net/mptcp/pm_netlink.c

   1 // SPDX-License-Identifier: GPL-2.0
   2 /* Multipath TCP
   3  *
   4  * Copyright (c) 2020, Red Hat, Inc.
   5  */
   6
   7 #define pr_fmt(fmt) "MPTCP: " fmt
   8
   9 #include <linux/inet.h>
  10 #include <linux/kernel.h>
  11 #include <net/tcp.h>
  12 #include <net/netns/generic.h>
  13 #include <net/mptcp.h>
  14 #include <net/genetlink.h>
  15 #include <uapi/linux/mptcp.h>
  16
  17 #include "protocol.h"
  18 #include "mib.h"
  19
  20 /* forward declaration */
  21 static struct genl_family mptcp_genl_family;
  22
  23 static int pm_nl_pernet_id;
  24
  25 struct mptcp_pm_addr_entry {
  26         struct list_head        list;
  27         struct mptcp_addr_info  addr;
  28         u8                      flags;
  29         int                     ifindex;
  30         struct socket           *lsk;
  31 };
  32
  33 struct mptcp_pm_add_entry {
  34         struct list_head        list;
  35         struct mptcp_addr_info  addr;
  36         struct timer_list       add_timer;
  37         struct mptcp_sock       *sock;
  38         u8                      retrans_times;
  39 };
  40
  41 struct pm_nl_pernet {
  42         /* protects pernet updates */
  43         spinlock_t              lock;
  44         struct list_head        local_addr_list;
  45         unsigned int            addrs;
  46         unsigned int            stale_loss_cnt;
  47         unsigned int            add_addr_signal_max;
  48         unsigned int            add_addr_accept_max;
  49         unsigned int            local_addr_max;
  50         unsigned int            subflows_max;
  51         unsigned int            next_id;
  52         DECLARE_BITMAP(id_bitmap, MPTCP_PM_MAX_ADDR_ID + 1);
  53 };
  54
  55 #define MPTCP_PM_ADDR_MAX       8
  56 #define ADD_ADDR_RETRANS_MAX    3
  57
  58 static bool addresses_equal(const struct mptcp_addr_info *a,
  59                             const struct mptcp_addr_info *b, bool use_port)
  60 {
  61         bool addr_equals = false;
  62
  63         if (a->family == b->family) {
  64                 if (a->family == AF_INET)
  65                         addr_equals = a->addr.s_addr == b->addr.s_addr;
  66 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
  67                 else
  68                         addr_equals = !ipv6_addr_cmp(&a->addr6, &b->addr6);
  69         } else if (a->family == AF_INET) {
  70                 if (ipv6_addr_v4mapped(&b->addr6))
  71                         addr_equals = a->addr.s_addr == b->addr6.s6_addr32[3];
  72         } else if (b->family == AF_INET) {
  73                 if (ipv6_addr_v4mapped(&a->addr6))
  74                         addr_equals = a->addr6.s6_addr32[3] == b->addr.s_addr;
  75 #endif
  76         }
  77
  78         if (!addr_equals)
  79                 return false;
  80         if (!use_port)
  81                 return true;
  82
  83         return a->port == b->port;
  84 }
  85
  86 static bool address_zero(const struct mptcp_addr_info *addr)
  87 {
  88         struct mptcp_addr_info zero;
  89
  90         memset(&zero, 0, sizeof(zero));
  91         zero.family = addr->family;
  92
  93         return addresses_equal(addr, &zero, true);
  94 }
  95
  96 static void local_address(const struct sock_common *skc,
  97                           struct mptcp_addr_info *addr)
  98 {
  99         addr->family = skc->skc_family;
 100         addr->port = htons(skc->skc_num);
 101         if (addr->family == AF_INET)
 102                 addr->addr.s_addr = skc->skc_rcv_saddr;
 103 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
 104         else if (addr->family == AF_INET6)
 105                 addr->addr6 = skc->skc_v6_rcv_saddr;
 106 #endif
 107 }
 108
 109 static void remote_address(const struct sock_common *skc,
 110                            struct mptcp_addr_info *addr)
 111 {
 112         addr->family = skc->skc_family;
 113         addr->port = skc->skc_dport;
 114         if (addr->family == AF_INET)
 115                 addr->addr.s_addr = skc->skc_daddr;
 116 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
 117         else if (addr->family == AF_INET6)
 118                 addr->addr6 = skc->skc_v6_daddr;
 119 #endif
 120 }
 121
 122 static bool lookup_subflow_by_saddr(const struct list_head *list,
 123                                     struct mptcp_addr_info *saddr)
 124 {
 125         struct mptcp_subflow_context *subflow;
 126         struct mptcp_addr_info cur;
 127         struct sock_common *skc;
 128
 129         list_for_each_entry(subflow, list, node) {
 130                 skc = (struct sock_common *)mptcp_subflow_tcp_sock(subflow);
 131
 132                 local_address(skc, &cur);
 133                 if (addresses_equal(&cur, saddr, saddr->port))
 134                         return true;
 135         }
 136
 137         return false;
 138 }
 139
 140 static bool lookup_subflow_by_daddr(const struct list_head *list,
 141                                     struct mptcp_addr_info *daddr)
 142 {
 143         struct mptcp_subflow_context *subflow;
 144         struct mptcp_addr_info cur;
 145         struct sock_common *skc;
 146
 147         list_for_each_entry(subflow, list, node) {
 148                 skc = (struct sock_common *)mptcp_subflow_tcp_sock(subflow);
 149
 150                 remote_address(skc, &cur);
 151                 if (addresses_equal(&cur, daddr, daddr->port))
 152                         return true;
 153         }
 154
 155         return false;
 156 }
 157
 158 static struct mptcp_pm_addr_entry *
 159 select_local_address(const struct pm_nl_pernet *pernet,
 160                      struct mptcp_sock *msk)
 161 {
 162         struct mptcp_pm_addr_entry *entry, *ret = NULL;
 163         struct sock *sk = (struct sock *)msk;
 164
 165         msk_owned_by_me(msk);
 166
 167         rcu_read_lock();
 168         list_for_each_entry_rcu(entry, &pernet->local_addr_list, list) {
 169                 if (!(entry->flags & MPTCP_PM_ADDR_FLAG_SUBFLOW))
 170                         continue;
 171
 172                 if (!test_bit(entry->addr.id, msk->pm.id_avail_bitmap))
 173                         continue;
 174
 175                 if (entry->addr.family != sk->sk_family) {
 176 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
 177                         if ((entry->addr.family == AF_INET &&
 178                              !ipv6_addr_v4mapped(&sk->sk_v6_daddr)) ||
 179                             (sk->sk_family == AF_INET &&
 180                              !ipv6_addr_v4mapped(&entry->addr.addr6)))
 181 #endif
 182                                 continue;
 183                 }
 184
 185                 ret = entry;
 186                 break;
 187         }
 188         rcu_read_unlock();
 189         return ret;
 190 }
 191
 192 static struct mptcp_pm_addr_entry *
 193 select_signal_address(struct pm_nl_pernet *pernet, struct mptcp_sock *msk)
 194 {
 195         struct mptcp_pm_addr_entry *entry, *ret = NULL;
 196
 197         rcu_read_lock();
 198         /* do not keep any additional per socket state, just signal
 199          * the address list in order.
 200          * Note: removal from the local address list during the msk life-cycle
 201          * can lead to additional addresses not being announced.
 202          */
 203         list_for_each_entry_rcu(entry, &pernet->local_addr_list, list) {
 204                 if (!test_bit(entry->addr.id, msk->pm.id_avail_bitmap))
 205                         continue;
 206
 207                 if (!(entry->flags & MPTCP_PM_ADDR_FLAG_SIGNAL))
 208                         continue;
 209
 210                 ret = entry;
 211                 break;
 212         }
 213         rcu_read_unlock();
 214         return ret;
 215 }
 216
 217 unsigned int mptcp_pm_get_add_addr_signal_max(struct mptcp_sock *msk)
 218 {
 219         struct pm_nl_pernet *pernet;
 220
 221         pernet = net_generic(sock_net((struct sock *)msk), pm_nl_pernet_id);
 222         return READ_ONCE(pernet->add_addr_signal_max);
 223 }
 224 EXPORT_SYMBOL_GPL(mptcp_pm_get_add_addr_signal_max);
 225
 226 unsigned int mptcp_pm_get_add_addr_accept_max(struct mptcp_sock *msk)
 227 {
 228         struct pm_nl_pernet *pernet;
 229
 230         pernet = net_generic(sock_net((struct sock *)msk), pm_nl_pernet_id);
 231         return READ_ONCE(pernet->add_addr_accept_max);
 232 }
 233 EXPORT_SYMBOL_GPL(mptcp_pm_get_add_addr_accept_max);
 234
 235 unsigned int mptcp_pm_get_subflows_max(struct mptcp_sock *msk)
 236 {
 237         struct pm_nl_pernet *pernet;
 238
 239         pernet = net_generic(sock_net((struct sock *)msk), pm_nl_pernet_id);
 240         return READ_ONCE(pernet->subflows_max);
 241 }
 242 EXPORT_SYMBOL_GPL(mptcp_pm_get_subflows_max);
 243
 244 unsigned int mptcp_pm_get_local_addr_max(struct mptcp_sock *msk)
 245 {
 246         struct pm_nl_pernet *pernet;
 247
 248         pernet = net_generic(sock_net((struct sock *)msk), pm_nl_pernet_id);
 249         return READ_ONCE(pernet->local_addr_max);
 250 }
 251 EXPORT_SYMBOL_GPL(mptcp_pm_get_local_addr_max);
 252
 253 bool mptcp_pm_nl_check_work_pending(struct mptcp_sock *msk)
 254 {
 255         struct pm_nl_pernet *pernet = net_generic(sock_net((struct sock *)msk), pm_nl_pernet_id);
 256
 257         if (msk->pm.subflows == mptcp_pm_get_subflows_max(msk) ||
 258             (find_next_and_bit(pernet->id_bitmap, msk->pm.id_avail_bitmap,
 259                                MPTCP_PM_MAX_ADDR_ID + 1, 0) == MPTCP_PM_MAX_ADDR_ID + 1)) {
 260                 WRITE_ONCE(msk->pm.work_pending, false);
 261                 return false;
 262         }
 263         return true;
 264 }
 265
 266 struct mptcp_pm_add_entry *
 267 mptcp_lookup_anno_list_by_saddr(struct mptcp_sock *msk,
 268                                 struct mptcp_addr_info *addr)
 269 {
 270         struct mptcp_pm_add_entry *entry;
 271
 272         lockdep_assert_held(&msk->pm.lock);
 273
 274         list_for_each_entry(entry, &msk->pm.anno_list, list) {
 275                 if (addresses_equal(&entry->addr, addr, true))
 276                         return entry;
 277         }
 278
 279         return NULL;
 280 }
 281
 282 bool mptcp_pm_sport_in_anno_list(struct mptcp_sock *msk, const struct sock *sk)
 283 {
