net/mptcp/pm_netlink.c

   1 // SPDX-License-Identifier: GPL-2.0
   2 /* Multipath TCP
   3  *
   4  * Copyright (c) 2020, Red Hat, Inc.
   5  */
   6
   7 #define pr_fmt(fmt) "MPTCP: " fmt
   8
   9 #include <linux/inet.h>
  10 #include <linux/kernel.h>
  11 #include <net/tcp.h>
  12 #include <net/netns/generic.h>
  13 #include <net/mptcp.h>
  14 #include <net/genetlink.h>
  15 #include <uapi/linux/mptcp.h>
  16
  17 #include "protocol.h"
  18 #include "mib.h"
  19
  20 /* forward declaration */
  21 static struct genl_family mptcp_genl_family;
  22
  23 static int pm_nl_pernet_id;
  24
  25 struct mptcp_pm_add_entry {
  26         struct list_head        list;
  27         struct mptcp_addr_info  addr;
  28         struct timer_list       add_timer;
  29         struct mptcp_sock       *sock;
  30         u8                      retrans_times;
  31 };
  32
  33 struct pm_nl_pernet {
  34         /* protects pernet updates */
  35         spinlock_t              lock;
  36         struct list_head        local_addr_list;
  37         unsigned int            addrs;
  38         unsigned int            stale_loss_cnt;
  39         unsigned int            add_addr_signal_max;
  40         unsigned int            add_addr_accept_max;
  41         unsigned int            local_addr_max;
  42         unsigned int            subflows_max;
  43         unsigned int            next_id;
  44         DECLARE_BITMAP(id_bitmap, MPTCP_PM_MAX_ADDR_ID + 1);
  45 };
  46
  47 #define MPTCP_PM_ADDR_MAX       8
  48 #define ADD_ADDR_RETRANS_MAX    3
  49
  50 static struct pm_nl_pernet *pm_nl_get_pernet(const struct net *net)
  51 {
  52         return net_generic(net, pm_nl_pernet_id);
  53 }
  54
  55 static struct pm_nl_pernet *
  56 pm_nl_get_pernet_from_msk(const struct mptcp_sock *msk)
  57 {
  58         return pm_nl_get_pernet(sock_net((struct sock *)msk));
  59 }
  60
  61 bool mptcp_addresses_equal(const struct mptcp_addr_info *a,
  62                            const struct mptcp_addr_info *b, bool use_port)
  63 {
  64         bool addr_equals = false;
  65
  66         if (a->family == b->family) {
  67                 if (a->family == AF_INET)
  68                         addr_equals = a->addr.s_addr == b->addr.s_addr;
  69 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
  70                 else
  71                         addr_equals = !ipv6_addr_cmp(&a->addr6, &b->addr6);
  72         } else if (a->family == AF_INET) {
  73                 if (ipv6_addr_v4mapped(&b->addr6))
  74                         addr_equals = a->addr.s_addr == b->addr6.s6_addr32[3];
  75         } else if (b->family == AF_INET) {
  76                 if (ipv6_addr_v4mapped(&a->addr6))
  77                         addr_equals = a->addr6.s6_addr32[3] == b->addr.s_addr;
  78 #endif
  79         }
  80
  81         if (!addr_equals)
  82                 return false;
  83         if (!use_port)
  84                 return true;
  85
  86         return a->port == b->port;
  87 }
  88
  89 static void local_address(const struct sock_common *skc,
  90                           struct mptcp_addr_info *addr)
  91 {
  92         addr->family = skc->skc_family;
  93         addr->port = htons(skc->skc_num);
  94         if (addr->family == AF_INET)
  95                 addr->addr.s_addr = skc->skc_rcv_saddr;
  96 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
  97         else if (addr->family == AF_INET6)
  98                 addr->addr6 = skc->skc_v6_rcv_saddr;
  99 #endif
 100 }
 101
 102 static void remote_address(const struct sock_common *skc,
 103                            struct mptcp_addr_info *addr)
 104 {
 105         addr->family = skc->skc_family;
 106         addr->port = skc->skc_dport;
 107         if (addr->family == AF_INET)
 108                 addr->addr.s_addr = skc->skc_daddr;
 109 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
 110         else if (addr->family == AF_INET6)
 111                 addr->addr6 = skc->skc_v6_daddr;
 112 #endif
 113 }
 114
 115 static bool lookup_subflow_by_saddr(const struct list_head *list,
 116                                     const struct mptcp_addr_info *saddr)
 117 {
 118         struct mptcp_subflow_context *subflow;
 119         struct mptcp_addr_info cur;
 120         struct sock_common *skc;
 121
 122         list_for_each_entry(subflow, list, node) {
 123                 skc = (struct sock_common *)mptcp_subflow_tcp_sock(subflow);
 124
 125                 local_address(skc, &cur);
 126                 if (mptcp_addresses_equal(&cur, saddr, saddr->port))
 127                         return true;
 128         }
 129
 130         return false;
 131 }
 132
 133 static bool lookup_subflow_by_daddr(const struct list_head *list,
 134                                     const struct mptcp_addr_info *daddr)
 135 {
 136         struct mptcp_subflow_context *subflow;
 137         struct mptcp_addr_info cur;
 138         struct sock_common *skc;
 139
 140         list_for_each_entry(subflow, list, node) {
 141                 skc = (struct sock_common *)mptcp_subflow_tcp_sock(subflow);
 142
 143                 remote_address(skc, &cur);
 144                 if (mptcp_addresses_equal(&cur, daddr, daddr->port))
 145                         return true;
 146         }
 147
 148         return false;
 149 }
 150
 151 static struct mptcp_pm_addr_entry *
 152 select_local_address(const struct pm_nl_pernet *pernet,
 153                      const struct mptcp_sock *msk)
 154 {
 155         const struct sock *sk = (const struct sock *)msk;
 156         struct mptcp_pm_addr_entry *entry, *ret = NULL;
 157
 158         msk_owned_by_me(msk);
 159
 160         rcu_read_lock();
 161         list_for_each_entry_rcu(entry, &pernet->local_addr_list, list) {
 162                 if (!(entry->flags & MPTCP_PM_ADDR_FLAG_SUBFLOW))
 163                         continue;
 164
 165                 if (!test_bit(entry->addr.id, msk->pm.id_avail_bitmap))
 166                         continue;
 167
 168                 if (entry->addr.family != sk->sk_family) {
 169 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
 170                         if ((entry->addr.family == AF_INET &&
 171                              !ipv6_addr_v4mapped(&sk->sk_v6_daddr)) ||
 172                             (sk->sk_family == AF_INET &&
 173                              !ipv6_addr_v4mapped(&entry->addr.addr6)))
 174 #endif
 175                                 continue;
 176                 }
 177
 178                 ret = entry;
 179                 break;
 180         }
 181         rcu_read_unlock();
 182         return ret;
 183 }
 184
 185 static struct mptcp_pm_addr_entry *
 186 select_signal_address(struct pm_nl_pernet *pernet, const struct mptcp_sock *msk)
 187 {
 188         struct mptcp_pm_addr_entry *entry, *ret = NULL;
 189
 190         rcu_read_lock();
 191         /* do not keep any additional per socket state, just signal
 192          * the address list in order.
 193          * Note: removal from the local address list during the msk life-cycle
 194          * can lead to additional addresses not being announced.
 195          */
 196         list_for_each_entry_rcu(entry, &pernet->local_addr_list, list) {
 197                 if (!test_bit(entry->addr.id, msk->pm.id_avail_bitmap))
 198                         continue;
 199
 200                 if (!(entry->flags & MPTCP_PM_ADDR_FLAG_SIGNAL))
 201                         continue;
 202
 203                 ret = entry;
 204                 break;
 205         }
 206         rcu_read_unlock();
 207         return ret;
 208 }
 209
 210 unsigned int mptcp_pm_get_add_addr_signal_max(const struct mptcp_sock *msk)
 211 {
 212         const struct pm_nl_pernet *pernet = pm_nl_get_pernet_from_msk(msk);
 213
 214         return READ_ONCE(pernet->add_addr_signal_max);
 215 }
 216 EXPORT_SYMBOL_GPL(mptcp_pm_get_add_addr_signal_max);
 217
 218 unsigned int mptcp_pm_get_add_addr_accept_max(const struct mptcp_sock *msk)
 219 {
 220         struct pm_nl_pernet *pernet = pm_nl_get_pernet_from_msk(msk);
 221
 222         return READ_ONCE(pernet->add_addr_accept_max);
 223 }
 224 EXPORT_SYMBOL_GPL(mptcp_pm_get_add_addr_accept_max);
 225
 226 unsigned int mptcp_pm_get_subflows_max(const struct mptcp_sock *msk)
 227 {
 228         struct pm_nl_pernet *pernet = pm_nl_get_pernet_from_msk(msk);
 229
 230         return READ_ONCE(pernet->subflows_max);
 231 }
 232 EXPORT_SYMBOL_GPL(mptcp_pm_get_subflows_max);
 233
 234 unsigned int mptcp_pm_get_local_addr_max(const struct mptcp_sock *msk)
 235 {
 236         struct pm_nl_pernet *pernet = pm_nl_get_pernet_from_msk(msk);
 237
 238         return READ_ONCE(pernet->local_addr_max);
 239 }
 240 EXPORT_SYMBOL_GPL(mptcp_pm_get_local_addr_max);
 241
 242 bool mptcp_pm_nl_check_work_pending(struct mptcp_sock *msk)
 243 {
 244         struct pm_nl_pernet *pernet = pm_nl_get_pernet_from_msk(msk);
 245
 246         if (msk->pm.subflows == mptcp_pm_get_subflows_max(msk) ||
 247             (find_next_and_bit(pernet->id_bitmap, msk->pm.id_avail_bitmap,
 248                                MPTCP_PM_MAX_ADDR_ID + 1, 0) == MPTCP_PM_MAX_ADDR_ID + 1)) {
 249                 WRITE_ONCE(msk->pm.work_pending, false);
 250                 return false;
 251         }
 252         return true;
 253 }
 254
 255 struct mptcp_pm_add_entry *
 256 mptcp_lookup_anno_list_by_saddr(const struct mptcp_sock *msk,
 257                                 const struct mptcp_addr_info *addr)
 258 {
 259         struct mptcp_pm_add_entry *entry;
 260
 261         lockdep_assert_held(&msk->pm.lock);
 262
 263         list_for_each_entry(entry, &msk->pm.anno_list, list) {
 264                 if (mptcp_addresses_equal(&entry->addr, addr, true))
 265                         return entry;
 266         }
 267
 268         return NULL;
 269 }
 270
 271 bool mptcp_pm_sport_in_anno_list(struct mptcp_sock *msk, const struct sock *sk)
 272 {
 273         struct mptcp_pm_add_entry *entry;
 274         struct mptcp_addr_info saddr;
 275         bool ret = false;
 276
 277         local_address((struct sock_common *)sk, &saddr);
 278
 279         spin_lock_bh(&msk->pm.lock);
 280         list_for_each_entry(entry, &msk->pm.anno_list, list) {
 281                 if (mptcp_addresses_equal(&entry->addr, &saddr, true)) {
 282                         ret = true;
 283                         goto out;
 284                 }
 285         }
 286
 287 out:
 288         spin_unlock_bh(&msk->pm.lock);
 289         return ret;
 290 }
 291
 292 static void mptcp_pm_add_timer(struct timer_list *timer)
 293 {
 294         struct mptcp_pm_add_entry *entry = from_timer(entry, timer, add_timer);
 295         struct mptcp_sock *msk = entry->sock;
 296         struct sock *sk = (struct sock *)msk;
 297
 298         pr_debug("msk=%p", msk);
 299
 300         if (!msk)
 301                 return;
 302
 303         if (inet_sk_state_load(sk) == TCP_CLOSE)
 304                 return;
 305
 306         if (!entry->addr.id)
 307                 return;
 308
 309         if (mptcp_pm_should_add_signal_addr(msk)) {
 310                 sk_reset_timer(sk, timer, jiffies + TCP_RTO_MAX / 8);
 311                 goto out;
 312         }
 313
 314         spin_lock_bh(&msk->pm.lock);
 315
 316         if (!mptcp_pm_should_add_signal_addr(msk)) {
 317                 pr_debug("retransmit ADD_ADDR id=%d", entry->addr.id);
 318                 mptcp_pm_announce_addr(msk, &entry->addr, false);
 319                 mptcp_pm_add_addr_send_ack(msk);
 320                 entry->retrans_times++;
 321         }
 322
 323         if (entry->retrans_times < ADD_ADDR_RETRANS_MAX)
 324                 sk_reset_timer(sk, timer,
 325                                jiffies + mptcp_get_add_addr_timeout(sock_net(sk)));
 326
 327         spin_unlock_bh(&msk->pm.lock);
 328
 329         if (entry->retrans_times == ADD_ADDR_RETRANS_MAX)
