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 static bool lookup_address_in_vec(const struct mptcp_addr_info *addrs, unsigned int nr,
 400                                   const struct mptcp_addr_info *addr)
 401 {
 402         int i;
 403
 404         for (i = 0; i < nr; i++) {
 405                 if (addrs[i].id == addr->id)
 406                         return true;
 407         }
 408
 409         return false;
 410 }
 411
 412 /* Fill all the remote addresses into the array addrs[],
 413  * and return the array size.
 414  */
 415 static unsigned int fill_remote_addresses_vec(struct mptcp_sock *msk,
 416                                               struct mptcp_addr_info *local,
 417                                               bool fullmesh,
 418                                               struct mptcp_addr_info *addrs)
 419 {
 420         bool deny_id0 = READ_ONCE(msk->pm.remote_deny_join_id0);
 421         struct sock *sk = (struct sock *)msk, *ssk;
 422         struct mptcp_subflow_context *subflow;
 423         struct mptcp_addr_info remote = { 0 };
 424         unsigned int subflows_max;
 425         int i = 0;
 426
 427         subflows_max = mptcp_pm_get_subflows_max(msk);
 428         remote_address((struct sock_common *)sk, &remote);
 429
 430         /* Non-fullmesh endpoint, fill in the single entry
 431          * corresponding to the primary MPC subflow remote address
 432          */
 433         if (!fullmesh) {
 434                 if (deny_id0)
 435                         return 0;
 436
 437                 if (!mptcp_pm_addr_families_match(sk, local, &remote))
 438                         return 0;
 439
 440                 msk->pm.subflows++;
 441                 addrs[i++] = remote;
 442         } else {
 443                 mptcp_for_each_subflow(msk, subflow) {
 444                         ssk = mptcp_subflow_tcp_sock(subflow);
 445                         remote_address((struct sock_common *)ssk, &addrs[i]);
 446                         addrs[i].id = subflow->remote_id;
 447                         if (deny_id0 && !addrs[i].id)
 448                                 continue;
 449
 450                         if (!mptcp_pm_addr_families_match(sk, local, &addrs[i]))
 451                                 continue;
 452
 453                         if (!lookup_address_in_vec(addrs, i, &addrs[i]) &&
 454                             msk->pm.subflows < subflows_max) {
 455                                 msk->pm.subflows++;
 456                                 i++;
 457                         }
 458                 }
 459         }
 460
 461         return i;
 462 }
 463
 464 static void __mptcp_pm_send_ack(struct mptcp_sock *msk, struct mptcp_subflow_context *subflow,
 465                                 bool prio, bool backup)
 466 {
 467         struct sock *ssk = mptcp_subflow_tcp_sock(subflow);
 468         bool slow;
 469
 470         pr_debug("send ack for %s",
 471                  prio ? "mp_prio" : (mptcp_pm_should_add_signal(msk) ? "add_addr" : "rm_addr"));
 472
 473         slow = lock_sock_fast(ssk);
 474         if (prio) {
 475                 subflow->send_mp_prio = 1;
 476                 subflow->backup = backup;
 477                 subflow->request_bkup = backup;
 478         }
 479
 480         __mptcp_subflow_send_ack(ssk);
 481         unlock_sock_fast(ssk, slow);
 482 }
 483
 484 static void mptcp_pm_send_ack(struct mptcp_sock *msk, struct mptcp_subflow_context *subflow,
 485                               bool prio, bool backup)
 486 {
 487         spin_unlock_bh(&msk->pm.lock);
 488         __mptcp_pm_send_ack(msk, subflow, prio, backup);
 489         spin_lock_bh(&msk->pm.lock);
 490 }
 491
 492 static struct mptcp_pm_addr_entry *
 493 __lookup_addr_by_id(struct pm_nl_pernet *pernet, unsigned int id)
 494 {
 495         struct mptcp_pm_addr_entry *entry;
 496
 497         list_for_each_entry(entry, &pernet->local_addr_list, list) {
 498                 if (entry->addr.id == id)
 499                         return entry;
 500         }
 501         return NULL;
 502 }
 503
 504 static struct mptcp_pm_addr_entry *
 505 __lookup_addr(struct pm_nl_pernet *pernet, const struct mptcp_addr_info *info,
 506               bool lookup_by_id)
 507 {
 508         struct mptcp_pm_addr_entry *entry;
 509
 510         list_for_each_entry(entry, &pernet->local_addr_list, list) {
 511                 if ((!lookup_by_id &&
 512                      mptcp_addresses_equal(&entry->addr, info, entry->addr.port)) ||
 513                     (lookup_by_id && entry->addr.id == info->id))
 514                         return entry;
 515         }
 516         return NULL;
 517 }
 518
 519 static void mptcp_pm_create_subflow_or_signal_addr(struct mptcp_sock *msk)
 520 {
 521         struct sock *sk = (struct sock *)msk;
 522         struct mptcp_pm_addr_entry *local;
 523         unsigned int add_addr_signal_max;
 524         unsigned int local_addr_max;
 525         struct pm_nl_pernet *pernet;
 526         unsigned int subflows_max;
 527
 528         pernet = pm_nl_get_pernet(sock_net(sk));
 529
 530         add_addr_signal_max = mptcp_pm_get_add_addr_signal_max(msk);
 531         local_addr_max = mptcp_pm_get_local_addr_max(msk);
 532         subflows_max = mptcp_pm_get_subflows_max(msk);
 533
 534         /* do lazy endpoint usage accounting for the MPC subflows */
 535         if (unlikely(!(msk->pm.status & BIT(MPTCP_PM_MPC_ENDPOINT_ACCOUNTED))) && msk->first) {
 536                 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(msk->first);
 537                 struct mptcp_pm_addr_entry *entry;
 538                 struct mptcp_addr_info mpc_addr;
 539                 bool backup = false;
 540
 541                 mptcp_local_address((struct sock_common *)msk->first, &mpc_addr);
 542                 rcu_read_lock();
 543                 entry = __lookup_addr(pernet, &mpc_addr, false);
 544                 if (entry) {
 545                         __clear_bit(entry->addr.id, msk->pm.id_avail_bitmap);
 546                         msk->mpc_endpoint_id = entry->addr.id;
 547                         backup = !!(entry->flags & MPTCP_PM_ADDR_FLAG_BACKUP);
 548                 }
 549                 rcu_read_unlock();
 550
 551                 if (backup)
 552                         mptcp_pm_send_ack(msk, subflow, true, backup);
 553
 554                 msk->pm.status |= BIT(MPTCP_PM_MPC_ENDPOINT_ACCOUNTED);
 555         }
 556
 557         pr_debug("local %d:%d signal %d:%d subflows %d:%d\n",
 558                  msk->pm.local_addr_used, local_addr_max,
 559                  msk->pm.add_addr_signaled, add_addr_signal_max,
 560                  msk->pm.subflows, subflows_max);
 561
 562         /* check first for announce */
 563         if (msk->pm.add_addr_signaled < add_addr_signal_max) {
 564                 local = select_signal_address(pernet, msk);
 565
 566                 /* due to racing events on both ends we can reach here while
 567                  * previous add address is still running: if we invoke now
 568                  * mptcp_pm_announce_addr(), that will fail and the
 569                  * corresponding id will be marked as used.
 570                  * Instead let the PM machinery reschedule us when the
 571                  * current address announce will be completed.
 572                  */
 573                 if (msk->pm.addr_signal & BIT(MPTCP_ADD_ADDR_SIGNAL))
 574                         return;
 575
 576                 if (local) {
 577                         if (mptcp_pm_alloc_anno_list(msk, &local->addr)) {
 578                                 __clear_bit(local->addr.id, msk->pm.id_avail_bitmap);
 579                                 msk->pm.add_addr_signaled++;
 580                                 mptcp_pm_announce_addr(msk, &local->addr, false);
 581                                 mptcp_pm_nl_addr_send_ack(msk);
 582                         }
 583                 }
 584         }
 585
 586         /* check if should create a new subflow */
 587         while (msk->pm.local_addr_used < local_addr_max &&
 588                msk->pm.subflows < subflows_max) {
 589                 struct mptcp_addr_info addrs[MPTCP_PM_ADDR_MAX];
 590                 bool fullmesh;
 591                 int i, nr;
 592
 593                 local = select_local_address(pernet, msk);
 594                 if (!local)
 595                         break;
 596
 597                 fullmesh = !!(local->flags & MPTCP_PM_ADDR_FLAG_FULLMESH);
 598
 599                 msk->pm.local_addr_used++;
 600                 __clear_bit(local->addr.id, msk->pm.id_avail_bitmap);
 601                 nr = fill_remote_addresses_vec(msk, &local->addr, fullmesh, addrs);
 602                 if (nr == 0)
 603                         continue;
 604
 605                 spin_unlock_bh(&msk->pm.lock);
 606                 for (i = 0; i < nr; i++)
 607                         __mptcp_subflow_connect(sk, &local->addr, &addrs[i]);
 608                 spin_lock_bh(&msk->pm.lock);
 609         }
 610         mptcp_pm_nl_check_work_pending(msk);
 611 }
 612
 613 static void mptcp_pm_nl_fully_established(struct mptcp_sock *msk)
 614 {
 615         mptcp_pm_create_subflow_or_signal_addr(msk);
 616 }
 617
 618 static void mptcp_pm_nl_subflow_established(struct mptcp_sock *msk)
 619 {
 620         mptcp_pm_create_subflow_or_signal_addr(msk);
 621 }
 622
 623 /* Fill all the local addresses into the array addrs[],
 624  * and return the array size.