 284         struct mptcp_pm_add_entry *entry;
 285         struct mptcp_addr_info saddr;
 286         bool ret = false;
 287
 288         local_address((struct sock_common *)sk, &saddr);
 289
 290         spin_lock_bh(&msk->pm.lock);
 291         list_for_each_entry(entry, &msk->pm.anno_list, list) {
 292                 if (addresses_equal(&entry->addr, &saddr, true)) {
 293                         ret = true;
 294                         goto out;
 295                 }
 296         }
 297
 298 out:
 299         spin_unlock_bh(&msk->pm.lock);
 300         return ret;
 301 }
 302
 303 static void mptcp_pm_add_timer(struct timer_list *timer)
 304 {
 305         struct mptcp_pm_add_entry *entry = from_timer(entry, timer, add_timer);
 306         struct mptcp_sock *msk = entry->sock;
 307         struct sock *sk = (struct sock *)msk;
 308
 309         pr_debug("msk=%p", msk);
 310
 311         if (!msk)
 312                 return;
 313
 314         if (inet_sk_state_load(sk) == TCP_CLOSE)
 315                 return;
 316
 317         if (!entry->addr.id)
 318                 return;
 319
 320         if (mptcp_pm_should_add_signal_addr(msk)) {
 321                 sk_reset_timer(sk, timer, jiffies + TCP_RTO_MAX / 8);
 322                 goto out;
 323         }
 324
 325         spin_lock_bh(&msk->pm.lock);
 326
 327         if (!mptcp_pm_should_add_signal_addr(msk)) {
 328                 pr_debug("retransmit ADD_ADDR id=%d", entry->addr.id);
 329                 mptcp_pm_announce_addr(msk, &entry->addr, false);
 330                 mptcp_pm_add_addr_send_ack(msk);
 331                 entry->retrans_times++;
 332         }
 333
 334         if (entry->retrans_times < ADD_ADDR_RETRANS_MAX)
 335                 sk_reset_timer(sk, timer,
 336                                jiffies + mptcp_get_add_addr_timeout(sock_net(sk)));
 337
 338         spin_unlock_bh(&msk->pm.lock);
 339
 340         if (entry->retrans_times == ADD_ADDR_RETRANS_MAX)
 341                 mptcp_pm_subflow_established(msk);
 342
 343 out:
 344         __sock_put(sk);
 345 }
 346
 347 struct mptcp_pm_add_entry *
 348 mptcp_pm_del_add_timer(struct mptcp_sock *msk,
 349                        struct mptcp_addr_info *addr, bool check_id)
 350 {
 351         struct mptcp_pm_add_entry *entry;
 352         struct sock *sk = (struct sock *)msk;
 353
 354         spin_lock_bh(&msk->pm.lock);
 355         entry = mptcp_lookup_anno_list_by_saddr(msk, addr);
 356         if (entry && (!check_id || entry->addr.id == addr->id))
 357                 entry->retrans_times = ADD_ADDR_RETRANS_MAX;
 358         spin_unlock_bh(&msk->pm.lock);
 359
 360         if (entry && (!check_id || entry->addr.id == addr->id))
 361                 sk_stop_timer_sync(sk, &entry->add_timer);
 362
 363         return entry;
 364 }
 365
 366 static bool mptcp_pm_alloc_anno_list(struct mptcp_sock *msk,
 367                                      struct mptcp_pm_addr_entry *entry)
 368 {
 369         struct mptcp_pm_add_entry *add_entry = NULL;
 370         struct sock *sk = (struct sock *)msk;
 371         struct net *net = sock_net(sk);
 372
 373         lockdep_assert_held(&msk->pm.lock);
 374
 375         if (mptcp_lookup_anno_list_by_saddr(msk, &entry->addr))
 376                 return false;
 377
 378         add_entry = kmalloc(sizeof(*add_entry), GFP_ATOMIC);
 379         if (!add_entry)
 380                 return false;
 381
 382         list_add(&add_entry->list, &msk->pm.anno_list);
 383
 384         add_entry->addr = entry->addr;
 385         add_entry->sock = msk;
 386         add_entry->retrans_times = 0;
 387
 388         timer_setup(&add_entry->add_timer, mptcp_pm_add_timer, 0);
 389         sk_reset_timer(sk, &add_entry->add_timer,
 390                        jiffies + mptcp_get_add_addr_timeout(net));
 391
 392         return true;
 393 }
 394
 395 void mptcp_pm_free_anno_list(struct mptcp_sock *msk)
 396 {
 397         struct mptcp_pm_add_entry *entry, *tmp;
 398         struct sock *sk = (struct sock *)msk;
 399         LIST_HEAD(free_list);
 400
 401         pr_debug("msk=%p", msk);
 402
 403         spin_lock_bh(&msk->pm.lock);
 404         list_splice_init(&msk->pm.anno_list, &free_list);
 405         spin_unlock_bh(&msk->pm.lock);
 406
 407         list_for_each_entry_safe(entry, tmp, &free_list, list) {
 408                 sk_stop_timer_sync(sk, &entry->add_timer);
 409                 kfree(entry);
 410         }
 411 }
 412
 413 static bool lookup_address_in_vec(struct mptcp_addr_info *addrs, unsigned int nr,
 414                                   struct mptcp_addr_info *addr)
 415 {
 416         int i;
 417
 418         for (i = 0; i < nr; i++) {
 419                 if (addresses_equal(&addrs[i], addr, addr->port))
 420                         return true;
 421         }
 422
 423         return false;
 424 }
 425
 426 /* Fill all the remote addresses into the array addrs[],
 427  * and return the array size.
 428  */
 429 static unsigned int fill_remote_addresses_vec(struct mptcp_sock *msk, bool fullmesh,
 430                                               struct mptcp_addr_info *addrs)
 431 {
 432         bool deny_id0 = READ_ONCE(msk->pm.remote_deny_join_id0);
 433         struct sock *sk = (struct sock *)msk, *ssk;
 434         struct mptcp_subflow_context *subflow;
 435         struct mptcp_addr_info remote = { 0 };
 436         unsigned int subflows_max;
 437         int i = 0;
 438
 439         subflows_max = mptcp_pm_get_subflows_max(msk);
 440         remote_address((struct sock_common *)sk, &remote);
 441
 442         /* Non-fullmesh endpoint, fill in the single entry
 443          * corresponding to the primary MPC subflow remote address
 444          */
 445         if (!fullmesh) {
 446                 if (deny_id0)
 447                         return 0;
 448
 449                 msk->pm.subflows++;
 450                 addrs[i++] = remote;
 451         } else {
 452                 mptcp_for_each_subflow(msk, subflow) {
 453                         ssk = mptcp_subflow_tcp_sock(subflow);
 454                         remote_address((struct sock_common *)ssk, &addrs[i]);
 455                         if (deny_id0 && addresses_equal(&addrs[i], &remote, false))
 456                                 continue;
 457
 458                         if (!lookup_address_in_vec(addrs, i, &addrs[i]) &&
 459                             msk->pm.subflows < subflows_max) {
 460                                 msk->pm.subflows++;
 461                                 i++;
 462                         }
 463                 }
 464         }
 465
 466         return i;
 467 }
 468
 469 static struct mptcp_pm_addr_entry *
 470 __lookup_addr_by_id(struct pm_nl_pernet *pernet, unsigned int id)
 471 {
 472         struct mptcp_pm_addr_entry *entry;
 473
 474         list_for_each_entry(entry, &pernet->local_addr_list, list) {
 475                 if (entry->addr.id == id)
 476                         return entry;
 477         }
 478         return NULL;
 479 }
 480
 481 static struct mptcp_pm_addr_entry *
 482 __lookup_addr(struct pm_nl_pernet *pernet, const struct mptcp_addr_info *info,
 483               bool lookup_by_id)
 484 {
 485         struct mptcp_pm_addr_entry *entry;
 486
 487         list_for_each_entry(entry, &pernet->local_addr_list, list) {
 488                 if ((!lookup_by_id && addresses_equal(&entry->addr, info, true)) ||
 489                     (lookup_by_id && entry->addr.id == info->id))
 490                         return entry;
 491         }
 492         return NULL;
 493 }
 494
 495 static int
 496 lookup_id_by_addr(struct pm_nl_pernet *pernet, const struct mptcp_addr_info *addr)
 497 {
 498         struct mptcp_pm_addr_entry *entry;
 499         int ret = -1;
 500
 501         rcu_read_lock();
 502         list_for_each_entry(entry, &pernet->local_addr_list, list) {
 503                 if (addresses_equal(&entry->addr, addr, entry->addr.port)) {
 504                         ret = entry->addr.id;
 505                         break;
 506                 }
 507         }
 508         rcu_read_unlock();
 509         return ret;
 510 }
 511
 512 static void mptcp_pm_create_subflow_or_signal_addr(struct mptcp_sock *msk)
 513 {
 514         struct sock *sk = (struct sock *)msk;
 515         struct mptcp_pm_addr_entry *local;
 516         unsigned int add_addr_signal_max;
 517         unsigned int local_addr_max;
 518         struct pm_nl_pernet *pernet;
 519         unsigned int subflows_max;
 520
 521         pernet = net_generic(sock_net(sk), pm_nl_pernet_id);
 522
 523         add_addr_signal_max = mptcp_pm_get_add_addr_signal_max(msk);
 524         local_addr_max = mptcp_pm_get_local_addr_max(msk);
 525         subflows_max = mptcp_pm_get_subflows_max(msk);
 526
 527         /* do lazy endpoint usage accounting for the MPC subflows */
 528         if (unlikely(!(msk->pm.status & BIT(MPTCP_PM_MPC_ENDPOINT_ACCOUNTED))) && msk->first) {
 529                 struct mptcp_addr_info mpc_addr;
 530                 int mpc_id;
 531
 532                 local_address((struct sock_common *)msk->first, &mpc_addr);
 533                 mpc_id = lookup_id_by_addr(pernet, &mpc_addr);
 534                 if (mpc_id >= 0)
 535                         __clear_bit(mpc_id, msk->pm.id_avail_bitmap);
 536
 537                 msk->pm.status |= BIT(MPTCP_PM_MPC_ENDPOINT_ACCOUNTED);
 538         }
 539
 540         pr_debug("local %d:%d signal %d:%d subflows %d:%d\n",
 541                  msk->pm.local_addr_used, local_addr_max,
 542                  msk->pm.add_addr_signaled, add_addr_signal_max,
 543                  msk->pm.subflows, subflows_max);
 544
 545         /* check first for announce */
 546         if (msk->pm.add_addr_signaled < add_addr_signal_max) {
 547                 local = select_signal_address(pernet, msk);
 548
 549                 /* due to racing events on both ends we can reach here while
 550                  * previous add address is still running: if we invoke now
 551                  * mptcp_pm_announce_addr(), that will fail and the
 552                  * corresponding id will be marked as used.
 553                  * Instead let the PM machinery reschedule us when the
 554                  * current address announce will be completed.