 330                 mptcp_pm_subflow_established(msk);
 331
 332 out:
 333         __sock_put(sk);
 334 }
 335
 336 struct mptcp_pm_add_entry *
 337 mptcp_pm_del_add_timer(struct mptcp_sock *msk,
 338                        const struct mptcp_addr_info *addr, bool check_id)
 339 {
 340         struct mptcp_pm_add_entry *entry;
 341         struct sock *sk = (struct sock *)msk;
 342
 343         spin_lock_bh(&msk->pm.lock);
 344         entry = mptcp_lookup_anno_list_by_saddr(msk, addr);
 345         if (entry && (!check_id || entry->addr.id == addr->id))
 346                 entry->retrans_times = ADD_ADDR_RETRANS_MAX;
 347         spin_unlock_bh(&msk->pm.lock);
 348
 349         if (entry && (!check_id || entry->addr.id == addr->id))
 350                 sk_stop_timer_sync(sk, &entry->add_timer);
 351
 352         return entry;
 353 }
 354
 355 bool mptcp_pm_alloc_anno_list(struct mptcp_sock *msk,
 356                               const struct mptcp_pm_addr_entry *entry)
 357 {
 358         struct mptcp_pm_add_entry *add_entry = NULL;
 359         struct sock *sk = (struct sock *)msk;
 360         struct net *net = sock_net(sk);
 361
 362         lockdep_assert_held(&msk->pm.lock);
 363
 364         add_entry = mptcp_lookup_anno_list_by_saddr(msk, &entry->addr);
 365
 366         if (add_entry) {
 367                 if (mptcp_pm_is_kernel(msk))
 368                         return false;
 369
 370                 sk_reset_timer(sk, &add_entry->add_timer,
 371                                jiffies + mptcp_get_add_addr_timeout(net));
 372                 return true;
 373         }
 374
 375         add_entry = kmalloc(sizeof(*add_entry), GFP_ATOMIC);
 376         if (!add_entry)
 377                 return false;
 378
 379         list_add(&add_entry->list, &msk->pm.anno_list);
 380
 381         add_entry->addr = entry->addr;
 382         add_entry->sock = msk;
 383         add_entry->retrans_times = 0;
 384
 385         timer_setup(&add_entry->add_timer, mptcp_pm_add_timer, 0);
 386         sk_reset_timer(sk, &add_entry->add_timer,
 387                        jiffies + mptcp_get_add_addr_timeout(net));
 388
 389         return true;
 390 }
 391
 392 void mptcp_pm_free_anno_list(struct mptcp_sock *msk)
 393 {
 394         struct mptcp_pm_add_entry *entry, *tmp;
 395         struct sock *sk = (struct sock *)msk;
 396         LIST_HEAD(free_list);
 397
 398         pr_debug("msk=%p", msk);
 399
 400         spin_lock_bh(&msk->pm.lock);
 401         list_splice_init(&msk->pm.anno_list, &free_list);
 402         spin_unlock_bh(&msk->pm.lock);
 403
 404         list_for_each_entry_safe(entry, tmp, &free_list, list) {
 405                 sk_stop_timer_sync(sk, &entry->add_timer);
 406                 kfree(entry);
 407         }
 408 }
 409
 410 static bool lookup_address_in_vec(const struct mptcp_addr_info *addrs, unsigned int nr,
 411                                   const struct mptcp_addr_info *addr)
 412 {
 413         int i;
 414
 415         for (i = 0; i < nr; i++) {
 416                 if (addrs[i].id == addr->id)
 417                         return true;
 418         }
 419
 420         return false;
 421 }
 422
 423 /* Fill all the remote addresses into the array addrs[],
 424  * and return the array size.
 425  */
 426 static unsigned int fill_remote_addresses_vec(struct mptcp_sock *msk, bool fullmesh,
 427                                               struct mptcp_addr_info *addrs)
 428 {
 429         bool deny_id0 = READ_ONCE(msk->pm.remote_deny_join_id0);
 430         struct sock *sk = (struct sock *)msk, *ssk;
 431         struct mptcp_subflow_context *subflow;
 432         struct mptcp_addr_info remote = { 0 };
 433         unsigned int subflows_max;
 434         int i = 0;
 435
 436         subflows_max = mptcp_pm_get_subflows_max(msk);
 437         remote_address((struct sock_common *)sk, &remote);
 438
 439         /* Non-fullmesh endpoint, fill in the single entry
 440          * corresponding to the primary MPC subflow remote address
 441          */
 442         if (!fullmesh) {
 443                 if (deny_id0)
 444                         return 0;
 445
 446                 msk->pm.subflows++;
 447                 addrs[i++] = remote;
 448         } else {
 449                 mptcp_for_each_subflow(msk, subflow) {
 450                         ssk = mptcp_subflow_tcp_sock(subflow);
 451                         remote_address((struct sock_common *)ssk, &addrs[i]);
 452                         addrs[i].id = subflow->remote_id;
 453                         if (deny_id0 && !addrs[i].id)
 454                                 continue;
 455
 456                         if (!lookup_address_in_vec(addrs, i, &addrs[i]) &&
 457                             msk->pm.subflows < subflows_max) {
 458                                 msk->pm.subflows++;
 459                                 i++;
 460                         }
 461                 }
 462         }
 463
 464         return i;
 465 }
 466
 467 static void __mptcp_pm_send_ack(struct mptcp_sock *msk, struct mptcp_subflow_context *subflow,
 468                                 bool prio, bool backup)
 469 {
 470         struct sock *ssk = mptcp_subflow_tcp_sock(subflow);
 471         bool slow;
 472
 473         pr_debug("send ack for %s",
 474                  prio ? "mp_prio" : (mptcp_pm_should_add_signal(msk) ? "add_addr" : "rm_addr"));
 475
 476         slow = lock_sock_fast(ssk);
 477         if (prio) {
 478                 if (subflow->backup != backup)
 479                         msk->last_snd = NULL;
 480
 481                 subflow->send_mp_prio = 1;
 482                 subflow->backup = backup;
 483                 subflow->request_bkup = backup;
 484         }
 485
 486         __mptcp_subflow_send_ack(ssk);
 487         unlock_sock_fast(ssk, slow);
 488 }
 489
 490 static void mptcp_pm_send_ack(struct mptcp_sock *msk, struct mptcp_subflow_context *subflow,
 491                               bool prio, bool backup)
 492 {
 493         spin_unlock_bh(&msk->pm.lock);
 494         __mptcp_pm_send_ack(msk, subflow, prio, backup);
 495         spin_lock_bh(&msk->pm.lock);
 496 }
 497
 498 static struct mptcp_pm_addr_entry *
 499 __lookup_addr_by_id(struct pm_nl_pernet *pernet, unsigned int id)
 500 {
 501         struct mptcp_pm_addr_entry *entry;
 502
 503         list_for_each_entry(entry, &pernet->local_addr_list, list) {
 504                 if (entry->addr.id == id)
 505                         return entry;
 506         }
 507         return NULL;
 508 }
 509
 510 static struct mptcp_pm_addr_entry *
 511 __lookup_addr(struct pm_nl_pernet *pernet, const struct mptcp_addr_info *info,
 512               bool lookup_by_id)
 513 {
 514         struct mptcp_pm_addr_entry *entry;
 515
 516         list_for_each_entry(entry, &pernet->local_addr_list, list) {
 517                 if ((!lookup_by_id &&
 518                      mptcp_addresses_equal(&entry->addr, info, entry->addr.port)) ||
 519                     (lookup_by_id && entry->addr.id == info->id))
 520                         return entry;
 521         }
 522         return NULL;
 523 }
 524
 525 static void mptcp_pm_create_subflow_or_signal_addr(struct mptcp_sock *msk)
 526 {
 527         struct sock *sk = (struct sock *)msk;
 528         struct mptcp_pm_addr_entry *local;
 529         unsigned int add_addr_signal_max;
 530         unsigned int local_addr_max;
 531         struct pm_nl_pernet *pernet;
 532         unsigned int subflows_max;
 533
 534         pernet = pm_nl_get_pernet(sock_net(sk));
 535
 536         add_addr_signal_max = mptcp_pm_get_add_addr_signal_max(msk);
 537         local_addr_max = mptcp_pm_get_local_addr_max(msk);
 538         subflows_max = mptcp_pm_get_subflows_max(msk);
 539
 540         /* do lazy endpoint usage accounting for the MPC subflows */
 541         if (unlikely(!(msk->pm.status & BIT(MPTCP_PM_MPC_ENDPOINT_ACCOUNTED))) && msk->first) {
 542                 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(msk->first);
 543                 struct mptcp_pm_addr_entry *entry;
 544                 struct mptcp_addr_info mpc_addr;
 545                 bool backup = false;
 546
 547                 local_address((struct sock_common *)msk->first, &mpc_addr);
 548                 rcu_read_lock();
 549                 entry = __lookup_addr(pernet, &mpc_addr, false);
 550                 if (entry) {
 551                         __clear_bit(entry->addr.id, msk->pm.id_avail_bitmap);
 552                         msk->mpc_endpoint_id = entry->addr.id;
 553                         backup = !!(entry->flags & MPTCP_PM_ADDR_FLAG_BACKUP);
 554                 }
 555                 rcu_read_unlock();
 556
 557                 if (backup)
 558                         mptcp_pm_send_ack(msk, subflow, true, backup);
 559
 560                 msk->pm.status |= BIT(MPTCP_PM_MPC_ENDPOINT_ACCOUNTED);
 561         }
 562
 563         pr_debug("local %d:%d signal %d:%d subflows %d:%d\n",
 564                  msk->pm.local_addr_used, local_addr_max,
 565                  msk->pm.add_addr_signaled, add_addr_signal_max,
 566                  msk->pm.subflows, subflows_max);
 567
 568         /* check first for announce */
 569         if (msk->pm.add_addr_signaled < add_addr_signal_max) {
 570                 local = select_signal_address(pernet, msk);
 571
 572                 /* due to racing events on both ends we can reach here while
 573                  * previous add address is still running: if we invoke now
 574                  * mptcp_pm_announce_addr(), that will fail and the
 575                  * corresponding id will be marked as used.
 576                  * Instead let the PM machinery reschedule us when the
 577                  * current address announce will be completed.
 578                  */
 579                 if (msk->pm.addr_signal & BIT(MPTCP_ADD_ADDR_SIGNAL))
 580                         return;
 581
 582                 if (local) {
 583                         if (mptcp_pm_alloc_anno_list(msk, local)) {
 584                                 __clear_bit(local->addr.id, msk->pm.id_avail_bitmap);
 585                                 msk->pm.add_addr_signaled++;
 586                                 mptcp_pm_announce_addr(msk, &local->addr, false);
 587                                 mptcp_pm_nl_addr_send_ack(msk);
 588                         }
 589                 }
 590         }
 591
 592         /* check if should create a new subflow */
 593         while (msk->pm.local_addr_used < local_addr_max &&
 594                msk->pm.subflows < subflows_max) {
 595                 struct mptcp_addr_info addrs[MPTCP_PM_ADDR_MAX];
 596                 bool fullmesh;
 597                 int i, nr;
 598
 599                 local = select_local_address(pernet, msk);
 600                 if (!local)
 601                         break;
 602
 603                 fullmesh = !!(local->flags & MPTCP_PM_ADDR_FLAG_FULLMESH);
 604
 605                 msk->pm.local_addr_used++;
 606                 nr = fill_remote_addresses_vec(msk, fullmesh, addrs);
 607                 if (nr)
 608                         __clear_bit(local->addr.id, msk->pm.id_avail_bitmap);
 609                 spin_unlock_bh(&msk->pm.lock);
 610                 for (i = 0; i < nr; i++)
 611                         __mptcp_subflow_connect(sk, &local->addr, &addrs[i]);
 612                 spin_lock_bh(&msk->pm.lock);
 613         }
 614         mptcp_pm_nl_check_work_pending(msk);
 615 }
 616
 617 static void mptcp_pm_nl_fully_established(struct mptcp_sock *msk)
 618 {
 619         mptcp_pm_create_subflow_or_signal_addr(msk);
 620 }
 621
 622 static void mptcp_pm_nl_subflow_established(struct mptcp_sock *msk)
 623 {
 624         mptcp_pm_create_subflow_or_signal_addr(msk);
 625 }
 626
 627 /* Fill all the local addresses into the array addrs[],
 628  * and return the array size.
