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                 goto out;
 917         if (test_bit(entry->addr.id, pernet->id_bitmap))
 918                 goto out;
 919
 920         /* do not insert duplicate address, differentiate on port only
 921          * singled addresses
 922          */
 923         if (!address_use_port(entry))
 924                 entry->addr.port = 0;
 925         list_for_each_entry(cur, &pernet->local_addr_list, list) {
 926                 if (mptcp_addresses_equal(&cur->addr, &entry->addr,
 927                                           cur->addr.port || entry->addr.port)) {
 928                         /* allow replacing the exiting endpoint only if such
 929                          * endpoint is an implicit one and the user-space
 930                          * did not provide an endpoint id
 931                          */
 932                         if (!(cur->flags & MPTCP_PM_ADDR_FLAG_IMPLICIT))
 933                                 goto out;
 934                         if (entry->addr.id)
 935                                 goto out;
 936
 937                         pernet->addrs--;
 938                         entry->addr.id = cur->addr.id;
 939                         list_del_rcu(&cur->list);
 940                         del_entry = cur;
 941                         break;
 942                 }
 943         }
 944
 945         if (!entry->addr.id) {
 946 find_next:
 947                 entry->addr.id = find_next_zero_bit(pernet->id_bitmap,
 948                                                     MPTCP_PM_MAX_ADDR_ID + 1,
 949                                                     pernet->next_id);
 950                 if (!entry->addr.id && pernet->next_id != 1) {
 951                         pernet->next_id = 1;
 952                         goto find_next;
 953                 }
 954         }
 955
 956         if (!entry->addr.id)
 957                 goto out;
 958
 959         __set_bit(entry->addr.id, pernet->id_bitmap);
 960         if (entry->addr.id > pernet->next_id)
 961                 pernet->next_id = entry->addr.id;
 962
 963         if (entry->flags & MPTCP_PM_ADDR_FLAG_SIGNAL) {
 964                 addr_max = pernet->add_addr_signal_max;
 965                 WRITE_ONCE(pernet->add_addr_signal_max, addr_max + 1);
 966         }
 967         if (entry->flags & MPTCP_PM_ADDR_FLAG_SUBFLOW) {
 968                 addr_max = pernet->local_addr_max;
 969                 WRITE_ONCE(pernet->local_addr_max, addr_max + 1);
 970         }
 971
 972         pernet->addrs++;
 973         if (!entry->addr.port)
 974                 list_add_tail_rcu(&entry->list, &pernet->local_addr_list);
 975         else
 976                 list_add_rcu(&entry->list, &pernet->local_addr_list);
 977         ret = entry->addr.id;
 978
 979 out:
 980         spin_unlock_bh(&pernet->lock);
 981
 982         /* just replaced an existing entry, free it */
 983         if (del_entry) {
 984                 synchronize_rcu();
 985                 __mptcp_pm_release_addr_entry(del_entry);
 986         }
 987         return ret;
 988 }
 989
 990 static struct lock_class_key mptcp_slock_keys[2];
 991 static struct lock_class_key mptcp_keys[2];
 992
 993 static int mptcp_pm_nl_create_listen_socket(struct sock *sk,
 994                                             struct mptcp_pm_addr_entry *entry)
 995 {
 996         bool is_ipv6 = sk->sk_family == AF_INET6;
 997         int addrlen = sizeof(struct sockaddr_in);
 998         struct sockaddr_storage addr;
 999         struct socket *ssock;
1000         struct sock *newsk;
1001         int backlog = 1024;
1002         int err;
1003
1004         err = sock_create_kern(sock_net(sk), entry->addr.family,
1005                                SOCK_STREAM, IPPROTO_MPTCP, &entry->lsk);
1006         if (err)
1007                 return err;
1008
1009         newsk = entry->lsk->sk;
1010         if (!newsk)
1011                 return -EINVAL;
1012
1013         /* The subflow socket lock is acquired in a nested to the msk one
1014          * in several places, even by the TCP stack, and this msk is a kernel
1015          * socket: lockdep complains. Instead of propagating the _nested
1016          * modifiers in several places, re-init the lock class for the msk
1017          * socket to an mptcp specific one.
1018          */
1019         sock_lock_init_class_and_name(newsk,
1020                                       is_ipv6 ? "mlock-AF_INET6" : "mlock-AF_INET",
1021                                       &mptcp_slock_keys[is_ipv6],
1022                                       is_ipv6 ? "msk_lock-AF_INET6" : "msk_lock-AF_INET",
1023                                       &mptcp_keys[is_ipv6]);
1024
1025         lock_sock(newsk);
1026         ssock = __mptcp_nmpc_socket(mptcp_sk(newsk));
1027         release_sock(newsk);
1028         if (!ssock)
1029                 return -EINVAL;
1030
1031         mptcp_info2sockaddr(&entry->addr, &addr, entry->addr.family);
1032 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
1033         if (entry->addr.family == AF_INET6)
1034                 addrlen = sizeof(struct sockaddr_in6);
1035 #endif
1036         err = kernel_bind(ssock, (struct sockaddr *)&addr, addrlen);
1037         if (err) {
1038                 pr_warn("kernel_bind error, err=%d", err);
1039                 return err;
1040         }
1041
1042         inet_sk_state_store(newsk, TCP_LISTEN);
1043         err = kernel_listen(ssock, backlog);
1044         if (err) {
1045                 pr_warn("kernel_listen error, err=%d", err);
1046                 return err;
1047         }
1048
1049         return 0;
1050 }
1051
1052 int mptcp_pm_nl_get_local_id(struct mptcp_sock *msk, struct sock_common *skc)
1053 {
1054         struct mptcp_pm_addr_entry *entry;
1055         struct mptcp_addr_info skc_local;
1056         struct mptcp_addr_info msk_local;
1057         struct pm_nl_pernet *pernet;
1058         int ret = -1;
1059
1060         if (WARN_ON_ONCE(!msk))
1061                 return -1;
1062
1063         /* The 0 ID mapping is defined by the first subflow, copied into the msk
1064          * addr
1065          */
1066         local_address((struct sock_common *)msk, &msk_local);
1067         local_address((struct sock_common *)skc, &skc_local);
1068         if (mptcp_addresses_equal(&msk_local, &skc_local, false))
1069                 return 0;
1070
1071         if (mptcp_pm_is_userspace(msk))
1072                 return mptcp_userspace_pm_get_local_id(msk, &skc_local);
1073
1074         pernet = pm_nl_get_pernet_from_msk(msk);
1075
1076         rcu_read_lock();
1077         list_for_each_entry_rcu(entry, &pernet->local_addr_list, list) {
1078                 if (mptcp_addresses_equal(&entry->addr, &skc_local, entry->addr.port)) {
1079                         ret = entry->addr.id;
1080                         break;
1081                 }