 555                  */
 556                 if (msk->pm.addr_signal & BIT(MPTCP_ADD_ADDR_SIGNAL))
 557                         return;
 558
 559                 if (local) {
 560                         if (mptcp_pm_alloc_anno_list(msk, local)) {
 561                                 __clear_bit(local->addr.id, msk->pm.id_avail_bitmap);
 562                                 msk->pm.add_addr_signaled++;
 563                                 mptcp_pm_announce_addr(msk, &local->addr, false);
 564                                 mptcp_pm_nl_addr_send_ack(msk);
 565                         }
 566                 }
 567         }
 568
 569         /* check if should create a new subflow */
 570         while (msk->pm.local_addr_used < local_addr_max &&
 571                msk->pm.subflows < subflows_max) {
 572                 struct mptcp_addr_info addrs[MPTCP_PM_ADDR_MAX];
 573                 bool fullmesh;
 574                 int i, nr;
 575
 576                 local = select_local_address(pernet, msk);
 577                 if (!local)
 578                         break;
 579
 580                 fullmesh = !!(local->flags & MPTCP_PM_ADDR_FLAG_FULLMESH);
 581
 582                 msk->pm.local_addr_used++;
 583                 nr = fill_remote_addresses_vec(msk, fullmesh, addrs);
 584                 if (nr)
 585                         __clear_bit(local->addr.id, msk->pm.id_avail_bitmap);
 586                 spin_unlock_bh(&msk->pm.lock);
 587                 for (i = 0; i < nr; i++)
 588                         __mptcp_subflow_connect(sk, &local->addr, &addrs[i]);
 589                 spin_lock_bh(&msk->pm.lock);
 590         }
 591         mptcp_pm_nl_check_work_pending(msk);
 592 }
 593
 594 static void mptcp_pm_nl_fully_established(struct mptcp_sock *msk)
 595 {
 596         mptcp_pm_create_subflow_or_signal_addr(msk);
 597 }
 598
 599 static void mptcp_pm_nl_subflow_established(struct mptcp_sock *msk)
 600 {
 601         mptcp_pm_create_subflow_or_signal_addr(msk);
 602 }
 603
 604 /* Fill all the local addresses into the array addrs[],
 605  * and return the array size.
 606  */
 607 static unsigned int fill_local_addresses_vec(struct mptcp_sock *msk,
 608                                              struct mptcp_addr_info *addrs)
 609 {
 610         struct sock *sk = (struct sock *)msk;
 611         struct mptcp_pm_addr_entry *entry;
 612         struct mptcp_addr_info local;
 613         struct pm_nl_pernet *pernet;
 614         unsigned int subflows_max;
 615         int i = 0;
 616
 617         pernet = net_generic(sock_net(sk), pm_nl_pernet_id);
 618         subflows_max = mptcp_pm_get_subflows_max(msk);
 619
 620         rcu_read_lock();
 621         list_for_each_entry_rcu(entry, &pernet->local_addr_list, list) {
 622                 if (!(entry->flags & MPTCP_PM_ADDR_FLAG_FULLMESH))
 623                         continue;
 624
 625                 if (entry->addr.family != sk->sk_family) {
 626 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
 627                         if ((entry->addr.family == AF_INET &&
 628                              !ipv6_addr_v4mapped(&sk->sk_v6_daddr)) ||
 629                             (sk->sk_family == AF_INET &&
 630                              !ipv6_addr_v4mapped(&entry->addr.addr6)))
 631 #endif
 632                                 continue;
 633                 }
 634
 635                 if (msk->pm.subflows < subflows_max) {
 636                         msk->pm.subflows++;
 637                         addrs[i++] = entry->addr;
 638                 }
 639         }
 640         rcu_read_unlock();
 641
 642         /* If the array is empty, fill in the single
 643          * 'IPADDRANY' local address
 644          */
 645         if (!i) {
 646                 memset(&local, 0, sizeof(local));
 647                 local.family = msk->pm.remote.family;
 648
 649                 msk->pm.subflows++;
 650                 addrs[i++] = local;
 651         }
 652
 653         return i;
 654 }
 655
 656 static void mptcp_pm_nl_add_addr_received(struct mptcp_sock *msk)
 657 {
 658         struct mptcp_addr_info addrs[MPTCP_PM_ADDR_MAX];
 659         struct sock *sk = (struct sock *)msk;
 660         unsigned int add_addr_accept_max;
 661         struct mptcp_addr_info remote;
 662         unsigned int subflows_max;
 663         int i, nr;
 664
 665         add_addr_accept_max = mptcp_pm_get_add_addr_accept_max(msk);
 666         subflows_max = mptcp_pm_get_subflows_max(msk);
 667
 668         pr_debug("accepted %d:%d remote family %d",
 669                  msk->pm.add_addr_accepted, add_addr_accept_max,
 670                  msk->pm.remote.family);
 671
 672         if (lookup_subflow_by_daddr(&msk->conn_list, &msk->pm.remote))
 673                 goto add_addr_echo;
 674
 675         /* connect to the specified remote address, using whatever
 676          * local address the routing configuration will pick.
 677          */
 678         remote = msk->pm.remote;
 679         if (!remote.port)
 680                 remote.port = sk->sk_dport;
 681         nr = fill_local_addresses_vec(msk, addrs);
 682
 683         msk->pm.add_addr_accepted++;
 684         if (msk->pm.add_addr_accepted >= add_addr_accept_max ||
 685             msk->pm.subflows >= subflows_max)
 686                 WRITE_ONCE(msk->pm.accept_addr, false);
 687
 688         spin_unlock_bh(&msk->pm.lock);
 689         for (i = 0; i < nr; i++)
 690                 __mptcp_subflow_connect(sk, &addrs[i], &remote);
 691         spin_lock_bh(&msk->pm.lock);
 692
 693 add_addr_echo:
 694         mptcp_pm_announce_addr(msk, &msk->pm.remote, true);
 695         mptcp_pm_nl_addr_send_ack(msk);
 696 }
 697
 698 void mptcp_pm_nl_addr_send_ack(struct mptcp_sock *msk)
 699 {
 700         struct mptcp_subflow_context *subflow;
 701
 702         msk_owned_by_me(msk);
 703         lockdep_assert_held(&msk->pm.lock);
 704
 705         if (!mptcp_pm_should_add_signal(msk) &&
 706             !mptcp_pm_should_rm_signal(msk))
 707                 return;
 708
 709         subflow = list_first_entry_or_null(&msk->conn_list, typeof(*subflow), node);
 710         if (subflow) {
 711                 struct sock *ssk = mptcp_subflow_tcp_sock(subflow);
 712
 713                 spin_unlock_bh(&msk->pm.lock);
 714                 pr_debug("send ack for %s",
 715                          mptcp_pm_should_add_signal(msk) ? "add_addr" : "rm_addr");
 716
 717                 mptcp_subflow_send_ack(ssk);
 718                 spin_lock_bh(&msk->pm.lock);
 719         }
 720 }
 721
 722 static int mptcp_pm_nl_mp_prio_send_ack(struct mptcp_sock *msk,
 723                                         struct mptcp_addr_info *addr,
 724                                         u8 bkup)
 725 {
 726         struct mptcp_subflow_context *subflow;
 727
 728         pr_debug("bkup=%d", bkup);
 729
 730         mptcp_for_each_subflow(msk, subflow) {
 731                 struct sock *ssk = mptcp_subflow_tcp_sock(subflow);
 732                 struct sock *sk = (struct sock *)msk;
 733                 struct mptcp_addr_info local;
 734
 735                 local_address((struct sock_common *)ssk, &local);
 736                 if (!addresses_equal(&local, addr, addr->port))
 737                         continue;
 738
 739                 subflow->backup = bkup;
 740                 subflow->send_mp_prio = 1;
 741                 subflow->request_bkup = bkup;
 742                 __MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_MPPRIOTX);
 743
 744                 spin_unlock_bh(&msk->pm.lock);
 745                 pr_debug("send ack for mp_prio");
 746                 mptcp_subflow_send_ack(ssk);
 747                 spin_lock_bh(&msk->pm.lock);
 748
 749                 return 0;
 750         }
 751
 752         return -EINVAL;
 753 }
 754
 755 static void mptcp_pm_nl_rm_addr_or_subflow(struct mptcp_sock *msk,
 756                                            const struct mptcp_rm_list *rm_list,
 757                                            enum linux_mptcp_mib_field rm_type)
 758 {
 759         struct mptcp_subflow_context *subflow, *tmp;
 760         struct sock *sk = (struct sock *)msk;
 761         u8 i;
 762
 763         pr_debug("%s rm_list_nr %d",
 764                  rm_type == MPTCP_MIB_RMADDR ? "address" : "subflow", rm_list->nr);
 765
 766         msk_owned_by_me(msk);
 767
 768         if (sk->sk_state == TCP_LISTEN)
 769                 return;
 770
 771         if (!rm_list->nr)
 772                 return;
 773
 774         if (list_empty(&msk->conn_list))
 775                 return;
 776
 777         for (i = 0; i < rm_list->nr; i++) {
 778                 bool removed = false;
 779
 780                 list_for_each_entry_safe(subflow, tmp, &msk->conn_list, node) {
 781                         struct sock *ssk = mptcp_subflow_tcp_sock(subflow);
 782                         int how = RCV_SHUTDOWN | SEND_SHUTDOWN;
 783                         u8 id = subflow->local_id;
 784
 785                         if (rm_type == MPTCP_MIB_RMADDR)
 786                                 id = subflow->remote_id;
 787
 788                         if (rm_list->ids[i] != id)
 789                                 continue;
 790
 791                         pr_debug(" -> %s rm_list_ids[%d]=%u local_id=%u remote_id=%u",
 792                                  rm_type == MPTCP_MIB_RMADDR ? "address" : "subflow",
 793                                  i, rm_list->ids[i], subflow->local_id, subflow->remote_id);
 794                         spin_unlock_bh(&msk->pm.lock);
 795                         mptcp_subflow_shutdown(sk, ssk, how);
 796
 797                         /* the following takes care of updating the subflows counter */
 798                         mptcp_close_ssk(sk, ssk, subflow);
 799                         spin_lock_bh(&msk->pm.lock);
 800
 801                         removed = true;
 802                         __MPTCP_INC_STATS(sock_net(sk), rm_type);
 803                 }
 804                 __set_bit(rm_list->ids[i], msk->pm.id_avail_bitmap);
 805                 if (!removed)
 806                         continue;
 807
 808                 if (rm_type == MPTCP_MIB_RMADDR) {
 809                         msk->pm.add_addr_accepted--;
 810                         WRITE_ONCE(msk->pm.accept_addr, true);
 811                 } else if (rm_type == MPTCP_MIB_RMSUBFLOW) {
 812                         msk->pm.local_addr_used--;
 813                 }
 814         }
 815 }
 816
 817 static void mptcp_pm_nl_rm_addr_received(struct mptcp_sock *msk)
 818 {
 819         mptcp_pm_nl_rm_addr_or_subflow(msk, &msk->pm.rm_list_rx, MPTCP_MIB_RMADDR);
 820 }
 821
 822 void mptcp_pm_nl_rm_subflow_received(struct mptcp_sock *msk,
 823                                      const struct mptcp_rm_list *rm_list)
 824 {
 825         mptcp_pm_nl_rm_addr_or_subflow(msk, rm_list, MPTCP_MIB_RMSUBFLOW);
 826 }
 827
 828 void mptcp_pm_nl_work(struct mptcp_sock *msk)
 829 {
 830         struct mptcp_pm_data *pm = &msk->pm;
 831
 832         msk_owned_by_me(msk);
 833
 834         if (!(pm->status & MPTCP_PM_WORK_MASK))
 835                 return;
 836
 837         spin_lock_bh(&msk->pm.lock);
 838
 839         pr_debug("msk=%p status=%x", msk, pm->status);
