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/inet_common.h>
  13 #include <net/netns/generic.h>
  14 #include <net/mptcp.h>
  15 #include <net/genetlink.h>
  16 #include <uapi/linux/mptcp.h>
  17
  18 #include "protocol.h"
  19 #include "mib.h"
  20
  21 /* forward declaration */
  22 static struct genl_family mptcp_genl_family;
  23
  24 static int pm_nl_pernet_id;
  25
  26 struct mptcp_pm_add_entry {
  27         struct list_head        list;
  28         struct mptcp_addr_info  addr;
  29         u8                      retrans_times;
  30         struct timer_list       add_timer;
  31         struct mptcp_sock       *sock;
  32 };
  33
  34 struct pm_nl_pernet {
  35         /* protects pernet updates */
  36         spinlock_t              lock;
  37         struct list_head        local_addr_list;
  38         unsigned int            addrs;
  39         unsigned int            stale_loss_cnt;
  40         unsigned int            add_addr_signal_max;
  41         unsigned int            add_addr_accept_max;
  42         unsigned int            local_addr_max;
  43         unsigned int            subflows_max;
  44         unsigned int            next_id;
  45         DECLARE_BITMAP(id_bitmap, MPTCP_PM_MAX_ADDR_ID + 1);
  46 };
  47
  48 #define MPTCP_PM_ADDR_MAX       8
  49 #define ADD_ADDR_RETRANS_MAX    3
  50
  51 static struct pm_nl_pernet *pm_nl_get_pernet(const struct net *net)
  52 {
  53         return net_generic(net, pm_nl_pernet_id);
  54 }
  55
  56 static struct pm_nl_pernet *
  57 pm_nl_get_pernet_from_msk(const struct mptcp_sock *msk)
  58 {
  59         return pm_nl_get_pernet(sock_net((struct sock *)msk));
  60 }
  61
  62 bool mptcp_addresses_equal(const struct mptcp_addr_info *a,
  63                            const struct mptcp_addr_info *b, bool use_port)
  64 {
  65         bool addr_equals = false;
  66
  67         if (a->family == b->family) {
  68                 if (a->family == AF_INET)
  69                         addr_equals = a->addr.s_addr == b->addr.s_addr;
  70 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
  71                 else
  72                         addr_equals = !ipv6_addr_cmp(&a->addr6, &b->addr6);
  73         } else if (a->family == AF_INET) {
  74                 if (ipv6_addr_v4mapped(&b->addr6))
  75                         addr_equals = a->addr.s_addr == b->addr6.s6_addr32[3];
  76         } else if (b->family == AF_INET) {
  77                 if (ipv6_addr_v4mapped(&a->addr6))
  78                         addr_equals = a->addr6.s6_addr32[3] == b->addr.s_addr;
  79 #endif
  80         }
  81
  82         if (!addr_equals)
  83                 return false;
  84         if (!use_port)
  85                 return true;
  86
  87         return a->port == b->port;
  88 }
  89
  90 void mptcp_local_address(const struct sock_common *skc, 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                 mptcp_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         struct mptcp_pm_addr_entry *entry, *ret = NULL;
 156
 157         msk_owned_by_me(msk);
 158
 159         rcu_read_lock();
 160         list_for_each_entry_rcu(entry, &pernet->local_addr_list, list) {
 161                 if (!(entry->flags & MPTCP_PM_ADDR_FLAG_SUBFLOW))
 162                         continue;
 163
 164                 if (!test_bit(entry->addr.id, msk->pm.id_avail_bitmap))
 165                         continue;
 166
 167                 ret = entry;
 168                 break;
 169         }
 170         rcu_read_unlock();
 171         return ret;
 172 }
 173
 174 static struct mptcp_pm_addr_entry *
 175 select_signal_address(struct pm_nl_pernet *pernet, const struct mptcp_sock *msk)
 176 {
 177         struct mptcp_pm_addr_entry *entry, *ret = NULL;
 178
 179         rcu_read_lock();
 180         /* do not keep any additional per socket state, just signal
 181          * the address list in order.
 182          * Note: removal from the local address list during the msk life-cycle
 183          * can lead to additional addresses not being announced.
 184          */
 185         list_for_each_entry_rcu(entry, &pernet->local_addr_list, list) {
 186                 if (!test_bit(entry->addr.id, msk->pm.id_avail_bitmap))
 187                         continue;
 188
 189                 if (!(entry->flags & MPTCP_PM_ADDR_FLAG_SIGNAL))
 190                         continue;
 191
 192                 ret = entry;
 193                 break;
 194         }
 195         rcu_read_unlock();
 196         return ret;
 197 }
 198
 199 unsigned int mptcp_pm_get_add_addr_signal_max(const struct mptcp_sock *msk)
 200 {
 201         const struct pm_nl_pernet *pernet = pm_nl_get_pernet_from_msk(msk);
 202
 203         return READ_ONCE(pernet->add_addr_signal_max);
 204 }
 205 EXPORT_SYMBOL_GPL(mptcp_pm_get_add_addr_signal_max);
 206
 207 unsigned int mptcp_pm_get_add_addr_accept_max(const struct mptcp_sock *msk)
 208 {
 209         struct pm_nl_pernet *pernet = pm_nl_get_pernet_from_msk(msk);
 210
 211         return READ_ONCE(pernet->add_addr_accept_max);
 212 }
 213 EXPORT_SYMBOL_GPL(mptcp_pm_get_add_addr_accept_max);
 214
 215 unsigned int mptcp_pm_get_subflows_max(const struct mptcp_sock *msk)
 216 {
 217         struct pm_nl_pernet *pernet = pm_nl_get_pernet_from_msk(msk);
 218
 219         return READ_ONCE(pernet->subflows_max);
 220 }
 221 EXPORT_SYMBOL_GPL(mptcp_pm_get_subflows_max);
 222
 223 unsigned int mptcp_pm_get_local_addr_max(const struct mptcp_sock *msk)
 224 {
 225         struct pm_nl_pernet *pernet = pm_nl_get_pernet_from_msk(msk);
 226
 227         return READ_ONCE(pernet->local_addr_max);
 228 }
 229 EXPORT_SYMBOL_GPL(mptcp_pm_get_local_addr_max);
 230
 231 bool mptcp_pm_nl_check_work_pending(struct mptcp_sock *msk)
 232 {
 233         struct pm_nl_pernet *pernet = pm_nl_get_pernet_from_msk(msk);
 234
 235         if (msk->pm.subflows == mptcp_pm_get_subflows_max(msk) ||
 236             (find_next_and_bit(pernet->id_bitmap, msk->pm.id_avail_bitmap,
 237                                MPTCP_PM_MAX_ADDR_ID + 1, 0) == MPTCP_PM_MAX_ADDR_ID + 1)) {
 238                 WRITE_ONCE(msk->pm.work_pending, false);
 239                 return false;
 240         }
 241         return true;
 242 }
 243
 244 struct mptcp_pm_add_entry *
 245 mptcp_lookup_anno_list_by_saddr(const struct mptcp_sock *msk,
 246                                 const struct mptcp_addr_info *addr)
 247 {
 248         struct mptcp_pm_add_entry *entry;
 249
 250         lockdep_assert_held(&msk->pm.lock);
 251
 252         list_for_each_entry(entry, &msk->pm.anno_list, list) {
 253                 if (mptcp_addresses_equal(&entry->addr, addr, true))
 254                         return entry;
 255         }
 256
 257         return NULL;
 258 }
 259
 260 bool mptcp_pm_sport_in_anno_list(struct mptcp_sock *msk, const struct sock *sk)
 261 {
 262         struct mptcp_pm_add_entry *entry;
 263         struct mptcp_addr_info saddr;
 264         bool ret = false;
 265
 266         mptcp_local_address((struct sock_common *)sk, &saddr);
 267
 268         spin_lock_bh(&msk->pm.lock);
 269         list_for_each_entry(entry, &msk->pm.anno_list, list) {
 270                 if (mptcp_addresses_equal(&entry->addr, &saddr, true)) {
 271                         ret = true;
 272                         goto out;
 273                 }
 274         }
 275
 276 out:
 277         spin_unlock_bh(&msk->pm.lock);
 278         return ret;
 279 }
 280
 281 static void mptcp_pm_add_timer(struct timer_list *timer)
 282 {
 283         struct mptcp_pm_add_entry *entry = from_timer(entry, timer, add_timer);
 284         struct mptcp_sock *msk = entry->sock;
 285         struct sock *sk = (struct sock *)msk;
 286
 287         pr_debug("msk=%p", msk);
 288
 289         if (!msk)
 290                 return;
 291
 292         if (inet_sk_state_load(sk) == TCP_CLOSE)
 293                 return;
 294
 295         if (!entry->addr.id)
 296                 return;
 297
 298         if (mptcp_pm_should_add_signal_addr(msk)) {
 299                 sk_reset_timer(sk, timer, jiffies + TCP_RTO_MAX / 8);
 300                 goto out;
 301         }
 302
 303         spin_lock_bh(&msk->pm.lock);
 304
 305         if (!mptcp_pm_should_add_signal_addr(msk)) {
 306                 pr_debug("retransmit ADD_ADDR id=%d", entry->addr.id);
 307                 mptcp_pm_announce_addr(msk, &entry->addr, false);
 308                 mptcp_pm_add_addr_send_ack(msk);
 309                 entry->retrans_times++;
 310         }
 311
 312         if (entry->retrans_times < ADD_ADDR_RETRANS_MAX)
 313                 sk_reset_timer(sk, timer,
 314                                jiffies + mptcp_get_add_addr_timeout(sock_net(sk)));
 315
 316         spin_unlock_bh(&msk->pm.lock);
 317
 318         if (entry->retrans_times == ADD_ADDR_RETRANS_MAX)
 319                 mptcp_pm_subflow_established(msk);
 320
 321 out:
 322         __sock_put(sk);
 323 }
 324
 325 struct mptcp_pm_add_entry *
 326 mptcp_pm_del_add_timer(struct mptcp_sock *msk,
 327                        const struct mptcp_addr_info *addr, bool check_id)
 328 {
 329         struct mptcp_pm_add_entry *entry;
 330         struct sock *sk = (struct sock *)msk;
 331
 332         spin_lock_bh(&msk->pm.lock);
 333         entry = mptcp_lookup_anno_list_by_saddr(msk, addr);
 334         if (entry && (!check_id || entry->addr.id == addr->id))
 335                 entry->retrans_times = ADD_ADDR_RETRANS_MAX;
 336         spin_unlock_bh(&msk->pm.lock);
 337
 338         if (entry && (!check_id || entry->addr.id == addr->id))
 339                 sk_stop_timer_sync(sk, &entry->add_timer);
 340
 341         return entry;
 342 }
 343
 344 bool mptcp_pm_alloc_anno_list(struct mptcp_sock *msk,
 345                               const struct mptcp_addr_info *addr)
 346 {
 347         struct mptcp_pm_add_entry *add_entry = NULL;
 348         struct sock *sk = (struct sock *)msk;
 349         struct net *net = sock_net(sk);
 350
 351         lockdep_assert_held(&msk->pm.lock);
 352
 353         add_entry = mptcp_lookup_anno_list_by_saddr(msk, addr);
 354
 355         if (add_entry) {
 356                 if (mptcp_pm_is_kernel(msk))
 357                         return false;
 358
 359                 sk_reset_timer(sk, &add_entry->add_timer,
 360                                jiffies + mptcp_get_add_addr_timeout(net));
 361                 return true;
 362         }
 363
 364         add_entry = kmalloc(sizeof(*add_entry), GFP_ATOMIC);
 365         if (!add_entry)
 366                 return false;
 367
 368         list_add(&add_entry->list, &msk->pm.anno_list);
 369
 370         add_entry->addr = *addr;
 371         add_entry->sock = msk;
 372         add_entry->retrans_times = 0;
 373
 374         timer_setup(&add_entry->add_timer, mptcp_pm_add_timer, 0);
 375         sk_reset_timer(sk, &add_entry->add_timer,
 376                        jiffies + mptcp_get_add_addr_timeout(net));
 377
 378         return true;
 379 }
 380
 381 void mptcp_pm_free_anno_list(struct mptcp_sock *msk)
 382 {
 383         struct mptcp_pm_add_entry *entry, *tmp;
 384         struct sock *sk = (struct sock *)msk;
 385         LIST_HEAD(free_list);
 386
 387         pr_debug("msk=%p", msk);
 388
 389         spin_lock_bh(&msk->pm.lock);
 390         list_splice_init(&msk->pm.anno_list, &free_list);
 391         spin_unlock_bh(&msk->pm.lock);
 392
 393         list_for_each_entry_safe(entry, tmp, &free_list, list) {
 394                 sk_stop_timer_sync(sk, &entry->add_timer);
 395                 kfree(entry);
 396         }
 397 }
 398
 399 /* Fill all the remote addresses into the array addrs[],
 400  * and return the array size.