1082         }
1083         rcu_read_unlock();
1084         if (ret >= 0)
1085                 return ret;
1086
1087         /* address not found, add to local list */
1088         entry = kmalloc(sizeof(*entry), GFP_ATOMIC);
1089         if (!entry)
1090                 return -ENOMEM;
1091
1092         entry->addr = skc_local;
1093         entry->addr.id = 0;
1094         entry->addr.port = 0;
1095         entry->ifindex = 0;
1096         entry->flags = MPTCP_PM_ADDR_FLAG_IMPLICIT;
1097         entry->lsk = NULL;
1098         ret = mptcp_pm_nl_append_new_local_addr(pernet, entry);
1099         if (ret < 0)
1100                 kfree(entry);
1101
1102         return ret;
1103 }
1104
1105 #define MPTCP_PM_CMD_GRP_OFFSET       0
1106 #define MPTCP_PM_EV_GRP_OFFSET        1
1107
1108 static const struct genl_multicast_group mptcp_pm_mcgrps[] = {
1109         [MPTCP_PM_CMD_GRP_OFFSET]       = { .name = MPTCP_PM_CMD_GRP_NAME, },
1110         [MPTCP_PM_EV_GRP_OFFSET]        = { .name = MPTCP_PM_EV_GRP_NAME,
1111                                             .flags = GENL_UNS_ADMIN_PERM,
1112                                           },
1113 };
1114
1115 static const struct nla_policy
1116 mptcp_pm_addr_policy[MPTCP_PM_ADDR_ATTR_MAX + 1] = {
1117         [MPTCP_PM_ADDR_ATTR_FAMILY]     = { .type       = NLA_U16,      },
1118         [MPTCP_PM_ADDR_ATTR_ID]         = { .type       = NLA_U8,       },
1119         [MPTCP_PM_ADDR_ATTR_ADDR4]      = { .type       = NLA_U32,      },
1120         [MPTCP_PM_ADDR_ATTR_ADDR6]      =
1121                 NLA_POLICY_EXACT_LEN(sizeof(struct in6_addr)),
1122         [MPTCP_PM_ADDR_ATTR_PORT]       = { .type       = NLA_U16       },
1123         [MPTCP_PM_ADDR_ATTR_FLAGS]      = { .type       = NLA_U32       },
1124         [MPTCP_PM_ADDR_ATTR_IF_IDX]     = { .type       = NLA_S32       },
1125 };
1126
1127 static const struct nla_policy mptcp_pm_policy[MPTCP_PM_ATTR_MAX + 1] = {
1128         [MPTCP_PM_ATTR_ADDR]            =
1129                                         NLA_POLICY_NESTED(mptcp_pm_addr_policy),
1130         [MPTCP_PM_ATTR_RCV_ADD_ADDRS]   = { .type       = NLA_U32,      },
1131         [MPTCP_PM_ATTR_SUBFLOWS]        = { .type       = NLA_U32,      },
1132         [MPTCP_PM_ATTR_TOKEN]           = { .type       = NLA_U32,      },
1133         [MPTCP_PM_ATTR_LOC_ID]          = { .type       = NLA_U8,       },
1134         [MPTCP_PM_ATTR_ADDR_REMOTE]     =
1135                                         NLA_POLICY_NESTED(mptcp_pm_addr_policy),
1136 };
1137
1138 void mptcp_pm_nl_subflow_chk_stale(const struct mptcp_sock *msk, struct sock *ssk)
1139 {
1140         struct mptcp_subflow_context *iter, *subflow = mptcp_subflow_ctx(ssk);
1141         struct sock *sk = (struct sock *)msk;
1142         unsigned int active_max_loss_cnt;
1143         struct net *net = sock_net(sk);
1144         unsigned int stale_loss_cnt;
1145         bool slow;
1146
1147         stale_loss_cnt = mptcp_stale_loss_cnt(net);
1148         if (subflow->stale || !stale_loss_cnt || subflow->stale_count <= stale_loss_cnt)
1149                 return;
1150
1151         /* look for another available subflow not in loss state */
1152         active_max_loss_cnt = max_t(int, stale_loss_cnt - 1, 1);
1153         mptcp_for_each_subflow(msk, iter) {
1154                 if (iter != subflow && mptcp_subflow_active(iter) &&
1155                     iter->stale_count < active_max_loss_cnt) {
1156                         /* we have some alternatives, try to mark this subflow as idle ...*/
1157                         slow = lock_sock_fast(ssk);
1158                         if (!tcp_rtx_and_write_queues_empty(ssk)) {
1159                                 subflow->stale = 1;
1160                                 __mptcp_retransmit_pending_data(sk);
1161                                 MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_SUBFLOWSTALE);
1162                         }
1163                         unlock_sock_fast(ssk, slow);
1164
1165                         /* always try to push the pending data regardless of re-injections:
1166                          * we can possibly use backup subflows now, and subflow selection
1167                          * is cheap under the msk socket lock
1168                          */
1169                         __mptcp_push_pending(sk, 0);
1170                         return;
1171                 }
1172         }
1173 }
1174
1175 static int mptcp_pm_family_to_addr(int family)
1176 {
1177 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
1178         if (family == AF_INET6)
1179                 return MPTCP_PM_ADDR_ATTR_ADDR6;
1180 #endif
1181         return MPTCP_PM_ADDR_ATTR_ADDR4;
1182 }
1183
1184 static int mptcp_pm_parse_pm_addr_attr(struct nlattr *tb[],
1185                                        const struct nlattr *attr,
1186                                        struct genl_info *info,
1187                                        struct mptcp_addr_info *addr,
1188                                        bool require_family)
1189 {
1190         int err, addr_addr;
1191
1192         if (!attr) {
1193                 GENL_SET_ERR_MSG(info, "missing address info");
1194                 return -EINVAL;
1195         }
1196
1197         /* no validation needed - was already done via nested policy */
1198         err = nla_parse_nested_deprecated(tb, MPTCP_PM_ADDR_ATTR_MAX, attr,
1199                                           mptcp_pm_addr_policy, info->extack);
1200         if (err)
1201                 return err;
1202
1203         if (tb[MPTCP_PM_ADDR_ATTR_ID])
1204                 addr->id = nla_get_u8(tb[MPTCP_PM_ADDR_ATTR_ID]);
1205
1206         if (!tb[MPTCP_PM_ADDR_ATTR_FAMILY]) {
1207                 if (!require_family)
1208                         return err;
1209
1210                 NL_SET_ERR_MSG_ATTR(info->extack, attr,
1211                                     "missing family");
1212                 return -EINVAL;
1213         }
1214
1215         addr->family = nla_get_u16(tb[MPTCP_PM_ADDR_ATTR_FAMILY]);
1216         if (addr->family != AF_INET
1217 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
1218             && addr->family != AF_INET6
1219 #endif
1220             ) {
1221                 NL_SET_ERR_MSG_ATTR(info->extack, attr,
1222                                     "unknown address family");
1223                 return -EINVAL;
1224         }
1225         addr_addr = mptcp_pm_family_to_addr(addr->family);
1226         if (!tb[addr_addr]) {
1227                 NL_SET_ERR_MSG_ATTR(info->extack, attr,
1228                                     "missing address data");
1229                 return -EINVAL;
1230         }
1231
1232 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
1233         if (addr->family == AF_INET6)
1234                 addr->addr6 = nla_get_in6_addr(tb[addr_addr]);
1235         else
1236 #endif