 629  */
 630 static unsigned int fill_local_addresses_vec(struct mptcp_sock *msk,
 631                                              struct mptcp_addr_info *addrs)
 632 {
 633         struct sock *sk = (struct sock *)msk;
 634         struct mptcp_pm_addr_entry *entry;
 635         struct mptcp_addr_info local;
 636         struct pm_nl_pernet *pernet;
 637         unsigned int subflows_max;
 638         int i = 0;
 639
 640         pernet = pm_nl_get_pernet_from_msk(msk);
 641         subflows_max = mptcp_pm_get_subflows_max(msk);
 642
 643         rcu_read_lock();
 644         list_for_each_entry_rcu(entry, &pernet->local_addr_list, list) {
 645                 if (!(entry->flags & MPTCP_PM_ADDR_FLAG_FULLMESH))
 646                         continue;
 647
 648                 if (entry->addr.family != sk->sk_family) {
 649 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
 650                         if ((entry->addr.family == AF_INET &&
 651                              !ipv6_addr_v4mapped(&sk->sk_v6_daddr)) ||
 652                             (sk->sk_family == AF_INET &&
 653                              !ipv6_addr_v4mapped(&entry->addr.addr6)))
 654 #endif
 655                                 continue;
 656                 }
 657
 658                 if (msk->pm.subflows < subflows_max) {
 659                         msk->pm.subflows++;
 660                         addrs[i++] = entry->addr;
 661                 }
 662         }
 663         rcu_read_unlock();
 664
 665         /* If the array is empty, fill in the single
 666          * 'IPADDRANY' local address
 667          */
 668         if (!i) {
 669                 memset(&local, 0, sizeof(local));
 670                 local.family = msk->pm.remote.family;
 671
 672                 msk->pm.subflows++;
 673                 addrs[i++] = local;
 674         }
 675
 676         return i;
 677 }
 678
 679 static void mptcp_pm_nl_add_addr_received(struct mptcp_sock *msk)
 680 {
 681         struct mptcp_addr_info addrs[MPTCP_PM_ADDR_MAX];
 682         struct sock *sk = (struct sock *)msk;
 683         unsigned int add_addr_accept_max;
 684         struct mptcp_addr_info remote;
 685         unsigned int subflows_max;
 686         int i, nr;
 687
 688         add_addr_accept_max = mptcp_pm_get_add_addr_accept_max(msk);
 689         subflows_max = mptcp_pm_get_subflows_max(msk);
 690
 691         pr_debug("accepted %d:%d remote family %d",
 692                  msk->pm.add_addr_accepted, add_addr_accept_max,
 693                  msk->pm.remote.family);
 694
 695         remote = msk->pm.remote;
 696         mptcp_pm_announce_addr(msk, &remote, true);
 697         mptcp_pm_nl_addr_send_ack(msk);
 698
 699         if (lookup_subflow_by_daddr(&msk->conn_list, &remote))
 700                 return;
 701
 702         /* pick id 0 port, if none is provided the remote address */
 703         if (!remote.port)
 704                 remote.port = sk->sk_dport;
 705
 706         /* connect to the specified remote address, using whatever
 707          * local address the routing configuration will pick.
 708          */
 709         nr = fill_local_addresses_vec(msk, addrs);
 710
 711         msk->pm.add_addr_accepted++;
 712         if (msk->pm.add_addr_accepted >= add_addr_accept_max ||
 713             msk->pm.subflows >= subflows_max)
 714                 WRITE_ONCE(msk->pm.accept_addr, false);
 715
 716         spin_unlock_bh(&msk->pm.lock);
 717         for (i = 0; i < nr; i++)
 718                 __mptcp_subflow_connect(sk, &addrs[i], &remote);
 719         spin_lock_bh(&msk->pm.lock);
 720 }
 721
 722 void mptcp_pm_nl_addr_send_ack(struct mptcp_sock *msk)
 723 {
 724         struct mptcp_subflow_context *subflow;
 725
 726         msk_owned_by_me(msk);
 727         lockdep_assert_held(&msk->pm.lock);
 728
 729         if (!mptcp_pm_should_add_signal(msk) &&
 730             !mptcp_pm_should_rm_signal(msk))
 731                 return;
 732
 733         subflow = list_first_entry_or_null(&msk->conn_list, typeof(*subflow), node);
 734         if (subflow)
 735                 mptcp_pm_send_ack(msk, subflow, false, false);
 736 }
 737
 738 int mptcp_pm_nl_mp_prio_send_ack(struct mptcp_sock *msk,
 739                                  struct mptcp_addr_info *addr,
 740                                  struct mptcp_addr_info *rem,
 741                                  u8 bkup)
 742 {
 743         struct mptcp_subflow_context *subflow;
 744
 745         pr_debug("bkup=%d", bkup);
 746
 747         mptcp_for_each_subflow(msk, subflow) {
 748                 struct sock *ssk = mptcp_subflow_tcp_sock(subflow);
 749                 struct mptcp_addr_info local, remote;
 750
 751                 local_address((struct sock_common *)ssk, &local);
 752                 if (!mptcp_addresses_equal(&local, addr, addr->port))
 753                         continue;
 754
 755                 if (rem && rem->family != AF_UNSPEC) {
 756                         remote_address((struct sock_common *)ssk, &remote);
 757                         if (!mptcp_addresses_equal(&remote, rem, rem->port))
 758                                 continue;
 759                 }
 760
 761                 __mptcp_pm_send_ack(msk, subflow, true, bkup);
 762                 return 0;
 763         }
 764
 765         return -EINVAL;
 766 }
 767
 768 static bool mptcp_local_id_match(const struct mptcp_sock *msk, u8 local_id, u8 id)
 769 {
 770         return local_id == id || (!local_id && msk->mpc_endpoint_id == id);
 771 }
 772
 773 static void mptcp_pm_nl_rm_addr_or_subflow(struct mptcp_sock *msk,
 774                                            const struct mptcp_rm_list *rm_list,
 775                                            enum linux_mptcp_mib_field rm_type)
 776 {
 777         struct mptcp_subflow_context *subflow, *tmp;
 778         struct sock *sk = (struct sock *)msk;
 779         u8 i;
 780
 781         pr_debug("%s rm_list_nr %d",
 782                  rm_type == MPTCP_MIB_RMADDR ? "address" : "subflow", rm_list->nr);
 783
 784         msk_owned_by_me(msk);
 785
 786         if (sk->sk_state == TCP_LISTEN)
 787                 return;
 788
 789         if (!rm_list->nr)
 790                 return;
 791
 792         if (list_empty(&msk->conn_list))
 793                 return;
 794
 795         for (i = 0; i < rm_list->nr; i++) {
 796                 u8 rm_id = rm_list->ids[i];
 797                 bool removed = false;
 798
 799                 mptcp_for_each_subflow_safe(msk, subflow, tmp) {
 800                         struct sock *ssk = mptcp_subflow_tcp_sock(subflow);
 801                         int how = RCV_SHUTDOWN | SEND_SHUTDOWN;
 802                         u8 id = subflow->local_id;
 803
 804                         if (rm_type == MPTCP_MIB_RMADDR && subflow->remote_id != rm_id)
 805                                 continue;
 806                         if (rm_type == MPTCP_MIB_RMSUBFLOW && !mptcp_local_id_match(msk, id, rm_id))
 807                                 continue;
 808
 809                         pr_debug(" -> %s rm_list_ids[%d]=%u local_id=%u remote_id=%u mpc_id=%u",
 810                                  rm_type == MPTCP_MIB_RMADDR ? "address" : "subflow",
 811                                  i, rm_id, subflow->local_id, subflow->remote_id,
 812                                  msk->mpc_endpoint_id);
 813                         spin_unlock_bh(&msk->pm.lock);
 814                         mptcp_subflow_shutdown(sk, ssk, how);
 815
 816                         /* the following takes care of updating the subflows counter */
 817                         mptcp_close_ssk(sk, ssk, subflow);
 818                         spin_lock_bh(&msk->pm.lock);
 819
 820                         removed = true;
 821                         __MPTCP_INC_STATS(sock_net(sk), rm_type);
 822                 }
 823                 if (rm_type == MPTCP_MIB_RMSUBFLOW)
 824                         __set_bit(rm_id ? rm_id : msk->mpc_endpoint_id, msk->pm.id_avail_bitmap);
 825                 if (!removed)
 826                         continue;
 827
 828                 if (!mptcp_pm_is_kernel(msk))
 829                         continue;
 830
 831                 if (rm_type == MPTCP_MIB_RMADDR) {
 832                         msk->pm.add_addr_accepted--;
 833                         WRITE_ONCE(msk->pm.accept_addr, true);
 834                 } else if (rm_type == MPTCP_MIB_RMSUBFLOW) {
 835                         msk->pm.local_addr_used--;
 836                 }
 837         }
 838 }
 839
 840 static void mptcp_pm_nl_rm_addr_received(struct mptcp_sock *msk)
 841 {
 842         mptcp_pm_nl_rm_addr_or_subflow(msk, &msk->pm.rm_list_rx, MPTCP_MIB_RMADDR);
 843 }
 844
 845 void mptcp_pm_nl_rm_subflow_received(struct mptcp_sock *msk,
 846                                      const struct mptcp_rm_list *rm_list)
 847 {
 848         mptcp_pm_nl_rm_addr_or_subflow(msk, rm_list, MPTCP_MIB_RMSUBFLOW);
 849 }
 850
 851 void mptcp_pm_nl_work(struct mptcp_sock *msk)
 852 {
 853         struct mptcp_pm_data *pm = &msk->pm;
 854
 855         msk_owned_by_me(msk);
 856
 857         if (!(pm->status & MPTCP_PM_WORK_MASK))
 858                 return;
 859
 860         spin_lock_bh(&msk->pm.lock);
 861
 862         pr_debug("msk=%p status=%x", msk, pm->status);
 863         if (pm->status & BIT(MPTCP_PM_ADD_ADDR_RECEIVED)) {
 864                 pm->status &= ~BIT(MPTCP_PM_ADD_ADDR_RECEIVED);
 865                 mptcp_pm_nl_add_addr_received(msk);
 866         }
 867         if (pm->status & BIT(MPTCP_PM_ADD_ADDR_SEND_ACK)) {
 868                 pm->status &= ~BIT(MPTCP_PM_ADD_ADDR_SEND_ACK);
 869                 mptcp_pm_nl_addr_send_ack(msk);
 870         }
 871         if (pm->status & BIT(MPTCP_PM_RM_ADDR_RECEIVED)) {
 872                 pm->status &= ~BIT(MPTCP_PM_RM_ADDR_RECEIVED);
 873                 mptcp_pm_nl_rm_addr_received(msk);
 874         }
 875         if (pm->status & BIT(MPTCP_PM_ESTABLISHED)) {
 876                 pm->status &= ~BIT(MPTCP_PM_ESTABLISHED);
 877                 mptcp_pm_nl_fully_established(msk);
 878         }
 879         if (pm->status & BIT(MPTCP_PM_SUBFLOW_ESTABLISHED)) {
 880                 pm->status &= ~BIT(MPTCP_PM_SUBFLOW_ESTABLISHED);
 881                 mptcp_pm_nl_subflow_established(msk);
 882         }
 883
 884         spin_unlock_bh(&msk->pm.lock);
 885 }
 886
 887 static bool address_use_port(struct mptcp_pm_addr_entry *entry)
 888 {
 889         return (entry->flags &
 890                 (MPTCP_PM_ADDR_FLAG_SIGNAL | MPTCP_PM_ADDR_FLAG_SUBFLOW)) ==
 891                 MPTCP_PM_ADDR_FLAG_SIGNAL;
 892 }
 893
 894 /* caller must ensure the RCU grace period is already elapsed */
 895 static void __mptcp_pm_release_addr_entry(struct mptcp_pm_addr_entry *entry)
 896 {
 897         if (entry->lsk)
 898                 sock_release(entry->lsk);
 899         kfree(entry);
 900 }
 901
 902 static int mptcp_pm_nl_append_new_local_addr(struct pm_nl_pernet *pernet,
 903                                              struct mptcp_pm_addr_entry *entry)
 904 {
 905         struct mptcp_pm_addr_entry *cur, *del_entry = NULL;
 906         unsigned int addr_max;
 907         int ret = -EINVAL;
 908
 909         spin_lock_bh(&pernet->lock);
 910         /* to keep the code simple, don't do IDR-like allocation for address ID,
 911          * just bail when we exceed limits
 912          */
 913         if (pernet->next_id == MPTCP_PM_MAX_ADDR_ID)
 914                 pernet->next_id = 1;
 915         if (pernet->addrs >= MPTCP_PM_ADDR_MAX) {
 916                 ret = -ERANGE;
 917                 goto out;
 918         }