 840         if (pm->status & BIT(MPTCP_PM_ADD_ADDR_RECEIVED)) {
 841                 pm->status &= ~BIT(MPTCP_PM_ADD_ADDR_RECEIVED);
 842                 mptcp_pm_nl_add_addr_received(msk);
 843         }
 844         if (pm->status & BIT(MPTCP_PM_ADD_ADDR_SEND_ACK)) {
 845                 pm->status &= ~BIT(MPTCP_PM_ADD_ADDR_SEND_ACK);
 846                 mptcp_pm_nl_addr_send_ack(msk);
 847         }
 848         if (pm->status & BIT(MPTCP_PM_RM_ADDR_RECEIVED)) {
 849                 pm->status &= ~BIT(MPTCP_PM_RM_ADDR_RECEIVED);
 850                 mptcp_pm_nl_rm_addr_received(msk);
 851         }
 852         if (pm->status & BIT(MPTCP_PM_ESTABLISHED)) {
 853                 pm->status &= ~BIT(MPTCP_PM_ESTABLISHED);
 854                 mptcp_pm_nl_fully_established(msk);
 855         }
 856         if (pm->status & BIT(MPTCP_PM_SUBFLOW_ESTABLISHED)) {
 857                 pm->status &= ~BIT(MPTCP_PM_SUBFLOW_ESTABLISHED);
 858                 mptcp_pm_nl_subflow_established(msk);
 859         }
 860
 861         spin_unlock_bh(&msk->pm.lock);
 862 }
 863
 864 static bool address_use_port(struct mptcp_pm_addr_entry *entry)
 865 {
 866         return (entry->flags &
 867                 (MPTCP_PM_ADDR_FLAG_SIGNAL | MPTCP_PM_ADDR_FLAG_SUBFLOW)) ==
 868                 MPTCP_PM_ADDR_FLAG_SIGNAL;
 869 }
 870
 871 static int mptcp_pm_nl_append_new_local_addr(struct pm_nl_pernet *pernet,
 872                                              struct mptcp_pm_addr_entry *entry)
 873 {
 874         struct mptcp_pm_addr_entry *cur;
 875         unsigned int addr_max;
 876         int ret = -EINVAL;
 877
 878         spin_lock_bh(&pernet->lock);
 879         /* to keep the code simple, don't do IDR-like allocation for address ID,
 880          * just bail when we exceed limits
 881          */
 882         if (pernet->next_id == MPTCP_PM_MAX_ADDR_ID)
 883                 pernet->next_id = 1;
 884         if (pernet->addrs >= MPTCP_PM_ADDR_MAX)
 885                 goto out;
 886         if (test_bit(entry->addr.id, pernet->id_bitmap))
 887                 goto out;
 888
 889         /* do not insert duplicate address, differentiate on port only
 890          * singled addresses
 891          */
 892         list_for_each_entry(cur, &pernet->local_addr_list, list) {
 893                 if (addresses_equal(&cur->addr, &entry->addr,
 894                                     address_use_port(entry) &&
 895                                     address_use_port(cur)))
 896                         goto out;
 897         }
 898
 899         if (!entry->addr.id) {
 900 find_next:
 901                 entry->addr.id = find_next_zero_bit(pernet->id_bitmap,
 902                                                     MPTCP_PM_MAX_ADDR_ID + 1,
 903                                                     pernet->next_id);
 904                 if (!entry->addr.id && pernet->next_id != 1) {
 905                         pernet->next_id = 1;
 906                         goto find_next;
 907                 }
 908         }
 909
 910         if (!entry->addr.id)
 911                 goto out;
 912
 913         __set_bit(entry->addr.id, pernet->id_bitmap);
 914         if (entry->addr.id > pernet->next_id)
 915                 pernet->next_id = entry->addr.id;
 916
 917         if (entry->flags & MPTCP_PM_ADDR_FLAG_SIGNAL) {
 918                 addr_max = pernet->add_addr_signal_max;
 919                 WRITE_ONCE(pernet->add_addr_signal_max, addr_max + 1);
 920         }
 921         if (entry->flags & MPTCP_PM_ADDR_FLAG_SUBFLOW) {
 922                 addr_max = pernet->local_addr_max;
 923                 WRITE_ONCE(pernet->local_addr_max, addr_max + 1);
 924         }
 925
 926         pernet->addrs++;
 927         list_add_tail_rcu(&entry->list, &pernet->local_addr_list);
 928         ret = entry->addr.id;
 929
 930 out:
 931         spin_unlock_bh(&pernet->lock);
 932         return ret;
 933 }
 934
 935 static int mptcp_pm_nl_create_listen_socket(struct sock *sk,
 936                                             struct mptcp_pm_addr_entry *entry)
 937 {
 938         int addrlen = sizeof(struct sockaddr_in);
 939         struct sockaddr_storage addr;
 940         struct mptcp_sock *msk;
 941         struct socket *ssock;
 942         int backlog = 1024;
 943         int err;
 944
 945         err = sock_create_kern(sock_net(sk), entry->addr.family,
 946                                SOCK_STREAM, IPPROTO_MPTCP, &entry->lsk);
 947         if (err)
 948                 return err;
 949
 950         msk = mptcp_sk(entry->lsk->sk);
 951         if (!msk) {
 952                 err = -EINVAL;
 953                 goto out;
 954         }
 955
 956         ssock = __mptcp_nmpc_socket(msk);
 957         if (!ssock) {
 958                 err = -EINVAL;
 959                 goto out;
 960         }
 961
 962         mptcp_info2sockaddr(&entry->addr, &addr, entry->addr.family);
 963 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
 964         if (entry->addr.family == AF_INET6)
 965                 addrlen = sizeof(struct sockaddr_in6);
 966 #endif
 967         err = kernel_bind(ssock, (struct sockaddr *)&addr, addrlen);
 968         if (err) {
 969                 pr_warn("kernel_bind error, err=%d", err);
 970                 goto out;
 971         }
 972
 973         err = kernel_listen(ssock, backlog);
 974         if (err) {
 975                 pr_warn("kernel_listen error, err=%d", err);
 976                 goto out;
 977         }
 978
 979         return 0;
 980
 981 out:
 982         sock_release(entry->lsk);
 983         return err;
 984 }
 985
 986 int mptcp_pm_nl_get_local_id(struct mptcp_sock *msk, struct sock_common *skc)
 987 {
 988         struct mptcp_pm_addr_entry *entry;
 989         struct mptcp_addr_info skc_local;
 990         struct mptcp_addr_info msk_local;
 991         struct pm_nl_pernet *pernet;
 992         int ret = -1;
 993
 994         if (WARN_ON_ONCE(!msk))
 995                 return -1;
 996
 997         /* The 0 ID mapping is defined by the first subflow, copied into the msk
 998          * addr
 999          */
1000         local_address((struct sock_common *)msk, &msk_local);
1001         local_address((struct sock_common *)skc, &skc_local);
1002         if (addresses_equal(&msk_local, &skc_local, false))
1003                 return 0;
1004
1005         if (address_zero(&skc_local))
1006                 return 0;
1007
1008         pernet = net_generic(sock_net((struct sock *)msk), pm_nl_pernet_id);
1009
1010         rcu_read_lock();
1011         list_for_each_entry_rcu(entry, &pernet->local_addr_list, list) {
1012                 if (addresses_equal(&entry->addr, &skc_local, entry->addr.port)) {
1013                         ret = entry->addr.id;
1014                         break;
1015                 }
1016         }
1017         rcu_read_unlock();
1018         if (ret >= 0)
1019                 return ret;
1020
1021         /* address not found, add to local list */
1022         entry = kmalloc(sizeof(*entry), GFP_ATOMIC);
1023         if (!entry)
1024                 return -ENOMEM;
1025
1026         entry->addr = skc_local;
1027         entry->addr.id = 0;
1028         entry->addr.port = 0;
1029         entry->ifindex = 0;
1030         entry->flags = 0;
1031         entry->lsk = NULL;
1032         ret = mptcp_pm_nl_append_new_local_addr(pernet, entry);
1033         if (ret < 0)
1034                 kfree(entry);
1035
1036         return ret;
1037 }
1038
1039 void mptcp_pm_nl_data_init(struct mptcp_sock *msk)
1040 {
1041         struct mptcp_pm_data *pm = &msk->pm;
1042         bool subflows;
1043
1044         subflows = !!mptcp_pm_get_subflows_max(msk);
1045         WRITE_ONCE(pm->work_pending, (!!mptcp_pm_get_local_addr_max(msk) && subflows) ||
1046                    !!mptcp_pm_get_add_addr_signal_max(msk));
1047         WRITE_ONCE(pm->accept_addr, !!mptcp_pm_get_add_addr_accept_max(msk) && subflows);
1048         WRITE_ONCE(pm->accept_subflow, subflows);
1049 }
1050
1051 #define MPTCP_PM_CMD_GRP_OFFSET       0
1052 #define MPTCP_PM_EV_GRP_OFFSET        1
1053
1054 static const struct genl_multicast_group mptcp_pm_mcgrps[] = {
1055         [MPTCP_PM_CMD_GRP_OFFSET]       = { .name = MPTCP_PM_CMD_GRP_NAME, },
1056         [MPTCP_PM_EV_GRP_OFFSET]        = { .name = MPTCP_PM_EV_GRP_NAME,
1057                                             .flags = GENL_UNS_ADMIN_PERM,
1058                                           },
1059 };
1060
1061 static const struct nla_policy
1062 mptcp_pm_addr_policy[MPTCP_PM_ADDR_ATTR_MAX + 1] = {
1063         [MPTCP_PM_ADDR_ATTR_FAMILY]     = { .type       = NLA_U16,      },
1064         [MPTCP_PM_ADDR_ATTR_ID]         = { .type       = NLA_U8,       },
1065         [MPTCP_PM_ADDR_ATTR_ADDR4]      = { .type       = NLA_U32,      },
1066         [MPTCP_PM_ADDR_ATTR_ADDR6]      =
1067                 NLA_POLICY_EXACT_LEN(sizeof(struct in6_addr)),
1068         [MPTCP_PM_ADDR_ATTR_PORT]       = { .type       = NLA_U16       },
1069         [MPTCP_PM_ADDR_ATTR_FLAGS]      = { .type       = NLA_U32       },
1070         [MPTCP_PM_ADDR_ATTR_IF_IDX]     = { .type       = NLA_S32       },
1071 };
1072
1073 static const struct nla_policy mptcp_pm_policy[MPTCP_PM_ATTR_MAX + 1] = {
1074         [MPTCP_PM_ATTR_ADDR]            =
1075                                         NLA_POLICY_NESTED(mptcp_pm_addr_policy),
1076         [MPTCP_PM_ATTR_RCV_ADD_ADDRS]   = { .type       = NLA_U32,      },
1077         [MPTCP_PM_ATTR_SUBFLOWS]        = { .type       = NLA_U32,      },
1078 };
1079
1080 void mptcp_pm_nl_subflow_chk_stale(const struct mptcp_sock *msk, struct sock *ssk)
1081 {
1082         struct mptcp_subflow_context *iter, *subflow = mptcp_subflow_ctx(ssk);
1083         struct sock *sk = (struct sock *)msk;
1084         unsigned int active_max_loss_cnt;
1085         struct net *net = sock_net(sk);
1086         unsigned int stale_loss_cnt;
1087         bool slow;
1088
1089         stale_loss_cnt = mptcp_stale_loss_cnt(net);
1090         if (subflow->stale || !stale_loss_cnt || subflow->stale_count <= stale_loss_cnt)
1091                 return;
1092
1093         /* look for another available subflow not in loss state */
1094         active_max_loss_cnt = max_t(int, stale_loss_cnt - 1, 1);
1095         mptcp_for_each_subflow(msk, iter) {
1096                 if (iter != subflow && mptcp_subflow_active(iter) &&
1097                     iter->stale_count < active_max_loss_cnt) {
1098                         /* we have some alternatives, try to mark this subflow as idle ...*/
1099                         slow = lock_sock_fast(ssk);
1100                         if (!tcp_rtx_and_write_queues_empty(ssk)) {
1101                                 subflow->stale = 1;
1102                                 __mptcp_retransmit_pending_data(sk);
1103                                 MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_SUBFLOWSTALE);
1104                         }
1105                         unlock_sock_fast(ssk, slow);
1106
1107                         /* always try to push the pending data regarless of re-injections:
1108                          * we can possibly use backup subflows now, and subflow selection
1109                          * is cheap under the msk socket lock
1110                          */