1237                 addr->addr.s_addr = nla_get_in_addr(tb[addr_addr]);
1238
1239         if (tb[MPTCP_PM_ADDR_ATTR_PORT])
1240                 addr->port = htons(nla_get_u16(tb[MPTCP_PM_ADDR_ATTR_PORT]));
1241
1242         return err;
1243 }
1244
1245 int mptcp_pm_parse_addr(struct nlattr *attr, struct genl_info *info,
1246                         struct mptcp_addr_info *addr)
1247 {
1248         struct nlattr *tb[MPTCP_PM_ADDR_ATTR_MAX + 1];
1249
1250         memset(addr, 0, sizeof(*addr));
1251
1252         return mptcp_pm_parse_pm_addr_attr(tb, attr, info, addr, true);
1253 }
1254
1255 int mptcp_pm_parse_entry(struct nlattr *attr, struct genl_info *info,
1256                          bool require_family,
1257                          struct mptcp_pm_addr_entry *entry)
1258 {
1259         struct nlattr *tb[MPTCP_PM_ADDR_ATTR_MAX + 1];
1260         int err;
1261
1262         memset(entry, 0, sizeof(*entry));
1263
1264         err = mptcp_pm_parse_pm_addr_attr(tb, attr, info, &entry->addr, require_family);
1265         if (err)
1266                 return err;
1267
1268         if (tb[MPTCP_PM_ADDR_ATTR_IF_IDX]) {
1269                 u32 val = nla_get_s32(tb[MPTCP_PM_ADDR_ATTR_IF_IDX]);
1270
1271                 entry->ifindex = val;
1272         }
1273
1274         if (tb[MPTCP_PM_ADDR_ATTR_FLAGS])
1275                 entry->flags = nla_get_u32(tb[MPTCP_PM_ADDR_ATTR_FLAGS]);
1276
1277         if (tb[MPTCP_PM_ADDR_ATTR_PORT])
1278                 entry->addr.port = htons(nla_get_u16(tb[MPTCP_PM_ADDR_ATTR_PORT]));
1279
1280         return 0;
1281 }
1282
1283 static struct pm_nl_pernet *genl_info_pm_nl(struct genl_info *info)
1284 {
1285         return pm_nl_get_pernet(genl_info_net(info));
1286 }
1287
1288 static int mptcp_nl_add_subflow_or_signal_addr(struct net *net)
1289 {
1290         struct mptcp_sock *msk;
1291         long s_slot = 0, s_num = 0;
1292
1293         while ((msk = mptcp_token_iter_next(net, &s_slot, &s_num)) != NULL) {
1294                 struct sock *sk = (struct sock *)msk;
1295
1296                 if (!READ_ONCE(msk->fully_established) ||
1297                     mptcp_pm_is_userspace(msk))
1298                         goto next;
1299
1300                 lock_sock(sk);
1301                 spin_lock_bh(&msk->pm.lock);
1302                 mptcp_pm_create_subflow_or_signal_addr(msk);
1303                 spin_unlock_bh(&msk->pm.lock);
1304                 release_sock(sk);
1305
1306 next:
1307                 sock_put(sk);
1308                 cond_resched();
1309         }
1310
1311         return 0;
1312 }
1313
1314 static int mptcp_nl_cmd_add_addr(struct sk_buff *skb, struct genl_info *info)
1315 {
1316         struct nlattr *attr = info->attrs[MPTCP_PM_ATTR_ADDR];
1317         struct pm_nl_pernet *pernet = genl_info_pm_nl(info);
1318         struct mptcp_pm_addr_entry addr, *entry;
1319         int ret;
1320
1321         ret = mptcp_pm_parse_entry(attr, info, true, &addr);
1322         if (ret < 0)
1323                 return ret;
1324
1325         if (addr.addr.port && !(addr.flags & MPTCP_PM_ADDR_FLAG_SIGNAL)) {
1326                 GENL_SET_ERR_MSG(info, "flags must have signal when using port");
1327                 return -EINVAL;
1328         }
1329
1330         if (addr.flags & MPTCP_PM_ADDR_FLAG_SIGNAL &&
1331             addr.flags & MPTCP_PM_ADDR_FLAG_FULLMESH) {
1332                 GENL_SET_ERR_MSG(info, "flags mustn't have both signal and fullmesh");
1333                 return -EINVAL;
1334         }
1335
1336         if (addr.flags & MPTCP_PM_ADDR_FLAG_IMPLICIT) {
1337                 GENL_SET_ERR_MSG(info, "can't create IMPLICIT endpoint");
1338                 return -EINVAL;
1339         }
1340
1341         entry = kzalloc(sizeof(*entry), GFP_KERNEL_ACCOUNT);
1342         if (!entry) {
1343                 GENL_SET_ERR_MSG(info, "can't allocate addr");
1344                 return -ENOMEM;
1345         }
1346
1347         *entry = addr;
1348         if (entry->addr.port) {
1349                 ret = mptcp_pm_nl_create_listen_socket(skb->sk, entry);
1350                 if (ret) {
1351                         GENL_SET_ERR_MSG(info, "create listen socket error");
1352                         goto out_free;
1353                 }
1354         }
1355         ret = mptcp_pm_nl_append_new_local_addr(pernet, entry);
1356         if (ret < 0) {
1357                 GENL_SET_ERR_MSG(info, "too many addresses or duplicate one");
1358                 goto out_free;
1359         }
1360
1361         mptcp_nl_add_subflow_or_signal_addr(sock_net(skb->sk));
1362         return 0;
1363
1364 out_free:
1365         __mptcp_pm_release_addr_entry(entry);
1366         return ret;
1367 }
1368
1369 int mptcp_pm_get_flags_and_ifindex_by_id(struct mptcp_sock *msk, unsigned int id,
1370                                          u8 *flags, int *ifindex)
1371 {
1372         struct mptcp_pm_addr_entry *entry;
1373         struct sock *sk = (struct sock *)msk;
1374         struct net *net = sock_net(sk);
1375
1376         *flags = 0;
1377         *ifindex = 0;
1378
1379         if (id) {
1380                 if (mptcp_pm_is_userspace(msk))
1381                         return mptcp_userspace_pm_get_flags_and_ifindex_by_id(msk,
1382                                                                               id,
1383                                                                               flags,
1384                                                                               ifindex);
1385
1386                 rcu_read_lock();
1387                 entry = __lookup_addr_by_id(pm_nl_get_pernet(net), id);
1388                 if (entry) {
1389                         *flags = entry->flags;
1390                         *ifindex = entry->ifindex;
1391                 }
1392                 rcu_read_unlock();
1393         }
1394
1395         return 0;
1396 }
1397
1398 static bool remove_anno_list_by_saddr(struct mptcp_sock *msk,
1399                                       const struct mptcp_addr_info *addr)
1400 {
1401         struct mptcp_pm_add_entry *entry;
1402
1403         entry = mptcp_pm_del_add_timer(msk, addr, false);
1404         if (entry) {
1405                 list_del(&entry->list);
1406                 kfree(entry);
1407                 return true;
1408         }
1409
1410         return false;
1411 }
1412
1413 static bool mptcp_pm_remove_anno_addr(struct mptcp_sock *msk,
1414                                       const struct mptcp_addr_info *addr,
1415                                       bool force)
1416 {
1417         struct mptcp_rm_list list = { .nr = 0 };
1418         bool ret;
1419
1420         list.ids[list.nr++] = addr->id;
1421
1422         ret = remove_anno_list_by_saddr(msk, addr);
1423         if (ret || force) {
1424                 spin_lock_bh(&msk->pm.lock);
1425                 mptcp_pm_remove_addr(msk, &list);
1426                 spin_unlock_bh(&msk->pm.lock);
