227                 WARN_ON_ONCE(1);
2228                 break;
2229         case MPTCP_EVENT_CREATED:
2230         case MPTCP_EVENT_ESTABLISHED:
2231                 if (mptcp_event_created(skb, msk, ssk) < 0)
2232                         goto nla_put_failure;
2233                 break;
2234         case MPTCP_EVENT_CLOSED:
2235                 if (nla_put_u32(skb, MPTCP_ATTR_TOKEN, msk->token) < 0)
2236                         goto nla_put_failure;
2237                 break;
2238         case MPTCP_EVENT_ANNOUNCED:
2239         case MPTCP_EVENT_REMOVED:
2240                 /* call mptcp_event_addr_announced()/removed instead */
2241                 WARN_ON_ONCE(1);
2242                 break;
2243         case MPTCP_EVENT_SUB_ESTABLISHED:
2244         case MPTCP_EVENT_SUB_PRIORITY:
2245                 if (mptcp_event_sub_established(skb, msk, ssk) < 0)
2246                         goto nla_put_failure;
2247                 break;
2248         case MPTCP_EVENT_SUB_CLOSED:
2249                 if (mptcp_event_sub_closed(skb, msk, ssk) < 0)
2250                         goto nla_put_failure;
2251                 break;
2252         }
2253
2254         genlmsg_end(skb, nlh);
2255         mptcp_nl_mcast_send(net, skb, gfp);
2256         return;
2257
2258 nla_put_failure:
2259         kfree_skb(skb);
2260 }
2261
2262 static const struct genl_small_ops mptcp_pm_ops[] = {
2263         {
2264                 .cmd    = MPTCP_PM_CMD_ADD_ADDR,
2265                 .doit   = mptcp_nl_cmd_add_addr,
2266                 .flags  = GENL_UNS_ADMIN_PERM,
2267         },
2268         {
2269                 .cmd    = MPTCP_PM_CMD_DEL_ADDR,
2270                 .doit   = mptcp_nl_cmd_del_addr,
2271                 .flags  = GENL_UNS_ADMIN_PERM,
2272         },
2273         {
2274                 .cmd    = MPTCP_PM_CMD_FLUSH_ADDRS,
2275                 .doit   = mptcp_nl_cmd_flush_addrs,
2276                 .flags  = GENL_UNS_ADMIN_PERM,
2277         },
2278         {
2279                 .cmd    = MPTCP_PM_CMD_GET_ADDR,
2280                 .doit   = mptcp_nl_cmd_get_addr,
2281                 .dumpit   = mptcp_nl_cmd_dump_addrs,
2282         },
2283         {
2284                 .cmd    = MPTCP_PM_CMD_SET_LIMITS,
2285                 .doit   = mptcp_nl_cmd_set_limits,
2286                 .flags  = GENL_UNS_ADMIN_PERM,
2287         },
2288         {
2289                 .cmd    = MPTCP_PM_CMD_GET_LIMITS,
2290                 .doit   = mptcp_nl_cmd_get_limits,
2291         },
2292         {
2293                 .cmd    = MPTCP_PM_CMD_SET_FLAGS,
2294                 .doit   = mptcp_nl_cmd_set_flags,
2295                 .flags  = GENL_UNS_ADMIN_PERM,
2296         },
2297         {
2298                 .cmd    = MPTCP_PM_CMD_ANNOUNCE,
2299                 .doit   = mptcp_nl_cmd_announce,
2300                 .flags  = GENL_UNS_ADMIN_PERM,
2301         },
2302         {
2303                 .cmd    = MPTCP_PM_CMD_REMOVE,
2304                 .doit   = mptcp_nl_cmd_remove,
2305                 .flags  = GENL_UNS_ADMIN_PERM,
2306         },
2307         {
2308                 .cmd    = MPTCP_PM_CMD_SUBFLOW_CREATE,
2309                 .doit   = mptcp_nl_cmd_sf_create,
2310                 .flags  = GENL_UNS_ADMIN_PERM,
2311         },
2312         {
2313                 .cmd    = MPTCP_PM_CMD_SUBFLOW_DESTROY,
2314                 .doit   = mptcp_nl_cmd_sf_destroy,
2315                 .flags  = GENL_UNS_ADMIN_PERM,
2316         },
2317 };
2318
2319 static struct genl_family mptcp_genl_family __ro_after_init = {
2320         .name           = MPTCP_PM_NAME,
2321         .version        = MPTCP_PM_VER,
2322         .maxattr        = MPTCP_PM_ATTR_MAX,
2323         .policy         = mptcp_pm_policy,
2324         .netnsok        = true,
2325         .module         = THIS_MODULE,
2326         .small_ops      = mptcp_pm_ops,
2327         .n_small_ops    = ARRAY_SIZE(mptcp_pm_ops),
2328         .resv_start_op  = MPTCP_PM_CMD_SUBFLOW_DESTROY + 1,
2329         .mcgrps         = mptcp_pm_mcgrps,
2330         .n_mcgrps       = ARRAY_SIZE(mptcp_pm_mcgrps),
2331 };
2332
2333 static int __net_init pm_nl_init_net(struct net *net)
2334 {
2335         struct pm_nl_pernet *pernet = pm_nl_get_pernet(net);
2336
2337         INIT_LIST_HEAD_RCU(&pernet->local_addr_list);
2338
2339         /* Cit. 2 subflows ought to be enough for anybody. */
2340         pernet->subflows_max = 2;
2341         pernet->next_id = 1;
2342         pernet->stale_loss_cnt = 4;
2343         spin_lock_init(&pernet->lock);
2344
2345         /* No need to initialize other pernet fields, the struct is zeroed at
2346          * allocation time.
2347          */
2348
2349         return 0;
2350 }
2351
2352 static void __net_exit pm_nl_exit_net(struct list_head *net_list)
2353 {
2354         struct net *net;
2355
2356         list_for_each_entry(net, net_list, exit_list) {
2357                 struct pm_nl_pernet *pernet = pm_nl_get_pernet(net);
2358
2359                 /* net is removed from namespace list, can't race with
2360                  * other modifiers, also netns core already waited for a
2361                  * RCU grace period.
2362                  */
2363                 __flush_addrs(&pernet->local_addr_list);
2364         }
2365 }
2366
2367 static struct pernet_operations mptcp_pm_pernet_ops = {
2368         .init = pm_nl_init_net,
2369         .exit_batch = pm_nl_exit_net,
2370         .id = &pm_nl_pernet_id,
2371         .size = sizeof(struct pm_nl_pernet),
2372 };
2373
2374 void __init mptcp_pm_nl_init(void)
2375 {
2376         if (register_pernet_subsys(&mptcp_pm_pernet_ops) < 0)
2377                 panic("Failed to register MPTCP PM pernet subsystem.\n");
2378
2379         if (genl_register_family(&mptcp_genl_family))
2380                 panic("Failed to register MPTCP PM netlink family\n");
2381 }