 401  */
 402 static unsigned int fill_remote_addresses_vec(struct mptcp_sock *msk,
 403                                               struct mptcp_addr_info *local,
 404                                               bool fullmesh,
 405                                               struct mptcp_addr_info *addrs)
 406 {
 407         bool deny_id0 = READ_ONCE(msk->pm.remote_deny_join_id0);
 408         struct sock *sk = (struct sock *)msk, *ssk;
 409         struct mptcp_subflow_context *subflow;
 410         struct mptcp_addr_info remote = { 0 };
 411         unsigned int subflows_max;
 412         int i = 0;
 413
 414         subflows_max = mptcp_pm_get_subflows_max(msk);
 415         remote_address((struct sock_common *)sk, &remote);
 416
 417         /* Non-fullmesh endpoint, fill in the single entry
 418          * corresponding to the primary MPC subflow remote address
 419          */
 420         if (!fullmesh) {
 421                 if (deny_id0)
 422                         return 0;
 423
 424                 if (!mptcp_pm_addr_families_match(sk, local, &remote))
 425                         return 0;
 426
 427                 msk->pm.subflows++;
 428                 addrs[i++] = remote;
 429         } else {
 430                 DECLARE_BITMAP(unavail_id, MPTCP_PM_MAX_ADDR_ID + 1);
 431
 432                 /* Forbid creation of new subflows matching existing
 433                  * ones, possibly already created by incoming ADD_ADDR
 434                  */
 435                 bitmap_zero(unavail_id, MPTCP_PM_MAX_ADDR_ID + 1);
 436                 mptcp_for_each_subflow(msk, subflow)
 437                         if (READ_ONCE(subflow->local_id) == local->id)
 438                                 __set_bit(subflow->remote_id, unavail_id);
 439
 440                 mptcp_for_each_subflow(msk, subflow) {
 441                         ssk = mptcp_subflow_tcp_sock(subflow);
 442                         remote_address((struct sock_common *)ssk, &addrs[i]);
 443                         addrs[i].id = READ_ONCE(subflow->remote_id);
 444                         if (deny_id0 && !addrs[i].id)
 445                                 continue;
 446
 447                         if (test_bit(addrs[i].id, unavail_id))
 448                                 continue;
 449
 450                         if (!mptcp_pm_addr_families_match(sk, local, &addrs[i]))
 451                                 continue;
 452
 453                         if (msk->pm.subflows < subflows_max) {
 454                                 /* forbid creating multiple address towards
 455                                  * this id
 456                                  */
 457                                 __set_bit(addrs[i].id, unavail_id);
 458                                 msk->pm.subflows++;
 459                                 i++;
 460                         }
 461                 }
 462         }
 463
 464         return i;
 465 }
 466
 467 static void __mptcp_pm_send_ack(struct mptcp_sock *msk, struct mptcp_subflow_context *subflow,
 468                                 bool prio, bool backup)
 469 {
 470         struct sock *ssk = mptcp_subflow_tcp_sock(subflow);
 471         bool slow;
 472
 473         pr_debug("send ack for %s",
 474                  prio ? "mp_prio" : (mptcp_pm_should_add_signal(msk) ? "add_addr" : "rm_addr"));
 475
 476         slow = lock_sock_fast(ssk);
 477         if (prio) {
 478                 subflow->send_mp_prio = 1;
 479                 subflow->backup = backup;
 480                 subflow->request_bkup = backup;
 481         }
 482
 483         __mptcp_subflow_send_ack(ssk);
 484         unlock_sock_fast(ssk, slow);
 485 }
 486
 487 static void mptcp_pm_send_ack(struct mptcp_sock *msk, struct mptcp_subflow_context *subflow,
 488                               bool prio, bool backup)
 489 {
 490         spin_unlock_bh(&msk->pm.lock);
 491         __mptcp_pm_send_ack(msk, subflow, prio, backup);
 492         spin_lock_bh(&msk->pm.lock);
 493 }
 494
 495 static struct mptcp_pm_addr_entry *
 496 __lookup_addr_by_id(struct pm_nl_pernet *pernet, unsigned int id)
 497 {
 498         struct mptcp_pm_addr_entry *entry;
 499
 500         list_for_each_entry(entry, &pernet->local_addr_list, list) {
 501                 if (entry->addr.id == id)
 502                         return entry;
 503         }
 504         return NULL;
 505 }
 506
 507 static struct mptcp_pm_addr_entry *
 508 __lookup_addr(struct pm_nl_pernet *pernet, const struct mptcp_addr_info *info,
 509               bool lookup_by_id)
 510 {
 511         struct mptcp_pm_addr_entry *entry;
 512
 513         list_for_each_entry(entry, &pernet->local_addr_list, list) {
 514                 if ((!lookup_by_id &&
 515                      mptcp_addresses_equal(&entry->addr, info, entry->addr.port)) ||
 516                     (lookup_by_id && entry->addr.id == info->id))
 517                         return entry;
 518         }
 519         return NULL;
 520 }
 521
 522 static void mptcp_pm_create_subflow_or_signal_addr(struct mptcp_sock *msk)
 523 {
 524         struct sock *sk = (struct sock *)msk;
 525         struct mptcp_pm_addr_entry *local;
 526         unsigned int add_addr_signal_max;
 527         unsigned int local_addr_max;
 528         struct pm_nl_pernet *pernet;
 529         unsigned int subflows_max;
 530
 531         pernet = pm_nl_get_pernet(sock_net(sk));
 532
 533         add_addr_signal_max = mptcp_pm_get_add_addr_signal_max(msk);
 534         local_addr_max = mptcp_pm_get_local_addr_max(msk);
 535         subflows_max = mptcp_pm_get_subflows_max(msk);
 536
 537         /* do lazy endpoint usage accounting for the MPC subflows */
 538         if (unlikely(!(msk->pm.status & BIT(MPTCP_PM_MPC_ENDPOINT_ACCOUNTED))) && msk->first) {
 539                 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(msk->first);
 540                 struct mptcp_pm_addr_entry *entry;
 541                 struct mptcp_addr_info mpc_addr;
 542                 bool backup = false;
 543
 544                 mptcp_local_address((struct sock_common *)msk->first, &mpc_addr);
 545                 rcu_read_lock();
 546                 entry = __lookup_addr(pernet, &mpc_addr, false);
 547                 if (entry) {
 548                         __clear_bit(entry->addr.id, msk->pm.id_avail_bitmap);
 549                         msk->mpc_endpoint_id = entry->addr.id;
 550                         backup = !!(entry->flags & MPTCP_PM_ADDR_FLAG_BACKUP);
 551                 }
 552                 rcu_read_unlock();
 553
 554                 if (backup)
 555                         mptcp_pm_send_ack(msk, subflow, true, backup);
 556
 557                 msk->pm.status |= BIT(MPTCP_PM_MPC_ENDPOINT_ACCOUNTED);
 558         }
 559
 560         pr_debug("local %d:%d signal %d:%d subflows %d:%d\n",
 561                  msk->pm.local_addr_used, local_addr_max,
 562                  msk->pm.add_addr_signaled, add_addr_signal_max,
 563                  msk->pm.subflows, subflows_max);
 564
 565         /* check first for announce */
 566         if (msk->pm.add_addr_signaled < add_addr_signal_max) {
 567                 local = select_signal_address(pernet, msk);
 568
 569                 /* due to racing events on both ends we can reach here while
 570                  * previous add address is still running: if we invoke now
 571                  * mptcp_pm_announce_addr(), that will fail and the
 572                  * corresponding id will be marked as used.
 573                  * Instead let the PM machinery reschedule us when the
 574                  * current address announce will be completed.
 575                  */
 576                 if (msk->pm.addr_signal & BIT(MPTCP_ADD_ADDR_SIGNAL))
 577                         return;
 578
 579                 if (local) {
 580                         if (mptcp_pm_alloc_anno_list(msk, &local->addr)) {
 581                                 __clear_bit(local->addr.id, msk->pm.id_avail_bitmap);
 582                                 msk->pm.add_addr_signaled++;
 583                                 mptcp_pm_announce_addr(msk, &local->addr, false);
 584                                 mptcp_pm_nl_addr_send_ack(msk);
 585                         }
 586                 }
 587         }
 588
 589         /* check if should create a new subflow */
 590         while (msk->pm.local_addr_used < local_addr_max &&
 591                msk->pm.subflows < subflows_max) {
 592                 struct mptcp_addr_info addrs[MPTCP_PM_ADDR_MAX];
 593                 bool fullmesh;
 594                 int i, nr;
 595
 596                 local = select_local_address(pernet, msk);
 597                 if (!local)
 598                         break;
 599
 600                 fullmesh = !!(local->flags & MPTCP_PM_ADDR_FLAG_FULLMESH);
 601
 602                 msk->pm.local_addr_used++;
 603                 __clear_bit(local->addr.id, msk->pm.id_avail_bitmap);
 604                 nr = fill_remote_addresses_vec(msk, &local->addr, fullmesh, addrs);
 605                 if (nr == 0)
 606                         continue;
 607
 608                 spin_unlock_bh(&msk->pm.lock);
 609                 for (i = 0; i < nr; i++)
 610                         __mptcp_subflow_connect(sk, &local->addr, &addrs[i]);
 611                 spin_lock_bh(&msk->pm.lock);
 612         }
 613         mptcp_pm_nl_check_work_pending(msk);
 614 }
 615
 616 static void mptcp_pm_nl_fully_established(struct mptcp_sock *msk)
 617 {
 618         mptcp_pm_create_subflow_or_signal_addr(msk);
 619 }
 620
 621 static void mptcp_pm_nl_subflow_established(struct mptcp_sock *msk)
 622 {
 623         mptcp_pm_create_subflow_or_signal_addr(msk);
 624 }
 625
 626 /* Fill all the local addresses into the array addrs[],
 627  * and return the array size.