 919         if (test_bit(entry->addr.id, pernet->id_bitmap)) {
 920                 ret = -EBUSY;
 921                 goto out;
 922         }
 923
 924         /* do not insert duplicate address, differentiate on port only
 925          * singled addresses
 926          */
 927         if (!address_use_port(entry))
 928                 entry->addr.port = 0;
 929         list_for_each_entry(cur, &pernet->local_addr_list, list) {
 930                 if (mptcp_addresses_equal(&cur->addr, &entry->addr,
 931                                           cur->addr.port || entry->addr.port)) {
 932                         /* allow replacing the exiting endpoint only if such
 933                          * endpoint is an implicit one and the user-space
 934                          * did not provide an endpoint id
 935                          */
 936                         if (!(cur->flags & MPTCP_PM_ADDR_FLAG_IMPLICIT)) {
 937                                 ret = -EEXIST;
 938                                 goto out;
 939                         }
 940                         if (entry->addr.id)
 941                                 goto out;
 942
 943                         pernet->addrs--;
 944                         entry->addr.id = cur->addr.id;
 945                         list_del_rcu(&cur->list);
 946                         del_entry = cur;
 947                         break;
 948                 }
 949         }
 950
 951         if (!entry->addr.id) {
 952 find_next:
 953                 entry->addr.id = find_next_zero_bit(pernet->id_bitmap,
 954                                                     MPTCP_PM_MAX_ADDR_ID + 1,
 955                                                     pernet->next_id);
 956                 if (!entry->addr.id && pernet->next_id != 1) {
 957                         pernet->next_id = 1;
 958                         goto find_next;
 959                 }
 960         }
 961
 962         if (!entry->addr.id)
 963                 goto out;
 964
 965         __set_bit(entry->addr.id, pernet->id_bitmap);
 966         if (entry->addr.id > pernet->next_id)
 967                 pernet->next_id = entry->addr.id;
 968
 969         if (entry->flags & MPTCP_PM_ADDR_FLAG_SIGNAL) {
 970                 addr_max = pernet->add_addr_signal_max;
 971                 WRITE_ONCE(pernet->add_addr_signal_max, addr_max + 1);
 972         }
 973         if (entry->flags & MPTCP_PM_ADDR_FLAG_SUBFLOW) {
 974                 addr_max = pernet->local_addr_max;
 975                 WRITE_ONCE(pernet->local_addr_max, addr_max + 1);
 976         }
 977
 978         pernet->addrs++;
 979         if (!entry->addr.port)
 980                 list_add_tail_rcu(&entry->list, &pernet->local_addr_list);
 981         else
 982                 list_add_rcu(&entry->list, &pernet->local_addr_list);
 983         ret = entry->addr.id;
 984
 985 out:
 986         spin_unlock_bh(&pernet->lock);
 987
 988         /* just replaced an existing entry, free it */
 989         if (del_entry) {
 990                 synchronize_rcu();
 991                 __mptcp_pm_release_addr_entry(del_entry);
 992         }
 993         return ret;
 994 }
 995
 996 static int mptcp_pm_nl_create_listen_socket(struct sock *sk,
 997                                             struct mptcp_pm_addr_entry *entry)
 998 {
 999         int addrlen = sizeof(struct sockaddr_in);
1000         struct sockaddr_storage addr;
1001         struct mptcp_sock *msk;
1002         struct socket *ssock;
1003         int backlog = 1024;
1004         int err;
1005
1006         err = sock_create_kern(sock_net(sk), entry->addr.family,
1007                                SOCK_STREAM, IPPROTO_MPTCP, &entry->lsk);
1008         if (err)
1009                 return err;
1010
1011         msk = mptcp_sk(entry->lsk->sk);
1012         if (!msk)
1013                 return -EINVAL;
1014
1015         ssock = __mptcp_nmpc_socket(msk);
1016         if (!ssock)
1017                 return -EINVAL;
1018
1019         mptcp_info2sockaddr(&entry->addr, &addr, entry->addr.family);
1020 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
1021         if (entry->addr.family == AF_INET6)
1022                 addrlen = sizeof(struct sockaddr_in6);
1023 #endif
1024         err = kernel_bind(ssock, (struct sockaddr *)&addr, addrlen);
1025         if (err)
1026                 return err;
1027
1028         err = kernel_listen(ssock, backlog);
1029         if (err)
1030                 return err;
1031
1032         mptcp_event_pm_listener(ssock->sk, MPTCP_EVENT_LISTENER_CREATED);
1033
1034         return 0;
1035 }
1036
1037 int mptcp_pm_nl_get_local_id(struct mptcp_sock *msk, struct sock_common *skc)
1038 {
1039         struct mptcp_pm_addr_entry *entry;
1040         struct mptcp_addr_info skc_local;
1041         struct mptcp_addr_info msk_local;
1042         struct pm_nl_pernet *pernet;
1043         int ret = -1;
1044
1045         if (WARN_ON_ONCE(!msk))
1046                 return -1;
1047
1048         /* The 0 ID mapping is defined by the first subflow, copied into the msk
1049          * addr
1050          */
1051         local_address((struct sock_common *)msk, &msk_local);
1052         local_address((struct sock_common *)skc, &skc_local);
1053         if (mptcp_addresses_equal(&msk_local, &skc_local, false))
1054                 return 0;
1055
1056         if (mptcp_pm_is_userspace(msk))
1057                 return mptcp_userspace_pm_get_local_id(msk, &skc_local);
1058
1059         pernet = pm_nl_get_pernet_from_msk(msk);
1060
1061         rcu_read_lock();
1062         list_for_each_entry_rcu(entry, &pernet->local_addr_list, list) {
1063                 if (mptcp_addresses_equal(&entry->addr, &skc_local, entry->addr.port)) {
1064                         ret = entry->addr.id;
1065                         break;
1066                 }
1067         }
1068         rcu_read_unlock();
1069         if (ret >= 0)
1070                 return ret;
1071
1072         /* address not found, add to local list */
1073         entry = kmalloc(sizeof(*entry), GFP_ATOMIC);
1074         if (!entry)
1075                 return -ENOMEM;
1076
1077         entry->addr = skc_local;
1078         entry->addr.id = 0;
1079         entry->addr.port = 0;
1080         entry->ifindex = 0;
1081         entry->flags = MPTCP_PM_ADDR_FLAG_IMPLICIT;
1082         entry->lsk = NULL;
1083         ret = mptcp_pm_nl_append_new_local_addr(pernet, entry);
1084         if (ret < 0)
1085                 kfree(entry);
1086
1087         return ret;
1088 }
1089
1090 #define MPTCP_PM_CMD_GRP_OFFSET       0
1091 #define MPTCP_PM_EV_GRP_OFFSET        1
1092
1093 static const struct genl_multicast_group mptcp_pm_mcgrps[] = {
1094         [MPTCP_PM_CMD_GRP_OFFSET]       = { .name = MPTCP_PM_CMD_GRP_NAME, },
1095         [MPTCP_PM_EV_GRP_OFFSET]        = { .name = MPTCP_PM_EV_GRP_NAME,
1096                                             .flags = GENL_UNS_ADMIN_PERM,
1097                                           },
1098 };
1099
1100 static const struct nla_policy
1101 mptcp_pm_addr_policy[MPTCP_PM_ADDR_ATTR_MAX + 1] = {
1102         [MPTCP_PM_ADDR_ATTR_FAMILY]     = { .type       = NLA_U16,      },
1103         [MPTCP_PM_ADDR_ATTR_ID]         = { .type       = NLA_U8,       },
1104         [MPTCP_PM_ADDR_ATTR_ADDR4]      = { .type       = NLA_U32,      },
1105         [MPTCP_PM_ADDR_ATTR_ADDR6]      =
1106                 NLA_POLICY_EXACT_LEN(sizeof(struct in6_addr)),
1107         [MPTCP_PM_ADDR_ATTR_PORT]       = { .type       = NLA_U16       },
1108         [MPTCP_PM_ADDR_ATTR_FLAGS]      = { .type       = NLA_U32       },
1109         [MPTCP_PM_ADDR_ATTR_IF_IDX]     = { .type       = NLA_S32       },
1110 };
1111
1112 static const struct nla_policy mptcp_pm_policy[MPTCP_PM_ATTR_MAX + 1] = {
1113         [MPTCP_PM_ATTR_ADDR]            =
1114                                         NLA_POLICY_NESTED(mptcp_pm_addr_policy),
1115         [MPTCP_PM_ATTR_RCV_ADD_ADDRS]   = { .type       = NLA_U32,      },
1116         [MPTCP_PM_ATTR_SUBFLOWS]        = { .type       = NLA_U32,      },
1117         [MPTCP_PM_ATTR_TOKEN]           = { .type       = NLA_U32,      },
1118         [MPTCP_PM_ATTR_LOC_ID]          = { .type       = NLA_U8,       },
1119         [MPTCP_PM_ATTR_ADDR_REMOTE]     =
1120                                         NLA_POLICY_NESTED(mptcp_pm_addr_policy),
1121 };
1122
1123 void mptcp_pm_nl_subflow_chk_stale(const struct mptcp_sock *msk, struct sock *ssk)
1124 {
1125         struct mptcp_subflow_context *iter, *subflow = mptcp_subflow_ctx(ssk);
1126         struct sock *sk = (struct sock *)msk;
1127         unsigned int active_max_loss_cnt;
1128         struct net *net = sock_net(sk);
1129         unsigned int stale_loss_cnt;
1130         bool slow;
1131
1132         stale_loss_cnt = mptcp_stale_loss_cnt(net);
1133         if (subflow->stale || !stale_loss_cnt || subflow->stale_count <= stale_loss_cnt)
1134                 return;
1135
1136         /* look for another available subflow not in loss state */
1137         active_max_loss_cnt = max_t(int, stale_loss_cnt - 1, 1);
1138         mptcp_for_each_subflow(msk, iter) {
1139                 if (iter != subflow && mptcp_subflow_active(iter) &&
1140                     iter->stale_count < active_max_loss_cnt) {
1141                         /* we have some alternatives, try to mark this subflow as idle ...*/
1142                         slow = lock_sock_fast(ssk);
1143                         if (!tcp_rtx_and_write_queues_empty(ssk)) {
1144                                 subflow->stale = 1;
1145                                 __mptcp_retransmit_pending_data(sk);
1146                                 MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_SUBFLOWSTALE);
1147                         }
1148                         unlock_sock_fast(ssk, slow);
1149
1150                         /* always try to push the pending data regardless of re-injections:
1151                          * we can possibly use backup subflows now, and subflow selection
1152                          * is cheap under the msk socket lock
1153                          */
1154                         __mptcp_push_pending(sk, 0);
1155                         return;
1156                 }
1157         }
1158 }
1159
1160 static int mptcp_pm_family_to_addr(int family)
1161 {
1162 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
1163         if (family == AF_INET6)
1164                 return MPTCP_PM_ADDR_ATTR_ADDR6;
1165 #endif
1166         return MPTCP_PM_ADDR_ATTR_ADDR4;
1167 }
1168
1169 static int mptcp_pm_parse_pm_addr_attr(struct nlattr *tb[],
1170                                        const struct nlattr *attr,
1171                                        struct genl_info *info,
1172                                        struct mptcp_addr_info *addr,
1173                                        bool require_family)
1174 {
1175         int err, addr_addr;
1176
1177         if (!attr) {
1178                 GENL_SET_ERR_MSG(info, "missing address info");
1179                 return -EINVAL;
1180         }
1181
1182         /* no validation needed - was already done via nested policy */
1183         err = nla_parse_nested_deprecated(tb, MPTCP_PM_ADDR_ATTR_MAX, attr,
1184                                           mptcp_pm_addr_policy, info->extack);
1185         if (err)
1186                 return err;
1187
1188         if (tb[MPTCP_PM_ADDR_ATTR_ID])
1189                 addr->id = nla_get_u8(tb[MPTCP_PM_ADDR_ATTR_ID]);
1190
1191         if (!tb[MPTCP_PM_ADDR_ATTR_FAMILY]) {
1192                 if (!require_family)
1193                         return 0;
1194