1427         }
1428         return ret;
1429 }
1430
1431 static int mptcp_nl_remove_subflow_and_signal_addr(struct net *net,
1432                                                    const struct mptcp_pm_addr_entry *entry)
1433 {
1434         const struct mptcp_addr_info *addr = &entry->addr;
1435         struct mptcp_rm_list list = { .nr = 0 };
1436         long s_slot = 0, s_num = 0;
1437         struct mptcp_sock *msk;
1438
1439         pr_debug("remove_id=%d", addr->id);
1440
1441         list.ids[list.nr++] = addr->id;
1442
1443         while ((msk = mptcp_token_iter_next(net, &s_slot, &s_num)) != NULL) {
1444                 struct sock *sk = (struct sock *)msk;
1445                 bool remove_subflow;
1446
1447                 if (mptcp_pm_is_userspace(msk))
1448                         goto next;
1449
1450                 if (list_empty(&msk->conn_list)) {
1451                         mptcp_pm_remove_anno_addr(msk, addr, false);
1452                         goto next;
1453                 }
1454
1455                 lock_sock(sk);
1456                 remove_subflow = lookup_subflow_by_saddr(&msk->conn_list, addr);
1457                 mptcp_pm_remove_anno_addr(msk, addr, remove_subflow &&
1458                                           !(entry->flags & MPTCP_PM_ADDR_FLAG_IMPLICIT));
1459                 if (remove_subflow)
1460                         mptcp_pm_remove_subflow(msk, &list);
1461                 release_sock(sk);
1462
1463 next:
1464                 sock_put(sk);
1465                 cond_resched();
1466         }
1467
1468         return 0;
1469 }
1470
1471 static int mptcp_nl_remove_id_zero_address(struct net *net,
1472                                            struct mptcp_addr_info *addr)
1473 {
1474         struct mptcp_rm_list list = { .nr = 0 };
1475         long s_slot = 0, s_num = 0;
1476         struct mptcp_sock *msk;
1477
1478         list.ids[list.nr++] = 0;
1479
1480         while ((msk = mptcp_token_iter_next(net, &s_slot, &s_num)) != NULL) {
1481                 struct sock *sk = (struct sock *)msk;
1482                 struct mptcp_addr_info msk_local;
1483
1484                 if (list_empty(&msk->conn_list) || mptcp_pm_is_userspace(msk))
1485                         goto next;
1486
1487                 local_address((struct sock_common *)msk, &msk_local);
1488                 if (!mptcp_addresses_equal(&msk_local, addr, addr->port))
1489                         goto next;
1490
1491                 lock_sock(sk);
1492                 spin_lock_bh(&msk->pm.lock);
1493                 mptcp_pm_remove_addr(msk, &list);
1494                 mptcp_pm_nl_rm_subflow_received(msk, &list);
1495                 spin_unlock_bh(&msk->pm.lock);
1496                 release_sock(sk);
1497
1498 next:
1499                 sock_put(sk);
1500                 cond_resched();
1501         }
1502
1503         return 0;
1504 }
1505
1506 static int mptcp_nl_cmd_del_addr(struct sk_buff *skb, struct genl_info *info)
1507 {
1508         struct nlattr *attr = info->attrs[MPTCP_PM_ATTR_ADDR];
1509         struct pm_nl_pernet *pernet = genl_info_pm_nl(info);
1510         struct mptcp_pm_addr_entry addr, *entry;
1511         unsigned int addr_max;
1512         int ret;
1513
1514         ret = mptcp_pm_parse_entry(attr, info, false, &addr);
1515         if (ret < 0)
1516                 return ret;
1517
1518         /* the zero id address is special: the first address used by the msk
1519          * always gets such an id, so different subflows can have different zero
1520          * id addresses. Additionally zero id is not accounted for in id_bitmap.
1521          * Let's use an 'mptcp_rm_list' instead of the common remove code.
1522          */
1523         if (addr.addr.id == 0)
1524                 return mptcp_nl_remove_id_zero_address(sock_net(skb->sk), &addr.addr);
1525
1526         spin_lock_bh(&pernet->lock);
1527         entry = __lookup_addr_by_id(pernet, addr.addr.id);
1528         if (!entry) {
1529                 GENL_SET_ERR_MSG(info, "address not found");
1530                 spin_unlock_bh(&pernet->lock);
1531                 return -EINVAL;
1532         }
1533         if (entry->flags & MPTCP_PM_ADDR_FLAG_SIGNAL) {
1534                 addr_max = pernet->add_addr_signal_max;
1535                 WRITE_ONCE(pernet->add_addr_signal_max, addr_max - 1);
1536         }
1537         if (entry->flags & MPTCP_PM_ADDR_FLAG_SUBFLOW) {
1538                 addr_max = pernet->local_addr_max;
1539                 WRITE_ONCE(pernet->local_addr_max, addr_max - 1);
1540         }
1541
1542         pernet->addrs--;
1543         list_del_rcu(&entry->list);
1544         __clear_bit(entry->addr.id, pernet->id_bitmap);
1545         spin_unlock_bh(&pernet->lock);
1546
1547         mptcp_nl_remove_subflow_and_signal_addr(sock_net(skb->sk), entry);
1548         synchronize_rcu();
1549         __mptcp_pm_release_addr_entry(entry);
1550
1551         return ret;
1552 }
1553
1554 void mptcp_pm_remove_addrs(struct mptcp_sock *msk, struct list_head *rm_list)
1555 {
1556         struct mptcp_rm_list alist = { .nr = 0 };
1557         struct mptcp_pm_addr_entry *entry;
1558
1559         list_for_each_entry(entry, rm_list, list) {
1560                 remove_anno_list_by_saddr(msk, &entry->addr);
1561                 if (alist.nr < MPTCP_RM_IDS_MAX)
1562                         alist.ids[alist.nr++] = entry->addr.id;
1563         }
1564
1565         if (alist.nr) {
1566                 spin_lock_bh(&msk->pm.lock);
1567                 mptcp_pm_remove_addr(msk, &alist);
1568                 spin_unlock_bh(&msk->pm.lock);
1569         }
1570 }
1571
1572 void mptcp_pm_remove_addrs_and_subflows(struct mptcp_sock *msk,
1573                                         struct list_head *rm_list)
1574 {
1575         struct mptcp_rm_list alist = { .nr = 0 }, slist = { .nr = 0 };
1576         struct mptcp_pm_addr_entry *entry;
1577
1578         list_for_each_entry(entry, rm_list, list) {
1579                 if (lookup_subflow_by_saddr(&msk->conn_list, &entry->addr) &&
1580                     slist.nr < MPTCP_RM_IDS_MAX)
1581                         slist.ids[slist.nr++] = entry->addr.id;
1582
1583                 if (remove_anno_list_by_saddr(msk, &entry->addr) &&
1584                     alist.nr < MPTCP_RM_IDS_MAX)
1585                         alist.ids[alist.nr++] = entry->addr.id;
1586         }
1587
1588         if (alist.nr) {
1589                 spin_lock_bh(&msk->pm.lock);
1590                 mptcp_pm_remove_addr(msk, &alist);
1591                 spin_unlock_bh(&msk->pm.lock);
1592         }
1593         if (slist.nr)
1594                 mptcp_pm_remove_subflow(msk, &slist);
1595 }
1596
1597 static void mptcp_nl_remove_addrs_list(struct net *net,
1598                                        struct list_head *rm_list)
1599 {
1600         long s_slot = 0, s_num = 0;