 628  */
 629 static unsigned int fill_local_addresses_vec(struct mptcp_sock *msk,
 630                                              struct mptcp_addr_info *remote,
 631                                              struct mptcp_addr_info *addrs)
 632 {
 633         struct sock *sk = (struct sock *)msk;
 634         struct mptcp_pm_addr_entry *entry;
 635         struct pm_nl_pernet *pernet;
 636         unsigned int subflows_max;
 637         int i = 0;
 638
 639         pernet = pm_nl_get_pernet_from_msk(msk);
 640         subflows_max = mptcp_pm_get_subflows_max(msk);
 641
 642         rcu_read_lock();
 643         list_for_each_entry_rcu(entry, &pernet->local_addr_list, list) {
 644                 if (!(entry->flags & MPTCP_PM_ADDR_FLAG_FULLMESH))
 645                         continue;
 646
 647                 if (!mptcp_pm_addr_families_match(sk, &entry->addr, remote))
 648                         continue;
 649
 650                 if (msk->pm.subflows < subflows_max) {
 651                         msk->pm.subflows++;
 652                         addrs[i++] = entry->addr;
 653                 }
 654         }
 655         rcu_read_unlock();
 656
 657         /* If the array is empty, fill in the single
 658          * 'IPADDRANY' local address
 659          */
 660         if (!i) {
 661                 struct mptcp_addr_info local;
 662
 663                 memset(&local, 0, sizeof(local));
 664                 local.family =
 665 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
 666                                remote->family == AF_INET6 &&
 667                                ipv6_addr_v4mapped(&remote->addr6) ? AF_INET :
 668 #endif
 669                                remote->family;
 670
 671                 if (!mptcp_pm_addr_families_match(sk, &local, remote))
 672                         return 0;
 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, &remote, addrs);
 712         if (nr == 0)
 713                 return;
 714
 715         msk->pm.add_addr_accepted++;
 716         if (msk->pm.add_addr_accepted >= add_addr_accept_max ||
 717             msk->pm.subflows >= subflows_max)
 718                 WRITE_ONCE(msk->pm.accept_addr, false);
 719
 720         spin_unlock_bh(&msk->pm.lock);
 721         for (i = 0; i < nr; i++)
 722                 __mptcp_subflow_connect(sk, &addrs[i], &remote);
 723         spin_lock_bh(&msk->pm.lock);
 724 }
 725
 726 void mptcp_pm_nl_addr_send_ack(struct mptcp_sock *msk)
 727 {
 728         struct mptcp_subflow_context *subflow;
 729
 730         msk_owned_by_me(msk);
 731         lockdep_assert_held(&msk->pm.lock);
 732
 733         if (!mptcp_pm_should_add_signal(msk) &&
 734             !mptcp_pm_should_rm_signal(msk))
 735                 return;
 736
 737         subflow = list_first_entry_or_null(&msk->conn_list, typeof(*subflow), node);
 738         if (subflow)
 739                 mptcp_pm_send_ack(msk, subflow, false, false);
 740 }
 741
 742 int mptcp_pm_nl_mp_prio_send_ack(struct mptcp_sock *msk,
 743                                  struct mptcp_addr_info *addr,
 744                                  struct mptcp_addr_info *rem,
 745                                  u8 bkup)
 746 {
 747         struct mptcp_subflow_context *subflow;
 748
 749         pr_debug("bkup=%d", bkup);
 750
 751         mptcp_for_each_subflow(msk, subflow) {
 752                 struct sock *ssk = mptcp_subflow_tcp_sock(subflow);
 753                 struct mptcp_addr_info local, remote;
 754
 755                 mptcp_local_address((struct sock_common *)ssk, &local);
 756                 if (!mptcp_addresses_equal(&local, addr, addr->port))
 757                         continue;
 758
 759                 if (rem && rem->family != AF_UNSPEC) {
 760                         remote_address((struct sock_common *)ssk, &remote);
 761                         if (!mptcp_addresses_equal(&remote, rem, rem->port))
 762                                 continue;
 763                 }
 764
 765                 __mptcp_pm_send_ack(msk, subflow, true, bkup);
 766                 return 0;
 767         }
 768
 769         return -EINVAL;
 770 }
 771
 772 static bool mptcp_local_id_match(const struct mptcp_sock *msk, u8 local_id, u8 id)
 773 {
 774         return local_id == id || (!local_id && msk->mpc_endpoint_id == id);
 775 }
 776
 777 static void mptcp_pm_nl_rm_addr_or_subflow(struct mptcp_sock *msk,
 778                                            const struct mptcp_rm_list *rm_list,
 779                                            enum linux_mptcp_mib_field rm_type)
 780 {
 781         struct mptcp_subflow_context *subflow, *tmp;
 782         struct sock *sk = (struct sock *)msk;
 783         u8 i;
 784
 785         pr_debug("%s rm_list_nr %d",
 786                  rm_type == MPTCP_MIB_RMADDR ? "address" : "subflow", rm_list->nr);
 787
 788         msk_owned_by_me(msk);
 789
 790         if (sk->sk_state == TCP_LISTEN)
 791                 return;
 792
 793         if (!rm_list->nr)
 794                 return;
 795
 796         if (list_empty(&msk->conn_list))
 797                 return;
 798
 799         for (i = 0; i < rm_list->nr; i++) {
 800                 u8 rm_id = rm_list->ids[i];
 801                 bool removed = false;
 802
 803                 mptcp_for_each_subflow_safe(msk, subflow, tmp) {
 804                         struct sock *ssk = mptcp_subflow_tcp_sock(subflow);
 805                         u8 remote_id = READ_ONCE(subflow->remote_id);
 806                         int how = RCV_SHUTDOWN | SEND_SHUTDOWN;
 807                         u8 id = subflow_get_local_id(subflow);
 808
 809                         if (rm_type == MPTCP_MIB_RMADDR && remote_id != rm_id)
 810                                 continue;
 811                         if (rm_type == MPTCP_MIB_RMSUBFLOW && !mptcp_local_id_match(msk, id, rm_id))
 812                                 continue;
 813
 814                         pr_debug(" -> %s rm_list_ids[%d]=%u local_id=%u remote_id=%u mpc_id=%u",
 815                                  rm_type == MPTCP_MIB_RMADDR ? "address" : "subflow",
 816                                  i, rm_id, id, remote_id, msk->mpc_endpoint_id);
 817                         spin_unlock_bh(&msk->pm.lock);
 818                         mptcp_subflow_shutdown(sk, ssk, how);
 819
 820                         /* the following takes care of updating the subflows counter */
 821                         mptcp_close_ssk(sk, ssk, subflow);
 822                         spin_lock_bh(&msk->pm.lock);
 823
 824                         removed = true;
 825                         __MPTCP_INC_STATS(sock_net(sk), rm_type);
 826                 }
 827                 if (rm_type == MPTCP_MIB_RMSUBFLOW)
 828                         __set_bit(rm_id ? rm_id : msk->mpc_endpoint_id, msk->pm.id_avail_bitmap);
 829                 if (!removed)
 830                         continue;
 831
 832                 if (!mptcp_pm_is_kernel(msk))
 833                         continue;
 834
 835                 if (rm_type == MPTCP_MIB_RMADDR) {
 836                         msk->pm.add_addr_accepted--;
 837                         WRITE_ONCE(msk->pm.accept_addr, true);
 838                 } else if (rm_type == MPTCP_MIB_RMSUBFLOW) {
 839                         msk->pm.local_addr_used--;
 840                 }
 841         }
 842 }
 843
 844 static void mptcp_pm_nl_rm_addr_received(struct mptcp_sock *msk)
 845 {
 846         mptcp_pm_nl_rm_addr_or_subflow(msk, &msk->pm.rm_list_rx, MPTCP_MIB_RMADDR);
 847 }
 848
 849 void mptcp_pm_nl_rm_subflow_received(struct mptcp_sock *msk,
 850                                      const struct mptcp_rm_list *rm_list)
 851 {
 852         mptcp_pm_nl_rm_addr_or_subflow(msk, rm_list, MPTCP_MIB_RMSUBFLOW);
 853 }
 854
 855 void mptcp_pm_nl_work(struct mptcp_sock *msk)
 856 {
 857         struct mptcp_pm_data *pm = &msk->pm;
 858
 859         msk_owned_by_me(msk);
 860
 861         if (!(pm->status & MPTCP_PM_WORK_MASK))
 862                 return;
 863
 864         spin_lock_bh(&msk->pm.lock);
 865
 866         pr_debug("msk=%p status=%x", msk, pm->status);
 867         if (pm->status & BIT(MPTCP_PM_ADD_ADDR_RECEIVED)) {
 868                 pm->status &= ~BIT(MPTCP_PM_ADD_ADDR_RECEIVED);
 869                 mptcp_pm_nl_add_addr_received(msk);
 870         }
 871         if (pm->status & BIT(MPTCP_PM_ADD_ADDR_SEND_ACK)) {
 872                 pm->status &= ~BIT(MPTCP_PM_ADD_ADDR_SEND_ACK);
 873                 mptcp_pm_nl_addr_send_ack(msk);
 874         }
 875         if (pm->status & BIT(MPTCP_PM_RM_ADDR_RECEIVED)) {
 876                 pm->status &= ~BIT(MPTCP_PM_RM_ADDR_RECEIVED);
 877                 mptcp_pm_nl_rm_addr_received(msk);
 878         }
 879         if (pm->status & BIT(MPTCP_PM_ESTABLISHED)) {
 880                 pm->status &= ~BIT(MPTCP_PM_ESTABLISHED);
 881                 mptcp_pm_nl_fully_established(msk);
 882         }
 883         if (pm->status & BIT(MPTCP_PM_SUBFLOW_ESTABLISHED)) {
 884                 pm->status &= ~BIT(MPTCP_PM_SUBFLOW_ESTABLISHED);
 885                 mptcp_pm_nl_subflow_established(msk);
 886         }
 887
 888         spin_unlock_bh(&msk->pm.lock);
 889 }
 890
 891 static bool address_use_port(struct mptcp_pm_addr_entry *entry)
 892 {
 893         return (entry->flags &
 894                 (MPTCP_PM_ADDR_FLAG_SIGNAL | MPTCP_PM_ADDR_FLAG_SUBFLOW)) ==
 895                 MPTCP_PM_ADDR_FLAG_SIGNAL;
 896 }
 897
 898 /* caller must ensure the RCU grace period is already elapsed */
 899 static void __mptcp_pm_release_addr_entry(struct mptcp_pm_addr_entry *entry)
 900 {
 901         if (entry->lsk)
 902                 sock_release(entry->lsk);
 903         kfree(entry);
 904 }
 905
 906 static int mptcp_pm_nl_append_new_local_addr(struct pm_nl_pernet *pernet,
 907                                              struct mptcp_pm_addr_entry *entry,
 908                                              bool needs_id)
 909 {
 910         struct mptcp_pm_addr_entry *cur, *del_entry = NULL;
 911         unsigned int addr_max;
 912         int ret = -EINVAL;
 913
 914         spin_lock_bh(&pernet->lock);
 915         /* to keep the code simple, don't do IDR-like allocation for address ID,
 916          * just bail when we exceed limits
 917          */
 918         if (pernet->next_id == MPTCP_PM_MAX_ADDR_ID)
 919                 pernet->next_id = 1;
 920         if (pernet->addrs >= MPTCP_PM_ADDR_MAX) {
 921                 ret = -ERANGE;
 922                 goto out;
 923         }
 924         if (test_bit(entry->addr.id, pernet->id_bitmap)) {
 925                 ret = -EBUSY;
 926                 goto out;
 927         }
 928
 929         /* do not insert duplicate address, differentiate on port only
 930          * singled addresses
 931          */
 932         if (!address_use_port(entry))
 933                 entry->addr.port = 0;
 934         list_for_each_entry(cur, &pernet->local_addr_list, list) {
 935                 if (mptcp_addresses_equal(&cur->addr, &entry->addr,
 936                                           cur->addr.port || entry->addr.port)) {
 937                         /* allow replacing the exiting endpoint only if such
 938                          * endpoint is an implicit one and the user-space
 939                          * did not provide an endpoint id
 940                          */
 941                         if (!(cur->flags & MPTCP_PM_ADDR_FLAG_IMPLICIT)) {
 942                                 ret = -EEXIST;
 943                                 goto out;
 944                         }
 945                         if (entry->addr.id)
 946                                 goto out;
 947
 948                         pernet->addrs--;
 949                         entry->addr.id = cur->addr.id;
 950                         list_del_rcu(&cur->list);
 951                         del_entry = cur;
 952                         break;
 953                 }
 954         }
 955
 956         if (!entry->addr.id && needs_id) {
 957 find_next:
 958                 entry->addr.id = find_next_zero_bit(pernet->id_bitmap,
 959                                                     MPTCP_PM_MAX_ADDR_ID + 1,
 960                                                     pernet->next_id);
 961                 if (!entry->addr.id && pernet->next_id != 1) {
 962                         pernet->next_id = 1;
 963                         goto find_next;
 964                 }
 965         }
 966
 967         if (!entry->addr.id && needs_id)
 968                 goto out;
 969
 970         __set_bit(entry->addr.id, pernet->id_bitmap);
 971         if (entry->addr.id > pernet->next_id)
 972                 pernet->next_id = entry->addr.id;
 973
 974         if (entry->flags & MPTCP_PM_ADDR_FLAG_SIGNAL) {
 975                 addr_max = pernet->add_addr_signal_max;
 976                 WRITE_ONCE(pernet->add_addr_signal_max, addr_max + 1);
 977         }
 978         if (entry->flags & MPTCP_PM_ADDR_FLAG_SUBFLOW) {
 979                 addr_max = pernet->local_addr_max;
 980                 WRITE_ONCE(pernet->local_addr_max, addr_max + 1);
 981         }
 982
 983         pernet->addrs++;
 984         if (!entry->addr.port)
 985                 list_add_tail_rcu(&entry->list, &pernet->local_addr_list);
 986         else
 987                 list_add_rcu(&entry->list, &pernet->local_addr_list);
 988         ret = entry->addr.id;
 989
 990 out:
 991         spin_unlock_bh(&pernet->lock);
 992
 993         /* just replaced an existing entry, free it */
 994         if (del_entry) {
 995                 synchronize_rcu();
 996                 __mptcp_pm_release_addr_entry(del_entry);
 997         }
 998         return ret;
 999 }
1000
1001 static struct lock_class_key mptcp_slock_keys[2];
1002 static struct lock_class_key mptcp_keys[2];
1003
1004 static int mptcp_pm_nl_create_listen_socket(struct sock *sk,
1005                                             struct mptcp_pm_addr_entry *entry)
1006 {
1007         bool is_ipv6 = sk->sk_family == AF_INET6;
1008         int addrlen = sizeof(struct sockaddr_in);
1009         struct sockaddr_storage addr;
1010         struct sock *newsk, *ssk;
1011         int backlog = 1024;
1012         int err;
1013
1014         err = sock_create_kern(sock_net(sk), entry->addr.family,
1015                                SOCK_STREAM, IPPROTO_MPTCP, &entry->lsk);
1016         if (err)
1017                 return err;
1018
1019         newsk = entry->lsk->sk;
1020         if (!newsk)
1021                 return -EINVAL;
1022
1023         /* The subflow socket lock is acquired in a nested to the msk one
1024          * in several places, even by the TCP stack, and this msk is a kernel
1025          * socket: lockdep complains. Instead of propagating the _nested
1026          * modifiers in several places, re-init the lock class for the msk
1027          * socket to an mptcp specific one.