1195                 NL_SET_ERR_MSG_ATTR(info->extack, attr,
1196                                     "missing family");
1197                 return -EINVAL;
1198         }
1199
1200         addr->family = nla_get_u16(tb[MPTCP_PM_ADDR_ATTR_FAMILY]);
1201         if (addr->family != AF_INET
1202 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
1203             && addr->family != AF_INET6
1204 #endif
1205             ) {
1206                 NL_SET_ERR_MSG_ATTR(info->extack, attr,
1207                                     "unknown address family");
1208                 return -EINVAL;
1209         }
1210         addr_addr = mptcp_pm_family_to_addr(addr->family);
1211         if (!tb[addr_addr]) {
1212                 NL_SET_ERR_MSG_ATTR(info->extack, attr,
1213                                     "missing address data");
1214                 return -EINVAL;
1215         }
1216
1217 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
1218         if (addr->family == AF_INET6)
1219                 addr->addr6 = nla_get_in6_addr(tb[addr_addr]);
1220         else
1221 #endif
1222                 addr->addr.s_addr = nla_get_in_addr(tb[addr_addr]);
1223
1224         if (tb[MPTCP_PM_ADDR_ATTR_PORT])
1225                 addr->port = htons(nla_get_u16(tb[MPTCP_PM_ADDR_ATTR_PORT]));
1226
1227         return 0;
1228 }
1229
1230 int mptcp_pm_parse_addr(struct nlattr *attr, struct genl_info *info,
1231                         struct mptcp_addr_info *addr)
1232 {
1233         struct nlattr *tb[MPTCP_PM_ADDR_ATTR_MAX + 1];
1234
1235         memset(addr, 0, sizeof(*addr));
1236
1237         return mptcp_pm_parse_pm_addr_attr(tb, attr, info, addr, true);
1238 }
1239
1240 int mptcp_pm_parse_entry(struct nlattr *attr, struct genl_info *info,
1241                          bool require_family,
1242                          struct mptcp_pm_addr_entry *entry)
1243 {
1244         struct nlattr *tb[MPTCP_PM_ADDR_ATTR_MAX + 1];
1245         int err;
1246
1247         memset(entry, 0, sizeof(*entry));
1248
1249         err = mptcp_pm_parse_pm_addr_attr(tb, attr, info, &entry->addr, require_family);
1250         if (err)
1251                 return err;
1252
1253         if (tb[MPTCP_PM_ADDR_ATTR_IF_IDX]) {
1254                 u32 val = nla_get_s32(tb[MPTCP_PM_ADDR_ATTR_IF_IDX]);
1255
1256                 entry->ifindex = val;
1257         }
1258
1259         if (tb[MPTCP_PM_ADDR_ATTR_FLAGS])
1260                 entry->flags = nla_get_u32(tb[MPTCP_PM_ADDR_ATTR_FLAGS]);
1261
1262         if (tb[MPTCP_PM_ADDR_ATTR_PORT])
1263                 entry->addr.port = htons(nla_get_u16(tb[MPTCP_PM_ADDR_ATTR_PORT]));
1264
1265         return 0;
1266 }
1267
1268 static struct pm_nl_pernet *genl_info_pm_nl(struct genl_info *info)
1269 {
1270         return pm_nl_get_pernet(genl_info_net(info));
1271 }
1272
1273 static int mptcp_nl_add_subflow_or_signal_addr(struct net *net)
1274 {
1275         struct mptcp_sock *msk;
1276         long s_slot = 0, s_num = 0;
1277
1278         while ((msk = mptcp_token_iter_next(net, &s_slot, &s_num)) != NULL) {
1279                 struct sock *sk = (struct sock *)msk;
1280
1281                 if (!READ_ONCE(msk->fully_established) ||
1282                     mptcp_pm_is_userspace(msk))
1283                         goto next;
1284
1285                 lock_sock(sk);
1286                 spin_lock_bh(&msk->pm.lock);
1287                 mptcp_pm_create_subflow_or_signal_addr(msk);
1288                 spin_unlock_bh(&msk->pm.lock);
1289                 release_sock(sk);
1290
1291 next:
1292                 sock_put(sk);
1293                 cond_resched();
1294         }
1295
1296         return 0;
1297 }
1298
1299 static int mptcp_nl_cmd_add_addr(struct sk_buff *skb, struct genl_info *info)
1300 {
1301         struct nlattr *attr = info->attrs[MPTCP_PM_ATTR_ADDR];
1302         struct pm_nl_pernet *pernet = genl_info_pm_nl(info);
1303         struct mptcp_pm_addr_entry addr, *entry;
1304         int ret;
1305
1306         ret = mptcp_pm_parse_entry(attr, info, true, &addr);
1307         if (ret < 0)
1308                 return ret;
1309
1310         if (addr.addr.port && !(addr.flags & MPTCP_PM_ADDR_FLAG_SIGNAL)) {
1311                 GENL_SET_ERR_MSG(info, "flags must have signal when using port");
1312                 return -EINVAL;
1313         }
1314
1315         if (addr.flags & MPTCP_PM_ADDR_FLAG_SIGNAL &&
1316             addr.flags & MPTCP_PM_ADDR_FLAG_FULLMESH) {
1317                 GENL_SET_ERR_MSG(info, "flags mustn't have both signal and fullmesh");
1318                 return -EINVAL;
1319         }
1320
1321         if (addr.flags & MPTCP_PM_ADDR_FLAG_IMPLICIT) {
1322                 GENL_SET_ERR_MSG(info, "can't create IMPLICIT endpoint");
1323                 return -EINVAL;
1324         }
1325
1326         entry = kzalloc(sizeof(*entry), GFP_KERNEL_ACCOUNT);
1327         if (!entry) {
1328                 GENL_SET_ERR_MSG(info, "can't allocate addr");
1329                 return -ENOMEM;
1330         }
1331
1332         *entry = addr;
1333         if (entry->addr.port) {
1334                 ret = mptcp_pm_nl_create_listen_socket(skb->sk, entry);
1335                 if (ret) {
1336                         GENL_SET_ERR_MSG_FMT(info, "create listen socket error: %d", ret);
1337                         goto out_free;
1338                 }
1339         }
1340         ret = mptcp_pm_nl_append_new_local_addr(pernet, entry);
1341         if (ret < 0) {
1342                 GENL_SET_ERR_MSG_FMT(info, "too many addresses or duplicate one: %d", ret);
1343                 goto out_free;
1344         }
1345
1346         mptcp_nl_add_subflow_or_signal_addr(sock_net(skb->sk));
1347         return 0;
1348
1349 out_free:
1350         __mptcp_pm_release_addr_entry(entry);
1351         return ret;
1352 }
1353
1354 int mptcp_pm_get_flags_and_ifindex_by_id(struct mptcp_sock *msk, unsigned int id,
1355                                          u8 *flags, int *ifindex)
1356 {
1357         struct mptcp_pm_addr_entry *entry;
1358         struct sock *sk = (struct sock *)msk;
1359         struct net *net = sock_net(sk);
1360
1361         *flags = 0;
1362         *ifindex = 0;
1363
1364         if (id) {
1365                 if (mptcp_pm_is_userspace(msk))
1366                         return mptcp_userspace_pm_get_flags_and_ifindex_by_id(msk,
1367                                                                               id,
1368                                                                               flags,
1369                                                                               ifindex);
1370
1371                 rcu_read_lock();
1372                 entry = __lookup_addr_by_id(pm_nl_get_pernet(net), id);
1373                 if (entry) {
1374                         *flags = entry->flags;
1375                         *ifindex = entry->ifindex;
1376                 }
1377                 rcu_read_unlock();
1378         }
1379
1380         return 0;
1381 }
1382
1383 static bool remove_anno_list_by_saddr(struct mptcp_sock *msk,
1384                                       const struct mptcp_addr_info *addr)
1385 {
1386         struct mptcp_pm_add_entry *entry;
1387
1388         entry = mptcp_pm_del_add_timer(msk, addr, false);
1389         if (entry) {
1390                 list_del(&entry->list);
1391                 kfree(entry);
1392                 return true;
1393         }
1394
1395         return false;
1396 }
1397
1398 static bool mptcp_pm_remove_anno_addr(struct mptcp_sock *msk,
1399                                       const struct mptcp_addr_info *addr,
1400                                       bool force)
1401 {
1402         struct mptcp_rm_list list = { .nr = 0 };
1403         bool ret;
1404
1405         list.ids[list.nr++] = addr->id;
1406
1407         ret = remove_anno_list_by_saddr(msk, addr);
1408         if (ret || force) {
1409                 spin_lock_bh(&msk->pm.lock);
1410                 mptcp_pm_remove_addr(msk, &list);
1411                 spin_unlock_bh(&msk->pm.lock);
1412         }
1413         return ret;
1414 }
1415
1416 static int mptcp_nl_remove_subflow_and_signal_addr(struct net *net,
1417                                                    const struct mptcp_pm_addr_entry *entry)
1418 {
1419         const struct mptcp_addr_info *addr = &entry->addr;
1420         struct mptcp_rm_list list = { .nr = 0 };
1421         long s_slot = 0, s_num = 0;
1422         struct mptcp_sock *msk;
1423
1424         pr_debug("remove_id=%d", addr->id);
1425
1426         list.ids[list.nr++] = addr->id;
1427
1428         while ((msk = mptcp_token_iter_next(net, &s_slot, &s_num)) != NULL) {
1429                 struct sock *sk = (struct sock *)msk;
1430                 bool remove_subflow;
1431
1432                 if (mptcp_pm_is_userspace(msk))
1433                         goto next;
1434
1435                 if (list_empty(&msk->conn_list)) {
1436                         mptcp_pm_remove_anno_addr(msk, addr, false);
1437                         goto next;
1438                 }
1439
1440                 lock_sock(sk);
1441                 remove_subflow = lookup_subflow_by_saddr(&msk->conn_list, addr);
1442                 mptcp_pm_remove_anno_addr(msk, addr, remove_subflow &&
1443                                           !(entry->flags & MPTCP_PM_ADDR_FLAG_IMPLICIT));
1444                 if (remove_subflow)
1445                         mptcp_pm_remove_subflow(msk, &list);
1446                 release_sock(sk);
1447
1448 next:
1449                 sock_put(sk);
1450                 cond_resched();
1451         }
1452
1453         return 0;
1454 }
1455
1456 static int mptcp_nl_remove_id_zero_address(struct net *net,
1457                                            struct mptcp_addr_info *addr)
1458 {
1459         struct mptcp_rm_list list = { .nr = 0 };
1460         long s_slot = 0, s_num = 0;
1461         struct mptcp_sock *msk;
1462
1463         list.ids[list.nr++] = 0;
1464
1465         while ((msk = mptcp_token_iter_next(net, &s_slot, &s_num)) != NULL) {
1466                 struct sock *sk = (struct sock *)msk;
1467                 struct mptcp_addr_info msk_local;
1468
1469                 if (list_empty(&msk->conn_list) || mptcp_pm_is_userspace(msk))
1470                         goto next;
1471
1472                 local_address((struct sock_common *)msk, &msk_local);
1473                 if (!mptcp_addresses_equal(&msk_local, addr, addr->port))
1474                         goto next;
1475
1476                 lock_sock(sk);
1477                 spin_lock_bh(&msk->pm.lock);
1478                 mptcp_pm_remove_addr(msk, &list);
1479                 mptcp_pm_nl_rm_subflow_received(msk, &list);
1480                 spin_unlock_bh(&msk->pm.lock);
1481                 release_sock(sk);
1482
1483 next:
1484                 sock_put(sk);
1485                 cond_resched();
1486         }
1487
1488         return 0;
1489 }
1490
1491 static int mptcp_nl_cmd_del_addr(struct sk_buff *skb, struct genl_info *info)
1492 {
1493         struct nlattr *attr = info->attrs[MPTCP_PM_ATTR_ADDR];
1494         struct pm_nl_pernet *pernet = genl_info_pm_nl(info);
1495         struct mptcp_pm_addr_entry addr, *entry;
1496         unsigned int addr_max;
1497         int ret;
1498
1499         ret = mptcp_pm_parse_entry(attr, info, false, &addr);
1500         if (ret < 0)
1501                 return ret;
1502
1503         /* the zero id address is special: the first address used by the msk
1504          * always gets such an id, so different subflows can have different zero
1505          * id addresses. Additionally zero id is not accounted for in id_bitmap.