1601         struct mptcp_sock *msk;
1602
1603         if (list_empty(rm_list))
1604                 return;
1605
1606         while ((msk = mptcp_token_iter_next(net, &s_slot, &s_num)) != NULL) {
1607                 struct sock *sk = (struct sock *)msk;
1608
1609                 if (!mptcp_pm_is_userspace(msk)) {
1610                         lock_sock(sk);
1611                         mptcp_pm_remove_addrs_and_subflows(msk, rm_list);
1612                         release_sock(sk);
1613                 }
1614
1615                 sock_put(sk);
1616                 cond_resched();
1617         }
1618 }
1619
1620 /* caller must ensure the RCU grace period is already elapsed */
1621 static void __flush_addrs(struct list_head *list)
1622 {
1623         while (!list_empty(list)) {
1624                 struct mptcp_pm_addr_entry *cur;
1625
1626                 cur = list_entry(list->next,
1627                                  struct mptcp_pm_addr_entry, list);
1628                 list_del_rcu(&cur->list);
1629                 __mptcp_pm_release_addr_entry(cur);
1630         }
1631 }
1632
1633 static void __reset_counters(struct pm_nl_pernet *pernet)
1634 {
1635         WRITE_ONCE(pernet->add_addr_signal_max, 0);
1636         WRITE_ONCE(pernet->add_addr_accept_max, 0);
1637         WRITE_ONCE(pernet->local_addr_max, 0);
1638         pernet->addrs = 0;
1639 }
1640
1641 static int mptcp_nl_cmd_flush_addrs(struct sk_buff *skb, struct genl_info *info)
1642 {
1643         struct pm_nl_pernet *pernet = genl_info_pm_nl(info);
1644         LIST_HEAD(free_list);
1645
1646         spin_lock_bh(&pernet->lock);
1647         list_splice_init(&pernet->local_addr_list, &free_list);
1648         __reset_counters(pernet);
1649         pernet->next_id = 1;
1650         bitmap_zero(pernet->id_bitmap, MPTCP_PM_MAX_ADDR_ID + 1);
1651         spin_unlock_bh(&pernet->lock);
1652         mptcp_nl_remove_addrs_list(sock_net(skb->sk), &free_list);
1653         synchronize_rcu();
1654         __flush_addrs(&free_list);
1655         return 0;
1656 }
1657
1658 static int mptcp_nl_fill_addr(struct sk_buff *skb,
1659                               struct mptcp_pm_addr_entry *entry)
1660 {
1661         struct mptcp_addr_info *addr = &entry->addr;
1662         struct nlattr *attr;
1663
1664         attr = nla_nest_start(skb, MPTCP_PM_ATTR_ADDR);
1665         if (!attr)
1666                 return -EMSGSIZE;
1667
1668         if (nla_put_u16(skb, MPTCP_PM_ADDR_ATTR_FAMILY, addr->family))
1669                 goto nla_put_failure;
1670         if (nla_put_u16(skb, MPTCP_PM_ADDR_ATTR_PORT, ntohs(addr->port)))
1671                 goto nla_put_failure;
1672         if (nla_put_u8(skb, MPTCP_PM_ADDR_ATTR_ID, addr->id))
1673                 goto nla_put_failure;
1674         if (nla_put_u32(skb, MPTCP_PM_ADDR_ATTR_FLAGS, entry->flags))
1675                 goto nla_put_failure;
1676         if (entry->ifindex &&
1677             nla_put_s32(skb, MPTCP_PM_ADDR_ATTR_IF_IDX, entry->ifindex))
1678                 goto nla_put_failure;
1679
1680         if (addr->family == AF_INET &&
1681             nla_put_in_addr(skb, MPTCP_PM_ADDR_ATTR_ADDR4,
1682                             addr->addr.s_addr))
1683                 goto nla_put_failure;
1684 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
1685         else if (addr->family == AF_INET6 &&
1686                  nla_put_in6_addr(skb, MPTCP_PM_ADDR_ATTR_ADDR6, &addr->addr6))
1687                 goto nla_put_failure;
1688 #endif
1689         nla_nest_end(skb, attr);
1690         return 0;
1691
1692 nla_put_failure:
1693         nla_nest_cancel(skb, attr);
1694         return -EMSGSIZE;
1695 }
1696
1697 static int mptcp_nl_cmd_get_addr(struct sk_buff *skb, struct genl_info *info)
1698 {
1699         struct nlattr *attr = info->attrs[MPTCP_PM_ATTR_ADDR];
1700         struct pm_nl_pernet *pernet = genl_info_pm_nl(info);
1701         struct mptcp_pm_addr_entry addr, *entry;
1702         struct sk_buff *msg;
1703         void *reply;
1704         int ret;
1705
1706         ret = mptcp_pm_parse_entry(attr, info, false, &addr);
1707         if (ret < 0)
1708                 return ret;
1709
1710         msg = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
1711         if (!msg)
1712                 return -ENOMEM;
1713
1714         reply = genlmsg_put_reply(msg, info, &mptcp_genl_family, 0,
1715                                   info->genlhdr->cmd);
1716         if (!reply) {
1717                 GENL_SET_ERR_MSG(info, "not enough space in Netlink message");
1718                 ret = -EMSGSIZE;
1719                 goto fail;
1720         }
1721
1722         spin_lock_bh(&pernet->lock);
1723         entry = __lookup_addr_by_id(pernet, addr.addr.id);
1724         if (!entry) {
1725                 GENL_SET_ERR_MSG(info, "address not found");
1726                 ret = -EINVAL;
1727                 goto unlock_fail;
1728         }
1729
1730         ret = mptcp_nl_fill_addr(msg, entry);
1731         if (ret)
1732                 goto unlock_fail;
1733
1734         genlmsg_end(msg, reply);
1735         ret = genlmsg_reply(msg, info);
1736         spin_unlock_bh(&pernet->lock);
1737         return ret;
1738
1739 unlock_fail:
1740         spin_unlock_bh(&pernet->lock);
1741
1742 fail:
1743         nlmsg_free(msg);
1744         return ret;
1745 }
1746
1747 static int mptcp_nl_cmd_dump_addrs(struct sk_buff *msg,
1748                                    struct netlink_callback *cb)
1749 {
1750         struct net *net = sock_net(msg->sk);
1751         struct mptcp_pm_addr_entry *entry;
1752         struct pm_nl_pernet *pernet;
1753         int id = cb->args[0];
1754         void *hdr;
1755         int i;
1756
1757         pernet = pm_nl_get_pernet(net);
1758
1759         spin_lock_bh(&pernet->lock);
1760         for (i = id; i < MPTCP_PM_MAX_ADDR_ID + 1; i++) {
1761                 if (test_bit(i, pernet->id_bitmap)) {
1762                         entry = __lookup_addr_by_id(pernet, i);
1763                         if (!entry)
1764                                 break;
1765
1766                         if (entry->addr.id <= id)
1767                                 continue;
1768
1769                         hdr = genlmsg_put(msg, NETLINK_CB(cb->skb).portid,
1770                                           cb->nlh->nlmsg_seq, &mptcp_genl_family,
1771                                           NLM_F_MULTI, MPTCP_PM_CMD_GET_ADDR);
1772                         if (!hdr)
1773                                 break;
1774
1775                         if (mptcp_nl_fill_addr(msg, entry) < 0) {
1776                                 genlmsg_cancel(msg, hdr);
1777                                 break;
1778                         }
1779
1780                         id = entry->addr.id;
1781                         genlmsg_end(msg, hdr);
1782                 }
1783         }
1784         spin_unlock_bh(&pernet->lock);
1785