1028          */
1029         sock_lock_init_class_and_name(newsk,
1030                                       is_ipv6 ? "mlock-AF_INET6" : "mlock-AF_INET",
1031                                       &mptcp_slock_keys[is_ipv6],
1032                                       is_ipv6 ? "msk_lock-AF_INET6" : "msk_lock-AF_INET",
1033                                       &mptcp_keys[is_ipv6]);
1034
1035         lock_sock(newsk);
1036         ssk = __mptcp_nmpc_sk(mptcp_sk(newsk));
1037         release_sock(newsk);
1038         if (IS_ERR(ssk))
1039                 return PTR_ERR(ssk);
1040
1041         mptcp_info2sockaddr(&entry->addr, &addr, entry->addr.family);
1042 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
1043         if (entry->addr.family == AF_INET6)
1044                 addrlen = sizeof(struct sockaddr_in6);
1045 #endif
1046         if (ssk->sk_family == AF_INET)
1047                 err = inet_bind_sk(ssk, (struct sockaddr *)&addr, addrlen);
1048 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
1049         else if (ssk->sk_family == AF_INET6)
1050                 err = inet6_bind_sk(ssk, (struct sockaddr *)&addr, addrlen);
1051 #endif
1052         if (err)
1053                 return err;
1054
1055         /* We don't use mptcp_set_state() here because it needs to be called
1056          * under the msk socket lock. For the moment, that will not bring
1057          * anything more than only calling inet_sk_state_store(), because the
1058          * old status is known (TCP_CLOSE).
1059          */
1060         inet_sk_state_store(newsk, TCP_LISTEN);
1061         lock_sock(ssk);
1062         err = __inet_listen_sk(ssk, backlog);
1063         if (!err)
1064                 mptcp_event_pm_listener(ssk, MPTCP_EVENT_LISTENER_CREATED);
1065         release_sock(ssk);
1066         return err;
1067 }
1068
1069 int mptcp_pm_nl_get_local_id(struct mptcp_sock *msk, struct mptcp_addr_info *skc)
1070 {
1071         struct mptcp_pm_addr_entry *entry;
1072         struct pm_nl_pernet *pernet;
1073         int ret = -1;
1074
1075         pernet = pm_nl_get_pernet_from_msk(msk);
1076
1077         rcu_read_lock();
1078         list_for_each_entry_rcu(entry, &pernet->local_addr_list, list) {
1079                 if (mptcp_addresses_equal(&entry->addr, skc, entry->addr.port)) {
1080                         ret = entry->addr.id;
1081                         break;
1082                 }
1083         }
1084         rcu_read_unlock();
1085         if (ret >= 0)
1086                 return ret;
1087
1088         /* address not found, add to local list */
1089         entry = kmalloc(sizeof(*entry), GFP_ATOMIC);
1090         if (!entry)
1091                 return -ENOMEM;
1092
1093         entry->addr = *skc;
1094         entry->addr.id = 0;
1095         entry->addr.port = 0;
1096         entry->ifindex = 0;
1097         entry->flags = MPTCP_PM_ADDR_FLAG_IMPLICIT;
1098         entry->lsk = NULL;
1099         ret = mptcp_pm_nl_append_new_local_addr(pernet, entry, true);
1100         if (ret < 0)
1101                 kfree(entry);
1102
1103         return ret;
1104 }
1105
1106 #define MPTCP_PM_CMD_GRP_OFFSET       0
1107 #define MPTCP_PM_EV_GRP_OFFSET        1
1108
1109 static const struct genl_multicast_group mptcp_pm_mcgrps[] = {
1110         [MPTCP_PM_CMD_GRP_OFFSET]       = { .name = MPTCP_PM_CMD_GRP_NAME, },
1111         [MPTCP_PM_EV_GRP_OFFSET]        = { .name = MPTCP_PM_EV_GRP_NAME,
1112                                             .flags = GENL_UNS_ADMIN_PERM,
1113                                           },
1114 };
1115
1116 static const struct nla_policy
1117 mptcp_pm_addr_policy[MPTCP_PM_ADDR_ATTR_MAX + 1] = {
1118         [MPTCP_PM_ADDR_ATTR_FAMILY]     = { .type       = NLA_U16,      },
1119         [MPTCP_PM_ADDR_ATTR_ID]         = { .type       = NLA_U8,       },
1120         [MPTCP_PM_ADDR_ATTR_ADDR4]      = { .type       = NLA_U32,      },
1121         [MPTCP_PM_ADDR_ATTR_ADDR6]      =
1122                 NLA_POLICY_EXACT_LEN(sizeof(struct in6_addr)),
1123         [MPTCP_PM_ADDR_ATTR_PORT]       = { .type       = NLA_U16       },
1124         [MPTCP_PM_ADDR_ATTR_FLAGS]      = { .type       = NLA_U32       },
1125         [MPTCP_PM_ADDR_ATTR_IF_IDX]     = { .type       = NLA_S32       },
1126 };
1127
1128 static const struct nla_policy mptcp_pm_policy[MPTCP_PM_ATTR_MAX + 1] = {
1129         [MPTCP_PM_ATTR_ADDR]            =
1130                                         NLA_POLICY_NESTED(mptcp_pm_addr_policy),
1131         [MPTCP_PM_ATTR_RCV_ADD_ADDRS]   = { .type       = NLA_U32,      },
1132         [MPTCP_PM_ATTR_SUBFLOWS]        = { .type       = NLA_U32,      },
1133         [MPTCP_PM_ATTR_TOKEN]           = { .type       = NLA_U32,      },
1134         [MPTCP_PM_ATTR_LOC_ID]          = { .type       = NLA_U8,       },
1135         [MPTCP_PM_ATTR_ADDR_REMOTE]     =
1136                                         NLA_POLICY_NESTED(mptcp_pm_addr_policy),
1137 };
1138
1139 void mptcp_pm_nl_subflow_chk_stale(const struct mptcp_sock *msk, struct sock *ssk)
1140 {
1141         struct mptcp_subflow_context *iter, *subflow = mptcp_subflow_ctx(ssk);
1142         struct sock *sk = (struct sock *)msk;
1143         unsigned int active_max_loss_cnt;
1144         struct net *net = sock_net(sk);
1145         unsigned int stale_loss_cnt;
1146         bool slow;
1147
1148         stale_loss_cnt = mptcp_stale_loss_cnt(net);
1149         if (subflow->stale || !stale_loss_cnt || subflow->stale_count <= stale_loss_cnt)
1150                 return;
1151
1152         /* look for another available subflow not in loss state */
1153         active_max_loss_cnt = max_t(int, stale_loss_cnt - 1, 1);
1154         mptcp_for_each_subflow(msk, iter) {
1155                 if (iter != subflow && mptcp_subflow_active(iter) &&
1156                     iter->stale_count < active_max_loss_cnt) {
1157                         /* we have some alternatives, try to mark this subflow as idle ...*/
1158                         slow = lock_sock_fast(ssk);
1159                         if (!tcp_rtx_and_write_queues_empty(ssk)) {
1160                                 subflow->stale = 1;
1161                                 __mptcp_retransmit_pending_data(sk);
1162                                 MPTCP_INC_STATS(net, MPTCP_MIB_SUBFLOWSTALE);
1163                         }
1164                         unlock_sock_fast(ssk, slow);
1165
1166                         /* always try to push the pending data regardless of re-injections:
1167                          * we can possibly use backup subflows now, and subflow selection
1168                          * is cheap under the msk socket lock
1169                          */
1170                         __mptcp_push_pending(sk, 0);
1171                         return;
1172                 }
1173         }
1174 }
1175
1176 static int mptcp_pm_family_to_addr(int family)
1177 {
1178 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
1179         if (family == AF_INET6)
1180                 return MPTCP_PM_ADDR_ATTR_ADDR6;
1181 #endif
1182         return MPTCP_PM_ADDR_ATTR_ADDR4;
1183 }
1184
1185 static int mptcp_pm_parse_pm_addr_attr(struct nlattr *tb[],
1186                                        const struct nlattr *attr,
1187                                        struct genl_info *info,
1188                                        struct mptcp_addr_info *addr,
1189                                        bool require_family)
1190 {
1191         int err, addr_addr;
1192
1193         if (!attr) {
1194                 GENL_SET_ERR_MSG(info, "missing address info");
1195                 return -EINVAL;
1196         }
1197
1198         /* no validation needed - was already done via nested policy */
1199         err = nla_parse_nested_deprecated(tb, MPTCP_PM_ADDR_ATTR_MAX, attr,
1200                                           mptcp_pm_addr_policy, info->extack);
1201         if (err)
1202                 return err;
1203
1204         if (tb[MPTCP_PM_ADDR_ATTR_ID])
1205                 addr->id = nla_get_u8(tb[MPTCP_PM_ADDR_ATTR_ID]);
1206
1207         if (!tb[MPTCP_PM_ADDR_ATTR_FAMILY]) {
1208                 if (!require_family)
1209                         return 0;
1210
1211                 NL_SET_ERR_MSG_ATTR(info->extack, attr,
1212                                     "missing family");
1213                 return -EINVAL;
1214         }
1215
1216         addr->family = nla_get_u16(tb[MPTCP_PM_ADDR_ATTR_FAMILY]);
1217         if (addr->family != AF_INET
1218 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
1219             && addr->family != AF_INET6
1220 #endif
1221             ) {
1222                 NL_SET_ERR_MSG_ATTR(info->extack, attr,
1223                                     "unknown address family");
1224                 return -EINVAL;
1225         }
1226         addr_addr = mptcp_pm_family_to_addr(addr->family);
1227         if (!tb[addr_addr]) {
1228                 NL_SET_ERR_MSG_ATTR(info->extack, attr,
1229                                     "missing address data");
1230                 return -EINVAL;
1231         }
1232
1233 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
1234         if (addr->family == AF_INET6)
1235                 addr->addr6 = nla_get_in6_addr(tb[addr_addr]);
1236         else
1237 #endif
1238                 addr->addr.s_addr = nla_get_in_addr(tb[addr_addr]);
1239
1240         if (tb[MPTCP_PM_ADDR_ATTR_PORT])
1241                 addr->port = htons(nla_get_u16(tb[MPTCP_PM_ADDR_ATTR_PORT]));
1242
1243         return 0;
1244 }
1245
1246 int mptcp_pm_parse_addr(struct nlattr *attr, struct genl_info *info,
1247                         struct mptcp_addr_info *addr)
1248 {
1249         struct nlattr *tb[MPTCP_PM_ADDR_ATTR_MAX + 1];
1250
1251         memset(addr, 0, sizeof(*addr));
1252
1253         return mptcp_pm_parse_pm_addr_attr(tb, attr, info, addr, true);
1254 }
1255
1256 int mptcp_pm_parse_entry(struct nlattr *attr, struct genl_info *info,
1257                          bool require_family,