1111                         __mptcp_push_pending(sk, 0);
1112                         return;
1113                 }
1114         }
1115 }
1116
1117 static int mptcp_pm_family_to_addr(int family)
1118 {
1119 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
1120         if (family == AF_INET6)
1121                 return MPTCP_PM_ADDR_ATTR_ADDR6;
1122 #endif
1123         return MPTCP_PM_ADDR_ATTR_ADDR4;
1124 }
1125
1126 static int mptcp_pm_parse_addr(struct nlattr *attr, struct genl_info *info,
1127                                bool require_family,
1128                                struct mptcp_pm_addr_entry *entry)
1129 {
1130         struct nlattr *tb[MPTCP_PM_ADDR_ATTR_MAX + 1];
1131         int err, addr_addr;
1132
1133         if (!attr) {
1134                 GENL_SET_ERR_MSG(info, "missing address info");
1135                 return -EINVAL;
1136         }
1137
1138         /* no validation needed - was already done via nested policy */
1139         err = nla_parse_nested_deprecated(tb, MPTCP_PM_ADDR_ATTR_MAX, attr,
1140                                           mptcp_pm_addr_policy, info->extack);
1141         if (err)
1142                 return err;
1143
1144         memset(entry, 0, sizeof(*entry));
1145         if (!tb[MPTCP_PM_ADDR_ATTR_FAMILY]) {
1146                 if (!require_family)
1147                         goto skip_family;
1148
1149                 NL_SET_ERR_MSG_ATTR(info->extack, attr,
1150                                     "missing family");
1151                 return -EINVAL;
1152         }
1153
1154         entry->addr.family = nla_get_u16(tb[MPTCP_PM_ADDR_ATTR_FAMILY]);
1155         if (entry->addr.family != AF_INET
1156 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
1157             && entry->addr.family != AF_INET6
1158 #endif
1159             ) {
1160                 NL_SET_ERR_MSG_ATTR(info->extack, attr,
1161                                     "unknown address family");
1162                 return -EINVAL;
1163         }
1164         addr_addr = mptcp_pm_family_to_addr(entry->addr.family);
1165         if (!tb[addr_addr]) {
1166                 NL_SET_ERR_MSG_ATTR(info->extack, attr,
1167                                     "missing address data");
1168                 return -EINVAL;
1169         }
1170
1171 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
1172         if (entry->addr.family == AF_INET6)
1173                 entry->addr.addr6 = nla_get_in6_addr(tb[addr_addr]);
1174         else
1175 #endif
1176                 entry->addr.addr.s_addr = nla_get_in_addr(tb[addr_addr]);
1177
1178 skip_family:
1179         if (tb[MPTCP_PM_ADDR_ATTR_IF_IDX]) {
1180                 u32 val = nla_get_s32(tb[MPTCP_PM_ADDR_ATTR_IF_IDX]);
1181
1182                 entry->ifindex = val;
1183         }
1184
1185         if (tb[MPTCP_PM_ADDR_ATTR_ID])
1186                 entry->addr.id = nla_get_u8(tb[MPTCP_PM_ADDR_ATTR_ID]);
1187
1188         if (tb[MPTCP_PM_ADDR_ATTR_FLAGS])
1189                 entry->flags = nla_get_u32(tb[MPTCP_PM_ADDR_ATTR_FLAGS]);
1190
1191         if (tb[MPTCP_PM_ADDR_ATTR_PORT]) {
1192                 if (!(entry->flags & MPTCP_PM_ADDR_FLAG_SIGNAL)) {
1193                         NL_SET_ERR_MSG_ATTR(info->extack, attr,
1194                                             "flags must have signal when using port");
1195                         return -EINVAL;
1196                 }
1197                 entry->addr.port = htons(nla_get_u16(tb[MPTCP_PM_ADDR_ATTR_PORT]));
1198         }
1199
1200         return 0;
1201 }
1202
1203 static struct pm_nl_pernet *genl_info_pm_nl(struct genl_info *info)
1204 {
1205         return net_generic(genl_info_net(info), pm_nl_pernet_id);
1206 }
1207
1208 static int mptcp_nl_add_subflow_or_signal_addr(struct net *net)
1209 {
1210         struct mptcp_sock *msk;
1211         long s_slot = 0, s_num = 0;
1212
1213         while ((msk = mptcp_token_iter_next(net, &s_slot, &s_num)) != NULL) {
1214                 struct sock *sk = (struct sock *)msk;
1215
1216                 if (!READ_ONCE(msk->fully_established))
1217                         goto next;
1218
1219                 lock_sock(sk);
1220                 spin_lock_bh(&msk->pm.lock);
1221                 mptcp_pm_create_subflow_or_signal_addr(msk);
1222                 spin_unlock_bh(&msk->pm.lock);
1223                 release_sock(sk);
1224
1225 next:
1226                 sock_put(sk);
1227                 cond_resched();
1228         }
1229
1230         return 0;
1231 }
1232
1233 static int mptcp_nl_cmd_add_addr(struct sk_buff *skb, struct genl_info *info)
1234 {
1235         struct nlattr *attr = info->attrs[MPTCP_PM_ATTR_ADDR];
1236         struct pm_nl_pernet *pernet = genl_info_pm_nl(info);
1237         struct mptcp_pm_addr_entry addr, *entry;
1238         int ret;
1239
1240         ret = mptcp_pm_parse_addr(attr, info, true, &addr);
1241         if (ret < 0)
1242                 return ret;
1243
1244         entry = kmalloc(sizeof(*entry), GFP_KERNEL);
1245         if (!entry) {
1246                 GENL_SET_ERR_MSG(info, "can't allocate addr");
1247                 return -ENOMEM;
1248         }
1249
1250         *entry = addr;
1251         if (entry->addr.port) {
1252                 ret = mptcp_pm_nl_create_listen_socket(skb->sk, entry);
1253                 if (ret) {
1254                         GENL_SET_ERR_MSG(info, "create listen socket error");
1255                         kfree(entry);
1256                         return ret;
1257                 }
1258         }
1259         ret = mptcp_pm_nl_append_new_local_addr(pernet, entry);
1260         if (ret < 0) {
1261                 GENL_SET_ERR_MSG(info, "too many addresses or duplicate one");
1262                 if (entry->lsk)
1263                         sock_release(entry->lsk);
1264                 kfree(entry);
1265                 return ret;
1266         }
1267
1268         mptcp_nl_add_subflow_or_signal_addr(sock_net(skb->sk));
1269
1270         return 0;
1271 }
1272
1273 int mptcp_pm_get_flags_and_ifindex_by_id(struct net *net, unsigned int id,
1274                                          u8 *flags, int *ifindex)
1275 {
1276         struct mptcp_pm_addr_entry *entry;
1277
1278         *flags = 0;
1279         *ifindex = 0;
1280
1281         if (id) {
1282                 rcu_read_lock();
1283                 entry = __lookup_addr_by_id(net_generic(net, pm_nl_pernet_id), id);
1284                 if (entry) {
1285                         *flags = entry->flags;
1286                         *ifindex = entry->ifindex;
1287                 }
1288                 rcu_read_unlock();
1289         }
1290
1291         return 0;
1292 }
1293
1294 static bool remove_anno_list_by_saddr(struct mptcp_sock *msk,
1295                                       struct mptcp_addr_info *addr)
1296 {
1297         struct mptcp_pm_add_entry *entry;
1298
1299         entry = mptcp_pm_del_add_timer(msk, addr, false);
1300         if (entry) {
1301                 list_del(&entry->list);
1302                 kfree(entry);
1303                 return true;
1304         }
1305
1306         return false;
1307 }
1308
1309 static bool mptcp_pm_remove_anno_addr(struct mptcp_sock *msk,
1310                                       struct mptcp_addr_info *addr,
1311                                       bool force)
1312 {
1313         struct mptcp_rm_list list = { .nr = 0 };
1314         bool ret;
1315
1316         list.ids[list.nr++] = addr->id;
1317
1318         ret = remove_anno_list_by_saddr(msk, addr);
1319         if (ret || force) {
1320                 spin_lock_bh(&msk->pm.lock);
1321                 mptcp_pm_remove_addr(msk, &list);
1322                 spin_unlock_bh(&msk->pm.lock);
1323         }
1324         return ret;
1325 }
1326
1327 static int mptcp_nl_remove_subflow_and_signal_addr(struct net *net,
1328                                                    struct mptcp_addr_info *addr)
1329 {
1330         struct mptcp_sock *msk;
1331         long s_slot = 0, s_num = 0;
1332         struct mptcp_rm_list list = { .nr = 0 };
1333
1334         pr_debug("remove_id=%d", addr->id);
1335
1336         list.ids[list.nr++] = addr->id;
1337
1338         while ((msk = mptcp_token_iter_next(net, &s_slot, &s_num)) != NULL) {
1339                 struct sock *sk = (struct sock *)msk;
1340                 bool remove_subflow;
1341
1342                 if (list_empty(&msk->conn_list)) {
1343                         mptcp_pm_remove_anno_addr(msk, addr, false);
1344                         goto next;
1345                 }
1346
1347                 lock_sock(sk);
1348                 remove_subflow = lookup_subflow_by_saddr(&msk->conn_list, addr);
1349                 mptcp_pm_remove_anno_addr(msk, addr, remove_subflow);
1350                 if (remove_subflow)
1351                         mptcp_pm_remove_subflow(msk, &list);
1352                 release_sock(sk);
1353
1354 next:
1355                 sock_put(sk);
1356                 cond_resched();
1357         }
1358
1359         return 0;
1360 }
1361
1362 /* caller must ensure the RCU grace period is already elapsed */
1363 static void __mptcp_pm_release_addr_entry(struct mptcp_pm_addr_entry *entry)
1364 {
1365         if (entry->lsk)
1366                 sock_release(entry->lsk);
1367         kfree(entry);
1368 }
1369
1370 static int mptcp_nl_remove_id_zero_address(struct net *net,
1371                                            struct mptcp_addr_info *addr)
1372 {
1373         struct mptcp_rm_list list = { .nr = 0 };
1374         long s_slot = 0, s_num = 0;
1375         struct mptcp_sock *msk;
1376
1377         list.ids[list.nr++] = 0;
1378
1379         while ((msk = mptcp_token_iter_next(net, &s_slot, &s_num)) != NULL) {
1380                 struct sock *sk = (struct sock *)msk;
1381                 struct mptcp_addr_info msk_local;
1382
1383                 if (list_empty(&msk->conn_list))
1384                         goto next;
1385
1386                 local_address((struct sock_common *)msk, &msk_local);
1387                 if (!addresses_equal(&msk_local, addr, addr->port))
1388                         goto next;
1389
1390                 lock_sock(sk);
1391                 spin_lock_bh(&msk->pm.lock);
1392                 mptcp_pm_remove_addr(msk, &list);
1393                 mptcp_pm_nl_rm_subflow_received(msk, &list);
1394                 spin_unlock_bh(&msk->pm.lock);
1395                 release_sock(sk);
1396
1397 next:
1398                 sock_put(sk);
1399                 cond_resched();
1400         }
1401
1402         return 0;
1403 }
1404
1405 static int mptcp_nl_cmd_del_addr(struct sk_buff *skb, struct genl_info *info)
1406 {
1407         struct nlattr *attr = info->attrs[MPTCP_PM_ATTR_ADDR];
1408         struct pm_nl_pernet *pernet = genl_info_pm_nl(info);
1409         struct mptcp_pm_addr_entry addr, *entry;
1410         unsigned int addr_max;
1411         int ret;
1412
1413         ret = mptcp_pm_parse_addr(attr, info, false, &addr);
1414         if (ret < 0)
1415                 return ret;
1416
1417         /* the zero id address is special: the first address used by the msk
1418          * always gets such an id, so different subflows can have different zero
1419          * id addresses. Additionally zero id is not accounted for in id_bitmap.