1786         cb->args[0] = id;
1787         return msg->len;
1788 }
1789
1790 static int parse_limit(struct genl_info *info, int id, unsigned int *limit)
1791 {
1792         struct nlattr *attr = info->attrs[id];
1793
1794         if (!attr)
1795                 return 0;
1796
1797         *limit = nla_get_u32(attr);
1798         if (*limit > MPTCP_PM_ADDR_MAX) {
1799                 GENL_SET_ERR_MSG(info, "limit greater than maximum");
1800                 return -EINVAL;
1801         }
1802         return 0;
1803 }
1804
1805 static int
1806 mptcp_nl_cmd_set_limits(struct sk_buff *skb, struct genl_info *info)
1807 {
1808         struct pm_nl_pernet *pernet = genl_info_pm_nl(info);
1809         unsigned int rcv_addrs, subflows;
1810         int ret;
1811
1812         spin_lock_bh(&pernet->lock);
1813         rcv_addrs = pernet->add_addr_accept_max;
1814         ret = parse_limit(info, MPTCP_PM_ATTR_RCV_ADD_ADDRS, &rcv_addrs);
1815         if (ret)
1816                 goto unlock;
1817
1818         subflows = pernet->subflows_max;
1819         ret = parse_limit(info, MPTCP_PM_ATTR_SUBFLOWS, &subflows);
1820         if (ret)
1821                 goto unlock;
1822
1823         WRITE_ONCE(pernet->add_addr_accept_max, rcv_addrs);
1824         WRITE_ONCE(pernet->subflows_max, subflows);
1825
1826 unlock:
1827         spin_unlock_bh(&pernet->lock);
1828         return ret;
1829 }
1830
1831 static int
1832 mptcp_nl_cmd_get_limits(struct sk_buff *skb, struct genl_info *info)
1833 {
1834         struct pm_nl_pernet *pernet = genl_info_pm_nl(info);
1835         struct sk_buff *msg;
1836         void *reply;
1837
1838         msg = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
1839         if (!msg)
1840                 return -ENOMEM;
1841
1842         reply = genlmsg_put_reply(msg, info, &mptcp_genl_family, 0,
1843                                   MPTCP_PM_CMD_GET_LIMITS);
1844         if (!reply)
1845                 goto fail;
1846
1847         if (nla_put_u32(msg, MPTCP_PM_ATTR_RCV_ADD_ADDRS,
1848                         READ_ONCE(pernet->add_addr_accept_max)))
1849                 goto fail;
1850
1851         if (nla_put_u32(msg, MPTCP_PM_ATTR_SUBFLOWS,
1852                         READ_ONCE(pernet->subflows_max)))
1853                 goto fail;
1854
1855         genlmsg_end(msg, reply);
1856         return genlmsg_reply(msg, info);
1857
1858 fail:
1859         GENL_SET_ERR_MSG(info, "not enough space in Netlink message");
1860         nlmsg_free(msg);
1861         return -EMSGSIZE;
1862 }
1863
1864 static void mptcp_pm_nl_fullmesh(struct mptcp_sock *msk,
1865                                  struct mptcp_addr_info *addr)
1866 {
1867         struct mptcp_rm_list list = { .nr = 0 };
1868
1869         list.ids[list.nr++] = addr->id;
1870
1871         spin_lock_bh(&msk->pm.lock);
1872         mptcp_pm_nl_rm_subflow_received(msk, &list);
1873         mptcp_pm_create_subflow_or_signal_addr(msk);
1874         spin_unlock_bh(&msk->pm.lock);
1875 }
1876
1877 static int mptcp_nl_set_flags(struct net *net,
1878                               struct mptcp_addr_info *addr,
1879                               u8 bkup, u8 changed)
1880 {
1881         long s_slot = 0, s_num = 0;
1882         struct mptcp_sock *msk;
1883         int ret = -EINVAL;
1884
1885         while ((msk = mptcp_token_iter_next(net, &s_slot, &s_num)) != NULL) {
1886                 struct sock *sk = (struct sock *)msk;
1887
1888                 if (list_empty(&msk->conn_list) || mptcp_pm_is_userspace(msk))
1889                         goto next;
1890
1891                 lock_sock(sk);
1892                 if (changed & MPTCP_PM_ADDR_FLAG_BACKUP)
1893                         ret = mptcp_pm_nl_mp_prio_send_ack(msk, addr, NULL, bkup);
1894                 if (changed & MPTCP_PM_ADDR_FLAG_FULLMESH)
1895                         mptcp_pm_nl_fullmesh(msk, addr);
1896                 release_sock(sk);
1897
1898 next:
1899                 sock_put(sk);
1900                 cond_resched();
1901         }
1902
1903         return ret;
1904 }
1905
1906 static int mptcp_nl_cmd_set_flags(struct sk_buff *skb, struct genl_info *info)
1907 {
1908         struct mptcp_pm_addr_entry addr = { .addr = { .family = AF_UNSPEC }, }, *entry;
1909         struct mptcp_pm_addr_entry remote = { .addr = { .family = AF_UNSPEC }, };
1910         struct nlattr *attr_rem = info->attrs[MPTCP_PM_ATTR_ADDR_REMOTE];
1911         struct nlattr *token = info->attrs[MPTCP_PM_ATTR_TOKEN];
1912         struct nlattr *attr = info->attrs[MPTCP_PM_ATTR_ADDR];
1913         struct pm_nl_pernet *pernet = genl_info_pm_nl(info);
1914         u8 changed, mask = MPTCP_PM_ADDR_FLAG_BACKUP |
1915                            MPTCP_PM_ADDR_FLAG_FULLMESH;
1916         struct net *net = sock_net(skb->sk);
1917         u8 bkup = 0, lookup_by_id = 0;
1918         int ret;
1919
1920         ret = mptcp_pm_parse_entry(attr, info, false, &addr);
1921         if (ret < 0)
1922                 return ret;
1923
1924         if (attr_rem) {
1925                 ret = mptcp_pm_parse_entry(attr_rem, info, false, &remote);
1926                 if (ret < 0)
1927                         return ret;
1928         }
1929
1930         if (addr.flags & MPTCP_PM_ADDR_FLAG_BACKUP)
1931                 bkup = 1;
1932         if (addr.addr.family == AF_UNSPEC) {
1933                 lookup_by_id = 1;
1934                 if (!addr.addr.id)
1935                         return -EOPNOTSUPP;
1936         }
1937
1938         if (token)
1939                 return mptcp_userspace_pm_set_flags(sock_net(skb->sk),
1940                                                     token, &addr, &remote, bkup);
1941
1942         spin_lock_bh(&pernet->lock);
1943         entry = __lookup_addr(pernet, &addr.addr, lookup_by_id);
1944         if (!entry) {
1945                 spin_unlock_bh(&pernet->lock);
1946                 return -EINVAL;
1947         }
1948         if ((addr.flags & MPTCP_PM_ADDR_FLAG_FULLMESH) &&
1949             (entry->flags & MPTCP_PM_ADDR_FLAG_SIGNAL)) {
1950                 spin_unlock_bh(&pernet->lock);
1951                 return -EINVAL;
1952         }
1953
1954         changed = (addr.flags ^ entry->flags) & mask;
1955         entry->flags = (entry->flags & ~mask) | (addr.flags & mask);
1956         addr = *entry;
1957         spin_unlock_bh(&pernet->lock);
1958
1959         mptcp_nl_set_flags(net, &addr.addr, bkup, changed);
1960         return 0;
1961 }
1962
1963 static void mptcp_nl_mcast_send(struct net *net, struct sk_buff *nlskb, gfp_t gfp)
1964 {
1965         genlmsg_multicast_netns(&mptcp_genl_family, net,
1966                                 nlskb, 0, MPTCP_PM_EV_GRP_OFFSET, gfp);
1967 }
1968
1969 bool mptcp_userspace_pm_active(const struct mptcp_sock *msk)
1970 {
1971         return genl_has_listeners(&mptcp_genl_family,
1972                                   sock_net((const struct sock *)msk),
