1 // SPDX-License-Identifier: GPL-2.0 2 /* Multipath TCP 3 * 4 * Copyright (c) 2020, Red Hat, Inc. 5 */ 6 7 #define pr_fmt(fmt) "MPTCP: " fmt 8 9 #include <linux/inet.h> 10 #include <linux/kernel.h> 11 #include <net/tcp.h> 12 #include <net/netns/generic.h> 13 #include <net/mptcp.h> 14 #include <net/genetlink.h> 15 #include <uapi/linux/mptcp.h> 16 17 #include "protocol.h" 18 #include "mib.h" 19 20 /* forward declaration */ 21 static struct genl_family mptcp_genl_family; 22 23 static int pm_nl_pernet_id; 24 25 struct mptcp_pm_addr_entry { 26 struct list_head list; 27 struct mptcp_addr_info addr; 28 u8 flags; 29 int ifindex; 30 struct socket *lsk; 31 }; 32 33 struct mptcp_pm_add_entry { 34 struct list_head list; 35 struct mptcp_addr_info addr; 36 struct timer_list add_timer; 37 struct mptcp_sock *sock; 38 u8 retrans_times; 39 }; 40 41 #define MAX_ADDR_ID 255 42 #define BITMAP_SZ DIV_ROUND_UP(MAX_ADDR_ID + 1, BITS_PER_LONG) 43 44 struct pm_nl_pernet { 45 /* protects pernet updates */ 46 spinlock_t lock; 47 struct list_head local_addr_list; 48 unsigned int addrs; 49 unsigned int stale_loss_cnt; 50 unsigned int add_addr_signal_max; 51 unsigned int add_addr_accept_max; 52 unsigned int local_addr_max; 53 unsigned int subflows_max; 54 unsigned int next_id; 55 unsigned long id_bitmap[BITMAP_SZ]; 56 }; 57 58 #define MPTCP_PM_ADDR_MAX 8 59 #define ADD_ADDR_RETRANS_MAX 3 60 61 static bool addresses_equal(const struct mptcp_addr_info *a, 62 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 bool address_zero(const struct mptcp_addr_info *addr) 90 { 91 struct mptcp_addr_info zero; 92 93 memset(&zero, 0, sizeof(zero)); 94 zero.family = addr->family; 95 96 return addresses_equal(addr, &zero, true); 97 } 98 99 static void local_address(const struct sock_common *skc, 100 struct mptcp_addr_info *addr) 101 { 102 addr->family = skc->skc_family; 103 addr->port = htons(skc->skc_num); 104 if (addr->family == AF_INET) 105 addr->addr.s_addr = skc->skc_rcv_saddr; 106 #if IS_ENABLED(CONFIG_MPTCP_IPV6) 107 else if (addr->family == AF_INET6) 108 addr->addr6 = skc->skc_v6_rcv_saddr; 109 #endif 110 } 111 112 static void remote_address(const struct sock_common *skc, 113 struct mptcp_addr_info *addr) 114 { 115 addr->family = skc->skc_family; 116 addr->port = skc->skc_dport; 117 if (addr->family == AF_INET) 118 addr->addr.s_addr = skc->skc_daddr; 119 #if IS_ENABLED(CONFIG_MPTCP_IPV6) 120 else if (addr->family == AF_INET6) 121 addr->addr6 = skc->skc_v6_daddr; 122 #endif 123 } 124 125 static bool lookup_subflow_by_saddr(const struct list_head *list, 126 struct mptcp_addr_info *saddr) 127 { 128 struct mptcp_subflow_context *subflow; 129 struct mptcp_addr_info cur; 130 struct sock_common *skc; 131 132 list_for_each_entry(subflow, list, node) { 133 skc = (struct sock_common *)mptcp_subflow_tcp_sock(subflow); 134 135 local_address(skc, &cur); 136 if (addresses_equal(&cur, saddr, saddr->port)) 137 return true; 138 } 139 140 return false; 141 } 142 143 static bool lookup_subflow_by_daddr(const struct list_head *list, 144 struct mptcp_addr_info *daddr) 145 { 146 struct mptcp_subflow_context *subflow; 147 struct mptcp_addr_info cur; 148 struct sock_common *skc; 149 150 list_for_each_entry(subflow, list, node) { 151 skc = (struct sock_common *)mptcp_subflow_tcp_sock(subflow); 152 153 remote_address(skc, &cur); 154 if (addresses_equal(&cur, daddr, daddr->port)) 155 return true; 156 } 157 158 return false; 159 } 160 161 static struct mptcp_pm_addr_entry * 162 select_local_address(const struct pm_nl_pernet *pernet, 163 struct mptcp_sock *msk) 164 { 165 struct mptcp_pm_addr_entry *entry, *ret = NULL; 166 struct sock *sk = (struct sock *)msk; 167 168 msk_owned_by_me(msk); 169 170 rcu_read_lock(); 171 __mptcp_flush_join_list(msk); 172 list_for_each_entry_rcu(entry, &pernet->local_addr_list, list) { 173 if (!(entry->flags & MPTCP_PM_ADDR_FLAG_SUBFLOW)) 174 continue; 175 176 if (entry->addr.family != sk->sk_family) { 177 #if IS_ENABLED(CONFIG_MPTCP_IPV6) 178 if ((entry->addr.family == AF_INET && 179 !ipv6_addr_v4mapped(&sk->sk_v6_daddr)) || 180 (sk->sk_family == AF_INET && 181 !ipv6_addr_v4mapped(&entry->addr.addr6))) 182 #endif 183 continue; 184 } 185 186 /* avoid any address already in use by subflows and 187 * pending join 188 */ 189 if (!lookup_subflow_by_saddr(&msk->conn_list, &entry->addr)) { 190 ret = entry; 191 break; 192 } 193 } 194 rcu_read_unlock(); 195 return ret; 196 } 197 198 static struct mptcp_pm_addr_entry * 199 select_signal_address(struct pm_nl_pernet *pernet, unsigned int pos) 200 { 201 struct mptcp_pm_addr_entry *entry, *ret = NULL; 202 int i = 0; 203 204 rcu_read_lock(); 205 /* do not keep any additional per socket state, just signal 206 * the address list in order. 207 * Note: removal from the local address list during the msk life-cycle 208 * can lead to additional addresses not being announced. 209 */ 210 list_for_each_entry_rcu(entry, &pernet->local_addr_list, list) { 211 if (!(entry->flags & MPTCP_PM_ADDR_FLAG_SIGNAL)) 212 continue; 213 if (i++ == pos) { 214 ret = entry; 215 break; 216 } 217 } 218 rcu_read_unlock(); 219 return ret; 220 } 221 222 unsigned int mptcp_pm_get_add_addr_signal_max(struct mptcp_sock *msk) 223 { 224 struct pm_nl_pernet *pernet; 225 226 pernet = net_generic(sock_net((struct sock *)msk), pm_nl_pernet_id); 227 return READ_ONCE(pernet->add_addr_signal_max); 228 } 229 EXPORT_SYMBOL_GPL(mptcp_pm_get_add_addr_signal_max); 230 231 unsigned int mptcp_pm_get_add_addr_accept_max(struct mptcp_sock *msk) 232 { 233 struct pm_nl_pernet *pernet; 234 235 pernet = net_generic(sock_net((struct sock *)msk), pm_nl_pernet_id); 236 return READ_ONCE(pernet->add_addr_accept_max); 237 } 238 EXPORT_SYMBOL_GPL(mptcp_pm_get_add_addr_accept_max); 239 240 unsigned int mptcp_pm_get_subflows_max(struct mptcp_sock *msk) 241 { 242 struct pm_nl_pernet *pernet; 243 244 pernet = net_generic(sock_net((struct sock *)msk), pm_nl_pernet_id); 245 return READ_ONCE(pernet->subflows_max); 246 } 247 EXPORT_SYMBOL_GPL(mptcp_pm_get_subflows_max); 248 249 unsigned int mptcp_pm_get_local_addr_max(struct mptcp_sock *msk) 250 { 251 struct pm_nl_pernet *pernet; 252 253 pernet = net_generic(sock_net((struct sock *)msk), pm_nl_pernet_id); 254 return READ_ONCE(pernet->local_addr_max); 255 } 256 EXPORT_SYMBOL_GPL(mptcp_pm_get_local_addr_max); 257 258 static void check_work_pending(struct mptcp_sock *msk) 259 { 260 if (msk->pm.add_addr_signaled == mptcp_pm_get_add_addr_signal_max(msk) && 261 (msk->pm.local_addr_used == mptcp_pm_get_local_addr_max(msk) || 262 msk->pm.subflows == mptcp_pm_get_subflows_max(msk))) 263 WRITE_ONCE(msk->pm.work_pending, false); 264 } 265 266 struct mptcp_pm_add_entry * 267 mptcp_lookup_anno_list_by_saddr(struct mptcp_sock *msk, 268 struct mptcp_addr_info *addr) 269 { 270 struct mptcp_pm_add_entry *entry; 271 272 lockdep_assert_held(&msk->pm.lock); 273 274 list_for_each_entry(entry, &msk->pm.anno_list, list) { 275 if (addresses_equal(&entry->addr, addr, true)) 276 return entry; 277 } 278 279 return NULL; 280 } 281 282 bool mptcp_pm_sport_in_anno_list(struct mptcp_sock *msk, const struct sock *sk) 283 { 284 struct mptcp_pm_add_entry *entry; 285 struct mptcp_addr_info saddr; 286 bool ret = false; 287 288 local_address((struct sock_common *)sk, &saddr); 289 290 spin_lock_bh(&msk->pm.lock); 291 list_for_each_entry(entry, &msk->pm.anno_list, list) { 292 if (addresses_equal(&entry->addr, &saddr, true)) { 293 ret = true; 294 goto out; 295 } 296 } 297 298 out: 299 spin_unlock_bh(&msk->pm.lock); 300 return ret; 301 } 302 303 static void mptcp_pm_add_timer(struct timer_list *timer) 304 { 305 struct mptcp_pm_add_entry *entry = from_timer(entry, timer, add_timer); 306 struct mptcp_sock *msk = entry->sock; 307 struct sock *sk = (struct sock *)msk; 308 309 pr_debug("msk=%p", msk); 310 311 if (!msk) 312 return; 313 314 if (inet_sk_state_load(sk) == TCP_CLOSE) 315 return; 316 317 if (!entry->addr.id) 318 return; 319 320 if (mptcp_pm_should_add_signal(msk)) { 321 sk_reset_timer(sk, timer, jiffies + TCP_RTO_MAX / 8); 322 goto out; 323 } 324 325 spin_lock_bh(&msk->pm.lock); 326 327 if (!mptcp_pm_should_add_signal(msk)) { 328 pr_debug("retransmit ADD_ADDR id=%d", entry->addr.id); 329 mptcp_pm_announce_addr(msk, &entry->addr, false); 330 mptcp_pm_add_addr_send_ack(msk); 331 entry->retrans_times++; 332 } 333 334 if (entry->retrans_times < ADD_ADDR_RETRANS_MAX) 335 sk_reset_timer(sk, timer, 336 jiffies + mptcp_get_add_addr_timeout(sock_net(sk))); 337 338 spin_unlock_bh(&msk->pm.lock); 339 340 if (entry->retrans_times == ADD_ADDR_RETRANS_MAX) 341 mptcp_pm_subflow_established(msk); 342 343 out: 344 __sock_put(sk); 345 } 346 347 struct mptcp_pm_add_entry * 348 mptcp_pm_del_add_timer(struct mptcp_sock *msk, 349 struct mptcp_addr_info *addr, bool check_id) 350 { 351 struct mptcp_pm_add_entry *entry; 352 struct sock *sk = (struct sock *)msk; 353 354 spin_lock_bh(&msk->pm.lock); 355 entry = mptcp_lookup_anno_list_by_saddr(msk, addr); 356 if (entry && (!check_id || entry->addr.id == addr->id)) 357 entry->retrans_times = ADD_ADDR_RETRANS_MAX; 358 spin_unlock_bh(&msk->pm.lock); 359 360 if (entry && (!check_id || entry->addr.id == addr->id)) 361 sk_stop_timer_sync(sk, &entry->add_timer); 362 363 return entry; 364 } 365 366 static bool mptcp_pm_alloc_anno_list(struct mptcp_sock *msk, 367 struct mptcp_pm_addr_entry *entry) 368 { 369 struct mptcp_pm_add_entry *add_entry = NULL; 370 struct sock *sk = (struct sock *)msk; 371 struct net *net = sock_net(sk); 372 373 lockdep_assert_held(&msk->pm.lock); 374 375 if (mptcp_lookup_anno_list_by_saddr(msk, &entry->addr)) 376 return false; 377 378 add_entry = kmalloc(sizeof(*add_entry), GFP_ATOMIC); 379 if (!add_entry) 380 return false; 381 382 list_add(&add_entry->list, &msk->pm.anno_list); 383 384 add_entry->addr = entry->addr; 385 add_entry->sock = msk; 386 add_entry->retrans_times = 0; 387 388 timer_setup(&add_entry->add_timer, mptcp_pm_add_timer, 0); 389 sk_reset_timer(sk, &add_entry->add_timer, 390 jiffies + mptcp_get_add_addr_timeout(net)); 391 392 return true; 393 } 394 395 void mptcp_pm_free_anno_list(struct mptcp_sock *msk) 396 { 397 struct mptcp_pm_add_entry *entry, *tmp; 398 struct sock *sk = (struct sock *)msk; 399 LIST_HEAD(free_list); 400 401 pr_debug("msk=%p", msk); 402 403 spin_lock_bh(&msk->pm.lock); 404 list_splice_init(&msk->pm.anno_list, &free_list); 405 spin_unlock_bh(&msk->pm.lock); 406 407 list_for_each_entry_safe(entry, tmp, &free_list, list) { 408 sk_stop_timer_sync(sk, &entry->add_timer); 409 kfree(entry); 410 } 411 } 412 413 static bool lookup_address_in_vec(struct mptcp_addr_info *addrs, unsigned int nr, 414 struct mptcp_addr_info *addr) 415 { 416 int i; 417 418 for (i = 0; i < nr; i++) { 419 if (addresses_equal(&addrs[i], addr, addr->port)) 420 return true; 421 } 422 423 return false; 424 } 425 426 /* Fill all the remote addresses into the array addrs[], 427 * and return the array size. 428 */ 429 static unsigned int fill_remote_addresses_vec(struct mptcp_sock *msk, bool fullmesh, 430 struct mptcp_addr_info *addrs) 431 { 432 struct sock *sk = (struct sock *)msk, *ssk; 433 struct mptcp_subflow_context *subflow; 434 struct mptcp_addr_info remote = { 0 }; 435 unsigned int subflows_max; 436 int i = 0; 437 438 subflows_max = mptcp_pm_get_subflows_max(msk); 439 440 /* Non-fullmesh endpoint, fill in the single entry 441 * corresponding to the primary MPC subflow remote address 442 */ 443 if (!fullmesh) { 444 remote_address((struct sock_common *)sk, &remote); 445 msk->pm.subflows++; 446 addrs[i++] = remote; 447 } else { 448 mptcp_for_each_subflow(msk, subflow) { 449 ssk = mptcp_subflow_tcp_sock(subflow); 450 remote_address((struct sock_common *)ssk, &remote); 451 if (!lookup_address_in_vec(addrs, i, &remote) && 452 msk->pm.subflows < subflows_max) { 453 msk->pm.subflows++; 454 addrs[i++] = remote; 455 } 456 } 457 } 458 459 return i; 460 } 461 462 static void mptcp_pm_create_subflow_or_signal_addr(struct mptcp_sock *msk) 463 { 464 struct sock *sk = (struct sock *)msk; 465 struct mptcp_pm_addr_entry *local; 466 unsigned int add_addr_signal_max; 467 unsigned int local_addr_max; 468 struct pm_nl_pernet *pernet; 469 unsigned int subflows_max; 470 471 pernet = net_generic(sock_net(sk), pm_nl_pernet_id); 472 473 add_addr_signal_max = mptcp_pm_get_add_addr_signal_max(msk); 474 local_addr_max = mptcp_pm_get_local_addr_max(msk); 475 subflows_max = mptcp_pm_get_subflows_max(msk); 476 477 pr_debug("local %d:%d signal %d:%d subflows %d:%d\n", 478 msk->pm.local_addr_used, local_addr_max, 479 msk->pm.add_addr_signaled, add_addr_signal_max, 480 msk->pm.subflows, subflows_max); 481 482 /* check first for announce */ 483 if (msk->pm.add_addr_signaled < add_addr_signal_max) { 484 local = select_signal_address(pernet, 485 msk->pm.add_addr_signaled); 486 487 if (local) { 488 if (mptcp_pm_alloc_anno_list(msk, local)) { 489 msk->pm.add_addr_signaled++; 490 mptcp_pm_announce_addr(msk, &local->addr, false); 491 mptcp_pm_nl_addr_send_ack(msk); 492 } 493 } else { 494 /* pick failed, avoid fourther attempts later */ 495 msk->pm.local_addr_used = add_addr_signal_max; 496 } 497 498 check_work_pending(msk); 499 } 500 501 /* check if should create a new subflow */ 502 if (msk->pm.local_addr_used < local_addr_max && 503 msk->pm.subflows < subflows_max && 504 !READ_ONCE(msk->pm.remote_deny_join_id0)) { 505 local = select_local_address(pernet, msk); 506 if (local) { 507 bool fullmesh = !!