 625  */
 626 static unsigned int fill_local_addresses_vec(struct mptcp_sock *msk,
 627                                              struct mptcp_addr_info *remote,
 628                                              struct mptcp_addr_info *addrs)
 629 {
 630         struct sock *sk = (struct sock *)msk;
 631         struct mptcp_pm_addr_entry *entry;
 632         struct pm_nl_pernet *pernet;
 633         unsigned int subflows_max;
 634         int i = 0;
 635
 636         pernet = pm_nl_get_pernet_from_msk(msk);
 637         subflows_max = mptcp_pm_get_subflows_max(msk);
 638
 639         rcu_read_lock();
 640         list_for_each_entry_rcu(entry, &pernet->local_addr_list, list) {
 641                 if (!(entry->flags & MPTCP_PM_ADDR_FLAG_FULLMESH))
 642                         continue;
 643
 644                 if (!mptcp_pm_addr_families_match(sk, &entry->addr, remote))
 645                         continue;
 646
 647                 if (msk->pm.subflows < subflows_max) {
 648                         msk->pm.subflows++;
 649                         addrs[i++] = entry->addr;
 650                 }
 651         }
 652         rcu_read_unlock();
 653
 654         /* If the array is empty, fill in the single
 655          * 'IPADDRANY' local address
 656          */
 657         if (!i) {
 658                 struct mptcp_addr_info local;
 659
 660                 memset(&local, 0, sizeof(local));
 661                 local.family =
 662 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
 663                                remote->family == AF_INET6 &&
 664                                ipv6_addr_v4mapped(&remote->addr6) ? AF_INET :
 665 #endif
 666                                remote->family;
 667
 668                 if (!mptcp_pm_addr_families_match(sk, &local, remote))
 669                         return 0;
 670
 671                 msk->pm.subflows++;
 672                 addrs[i++] = local;
 673         }
 674
 675         return i;
 676 }
 677
 678 static void mptcp_pm_nl_add_addr_received(struct mptcp_sock *msk)
 679 {
 680         struct mptcp_addr_info addrs[MPTCP_PM_ADDR_MAX];
 681         struct sock *sk = (struct sock *)msk;
 682         unsigned int add_addr_accept_max;
 683         struct mptcp_addr_info remote;
 684         unsigned int subflows_max;
 685         int i, nr;
 686
 687         add_addr_accept_max = mptcp_pm_get_add_addr_accept_max(msk);
 688         subflows_max = mptcp_pm_get_subflows_max(msk);
 689
 690         pr_debug("accepted %d:%d remote family %d",
 691                  msk->pm.add_addr_accepted, add_addr_accept_max,
 692                  msk->pm.remote.family);
 693
 694         remote = msk->pm.remote;
 695         mptcp_pm_announce_addr(msk, &remote, true);
 696         mptcp_pm_nl_addr_send_ack(msk);
 697
 698         if (lookup_subflow_by_daddr(&msk->conn_list, &remote))
 699                 return;
 700
 701         /* pick id 0 port, if none is provided the remote address */
 702         if (!remote.port)
 703                 remote.port = sk->sk_dport;
 704
 705         /* connect to the specified remote address, using whatever
 706          * local address the routing configuration will pick.
 707          */
 708         nr = fill_local_addresses_vec(msk, &remote, addrs);
 709         if (nr == 0)
 710                 return;
 711
 712         msk->pm.add_addr_accepted++;
 713         if (msk->pm.add_addr_accepted >= add_addr_accept_max ||
 714             msk->pm.subflows >= subflows_max)
 715                 WRITE_ONCE(msk->pm.accept_addr, false);
 716
 717         spin_unlock_bh(&msk->pm.lock);
 718         for (i = 0; i < nr; i++)
 719                 __mptcp_subflow_connect(sk, &addrs[i], &remote);
 720         spin_lock_bh(&msk->pm.lock);
 721 }
 722
 723 void mptcp_pm_nl_addr_send_ack(struct mptcp_sock *msk)
 724 {
 725         struct mptcp_subflow_context *subflow;
 726
 727         msk_owned_by_me(msk);
 728         lockdep_assert_held(&msk->pm.lock);
 729
 730         if (!mptcp_pm_should_add_signal(msk) &&
 731             !mptcp_pm_should_rm_signal(msk))
 732                 return;
 733
 734         subflow = list_first_entry_or_null(&msk->conn_list, typeof(*subflow), node);
 735         if (subflow)
 736                 mptcp_pm_send_ack(msk, subflow, false, false);
 737 }
 738
 739 int mptcp_pm_nl_mp_prio_send_ack(struct mptcp_sock *msk,
 740                                  struct mptcp_addr_info *addr,
 741                                  struct mptcp_addr_info *rem,
 742                                  u8 bkup)
 743 {
 744         struct mptcp_subflow_context *subflow;
 745
 746         pr_debug("bkup=%d", bkup);
 747
 748         mptcp_for_each_subflow(msk, subflow) {
 749                 struct sock *ssk = mptcp_subflow_tcp_sock(subflow);
 750                 struct mptcp_addr_info local, remote;
 751
 752                 mptcp_local_address((struct sock_common *)ssk, &local);
 753                 if (!mptcp_addresses_equal(&local, addr, addr->port))
 754                         continue;
 755
 756                 if (rem && rem->family != AF_UNSPEC) {
 757                         remote_address((struct sock_common *)ssk, &remote);
 758                         if (!mptcp_addresses_equal(&remote, rem, rem->port))
 759                                 continue;
 760                 }
 761
 762                 __mptcp_pm_send_ack(msk, subflow, true, bkup);
 763                 return 0;
 764         }
 765
 766         return -EINVAL;
 767 }
 768
 769 static bool mptcp_local_id_match(const struct mptcp_sock *msk, u8 local_id, u8 id)
 770 {
 771         return local_id == id || (!local_id && msk->mpc_endpoint_id == id);
 772 }
 773
 774 static void mptcp_pm_nl_rm_addr_or_subflow(struct mptcp_sock *msk,
 775                                            const struct mptcp_rm_list *rm_list,
 776                                            enum linux_mptcp_mib_field rm_type)
 777 {
 778         struct mptcp_subflow_context *subflow, *tmp;
 779         struct sock *sk = (struct sock *)msk;
 780         u8 i;
 781
 782         pr_debug("%s rm_list_nr %d",
 783                  rm_type == MPTCP_MIB_RMADDR ? "address" : "subflow", rm_list->nr);
 784
 785         msk_owned_by_me(msk);
 786
 787         if (sk->sk_state == TCP_LISTEN)
 788                 return;
 789
 790         if (!rm_list->nr)
 791                 return;
 792
 793         if (list_empty(&msk->conn_list))
 794                 return;
 795
 796         for (i = 0; i < rm_list->nr; i++) {
 797                 u8 rm_id = rm_list->ids[i];
 798                 bool removed = false;
 799
 800                 mptcp_for_each_subflow_safe(msk, subflow, tmp) {
 801                         struct sock *ssk = mptcp_subflow_tcp_sock(subflow);
 802                         int how = RCV_SHUTDOWN | SEND_SHUTDOWN;
 803                         u8 id = subflow->local_id;
 804
 805                         if (rm_type == MPTCP_MIB_RMADDR && subflow->remote_id != rm_id)
 806                                 continue;
 807                         if (rm_type == MPTCP_MIB_RMSUBFLOW && !mptcp_local_id_match(msk, id, rm_id))
 808                                 continue;
 809
 810                         pr_debug(" -> %s rm_list_ids[%d]=%u local_id=%u remote_id=%u mpc_id=%u",
 811                                  rm_type == MPTCP_MIB_RMADDR ? "address" : "subflow",
 812                                  i, rm_id, subflow->local_id, subflow->remote_id,
 813                                  msk->mpc_endpoint_id);
 814                         spin_unlock_bh(&msk->pm.lock);
 815                         mptcp_subflow_shutdown(sk, ssk, how);
 816
 817                         /* the following takes care of updating the subflows counter */
 818                         mptcp_close_ssk(sk, ssk, subflow);
 819                         spin_lock_bh(&msk->pm.lock);
 820
 821                         removed = true;
 822                         __MPTCP_INC_STATS(sock_net(sk), rm_type);
 823                 }
 824                 if (rm_type == MPTCP_MIB_RMSUBFLOW)
 825                         __set_bit(rm_id ? rm_id : msk->mpc_endpoint_id, msk->pm.id_avail_bitmap);
 826                 if (!removed)
 827                         continue;
 828
 829                 if (!mptcp_pm_is_kernel(msk))
 830                         continue;
 831
 832                 if (rm_type == MPTCP_MIB_RMADDR) {
 833                         msk->pm.add_addr_accepted--;
 834                         WRITE_ONCE(msk->pm.accept_addr, true);
 835                 } else if (rm_type == MPTCP_MIB_RMSUBFLOW) {
 836                         msk->pm.local_addr_used--;
 837                 }
 838         }
 839 }
 840
 841 static void mptcp_pm_nl_rm_addr_received(struct mptcp_sock *msk)
 842 {
 843         mptcp_pm_nl_rm_addr_or_subflow(msk, &msk->pm.rm_list_rx, MPTCP_MIB_RMADDR);
 844 }
 845
 846 void mptcp_pm_nl_rm_subflow_received(struct mptcp_sock *msk,
 847                                      const struct mptcp_rm_list *rm_list)
 848 {
 849         mptcp_pm_nl_rm_addr_or_subflow(msk, rm_list, MPTCP_MIB_RMSUBFLOW);
 850 }
 851
 852 void mptcp_pm_nl_work(struct mptcp_sock *msk)
 853 {
 854         struct mptcp_pm_data *pm = &msk->pm;
 855
 856         msk_owned_by_me(msk);
 857
 858         if (!(pm->status & MPTCP_PM_WORK_MASK))
 859                 return;
 860
 861         spin_lock_bh(&msk->pm.lock);
 862
 863         pr_debug("msk=%p status=%x", msk, pm->status);
 864         if (pm->status & BIT(MPTCP_PM_ADD_ADDR_RECEIVED)) {
 865                 pm->status &= ~BIT(MPTCP_PM_ADD_ADDR_RECEIVED);
 866                 mptcp_pm_nl_add_addr_received(msk);