1973                                   MPTCP_PM_EV_GRP_OFFSET);
1974 }
1975
1976 static int mptcp_event_add_subflow(struct sk_buff *skb, const struct sock *ssk)
1977 {
1978         const struct inet_sock *issk = inet_sk(ssk);
1979         const struct mptcp_subflow_context *sf;
1980
1981         if (nla_put_u16(skb, MPTCP_ATTR_FAMILY, ssk->sk_family))
1982                 return -EMSGSIZE;
1983
1984         switch (ssk->sk_family) {
1985         case AF_INET:
1986                 if (nla_put_in_addr(skb, MPTCP_ATTR_SADDR4, issk->inet_saddr))
1987                         return -EMSGSIZE;
1988                 if (nla_put_in_addr(skb, MPTCP_ATTR_DADDR4, issk->inet_daddr))
1989                         return -EMSGSIZE;
1990                 break;
1991 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
1992         case AF_INET6: {
1993                 const struct ipv6_pinfo *np = inet6_sk(ssk);
1994
1995                 if (nla_put_in6_addr(skb, MPTCP_ATTR_SADDR6, &np->saddr))
1996                         return -EMSGSIZE;
1997                 if (nla_put_in6_addr(skb, MPTCP_ATTR_DADDR6, &ssk->sk_v6_daddr))
1998                         return -EMSGSIZE;
1999                 break;
2000         }
2001 #endif
2002         default:
2003                 WARN_ON_ONCE(1);
2004                 return -EMSGSIZE;
2005         }
2006
2007         if (nla_put_be16(skb, MPTCP_ATTR_SPORT, issk->inet_sport))
2008                 return -EMSGSIZE;
2009         if (nla_put_be16(skb, MPTCP_ATTR_DPORT, issk->inet_dport))
2010                 return -EMSGSIZE;
2011
2012         sf = mptcp_subflow_ctx(ssk);
2013         if (WARN_ON_ONCE(!sf))
2014                 return -EINVAL;
2015
2016         if (nla_put_u8(skb, MPTCP_ATTR_LOC_ID, sf->local_id))
2017                 return -EMSGSIZE;
2018
2019         if (nla_put_u8(skb, MPTCP_ATTR_REM_ID, sf->remote_id))
2020                 return -EMSGSIZE;
2021
2022         return 0;
2023 }
2024
2025 static int mptcp_event_put_token_and_ssk(struct sk_buff *skb,
2026                                          const struct mptcp_sock *msk,
2027                                          const struct sock *ssk)
2028 {
2029         const struct sock *sk = (const struct sock *)msk;
2030         const struct mptcp_subflow_context *sf;
2031         u8 sk_err;
2032
2033         if (nla_put_u32(skb, MPTCP_ATTR_TOKEN, msk->token))
2034                 return -EMSGSIZE;
2035
2036         if (mptcp_event_add_subflow(skb, ssk))
2037                 return -EMSGSIZE;
2038
2039         sf = mptcp_subflow_ctx(ssk);
2040         if (WARN_ON_ONCE(!sf))
2041                 return -EINVAL;
2042
2043         if (nla_put_u8(skb, MPTCP_ATTR_BACKUP, sf->backup))
2044                 return -EMSGSIZE;
2045
2046         if (ssk->sk_bound_dev_if &&
2047             nla_put_s32(skb, MPTCP_ATTR_IF_IDX, ssk->sk_bound_dev_if))
2048                 return -EMSGSIZE;
2049
2050         sk_err = READ_ONCE(ssk->sk_err);
2051         if (sk_err && sk->sk_state == TCP_ESTABLISHED &&
2052             nla_put_u8(skb, MPTCP_ATTR_ERROR, sk_err))
2053                 return -EMSGSIZE;
2054
2055         return 0;
2056 }
2057
2058 static int mptcp_event_sub_established(struct sk_buff *skb,
2059                                        const struct mptcp_sock *msk,
2060                                        const struct sock *ssk)
2061 {
2062         return mptcp_event_put_token_and_ssk(skb, msk, ssk);
2063 }
2064
2065 static int mptcp_event_sub_closed(struct sk_buff *skb,
2066                                   const struct mptcp_sock *msk,
2067                                   const struct sock *ssk)
2068 {
2069         const struct mptcp_subflow_context *sf;
2070
2071         if (mptcp_event_put_token_and_ssk(skb, msk, ssk))
2072                 return -EMSGSIZE;
2073
2074         sf = mptcp_subflow_ctx(ssk);
2075         if (!sf->reset_seen)
2076                 return 0;
2077
2078         if (nla_put_u32(skb, MPTCP_ATTR_RESET_REASON, sf->reset_reason))
2079                 return -EMSGSIZE;
2080
2081         if (nla_put_u32(skb, MPTCP_ATTR_RESET_FLAGS, sf->reset_transient))
2082                 return -EMSGSIZE;
2083
2084         return 0;
2085 }
2086
2087 static int mptcp_event_created(struct sk_buff *skb,
2088                                const struct mptcp_sock *msk,
2089                                const struct sock *ssk)
2090 {
2091         int err = nla_put_u32(skb, MPTCP_ATTR_TOKEN, msk->token);
2092
2093         if (err)
2094                 return err;
2095
2096         if (nla_put_u8(skb, MPTCP_ATTR_SERVER_SIDE, READ_ONCE(msk->pm.server_side)))
2097                 return -EMSGSIZE;
2098
2099         return mptcp_event_add_subflow(skb, ssk);
2100 }
2101
2102 void mptcp_event_addr_removed(const struct mptcp_sock *msk, uint8_t id)
2103 {
2104         struct net *net = sock_net((const struct sock *)msk);
2105         struct nlmsghdr *nlh;
2106         struct sk_buff *skb;
2107
2108         if (!genl_has_listeners(&mptcp_genl_family, net, MPTCP_PM_EV_GRP_OFFSET))
2109                 return;
2110
2111         skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_ATOMIC);
2112         if (!skb)
2113                 return;
2114
2115         nlh = genlmsg_put(skb, 0, 0, &mptcp_genl_family, 0, MPTCP_EVENT_REMOVED);
2116         if (!nlh)
2117                 goto nla_put_failure;
2118
2119         if (nla_put_u32(skb, MPTCP_ATTR_TOKEN, msk->token))
2120                 goto nla_put_failure;
2121
2122         if (nla_put_u8(skb, MPTCP_ATTR_REM_ID, id))
2123                 goto nla_put_failure;
2124
2125         genlmsg_end(skb, nlh);
2126         mptcp_nl_mcast_send(net, skb, GFP_ATOMIC);
2127         return;
2128
2129 nla_put_failure:
2130         kfree_skb(skb);
2131 }
2132
2133 void mptcp_event_addr_announced(const struct sock *ssk,
2134                                 const struct mptcp_addr_info *info)
2135 {
2136         struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk);
2137         struct mptcp_sock *msk = mptcp_sk(subflow->conn);
2138         struct net *net = sock_net(ssk);
2139         struct nlmsghdr *nlh;
2140         struct sk_buff *skb;
2141
2142         if (!genl_has_listeners(&mptcp_genl_family, net, MPTCP_PM_EV_GRP_OFFSET))
2143                 return;
2144
2145         skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_ATOMIC);
2146         if (!skb)
2147                 return;
2148
2149         nlh = genlmsg_put(skb, 0, 0, &mptcp_genl_family, 0,
2150                           MPTCP_EVENT_ANNOUNCED);
2151         if (!nlh)
2152                 goto nla_put_failure;
2153
2154         if (nla_put_u32(skb, MPTCP_ATTR_TOKEN, msk->token))
2155                 goto nla_put_failure;
2156
2157         if (nla_put_u8(skb, MPTCP_ATTR_REM_ID, info->id))
2158                 goto nla_put_failure;
2159
2160         if (nla_put_be16(skb, MPTCP_ATTR_DPORT,
2161                          info->port == 0 ?