1258                          struct mptcp_pm_addr_entry *entry)
1259 {
1260         struct nlattr *tb[MPTCP_PM_ADDR_ATTR_MAX + 1];
1261         int err;
1262
1263         memset(entry, 0, sizeof(*entry));
1264
1265         err = mptcp_pm_parse_pm_addr_attr(tb, attr, info, &entry->addr, require_family);
1266         if (err)
1267                 return err;
1268
1269         if (tb[MPTCP_PM_ADDR_ATTR_IF_IDX]) {
1270                 u32 val = nla_get_s32(tb[MPTCP_PM_ADDR_ATTR_IF_IDX]);
1271
1272                 entry->ifindex = val;
1273         }
1274
1275         if (tb[MPTCP_PM_ADDR_ATTR_FLAGS])
1276                 entry->flags = nla_get_u32(tb[MPTCP_PM_ADDR_ATTR_FLAGS]);
1277
1278         if (tb[MPTCP_PM_ADDR_ATTR_PORT])
1279                 entry->addr.port = htons(nla_get_u16(tb[MPTCP_PM_ADDR_ATTR_PORT]));
1280
1281         return 0;
1282 }
1283
1284 static struct pm_nl_pernet *genl_info_pm_nl(struct genl_info *info)
1285 {
1286         return pm_nl_get_pernet(genl_info_net(info));
1287 }
1288
1289 static int mptcp_nl_add_subflow_or_signal_addr(struct net *net)
1290 {
1291         struct mptcp_sock *msk;
1292         long s_slot = 0, s_num = 0;
1293
1294         while ((msk = mptcp_token_iter_next(net, &s_slot, &s_num)) != NULL) {
1295                 struct sock *sk = (struct sock *)msk;
1296
1297                 if (!READ_ONCE(msk->fully_established) ||
1298                     mptcp_pm_is_userspace(msk))
1299                         goto next;
1300
1301                 lock_sock(sk);
1302                 spin_lock_bh(&msk->pm.lock);
1303                 mptcp_pm_create_subflow_or_signal_addr(msk);
1304                 spin_unlock_bh(&msk->pm.lock);
1305                 release_sock(sk);
1306
1307 next:
1308                 sock_put(sk);
1309                 cond_resched();
1310         }
1311
1312         return 0;
1313 }
1314
1315 static bool mptcp_pm_has_addr_attr_id(const struct nlattr *attr,
1316                                       struct genl_info *info)
1317 {
1318         struct nlattr *tb[MPTCP_PM_ADDR_ATTR_MAX + 1];
1319
1320         if (!nla_parse_nested_deprecated(tb, MPTCP_PM_ADDR_ATTR_MAX, attr,
1321                                          mptcp_pm_addr_policy, info->extack) &&
1322             tb[MPTCP_PM_ADDR_ATTR_ID])
1323                 return true;
1324         return false;
1325 }
1326
1327 static int mptcp_nl_cmd_add_addr(struct sk_buff *skb, struct genl_info *info)
1328 {
1329         struct nlattr *attr = info->attrs[MPTCP_PM_ATTR_ADDR];
1330         struct pm_nl_pernet *pernet = genl_info_pm_nl(info);
1331         struct mptcp_pm_addr_entry addr, *entry;
1332         int ret;
1333
1334         ret = mptcp_pm_parse_entry(attr, info, true, &addr);
1335         if (ret < 0)
1336                 return ret;
1337
1338         if (addr.addr.port && !(addr.flags & MPTCP_PM_ADDR_FLAG_SIGNAL)) {
1339                 GENL_SET_ERR_MSG(info, "flags must have signal when using port");
1340                 return -EINVAL;
1341         }
1342
1343         if (addr.flags & MPTCP_PM_ADDR_FLAG_SIGNAL &&
1344             addr.flags & MPTCP_PM_ADDR_FLAG_FULLMESH) {
1345                 GENL_SET_ERR_MSG(info, "flags mustn't have both signal and fullmesh");
1346                 return -EINVAL;
1347         }
1348
1349         if (addr.flags & MPTCP_PM_ADDR_FLAG_IMPLICIT) {
1350                 GENL_SET_ERR_MSG(info, "can't create IMPLICIT endpoint");
1351                 return -EINVAL;
1352         }
1353
1354         entry = kzalloc(sizeof(*entry), GFP_KERNEL_ACCOUNT);
1355         if (!entry) {
1356                 GENL_SET_ERR_MSG(info, "can't allocate addr");
1357                 return -ENOMEM;
1358         }
1359
1360         *entry = addr;
1361         if (entry->addr.port) {
1362                 ret = mptcp_pm_nl_create_listen_socket(skb->sk, entry);
1363                 if (ret) {
1364                         GENL_SET_ERR_MSG_FMT(info, "create listen socket error: %d", ret);
1365                         goto out_free;
1366                 }
1367         }
1368         ret = mptcp_pm_nl_append_new_local_addr(pernet, entry,
1369                                                 !mptcp_pm_has_addr_attr_id(attr, info));
1370         if (ret < 0) {
1371                 GENL_SET_ERR_MSG_FMT(info, "too many addresses or duplicate one: %d", ret);
1372                 goto out_free;
1373         }
1374
1375         mptcp_nl_add_subflow_or_signal_addr(sock_net(skb->sk));
1376         return 0;
1377
1378 out_free:
1379         __mptcp_pm_release_addr_entry(entry);
1380         return ret;
1381 }
1382
1383 int mptcp_pm_nl_get_flags_and_ifindex_by_id(struct mptcp_sock *msk, unsigned int id,
1384                                             u8 *flags, int *ifindex)
1385 {
1386         struct mptcp_pm_addr_entry *entry;
1387         struct sock *sk = (struct sock *)msk;
1388         struct net *net = sock_net(sk);
1389
1390         rcu_read_lock();
1391         entry = __lookup_addr_by_id(pm_nl_get_pernet(net), id);
1392         if (entry) {
1393                 *flags = entry->flags;
1394                 *ifindex = entry->ifindex;
1395         }
1396         rcu_read_unlock();
1397
1398         return 0;
1399 }
1400
1401 static bool remove_anno_list_by_saddr(struct mptcp_sock *msk,
1402                                       const struct mptcp_addr_info *addr)
1403 {
1404         struct mptcp_pm_add_entry *entry;
1405
1406         entry = mptcp_pm_del_add_timer(msk, addr, false);
1407         if (entry) {
1408                 list_del(&entry->list);
1409                 kfree(entry);
1410                 return true;
1411         }
1412
1413         return false;
1414 }
1415
1416 static bool mptcp_pm_remove_anno_addr(struct mptcp_sock *msk,
1417                                       const struct mptcp_addr_info *addr,
1418                                       bool force)
1419 {
1420         struct mptcp_rm_list list = { .nr = 0 };
1421         bool ret;
1422
1423         list.ids[list.nr++] = addr->id;
1424
1425         ret = remove_anno_list_by_saddr(msk, addr);
1426         if (ret || force) {
1427                 spin_lock_bh(&msk->pm.lock);
1428                 mptcp_pm_remove_addr(msk, &list);
1429                 spin_unlock_bh(&msk->pm.lock);
1430         }
1431         return ret;
1432 }
1433
1434 static int mptcp_nl_remove_subflow_and_signal_addr(struct net *net,
1435                                                    const struct mptcp_pm_addr_entry *entry)
1436 {
1437         const struct mptcp_addr_info *addr = &entry->addr;
1438         struct mptcp_rm_list list = { .nr = 0 };
1439         long s_slot = 0, s_num = 0;
1440         struct mptcp_sock *msk;
1441
1442         pr_debug("remove_id=%d", addr->id);
1443
1444         list.ids[list.nr++] = addr->id;
1445
1446         while ((msk = mptcp_token_iter_next(net, &s_slot, &s_num)) != NULL) {
1447                 struct sock *sk = (struct sock *)msk;
1448                 bool remove_subflow;
1449
1450                 if (mptcp_pm_is_userspace(msk))
1451                         goto next;
1452
1453                 if (list_empty(&msk->conn_list)) {
1454                         mptcp_pm_remove_anno_addr(msk, addr, false);
1455                         goto next;
1456                 }
1457
1458                 lock_sock(sk);
1459                 remove_subflow = lookup_subflow_by_saddr(&msk->conn_list, addr);
1460                 mptcp_pm_remove_anno_addr(msk, addr, remove_subflow &&
1461                                           !(entry->flags & MPTCP_PM_ADDR_FLAG_IMPLICIT));
1462                 if (remove_subflow)
1463                         mptcp_pm_remove_subflow(msk, &list);
1464                 release_sock(sk);
1465
1466 next:
1467                 sock_put(sk);
1468                 cond_resched();
1469         }
1470
1471         return 0;
1472 }
1473
1474 static int mptcp_nl_remove_id_zero_address(struct net *net,
1475                                            struct mptcp_addr_info *addr)
1476 {
1477         struct mptcp_rm_list list = { .nr = 0 };
1478         long s_slot = 0, s_num = 0;
1479         struct mptcp_sock *msk;
1480
1481         list.ids[list.nr++] = 0;
1482
1483         while ((msk = mptcp_token_iter_next(net, &s_slot, &s_num)) != NULL) {
1484                 struct sock *sk = (struct sock *)msk;
1485                 struct mptcp_addr_info msk_local;
1486
1487                 if (list_empty(&msk->conn_list) || mptcp_pm_is_userspace(msk))
1488                         goto next;
1489
1490                 mptcp_local_address((struct sock_common *)msk, &msk_local);
1491                 if (!mptcp_addresses_equal(&msk_local, addr, addr->port))
1492                         goto next;
1493
1494                 lock_sock(sk);
1495                 spin_lock_bh(&msk->pm.lock);
1496                 mptcp_pm_remove_addr(msk, &list);
1497                 mptcp_pm_nl_rm_subflow_received(msk, &list);
1498                 spin_unlock_bh(&msk->pm.lock);
1499                 release_sock(sk);
1500
1501 next:
1502                 sock_put(sk);
1503                 cond_resched();
1504         }
1505
1506         return 0;
1507 }
1508
1509 static int mptcp_nl_cmd_del_addr(struct sk_buff *skb, struct genl_info *info)
1510 {
1511         struct nlattr *attr = info->attrs[MPTCP_PM_ATTR_ADDR];
1512         struct pm_nl_pernet *pernet = genl_info_pm_nl(info);
1513         struct mptcp_pm_addr_entry addr, *entry;
1514         unsigned int addr_max;
1515         int ret;
1516
1517         ret = mptcp_pm_parse_entry(attr, info, false, &addr);
1518         if (ret < 0)
1519                 return ret;
1520
1521         /* the zero id address is special: the first address used by the msk
1522          * always gets such an id, so different subflows can have different zero
1523          * id addresses. Additionally zero id is not accounted for in id_bitmap.
1524          * Let's use an 'mptcp_rm_list' instead of the common remove code.