 867         }
 868         if (pm->status & BIT(MPTCP_PM_ADD_ADDR_SEND_ACK)) {
 869                 pm->status &= ~BIT(MPTCP_PM_ADD_ADDR_SEND_ACK);
 870                 mptcp_pm_nl_addr_send_ack(msk);
 871         }
 872         if (pm->status & BIT(MPTCP_PM_RM_ADDR_RECEIVED)) {
 873                 pm->status &= ~BIT(MPTCP_PM_RM_ADDR_RECEIVED);
 874                 mptcp_pm_nl_rm_addr_received(msk);
 875         }
 876         if (pm->status & BIT(MPTCP_PM_ESTABLISHED)) {
 877                 pm->status &= ~BIT(MPTCP_PM_ESTABLISHED);
 878                 mptcp_pm_nl_fully_established(msk);
 879         }
 880         if (pm->status & BIT(MPTCP_PM_SUBFLOW_ESTABLISHED)) {
 881                 pm->status &= ~BIT(MPTCP_PM_SUBFLOW_ESTABLISHED);
 882                 mptcp_pm_nl_subflow_established(msk);
 883         }
 884
 885         spin_unlock_bh(&msk->pm.lock);
 886 }
 887
 888 static bool address_use_port(struct mptcp_pm_addr_entry *entry)
 889 {
 890         return (entry->flags &
 891                 (MPTCP_PM_ADDR_FLAG_SIGNAL | MPTCP_PM_ADDR_FLAG_SUBFLOW)) ==
 892                 MPTCP_PM_ADDR_FLAG_SIGNAL;
 893 }
 894
 895 /* caller must ensure the RCU grace period is already elapsed */
 896 static void __mptcp_pm_release_addr_entry(struct mptcp_pm_addr_entry *entry)
 897 {
 898         if (entry->lsk)
 899                 sock_release(entry->lsk);
 900         kfree(entry);
 901 }
 902
 903 static int mptcp_pm_nl_append_new_local_addr(struct pm_nl_pernet *pernet,
 904                                              struct mptcp_pm_addr_entry *entry)
 905 {
 906         struct mptcp_pm_addr_entry *cur, *del_entry = NULL;
 907         unsigned int addr_max;
 908         int ret = -EINVAL;
 909
 910         spin_lock_bh(&pernet->lock);
 911         /* to keep the code simple, don't do IDR-like allocation for address ID,
 912          * just bail when we exceed limits
 913          */
 914         if (pernet->next_id == MPTCP_PM_MAX_ADDR_ID)
 915                 pernet->next_id = 1;
 916         if (pernet->addrs >= MPTCP_PM_ADDR_MAX) {
 917                 ret = -ERANGE;
 918                 goto out;
 919         }
 920         if (test_bit(entry->addr.id, pernet->id_bitmap)) {
 921                 ret = -EBUSY;
 922                 goto out;
 923         }
 924
 925         /* do not insert duplicate address, differentiate on port only
 926          * singled addresses
 927          */
 928         if (!address_use_port(entry))
 929                 entry->addr.port = 0;
 930         list_for_each_entry(cur, &pernet->local_addr_list, list) {
 931                 if (mptcp_addresses_equal(&cur->addr, &entry->addr,
 932                                           cur->addr.port || entry->addr.port)) {
 933                         /* allow replacing the exiting endpoint only if such
 934                          * endpoint is an implicit one and the user-space
 935                          * did not provide an endpoint id
 936                          */
 937                         if (!(cur->flags & MPTCP_PM_ADDR_FLAG_IMPLICIT)) {
 938                                 ret = -EEXIST;
 939                                 goto out;
 940                         }
 941                         if (entry->addr.id)
 942                                 goto out;
 943
 944                         pernet->addrs--;
 945                         entry->addr.id = cur->addr.id;
 946                         list_del_rcu(&cur->list);
 947                         del_entry = cur;
 948                         break;
 949                 }
 950         }
 951
 952         if (!entry->addr.id) {
 953 find_next:
 954                 entry->addr.id = find_next_zero_bit(pernet->id_bitmap,
 955                                                     MPTCP_PM_MAX_ADDR_ID + 1,
 956                                                     pernet->next_id);
 957                 if (!entry->addr.id && pernet->next_id != 1) {
 958                         pernet->next_id = 1;
 959                         goto find_next;
 960                 }
 961         }
 962
 963         if (!entry->addr.id)
 964                 goto out;
 965
 966         __set_bit(entry->addr.id, pernet->id_bitmap);
 967         if (entry->addr.id > pernet->next_id)
 968                 pernet->next_id = entry->addr.id;
 969
 970         if (entry->flags & MPTCP_PM_ADDR_FLAG_SIGNAL) {
 971                 addr_max = pernet->add_addr_signal_max;
 972                 WRITE_ONCE(pernet->add_addr_signal_max, addr_max + 1);
 973         }
 974         if (entry->flags & MPTCP_PM_ADDR_FLAG_SUBFLOW) {
 975                 addr_max = pernet->local_addr_max;
 976                 WRITE_ONCE(pernet->local_addr_max, addr_max + 1);
 977         }
 978
 979         pernet->addrs++;
 980         if (!entry->addr.port)
 981                 list_add_tail_rcu(&entry->list, &pernet->local_addr_list);
 982         else
 983                 list_add_rcu(&entry->list, &pernet->local_addr_list);
 984         ret = entry->addr.id;
 985
 986 out:
 987         spin_unlock_bh(&pernet->lock);
 988
 989         /* just replaced an existing entry, free it */
 990         if (del_entry) {
 991                 synchronize_rcu();
 992                 __mptcp_pm_release_addr_entry(del_entry);
 993         }
 994         return ret;
 995 }
 996
 997 static struct lock_class_key mptcp_slock_keys[2];
 998 static struct lock_class_key mptcp_keys[2];
 999
1000 static int mptcp_pm_nl_create_listen_socket(struct sock *sk,
1001                                             struct mptcp_pm_addr_entry *entry)
1002 {
1003         bool is_ipv6 = sk->sk_family == AF_INET6;
1004         int addrlen = sizeof(struct sockaddr_in);
1005         struct sockaddr_storage addr;
1006         struct sock *newsk, *ssk;
1007         int backlog = 1024;
1008         int err;
1009
1010         err = sock_create_kern(sock_net(sk), entry->addr.family,
1011                                SOCK_STREAM, IPPROTO_MPTCP, &entry->lsk);
1012         if (err)
1013                 return err;
1014
1015         newsk = entry->lsk->sk;
1016         if (!newsk)
1017                 return -EINVAL;
1018
1019         /* The subflow socket lock is acquired in a nested to the msk one
1020          * in several places, even by the TCP stack, and this msk is a kernel
1021          * socket: lockdep complains. Instead of propagating the _nested
1022          * modifiers in several places, re-init the lock class for the msk
1023          * socket to an mptcp specific one.
1024          */
1025         sock_lock_init_class_and_name(newsk,
1026                                       is_ipv6 ? "mlock-AF_INET6" : "mlock-AF_INET",
1027                                       &mptcp_slock_keys[is_ipv6],
1028                                       is_ipv6 ? "msk_lock-AF_INET6" : "msk_lock-AF_INET",
1029                                       &mptcp_keys[is_ipv6]);
1030
1031         lock_sock(newsk);
1032         ssk = __mptcp_nmpc_sk(mptcp_sk(newsk));
1033         release_sock(newsk);
1034         if (IS_ERR(ssk))
1035                 return PTR_ERR(ssk);
1036
1037         mptcp_info2sockaddr(&entry->addr, &addr, entry->addr.family);
1038 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
1039         if (entry->addr.family == AF_INET6)
1040                 addrlen = sizeof(struct sockaddr_in6);
1041 #endif
1042         if (ssk->sk_family == AF_INET)
1043                 err = inet_bind_sk(ssk, (struct sockaddr *)&addr, addrlen);
1044 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
1045         else if (ssk->sk_family == AF_INET6)
1046                 err = inet6_bind_sk(ssk, (struct sockaddr *)&addr, addrlen);
1047 #endif
1048         if (err)
1049                 return err;
1050
1051         inet_sk_state_store(newsk, TCP_LISTEN);
1052         lock_sock(ssk);
1053         err = __inet_listen_sk(ssk, backlog);
1054         if (!err)
1055                 mptcp_event_pm_listener(ssk, MPTCP_EVENT_LISTENER_CREATED);
1056         release_sock(ssk);
1057         return err;
1058 }
1059
1060 int mptcp_pm_nl_get_local_id(struct mptcp_sock *msk, struct mptcp_addr_info *skc)
1061 {
1062         struct mptcp_pm_addr_entry *entry;
1063         struct pm_nl_pernet *pernet;
1064         int ret = -1;
1065
1066         pernet = pm_nl_get_pernet_from_msk(msk);
1067
1068         rcu_read_lock();
1069         list_for_each_entry_rcu(entry, &pernet->local_addr_list, list) {
1070                 if (mptcp_addresses_equal(&entry->addr, skc, entry->addr.port)) {
1071                         ret = entry->addr.id;
1072                         break;
1073                 }
1074         }
1075         rcu_read_unlock();
1076         if (ret >= 0)
1077                 return ret;
1078
1079         /* address not found, add to local list */
1080         entry = kmalloc(sizeof(*entry), GFP_ATOMIC);
1081         if (!entry)
1082                 return -ENOMEM;
1083
1084         entry->addr = *skc;
1085         entry->addr.id = 0;
1086         entry->addr.port = 0;
1087         entry->ifindex = 0;
1088         entry->flags = MPTCP_PM_ADDR_FLAG_IMPLICIT;
1089         entry->lsk = NULL;
1090         ret = mptcp_pm_nl_append_new_local_addr(pernet, entry);
1091         if (ret < 0)
1092                 kfree(entry);
1093
1094         return ret;
1095 }
1096
1097 #define MPTCP_PM_CMD_GRP_OFFSET       0
1098 #define MPTCP_PM_EV_GRP_OFFSET        1
1099
1100 static const struct genl_multicast_group mptcp_pm_mcgrps[] = {
1101         [MPTCP_PM_CMD_GRP_OFFSET]       = { .name = MPTCP_PM_CMD_GRP_NAME, },
1102         [MPTCP_PM_EV_GRP_OFFSET]        = { .name = MPTCP_PM_EV_GRP_NAME,