(local->flags & MPTCP_PM_ADDR_FLAG_FULLMESH); 508 struct mptcp_addr_info addrs[MPTCP_PM_ADDR_MAX]; 509 int i, nr; 510 511 msk->pm.local_addr_used++; 512 check_work_pending(msk); 513 nr = fill_remote_addresses_vec(msk, fullmesh, addrs); 514 spin_unlock_bh(&msk->pm.lock); 515 for (i = 0; i < nr; i++) 516 __mptcp_subflow_connect(sk, &local->addr, &addrs[i]); 517 spin_lock_bh(&msk->pm.lock); 518 return; 519 } 520 521 /* lookup failed, avoid fourther attempts later */ 522 msk->pm.local_addr_used = local_addr_max; 523 check_work_pending(msk); 524 } 525 } 526 527 static void mptcp_pm_nl_fully_established(struct mptcp_sock *msk) 528 { 529 mptcp_pm_create_subflow_or_signal_addr(msk); 530 } 531 532 static void mptcp_pm_nl_subflow_established(struct mptcp_sock *msk) 533 { 534 mptcp_pm_create_subflow_or_signal_addr(msk); 535 } 536 537 /* Fill all the local addresses into the array addrs[], 538 * and return the array size. 539 */ 540 static unsigned int fill_local_addresses_vec(struct mptcp_sock *msk, 541 struct mptcp_addr_info *addrs) 542 { 543 struct sock *sk = (struct sock *)msk; 544 struct mptcp_pm_addr_entry *entry; 545 struct mptcp_addr_info local; 546 struct pm_nl_pernet *pernet; 547 unsigned int subflows_max; 548 int i = 0; 549 550 pernet = net_generic(sock_net(sk), pm_nl_pernet_id); 551 subflows_max = mptcp_pm_get_subflows_max(msk); 552 553 rcu_read_lock(); 554 __mptcp_flush_join_list(msk); 555 list_for_each_entry_rcu(entry, &pernet->local_addr_list, list) { 556 if (!(entry->flags & MPTCP_PM_ADDR_FLAG_FULLMESH)) 557 continue; 558 559 if (entry->addr.family != sk->sk_family) { 560 #if IS_ENABLED(CONFIG_MPTCP_IPV6) 561 if ((entry->addr.family == AF_INET && 562 !ipv6_addr_v4mapped(&sk->sk_v6_daddr)) || 563 (sk->sk_family == AF_INET && 564 !ipv6_addr_v4mapped(&entry->addr.addr6))) 565 #endif 566 continue; 567 } 568 569 if (msk->pm.subflows < subflows_max) { 570 msk->pm.subflows++; 571 addrs[i++] = entry->addr; 572 } 573 } 574 rcu_read_unlock(); 575 576 /* If the array is empty, fill in the single 577 * 'IPADDRANY' local address 578 */ 579 if (!i) { 580 memset(&local, 0, sizeof(local)); 581 local.family = msk->pm.remote.family; 582 583 msk->pm.subflows++; 584 addrs[i++] = local; 585 } 586 587 return i; 588 } 589 590 static void mptcp_pm_nl_add_addr_received(struct mptcp_sock *msk) 591 { 592 struct mptcp_addr_info addrs[MPTCP_PM_ADDR_MAX]; 593 struct sock *sk = (struct sock *)msk; 594 unsigned int add_addr_accept_max; 595 struct mptcp_addr_info remote; 596 unsigned int subflows_max; 597 int i, nr; 598 599 add_addr_accept_max = mptcp_pm_get_add_addr_accept_max(msk); 600 subflows_max = mptcp_pm_get_subflows_max(msk); 601 602 pr_debug("accepted %d:%d remote family %d", 603 msk->pm.add_addr_accepted, add_addr_accept_max, 604 msk->pm.remote.family); 605 606 if (lookup_subflow_by_daddr(&msk->conn_list, &msk->pm.remote)) 607 goto add_addr_echo; 608 609 /* connect to the specified remote address, using whatever 610 * local address the routing configuration will pick. 611 */ 612 remote = msk->pm.remote; 613 if (!remote.port) 614 remote.port = sk->sk_dport; 615 nr = fill_local_addresses_vec(msk, addrs); 616 617 msk->pm.add_addr_accepted++; 618 if (msk->pm.add_addr_accepted >= add_addr_accept_max || 619 msk->pm.subflows >= subflows_max) 620 WRITE_ONCE(msk->pm.accept_addr, false); 621 622 spin_unlock_bh(&msk->pm.lock); 623 for (i = 0; i < nr; i++) 624 __mptcp_subflow_connect(sk, &addrs[i], &remote); 625 spin_lock_bh(&msk->pm.lock); 626 627 add_addr_echo: 628 mptcp_pm_announce_addr(msk, &msk->pm.remote, true); 629 mptcp_pm_nl_addr_send_ack(msk); 630 } 631 632 void mptcp_pm_nl_addr_send_ack(struct mptcp_sock *msk) 633 { 634 struct mptcp_subflow_context *subflow; 635 636 msk_owned_by_me(msk); 637 lockdep_assert_held(&msk->pm.lock); 638 639 if (!mptcp_pm_should_add_signal(msk) && 640 !mptcp_pm_should_rm_signal(msk)) 641 return; 642 643 __mptcp_flush_join_list(msk); 644 subflow = list_first_entry_or_null(&msk->conn_list, typeof(*subflow), node); 645 if (subflow) { 646 struct sock *ssk = mptcp_subflow_tcp_sock(subflow); 647 bool slow; 648 649 spin_unlock_bh(&msk->pm.lock); 650 pr_debug("send ack for %s%s%s", 651 mptcp_pm_should_add_signal(msk) ? "add_addr" : "rm_addr", 652 mptcp_pm_should_add_signal_ipv6(msk) ? " [ipv6]" : "", 653 mptcp_pm_should_add_signal_port(msk) ? " [port]" : ""); 654 655 slow = lock_sock_fast(ssk); 656 tcp_send_ack(ssk); 657 unlock_sock_fast(ssk, slow); 658 spin_lock_bh(&msk->pm.lock); 659 } 660 } 661 662 int mptcp_pm_nl_mp_prio_send_ack(struct mptcp_sock *msk, 663 struct mptcp_addr_info *addr, 664 u8 bkup) 665 { 666 struct mptcp_subflow_context *subflow; 667 668 pr_debug("bkup=%d", bkup); 669 670 mptcp_for_each_subflow(msk, subflow) { 671 struct sock *ssk = mptcp_subflow_tcp_sock(subflow); 672 struct sock *sk = (struct sock *)msk; 673 struct mptcp_addr_info local; 674 bool slow; 675 676 local_address((struct sock_common *)ssk, &local); 677 if (!addresses_equal(&local, addr, addr->port)) 678 continue; 679 680 subflow->backup = bkup; 681 subflow->send_mp_prio = 1; 682 subflow->request_bkup = bkup; 683 __MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_MPPRIOTX); 684 685 spin_unlock_bh(&msk->pm.lock); 686 pr_debug("send ack for mp_prio"); 687 slow = lock_sock_fast(ssk); 688 tcp_send_ack(ssk); 689 unlock_sock_fast(ssk, slow); 690 spin_lock_bh(&msk->pm.lock); 691 692 return 0; 693 } 694 695 return -EINVAL; 696 } 697 698 static void mptcp_pm_nl_rm_addr_or_subflow(struct mptcp_sock *msk, 699 const struct mptcp_rm_list *rm_list, 700 enum linux_mptcp_mib_field rm_type) 701 { 702 struct mptcp_subflow_context *subflow, *tmp; 703 struct sock *sk = (struct sock *)msk; 704 u8 i; 705 706 pr_debug("%s rm_list_nr %d", 707 rm_type == MPTCP_MIB_RMADDR ? "address" : "subflow", rm_list->nr); 708 709 msk_owned_by_me(msk); 710 711 if (!rm_list->nr) 712 return; 713 714 if (list_empty(&msk->conn_list)) 715 return; 716 717 for (i = 0; i < rm_list->nr; i++) { 718 list_for_each_entry_safe(subflow, tmp, &msk->conn_list, node) { 719 struct sock *ssk = mptcp_subflow_tcp_sock(subflow); 720 int how = RCV_SHUTDOWN | SEND_SHUTDOWN; 721 u8 id = subflow->local_id; 722 723 if (rm_type == MPTCP_MIB_RMADDR) 724 id = subflow->remote_id; 725 726 if (rm_list->ids[i] != id) 727 continue; 728 729 pr_debug(" -> %s rm_list_ids[%d]=%u local_id=%u remote_id=%u", 730 rm_type == MPTCP_MIB_RMADDR ? "address" : "subflow", 731 i, rm_list->ids[i], subflow->local_id, subflow->remote_id); 732 spin_unlock_bh(&msk->pm.lock); 733 mptcp_subflow_shutdown(sk, ssk, how); 734 mptcp_close_ssk(sk, ssk, subflow); 735 spin_lock_bh(&msk->pm.lock); 736 737 if (rm_type == MPTCP_MIB_RMADDR) { 738 msk->pm.add_addr_accepted--; 739 WRITE_ONCE(msk->pm.accept_addr, true); 740 } else if (rm_type == MPTCP_MIB_RMSUBFLOW) { 741 msk->pm.local_addr_used--; 742 } 743 msk->pm.subflows--; 744 __MPTCP_INC_STATS(sock_net(sk), rm_type); 745 } 746 } 747 } 748 749 static void mptcp_pm_nl_rm_addr_received(struct mptcp_sock *msk) 750 { 751 mptcp_pm_nl_rm_addr_or_subflow(msk, &msk->pm.rm_list_rx, MPTCP_MIB_RMADDR); 752 } 753 754 void mptcp_pm_nl_rm_subflow_received(struct mptcp_sock *msk, 755 const struct mptcp_rm_list *rm_list) 756 { 757 mptcp_pm_nl_rm_addr_or_subflow(msk, rm_list, MPTCP_MIB_RMSUBFLOW); 758 } 759 760 void mptcp_pm_nl_work(struct mptcp_sock *msk) 761 { 762 struct mptcp_pm_data *pm = &msk->pm; 763 764 msk_owned_by_me(msk); 765 766 spin_lock_bh(&msk->pm.lock); 767 768 pr_debug("msk=%p status=%x", msk, pm->status); 769 if (pm->status & BIT(MPTCP_PM_ADD_ADDR_RECEIVED)) { 770 pm->status &= ~BIT(MPTCP_PM_ADD_ADDR_RECEIVED); 771 mptcp_pm_nl_add_addr_received(msk); 772 } 773 if (pm->status & BIT(MPTCP_PM_ADD_ADDR_SEND_ACK)) { 774 pm->status &= ~BIT(MPTCP_PM_ADD_ADDR_SEND_ACK); 775 mptcp_pm_nl_addr_send_ack(msk); 776 } 777 if (pm->status & BIT(MPTCP_PM_RM_ADDR_RECEIVED)) { 778 pm->status &= ~BIT(MPTCP_PM_RM_ADDR_RECEIVED); 779 mptcp_pm_nl_rm_addr_received(msk); 780 } 781 if (pm->status & BIT(MPTCP_PM_ESTABLISHED)) { 782 pm->status &= ~BIT(MPTCP_PM_ESTABLISHED); 783 mptcp_pm_nl_fully_established(msk); 784 } 785 if (pm->status & BIT(MPTCP_PM_SUBFLOW_ESTABLISHED)) { 786 pm->status &= ~BIT(MPTCP_PM_SUBFLOW_ESTABLISHED); 787 mptcp_pm_nl_subflow_established(msk); 788 } 789 790 spin_unlock_bh(&msk->pm.lock); 791 } 792 793 static bool address_use_port(struct mptcp_pm_addr_entry *entry) 794 { 795 return (entry->flags & 796 (MPTCP_PM_ADDR_FLAG_SIGNAL | MPTCP_PM_ADDR_FLAG_SUBFLOW)) == 797 MPTCP_PM_ADDR_FLAG_SIGNAL; 798 } 799 800 static int mptcp_pm_nl_append_new_local_addr(struct pm_nl_pernet *pernet, 801 struct mptcp_pm_addr_entry *entry) 802 { 803 struct mptcp_pm_addr_entry *cur; 804 unsigned int addr_max; 805 int ret = -EINVAL; 806 807 spin_lock_bh(&pernet->lock); 808 /* to keep the code simple, don't do IDR-like allocation for address ID, 809 * just bail when we exceed limits 810 */ 811 if (pernet->next_id == MAX_ADDR_ID) 812 pernet->next_id = 1; 813 if (pernet->addrs >= MPTCP_PM_ADDR_MAX) 814 goto out; 815 if (test_bit(entry->addr.id, pernet->id_bitmap)) 816 goto out; 817 818 /* do not insert duplicate address, differentiate on port only 819 * singled addresses 820 */ 821 list_for_each_entry(cur, &pernet->local_addr_list, list) { 822 if (addresses_equal(&cur->addr, &entry->addr, 823 address_use_port(entry) && 824 address_use_port(cur))) 825 goto out; 826 } 827 828 if (!entry->addr.id) { 829 find_next: 830 entry->addr.id = find_next_zero_bit(pernet->id_bitmap, 831 MAX_ADDR_ID + 1, 832 pernet->next_id); 833 if ((!entry->addr.id || entry->addr.id > MAX_ADDR_ID) && 834 pernet->next_id != 1) { 835 pernet->next_id = 1; 836 goto find_next; 837 } 838 } 839 840 if (!entry->addr.id || entry->addr.id > MAX_ADDR_ID) 841 goto out; 842 843 __set_bit(entry->addr.id, pernet->id_bitmap); 844 if (entry->addr.id > pernet->next_id) 845 pernet->next_id = entry->addr.id; 846 847 if (entry->flags & MPTCP_PM_ADDR_FLAG_SIGNAL) { 848 addr_max = pernet->add_addr_signal_max; 849 WRITE_ONCE(pernet->add_addr_signal_max, addr_max + 1); 850 } 851 if (entry->flags & MPTCP_PM_ADDR_FLAG_SUBFLOW) { 852 addr_max = pernet->local_addr_max; 853 WRITE_ONCE(pernet->local_addr_max, addr_max + 1); 854 } 855 856 pernet->addrs++; 857 list_add_tail_rcu(&entry->list, &pernet->local_addr_list); 858 ret = entry->addr.id; 859 860 out: 861 spin_unlock_bh(&pernet->lock); 862 return ret; 863 } 864 865 static int mptcp_pm_nl_create_listen_socket(struct sock *sk, 866 struct mptcp_pm_addr_entry *entry) 867 { 868 struct sockaddr_storage addr; 869 struct mptcp_sock *msk; 870 struct socket *ssock; 871 int backlog = 1024; 872 int err; 873 874 err = sock_create_kern(sock_net(sk), entry->addr.family, 875 SOCK_STREAM, IPPROTO_MPTCP, &entry->lsk); 876 if (err) 877 return err; 878 879 msk = mptcp_sk(entry->lsk->sk); 880 if (!msk) { 881 err = -EINVAL; 882 goto out; 883 } 884 885 ssock = __mptcp_nmpc_socket(msk); 886 if (!ssock) { 887 err = -EINVAL; 888 goto out; 889 } 890 891 mptcp_info2sockaddr(&entry->addr, &addr, entry->addr.family); 892 err = kernel_bind(ssock, (struct sockaddr *)&addr, 893 sizeof(struct sockaddr_in)); 894 if (err) { 895 pr_warn("kernel_bind error, err=%d", err); 896 goto out; 897 } 898 899 err = kernel_listen(ssock, backlog); 900 if (err) { 901 pr_warn("kernel_listen error, err=%d", err); 902 goto out; 903 } 904 905 return 0; 906 907 out: 908 sock_release(entry->lsk); 909 return err; 910 } 911 912 int mptcp_pm_nl_get_local_id(struct mptcp_sock *msk, struct sock_common *skc) 913 { 914 struct mptcp_pm_addr_entry *entry; 915 struct mptcp_addr_info skc_local; 916 struct mptcp_addr_info msk_local; 917 struct pm_nl_pernet *pernet; 918 int ret = -1; 919 920 if (WARN_ON_ONCE(!msk)) 921 return -1; 922 923 /* The 0 ID mapping is defined by the first subflow, copied into the msk 924 * addr 925 */ 926 local_address((struct sock_common *)msk, &msk_local); 927 local_address((struct sock_common *)skc, &skc_local); 928 if (addresses_equal(&msk_local, &skc_local, false)) 929 return 0; 930 931 if (address_zero(&skc_local)) 932 return 0; 933 934 pernet = net_generic(sock_net((struct sock *)msk), pm_nl_pernet_id); 935 936 rcu_read_lock(); 937 list_for_each_entry_rcu(entry, &pernet->local_addr_list, list) { 938 if (addresses_equal(&entry->addr, &skc_local, entry->addr.port)) { 939 ret = entry->addr.id; 940 break; 941 } 942 } 943 rcu_read_unlock(); 944 if (ret >= 0) 945 return ret; 946 947 /* address not found, add to local list */ 948 entry = kmalloc(sizeof(*entry), GFP_ATOMIC); 949 if (!entry) 950 return -ENOMEM; 951 952 entry->addr = skc_local; 953 entry->addr.id = 0; 954 entry->addr.port = 0; 955 entry->ifindex = 0; 956 entry->flags = 0; 957 entry->lsk = NULL; 958 ret = mptcp_pm_nl_append_new_local_addr(pernet, entry); 959 if (ret < 0) 960 kfree(entry); 961 962 return ret; 963 } 964 965 void mptcp_pm_nl_data_init(struct mptcp_sock *msk) 966 { 967 struct mptcp_pm_data *pm = &msk->pm; 968 bool subflows; 969 970 subflows = !!mptcp_pm_get_subflows_max(msk); 971 WRITE_ONCE(pm->work_pending, (!!mptcp_pm_get_local_addr_max(msk) && subflows) || 972 !!mptcp_pm_get_add_addr_signal_max(msk)); 973 WRITE_ONCE(pm->accept_addr, !!mptcp_pm_get_add_addr_accept_max(msk) && subflows); 974 WRITE_ONCE(pm->accept_subflow, subflows); 975 } 976 977 #define MPTCP_PM_CMD_GRP_OFFSET 0 978 #define MPTCP_PM_EV_GRP_OFFSET 1 979 980 static const struct genl_multicast_group mptcp_pm_mcgrps[] = { 981 [MPTCP_PM_CMD_GRP_OFFSET] = { .name = MPTCP_PM_CMD_GRP_NAME, }, 982 [MPTCP_PM_EV_GRP_OFFSET] = { .name = MPTCP_PM_EV_GRP_NAME, 983 .flags = GENL_UNS_ADMIN_PERM, 984 }, 985 }; 986 987 static const struct nla_policy 988 mptcp_pm_addr_policy[MPTCP_PM_ADDR_ATTR_MAX + 1] = { 989 [MPTCP_PM_ADDR_ATTR_FAMILY] = { .type = NLA_U16, }, 990 [MPTCP_PM_ADDR_ATTR_ID] = { .type = NLA_U8, }, 991 [MPTCP_PM_ADDR_ATTR_ADDR4] = { .type = NLA_U32, }, 992 [MPTCP_PM_ADDR_ATTR_ADDR6] = 993 NLA_POLICY_EXACT_LEN(sizeof(struct in6_addr)), 994 [MPTCP_PM_ADDR_ATTR_PORT] = { .type = NLA_U16 }, 995 [MPTCP_PM_ADDR_ATTR_FLAGS] = { .type = NLA_U32 }, 996 [MPTCP_PM_ADDR_ATTR_IF_IDX] = { .type = NLA_S32 }, 997 }; 998 999 static const struct nla_policy mptcp_pm_policy[MPTCP_PM_ATTR_MAX + 1] = { 1000 [MPTCP_PM_ATTR_ADDR] = 1001 NLA_POLICY_NESTED(mptcp_pm_addr_policy), 1002 [MPTCP_PM_ATTR_RCV_ADD_ADDRS] = { .type = NLA_U32, }, 1003 [MPTCP_PM_ATTR_SUBFLOWS] = { .type = NLA_U32, }, 1004 }; 1005 1006 void mptcp_pm_nl_subflow_chk_stale(const struct mptcp_sock *msk, struct sock *ssk) 1007 { 1008 struct mptcp_subflow_context *iter, *subflow = mptcp_subflow_ctx(ssk); 1009 struct sock *sk = (struct sock *)msk; 1010 unsigned int active_max_loss_cnt; 1011 struct net *net = sock_net(sk); 1012 unsigned int stale_loss_cnt; 1013 bool slow; 1014 1015 stale_loss_cnt = mptcp_stale_loss_cnt(net); 1016 if (subflow->stale || !stale_loss_cnt || subflow->stale_count <= stale_loss_cnt) 1017 return; 1018 1019 /* look for another available subflow not in loss state */ 1020 active_max_loss_cnt = max_t(int, stale_loss_cnt - 1, 1); 1021 mptcp_for_each_subflow(msk, iter) { 1022 if (iter != subflow && mptcp_subflow_active(iter) && 1023 iter->stale_count < active_max_loss_cnt) { 1024 /* we have some alternatives, try to mark this subflow as idle ...*/ 1025 slow = lock_sock_fast(ssk); 1026 if (!tcp_rtx_and_write_queues_empty(ssk)) { 1027 subflow->stale = 1; 1028 __mptcp_retransmit_pending_data(sk); 1029 MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_SUBFLOWSTALE); 1030 } 1031 unlock_sock_fast(ssk, slow); 1032 1033 /* always try to push the pending data regarless of re-injections: 1034 * we can possibly use backup subflows now, and subflow selection 1035 * is cheap under the msk socket lock 1036 */ 1037 __mptcp_push_pending(sk, 0); 1038 return; 1039 } 1040 } 1041 } 1042 1043 static int mptcp_pm_family_to_addr(int family) 1044 { 1045 #if IS_ENABLED(CONFIG_MPTCP_IPV6) 1046 if (family == AF_INET6) 1047 return MPTCP_PM_ADDR_ATTR_ADDR6; 1048 #endif 1049 return MPTCP_PM_ADDR_ATTR_ADDR4; 1050 } 1051 1052 static int mptcp_pm_parse_addr(struct nlattr *attr, struct genl_info *info, 1053 bool require_family, 1054 struct mptcp_pm_addr_entry *entry) 1055 { 1056 struct nlattr *tb[MPTCP_PM_ADDR_ATTR_MAX + 1]; 1057 int err, addr_addr; 1058 1059 if (!attr) { 1060 GENL_SET_ERR_MSG(info, "missing address info"); 1061 return -EINVAL; 1062 } 1063 1064 /* no validation needed - was already done via nested policy */ 1065 err = nla_parse_nested_deprecated(tb, MPTCP_PM_ADDR_ATTR_MAX, attr, 1066 mptcp_pm_addr_policy, info->extack); 1067 if (err) 1068 return err; 1069 1070 memset(entry, 0, sizeof(*entry)); 1071 if (!tb[MPTCP_PM_ADDR_ATTR_FAMILY]) { 1072 if (!require_family) 1073 goto skip_family; 1074 1075 NL_SET_ERR_MSG_ATTR(info->extack, attr, 1076 "missing family"); 1077 return -EINVAL; 1078 } 1079 1080 entry->addr.family = nla_get_u16(tb[MPTCP_PM_ADDR_ATTR_FAMILY]); 1081 if (entry->addr.family != AF_INET 1082 #if IS_ENABLED(CONFIG_MPTCP_IPV6) 1083 && entry->addr.family != AF_INET6 1084 #endif 1085 ) { 1086 NL_SET_ERR_MSG_ATTR(info->extack, attr, 1087 "unknown address family"); 1088 return -EINVAL; 1089 } 1090 addr_addr = mptcp_pm_family_to_addr(entry->addr.family); 1091 if (!tb[addr_addr]) { 1092 NL_SET_ERR_MSG_ATTR(info->extack, attr, 1093 "missing address data"); 1094 return -EINVAL; 1095 } 1096 1097 #if IS_ENABLED(CONFIG_MPTCP_IPV6) 1098 if (entry->addr.family == AF_INET6) 1099 entry->addr.addr6 = nla_get_in6_addr(tb[addr_addr]); 1100 else 1101 #endif 1102 entry->addr.addr.s_addr = nla_get_in_addr(tb[addr_addr]); 1103 1104 skip_family: 1105 if (tb[MPTCP_PM_ADDR_ATTR_IF_IDX]) { 1106 u32 val = nla_get_s32(tb[MPTCP_PM_ADDR_ATTR_IF_IDX]); 1107 1108 entry->ifindex = val; 1109 } 1110 1111 if (tb[MPTCP_PM_ADDR_ATTR_ID]) 1112 entry->addr.id = nla_get_u8(tb[MPTCP_PM_ADDR_ATTR_ID]); 1113 1114 if (tb[MPTCP_PM_ADDR_ATTR_FLAGS]) 1115 entry->flags = nla_get_u32(tb[MPTCP_PM_ADDR_ATTR_FLAGS]); 1116 1117 if (tb[MPTCP_PM_ADDR_ATTR_PORT]) { 1118 if (!(entry->flags & MPTCP_PM_ADDR_FLAG_SIGNAL)) { 1119 NL_SET_ERR_MSG_ATTR(info->extack, attr, 1120 "flags must have signal when using port"); 1121 return -EINVAL; 1122 } 1123 entry->addr.port = htons(nla_get_u16(tb[MPTCP_PM_ADDR_ATTR_PORT])); 1124 } 1125 1126 return 0; 1127 } 1128 1129 static struct pm_nl_pernet *genl_info_pm_nl(struct genl_info *info) 1130 { 1131 return net_generic(genl_info_net(info), pm_nl_pernet_id); 1132 } 1133 1134 static int mptcp_nl_add_subflow_or_signal_addr(struct net *net) 1135 { 1136 struct mptcp_sock *msk; 1137 long s_slot = 0, s_num = 0; 1138 1139 while ((msk = mptcp_token_iter_next(net, &s_slot, &s_num)) != NULL) { 1140 struct sock *sk = (struct sock *)msk; 1141 1142 if (!READ_ONCE(msk->fully_established)) 1143 goto next; 1144 1145 lock_sock(sk); 1146 spin_lock_bh(&msk->pm.lock); 1147 mptcp_pm_create_subflow_or_signal_addr(msk); 1148 spin_unlock_bh(&msk->pm.lock); 1149 release_sock(sk); 1150 1151 next: 1152 sock_put(sk); 1153 cond_resched(); 1154 } 1155 1156 return 0; 1157 } 1158 1159 static int mptcp_nl_cmd_add_addr(struct sk_buff *skb, struct genl_info *info) 1160 { 1161 struct nlattr *attr = info->attrs[MPTCP_PM_ATTR_ADDR]; 1162 struct pm_nl_pernet *pernet = genl_info_pm_nl(info); 1163 struct mptcp_pm_addr_entry addr, *entry; 1164 int ret; 1165 1166 ret = mptcp_pm_parse_addr(attr, info, true, &addr); 1167 if (ret < 0) 1168 return ret; 1169 1170 entry = kmalloc(sizeof(*entry), GFP_KERNEL); 1171 if (!entry) { 1172 GENL_SET_ERR_MSG(info, "can't allocate addr"); 1173 return -ENOMEM; 1174 } 1175 1176 *entry = addr; 1177 if (entry->addr.port) { 1178 ret = mptcp_pm_nl_create_listen_socket(skb->sk, entry); 1179 if (ret) { 1180 GENL_SET_ERR_MSG(info, "create listen socket error"); 1181 kfree(entry); 1182 return ret; 1183 } 1184 } 1185 ret = mptcp_pm_nl_append_new_local_addr(pernet, entry); 1186 if (ret < 0) { 1187 GENL_SET_ERR_MSG(info, "too many addresses or duplicate one"); 1188 if (entry->lsk) 1189 sock_release(entry->lsk); 1190 kfree(entry); 1191 return ret; 1192 } 1193 1194 mptcp_nl_add_subflow_or_signal_addr(sock_net(skb->sk)); 1195 1196 return 0; 1197 } 1198 1199 static struct mptcp_pm_addr_entry * 1200 __lookup_addr_by_id(struct pm_nl_pernet *pernet, unsigned int id) 1201 { 1202 struct mptcp_pm_addr_entry *entry; 1203 1204 