1525          */
1526         if (addr.addr.id == 0)
1527                 return mptcp_nl_remove_id_zero_address(sock_net(skb->sk), &addr.addr);
1528
1529         spin_lock_bh(&pernet->lock);
1530         entry = __lookup_addr_by_id(pernet, addr.addr.id);
1531         if (!entry) {
1532                 GENL_SET_ERR_MSG(info, "address not found");
1533                 spin_unlock_bh(&pernet->lock);
1534                 return -EINVAL;
1535         }
1536         if (entry->flags & MPTCP_PM_ADDR_FLAG_SIGNAL) {
1537                 addr_max = pernet->add_addr_signal_max;
1538                 WRITE_ONCE(pernet->add_addr_signal_max, addr_max - 1);
1539         }
1540         if (entry->flags & MPTCP_PM_ADDR_FLAG_SUBFLOW) {
1541                 addr_max = pernet->local_addr_max;
1542                 WRITE_ONCE(pernet->local_addr_max, addr_max - 1);
1543         }
1544
1545         pernet->addrs--;
1546         list_del_rcu(&entry->list);
1547         __clear_bit(entry->addr.id, pernet->id_bitmap);
1548         spin_unlock_bh(&pernet->lock);
1549
1550         mptcp_nl_remove_subflow_and_signal_addr(sock_net(skb->sk), entry);
1551         synchronize_rcu();
1552         __mptcp_pm_release_addr_entry(entry);
1553
1554         return ret;
1555 }
1556
1557 void mptcp_pm_remove_addrs(struct mptcp_sock *msk, struct list_head *rm_list)
1558 {
1559         struct mptcp_rm_list alist = { .nr = 0 };
1560         struct mptcp_pm_addr_entry *entry;
1561
1562         list_for_each_entry(entry, rm_list, list) {
1563                 if ((remove_anno_list_by_saddr(msk, &entry->addr) ||
1564                      lookup_subflow_by_saddr(&msk->conn_list, &entry->addr)) &&
1565                     alist.nr < MPTCP_RM_IDS_MAX)
1566                         alist.ids[alist.nr++] = entry->addr.id;
1567         }
1568
1569         if (alist.nr) {
1570                 spin_lock_bh(&msk->pm.lock);
1571                 mptcp_pm_remove_addr(msk, &alist);
1572                 spin_unlock_bh(&msk->pm.lock);
1573         }
1574 }
1575
1576 void mptcp_pm_remove_addrs_and_subflows(struct mptcp_sock *msk,
1577                                         struct list_head *rm_list)
1578 {
1579         struct mptcp_rm_list alist = { .nr = 0 }, slist = { .nr = 0 };
1580         struct mptcp_pm_addr_entry *entry;
1581
1582         list_for_each_entry(entry, rm_list, list) {
1583                 if (lookup_subflow_by_saddr(&msk->conn_list, &entry->addr) &&
1584                     slist.nr < MPTCP_RM_IDS_MAX)
1585                         slist.ids[slist.nr++] = entry->addr.id;
1586
1587                 if (remove_anno_list_by_saddr(msk, &entry->addr) &&
1588                     alist.nr < MPTCP_RM_IDS_MAX)
1589                         alist.ids[alist.nr++] = entry->addr.id;
1590         }
1591
1592         if (alist.nr) {
1593                 spin_lock_bh(&msk->pm.lock);
1594                 mptcp_pm_remove_addr(msk, &alist);
1595                 spin_unlock_bh(&msk->pm.lock);
1596         }
1597         if (slist.nr)
1598                 mptcp_pm_remove_subflow(msk, &slist);
1599 }
1600
1601 static void mptcp_nl_remove_addrs_list(struct net *net,
1602                                        struct list_head *rm_list)
1603 {
1604         long s_slot = 0, s_num = 0;
1605         struct mptcp_sock *msk;
1606
1607         if (list_empty(rm_list))
1608                 return;
1609
1610         while ((msk = mptcp_token_iter_next(net, &s_slot, &s_num)) != NULL) {
1611                 struct sock *sk = (struct sock *)msk;
1612
1613                 if (!mptcp_pm_is_userspace(msk)) {
1614                         lock_sock(sk);
1615                         mptcp_pm_remove_addrs_and_subflows(msk, rm_list);
1616                         release_sock(sk);
1617                 }
1618
1619                 sock_put(sk);
1620                 cond_resched();
1621         }
1622 }
1623
1624 /* caller must ensure the RCU grace period is already elapsed */
1625 static void __flush_addrs(struct list_head *list)
1626 {
1627         while (!list_empty(list)) {
1628                 struct mptcp_pm_addr_entry *cur;
1629
1630                 cur = list_entry(list->next,
1631                                  struct mptcp_pm_addr_entry, list);
1632                 list_del_rcu(&cur->list);
1633                 __mptcp_pm_release_addr_entry(cur);
1634         }
1635 }
1636
1637 static void __reset_counters(struct pm_nl_pernet *pernet)
1638 {
1639         WRITE_ONCE(pernet->add_addr_signal_max, 0);
1640         WRITE_ONCE(pernet->add_addr_accept_max, 0);
1641         WRITE_ONCE(pernet->local_addr_max, 0);
1642         pernet->addrs = 0;
1643 }
1644
1645 static int mptcp_nl_cmd_flush_addrs(struct sk_buff *skb, struct genl_info *info)
1646 {
1647         struct pm_nl_pernet *pernet = genl_info_pm_nl(info);
1648         LIST_HEAD(free_list);
1649
1650         spin_lock_bh(&pernet->lock);
1651         list_splice_init(&pernet->local_addr_list, &free_list);
1652         __reset_counters(pernet);
1653         pernet->next_id = 1;
1654         bitmap_zero(pernet->id_bitmap, MPTCP_PM_MAX_ADDR_ID + 1);
1655         spin_unlock_bh(&pernet->lock);
1656         mptcp_nl_remove_addrs_list(sock_net(skb->sk), &free_list);
1657         synchronize_rcu();
1658         __flush_addrs(&free_list);
1659         return 0;
1660 }
1661
1662 static int mptcp_nl_fill_addr(struct sk_buff *skb,
1663                               struct mptcp_pm_addr_entry *entry)
1664 {
1665         struct mptcp_addr_info *addr = &entry->addr;
1666         struct nlattr *attr;
1667
1668         attr = nla_nest_start(skb, MPTCP_PM_ATTR_ADDR);
1669         if (!attr)
1670                 return -EMSGSIZE;
1671
1672         if (nla_put_u16(skb, MPTCP_PM_ADDR_ATTR_FAMILY, addr->family))
1673                 goto nla_put_failure;
1674         if (nla_put_u16(skb, MPTCP_PM_ADDR_ATTR_PORT, ntohs(addr->port)))
1675                 goto nla_put_failure;
1676         if (nla_put_u8(skb, MPTCP_PM_ADDR_ATTR_ID, addr->id))
1677                 goto nla_put_failure;
1678         if (nla_put_u32(skb, MPTCP_PM_ADDR_ATTR_FLAGS, entry->flags))
1679                 goto nla_put_failure;
1680         if (entry->ifindex &&
1681             nla_put_s32(skb, MPTCP_PM_ADDR_ATTR_IF_IDX, entry->ifindex))
1682                 goto nla_put_failure;
1683
1684         if (addr->family == AF_INET &&
1685             nla_put_in_addr(skb, MPTCP_PM_ADDR_ATTR_ADDR4,
1686                             addr->addr.s_addr))
1687                 goto nla_put_failure;
1688 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
1689         else if (addr->family == AF_INET6 &&
1690                  nla_put_in6_addr(skb, MPTCP_PM_ADDR_ATTR_ADDR6, &addr->addr6))
1691                 goto nla_put_failure;
1692 #endif
1693         nla_nest_end(skb, attr);
1694         return 0;
1695
1696 nla_put_failure:
1697         nla_nest_cancel(skb, attr);
1698         return -EMSGSIZE;
1699 }
1700
1701 static int mptcp_nl_cmd_get_addr(struct sk_buff *skb, struct genl_info *info)
1702 {
1703         struct nlattr *attr = info->attrs[MPTCP_PM_ATTR_ADDR];
1704         struct pm_nl_pernet *pernet = genl_info_pm_nl(info);
1705         struct mptcp_pm_addr_entry addr, *entry;
1706         struct sk_buff *msg;
1707         void *reply;
1708         int ret;
1709
1710         ret = mptcp_pm_parse_entry(attr, info, false, &addr);
1711         if (ret < 0)
1712                 return ret;
1713
1714         msg = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
1715         if (!msg)
1716                 return -ENOMEM;
1717
1718         reply = genlmsg_put_reply(msg, info, &mptcp_genl_family, 0,
1719                                   info->genlhdr->cmd);
1720         if (!reply) {
1721                 GENL_SET_ERR_MSG(info, "not enough space in Netlink message");
1722                 ret = -EMSGSIZE;
1723                 goto fail;
1724         }
1725
1726         spin_lock_bh(&pernet->lock);
1727         entry = __lookup_addr_by_id(pernet, addr.addr.id);
1728         if (!entry) {
1729                 GENL_SET_ERR_MSG(info, "address not found");
1730                 ret = -EINVAL;
1731                 goto unlock_fail;
1732         }
1733
1734         ret = mptcp_nl_fill_addr(msg, entry);
1735         if (ret)
1736                 goto unlock_fail;
1737
1738         genlmsg_end(msg, reply);
1739         ret = genlmsg_reply(msg, info);
1740         spin_unlock_bh(&pernet->lock);
1741         return ret;
1742
1743 unlock_fail:
1744         spin_unlock_bh(&pernet->lock);
1745
1746 fail:
1747         nlmsg_free(msg);
1748         return ret;
1749 }
1750
1751 static int mptcp_nl_cmd_dump_addrs(struct sk_buff *msg,
1752                                    struct netlink_callback *cb)
1753 {
1754         struct net *net = sock_net(msg->sk);
1755         struct mptcp_pm_addr_entry *entry;
1756         struct pm_nl_pernet *pernet;
1757         int id = cb->args[0];
1758         void *hdr;
1759         int i;
1760
1761         pernet = pm_nl_get_pernet(net);
1762
1763         spin_lock_bh(&pernet->lock);
1764         for (i = id; i < MPTCP_PM_MAX_ADDR_ID + 1; i++) {
1765                 if (test_bit(i, pernet->id_bitmap)) {
1766                         entry = __lookup_addr_by_id(pernet, i);
1767                         if (!entry)
1768                                 break;
1769
1770                         if (entry->addr.id <= id)
1771                                 continue;
1772
1773                         hdr = genlmsg_put(msg, NETLINK_CB(cb->skb).portid,
1774                                           cb->nlh->nlmsg_seq, &mptcp_genl_family,
1775                                           NLM_F_MULTI, MPTCP_PM_CMD_GET_ADDR);
1776                         if (!hdr)
1777                                 break;
1778
1779                         if (mptcp_nl_fill_addr(msg, entry) < 0) {
1780                                 genlmsg_cancel(msg, hdr);
1781                                 break;
1782                         }
1783
1784                         id = entry->addr.id;
1785                         genlmsg_end(msg, hdr);
1786                 }
1787         }
1788         spin_unlock_bh(&pernet->lock);
1789
1790         cb->args[0] = id;
1791         return msg->len;
1792 }
1793
1794 static int parse_limit(struct genl_info *info, int id, unsigned int *limit)
1795 {
1796         struct nlattr *attr = info->attrs[id];
1797
1798         if (!attr)
1799                 return 0;
1800
1801         *limit = nla_get_u32(attr);
1802         if (*limit > MPTCP_PM_ADDR_MAX) {
1803                 GENL_SET_ERR_MSG(info, "limit greater than maximum");
1804                 return -EINVAL;
1805         }
1806         return 0;
1807 }
1808
1809 static int
1810 mptcp_nl_cmd_set_limits(struct sk_buff *skb, struct genl_info *info)
1811 {
1812         struct pm_nl_pernet *pernet = genl_info_pm_nl(info);
1813         unsigned int rcv_addrs, subflows;
1814         int ret;
1815
1816         spin_lock_bh(&pernet->lock);
1817         rcv_addrs = pernet->add_addr_accept_max;
1818         ret = parse_limit(info, MPTCP_PM_ATTR_RCV_ADD_ADDRS, &rcv_addrs);
1819         if (ret)
1820                 goto unlock;
1821
1822         subflows = pernet->subflows_max;
1823         ret = parse_limit(info, MPTCP_PM_ATTR_SUBFLOWS, &subflows);
1824         if (ret)
1825                 goto unlock;
1826
1827         WRITE_ONCE(pernet->add_addr_accept_max, rcv_addrs);
1828         WRITE_ONCE(pernet->subflows_max, subflows);
1829
1830 unlock:
1831         spin_unlock_bh(&pernet->lock);
1832         return ret;
1833 }
1834
1835 static int
1836 mptcp_nl_cmd_get_limits(struct sk_buff *skb, struct genl_info *info)
1837 {
1838         struct pm_nl_pernet *pernet = genl_info_pm_nl(info);
1839         struct sk_buff *msg;
1840         void *reply;
1841
1842         msg = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
1843         if (!msg)
1844                 return -ENOMEM;
1845
1846         reply = genlmsg_put_reply(msg, info, &mptcp_genl_family, 0,