1103                                             .flags = GENL_UNS_ADMIN_PERM,
1104                                           },
1105 };
1106
1107 void mptcp_pm_nl_subflow_chk_stale(const struct mptcp_sock *msk, struct sock *ssk)
1108 {
1109         struct mptcp_subflow_context *iter, *subflow = mptcp_subflow_ctx(ssk);
1110         struct sock *sk = (struct sock *)msk;
1111         unsigned int active_max_loss_cnt;
1112         struct net *net = sock_net(sk);
1113         unsigned int stale_loss_cnt;
1114         bool slow;
1115
1116         stale_loss_cnt = mptcp_stale_loss_cnt(net);
1117         if (subflow->stale || !stale_loss_cnt || subflow->stale_count <= stale_loss_cnt)
1118                 return;
1119
1120         /* look for another available subflow not in loss state */
1121         active_max_loss_cnt = max_t(int, stale_loss_cnt - 1, 1);
1122         mptcp_for_each_subflow(msk, iter) {
1123                 if (iter != subflow && mptcp_subflow_active(iter) &&
1124                     iter->stale_count < active_max_loss_cnt) {
1125                         /* we have some alternatives, try to mark this subflow as idle ...*/
1126                         slow = lock_sock_fast(ssk);
1127                         if (!tcp_rtx_and_write_queues_empty(ssk)) {
1128                                 subflow->stale = 1;
1129                                 __mptcp_retransmit_pending_data(sk);
1130                                 MPTCP_INC_STATS(net, MPTCP_MIB_SUBFLOWSTALE);
1131                         }
1132                         unlock_sock_fast(ssk, slow);
1133
1134                         /* always try to push the pending data regardless of re-injections:
1135                          * we can possibly use backup subflows now, and subflow selection
1136                          * is cheap under the msk socket lock
1137                          */
1138                         __mptcp_push_pending(sk, 0);
1139                         return;
1140                 }
1141         }
1142 }
1143
1144 static int mptcp_pm_family_to_addr(int family)
1145 {
1146 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
1147         if (family == AF_INET6)
1148                 return MPTCP_PM_ADDR_ATTR_ADDR6;
1149 #endif
1150         return MPTCP_PM_ADDR_ATTR_ADDR4;
1151 }
1152
1153 static int mptcp_pm_parse_pm_addr_attr(struct nlattr *tb[],
1154                                        const struct nlattr *attr,
1155                                        struct genl_info *info,
1156                                        struct mptcp_addr_info *addr,
1157                                        bool require_family)
1158 {
1159         int err, addr_addr;
1160
1161         if (!attr) {
1162                 GENL_SET_ERR_MSG(info, "missing address info");
1163                 return -EINVAL;
1164         }
1165
1166         /* no validation needed - was already done via nested policy */
1167         err = nla_parse_nested_deprecated(tb, MPTCP_PM_ADDR_ATTR_MAX, attr,
1168                                           mptcp_pm_address_nl_policy, info->extack);
1169         if (err)
1170                 return err;
1171
1172         if (tb[MPTCP_PM_ADDR_ATTR_ID])
1173                 addr->id = nla_get_u8(tb[MPTCP_PM_ADDR_ATTR_ID]);
1174
1175         if (!tb[MPTCP_PM_ADDR_ATTR_FAMILY]) {
1176                 if (!require_family)
1177                         return 0;
1178
1179                 NL_SET_ERR_MSG_ATTR(info->extack, attr,
1180                                     "missing family");
1181                 return -EINVAL;
1182         }
1183
1184         addr->family = nla_get_u16(tb[MPTCP_PM_ADDR_ATTR_FAMILY]);
1185         if (addr->family != AF_INET
1186 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
1187             && addr->family != AF_INET6
1188 #endif
1189             ) {
1190                 NL_SET_ERR_MSG_ATTR(info->extack, attr,
1191                                     "unknown address family");
1192                 return -EINVAL;
1193         }
1194         addr_addr = mptcp_pm_family_to_addr(addr->family);
1195         if (!tb[addr_addr]) {
1196                 NL_SET_ERR_MSG_ATTR(info->extack, attr,
1197                                     "missing address data");
1198                 return -EINVAL;
1199         }
1200
1201 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
1202         if (addr->family == AF_INET6)
1203                 addr->addr6 = nla_get_in6_addr(tb[addr_addr]);
1204         else
1205 #endif
1206                 addr->addr.s_addr = nla_get_in_addr(tb[addr_addr]);
1207
1208         if (tb[MPTCP_PM_ADDR_ATTR_PORT])
1209                 addr->port = htons(nla_get_u16(tb[MPTCP_PM_ADDR_ATTR_PORT]));
1210
1211         return 0;
1212 }
1213
1214 int mptcp_pm_parse_addr(struct nlattr *attr, struct genl_info *info,
1215                         struct mptcp_addr_info *addr)
1216 {
1217         struct nlattr *tb[MPTCP_PM_ADDR_ATTR_MAX + 1];
1218
1219         memset(addr, 0, sizeof(*addr));
1220
1221         return mptcp_pm_parse_pm_addr_attr(tb, attr, info, addr, true);
1222 }
1223
1224 int mptcp_pm_parse_entry(struct nlattr *attr, struct genl_info *info,
1225                          bool require_family,
1226                          struct mptcp_pm_addr_entry *entry)
1227 {
1228         struct nlattr *tb[MPTCP_PM_ADDR_ATTR_MAX + 1];
1229         int err;
1230
1231         memset(entry, 0, sizeof(*entry));
1232
1233         err = mptcp_pm_parse_pm_addr_attr(tb, attr, info, &entry->addr, require_family);
1234         if (err)
1235                 return err;
1236
1237         if (tb[MPTCP_PM_ADDR_ATTR_IF_IDX]) {
1238                 u32 val = nla_get_s32(tb[MPTCP_PM_ADDR_ATTR_IF_IDX]);
1239
1240                 entry->ifindex = val;
1241         }
1242
1243         if (tb[MPTCP_PM_ADDR_ATTR_FLAGS])
1244                 entry->flags = nla_get_u32(tb[MPTCP_PM_ADDR_ATTR_FLAGS]);
1245
1246         if (tb[MPTCP_PM_ADDR_ATTR_PORT])
1247                 entry->addr.port = htons(nla_get_u16(tb[MPTCP_PM_ADDR_ATTR_PORT]));
1248
1249         return 0;
1250 }
1251
1252 static struct pm_nl_pernet *genl_info_pm_nl(struct genl_info *info)
1253 {
1254         return pm_nl_get_pernet(genl_info_net(info));
1255 }
1256
1257 static int mptcp_nl_add_subflow_or_signal_addr(struct net *net)
1258 {
1259         struct mptcp_sock *msk;
1260         long s_slot = 0, s_num = 0;
1261
1262         while ((msk = mptcp_token_iter_next(net, &s_slot, &s_num)) != NULL) {
1263                 struct sock *sk = (struct sock *)msk;
1264
1265                 if (!READ_ONCE(msk->fully_established) ||
1266                     mptcp_pm_is_userspace(msk))
1267                         goto next;
1268
1269                 lock_sock(sk);
1270                 spin_lock_bh(&msk->pm.lock);
1271                 mptcp_pm_create_subflow_or_signal_addr(msk);
1272                 spin_unlock_bh(&msk->pm.lock);
1273                 release_sock(sk);
1274
1275 next:
1276                 sock_put(sk);
1277                 cond_resched();
1278         }
1279
1280         return 0;
1281 }
1282
1283 int mptcp_pm_nl_add_addr_doit(struct sk_buff *skb, struct genl_info *info)
1284 {
1285         struct nlattr *attr = info->attrs[MPTCP_PM_ENDPOINT_ADDR];
1286         struct pm_nl_pernet *pernet = genl_info_pm_nl(info);
1287         struct mptcp_pm_addr_entry addr, *entry;
1288         int ret;
1289
1290         ret = mptcp_pm_parse_entry(attr, info, true, &addr);
1291         if (ret < 0)
1292                 return ret;
1293
1294         if (addr.addr.port && !(addr.flags & MPTCP_PM_ADDR_FLAG_SIGNAL)) {
1295                 GENL_SET_ERR_MSG(info, "flags must have signal when using port");
1296                 return -EINVAL;
1297         }
1298
1299         if (addr.flags & MPTCP_PM_ADDR_FLAG_SIGNAL &&
1300             addr.flags & MPTCP_PM_ADDR_FLAG_FULLMESH) {
1301                 GENL_SET_ERR_MSG(info, "flags mustn't have both signal and fullmesh");
1302                 return -EINVAL;
1303         }
1304
1305         if (addr.flags & MPTCP_PM_ADDR_FLAG_IMPLICIT) {
1306                 GENL_SET_ERR_MSG(info, "can't create IMPLICIT endpoint");
1307                 return -EINVAL;
1308         }
1309
1310         entry = kzalloc(sizeof(*entry), GFP_KERNEL_ACCOUNT);
1311         if (!entry) {
1312                 GENL_SET_ERR_MSG(info, "can't allocate addr");
1313                 return -ENOMEM;
1314         }
1315
1316         *entry = addr;
1317         if (entry->addr.port) {
1318                 ret = mptcp_pm_nl_create_listen_socket(skb->sk, entry);
1319                 if (ret) {
1320                         GENL_SET_ERR_MSG_FMT(info, "create listen socket error: %d", ret);
1321                         goto out_free;
1322                 }
1323         }
1324         ret = mptcp_pm_nl_append_new_local_addr(pernet, entry);
1325         if (ret < 0) {
1326                 GENL_SET_ERR_MSG_FMT(info, "too many addresses or duplicate one: %d", ret);
1327                 goto out_free;
1328         }
1329
1330         mptcp_nl_add_subflow_or_signal_addr(sock_net(skb->sk));
1331         return 0;
1332
1333 out_free:
1334         __mptcp_pm_release_addr_entry(entry);
1335         return ret;
1336 }
1337
1338 int mptcp_pm_nl_get_flags_and_ifindex_by_id(struct mptcp_sock *msk, unsigned int id,
1339                                             u8 *flags, int *ifindex)
1340 {
1341         struct mptcp_pm_addr_entry *entry;
1342         struct sock *sk = (struct sock *)msk;
1343         struct net *net = sock_net(sk);
1344
1345         rcu_read_lock();
1346         entry = __lookup_addr_by_id(pm_nl_get_pernet(net), id);
1347         if (entry) {