list_for_each_entry(entry, &pernet->local_addr_list, list) { 1205 if (entry->addr.id == id) 1206 return entry; 1207 } 1208 return NULL; 1209 } 1210 1211 int mptcp_pm_get_flags_and_ifindex_by_id(struct net *net, unsigned int id, 1212 u8 *flags, int *ifindex) 1213 { 1214 struct mptcp_pm_addr_entry *entry; 1215 1216 *flags = 0; 1217 *ifindex = 0; 1218 1219 if (id) { 1220 rcu_read_lock(); 1221 entry = __lookup_addr_by_id(net_generic(net, pm_nl_pernet_id), id); 1222 if (entry) { 1223 *flags = entry->flags; 1224 *ifindex = entry->ifindex; 1225 } 1226 rcu_read_unlock(); 1227 } 1228 1229 return 0; 1230 } 1231 1232 static bool remove_anno_list_by_saddr(struct mptcp_sock *msk, 1233 struct mptcp_addr_info *addr) 1234 { 1235 struct mptcp_pm_add_entry *entry; 1236 1237 entry = mptcp_pm_del_add_timer(msk, addr, false); 1238 if (entry) { 1239 list_del(&entry->list); 1240 kfree(entry); 1241 return true; 1242 } 1243 1244 return false; 1245 } 1246 1247 static bool mptcp_pm_remove_anno_addr(struct mptcp_sock *msk, 1248 struct mptcp_addr_info *addr, 1249 bool force) 1250 { 1251 struct mptcp_rm_list list = { .nr = 0 }; 1252 bool ret; 1253 1254 list.ids[list.nr++] = addr->id; 1255 1256 ret = remove_anno_list_by_saddr(msk, addr); 1257 if (ret || force) { 1258 spin_lock_bh(&msk->pm.lock); 1259 mptcp_pm_remove_addr(msk, &list); 1260 spin_unlock_bh(&msk->pm.lock); 1261 } 1262 return ret; 1263 } 1264 1265 static int mptcp_nl_remove_subflow_and_signal_addr(struct net *net, 1266 struct mptcp_addr_info *addr) 1267 { 1268 struct mptcp_sock *msk; 1269 long s_slot = 0, s_num = 0; 1270 struct mptcp_rm_list list = { .nr = 0 }; 1271 1272 pr_debug("remove_id=%d", addr->id); 1273 1274 list.ids[list.nr++] = addr->id; 1275 1276 while ((msk = mptcp_token_iter_next(net, &s_slot, &s_num)) != NULL) { 1277 struct sock *sk = (struct sock *)msk; 1278 bool remove_subflow; 1279 1280 if (list_empty(&msk->conn_list)) { 1281 mptcp_pm_remove_anno_addr(msk, addr, false); 1282 goto next; 1283 } 1284 1285 lock_sock(sk); 1286 remove_subflow = lookup_subflow_by_saddr(&msk->conn_list, addr); 1287 mptcp_pm_remove_anno_addr(msk, addr, remove_subflow); 1288 if (remove_subflow) 1289 mptcp_pm_remove_subflow(msk, &list); 1290 release_sock(sk); 1291 1292 next: 1293 sock_put(sk); 1294 cond_resched(); 1295 } 1296 1297 return 0; 1298 } 1299 1300 /* caller must ensure the RCU grace period is already elapsed */ 1301 static void __mptcp_pm_release_addr_entry(struct mptcp_pm_addr_entry *entry) 1302 { 1303 if (entry->lsk) 1304 sock_release(entry->lsk); 1305 kfree(entry); 1306 } 1307 1308 static int mptcp_nl_remove_id_zero_address(struct net *net, 1309 struct mptcp_addr_info *addr) 1310 { 1311 struct mptcp_rm_list list = { .nr = 0 }; 1312 long s_slot = 0, s_num = 0; 1313 struct mptcp_sock *msk; 1314 1315 list.ids[list.nr++] = 0; 1316 1317 while ((msk = mptcp_token_iter_next(net, &s_slot, &s_num)) != NULL) { 1318 struct sock *sk = (struct sock *)msk; 1319 struct mptcp_addr_info msk_local; 1320 1321 if (list_empty(&msk->conn_list)) 1322 goto next; 1323 1324 local_address((struct sock_common *)msk, &msk_local); 1325 if (!addresses_equal(&msk_local, addr, addr->port)) 1326 goto next; 1327 1328 lock_sock(sk); 1329 spin_lock_bh(&msk->pm.lock); 1330 mptcp_pm_remove_addr(msk, &list); 1331 mptcp_pm_nl_rm_subflow_received(msk, &list); 1332 spin_unlock_bh(&msk->pm.lock); 1333 release_sock(sk); 1334 1335 next: 1336 sock_put(sk); 1337 cond_resched(); 1338 } 1339 1340 return 0; 1341 } 1342 1343 static int mptcp_nl_cmd_del_addr(struct sk_buff *skb, struct genl_info *info) 1344 { 1345 struct nlattr *attr = info->attrs[MPTCP_PM_ATTR_ADDR]; 1346 struct pm_nl_pernet *pernet = genl_info_pm_nl(info); 1347 struct mptcp_pm_addr_entry addr, *entry; 1348 unsigned int addr_max; 1349 int ret; 1350 1351 ret = mptcp_pm_parse_addr(attr, info, false, &addr); 1352 if (ret < 0) 1353 return ret; 1354 1355 /* the zero id address is special: the first address used by the msk 1356 * always gets such an id, so different subflows can have different zero 1357 * id addresses. Additionally zero id is not accounted for in id_bitmap. 1358 * Let's use an 'mptcp_rm_list' instead of the common remove code. 1359 */ 1360 if (addr.addr.id == 0) 1361 return mptcp_nl_remove_id_zero_address(sock_net(skb->sk), &addr.addr); 1362 1363 spin_lock_bh(&pernet->lock); 1364 entry = __lookup_addr_by_id(pernet, addr.addr.id); 1365 if (!entry) { 1366 GENL_SET_ERR_MSG(info, "address not found"); 1367 spin_unlock_bh(&pernet->lock); 1368 return -EINVAL; 1369 } 1370 if (entry->flags & MPTCP_PM_ADDR_FLAG_SIGNAL) { 1371 addr_max = pernet->add_addr_signal_max; 1372 WRITE_ONCE(pernet->add_addr_signal_max, addr_max - 1); 1373 } 1374 if (entry->flags & MPTCP_PM_ADDR_FLAG_SUBFLOW) { 1375 addr_max = pernet->local_addr_max; 1376 WRITE_ONCE(pernet->local_addr_max, addr_max - 1); 1377 } 1378 1379 pernet->addrs--; 1380 list_del_rcu(&entry->list); 1381 __clear_bit(entry->addr.id, pernet->id_bitmap); 1382 spin_unlock_bh(&pernet->lock); 1383 1384 mptcp_nl_remove_subflow_and_signal_addr(sock_net(skb->sk), &entry->addr); 1385 synchronize_rcu(); 1386 __mptcp_pm_release_addr_entry(entry); 1387 1388 return ret; 1389 } 1390 1391 static void mptcp_pm_remove_addrs_and_subflows(struct mptcp_sock *msk, 1392 struct list_head *rm_list) 1393 { 1394 struct mptcp_rm_list alist = { .nr = 0 }, slist = { .nr = 0 }; 1395 struct mptcp_pm_addr_entry *entry; 1396 1397 list_for_each_entry(entry, rm_list, list) { 1398 if (lookup_subflow_by_saddr(&msk->conn_list, &entry->addr) && 1399 alist.nr < MPTCP_RM_IDS_MAX && 1400 slist.nr < MPTCP_RM_IDS_MAX) { 1401 alist.ids[alist.nr++] = entry->addr.id; 1402 slist.ids[slist.nr++] = entry->addr.id; 1403 } else if (remove_anno_list_by_saddr(msk, &entry->addr) && 1404 alist.nr < MPTCP_RM_IDS_MAX) { 1405 alist.ids[alist.nr++] = entry->addr.id; 1406 } 1407 } 1408 1409 if (alist.nr) { 1410 spin_lock_bh(&msk->pm.lock); 1411 mptcp_pm_remove_addr(msk, &alist); 1412 spin_unlock_bh(&msk->pm.lock); 1413 } 1414 if (slist.nr) 1415 mptcp_pm_remove_subflow(msk, &slist); 1416 } 1417 1418 static void mptcp_nl_remove_addrs_list(struct net *net, 1419 struct list_head *rm_list) 1420 { 1421 long s_slot = 0, s_num = 0; 1422 struct mptcp_sock *msk; 1423 1424 if (list_empty(rm_list)) 1425 return; 1426 1427 while ((msk = mptcp_token_iter_next(net, &s_slot, &s_num)) != NULL) { 1428 struct sock *sk = (struct sock *)msk; 1429 1430 lock_sock(sk); 1431 mptcp_pm_remove_addrs_and_subflows(msk, rm_list); 1432 release_sock(sk); 1433 1434 sock_put(sk); 1435 cond_resched(); 1436 } 1437 } 1438 1439 /* caller must ensure the RCU grace period is already elapsed */ 1440 static void __flush_addrs(struct list_head *list) 1441 { 1442 while (!list_empty(list)) { 1443 struct