1506          * Let's use an 'mptcp_rm_list' instead of the common remove code.
1507          */
1508         if (addr.addr.id == 0)
1509                 return mptcp_nl_remove_id_zero_address(sock_net(skb->sk), &addr.addr);
1510
1511         spin_lock_bh(&pernet->lock);
1512         entry = __lookup_addr_by_id(pernet, addr.addr.id);
1513         if (!entry) {
1514                 GENL_SET_ERR_MSG(info, "address not found");
1515                 spin_unlock_bh(&pernet->lock);
1516                 return -EINVAL;
1517         }
1518         if (entry->flags & MPTCP_PM_ADDR_FLAG_SIGNAL) {
1519                 addr_max = pernet->add_addr_signal_max;
1520                 WRITE_ONCE(pernet->add_addr_signal_max, addr_max - 1);
1521         }
1522         if (entry->flags & MPTCP_PM_ADDR_FLAG_SUBFLOW) {
1523                 addr_max = pernet->local_addr_max;
1524                 WRITE_ONCE(pernet->local_addr_max, addr_max - 1);
1525         }
1526
1527         pernet->addrs--;
1528         list_del_rcu(&entry->list);
1529         __clear_bit(entry->addr.id, pernet->id_bitmap);
1530         spin_unlock_bh(&pernet->lock);
1531
1532         mptcp_nl_remove_subflow_and_signal_addr(sock_net(skb->sk), entry);
1533         synchronize_rcu();
1534         __mptcp_pm_release_addr_entry(entry);
1535
1536         return ret;
1537 }
1538
1539 void mptcp_pm_remove_addrs_and_subflows(struct mptcp_sock *msk,
1540                                         struct list_head *rm_list)
1541 {
1542         struct mptcp_rm_list alist = { .nr = 0 }, slist = { .nr = 0 };
1543         struct mptcp_pm_addr_entry *entry;
1544
1545         list_for_each_entry(entry, rm_list, list) {
1546                 if (lookup_subflow_by_saddr(&msk->conn_list, &entry->addr) &&
1547                     slist.nr < MPTCP_RM_IDS_MAX)
1548                         slist.ids[slist.nr++] = entry->addr.id;
1549
1550                 if (remove_anno_list_by_saddr(msk, &entry->addr) &&
1551                     alist.nr < MPTCP_RM_IDS_MAX)
1552                         alist.ids[alist.nr++] = entry->addr.id;
1553         }
1554
1555         if (alist.nr) {
1556                 spin_lock_bh(&msk->pm.lock);
1557                 mptcp_pm_remove_addr(msk, &alist);
1558                 spin_unlock_bh(&msk->pm.lock);
1559         }
1560         if (slist.nr)
1561                 mptcp_pm_remove_subflow(msk, &slist);
1562 }
1563
1564 static void mptcp_nl_remove_addrs_list(struct net *net,
1565                                        struct list_head *rm_list)
1566 {
1567         long s_slot = 0, s_num = 0;
1568         struct mptcp_sock *msk;
1569
1570         if (list_empty(rm_list))
1571                 return;
1572
1573         while ((msk = mptcp_token_iter_next(net, &s_slot, &s_num)) != NULL) {
1574                 struct sock *sk = (struct sock *)msk;
1575
1576                 if (!mptcp_pm_is_userspace(msk)) {
1577                         lock_sock(sk);
1578                         mptcp_pm_remove_addrs_and_subflows(msk, rm_list);
1579                         release_sock(sk);
1580                 }
1581
1582                 sock_put(sk);
1583                 cond_resched();
1584         }
1585 }
1586
1587 /* caller must ensure the RCU grace period is already elapsed */
1588 static void __flush_addrs(struct list_head *list)
1589 {
1590         while (!list_empty(list)) {
1591                 struct mptcp_pm_addr_entry *cur;
1592
1593                 cur = list_entry(list->next,
1594                                  struct mptcp_pm_addr_entry, list);
1595                 list_del_rcu(&cur->list);
1596                 __mptcp_pm_release_addr_entry(cur);
1597         }
1598 }
1599
1600 static void __reset_counters(struct pm_nl_pernet *pernet)
1601 {
1602         WRITE_ONCE(pernet->add_addr_signal_max, 0);
1603         WRITE_ONCE(pernet->add_addr_accept_max, 0);
1604         WRITE_ONCE(pernet->local_addr_max, 0);
1605         pernet->addrs = 0;
1606 }
1607
1608 static int mptcp_nl_cmd_flush_addrs(struct sk_buff *skb, struct genl_info *info)
1609 {
1610         struct pm_nl_pernet *pernet = genl_info_pm_nl(info);
1611         LIST_HEAD(free_list);
1612
1613         spin_lock_bh(&pernet->lock);
1614         list_splice_init(&pernet->local_addr_list, &free_list);
1615         __reset_counters(pernet);
1616         pernet->next_id = 1;
1617         bitmap_zero(pernet->id_bitmap, MPTCP_PM_MAX_ADDR_ID + 1);
1618         spin_unlock_bh(&pernet->lock);
1619         mptcp_nl_remove_addrs_list(sock_net(skb->sk), &free_list);
1620         synchronize_rcu();
1621         __flush_addrs(&free_list);
1622         return 0;
1623 }
1624
1625 static int mptcp_nl_fill_addr(struct sk_buff *skb,
1626                               struct mptcp_pm_addr_entry *entry)
1627 {
1628         struct mptcp_addr_info *addr = &entry->addr;
1629         struct nlattr *attr;
1630
1631         attr = nla_nest_start(skb, MPTCP_PM_ATTR_ADDR);
1632         if (!attr)
1633                 return -EMSGSIZE;
1634
1635         if (nla_put_u16(skb, MPTCP_PM_ADDR_ATTR_FAMILY, addr->family))
1636                 goto nla_put_failure;
1637         if (nla_put_u16(skb, MPTCP_PM_ADDR_ATTR_PORT, ntohs(addr->port)))
1638                 goto nla_put_failure;
1639         if (nla_put_u8(skb, MPTCP_PM_ADDR_ATTR_ID, addr->id))
1640                 goto nla_put_failure;
1641         if (nla_put_u32(skb, MPTCP_PM_ADDR_ATTR_FLAGS, entry->flags))
1642                 goto nla_put_failure;
1643         if (entry->ifindex &&
1644             nla_put_s32(skb, MPTCP_PM_ADDR_ATTR_IF_IDX, entry->ifindex))
1645                 goto nla_put_failure;
1646
1647         if (addr->family == AF_INET &&
1648             nla_put_in_addr(skb, MPTCP_PM_ADDR_ATTR_ADDR4,
1649                             addr->addr.s_addr))
1650                 goto nla_put_failure;
1651 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
1652         else if (addr->family == AF_INET6 &&
1653                  nla_put_in6_addr(skb, MPTCP_PM_ADDR_ATTR_ADDR6, &addr->addr6))
1654                 goto nla_put_failure;
1655 #endif
1656         nla_nest_end(skb, attr);
1657         return 0;
1658
1659 nla_put_failure:
1660         nla_nest_cancel(skb, attr);
1661         return -EMSGSIZE;
1662 }
1663
1664 static int mptcp_nl_cmd_get_addr(struct sk_buff *skb, struct genl_info *info)
1665 {
1666         struct nlattr *attr = info->attrs[MPTCP_PM_ATTR_ADDR];
1667         struct pm_nl_pernet *pernet = genl_info_pm_nl(info);
1668         struct mptcp_pm_addr_entry addr, *entry;
1669         struct sk_buff *msg;
1670         void *reply;
1671         int ret;
1672
1673         ret = mptcp_pm_parse_entry(attr, info, false, &addr);
1674         if (ret < 0)
1675                 return ret;
1676
1677         msg = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
1678         if (!msg)
1679                 return -ENOMEM;
1680
1681         reply = genlmsg_put_reply(msg, info, &mptcp_genl_family, 0,
1682                                   info->genlhdr->cmd);
1683         if (!reply) {
1684                 GENL_SET_ERR_MSG(info, "not enough space in Netlink message");
1685                 ret = -EMSGSIZE;
1686                 goto fail;
1687         }
1688
1689         spin_lock_bh(&pernet->lock);
1690         entry = __lookup_addr_by_id(pernet, addr.addr.id);
1691         if (!entry) {
1692                 GENL_SET_ERR_MSG(info, "address not found");
1693                 ret = -EINVAL;
1694                 goto unlock_fail;
1695         }
1696
1697         ret = mptcp_nl_fill_addr(msg, entry);
1698         if (ret)
1699                 goto unlock_fail;
1700
1701         genlmsg_end(msg, reply);
1702         ret = genlmsg_reply(msg, info);
1703         spin_unlock_bh(&pernet->lock);
1704         return ret;
1705
1706 unlock_fail:
1707         spin_unlock_bh(&pernet->lock);
1708
1709 fail:
1710         nlmsg_free(msg);
1711         return ret;
1712 }
1713
1714 static int mptcp_nl_cmd_dump_addrs(struct sk_buff *msg,
1715                                    struct netlink_callback *cb)
1716 {
1717         struct net *net = sock_net(msg->sk);
1718         struct mptcp_pm_addr_entry *entry;
1719         struct pm_nl_pernet *pernet;
1720         int id = cb->args[0];
1721         void *hdr;
1722         int i;
1723
1724         pernet = pm_nl_get_pernet(net);
1725
1726         spin_lock_bh(&pernet->lock);
1727         for (i = id; i < MPTCP_PM_MAX_ADDR_ID + 1; i++) {
1728                 if (test_bit(i, pernet->id_bitmap)) {
1729                         entry = __lookup_addr_by_id(pernet, i);
1730                         if (!entry)
1731                                 break;
1732
1733                         if (entry->addr.id <= id)
1734                                 continue;
1735
1736                         hdr = genlmsg_put(msg, NETLINK_CB(cb->skb).portid,
1737                                           cb->nlh->nlmsg_seq, &mptcp_genl_family,
1738                                           NLM_F_MULTI, MPTCP_PM_CMD_GET_ADDR);
1739                         if (!hdr)
1740                                 break;
1741
1742                         if (mptcp_nl_fill_addr(msg, entry) < 0) {
1743                                 genlmsg_cancel(msg, hdr);
1744                                 break;
1745                         }
1746
1747                         id = entry->addr.id;
1748                         genlmsg_end(msg, hdr);
1749                 }
1750         }
1751         spin_unlock_bh(&pernet->lock);
1752
1753         cb->args[0] = id;
1754         return msg->len;
1755 }
1756
1757 static int parse_limit(struct genl_info *info, int id, unsigned int *limit)
1758 {
1759         struct nlattr *attr = info->attrs[id];
1760
1761         if (!attr)
1762                 return 0;
1763
1764         *limit = nla_get_u32(attr);
1765         if (*limit > MPTCP_PM_ADDR_MAX) {
1766                 GENL_SET_ERR_MSG(info, "limit greater than maximum");
1767                 return -EINVAL;
1768         }
1769         return 0;
1770 }
1771
1772 static int
1773 mptcp_nl_cmd_set_limits(struct sk_buff *skb, struct genl_info *info)
1774 {
1775         struct pm_nl_pernet *pernet = genl_info_pm_nl(info);
1776         unsigned int rcv_addrs, subflows;
1777         int ret;
1778
1779         spin_lock_bh(&pernet->lock);
1780         rcv_addrs = pernet->add_addr_accept_max;
1781         ret = parse_limit(info, MPTCP_PM_ATTR_RCV_ADD_ADDRS, &rcv_addrs);
1782         if (ret)
1783                 goto unlock;
1784
1785         subflows = pernet->subflows_max;
1786         ret = parse_limit(info, MPTCP_PM_ATTR_SUBFLOWS, &subflows);
1787         if (ret)
1788                 goto unlock;
1789
1790         WRITE_ONCE(pernet->add_addr_accept_max, rcv_addrs);
1791         WRITE_ONCE(pernet->subflows_max, subflows);
1792
1793 unlock:
1794         spin_unlock_bh(&pernet->lock);
1795         return ret;
1796 }
1797
1798 static int
1799 mptcp_nl_cmd_get_limits(struct sk_buff *skb, struct genl_info *info)
1800 {
1801         struct pm_nl_pernet *pernet = genl_info_pm_nl(info);
1802         struct sk_buff *msg;
1803         void *reply;
1804
1805         msg = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
1806         if (!msg)
1807                 return -ENOMEM;
1808
1809         reply = genlmsg_put_reply(msg, info, &mptcp_genl_family, 0,
1810                                   MPTCP_PM_CMD_GET_LIMITS);
1811         if (!reply)
1812                 goto fail;
1813
1814         if (nla_put_u32(msg, MPTCP_PM_ATTR_RCV_ADD_ADDRS,
1815                         READ_ONCE(pernet->add_addr_accept_max)))
1816                 goto fail;
1817
1818         if (nla_put_u32(msg, MPTCP_PM_ATTR_SUBFLOWS,