1348                 *flags = entry->flags;
1349                 *ifindex = entry->ifindex;
1350         }
1351         rcu_read_unlock();
1352
1353         return 0;
1354 }
1355
1356 static bool remove_anno_list_by_saddr(struct mptcp_sock *msk,
1357                                       const struct mptcp_addr_info *addr)
1358 {
1359         struct mptcp_pm_add_entry *entry;
1360
1361         entry = mptcp_pm_del_add_timer(msk, addr, false);
1362         if (entry) {
1363                 list_del(&entry->list);
1364                 kfree(entry);
1365                 return true;
1366         }
1367
1368         return false;
1369 }
1370
1371 static bool mptcp_pm_remove_anno_addr(struct mptcp_sock *msk,
1372                                       const struct mptcp_addr_info *addr,
1373                                       bool force)
1374 {
1375         struct mptcp_rm_list list = { .nr = 0 };
1376         bool ret;
1377
1378         list.ids[list.nr++] = addr->id;
1379
1380         ret = remove_anno_list_by_saddr(msk, addr);
1381         if (ret || force) {
1382                 spin_lock_bh(&msk->pm.lock);
1383                 mptcp_pm_remove_addr(msk, &list);
1384                 spin_unlock_bh(&msk->pm.lock);
1385         }
1386         return ret;
1387 }
1388
1389 static int mptcp_nl_remove_subflow_and_signal_addr(struct net *net,
1390                                                    const struct mptcp_pm_addr_entry *entry)
1391 {
1392         const struct mptcp_addr_info *addr = &entry->addr;
1393         struct mptcp_rm_list list = { .nr = 0 };
1394         long s_slot = 0, s_num = 0;
1395         struct mptcp_sock *msk;
1396
1397         pr_debug("remove_id=%d", addr->id);
1398
1399         list.ids[list.nr++] = addr->id;
1400
1401         while ((msk = mptcp_token_iter_next(net, &s_slot, &s_num)) != NULL) {
1402                 struct sock *sk = (struct sock *)msk;
1403                 bool remove_subflow;
1404
1405                 if (mptcp_pm_is_userspace(msk))
1406                         goto next;
1407
1408                 if (list_empty(&msk->conn_list)) {
1409                         mptcp_pm_remove_anno_addr(msk, addr, false);
1410                         goto next;
1411                 }
1412
1413                 lock_sock(sk);
1414                 remove_subflow = lookup_subflow_by_saddr(&msk->conn_list, addr);
1415                 mptcp_pm_remove_anno_addr(msk, addr, remove_subflow &&
1416                                           !(entry->flags & MPTCP_PM_ADDR_FLAG_IMPLICIT));
1417                 if (remove_subflow)
1418                         mptcp_pm_remove_subflow(msk, &list);
1419                 release_sock(sk);
1420
1421 next:
1422                 sock_put(sk);
1423                 cond_resched();
1424         }
1425
1426         return 0;
1427 }
1428
1429 static int mptcp_nl_remove_id_zero_address(struct net *net,
1430                                            struct mptcp_addr_info *addr)
1431 {
1432         struct mptcp_rm_list list = { .nr = 0 };
1433         long s_slot = 0, s_num = 0;
1434         struct mptcp_sock *msk;
1435
1436         list.ids[list.nr++] = 0;
1437
1438         while ((msk = mptcp_token_iter_next(net, &s_slot, &s_num)) != NULL) {
1439                 struct sock *sk = (struct sock *)msk;
1440                 struct mptcp_addr_info msk_local;
1441
1442                 if (list_empty(&msk->conn_list) || mptcp_pm_is_userspace(msk))
1443                         goto next;
1444
1445                 mptcp_local_address((struct sock_common *)msk, &msk_local);
1446                 if (!mptcp_addresses_equal(&msk_local, addr, addr->port))
1447                         goto next;
1448
1449                 lock_sock(sk);
1450                 spin_lock_bh(&msk->pm.lock);
1451                 mptcp_pm_remove_addr(msk, &list);
1452                 mptcp_pm_nl_rm_subflow_received(msk, &list);
1453                 spin_unlock_bh(&msk->pm.lock);
1454                 release_sock(sk);
1455
1456 next:
1457                 sock_put(sk);
1458                 cond_resched();
1459         }
1460
1461         return 0;
1462 }
1463
1464 int mptcp_pm_nl_del_addr_doit(struct sk_buff *skb, struct genl_info *info)
1465 {
1466         struct nlattr *attr = info->attrs[MPTCP_PM_ENDPOINT_ADDR];
1467         struct pm_nl_pernet *pernet = genl_info_pm_nl(info);
1468         struct mptcp_pm_addr_entry addr, *entry;
1469         unsigned int addr_max;
1470         int ret;
1471
1472         ret = mptcp_pm_parse_entry(attr, info, false, &addr);
1473         if (ret < 0)
1474                 return ret;
1475
1476         /* the zero id address is special: the first address used by the msk
1477          * always gets such an id, so different subflows can have different zero
1478          * id addresses. Additionally zero id is not accounted for in id_bitmap.
1479          * Let's use an 'mptcp_rm_list' instead of the common remove code.
1480          */
1481         if (addr.addr.id == 0)
1482                 return mptcp_nl_remove_id_zero_address(sock_net(skb->sk), &addr.addr);
1483
1484         spin_lock_bh(&pernet->lock);
1485         entry = __lookup_addr_by_id(pernet, addr.addr.id);
1486         if (!entry) {
1487                 GENL_SET_ERR_MSG(info, "address not found");
1488                 spin_unlock_bh(&pernet->lock);
1489                 return -EINVAL;
1490         }
1491         if (entry->flags & MPTCP_PM_ADDR_FLAG_SIGNAL) {
1492                 addr_max = pernet->add_addr_signal_max;
1493                 WRITE_ONCE(pernet->add_addr_signal_max, addr_max - 1);
1494         }
1495         if (entry->flags & MPTCP_PM_ADDR_FLAG_SUBFLOW) {
1496                 addr_max = pernet->local_addr_max;
1497                 WRITE_ONCE(pernet->local_addr_max, addr_max - 1);
1498         }
1499
1500         pernet->addrs--;
1501         list_del_rcu(&entry->list);
1502         __clear_bit(entry->addr.id, pernet->id_bitmap);
1503         spin_unlock_bh(&pernet->lock);
1504
1505         mptcp_nl_remove_subflow_and_signal_addr(sock_net(skb->sk), entry);
1506         synchronize_rcu();
1507         __mptcp_pm_release_addr_entry(entry);
1508
1509         return ret;
1510 }
1511
1512 void mptcp_pm_remove_addrs(struct mptcp_sock *msk, struct list_head *rm_list)
1513 {
1514         struct mptcp_rm_list alist = { .nr = 0 };
1515         struct mptcp_pm_addr_entry *entry;
1516
1517         list_for_each_entry(entry, rm_list, list) {
1518                 if ((remove_anno_list_by_saddr(msk, &entry->addr) ||
1519                      lookup_subflow_by_saddr(&msk->conn_list, &entry->addr)) &&
1520                     alist.nr < MPTCP_RM_IDS_MAX)
1521                         alist.ids[alist.nr++] = entry->addr.id;
1522         }
1523
1524         if (alist.nr) {
1525                 spin_lock_bh(&msk->pm.lock);
1526                 mptcp_pm_remove_addr(msk, &alist);
1527                 spin_unlock_bh(&msk->pm.lock);
1528         }
1529 }
1530
1531 void mptcp_pm_remove_addrs_and_subflows(struct mptcp_sock *msk,
1532                                         struct list_head *rm_list)
1533 {
1534         struct mptcp_rm_list alist = { .nr = 0 }, slist = { .nr = 0 };
1535         struct mptcp_pm_addr_entry *entry;
1536
1537         list_for_each_entry(entry, rm_list, list) {
1538                 if (lookup_subflow_by_saddr(&msk->conn_list, &entry->addr) &&
1539                     slist.nr < MPTCP_RM_IDS_MAX)
1540                         slist.ids[slist.nr++] = entry->addr.id;
1541
1542                 if (remove_anno_list_by_saddr(msk, &entry->addr) &&
1543                     alist.nr < MPTCP_RM_IDS_MAX)
1544                         alist.ids[alist.nr++] = entry->addr.id;
1545         }
1546
1547         if (alist.nr) {
1548                 spin_lock_bh(&msk->pm.lock);
1549                 mptcp_pm_remove_addr(msk, &alist);
1550                 spin_unlock_bh(&msk->pm.lock);
1551         }
1552         if (slist.nr)
1553                 mptcp_pm_remove_subflow(msk, &slist);
1554 }
1555
1556 static void mptcp_nl_remove_addrs_list(struct net *net,
1557                                        struct list_head *rm_list)
1558 {
1559         long s_slot = 0, s_num = 0;
1560         struct mptcp_sock *msk;
1561
1562         if (list_empty(rm_list))
1563                 return;
1564
1565         while ((msk = mptcp_token_iter_next(net, &s_slot, &s_num)) != NULL) {
1566                 struct sock *sk = (struct sock *)msk;
1567
1568                 if (!mptcp_pm_is_userspace(msk)) {
1569                         lock_sock(sk);
1570                         mptcp_pm_remove_addrs_and_subflows(msk, rm_list);
1571                         release_sock(sk);
1572                 }
1573
1574                 sock_put(sk);
1575                 cond_resched();
1576         }
1577 }
1578
1579 /* caller must ensure the RCU grace period is already elapsed */
1580 static void __flush_addrs(struct list_head *list)
1581 {
1582         while (!list_empty(list)) {
1583                 struct mptcp_pm_addr_entry *cur;
1584
1585                 cur = list_entry(list->next,
1586                                  struct mptcp_pm_addr_entry, list);
1587                 list_del_rcu(&cur->list);
1588                 __mptcp_pm_release_addr_entry(cur);
1589         }
1590 }
1591
1592 static void __reset_counters(struct pm_nl_pernet *pernet)
1593 {
1594         WRITE_ONCE(pernet->add_addr_signal_max, 0);
1595         WRITE_ONCE(pernet->add_addr_accept_max, 0);
1596         WRITE_ONCE(pernet->local_addr_max, 0);
1597         pernet->addrs = 0;
1598 }
1599
1600 int mptcp_pm_nl_flush_addrs_doit(struct sk_buff *skb, struct genl_info *info)
1601 {