1420          * Let's use an 'mptcp_rm_list' instead of the common remove code.
1421          */
1422         if (addr.addr.id == 0)
1423                 return mptcp_nl_remove_id_zero_address(sock_net(skb->sk), &addr.addr);
1424
1425         spin_lock_bh(&pernet->lock);
1426         entry = __lookup_addr_by_id(pernet, addr.addr.id);
1427         if (!entry) {
1428                 GENL_SET_ERR_MSG(info, "address not found");
1429                 spin_unlock_bh(&pernet->lock);
1430                 return -EINVAL;
1431         }
1432         if (entry->flags & MPTCP_PM_ADDR_FLAG_SIGNAL) {
1433                 addr_max = pernet->add_addr_signal_max;
1434                 WRITE_ONCE(pernet->add_addr_signal_max, addr_max - 1);
1435         }
1436         if (entry->flags & MPTCP_PM_ADDR_FLAG_SUBFLOW) {
1437                 addr_max = pernet->local_addr_max;
1438                 WRITE_ONCE(pernet->local_addr_max, addr_max - 1);
1439         }
1440
1441         pernet->addrs--;
1442         list_del_rcu(&entry->list);
1443         __clear_bit(entry->addr.id, pernet->id_bitmap);
1444         spin_unlock_bh(&pernet->lock);
1445
1446         mptcp_nl_remove_subflow_and_signal_addr(sock_net(skb->sk), &entry->addr);
1447         synchronize_rcu();
1448         __mptcp_pm_release_addr_entry(entry);
1449
1450         return ret;
1451 }
1452
1453 static void mptcp_pm_remove_addrs_and_subflows(struct mptcp_sock *msk,
1454                                                struct list_head *rm_list)
1455 {
1456         struct mptcp_rm_list alist = { .nr = 0 }, slist = { .nr = 0 };
1457         struct mptcp_pm_addr_entry *entry;
1458
1459         list_for_each_entry(entry, rm_list, list) {
1460                 if (lookup_subflow_by_saddr(&msk->conn_list, &entry->addr) &&
1461                     alist.nr < MPTCP_RM_IDS_MAX &&
1462                     slist.nr < MPTCP_RM_IDS_MAX) {
1463                         alist.ids[alist.nr++] = entry->addr.id;
1464                         slist.ids[slist.nr++] = entry->addr.id;
1465                 } else if (remove_anno_list_by_saddr(msk, &entry->addr) &&
1466                          alist.nr < MPTCP_RM_IDS_MAX) {
1467                         alist.ids[alist.nr++] = entry->addr.id;
1468                 }
1469         }
1470
1471         if (alist.nr) {
1472                 spin_lock_bh(&msk->pm.lock);
1473                 mptcp_pm_remove_addr(msk, &alist);
1474                 spin_unlock_bh(&msk->pm.lock);
1475         }
1476         if (slist.nr)
1477                 mptcp_pm_remove_subflow(msk, &slist);
1478 }
1479
1480 static void mptcp_nl_remove_addrs_list(struct net *net,
1481                                        struct list_head *rm_list)
1482 {
1483         long s_slot = 0, s_num = 0;
1484         struct mptcp_sock *msk;
1485
1486         if (list_empty(rm_list))
1487                 return;
1488
1489         while ((msk = mptcp_token_iter_next(net, &s_slot, &s_num)) != NULL) {
1490                 struct sock *sk = (struct sock *)msk;
1491
1492                 lock_sock(sk);
1493                 mptcp_pm_remove_addrs_and_subflows(msk, rm_list);
1494                 release_sock(sk);
1495
1496                 sock_put(sk);
1497                 cond_resched();
1498         }
1499 }
1500
1501 /* caller must ensure the RCU grace period is already elapsed */
1502 static void __flush_addrs(struct list_head *list)
1503 {
1504         while (!list_empty(list)) {
1505                 struct mptcp_pm_addr_entry *cur;
1506
1507                 cur = list_entry(list->next,
1508                                  struct mptcp_pm_addr_entry, list);
1509                 list_del_rcu(&cur->list);
1510                 __mptcp_pm_release_addr_entry(cur);
1511         }
1512 }
1513
1514 static void __reset_counters(struct pm_nl_pernet *pernet)
1515 {
1516         WRITE_ONCE(pernet->add_addr_signal_max, 0);
1517         WRITE_ONCE(pernet->add_addr_accept_max, 0);
1518         WRITE_ONCE(pernet->local_addr_max, 0);
1519         pernet->addrs = 0;
1520 }
1521
1522 static int mptcp_nl_cmd_flush_addrs(struct sk_buff *skb, struct genl_info *info)
1523 {
1524         struct pm_nl_pernet *pernet = genl_info_pm_nl(info);
1525         LIST_HEAD(free_list);
1526
1527         spin_lock_bh(&pernet->lock);
1528         list_splice_init(&pernet->local_addr_list, &free_list);
1529         __reset_counters(pernet);
1530         pernet->next_id = 1;
1531         bitmap_zero(pernet->id_bitmap, MPTCP_PM_MAX_ADDR_ID + 1);
1532         spin_unlock_bh(&pernet->lock);
1533         mptcp_nl_remove_addrs_list(sock_net(skb->sk), &free_list);
1534         synchronize_rcu();
1535         __flush_addrs(&free_list);
1536         return 0;
1537 }
1538
1539 static int mptcp_nl_fill_addr(struct sk_buff *skb,
1540                               struct mptcp_pm_addr_entry *entry)
1541 {
1542         struct mptcp_addr_info *addr = &entry->addr;
1543         struct nlattr *attr;
1544
1545         attr = nla_nest_start(skb, MPTCP_PM_ATTR_ADDR);
1546         if (!attr)
1547                 return -EMSGSIZE;
1548
1549         if (nla_put_u16(skb, MPTCP_PM_ADDR_ATTR_FAMILY, addr->family))
1550                 goto nla_put_failure;
1551         if (nla_put_u16(skb, MPTCP_PM_ADDR_ATTR_PORT, ntohs(addr->port)))
1552                 goto nla_put_failure;
1553         if (nla_put_u8(skb, MPTCP_PM_ADDR_ATTR_ID, addr->id))
1554                 goto nla_put_failure;
1555         if (nla_put_u32(skb, MPTCP_PM_ADDR_ATTR_FLAGS, entry->flags))
1556                 goto nla_put_failure;
1557         if (entry->ifindex &&
1558             nla_put_s32(skb, MPTCP_PM_ADDR_ATTR_IF_IDX, entry->ifindex))
1559                 goto nla_put_failure;
1560
1561         if (addr->family == AF_INET &&
1562             nla_put_in_addr(skb, MPTCP_PM_ADDR_ATTR_ADDR4,
1563                             addr->addr.s_addr))
1564                 goto nla_put_failure;
1565 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
1566         else if (addr->family == AF_INET6 &&
1567                  nla_put_in6_addr(skb, MPTCP_PM_ADDR_ATTR_ADDR6, &addr->addr6))
1568                 goto nla_put_failure;
1569 #endif
1570         nla_nest_end(skb, attr);
1571         return 0;
1572
1573 nla_put_failure:
1574         nla_nest_cancel(skb, attr);
1575         return -EMSGSIZE;
1576 }
1577
1578 static int mptcp_nl_cmd_get_addr(struct sk_buff *skb, struct genl_info *info)
1579 {
1580         struct nlattr *attr = info->attrs[MPTCP_PM_ATTR_ADDR];
1581         struct pm_nl_pernet *pernet = genl_info_pm_nl(info);
1582         struct mptcp_pm_addr_entry addr, *entry;
1583         struct sk_buff *msg;
1584         void *reply;
1585         int ret;
1586
1587         ret = mptcp_pm_parse_addr(attr, info, false, &addr);
1588         if (ret < 0)
1589                 return ret;
1590
1591         msg = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
1592         if (!msg)
1593                 return -ENOMEM;
1594
1595         reply = genlmsg_put_reply(msg, info, &mptcp_genl_family, 0,
1596                                   info->genlhdr->cmd);
1597         if (!reply) {
1598                 GENL_SET_ERR_MSG(info, "not enough space in Netlink message");
1599                 ret = -EMSGSIZE;
1600                 goto fail;
1601         }
1602
1603         spin_lock_bh(&pernet->lock);
1604         entry = __lookup_addr_by_id(pernet, addr.addr.id);
1605         if (!entry) {
1606                 GENL_SET_ERR_MSG(info, "address not found");
1607                 ret = -EINVAL;
1608                 goto unlock_fail;
1609         }
1610
1611         ret = mptcp_nl_fill_addr(msg, entry);
1612         if (ret)
1613                 goto unlock_fail;
1614
1615         genlmsg_end(msg, reply);
1616         ret = genlmsg_reply(msg, info);
1617         spin_unlock_bh(&pernet->lock);
1618         return ret;
1619
1620 unlock_fail:
1621         spin_unlock_bh(&pernet->lock);
1622
1623 fail:
1624         nlmsg_free(msg);
1625         return ret;
1626 }
1627
1628 static int mptcp_nl_cmd_dump_addrs(struct sk_buff *msg,
1629                                    struct netlink_callback *cb)
1630 {
1631         struct net *net = sock_net(msg->sk);
1632         struct mptcp_pm_addr_entry *entry;
1633         struct pm_nl_pernet *pernet;
1634         int id = cb->args[0];
1635         void *hdr;
1636         int i;
1637
1638         pernet = net_generic(net, pm_nl_pernet_id);
1639
1640         spin_lock_bh(&pernet->lock);
1641         for (i = id; i < MPTCP_PM_MAX_ADDR_ID + 1; i++) {
1642                 if (test_bit(i, pernet->id_bitmap)) {
1643                         entry = __lookup_addr_by_id(pernet, i);
1644                         if (!entry)
1645                                 break;
1646
1647                         if (entry->addr.id <= id)
1648                                 continue;
1649
1650                         hdr = genlmsg_put(msg, NETLINK_CB(cb->skb).portid,
1651                                           cb->nlh->nlmsg_seq, &mptcp_genl_family,
1652                                           NLM_F_MULTI, MPTCP_PM_CMD_GET_ADDR);
1653                         if (!hdr)
1654                                 break;
1655
1656                         if (mptcp_nl_fill_addr(msg, entry) < 0) {
1657                                 genlmsg_cancel(msg, hdr);