1819                         READ_ONCE(pernet->subflows_max)))
1820                 goto fail;
1821
1822         genlmsg_end(msg, reply);
1823         return genlmsg_reply(msg, info);
1824
1825 fail:
1826         GENL_SET_ERR_MSG(info, "not enough space in Netlink message");
1827         nlmsg_free(msg);
1828         return -EMSGSIZE;
1829 }
1830
1831 static void mptcp_pm_nl_fullmesh(struct mptcp_sock *msk,
1832                                  struct mptcp_addr_info *addr)
1833 {
1834         struct mptcp_rm_list list = { .nr = 0 };
1835
1836         list.ids[list.nr++] = addr->id;
1837
1838         spin_lock_bh(&msk->pm.lock);
1839         mptcp_pm_nl_rm_subflow_received(msk, &list);
1840         mptcp_pm_create_subflow_or_signal_addr(msk);
1841         spin_unlock_bh(&msk->pm.lock);
1842 }
1843
1844 static int mptcp_nl_set_flags(struct net *net,
1845                               struct mptcp_addr_info *addr,
1846                               u8 bkup, u8 changed)
1847 {
1848         long s_slot = 0, s_num = 0;
1849         struct mptcp_sock *msk;
1850         int ret = -EINVAL;
1851
1852         while ((msk = mptcp_token_iter_next(net, &s_slot, &s_num)) != NULL) {
1853                 struct sock *sk = (struct sock *)msk;
1854
1855                 if (list_empty(&msk->conn_list) || mptcp_pm_is_userspace(msk))
1856                         goto next;
1857
1858                 lock_sock(sk);
1859                 if (changed & MPTCP_PM_ADDR_FLAG_BACKUP)
1860                         ret = mptcp_pm_nl_mp_prio_send_ack(msk, addr, NULL, bkup);
1861                 if (changed & MPTCP_PM_ADDR_FLAG_FULLMESH)
1862                         mptcp_pm_nl_fullmesh(msk, addr);
1863                 release_sock(sk);
1864
1865 next:
1866                 sock_put(sk);
1867                 cond_resched();
1868         }
1869
1870         return ret;
1871 }
1872
1873 static int mptcp_nl_cmd_set_flags(struct sk_buff *skb, struct genl_info *info)
1874 {
1875         struct mptcp_pm_addr_entry addr = { .addr = { .family = AF_UNSPEC }, }, *entry;
1876         struct mptcp_pm_addr_entry remote = { .addr = { .family = AF_UNSPEC }, };
1877         struct nlattr *attr_rem = info->attrs[MPTCP_PM_ATTR_ADDR_REMOTE];
1878         struct nlattr *token = info->attrs[MPTCP_PM_ATTR_TOKEN];
1879         struct nlattr *attr = info->attrs[MPTCP_PM_ATTR_ADDR];
1880         struct pm_nl_pernet *pernet = genl_info_pm_nl(info);
1881         u8 changed, mask = MPTCP_PM_ADDR_FLAG_BACKUP |
1882                            MPTCP_PM_ADDR_FLAG_FULLMESH;
1883         struct net *net = sock_net(skb->sk);
1884         u8 bkup = 0, lookup_by_id = 0;
1885         int ret;
1886
1887         ret = mptcp_pm_parse_entry(attr, info, false, &addr);
1888         if (ret < 0)
1889                 return ret;
1890
1891         if (attr_rem) {
1892                 ret = mptcp_pm_parse_entry(attr_rem, info, false, &remote);
1893                 if (ret < 0)
1894                         return ret;
1895         }
1896
1897         if (addr.flags & MPTCP_PM_ADDR_FLAG_BACKUP)
1898                 bkup = 1;
1899         if (addr.addr.family == AF_UNSPEC) {
1900                 lookup_by_id = 1;
1901                 if (!addr.addr.id)
1902                         return -EOPNOTSUPP;
1903         }
1904
1905         if (token)
1906                 return mptcp_userspace_pm_set_flags(sock_net(skb->sk),
1907                                                     token, &addr, &remote, bkup);
1908
1909         spin_lock_bh(&pernet->lock);
1910         entry = __lookup_addr(pernet, &addr.addr, lookup_by_id);
1911         if (!entry) {
1912                 spin_unlock_bh(&pernet->lock);
1913                 return -EINVAL;
1914         }
1915         if ((addr.flags & MPTCP_PM_ADDR_FLAG_FULLMESH) &&
1916             (entry->flags & MPTCP_PM_ADDR_FLAG_SIGNAL)) {
1917                 spin_unlock_bh(&pernet->lock);
1918                 return -EINVAL;
1919         }
1920
1921         changed = (addr.flags ^ entry->flags) & mask;
1922         entry->flags = (entry->flags & ~mask) | (addr.flags & mask);
1923         addr = *entry;
1924         spin_unlock_bh(&pernet->lock);
1925
1926         mptcp_nl_set_flags(net, &addr.addr, bkup, changed);
1927         return 0;
1928 }
1929
1930 static void mptcp_nl_mcast_send(struct net *net, struct sk_buff *nlskb, gfp_t gfp)
1931 {
1932         genlmsg_multicast_netns(&mptcp_genl_family, net,
1933                                 nlskb, 0, MPTCP_PM_EV_GRP_OFFSET, gfp);
1934 }
1935
1936 bool mptcp_userspace_pm_active(const struct mptcp_sock *msk)
1937 {
1938         return genl_has_listeners(&mptcp_genl_family,
1939                                   sock_net((const struct sock *)msk),
1940                                   MPTCP_PM_EV_GRP_OFFSET);
1941 }
1942
1943 static int mptcp_event_add_subflow(struct sk_buff *skb, const struct sock *ssk)
1944 {
1945         const struct inet_sock *issk = inet_sk(ssk);
1946         const struct mptcp_subflow_context *sf;
1947
1948         if (nla_put_u16(skb, MPTCP_ATTR_FAMILY, ssk->sk_family))
1949                 return -EMSGSIZE;
1950
1951         switch (ssk->sk_family) {
1952         case AF_INET:
1953                 if (nla_put_in_addr(skb, MPTCP_ATTR_SADDR4, issk->inet_saddr))
1954                         return -EMSGSIZE;
1955                 if (nla_put_in_addr(skb, MPTCP_ATTR_DADDR4, issk->inet_daddr))
1956                         return -EMSGSIZE;
1957                 break;
1958 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
1959         case AF_INET6: {
1960                 const struct ipv6_pinfo *np = inet6_sk(ssk);
1961
1962                 if (nla_put_in6_addr(skb, MPTCP_ATTR_SADDR6, &np->saddr))
1963                         return -EMSGSIZE;
1964                 if (nla_put_in6_addr(skb, MPTCP_ATTR_DADDR6, &ssk->sk_v6_daddr))
1965                         return -EMSGSIZE;
1966                 break;
1967         }
1968 #endif
1969         default:
1970                 WARN_ON_ONCE(1);
1971                 return -EMSGSIZE;
1972         }
1973
1974         if (nla_put_be16(skb, MPTCP_ATTR_SPORT, issk->inet_sport))
1975                 return -EMSGSIZE;
1976         if (nla_put_be16(skb, MPTCP_ATTR_DPORT, issk->inet_dport))
1977                 return -EMSGSIZE;
1978
1979         sf = mptcp_subflow_ctx(ssk);
1980         if (WARN_ON_ONCE(!sf))
1981                 return -EINVAL;
1982
1983         if (nla_put_u8(skb, MPTCP_ATTR_LOC_ID, sf->local_id))
1984                 return -EMSGSIZE;
1985
1986         if (nla_put_u8(skb, MPTCP_ATTR_REM_ID, sf->remote_id))
1987                 return -EMSGSIZE;
1988
1989         return 0;
1990 }
1991
1992 static int mptcp_event_put_token_and_ssk(struct sk_buff *skb,
1993                                          const struct mptcp_sock *msk,
1994                                          const struct sock *ssk)
1995 {
1996         const struct sock *sk = (const struct sock *)msk;
1997         const struct mptcp_subflow_context *sf;
1998         u8 sk_err;
1999
2000         if (nla_put_u32(skb, MPTCP_ATTR_TOKEN, msk->token))
2001                 return -EMSGSIZE;
2002
2003         if (mptcp_event_add_subflow(skb, ssk))
2004                 return -EMSGSIZE;
2005
2006         sf = mptcp_subflow_ctx(ssk);
2007         if (WARN_ON_ONCE(!sf))
2008                 return -EINVAL;
2009
2010         if (nla_put_u8(skb, MPTCP_ATTR_BACKUP, sf->backup))
2011                 return -EMSGSIZE;
2012
2013         if (ssk->sk_bound_dev_if &&
2014             nla_put_s32(skb, MPTCP_ATTR_IF_IDX, ssk->sk_bound_dev_if))
2015                 return -EMSGSIZE;
2016
2017         sk_err = ssk->sk_err;
2018         if (sk_err && sk->sk_state == TCP_ESTABLISHED &&
2019             nla_put_u8(skb, MPTCP_ATTR_ERROR, sk_err))
2020                 return -EMSGSIZE;
2021
2022         return 0;
2023 }
2024
2025 static int mptcp_event_sub_established(struct sk_buff *skb,
2026                                        const struct mptcp_sock *msk,
2027                                        const struct sock *ssk)
2028 {
2029         return mptcp_event_put_token_and_ssk(skb, msk, ssk);
2030 }
2031
2032 static int mptcp_event_sub_closed(struct sk_buff *skb,
2033                                   const struct mptcp_sock *msk,
2034                                   const struct sock *ssk)
2035 {
2036         const struct mptcp_subflow_context *sf;
2037
2038         if (mptcp_event_put_token_and_ssk(skb, msk, ssk))
2039                 return -EMSGSIZE;
2040
2041         sf = mptcp_subflow_ctx(ssk);
2042         if (!sf->reset_seen)
2043                 return 0;
2044
2045         if (nla_put_u32(skb, MPTCP_ATTR_RESET_REASON, sf->reset_reason))
2046                 return -EMSGSIZE;
2047
2048         if (nla_put_u32(skb, MPTCP_ATTR_RESET_FLAGS, sf->reset_transient))
2049                 return -EMSGSIZE;
2050
2051         return 0;
2052 }
2053
2054 static int mptcp_event_created(struct sk_buff *skb,
2055                                const struct mptcp_sock *msk,
2056                                const struct sock *ssk)
2057 {
2058         int err = nla_put_u32(skb, MPTCP_ATTR_TOKEN, msk->token);
2059
2060         if (err)
2061                 return err;
2062
2063         if (nla_put_u8(skb, MPTCP_ATTR_SERVER_SIDE, READ_ONCE(msk->pm.server_side)))
2064                 return -EMSGSIZE;
2065
2066         return mptcp_event_add_subflow(skb, ssk);
2067 }
2068
2069 void mptcp_event_addr_removed(const struct mptcp_sock *msk, uint8_t id)
2070 {
2071         struct net *net = sock_net((const struct sock *)msk);
2072         struct nlmsghdr *nlh;
2073         struct sk_buff *skb;
2074
2075         if (!genl_has_listeners(&mptcp_genl_family, net, MPTCP_PM_EV_GRP_OFFSET))
2076                 return;
2077
2078         skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_ATOMIC);
2079         if (!skb)
2080                 return;
2081
2082         nlh = genlmsg_put(skb, 0, 0, &mptcp_genl_family, 0, MPTCP_EVENT_REMOVED);
2083         if (!nlh)
2084                 goto nla_put_failure;
2085
2086         if (nla_put_u32(skb, MPTCP_ATTR_TOKEN, msk->token))
2087                 goto nla_put_failure;
2088
2089         if (nla_put_u8(skb, MPTCP_ATTR_REM_ID, id))
2090                 goto nla_put_failure;
2091
2092         genlmsg_end(skb, nlh);
2093         mptcp_nl_mcast_send(net, skb, GFP_ATOMIC);
2094         return;
2095
2096 nla_put_failure:
2097         nlmsg_free(skb);
2098 }
2099
2100 void mptcp_event_addr_announced(const struct sock *ssk,
2101                                 const struct mptcp_addr_info *info)
2102 {
2103         struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk);
2104         struct mptcp_sock *msk = mptcp_sk(subflow->conn);
2105         struct net *net = sock_net(ssk);
2106         struct nlmsghdr *nlh;
2107         struct sk_buff *skb;
2108
2109         if (!genl_has_listeners(&mptcp_genl_family, net, MPTCP_PM_EV_GRP_OFFSET))
2110                 return;
2111
2112         skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_ATOMIC);
2113         if (!skb)
2114                 return;
2115
2116         nlh = genlmsg_put(skb, 0, 0, &mptcp_genl_family, 0,
2117                           MPTCP_EVENT_ANNOUNCED);
2118         if (!nlh)
2119                 goto nla_put_failure;
2120
2121         if (nla_put_u32(skb, MPTCP_ATTR_TOKEN, msk->token))
2122                 goto nla_put_failure;
2123
2124         if (nla_put_u8(skb, MPTCP_ATTR_REM_ID, info->id))
2125                 goto nla_put_failure;
2126
2127         if (nla_put_be16(skb, MPTCP_ATTR_DPORT,
2128                          info->port == 0 ?