1602         struct pm_nl_pernet *pernet = genl_info_pm_nl(info);
1603         LIST_HEAD(free_list);
1604
1605         spin_lock_bh(&pernet->lock);
1606         list_splice_init(&pernet->local_addr_list, &free_list);
1607         __reset_counters(pernet);
1608         pernet->next_id = 1;
1609         bitmap_zero(pernet->id_bitmap, MPTCP_PM_MAX_ADDR_ID + 1);
1610         spin_unlock_bh(&pernet->lock);
1611         mptcp_nl_remove_addrs_list(sock_net(skb->sk), &free_list);
1612         synchronize_rcu();
1613         __flush_addrs(&free_list);
1614         return 0;
1615 }
1616
1617 static int mptcp_nl_fill_addr(struct sk_buff *skb,
1618                               struct mptcp_pm_addr_entry *entry)
1619 {
1620         struct mptcp_addr_info *addr = &entry->addr;
1621         struct nlattr *attr;
1622
1623         attr = nla_nest_start(skb, MPTCP_PM_ATTR_ADDR);
1624         if (!attr)
1625                 return -EMSGSIZE;
1626
1627         if (nla_put_u16(skb, MPTCP_PM_ADDR_ATTR_FAMILY, addr->family))
1628                 goto nla_put_failure;
1629         if (nla_put_u16(skb, MPTCP_PM_ADDR_ATTR_PORT, ntohs(addr->port)))
1630                 goto nla_put_failure;
1631         if (nla_put_u8(skb, MPTCP_PM_ADDR_ATTR_ID, addr->id))
1632                 goto nla_put_failure;
1633         if (nla_put_u32(skb, MPTCP_PM_ADDR_ATTR_FLAGS, entry->flags))
1634                 goto nla_put_failure;
1635         if (entry->ifindex &&
1636             nla_put_s32(skb, MPTCP_PM_ADDR_ATTR_IF_IDX, entry->ifindex))
1637                 goto nla_put_failure;
1638
1639         if (addr->family == AF_INET &&
1640             nla_put_in_addr(skb, MPTCP_PM_ADDR_ATTR_ADDR4,
1641                             addr->addr.s_addr))
1642                 goto nla_put_failure;
1643 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
1644         else if (addr->family == AF_INET6 &&
1645                  nla_put_in6_addr(skb, MPTCP_PM_ADDR_ATTR_ADDR6, &addr->addr6))
1646                 goto nla_put_failure;
1647 #endif
1648         nla_nest_end(skb, attr);
1649         return 0;
1650
1651 nla_put_failure:
1652         nla_nest_cancel(skb, attr);
1653         return -EMSGSIZE;
1654 }
1655
1656 int mptcp_pm_nl_get_addr_doit(struct sk_buff *skb, struct genl_info *info)
1657 {
1658         struct nlattr *attr = info->attrs[MPTCP_PM_ENDPOINT_ADDR];
1659         struct pm_nl_pernet *pernet = genl_info_pm_nl(info);
1660         struct mptcp_pm_addr_entry addr, *entry;
1661         struct sk_buff *msg;
1662         void *reply;
1663         int ret;
1664
1665         ret = mptcp_pm_parse_entry(attr, info, false, &addr);
1666         if (ret < 0)
1667                 return ret;
1668
1669         msg = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
1670         if (!msg)
1671                 return -ENOMEM;
1672
1673         reply = genlmsg_put_reply(msg, info, &mptcp_genl_family, 0,
1674                                   info->genlhdr->cmd);
1675         if (!reply) {
1676                 GENL_SET_ERR_MSG(info, "not enough space in Netlink message");
1677                 ret = -EMSGSIZE;
1678                 goto fail;
1679         }
1680
1681         spin_lock_bh(&pernet->lock);
1682         entry = __lookup_addr_by_id(pernet, addr.addr.id);
1683         if (!entry) {
1684                 GENL_SET_ERR_MSG(info, "address not found");
1685                 ret = -EINVAL;
1686                 goto unlock_fail;
1687         }
1688
1689         ret = mptcp_nl_fill_addr(msg, entry);
1690         if (ret)
1691                 goto unlock_fail;
1692
1693         genlmsg_end(msg, reply);
1694         ret = genlmsg_reply(msg, info);
1695         spin_unlock_bh(&pernet->lock);
1696         return ret;
1697
1698 unlock_fail:
1699         spin_unlock_bh(&pernet->lock);
1700
1701 fail:
1702         nlmsg_free(msg);
1703         return ret;
1704 }
1705
1706 int mptcp_pm_nl_get_addr_dumpit(struct sk_buff *msg,
1707                                 struct netlink_callback *cb)
1708 {
1709         struct net *net = sock_net(msg->sk);
1710         struct mptcp_pm_addr_entry *entry;
1711         struct pm_nl_pernet *pernet;
1712         int id = cb->args[0];
1713         void *hdr;
1714         int i;
1715
1716         pernet = pm_nl_get_pernet(net);
1717
1718         spin_lock_bh(&pernet->lock);
1719         for (i = id; i < MPTCP_PM_MAX_ADDR_ID + 1; i++) {
1720                 if (test_bit(i, pernet->id_bitmap)) {
1721                         entry = __lookup_addr_by_id(pernet, i);
1722                         if (!entry)
1723                                 break;
1724
1725                         if (entry->addr.id <= id)
1726                                 continue;
1727
1728                         hdr = genlmsg_put(msg, NETLINK_CB(cb->skb).portid,
1729                                           cb->nlh->nlmsg_seq, &mptcp_genl_family,
1730                                           NLM_F_MULTI, MPTCP_PM_CMD_GET_ADDR);
1731                         if (!hdr)
1732                                 break;
1733
1734                         if (mptcp_nl_fill_addr(msg, entry) < 0) {
1735                                 genlmsg_cancel(msg, hdr);
1736                                 break;
1737                         }
1738
1739                         id = entry->addr.id;
1740                         genlmsg_end(msg, hdr);
1741                 }
1742         }
1743         spin_unlock_bh(&pernet->lock);
1744
1745         cb->args[0] = id;
1746         return msg->len;
1747 }
1748
1749 static int parse_limit(struct genl_info *info, int id, unsigned int *limit)
1750 {
1751         struct nlattr *attr = info->attrs[id];
1752
1753         if (!attr)
1754                 return 0;
1755
1756         *limit = nla_get_u32(attr);
1757         if (*limit > MPTCP_PM_ADDR_MAX) {
1758                 GENL_SET_ERR_MSG(info, "limit greater than maximum");
1759                 return -EINVAL;
1760         }
1761         return 0;
1762 }
1763
1764 int mptcp_pm_nl_set_limits_doit(struct sk_buff *skb, struct genl_info *info)
1765 {
1766         struct pm_nl_pernet *pernet = genl_info_pm_nl(info);
1767         unsigned int rcv_addrs, subflows;
1768         int ret;
1769
1770         spin_lock_bh(&pernet->lock);
1771         rcv_addrs = pernet->add_addr_accept_max;
1772         ret = parse_limit(info, MPTCP_PM_ATTR_RCV_ADD_ADDRS, &rcv_addrs);
1773         if (ret)
1774                 goto unlock;
1775
1776         subflows = pernet->subflows_max;
1777         ret = parse_limit(info, MPTCP_PM_ATTR_SUBFLOWS, &subflows);
1778         if (ret)
1779                 goto unlock;
1780
1781         WRITE_ONCE(pernet->add_addr_accept_max, rcv_addrs);
1782         WRITE_ONCE(pernet->subflows_max, subflows);
1783
1784 unlock:
1785         spin_unlock_bh(&pernet->lock);
1786         return ret;
1787 }
1788
1789 int mptcp_pm_nl_get_limits_doit(struct sk_buff *skb, struct genl_info *info)
1790 {
1791         struct pm_nl_pernet *pernet = genl_info_pm_nl(info);
1792         struct sk_buff *msg;
1793         void *reply;
1794
1795         msg = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
1796         if (!msg)
1797                 return -ENOMEM;
1798
1799         reply = genlmsg_put_reply(msg, info, &mptcp_genl_family, 0,
1800                                   MPTCP_PM_CMD_GET_LIMITS);
1801         if (!reply)
1802                 goto fail;
1803
1804         if (nla_put_u32(msg, MPTCP_PM_ATTR_RCV_ADD_ADDRS,
1805                         READ_ONCE(pernet->add_addr_accept_max)))
1806                 goto fail;
1807
1808         if (nla_put_u32(msg, MPTCP_PM_ATTR_SUBFLOWS,
1809                         READ_ONCE(pernet->subflows_max)))
1810                 goto fail;
1811
1812         genlmsg_end(msg, reply);
1813         return genlmsg_reply(msg, info);
1814
1815 fail:
1816         GENL_SET_ERR_MSG(info, "not enough space in Netlink message");
1817         nlmsg_free(msg);
1818         return -EMSGSIZE;
1819 }
1820
1821 static void mptcp_pm_nl_fullmesh(struct mptcp_sock *msk,
1822                                  struct mptcp_addr_info *addr)
1823 {
1824         struct mptcp_rm_list list = { .nr = 0 };
1825
1826         list.ids[list.nr++] = addr->id;
1827
1828         spin_lock_bh(&msk->pm.lock);
1829         mptcp_pm_nl_rm_subflow_received(msk, &list);
1830         mptcp_pm_create_subflow_or_signal_addr(msk);
1831         spin_unlock_bh(&msk->pm.lock);
1832 }
1833
1834 static int mptcp_nl_set_flags(struct net *net,
1835                               struct mptcp_addr_info *addr,
1836                               u8 bkup, u8 changed)
1837 {
1838         long s_slot = 0, s_num = 0;
1839         struct mptcp_sock *msk;
1840         int ret = -EINVAL;
1841
1842         while ((msk = mptcp_token_iter_next(net, &s_slot, &s_num)) != NULL) {
1843                 struct sock *sk = (struct sock *)msk;
1844
1845                 if (list_empty(&msk->conn_list) || mptcp_pm_is_userspace(msk))
1846                         goto next;
1847
1848                 lock_sock(sk);
1849                 if (changed & MPTCP_PM_ADDR_FLAG_BACKUP)
1850                         ret = mptcp_pm_nl_mp_prio_send_ack(msk, addr, NULL, bkup);
1851                 if (changed & MPTCP_PM_ADDR_FLAG_FULLMESH)
1852                         mptcp_pm_nl_fullmesh(msk, addr);
1853                 release_sock(sk);
1854
1855 next:
1856                 sock_put(sk);
1857                 cond_resched();
1858         }
1859
1860         return ret;
1861 }
1862
1863 int mptcp_pm_nl_set_flags(struct net *net, struct mptcp_pm_addr_entry *addr, u8 bkup)
1864 {
1865         struct pm_nl_pernet *pernet = pm_nl_get_pernet(net);