1658                                 break;
1659                         }
1660
1661                         id = entry->addr.id;
1662                         genlmsg_end(msg, hdr);
1663                 }
1664         }
1665         spin_unlock_bh(&pernet->lock);
1666
1667         cb->args[0] = id;
1668         return msg->len;
1669 }
1670
1671 static int parse_limit(struct genl_info *info, int id, unsigned int *limit)
1672 {
1673         struct nlattr *attr = info->attrs[id];
1674
1675         if (!attr)
1676                 return 0;
1677
1678         *limit = nla_get_u32(attr);
1679         if (*limit > MPTCP_PM_ADDR_MAX) {
1680                 GENL_SET_ERR_MSG(info, "limit greater than maximum");
1681                 return -EINVAL;
1682         }
1683         return 0;
1684 }
1685
1686 static int
1687 mptcp_nl_cmd_set_limits(struct sk_buff *skb, struct genl_info *info)
1688 {
1689         struct pm_nl_pernet *pernet = genl_info_pm_nl(info);
1690         unsigned int rcv_addrs, subflows;
1691         int ret;
1692
1693         spin_lock_bh(&pernet->lock);
1694         rcv_addrs = pernet->add_addr_accept_max;
1695         ret = parse_limit(info, MPTCP_PM_ATTR_RCV_ADD_ADDRS, &rcv_addrs);
1696         if (ret)
1697                 goto unlock;
1698
1699         subflows = pernet->subflows_max;
1700         ret = parse_limit(info, MPTCP_PM_ATTR_SUBFLOWS, &subflows);
1701         if (ret)
1702                 goto unlock;
1703
1704         WRITE_ONCE(pernet->add_addr_accept_max, rcv_addrs);
1705         WRITE_ONCE(pernet->subflows_max, subflows);
1706
1707 unlock:
1708         spin_unlock_bh(&pernet->lock);
1709         return ret;
1710 }
1711
1712 static int
1713 mptcp_nl_cmd_get_limits(struct sk_buff *skb, struct genl_info *info)
1714 {
1715         struct pm_nl_pernet *pernet = genl_info_pm_nl(info);
1716         struct sk_buff *msg;
1717         void *reply;
1718
1719         msg = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
1720         if (!msg)
1721                 return -ENOMEM;
1722
1723         reply = genlmsg_put_reply(msg, info, &mptcp_genl_family, 0,
1724                                   MPTCP_PM_CMD_GET_LIMITS);
1725         if (!reply)
1726                 goto fail;
1727
1728         if (nla_put_u32(msg, MPTCP_PM_ATTR_RCV_ADD_ADDRS,
1729                         READ_ONCE(pernet->add_addr_accept_max)))
1730                 goto fail;
1731
1732         if (nla_put_u32(msg, MPTCP_PM_ATTR_SUBFLOWS,
1733                         READ_ONCE(pernet->subflows_max)))
1734                 goto fail;
1735
1736         genlmsg_end(msg, reply);
1737         return genlmsg_reply(msg, info);
1738
1739 fail:
1740         GENL_SET_ERR_MSG(info, "not enough space in Netlink message");
1741         nlmsg_free(msg);
1742         return -EMSGSIZE;
1743 }
1744
1745 static int mptcp_nl_addr_backup(struct net *net,
1746                                 struct mptcp_addr_info *addr,
1747                                 u8 bkup)
1748 {
1749         long s_slot = 0, s_num = 0;
1750         struct mptcp_sock *msk;
1751         int ret = -EINVAL;
1752
1753         while ((msk = mptcp_token_iter_next(net, &s_slot, &s_num)) != NULL) {
1754                 struct sock *sk = (struct sock *)msk;
1755
1756                 if (list_empty(&msk->conn_list))
1757                         goto next;
1758
1759                 lock_sock(sk);
1760                 spin_lock_bh(&msk->pm.lock);
1761                 ret = mptcp_pm_nl_mp_prio_send_ack(msk, addr, bkup);
1762                 spin_unlock_bh(&msk->pm.lock);
1763                 release_sock(sk);
1764
1765 next:
1766                 sock_put(sk);
1767                 cond_resched();
1768         }
1769
1770         return ret;
1771 }
1772
1773 static int mptcp_nl_cmd_set_flags(struct sk_buff *skb, struct genl_info *info)
1774 {
1775         struct mptcp_pm_addr_entry addr = { .addr = { .family = AF_UNSPEC }, }, *entry;
1776         struct nlattr *attr = info->attrs[MPTCP_PM_ATTR_ADDR];
1777         struct pm_nl_pernet *pernet = genl_info_pm_nl(info);
1778         struct net *net = sock_net(skb->sk);
1779         u8 bkup = 0, lookup_by_id = 0;
1780         int ret;
1781
1782         ret = mptcp_pm_parse_addr(attr, info, false, &addr);
1783         if (ret < 0)
1784                 return ret;
1785
1786         if (addr.flags & MPTCP_PM_ADDR_FLAG_BACKUP)
1787                 bkup = 1;
1788         if (addr.addr.family == AF_UNSPEC) {
1789                 lookup_by_id = 1;
1790                 if (!addr.addr.id)
1791                         return -EOPNOTSUPP;
1792         }
1793
1794         spin_lock_bh(&pernet->lock);
1795         entry = __lookup_addr(pernet, &addr.addr, lookup_by_id);
1796         if (!entry) {
1797                 spin_unlock_bh(&pernet->lock);
1798                 return -EINVAL;
1799         }
1800
1801         if (bkup)
1802                 entry->flags |= MPTCP_PM_ADDR_FLAG_BACKUP;
1803         else
1804                 entry->flags &= ~MPTCP_PM_ADDR_FLAG_BACKUP;
1805         addr = *entry;
1806         spin_unlock_bh(&pernet->lock);
1807
1808         mptcp_nl_addr_backup(net, &addr.addr, bkup);
1809         return 0;
1810 }
1811
1812 static void mptcp_nl_mcast_send(struct net *net, struct sk_buff *nlskb, gfp_t gfp)
1813 {
1814         genlmsg_multicast_netns(&mptcp_genl_family, net,
1815                                 nlskb, 0, MPTCP_PM_EV_GRP_OFFSET, gfp);
1816 }
1817
1818 static int mptcp_event_add_subflow(struct sk_buff *skb, const struct sock *ssk)
1819 {
1820         const struct inet_sock *issk = inet_sk(ssk);
1821         const struct mptcp_subflow_context *sf;
1822
1823         if (nla_put_u16(skb, MPTCP_ATTR_FAMILY, ssk->sk_family))
1824                 return -EMSGSIZE;
1825
1826         switch (ssk->sk_family) {
1827         case AF_INET:
1828                 if (nla_put_in_addr(skb, MPTCP_ATTR_SADDR4, issk->inet_saddr))
1829                         return -EMSGSIZE;
1830                 if (nla_put_in_addr(skb, MPTCP_ATTR_DADDR4, issk->inet_daddr))
1831                         return -EMSGSIZE;
1832                 break;
1833 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
1834         case AF_INET6: {
1835                 const struct ipv6_pinfo *np = inet6_sk(ssk);
1836
1837                 if (nla_put_in6_addr(skb, MPTCP_ATTR_SADDR6, &np->saddr))
1838                         return -EMSGSIZE;
1839                 if (nla_put_in6_addr(skb, MPTCP_ATTR_DADDR6, &ssk->sk_v6_daddr))
1840                         return -EMSGSIZE;
1841                 break;
1842         }
1843 #endif
1844         default:
1845                 WARN_ON_ONCE(1);
1846                 return -EMSGSIZE;
1847         }
1848
1849         if (nla_put_be16(skb, MPTCP_ATTR_SPORT, issk->inet_sport))
1850                 return -EMSGSIZE;
1851         if (nla_put_be16(skb, MPTCP_ATTR_DPORT, issk->inet_dport))
1852                 return -EMSGSIZE;
1853
1854         sf = mptcp_subflow_ctx(ssk);
1855         if (WARN_ON_ONCE(!sf))
1856                 return -EINVAL;
1857
1858         if (nla_put_u8(skb, MPTCP_ATTR_LOC_ID, sf->local_id))
1859                 return -EMSGSIZE;
1860
1861         if (nla_put_u8(skb, MPTCP_ATTR_REM_ID, sf->remote_id))
1862                 return -EMSGSIZE;
1863
1864         return 0;
1865 }
1866
1867 static int mptcp_event_put_token_and_ssk(struct sk_buff *skb,
1868                                          const struct mptcp_sock *msk,
1869                                          const struct sock *ssk)
1870 {
1871         const struct sock *sk = (const struct sock *)msk;
1872         const struct mptcp_subflow_context *sf;
1873         u8 sk_err;
1874
1875         if (nla_put_u32(skb, MPTCP_ATTR_TOKEN, msk->token))
1876                 return -EMSGSIZE;
1877
1878         if (mptcp_event_add_subflow(skb, ssk))
1879                 return -EMSGSIZE;
1880
1881         sf = mptcp_subflow_ctx(ssk);
1882         if (WARN_ON_ONCE(!sf))
1883                 return -EINVAL;
1884
1885         if (nla_put_u8(skb, MPTCP_ATTR_BACKUP, sf->backup))
1886                 return -EMSGSIZE;
1887
1888         if (ssk->sk_bound_dev_if &&
1889             nla_put_s32(skb, MPTCP_ATTR_IF_IDX, ssk->sk_bound_dev_if))
1890                 return -EMSGSIZE;
1891
1892         sk_err = ssk->sk_err;
1893         if (sk_err && sk->sk_state == TCP_ESTABLISHED &&
1894             nla_put_u8(skb, MPTCP_ATTR_ERROR, sk_err))
1895                 return -EMSGSIZE;
1896
1897         return 0;
1898 }
1899
1900 static int mptcp_event_sub_established(struct sk_buff *skb,
1901                                        const struct mptcp_sock *msk,
1902                                        const struct sock *ssk)
1903 {
1904         return mptcp_event_put_token_and_ssk(skb, msk, ssk);
1905 }
1906
1907 static int mptcp_event_sub_closed(struct sk_buff *skb,
1908                                   const struct mptcp_sock *msk,
1909                                   const struct sock *ssk)
1910 {
1911         const struct mptcp_subflow_context *sf;
1912
1913         if (mptcp_event_put_token_and_ssk(skb, msk, ssk))
1914                 return -EMSGSIZE;
1915
1916         sf = mptcp_subflow_ctx(ssk);
1917         if (!sf->reset_seen)
1918                 return 0;
1919