mptcp_pm_addr_entry *cur; 1444 1445 cur = list_entry(list->next, 1446 struct mptcp_pm_addr_entry, list); 1447 list_del_rcu(&cur->list); 1448 __mptcp_pm_release_addr_entry(cur); 1449 } 1450 } 1451 1452 static void __reset_counters(struct pm_nl_pernet *pernet) 1453 { 1454 WRITE_ONCE(pernet->add_addr_signal_max, 0); 1455 WRITE_ONCE(pernet->add_addr_accept_max, 0); 1456 WRITE_ONCE(pernet->local_addr_max, 0); 1457 pernet->addrs = 0; 1458 } 1459 1460 static int mptcp_nl_cmd_flush_addrs(struct sk_buff *skb, struct genl_info *info) 1461 { 1462 struct pm_nl_pernet *pernet = genl_info_pm_nl(info); 1463 LIST_HEAD(free_list); 1464 1465 spin_lock_bh(&pernet->lock); 1466 list_splice_init(&pernet->local_addr_list, &free_list); 1467 __reset_counters(pernet); 1468 pernet->next_id = 1; 1469 bitmap_zero(pernet->id_bitmap, MAX_ADDR_ID + 1); 1470 spin_unlock_bh(&pernet->lock); 1471 mptcp_nl_remove_addrs_list(sock_net(skb->sk), &free_list); 1472 synchronize_rcu(); 1473 __flush_addrs(&free_list); 1474 return 0; 1475 } 1476 1477 static int mptcp_nl_fill_addr(struct sk_buff *skb, 1478 struct mptcp_pm_addr_entry *entry) 1479 { 1480 struct mptcp_addr_info *addr = &entry->addr; 1481 struct nlattr *attr; 1482 1483 attr = nla_nest_start(skb, MPTCP_PM_ATTR_ADDR); 1484 if (!attr) 1485 return -EMSGSIZE; 1486 1487 if (nla_put_u16(skb, MPTCP_PM_ADDR_ATTR_FAMILY, addr->family)) 1488 goto nla_put_failure; 1489 if (nla_put_u16(skb, MPTCP_PM_ADDR_ATTR_PORT, ntohs(addr->port))) 1490 goto nla_put_failure; 1491 if (nla_put_u8(skb, MPTCP_PM_ADDR_ATTR_ID, addr->id)) 1492 goto nla_put_failure; 1493 if (nla_put_u32(skb, MPTCP_PM_ADDR_ATTR_FLAGS, entry->flags)) 1494 goto nla_put_failure; 1495 if (entry->ifindex && 1496 nla_put_s32(skb, MPTCP_PM_ADDR_ATTR_IF_IDX, entry->ifindex)) 1497 goto nla_put_failure; 1498 1499 if (addr->family == AF_INET && 1500 nla_put_in_addr(skb, MPTCP_PM_ADDR_ATTR_ADDR4, 1501 addr->addr.s_addr)) 1502 goto nla_put_failure; 1503 #if IS_ENABLED(CONFIG_MPTCP_IPV6) 1504 else if (addr->family == AF_INET6 && 1505 nla_put_in6_addr(skb, MPTCP_PM_ADDR_ATTR_ADDR6, &addr->addr6)) 1506 goto nla_put_failure; 1507 #endif 1508 nla_nest_end(skb, attr); 1509 return 0; 1510 1511 nla_put_failure: 1512 nla_nest_cancel(skb, attr); 1513 return -EMSGSIZE; 1514 } 1515 1516 static int mptcp_nl_cmd_get_addr(struct sk_buff *skb, struct genl_info *info) 1517 { 1518 struct nlattr *attr = info->attrs[MPTCP_PM_ATTR_ADDR]; 1519 struct pm_nl_pernet *pernet = genl_info_pm_nl(info); 1520 struct mptcp_pm_addr_entry addr, *entry; 1521 struct sk_buff *msg; 1522 void *reply; 1523 int ret; 1524 1525 ret = mptcp_pm_parse_addr(attr, info, false, &addr); 1526 if (ret < 0) 1527 return ret; 1528 1529 msg = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL); 1530 if (!msg) 1531 return -ENOMEM; 1532 1533 reply = genlmsg_put_reply(msg, info, &mptcp_genl_family, 0, 1534 info->genlhdr->cmd); 1535 if (!reply) { 1536 GENL_SET_ERR_MSG(info, "not enough space in Netlink message"); 1537 ret = -EMSGSIZE; 1538 goto fail; 1539 } 1540 1541 spin_lock_bh(&pernet->lock); 1542 entry = __lookup_addr_by_id(pernet, addr.addr.id); 1543 if (!entry) { 1544 GENL_SET_ERR_MSG(info, "address not found"); 1545 ret = -EINVAL; 1546 goto unlock_fail; 1547 } 1548 1549 ret = mptcp_nl_fill_addr(msg, entry); 1550 if (ret) 1551 goto unlock_fail; 1552 1553 genlmsg_end(msg, reply); 1554 ret = genlmsg_reply(msg, info); 1555 spin_unlock_bh(&pernet->lock); 1556 return ret; 1557 1558 unlock_fail: 1559 spin_unlock_bh(&pernet->lock); 1560 1561 fail: 1562 nlmsg_free(msg); 1563 return ret; 1564 } 1565 1566 static int mptcp_nl_cmd_dump_addrs(struct sk_buff *msg, 1567 struct netlink_callback *cb) 1568 { 1569 struct net *net = sock_net(msg->sk); 1570 struct mptcp_pm_addr_entry *entry; 1571 struct pm_nl_pernet *pernet; 1572 int id = cb->args[0]; 1573 void *hdr; 1574 int i; 1575 1576 pernet = net_generic(net, pm_nl_pernet_id); 1577 1578 spin_lock_bh(&pernet->lock); 1579 for (i = id; i < MAX_ADDR_ID + 1; i++) { 1580 if (test_bit(i, pernet->id_bitmap)) { 1581 entry = __lookup_addr_by_id(pernet, i); 1582 if (!entry) 1583 break; 1584 1585 if (entry->addr.id <= id) 1586 continue; 1587 1588 hdr = genlmsg_put(msg, NETLINK_CB(cb->skb).portid, 1589 cb->nlh->nlmsg_seq, &mptcp_genl_family, 1590 NLM_F_MULTI, MPTCP_PM_CMD_GET_ADDR); 1591 if (!hdr) 1592 break; 1593 1594 if (mptcp_nl_fill_addr(msg, entry) < 0) { 1595 genlmsg_cancel(msg, hdr); 1596 break; 1597 } 1598 1599 id = entry->addr.id; 1600 genlmsg_end(msg, hdr); 1601 } 1602 } 1603 spin_unlock_bh(&pernet->lock); 1604 1605 cb->args[0] = id; 1606 return msg->len; 1607 } 1608 1609 static int parse_limit(struct genl_info *info, int id, unsigned int *limit) 1610 { 1611 struct nlattr *attr = info->attrs[id]; 1612 1613 if (!attr) 1614 return 0; 1615 1616 *limit = nla_get_u32(attr); 1617 if (*limit > MPTCP_PM_ADDR_MAX) { 1618 GENL_SET_ERR_MSG(info, "limit greater than maximum"); 1619 return -EINVAL; 1620 } 1621 return 0; 1622 } 1623 1624 static int 1625 mptcp_nl_cmd_set_limits(struct sk_buff *skb, struct genl_info *info) 1626 { 1627 struct pm_nl_pernet *pernet = genl_info_pm_nl(info); 1628 unsigned int rcv_addrs, subflows; 1629 int ret; 1630 1631 spin_lock_bh(&pernet->lock); 1632 rcv_addrs = pernet->add_addr_accept_max; 1633 ret = parse_limit(info, MPTCP_PM_ATTR_RCV_ADD_ADDRS, &rcv_addrs); 1634 if (ret) 1635 goto unlock; 1636 1637 subflows = pernet->subflows_max; 1638 ret = parse_limit(info, MPTCP_PM_ATTR_SUBFLOWS, &subflows); 1639 if (ret) 1640 goto unlock; 1641 1642 WRITE_ONCE(pernet->add_addr_accept_max, rcv_addrs); 1643 WRITE_ONCE(pernet->subflows_max, subflows); 1644 1645 unlock: 1646 spin_unlock_bh(&pernet->lock); 1647 return ret; 1648 } 1649 1650 static int 1651 mptcp_nl_cmd_get_limits(struct sk_buff *skb, struct genl_info *info) 1652 { 1653 struct pm_nl_pernet *pernet = genl_info_pm_nl(info); 1654 struct sk_buff *msg; 1655 void *reply; 1656 1657 msg = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL); 1658 if (!msg) 1659 return -ENOMEM; 1660 1661 reply = genlmsg_put_reply(msg, info, &mptcp_genl_family, 0, 1662 MPTCP_PM_CMD_GET_LIMITS); 1663 if (!reply) 1664 goto fail; 1665 1666 if (nla_put_u32(msg, MPTCP_PM_ATTR_RCV_ADD_ADDRS, 1667 READ_ONCE(pernet->add_addr_accept_max))) 1668 goto fail; 1669 1670 if (nla_put_u32(msg, MPTCP_PM_ATTR_SUBFLOWS, 1671 READ_ONCE(pernet->subflows_max))) 1672 goto fail; 1673 1674 genlmsg_end(msg, reply); 1675 return genlmsg_reply(msg, info); 1676 1677 fail: 1678 GENL_SET_ERR_MSG(info, "not enough space in Netlink message"); 1679 nlmsg_free(msg); 1680 return -EMSGSIZE; 1681 } 1682 1683 static int mptcp_nl_addr_backup(struct net *net, 1684 struct mptcp_addr_info *addr, 1685 u8 bkup) 1686 { 1687 long s_slot = 0, s_num = 0; 1688 struct