1866         u8 changed, mask = MPTCP_PM_ADDR_FLAG_BACKUP |
1867                            MPTCP_PM_ADDR_FLAG_FULLMESH;
1868         struct mptcp_pm_addr_entry *entry;
1869         u8 lookup_by_id = 0;
1870
1871         if (addr->addr.family == AF_UNSPEC) {
1872                 lookup_by_id = 1;
1873                 if (!addr->addr.id)
1874                         return -EOPNOTSUPP;
1875         }
1876
1877         spin_lock_bh(&pernet->lock);
1878         entry = __lookup_addr(pernet, &addr->addr, lookup_by_id);
1879         if (!entry) {
1880                 spin_unlock_bh(&pernet->lock);
1881                 return -EINVAL;
1882         }
1883         if ((addr->flags & MPTCP_PM_ADDR_FLAG_FULLMESH) &&
1884             (entry->flags & MPTCP_PM_ADDR_FLAG_SIGNAL)) {
1885                 spin_unlock_bh(&pernet->lock);
1886                 return -EINVAL;
1887         }
1888
1889         changed = (addr->flags ^ entry->flags) & mask;
1890         entry->flags = (entry->flags & ~mask) | (addr->flags & mask);
1891         *addr = *entry;
1892         spin_unlock_bh(&pernet->lock);
1893
1894         mptcp_nl_set_flags(net, &addr->addr, bkup, changed);
1895         return 0;
1896 }
1897
1898 int mptcp_pm_nl_set_flags_doit(struct sk_buff *skb, struct genl_info *info)
1899 {
1900         struct mptcp_pm_addr_entry remote = { .addr = { .family = AF_UNSPEC }, };
1901         struct mptcp_pm_addr_entry addr = { .addr = { .family = AF_UNSPEC }, };
1902         struct nlattr *attr_rem = info->attrs[MPTCP_PM_ATTR_ADDR_REMOTE];
1903         struct nlattr *token = info->attrs[MPTCP_PM_ATTR_TOKEN];
1904         struct nlattr *attr = info->attrs[MPTCP_PM_ATTR_ADDR];
1905         struct net *net = sock_net(skb->sk);
1906         u8 bkup = 0;
1907         int ret;
1908
1909         ret = mptcp_pm_parse_entry(attr, info, false, &addr);
1910         if (ret < 0)
1911                 return ret;
1912
1913         if (attr_rem) {
1914                 ret = mptcp_pm_parse_entry(attr_rem, info, false, &remote);
1915                 if (ret < 0)
1916                         return ret;
1917         }
1918
1919         if (addr.flags & MPTCP_PM_ADDR_FLAG_BACKUP)
1920                 bkup = 1;
1921
1922         return mptcp_pm_set_flags(net, token, &addr, &remote, bkup);
1923 }
1924
1925 static void mptcp_nl_mcast_send(struct net *net, struct sk_buff *nlskb, gfp_t gfp)
1926 {
1927         genlmsg_multicast_netns(&mptcp_genl_family, net,
1928                                 nlskb, 0, MPTCP_PM_EV_GRP_OFFSET, gfp);
1929 }
1930
1931 bool mptcp_userspace_pm_active(const struct mptcp_sock *msk)
1932 {
1933         return genl_has_listeners(&mptcp_genl_family,
1934                                   sock_net((const struct sock *)msk),
1935                                   MPTCP_PM_EV_GRP_OFFSET);
1936 }
1937
1938 static int mptcp_event_add_subflow(struct sk_buff *skb, const struct sock *ssk)
1939 {
1940         const struct inet_sock *issk = inet_sk(ssk);
1941         const struct mptcp_subflow_context *sf;
1942
1943         if (nla_put_u16(skb, MPTCP_ATTR_FAMILY, ssk->sk_family))
1944                 return -EMSGSIZE;
1945
1946         switch (ssk->sk_family) {
1947         case AF_INET:
1948                 if (nla_put_in_addr(skb, MPTCP_ATTR_SADDR4, issk->inet_saddr))
1949                         return -EMSGSIZE;
1950                 if (nla_put_in_addr(skb, MPTCP_ATTR_DADDR4, issk->inet_daddr))
1951                         return -EMSGSIZE;
1952                 break;
1953 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
1954         case AF_INET6: {
1955                 const struct ipv6_pinfo *np = inet6_sk(ssk);
1956
1957                 if (nla_put_in6_addr(skb, MPTCP_ATTR_SADDR6, &np->saddr))
1958                         return -EMSGSIZE;
1959                 if (nla_put_in6_addr(skb, MPTCP_ATTR_DADDR6, &ssk->sk_v6_daddr))
1960                         return -EMSGSIZE;
1961                 break;
1962         }
1963 #endif
1964         default:
1965                 WARN_ON_ONCE(1);
1966                 return -EMSGSIZE;
1967         }
1968
1969         if (nla_put_be16(skb, MPTCP_ATTR_SPORT, issk->inet_sport))
1970                 return -EMSGSIZE;
1971         if (nla_put_be16(skb, MPTCP_ATTR_DPORT, issk->inet_dport))
1972                 return -EMSGSIZE;
1973
1974         sf = mptcp_subflow_ctx(ssk);
1975         if (WARN_ON_ONCE(!sf))
1976                 return -EINVAL;
1977
1978         if (nla_put_u8(skb, MPTCP_ATTR_LOC_ID, sf->local_id))
1979                 return -EMSGSIZE;
1980
1981         if (nla_put_u8(skb, MPTCP_ATTR_REM_ID, sf->remote_id))
1982                 return -EMSGSIZE;
1983
1984         return 0;
1985 }
1986
1987 static int mptcp_event_put_token_and_ssk(struct sk_buff *skb,
1988                                          const struct mptcp_sock *msk,
1989                                          const struct sock *ssk)
1990 {
1991         const struct sock *sk = (const struct sock *)msk;
1992         const struct mptcp_subflow_context *sf;
1993         u8 sk_err;
1994
1995         if (nla_put_u32(skb, MPTCP_ATTR_TOKEN, msk->token))
1996                 return -EMSGSIZE;
1997
1998         if (mptcp_event_add_subflow(skb, ssk))
1999                 return -EMSGSIZE;
2000
2001         sf = mptcp_subflow_ctx(ssk);
2002         if (WARN_ON_ONCE(!sf))
2003                 return -EINVAL;
2004
2005         if (nla_put_u8(skb, MPTCP_ATTR_BACKUP, sf->backup))
2006                 return -EMSGSIZE;
2007
2008         if (ssk->sk_bound_dev_if &&
2009             nla_put_s32(skb, MPTCP_ATTR_IF_IDX, ssk->sk_bound_dev_if))
2010                 return -EMSGSIZE;
2011
2012         sk_err = READ_ONCE(ssk->sk_err);
2013         if (sk_err && sk->sk_state == TCP_ESTABLISHED &&
2014             nla_put_u8(skb, MPTCP_ATTR_ERROR, sk_err))
2015                 return -EMSGSIZE;
2016
2017         return 0;
2018 }
2019
2020 static int mptcp_event_sub_established(struct sk_buff *skb,
2021                                        const struct mptcp_sock *msk,
2022                                        const struct sock *ssk)
2023 {
2024         return mptcp_event_put_token_and_ssk(skb, msk, ssk);
2025 }
2026
2027 static int mptcp_event_sub_closed(struct sk_buff *skb,
2028                                   const struct mptcp_sock *msk,
2029                                   const struct sock *ssk)
2030 {
2031         const struct mptcp_subflow_context *sf;
2032
2033         if (mptcp_event_put_token_and_ssk(skb, msk, ssk))
2034                 return -EMSGSIZE;
2035
2036         sf = mptcp_subflow_ctx(ssk);
2037         if (!sf->reset_seen)
2038                 return 0;
2039
2040         if (nla_put_u32(skb, MPTCP_ATTR_RESET_REASON, sf->reset_reason))
2041                 return -EMSGSIZE;
2042
2043         if (nla_put_u32(skb, MPTCP_ATTR_RESET_FLAGS, sf->reset_transient))
2044                 return -EMSGSIZE;
2045
2046         return 0;
2047 }
2048
2049 static int mptcp_event_created(struct sk_buff *skb,
2050                                const struct mptcp_sock *msk,
2051                                const struct sock *ssk)
2052 {
2053         int err = nla_put_u32(skb, MPTCP_ATTR_TOKEN, msk->token);
2054
2055         if (err)
2056                 return err;
2057
2058         if (nla_put_u8(skb, MPTCP_ATTR_SERVER_SIDE, READ_ONCE(msk->pm.server_side)))
2059                 return -EMSGSIZE;
2060
2061         return mptcp_event_add_subflow(skb, ssk);
2062 }
2063
2064 void mptcp_event_addr_removed(const struct mptcp_sock *msk, uint8_t id)
2065 {
2066         struct net *net = sock_net((const struct sock *)msk);
2067         struct nlmsghdr *nlh;
2068         struct sk_buff *skb;
2069
2070         if (!genl_has_listeners(&mptcp_genl_family, net, MPTCP_PM_EV_GRP_OFFSET))
2071                 return;
2072
2073         skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_ATOMIC);
2074         if (!skb)
2075                 return;
2076
2077         nlh = genlmsg_put(skb, 0, 0, &mptcp_genl_family, 0, MPTCP_EVENT_REMOVED);
2078         if (!nlh)
2079                 goto nla_put_failure;
2080
2081         if (nla_put_u32(skb, MPTCP_ATTR_TOKEN, msk->token))
2082                 goto nla_put_failure;
2083
2084         if (nla_put_u8(skb, MPTCP_ATTR_REM_ID, id))
2085                 goto nla_put_failure;
2086
2087         genlmsg_end(skb, nlh);
2088         mptcp_nl_mcast_send(net, skb, GFP_ATOMIC);
2089         return;
2090
2091 nla_put_failure:
2092         nlmsg_free(skb);
2093 }
2094
2095 void mptcp_event_addr_announced(const struct sock *ssk,
2096                                 const struct mptcp_addr_info *info)
2097 {
2098         struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk);
2099         struct mptcp_sock *msk = mptcp_sk(subflow->conn);
2100         struct net *net = sock_net(ssk);
2101         struct nlmsghdr *nlh;
2102         struct sk_buff *skb;
2103
2104         if (!genl_has_listeners(&mptcp_genl_family, net, MPTCP_PM_EV_GRP_OFFSET))
2105                 return;
2106
2107         skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_ATOMIC);
2108         if (!skb)
2109                 return;
2110
2111         nlh = genlmsg_put(skb, 0, 0, &mptcp_genl_family, 0,
2112                           MPTCP_EVENT_ANNOUNCED);
2113         if (!nlh)
2114                 goto nla_put_failure;
2115
2116         if (nla_put_u32(skb, MPTCP_ATTR_TOKEN, msk->token))
2117                 goto nla_put_failure;
2118
2119         if (nla_put_u8(skb, MPTCP_ATTR_REM_ID, info->id))
2120                 goto nla_put_failure;
2121
2122         if (nla_put_be16(skb, MPTCP_ATTR_DPORT,
2123                          info->port == 0 ?