1847                                   MPTCP_PM_CMD_GET_LIMITS);
1848         if (!reply)
1849                 goto fail;
1850
1851         if (nla_put_u32(msg, MPTCP_PM_ATTR_RCV_ADD_ADDRS,
1852                         READ_ONCE(pernet->add_addr_accept_max)))
1853                 goto fail;
1854
1855         if (nla_put_u32(msg, MPTCP_PM_ATTR_SUBFLOWS,
1856                         READ_ONCE(pernet->subflows_max)))
1857                 goto fail;
1858
1859         genlmsg_end(msg, reply);
1860         return genlmsg_reply(msg, info);
1861
1862 fail:
1863         GENL_SET_ERR_MSG(info, "not enough space in Netlink message");
1864         nlmsg_free(msg);
1865         return -EMSGSIZE;
1866 }
1867
1868 static void mptcp_pm_nl_fullmesh(struct mptcp_sock *msk,
1869                                  struct mptcp_addr_info *addr)
1870 {
1871         struct mptcp_rm_list list = { .nr = 0 };
1872
1873         list.ids[list.nr++] = addr->id;
1874
1875         spin_lock_bh(&msk->pm.lock);
1876         mptcp_pm_nl_rm_subflow_received(msk, &list);
1877         mptcp_pm_create_subflow_or_signal_addr(msk);
1878         spin_unlock_bh(&msk->pm.lock);
1879 }
1880
1881 static int mptcp_nl_set_flags(struct net *net,
1882                               struct mptcp_addr_info *addr,
1883                               u8 bkup, u8 changed)
1884 {
1885         long s_slot = 0, s_num = 0;
1886         struct mptcp_sock *msk;
1887         int ret = -EINVAL;
1888
1889         while ((msk = mptcp_token_iter_next(net, &s_slot, &s_num)) != NULL) {
1890                 struct sock *sk = (struct sock *)msk;
1891
1892                 if (list_empty(&msk->conn_list) || mptcp_pm_is_userspace(msk))
1893                         goto next;
1894
1895                 lock_sock(sk);
1896                 if (changed & MPTCP_PM_ADDR_FLAG_BACKUP)
1897                         ret = mptcp_pm_nl_mp_prio_send_ack(msk, addr, NULL, bkup);
1898                 if (changed & MPTCP_PM_ADDR_FLAG_FULLMESH)
1899                         mptcp_pm_nl_fullmesh(msk, addr);
1900                 release_sock(sk);
1901
1902 next:
1903                 sock_put(sk);
1904                 cond_resched();
1905         }
1906
1907         return ret;
1908 }
1909
1910 int mptcp_pm_nl_set_flags(struct net *net, struct mptcp_pm_addr_entry *addr, u8 bkup)
1911 {
1912         struct pm_nl_pernet *pernet = pm_nl_get_pernet(net);
1913         u8 changed, mask = MPTCP_PM_ADDR_FLAG_BACKUP |
1914                            MPTCP_PM_ADDR_FLAG_FULLMESH;
1915         struct mptcp_pm_addr_entry *entry;
1916         u8 lookup_by_id = 0;
1917
1918         if (addr->addr.family == AF_UNSPEC) {
1919                 lookup_by_id = 1;
1920                 if (!addr->addr.id)
1921                         return -EOPNOTSUPP;
1922         }
1923
1924         spin_lock_bh(&pernet->lock);
1925         entry = __lookup_addr(pernet, &addr->addr, lookup_by_id);
1926         if (!entry) {
1927                 spin_unlock_bh(&pernet->lock);
1928                 return -EINVAL;
1929         }
1930         if ((addr->flags & MPTCP_PM_ADDR_FLAG_FULLMESH) &&
1931             (entry->flags & MPTCP_PM_ADDR_FLAG_SIGNAL)) {
1932                 spin_unlock_bh(&pernet->lock);
1933                 return -EINVAL;
1934         }
1935
1936         changed = (addr->flags ^ entry->flags) & mask;
1937         entry->flags = (entry->flags & ~mask) | (addr->flags & mask);
1938         *addr = *entry;
1939         spin_unlock_bh(&pernet->lock);
1940
1941         mptcp_nl_set_flags(net, &addr->addr, bkup, changed);
1942         return 0;
1943 }
1944
1945 static int mptcp_nl_cmd_set_flags(struct sk_buff *skb, struct genl_info *info)
1946 {
1947         struct mptcp_pm_addr_entry remote = { .addr = { .family = AF_UNSPEC }, };
1948         struct mptcp_pm_addr_entry addr = { .addr = { .family = AF_UNSPEC }, };
1949         struct nlattr *attr_rem = info->attrs[MPTCP_PM_ATTR_ADDR_REMOTE];
1950         struct nlattr *token = info->attrs[MPTCP_PM_ATTR_TOKEN];
1951         struct nlattr *attr = info->attrs[MPTCP_PM_ATTR_ADDR];
1952         struct net *net = sock_net(skb->sk);
1953         u8 bkup = 0;
1954         int ret;
1955
1956         ret = mptcp_pm_parse_entry(attr, info, false, &addr);
1957         if (ret < 0)
1958                 return ret;
1959
1960         if (attr_rem) {
1961                 ret = mptcp_pm_parse_entry(attr_rem, info, false, &remote);
1962                 if (ret < 0)
1963                         return ret;
1964         }
1965
1966         if (addr.flags & MPTCP_PM_ADDR_FLAG_BACKUP)
1967                 bkup = 1;
1968
1969         return mptcp_pm_set_flags(net, token, &addr, &remote, bkup);
1970 }
1971
1972 static void mptcp_nl_mcast_send(struct net *net, struct sk_buff *nlskb, gfp_t gfp)
1973 {
1974         genlmsg_multicast_netns(&mptcp_genl_family, net,
1975                                 nlskb, 0, MPTCP_PM_EV_GRP_OFFSET, gfp);
1976 }
1977
1978 bool mptcp_userspace_pm_active(const struct mptcp_sock *msk)
1979 {
1980         return genl_has_listeners(&mptcp_genl_family,
1981                                   sock_net((const struct sock *)msk),
1982                                   MPTCP_PM_EV_GRP_OFFSET);
1983 }
1984
1985 static int mptcp_event_add_subflow(struct sk_buff *skb, const struct sock *ssk)
1986 {
1987         const struct inet_sock *issk = inet_sk(ssk);
1988         const struct mptcp_subflow_context *sf;
1989
1990         if (nla_put_u16(skb, MPTCP_ATTR_FAMILY, ssk->sk_family))
1991                 return -EMSGSIZE;
1992
1993         switch (ssk->sk_family) {
1994         case AF_INET:
1995                 if (nla_put_in_addr(skb, MPTCP_ATTR_SADDR4, issk->inet_saddr))
1996                         return -EMSGSIZE;
1997                 if (nla_put_in_addr(skb, MPTCP_ATTR_DADDR4, issk->inet_daddr))
1998                         return -EMSGSIZE;
1999                 break;
2000 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
2001         case AF_INET6: {
2002                 const struct ipv6_pinfo *np = inet6_sk(ssk);
2003
2004                 if (nla_put_in6_addr(skb, MPTCP_ATTR_SADDR6, &np->saddr))
2005                         return -EMSGSIZE;
2006                 if (nla_put_in6_addr(skb, MPTCP_ATTR_DADDR6, &ssk->sk_v6_daddr))
2007                         return -EMSGSIZE;
2008                 break;
2009         }
2010 #endif
2011         default:
2012                 WARN_ON_ONCE(1);
2013                 return -EMSGSIZE;
2014         }
2015
2016         if (nla_put_be16(skb, MPTCP_ATTR_SPORT, issk->inet_sport))
2017                 return -EMSGSIZE;
2018         if (nla_put_be16(skb, MPTCP_ATTR_DPORT, issk->inet_dport))
2019                 return -EMSGSIZE;
2020
2021         sf = mptcp_subflow_ctx(ssk);
2022         if (WARN_ON_ONCE(!sf))
2023                 return -EINVAL;
2024
2025         if (nla_put_u8(skb, MPTCP_ATTR_LOC_ID, subflow_get_local_id(sf)))
2026                 return -EMSGSIZE;
2027
2028         if (nla_put_u8(skb, MPTCP_ATTR_REM_ID, sf->remote_id))
2029                 return -EMSGSIZE;
2030
2031         return 0;
2032 }
2033
2034 static int mptcp_event_put_token_and_ssk(struct sk_buff *skb,
2035                                          const struct mptcp_sock *msk,
2036                                          const struct sock *ssk)
2037 {
2038         const struct sock *sk = (const struct sock *)msk;
2039         const struct mptcp_subflow_context *sf;
2040         u8 sk_err;
2041
2042         if (nla_put_u32(skb, MPTCP_ATTR_TOKEN, msk->token))
2043                 return -EMSGSIZE;
2044
2045         if (mptcp_event_add_subflow(skb, ssk))
2046                 return -EMSGSIZE;
2047
2048         sf = mptcp_subflow_ctx(ssk);
2049         if (WARN_ON_ONCE(!sf))
2050                 return -EINVAL;
2051
2052         if (nla_put_u8(skb, MPTCP_ATTR_BACKUP, sf->backup))
2053                 return -EMSGSIZE;
2054
2055         if (ssk->sk_bound_dev_if &&
2056             nla_put_s32(skb, MPTCP_ATTR_IF_IDX, ssk->sk_bound_dev_if))
2057                 return -EMSGSIZE;
2058
2059         sk_err = READ_ONCE(ssk->sk_err);
2060         if (sk_err && sk->sk_state == TCP_ESTABLISHED &&
2061             nla_put_u8(skb, MPTCP_ATTR_ERROR, sk_err))
2062                 return -EMSGSIZE;
2063
2064         return 0;
2065 }
2066
2067 static int mptcp_event_sub_established(struct sk_buff *skb,
2068                                        const struct mptcp_sock *msk,
2069                                        const struct sock *ssk)
2070 {
2071         return mptcp_event_put_token_and_ssk(skb, msk, ssk);
2072 }
2073
2074 static int mptcp_event_sub_closed(struct sk_buff *skb,
2075                                   const struct mptcp_sock *msk,
2076                                   const struct sock *ssk)
2077 {
2078         const struct mptcp_subflow_context *sf;
2079
2080         if (mptcp_event_put_token_and_ssk(skb, msk, ssk))
2081                 return -EMSGSIZE;
2082
2083         sf = mptcp_subflow_ctx(ssk);
2084         if (!sf->reset_seen)
2085                 return 0;
2086
2087         if (nla_put_u32(skb, MPTCP_ATTR_RESET_REASON, sf->reset_reason))
2088                 return -EMSGSIZE;
2089
2090         if (nla_put_u32(skb, MPTCP_ATTR_RESET_FLAGS, sf->reset_transient))
2091                 return -EMSGSIZE;
2092
2093         return 0;
2094 }
2095
2096 static int mptcp_event_created(struct sk_buff *skb,
2097                                const struct mptcp_sock *msk,
2098                                const struct sock *ssk)
2099 {
2100         int err = nla_put_u32(skb, MPTCP_ATTR_TOKEN, msk->token);
2101
2102         if (err)
2103                 return err;
2104
2105         if (nla_put_u8(skb, MPTCP_ATTR_SERVER_SIDE, READ_ONCE(msk->pm.server_side)))
2106                 return -EMSGSIZE;
2107
2108         return mptcp_event_add_subflow(skb, ssk);
2109 }
2110
2111 void mptcp_event_addr_removed(const struct mptcp_sock *msk, uint8_t id)
2112 {
2113         struct net *net = sock_net((const struct sock *)msk);
2114         struct nlmsghdr *nlh;
2115         struct sk_buff *skb;
2116
2117         if (!genl_has_listeners(&mptcp_genl_family, net, MPTCP_PM_EV_GRP_OFFSET))
2118                 return;
2119
2120         skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_ATOMIC);
2121         if (!skb)
2122                 return;
2123
2124         nlh = genlmsg_put(skb, 0, 0, &mptcp_genl_family, 0, MPTCP_EVENT_REMOVED);
2125         if (!nlh)
2126                 goto nla_put_failure;
2127
2128         if (nla_put_u32(skb, MPTCP_ATTR_TOKEN, msk->token))
2129                 goto nla_put_failure;
2130
2131         if (nla_put_u8(skb, MPTCP_ATTR_REM_ID, id))
2132                 goto nla_put_failure;
2133
2134         genlmsg_end(skb, nlh);
2135         mptcp_nl_mcast_send(net, skb, GFP_ATOMIC);
2136         return;
2137
2138 nla_put_failure:
2139         nlmsg_free(skb);
2140 }
2141
2142 void mptcp_event_addr_announced(const struct sock *ssk,
2143                                 const struct mptcp_addr_info *info)
2144 {
2145         struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk);
2146         struct mptcp_sock *msk = mptcp_sk(subflow->conn);
2147         struct net *net = sock_net(ssk);
2148         struct nlmsghdr *nlh;
2149         struct sk_buff *skb;
2150
2151         if (!genl_has_listeners(&mptcp_genl_family, net, MPTCP_PM_EV_GRP_OFFSET))
2152                 return;
2153
2154         skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_ATOMIC);
2155         if (!skb)
2156                 return;
2157
2158         nlh = genlmsg_put(skb, 0, 0, &mptcp_genl_family, 0,
2159                           MPTCP_EVENT_ANNOUNCED);
2160         if (!nlh)
2161                 goto nla_put_failure;
2162
2163         if (nla_put_u32(skb, MPTCP_ATTR_TOKEN, msk->token))
2164                 goto nla_put_failure;
2165
2166         if (nla_put_u8(skb, MPTCP_ATTR_REM_ID, info->id))
2167                 goto nla_put_failure;
2168
2169         if (nla_put_be16(skb, MPTCP_ATTR_DPORT,
2170                          info->port == 0 ?