1920         if (nla_put_u32(skb, MPTCP_ATTR_RESET_REASON, sf->reset_reason))
1921                 return -EMSGSIZE;
1922
1923         if (nla_put_u32(skb, MPTCP_ATTR_RESET_FLAGS, sf->reset_transient))
1924                 return -EMSGSIZE;
1925
1926         return 0;
1927 }
1928
1929 static int mptcp_event_created(struct sk_buff *skb,
1930                                const struct mptcp_sock *msk,
1931                                const struct sock *ssk)
1932 {
1933         int err = nla_put_u32(skb, MPTCP_ATTR_TOKEN, msk->token);
1934
1935         if (err)
1936                 return err;
1937
1938         return mptcp_event_add_subflow(skb, ssk);
1939 }
1940
1941 void mptcp_event_addr_removed(const struct mptcp_sock *msk, uint8_t id)
1942 {
1943         struct net *net = sock_net((const struct sock *)msk);
1944         struct nlmsghdr *nlh;
1945         struct sk_buff *skb;
1946
1947         if (!genl_has_listeners(&mptcp_genl_family, net, MPTCP_PM_EV_GRP_OFFSET))
1948                 return;
1949
1950         skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_ATOMIC);
1951         if (!skb)
1952                 return;
1953
1954         nlh = genlmsg_put(skb, 0, 0, &mptcp_genl_family, 0, MPTCP_EVENT_REMOVED);
1955         if (!nlh)
1956                 goto nla_put_failure;
1957
1958         if (nla_put_u32(skb, MPTCP_ATTR_TOKEN, msk->token))
1959                 goto nla_put_failure;
1960
1961         if (nla_put_u8(skb, MPTCP_ATTR_REM_ID, id))
1962                 goto nla_put_failure;
1963
1964         genlmsg_end(skb, nlh);
1965         mptcp_nl_mcast_send(net, skb, GFP_ATOMIC);
1966         return;
1967
1968 nla_put_failure:
1969         kfree_skb(skb);
1970 }
1971
1972 void mptcp_event_addr_announced(const struct mptcp_sock *msk,
1973                                 const struct mptcp_addr_info *info)
1974 {
1975         struct net *net = sock_net((const struct sock *)msk);
1976         struct nlmsghdr *nlh;
1977         struct sk_buff *skb;
1978
1979         if (!genl_has_listeners(&mptcp_genl_family, net, MPTCP_PM_EV_GRP_OFFSET))
1980                 return;
1981
1982         skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_ATOMIC);
1983         if (!skb)
1984                 return;
1985
1986         nlh = genlmsg_put(skb, 0, 0, &mptcp_genl_family, 0,
1987                           MPTCP_EVENT_ANNOUNCED);
1988         if (!nlh)
1989                 goto nla_put_failure;
1990
1991         if (nla_put_u32(skb, MPTCP_ATTR_TOKEN, msk->token))
1992                 goto nla_put_failure;
1993
1994         if (nla_put_u8(skb, MPTCP_ATTR_REM_ID, info->id))
1995                 goto nla_put_failure;
1996
1997         if (nla_put_be16(skb, MPTCP_ATTR_DPORT, info->port))
1998                 goto nla_put_failure;
1999
2000         switch (info->family) {
2001         case AF_INET:
2002                 if (nla_put_in_addr(skb, MPTCP_ATTR_DADDR4, info->addr.s_addr))
2003                         goto nla_put_failure;
2004                 break;
2005 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
2006         case AF_INET6:
2007                 if (nla_put_in6_addr(skb, MPTCP_ATTR_DADDR6, &info->addr6))
2008                         goto nla_put_failure;
2009                 break;
2010 #endif
2011         default:
2012                 WARN_ON_ONCE(1);
2013                 goto nla_put_failure;
2014         }
2015
2016         genlmsg_end(skb, nlh);
2017         mptcp_nl_mcast_send(net, skb, GFP_ATOMIC);
2018         return;
2019
2020 nla_put_failure:
2021         kfree_skb(skb);
2022 }
2023
2024 void mptcp_event(enum mptcp_event_type type, const struct mptcp_sock *msk,
2025                  const struct sock *ssk, gfp_t gfp)
2026 {
2027         struct net *net = sock_net((const struct sock *)msk);
2028         struct nlmsghdr *nlh;
2029         struct sk_buff *skb;
2030
2031         if (!genl_has_listeners(&mptcp_genl_family, net, MPTCP_PM_EV_GRP_OFFSET))
2032                 return;
2033
2034         skb = nlmsg_new(NLMSG_DEFAULT_SIZE, gfp);
2035         if (!skb)
2036                 return;
2037
2038         nlh = genlmsg_put(skb, 0, 0, &mptcp_genl_family, 0, type);
2039         if (!nlh)
2040                 goto nla_put_failure;
2041
2042         switch (type) {
2043         case MPTCP_EVENT_UNSPEC:
2044                 WARN_ON_ONCE(1);
2045                 break;
2046         case MPTCP_EVENT_CREATED:
2047         case MPTCP_EVENT_ESTABLISHED:
2048                 if (mptcp_event_created(skb, msk, ssk) < 0)
2049                         goto nla_put_failure;
2050                 break;
2051         case MPTCP_EVENT_CLOSED:
2052                 if (nla_put_u32(skb, MPTCP_ATTR_TOKEN, msk->token) < 0)
2053                         goto nla_put_failure;
2054                 break;
2055         case MPTCP_EVENT_ANNOUNCED:
2056         case MPTCP_EVENT_REMOVED:
2057                 /* call mptcp_event_addr_announced()/removed instead */
2058                 WARN_ON_ONCE(1);
2059                 break;
2060         case MPTCP_EVENT_SUB_ESTABLISHED:
2061         case MPTCP_EVENT_SUB_PRIORITY:
2062                 if (mptcp_event_sub_established(skb, msk, ssk) < 0)
2063                         goto nla_put_failure;
2064                 break;
2065         case MPTCP_EVENT_SUB_CLOSED:
2066                 if (mptcp_event_sub_closed(skb, msk, ssk) < 0)
2067                         goto nla_put_failure;
2068                 break;
2069         }
2070
2071         genlmsg_end(skb, nlh);
2072         mptcp_nl_mcast_send(net, skb, gfp);
2073         return;
2074
2075 nla_put_failure:
2076         kfree_skb(skb);
2077 }
2078
2079 static const struct genl_small_ops mptcp_pm_ops[] = {
2080         {
2081                 .cmd    = MPTCP_PM_CMD_ADD_ADDR,
2082                 .doit   = mptcp_nl_cmd_add_addr,
2083                 .flags  = GENL_ADMIN_PERM,
2084         },
2085         {
2086                 .cmd    = MPTCP_PM_CMD_DEL_ADDR,
2087                 .doit   = mptcp_nl_cmd_del_addr,
2088                 .flags  = GENL_ADMIN_PERM,
2089         },
2090         {
2091                 .cmd    = MPTCP_PM_CMD_FLUSH_ADDRS,
2092                 .doit   = mptcp_nl_cmd_flush_addrs,
2093                 .flags  = GENL_ADMIN_PERM,
2094         },
2095         {
2096                 .cmd    = MPTCP_PM_CMD_GET_ADDR,
2097                 .doit   = mptcp_nl_cmd_get_addr,
2098                 .dumpit   = mptcp_nl_cmd_dump_addrs,
2099         },
2100         {
2101                 .cmd    = MPTCP_PM_CMD_SET_LIMITS,
2102                 .doit   = mptcp_nl_cmd_set_limits,
2103                 .flags  = GENL_ADMIN_PERM,
2104         },
2105         {
2106                 .cmd    = MPTCP_PM_CMD_GET_LIMITS,
2107                 .doit   = mptcp_nl_cmd_get_limits,
2108         },
2109         {
2110                 .cmd    = MPTCP_PM_CMD_SET_FLAGS,
2111                 .doit   = mptcp_nl_cmd_set_flags,
2112                 .flags  = GENL_ADMIN_PERM,
2113         },
2114 };
2115
2116 static struct genl_family mptcp_genl_family __ro_after_init = {
2117         .name           = MPTCP_PM_NAME,
2118         .version        = MPTCP_PM_VER,
2119         .maxattr        = MPTCP_PM_ATTR_MAX,
2120         .policy         = mptcp_pm_policy,
2121         .netnsok        = true,
2122         .module         = THIS_MODULE,
2123         .small_ops      = mptcp_pm_ops,
2124         .n_small_ops    = ARRAY_SIZE(mptcp_pm_ops),
2125         .mcgrps         = mptcp_pm_mcgrps,
2126         .n_mcgrps       = ARRAY_SIZE(mptcp_pm_mcgrps),
2127 };
2128
2129 static int __net_init pm_nl_init_net(struct net *net)
2130 {
2131         struct pm_nl_pernet *pernet = net_generic(net, pm_nl_pernet_id);
2132
2133         INIT_LIST_HEAD_RCU(&pernet->local_addr_list);
2134
2135         /* Cit. 2 subflows ought to be enough for anybody. */
2136         pernet->subflows_max = 2;
2137         pernet->next_id = 1;
2138         pernet->stale_loss_cnt = 4;
2139         spin_lock_init(&pernet->lock);
2140
2141         /* No need to initialize other pernet fields, the struct is zeroed at
2142          * allocation time.
2143          */
2144
2145         return 0;
2146 }
2147
2148 static void __net_exit pm_nl_exit_net(struct list_head *net_list)
2149 {
2150         struct net *net;
2151
2152         list_for_each_entry(net, net_list, exit_list) {
2153                 struct pm_nl_pernet *pernet = net_generic(net, pm_nl_pernet_id);
2154
2155                 /* net is removed from namespace list, can't race with
2156                  * other modifiers, also netns core already waited for a
2157                  * RCU grace period.
2158                  */
2159                 __flush_addrs(&pernet->local_addr_list);
2160         }
2161 }
2162
2163 static struct pernet_operations mptcp_pm_pernet_ops = {
2164         .init = pm_nl_init_net,
2165         .exit_batch = pm_nl_exit_net,
2166         .id = &pm_nl_pernet_id,
2167         .size = sizeof(struct pm_nl_pernet),
2168 };
2169
2170 void __init mptcp_pm_nl_init(void)
2171 {
2172         if (register_pernet_subsys(&mptcp_pm_pernet_ops) < 0)
2173                 panic("Failed to register MPTCP PM pernet subsystem.\n");
2174
2175         if (genl_register_family(&mptcp_genl_family))
2176                 panic("Failed to register MPTCP PM netlink family\n");
2177 }