2124                          inet_sk(ssk)->inet_dport :
2125                          info->port))
2126                 goto nla_put_failure;
2127
2128         switch (info->family) {
2129         case AF_INET:
2130                 if (nla_put_in_addr(skb, MPTCP_ATTR_DADDR4, info->addr.s_addr))
2131                         goto nla_put_failure;
2132                 break;
2133 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
2134         case AF_INET6:
2135                 if (nla_put_in6_addr(skb, MPTCP_ATTR_DADDR6, &info->addr6))
2136                         goto nla_put_failure;
2137                 break;
2138 #endif
2139         default:
2140                 WARN_ON_ONCE(1);
2141                 goto nla_put_failure;
2142         }
2143
2144         genlmsg_end(skb, nlh);
2145         mptcp_nl_mcast_send(net, skb, GFP_ATOMIC);
2146         return;
2147
2148 nla_put_failure:
2149         nlmsg_free(skb);
2150 }
2151
2152 void mptcp_event_pm_listener(const struct sock *ssk,
2153                              enum mptcp_event_type event)
2154 {
2155         const struct inet_sock *issk = inet_sk(ssk);
2156         struct net *net = sock_net(ssk);
2157         struct nlmsghdr *nlh;
2158         struct sk_buff *skb;
2159
2160         if (!genl_has_listeners(&mptcp_genl_family, net, MPTCP_PM_EV_GRP_OFFSET))
2161                 return;
2162
2163         skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
2164         if (!skb)
2165                 return;
2166
2167         nlh = genlmsg_put(skb, 0, 0, &mptcp_genl_family, 0, event);
2168         if (!nlh)
2169                 goto nla_put_failure;
2170
2171         if (nla_put_u16(skb, MPTCP_ATTR_FAMILY, ssk->sk_family))
2172                 goto nla_put_failure;
2173
2174         if (nla_put_be16(skb, MPTCP_ATTR_SPORT, issk->inet_sport))
2175                 goto nla_put_failure;
2176
2177         switch (ssk->sk_family) {
2178         case AF_INET:
2179                 if (nla_put_in_addr(skb, MPTCP_ATTR_SADDR4, issk->inet_saddr))
2180                         goto nla_put_failure;
2181                 break;
2182 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
2183         case AF_INET6: {
2184                 const struct ipv6_pinfo *np = inet6_sk(ssk);
2185
2186                 if (nla_put_in6_addr(skb, MPTCP_ATTR_SADDR6, &np->saddr))
2187                         goto nla_put_failure;
2188                 break;
2189         }
2190 #endif
2191         default:
2192                 WARN_ON_ONCE(1);
2193                 goto nla_put_failure;
2194         }
2195
2196         genlmsg_end(skb, nlh);
2197         mptcp_nl_mcast_send(net, skb, GFP_KERNEL);
2198         return;
2199
2200 nla_put_failure:
2201         nlmsg_free(skb);
2202 }
2203
2204 void mptcp_event(enum mptcp_event_type type, const struct mptcp_sock *msk,
2205                  const struct sock *ssk, gfp_t gfp)
2206 {
2207         struct net *net = sock_net((const struct sock *)msk);
2208         struct nlmsghdr *nlh;
2209         struct sk_buff *skb;
2210
2211         if (!genl_has_listeners(&mptcp_genl_family, net, MPTCP_PM_EV_GRP_OFFSET))
2212                 return;
2213
2214         skb = nlmsg_new(NLMSG_DEFAULT_SIZE, gfp);
2215         if (!skb)
2216                 return;
2217
2218         nlh = genlmsg_put(skb, 0, 0, &mptcp_genl_family, 0, type);
2219         if (!nlh)
2220                 goto nla_put_failure;
2221
2222         switch (type) {
2223         case MPTCP_EVENT_UNSPEC:
2224                 WARN_ON_ONCE(1);
2225                 break;
2226         case MPTCP_EVENT_CREATED:
2227         case MPTCP_EVENT_ESTABLISHED:
2228                 if (mptcp_event_created(skb, msk, ssk) < 0)
2229                         goto nla_put_failure;
2230                 break;
2231         case MPTCP_EVENT_CLOSED:
2232                 if (nla_put_u32(skb, MPTCP_ATTR_TOKEN, msk->token) < 0)
2233                         goto nla_put_failure;
2234                 break;
2235         case MPTCP_EVENT_ANNOUNCED:
2236         case MPTCP_EVENT_REMOVED:
2237                 /* call mptcp_event_addr_announced()/removed instead */
2238                 WARN_ON_ONCE(1);
2239                 break;
2240         case MPTCP_EVENT_SUB_ESTABLISHED:
2241         case MPTCP_EVENT_SUB_PRIORITY:
2242                 if (mptcp_event_sub_established(skb, msk, ssk) < 0)
2243                         goto nla_put_failure;
2244                 break;
2245         case MPTCP_EVENT_SUB_CLOSED:
2246                 if (mptcp_event_sub_closed(skb, msk, ssk) < 0)
2247                         goto nla_put_failure;
2248                 break;
2249         case MPTCP_EVENT_LISTENER_CREATED:
2250         case MPTCP_EVENT_LISTENER_CLOSED:
2251                 break;
2252         }
2253
2254         genlmsg_end(skb, nlh);
2255         mptcp_nl_mcast_send(net, skb, gfp);
2256         return;
2257
2258 nla_put_failure:
2259         nlmsg_free(skb);
2260 }
2261
2262 static struct genl_family mptcp_genl_family __ro_after_init = {
2263         .name           = MPTCP_PM_NAME,
2264         .version        = MPTCP_PM_VER,
2265         .netnsok        = true,
2266         .module         = THIS_MODULE,
2267         .ops            = mptcp_pm_nl_ops,
2268         .n_ops          = ARRAY_SIZE(mptcp_pm_nl_ops),
2269         .resv_start_op  = MPTCP_PM_CMD_SUBFLOW_DESTROY + 1,
2270         .mcgrps         = mptcp_pm_mcgrps,
2271         .n_mcgrps       = ARRAY_SIZE(mptcp_pm_mcgrps),
2272 };
2273
2274 static int __net_init pm_nl_init_net(struct net *net)
2275 {
2276         struct pm_nl_pernet *pernet = pm_nl_get_pernet(net);
2277
2278         INIT_LIST_HEAD_RCU(&pernet->local_addr_list);
2279
2280         /* Cit. 2 subflows ought to be enough for anybody. */
2281         pernet->subflows_max = 2;
2282         pernet->next_id = 1;
2283         pernet->stale_loss_cnt = 4;
2284         spin_lock_init(&pernet->lock);
2285
2286         /* No need to initialize other pernet fields, the struct is zeroed at
2287          * allocation time.
2288          */
2289
2290         return 0;
2291 }
2292
2293 static void __net_exit pm_nl_exit_net(struct list_head *net_list)
2294 {
2295         struct net *net;
2296
2297         list_for_each_entry(net, net_list, exit_list) {
2298                 struct pm_nl_pernet *pernet = pm_nl_get_pernet(net);
2299
2300                 /* net is removed from namespace list, can't race with
2301                  * other modifiers, also netns core already waited for a
2302                  * RCU grace period.
2303                  */
2304                 __flush_addrs(&pernet->local_addr_list);
2305         }
2306 }
2307
2308 static struct pernet_operations mptcp_pm_pernet_ops = {
2309         .init = pm_nl_init_net,
2310         .exit_batch = pm_nl_exit_net,
2311         .id = &pm_nl_pernet_id,
2312         .size = sizeof(struct pm_nl_pernet),
2313 };
2314
2315 void __init mptcp_pm_nl_init(void)
2316 {
2317         if (register_pernet_subsys(&mptcp_pm_pernet_ops) < 0)
2318                 panic("Failed to register MPTCP PM pernet subsystem.\n");
2319
2320         if (genl_register_family(&mptcp_genl_family))
2321                 panic("Failed to register MPTCP PM netlink family\n");
2322 }