2162                          inet_sk(ssk)->inet_dport :
2163                          info->port))
2164                 goto nla_put_failure;
2165
2166         switch (info->family) {
2167         case AF_INET:
2168                 if (nla_put_in_addr(skb, MPTCP_ATTR_DADDR4, info->addr.s_addr))
2169                         goto nla_put_failure;
2170                 break;
2171 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
2172         case AF_INET6:
2173                 if (nla_put_in6_addr(skb, MPTCP_ATTR_DADDR6, &info->addr6))
2174                         goto nla_put_failure;
2175                 break;
2176 #endif
2177         default:
2178                 WARN_ON_ONCE(1);
2179                 goto nla_put_failure;
2180         }
2181
2182         genlmsg_end(skb, nlh);
2183         mptcp_nl_mcast_send(net, skb, GFP_ATOMIC);
2184         return;
2185
2186 nla_put_failure:
2187         kfree_skb(skb);
2188 }
2189
2190 void mptcp_event(enum mptcp_event_type type, const struct mptcp_sock *msk,
2191                  const struct sock *ssk, gfp_t gfp)
2192 {
2193         struct net *net = sock_net((const struct sock *)msk);
2194         struct nlmsghdr *nlh;
2195         struct sk_buff *skb;
2196
2197         if (!genl_has_listeners(&mptcp_genl_family, net, MPTCP_PM_EV_GRP_OFFSET))
2198                 return;
2199
2200         skb = nlmsg_new(NLMSG_DEFAULT_SIZE, gfp);
2201         if (!skb)
2202                 return;
2203
2204         nlh = genlmsg_put(skb, 0, 0, &mptcp_genl_family, 0, type);
2205         if (!nlh)
2206                 goto nla_put_failure;
2207
2208         switch (type) {
2209         case MPTCP_EVENT_UNSPEC:
2210                 WARN_ON_ONCE(1);
2211                 break;
2212         case MPTCP_EVENT_CREATED:
2213         case MPTCP_EVENT_ESTABLISHED:
2214                 if (mptcp_event_created(skb, msk, ssk) < 0)
2215                         goto nla_put_failure;
2216                 break;
2217         case MPTCP_EVENT_CLOSED:
2218                 if (nla_put_u32(skb, MPTCP_ATTR_TOKEN, msk->token) < 0)
2219                         goto nla_put_failure;
2220                 break;
2221         case MPTCP_EVENT_ANNOUNCED:
2222         case MPTCP_EVENT_REMOVED:
2223                 /* call mptcp_event_addr_announced()/removed instead */
2224                 WARN_ON_ONCE(1);
2225                 break;
2226         case MPTCP_EVENT_SUB_ESTABLISHED:
2227         case MPTCP_EVENT_SUB_PRIORITY:
2228                 if (mptcp_event_sub_established(skb, msk, ssk) < 0)
2229                         goto nla_put_failure;
2230                 break;
2231         case MPTCP_EVENT_SUB_CLOSED:
2232                 if (mptcp_event_sub_closed(skb, msk, ssk) < 0)
2233                         goto nla_put_failure;
2234                 break;
2235         }
2236
2237         genlmsg_end(skb, nlh);
2238         mptcp_nl_mcast_send(net, skb, gfp);
2239         return;
2240
2241 nla_put_failure:
2242         kfree_skb(skb);
2243 }
2244
2245 static const struct genl_small_ops mptcp_pm_ops[] = {
2246         {
2247                 .cmd    = MPTCP_PM_CMD_ADD_ADDR,
2248                 .doit   = mptcp_nl_cmd_add_addr,
2249                 .flags  = GENL_UNS_ADMIN_PERM,
2250         },
2251         {
2252                 .cmd    = MPTCP_PM_CMD_DEL_ADDR,
2253                 .doit   = mptcp_nl_cmd_del_addr,
2254                 .flags  = GENL_UNS_ADMIN_PERM,
2255         },
2256         {
2257                 .cmd    = MPTCP_PM_CMD_FLUSH_ADDRS,
2258                 .doit   = mptcp_nl_cmd_flush_addrs,
2259                 .flags  = GENL_UNS_ADMIN_PERM,
2260         },
2261         {
2262                 .cmd    = MPTCP_PM_CMD_GET_ADDR,
2263                 .doit   = mptcp_nl_cmd_get_addr,
2264                 .dumpit   = mptcp_nl_cmd_dump_addrs,
2265         },
2266         {
2267                 .cmd    = MPTCP_PM_CMD_SET_LIMITS,
2268                 .doit   = mptcp_nl_cmd_set_limits,
2269                 .flags  = GENL_UNS_ADMIN_PERM,
2270         },
2271         {
2272                 .cmd    = MPTCP_PM_CMD_GET_LIMITS,
2273                 .doit   = mptcp_nl_cmd_get_limits,
2274         },
2275         {
2276                 .cmd    = MPTCP_PM_CMD_SET_FLAGS,
2277                 .doit   = mptcp_nl_cmd_set_flags,
2278                 .flags  = GENL_UNS_ADMIN_PERM,
2279         },
2280         {
2281                 .cmd    = MPTCP_PM_CMD_ANNOUNCE,
2282                 .doit   = mptcp_nl_cmd_announce,
2283                 .flags  = GENL_UNS_ADMIN_PERM,
2284         },
2285         {
2286                 .cmd    = MPTCP_PM_CMD_REMOVE,
2287                 .doit   = mptcp_nl_cmd_remove,
2288                 .flags  = GENL_UNS_ADMIN_PERM,
2289         },
2290         {
2291                 .cmd    = MPTCP_PM_CMD_SUBFLOW_CREATE,
2292                 .doit   = mptcp_nl_cmd_sf_create,
2293                 .flags  = GENL_UNS_ADMIN_PERM,
2294         },
2295         {
2296                 .cmd    = MPTCP_PM_CMD_SUBFLOW_DESTROY,
2297                 .doit   = mptcp_nl_cmd_sf_destroy,
2298                 .flags  = GENL_UNS_ADMIN_PERM,
2299         },
2300 };
2301
2302 static struct genl_family mptcp_genl_family __ro_after_init = {
2303         .name           = MPTCP_PM_NAME,
2304         .version        = MPTCP_PM_VER,
2305         .maxattr        = MPTCP_PM_ATTR_MAX,
2306         .policy         = mptcp_pm_policy,
2307         .netnsok        = true,
2308         .module         = THIS_MODULE,
2309         .small_ops      = mptcp_pm_ops,
2310         .n_small_ops    = ARRAY_SIZE(mptcp_pm_ops),
2311         .resv_start_op  = MPTCP_PM_CMD_SUBFLOW_DESTROY + 1,
2312         .mcgrps         = mptcp_pm_mcgrps,
2313         .n_mcgrps       = ARRAY_SIZE(mptcp_pm_mcgrps),
2314 };
2315
2316 static int __net_init pm_nl_init_net(struct net *net)
2317 {
2318         struct pm_nl_pernet *pernet = pm_nl_get_pernet(net);
2319
2320         INIT_LIST_HEAD_RCU(&pernet->local_addr_list);
2321
2322         /* Cit. 2 subflows ought to be enough for anybody. */
2323         pernet->subflows_max = 2;
2324         pernet->next_id = 1;
2325         pernet->stale_loss_cnt = 4;
2326         spin_lock_init(&pernet->lock);
2327
2328         /* No need to initialize other pernet fields, the struct is zeroed at
2329          * allocation time.
2330          */
2331
2332         return 0;
2333 }
2334
2335 static void __net_exit pm_nl_exit_net(struct list_head *net_list)
2336 {
2337         struct net *net;
2338
2339         list_for_each_entry(net, net_list, exit_list) {
2340                 struct pm_nl_pernet *pernet = pm_nl_get_pernet(net);
2341
2342                 /* net is removed from namespace list, can't race with
2343                  * other modifiers, also netns core already waited for a
2344                  * RCU grace period.
2345                  */
2346                 __flush_addrs(&pernet->local_addr_list);
2347         }
2348 }
2349
2350 static struct pernet_operations mptcp_pm_pernet_ops = {
2351         .init = pm_nl_init_net,
2352         .exit_batch = pm_nl_exit_net,
2353         .id = &pm_nl_pernet_id,
2354         .size = sizeof(struct pm_nl_pernet),
2355 };
2356
2357 void __init mptcp_pm_nl_init(void)
2358 {
2359         if (register_pernet_subsys(&mptcp_pm_pernet_ops) < 0)
2360                 panic("Failed to register MPTCP PM pernet subsystem.\n");
2361
2362         if (genl_register_family(&mptcp_genl_family))
2363                 panic("Failed to register MPTCP PM netlink family\n");
2364 }