2171                          inet_sk(ssk)->inet_dport :
2172                          info->port))
2173                 goto nla_put_failure;
2174
2175         switch (info->family) {
2176         case AF_INET:
2177                 if (nla_put_in_addr(skb, MPTCP_ATTR_DADDR4, info->addr.s_addr))
2178                         goto nla_put_failure;
2179                 break;
2180 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
2181         case AF_INET6:
2182                 if (nla_put_in6_addr(skb, MPTCP_ATTR_DADDR6, &info->addr6))
2183                         goto nla_put_failure;
2184                 break;
2185 #endif
2186         default:
2187                 WARN_ON_ONCE(1);
2188                 goto nla_put_failure;
2189         }
2190
2191         genlmsg_end(skb, nlh);
2192         mptcp_nl_mcast_send(net, skb, GFP_ATOMIC);
2193         return;
2194
2195 nla_put_failure:
2196         nlmsg_free(skb);
2197 }
2198
2199 void mptcp_event_pm_listener(const struct sock *ssk,
2200                              enum mptcp_event_type event)
2201 {
2202         const struct inet_sock *issk = inet_sk(ssk);
2203         struct net *net = sock_net(ssk);
2204         struct nlmsghdr *nlh;
2205         struct sk_buff *skb;
2206
2207         if (!genl_has_listeners(&mptcp_genl_family, net, MPTCP_PM_EV_GRP_OFFSET))
2208                 return;
2209
2210         skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
2211         if (!skb)
2212                 return;
2213
2214         nlh = genlmsg_put(skb, 0, 0, &mptcp_genl_family, 0, event);
2215         if (!nlh)
2216                 goto nla_put_failure;
2217
2218         if (nla_put_u16(skb, MPTCP_ATTR_FAMILY, ssk->sk_family))
2219                 goto nla_put_failure;
2220
2221         if (nla_put_be16(skb, MPTCP_ATTR_SPORT, issk->inet_sport))
2222                 goto nla_put_failure;
2223
2224         switch (ssk->sk_family) {
2225         case AF_INET:
2226                 if (nla_put_in_addr(skb, MPTCP_ATTR_SADDR4, issk->inet_saddr))
2227                         goto nla_put_failure;
2228                 break;
2229 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
2230         case AF_INET6: {
2231                 const struct ipv6_pinfo *np = inet6_sk(ssk);
2232
2233                 if (nla_put_in6_addr(skb, MPTCP_ATTR_SADDR6, &np->saddr))
2234                         goto nla_put_failure;
2235                 break;
2236         }
2237 #endif
2238         default:
2239                 WARN_ON_ONCE(1);
2240                 goto nla_put_failure;
2241         }
2242
2243         genlmsg_end(skb, nlh);
2244         mptcp_nl_mcast_send(net, skb, GFP_KERNEL);
2245         return;
2246
2247 nla_put_failure:
2248         nlmsg_free(skb);
2249 }
2250
2251 void mptcp_event(enum mptcp_event_type type, const struct mptcp_sock *msk,
2252                  const struct sock *ssk, gfp_t gfp)
2253 {
2254         struct net *net = sock_net((const struct sock *)msk);
2255         struct nlmsghdr *nlh;
2256         struct sk_buff *skb;
2257
2258         if (!genl_has_listeners(&mptcp_genl_family, net, MPTCP_PM_EV_GRP_OFFSET))
2259                 return;
2260
2261         skb = nlmsg_new(NLMSG_DEFAULT_SIZE, gfp);
2262         if (!skb)
2263                 return;
2264
2265         nlh = genlmsg_put(skb, 0, 0, &mptcp_genl_family, 0, type);
2266         if (!nlh)
2267                 goto nla_put_failure;
2268
2269         switch (type) {
2270         case MPTCP_EVENT_UNSPEC:
2271                 WARN_ON_ONCE(1);
2272                 break;
2273         case MPTCP_EVENT_CREATED:
2274         case MPTCP_EVENT_ESTABLISHED:
2275                 if (mptcp_event_created(skb, msk, ssk) < 0)
2276                         goto nla_put_failure;
2277                 break;
2278         case MPTCP_EVENT_CLOSED:
2279                 if (nla_put_u32(skb, MPTCP_ATTR_TOKEN, msk->token) < 0)
2280                         goto nla_put_failure;
2281                 break;
2282         case MPTCP_EVENT_ANNOUNCED:
2283         case MPTCP_EVENT_REMOVED:
2284                 /* call mptcp_event_addr_announced()/removed instead */
2285                 WARN_ON_ONCE(1);
2286                 break;
2287         case MPTCP_EVENT_SUB_ESTABLISHED:
2288         case MPTCP_EVENT_SUB_PRIORITY:
2289                 if (mptcp_event_sub_established(skb, msk, ssk) < 0)
2290                         goto nla_put_failure;
2291                 break;
2292         case MPTCP_EVENT_SUB_CLOSED:
2293                 if (mptcp_event_sub_closed(skb, msk, ssk) < 0)
2294                         goto nla_put_failure;
2295                 break;
2296         case MPTCP_EVENT_LISTENER_CREATED:
2297         case MPTCP_EVENT_LISTENER_CLOSED:
2298                 break;
2299         }
2300
2301         genlmsg_end(skb, nlh);
2302         mptcp_nl_mcast_send(net, skb, gfp);
2303         return;
2304
2305 nla_put_failure:
2306         nlmsg_free(skb);
2307 }
2308
2309 static const struct genl_small_ops mptcp_pm_ops[] = {
2310         {
2311                 .cmd    = MPTCP_PM_CMD_ADD_ADDR,
2312                 .doit   = mptcp_nl_cmd_add_addr,
2313                 .flags  = GENL_UNS_ADMIN_PERM,
2314         },
2315         {
2316                 .cmd    = MPTCP_PM_CMD_DEL_ADDR,
2317                 .doit   = mptcp_nl_cmd_del_addr,
2318                 .flags  = GENL_UNS_ADMIN_PERM,
2319         },
2320         {
2321                 .cmd    = MPTCP_PM_CMD_FLUSH_ADDRS,
2322                 .doit   = mptcp_nl_cmd_flush_addrs,
2323                 .flags  = GENL_UNS_ADMIN_PERM,
2324         },
2325         {
2326                 .cmd    = MPTCP_PM_CMD_GET_ADDR,
2327                 .doit   = mptcp_nl_cmd_get_addr,
2328                 .dumpit   = mptcp_nl_cmd_dump_addrs,
2329         },
2330         {
2331                 .cmd    = MPTCP_PM_CMD_SET_LIMITS,
2332                 .doit   = mptcp_nl_cmd_set_limits,
2333                 .flags  = GENL_UNS_ADMIN_PERM,
2334         },
2335         {
2336                 .cmd    = MPTCP_PM_CMD_GET_LIMITS,
2337                 .doit   = mptcp_nl_cmd_get_limits,
2338         },
2339         {
2340                 .cmd    = MPTCP_PM_CMD_SET_FLAGS,
2341                 .doit   = mptcp_nl_cmd_set_flags,
2342                 .flags  = GENL_UNS_ADMIN_PERM,
2343         },
2344         {
2345                 .cmd    = MPTCP_PM_CMD_ANNOUNCE,
2346                 .doit   = mptcp_nl_cmd_announce,
2347                 .flags  = GENL_UNS_ADMIN_PERM,
2348         },
2349         {
2350                 .cmd    = MPTCP_PM_CMD_REMOVE,
2351                 .doit   = mptcp_nl_cmd_remove,
2352                 .flags  = GENL_UNS_ADMIN_PERM,
2353         },
2354         {
2355                 .cmd    = MPTCP_PM_CMD_SUBFLOW_CREATE,
2356                 .doit   = mptcp_nl_cmd_sf_create,
2357                 .flags  = GENL_UNS_ADMIN_PERM,
2358         },
2359         {
2360                 .cmd    = MPTCP_PM_CMD_SUBFLOW_DESTROY,
2361                 .doit   = mptcp_nl_cmd_sf_destroy,
2362                 .flags  = GENL_UNS_ADMIN_PERM,
2363         },
2364 };
2365
2366 static struct genl_family mptcp_genl_family __ro_after_init = {
2367         .name           = MPTCP_PM_NAME,
2368         .version        = MPTCP_PM_VER,
2369         .maxattr        = MPTCP_PM_ATTR_MAX,
2370         .policy         = mptcp_pm_policy,
2371         .netnsok        = true,
2372         .module         = THIS_MODULE,
2373         .small_ops      = mptcp_pm_ops,
2374         .n_small_ops    = ARRAY_SIZE(mptcp_pm_ops),
2375         .resv_start_op  = MPTCP_PM_CMD_SUBFLOW_DESTROY + 1,
2376         .mcgrps         = mptcp_pm_mcgrps,
2377         .n_mcgrps       = ARRAY_SIZE(mptcp_pm_mcgrps),
2378 };
2379
2380 static int __net_init pm_nl_init_net(struct net *net)
2381 {
2382         struct pm_nl_pernet *pernet = pm_nl_get_pernet(net);
2383
2384         INIT_LIST_HEAD_RCU(&pernet->local_addr_list);
2385
2386         /* Cit. 2 subflows ought to be enough for anybody. */
2387         pernet->subflows_max = 2;
2388         pernet->next_id = 1;
2389         pernet->stale_loss_cnt = 4;
2390         spin_lock_init(&pernet->lock);
2391
2392         /* No need to initialize other pernet fields, the struct is zeroed at
2393          * allocation time.
2394          */
2395
2396         return 0;
2397 }
2398
2399 static void __net_exit pm_nl_exit_net(struct list_head *net_list)
2400 {
2401         struct net *net;
2402
2403         list_for_each_entry(net, net_list, exit_list) {
2404                 struct pm_nl_pernet *pernet = pm_nl_get_pernet(net);
2405
2406                 /* net is removed from namespace list, can't race with
2407                  * other modifiers, also netns core already waited for a
2408                  * RCU grace period.
2409                  */
2410                 __flush_addrs(&pernet->local_addr_list);
2411         }
2412 }
2413
2414 static struct pernet_operations mptcp_pm_pernet_ops = {
2415         .init = pm_nl_init_net,
2416         .exit_batch = pm_nl_exit_net,
2417         .id = &pm_nl_pernet_id,
2418         .size = sizeof(struct pm_nl_pernet),
2419 };
2420
2421 void __init mptcp_pm_nl_init(void)
2422 {
2423         if (register_pernet_subsys(&mptcp_pm_pernet_ops) < 0)
2424                 panic("Failed to register MPTCP PM pernet subsystem.\n");
2425
2426         if (genl_register_family(&mptcp_genl_family))
2427                 panic("Failed to register MPTCP PM netlink family\n");
2428 }