2129                          inet_sk(ssk)->inet_dport :
2130                          info->port))
2131                 goto nla_put_failure;
2132
2133         switch (info->family) {
2134         case AF_INET:
2135                 if (nla_put_in_addr(skb, MPTCP_ATTR_DADDR4, info->addr.s_addr))
2136                         goto nla_put_failure;
2137                 break;
2138 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
2139         case AF_INET6:
2140                 if (nla_put_in6_addr(skb, MPTCP_ATTR_DADDR6, &info->addr6))
2141                         goto nla_put_failure;
2142                 break;
2143 #endif
2144         default:
2145                 WARN_ON_ONCE(1);
2146                 goto nla_put_failure;
2147         }
2148
2149         genlmsg_end(skb, nlh);
2150         mptcp_nl_mcast_send(net, skb, GFP_ATOMIC);
2151         return;
2152
2153 nla_put_failure:
2154         nlmsg_free(skb);
2155 }
2156
2157 void mptcp_event_pm_listener(const struct sock *ssk,
2158                              enum mptcp_event_type event)
2159 {
2160         const struct inet_sock *issk = inet_sk(ssk);
2161         struct net *net = sock_net(ssk);
2162         struct nlmsghdr *nlh;
2163         struct sk_buff *skb;
2164
2165         if (!genl_has_listeners(&mptcp_genl_family, net, MPTCP_PM_EV_GRP_OFFSET))
2166                 return;
2167
2168         skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
2169         if (!skb)
2170                 return;
2171
2172         nlh = genlmsg_put(skb, 0, 0, &mptcp_genl_family, 0, event);
2173         if (!nlh)
2174                 goto nla_put_failure;
2175
2176         if (nla_put_u16(skb, MPTCP_ATTR_FAMILY, ssk->sk_family))
2177                 goto nla_put_failure;
2178
2179         if (nla_put_be16(skb, MPTCP_ATTR_SPORT, issk->inet_sport))
2180                 goto nla_put_failure;
2181
2182         switch (ssk->sk_family) {
2183         case AF_INET:
2184                 if (nla_put_in_addr(skb, MPTCP_ATTR_SADDR4, issk->inet_saddr))
2185                         goto nla_put_failure;
2186                 break;
2187 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
2188         case AF_INET6: {
2189                 const struct ipv6_pinfo *np = inet6_sk(ssk);
2190
2191                 if (nla_put_in6_addr(skb, MPTCP_ATTR_SADDR6, &np->saddr))
2192                         goto nla_put_failure;
2193                 break;
2194         }
2195 #endif
2196         default:
2197                 WARN_ON_ONCE(1);
2198                 goto nla_put_failure;
2199         }
2200
2201         genlmsg_end(skb, nlh);
2202         mptcp_nl_mcast_send(net, skb, GFP_KERNEL);
2203         return;
2204
2205 nla_put_failure:
2206         nlmsg_free(skb);
2207 }
2208
2209 void mptcp_event(enum mptcp_event_type type, const struct mptcp_sock *msk,
2210                  const struct sock *ssk, gfp_t gfp)
2211 {
2212         struct net *net = sock_net((const struct sock *)msk);
2213         struct nlmsghdr *nlh;
2214         struct sk_buff *skb;
2215
2216         if (!genl_has_listeners(&mptcp_genl_family, net, MPTCP_PM_EV_GRP_OFFSET))
2217                 return;
2218
2219         skb = nlmsg_new(NLMSG_DEFAULT_SIZE, gfp);
2220         if (!skb)
2221                 return;
2222
2223         nlh = genlmsg_put(skb, 0, 0, &mptcp_genl_family, 0, type);
2224         if (!nlh)
2225                 goto nla_put_failure;
2226
2227         switch (type) {
2228         case MPTCP_EVENT_UNSPEC:
2229                 WARN_ON_ONCE(1);
2230                 break;
2231         case MPTCP_EVENT_CREATED:
2232         case MPTCP_EVENT_ESTABLISHED:
2233                 if (mptcp_event_created(skb, msk, ssk) < 0)
2234                         goto nla_put_failure;
2235                 break;
2236         case MPTCP_EVENT_CLOSED:
2237                 if (nla_put_u32(skb, MPTCP_ATTR_TOKEN, msk->token) < 0)
2238                         goto nla_put_failure;
2239                 break;
2240         case MPTCP_EVENT_ANNOUNCED:
2241         case MPTCP_EVENT_REMOVED:
2242                 /* call mptcp_event_addr_announced()/removed instead */
2243                 WARN_ON_ONCE(1);
2244                 break;
2245         case MPTCP_EVENT_SUB_ESTABLISHED:
2246         case MPTCP_EVENT_SUB_PRIORITY:
2247                 if (mptcp_event_sub_established(skb, msk, ssk) < 0)
2248                         goto nla_put_failure;
2249                 break;
2250         case MPTCP_EVENT_SUB_CLOSED:
2251                 if (mptcp_event_sub_closed(skb, msk, ssk) < 0)
2252                         goto nla_put_failure;
2253                 break;
2254         case MPTCP_EVENT_LISTENER_CREATED:
2255         case MPTCP_EVENT_LISTENER_CLOSED:
2256                 break;
2257         }
2258
2259         genlmsg_end(skb, nlh);
2260         mptcp_nl_mcast_send(net, skb, gfp);
2261         return;
2262
2263 nla_put_failure:
2264         nlmsg_free(skb);
2265 }
2266
2267 static const struct genl_small_ops mptcp_pm_ops[] = {
2268         {
2269                 .cmd    = MPTCP_PM_CMD_ADD_ADDR,
2270                 .doit   = mptcp_nl_cmd_add_addr,
2271                 .flags  = GENL_UNS_ADMIN_PERM,
2272         },
2273         {
2274                 .cmd    = MPTCP_PM_CMD_DEL_ADDR,
2275                 .doit   = mptcp_nl_cmd_del_addr,
2276                 .flags  = GENL_UNS_ADMIN_PERM,
2277         },
2278         {
2279                 .cmd    = MPTCP_PM_CMD_FLUSH_ADDRS,
2280                 .doit   = mptcp_nl_cmd_flush_addrs,
2281                 .flags  = GENL_UNS_ADMIN_PERM,
2282         },
2283         {
2284                 .cmd    = MPTCP_PM_CMD_GET_ADDR,
2285                 .doit   = mptcp_nl_cmd_get_addr,
2286                 .dumpit   = mptcp_nl_cmd_dump_addrs,
2287         },
2288         {
2289                 .cmd    = MPTCP_PM_CMD_SET_LIMITS,
2290                 .doit   = mptcp_nl_cmd_set_limits,
2291                 .flags  = GENL_UNS_ADMIN_PERM,
2292         },
2293         {
2294                 .cmd    = MPTCP_PM_CMD_GET_LIMITS,
2295                 .doit   = mptcp_nl_cmd_get_limits,
2296         },
2297         {
2298                 .cmd    = MPTCP_PM_CMD_SET_FLAGS,
2299                 .doit   = mptcp_nl_cmd_set_flags,
2300                 .flags  = GENL_UNS_ADMIN_PERM,
2301         },
2302         {
2303                 .cmd    = MPTCP_PM_CMD_ANNOUNCE,
2304                 .doit   = mptcp_nl_cmd_announce,
2305                 .flags  = GENL_UNS_ADMIN_PERM,
2306         },
2307         {
2308                 .cmd    = MPTCP_PM_CMD_REMOVE,
2309                 .doit   = mptcp_nl_cmd_remove,
2310                 .flags  = GENL_UNS_ADMIN_PERM,
2311         },
2312         {
2313                 .cmd    = MPTCP_PM_CMD_SUBFLOW_CREATE,
2314                 .doit   = mptcp_nl_cmd_sf_create,
2315                 .flags  = GENL_UNS_ADMIN_PERM,
2316         },
2317         {
2318                 .cmd    = MPTCP_PM_CMD_SUBFLOW_DESTROY,
2319                 .doit   = mptcp_nl_cmd_sf_destroy,
2320                 .flags  = GENL_UNS_ADMIN_PERM,
2321         },
2322 };
2323
2324 static struct genl_family mptcp_genl_family __ro_after_init = {
2325         .name           = MPTCP_PM_NAME,
2326         .version        = MPTCP_PM_VER,
2327         .maxattr        = MPTCP_PM_ATTR_MAX,
2328         .policy         = mptcp_pm_policy,
2329         .netnsok        = true,
2330         .module         = THIS_MODULE,
2331         .small_ops      = mptcp_pm_ops,
2332         .n_small_ops    = ARRAY_SIZE(mptcp_pm_ops),
2333         .resv_start_op  = MPTCP_PM_CMD_SUBFLOW_DESTROY + 1,
2334         .mcgrps         = mptcp_pm_mcgrps,
2335         .n_mcgrps       = ARRAY_SIZE(mptcp_pm_mcgrps),
2336 };
2337
2338 static int __net_init pm_nl_init_net(struct net *net)
2339 {
2340         struct pm_nl_pernet *pernet = pm_nl_get_pernet(net);
2341
2342         INIT_LIST_HEAD_RCU(&pernet->local_addr_list);
2343
2344         /* Cit. 2 subflows ought to be enough for anybody. */
2345         pernet->subflows_max = 2;
2346         pernet->next_id = 1;
2347         pernet->stale_loss_cnt = 4;
2348         spin_lock_init(&pernet->lock);
2349
2350         /* No need to initialize other pernet fields, the struct is zeroed at
2351          * allocation time.
2352          */
2353
2354         return 0;
2355 }
2356
2357 static void __net_exit pm_nl_exit_net(struct list_head *net_list)
2358 {
2359         struct net *net;
2360
2361         list_for_each_entry(net, net_list, exit_list) {
2362                 struct pm_nl_pernet *pernet = pm_nl_get_pernet(net);
2363
2364                 /* net is removed from namespace list, can't race with
2365                  * other modifiers, also netns core already waited for a
2366                  * RCU grace period.
2367                  */
2368                 __flush_addrs(&pernet->local_addr_list);
2369         }
2370 }
2371
2372 static struct pernet_operations mptcp_pm_pernet_ops = {
2373         .init = pm_nl_init_net,
2374         .exit_batch = pm_nl_exit_net,
2375         .id = &pm_nl_pernet_id,
2376         .size = sizeof(struct pm_nl_pernet),
2377 };
2378
2379 void __init mptcp_pm_nl_init(void)
2380 {
2381         if (register_pernet_subsys(&mptcp_pm_pernet_ops) < 0)
2382                 panic("Failed to register MPTCP PM pernet subsystem.\n");
2383
2384         if (genl_register_family(&mptcp_genl_family))
2385                 panic("Failed to register MPTCP PM netlink family\n");
2386 }