mptcp_sock *msk; 1689 int ret = -EINVAL; 1690 1691 while ((msk = mptcp_token_iter_next(net, &s_slot, &s_num)) != NULL) { 1692 struct sock *sk = (struct sock *)msk; 1693 1694 if (list_empty(&msk->conn_list)) 1695 goto next; 1696 1697 lock_sock(sk); 1698 spin_lock_bh(&msk->pm.lock); 1699 ret = mptcp_pm_nl_mp_prio_send_ack(msk, addr, bkup); 1700 spin_unlock_bh(&msk->pm.lock); 1701 release_sock(sk); 1702 1703 next: 1704 sock_put(sk); 1705 cond_resched(); 1706 } 1707 1708 return ret; 1709 } 1710 1711 static int mptcp_nl_cmd_set_flags(struct sk_buff *skb, struct genl_info *info) 1712 { 1713 struct nlattr *attr = info->attrs[MPTCP_PM_ATTR_ADDR]; 1714 struct pm_nl_pernet *pernet = genl_info_pm_nl(info); 1715 struct mptcp_pm_addr_entry addr, *entry; 1716 struct net *net = sock_net(skb->sk); 1717 u8 bkup = 0; 1718 int ret; 1719 1720 ret = mptcp_pm_parse_addr(attr, info, true, &addr); 1721 if (ret < 0) 1722 return ret; 1723 1724 if (addr.flags & MPTCP_PM_ADDR_FLAG_BACKUP) 1725 bkup = 1; 1726 1727 list_for_each_entry(entry, &pernet->local_addr_list, list) { 1728 if (addresses_equal(&entry->addr, &addr.addr, true)) { 1729 ret = mptcp_nl_addr_backup(net, &entry->addr, bkup); 1730 if (ret) 1731 return ret; 1732 1733 if (bkup) 1734 entry->flags |= MPTCP_PM_ADDR_FLAG_BACKUP; 1735 else 1736 entry->flags &= ~MPTCP_PM_ADDR_FLAG_BACKUP; 1737 } 1738 } 1739 1740 return 0; 1741 } 1742 1743 static void mptcp_nl_mcast_send(struct net *net, struct sk_buff *nlskb, gfp_t gfp) 1744 { 1745 genlmsg_multicast_netns(&mptcp_genl_family, net, 1746 nlskb, 0, MPTCP_PM_EV_GRP_OFFSET, gfp); 1747 } 1748 1749 static int mptcp_event_add_subflow(struct sk_buff *skb, const struct sock *ssk) 1750 { 1751 const struct inet_sock *issk = inet_sk(ssk); 1752 const struct mptcp_subflow_context *sf; 1753 1754 if (nla_put_u16(skb, MPTCP_ATTR_FAMILY, ssk->sk_family)) 1755 return -EMSGSIZE; 1756 1757 switch (ssk->sk_family) { 1758 case AF_INET: 1759 if (nla_put_in_addr(skb, MPTCP_ATTR_SADDR4, issk->inet_saddr)) 1760 return -EMSGSIZE; 1761 if (nla_put_in_addr(skb, MPTCP_ATTR_DADDR4, issk->inet_daddr)) 1762 return -EMSGSIZE; 1763 break; 1764 #if IS_ENABLED(CONFIG_MPTCP_IPV6) 1765 case AF_INET6: { 1766 const struct ipv6_pinfo *np = inet6_sk(ssk); 1767 1768 if (nla_put_in6_addr(skb, MPTCP_ATTR_SADDR6, &np->saddr)) 1769 return -EMSGSIZE; 1770 if (nla_put_in6_addr(skb, MPTCP_ATTR_DADDR6, &ssk->sk_v6_daddr)) 1771 return -EMSGSIZE; 1772 break; 1773 } 1774 #endif 1775 default: 1776 WARN_ON_ONCE(1); 1777 return -EMSGSIZE; 1778 } 1779 1780 if (nla_put_be16(skb, MPTCP_ATTR_SPORT, issk->inet_sport)) 1781 return -EMSGSIZE; 1782 if (nla_put_be16(skb, MPTCP_ATTR_DPORT, issk->inet_dport)) 1783 return -EMSGSIZE; 1784 1785 sf = mptcp_subflow_ctx(ssk); 1786 if (WARN_ON_ONCE(!sf)) 1787 return -EINVAL; 1788 1789 if (nla_put_u8(skb, MPTCP_ATTR_LOC_ID, sf->local_id)) 1790 return -EMSGSIZE; 1791 1792 if (nla_put_u8(skb, MPTCP_ATTR_REM_ID, sf->remote_id)) 1793 return -EMSGSIZE; 1794 1795 return 0; 1796 } 1797 1798 static int mptcp_event_put_token_and_ssk(struct sk_buff *skb, 1799 const struct mptcp_sock *msk, 1800 const struct sock *ssk) 1801 { 1802 const struct sock *sk = (const struct sock *)msk; 1803 const struct mptcp_subflow_context *sf; 1804 u8 sk_err; 1805 1806 if (nla_put_u32(skb, MPTCP_ATTR_TOKEN, msk->token)) 1807 return -EMSGSIZE; 1808 1809 if (mptcp_event_add_subflow(skb, ssk)) 1810 return -EMSGSIZE; 1811 1812 sf = mptcp_subflow_ctx(ssk); 1813 if (WARN_ON_ONCE(!sf)) 1814 return -EINVAL; 1815 1816 if (nla_put_u8(skb, MPTCP_ATTR_BACKUP, sf->backup)) 1817 return -EMSGSIZE; 1818 1819 if (ssk->sk_bound_dev_if && 1820 nla_put_s32(skb, MPTCP_ATTR_IF_IDX, ssk->sk_bound_dev_if)) 1821 return -EMSGSIZE; 1822 1823 sk_err = ssk->sk_err; 1824 if (sk_err && sk->sk_state == TCP_ESTABLISHED && 1825 nla_put_u8(skb, MPTCP_ATTR_ERROR, sk_err)) 1826 return -EMSGSIZE; 1827 1828 return 0; 1829 } 1830 1831 static int mptcp_event_sub_established(struct sk_buff *skb, 1832 const struct mptcp_sock *msk, 1833 const struct sock *ssk) 1834 { 1835 return mptcp_event_put_token_and_ssk(skb, msk, ssk); 1836 } 1837 1838 static int mptcp_event_sub_closed(struct sk_buff *skb, 1839 const struct mptcp_sock *msk, 1840 const struct sock *ssk) 1841 { 1842 const struct mptcp_subflow_context *sf; 1843 1844 if (mptcp_event_put_token_and_ssk(skb, msk, ssk)) 1845 return -EMSGSIZE; 1846 1847 sf = mptcp_subflow_ctx(ssk); 1848 if (!sf->reset_seen) 1849 return 0; 1850 1851 if (nla_put_u32(skb, MPTCP_ATTR_RESET_REASON, sf->reset_reason)) 1852 return -EMSGSIZE; 1853 1854 if (nla_put_u32(skb, MPTCP_ATTR_RESET_FLAGS, sf->reset_transient)) 1855 return -EMSGSIZE; 1856 1857 return 0; 1858 } 1859 1860 static int mptcp_event_created(struct sk_buff *skb, 1861 const struct mptcp_sock *msk, 1862 const struct sock *ssk) 1863 { 1864 int err = nla_put_u32(skb, MPTCP_ATTR_TOKEN, msk->token); 1865 1866 if (err) 1867 return err; 1868 1869 return mptcp_event_add_subflow(skb, ssk); 1870 } 1871 1872 void mptcp_event_addr_removed(const struct mptcp_sock *msk, uint8_t id) 1873 { 1874 struct net *net = sock_net((const struct sock *)msk); 1875 struct nlmsghdr *nlh; 1876 struct sk_buff *skb; 1877 1878 if (!genl_has_listeners(&mptcp_genl_family, net, MPTCP_PM_EV_GRP_OFFSET)) 1879 return; 1880 1881 skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_ATOMIC); 1882 if (!skb) 1883 return; 1884 1885 nlh = genlmsg_put(skb, 0, 0, &mptcp_genl_family, 0, MPTCP_EVENT_REMOVED); 1886 if (!nlh) 1887 goto nla_put_failure; 1888 1889 if (nla_put_u32(skb, MPTCP_ATTR_TOKEN, msk->token)) 1890 goto nla_put_failure; 1891 1892 if (nla_put_u8(skb, MPTCP_ATTR_REM_ID, id)) 1893 goto nla_put_failure; 1894 1895 genlmsg_end(skb, nlh); 1896 mptcp_nl_mcast_send(net, skb, GFP_ATOMIC); 1897 return; 1898 1899 nla_put_failure: 1900 kfree_skb(skb); 1901 } 1902 1903 void mptcp_event_addr_announced(const struct mptcp_sock *msk, 1904 const struct mptcp_addr_info *info) 1905 { 1906 struct net *net = sock_net((const struct sock *)msk); 1907 struct nlmsghdr *nlh; 1908 struct sk_buff *skb; 1909 1910 if (!genl_has_listeners(&mptcp_genl_family, net, MPTCP_PM_EV_GRP_OFFSET)) 1911 return; 1912 1913 skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_ATOMIC); 1914 if (!skb) 1915 return; 1916 1917 nlh = genlmsg_put(skb, 0, 0, &mptcp_genl_family, 0, 1918 MPTCP_EVENT_ANNOUNCED); 1919 if (!nlh) 1920 goto nla_put_failure; 1921 1922 if (nla_put_u32(skb, MPTCP_ATTR_TOKEN, msk->token)) 1923 goto nla_put_failure; 1924 1925 if (nla_put_u8(skb, MPTCP_ATTR_REM_ID, info->id)) 1926 goto nla_put_failure; 1927 1928 if (nla_put_be16(skb, MPTCP_ATTR_DPORT, info->port)) 1929 goto nla_put_failure; 1930 1931 switch (info->family) { 1932 case AF_INET: 1933 if (nla_put_in_addr(skb, MPTCP_ATTR_DADDR4, info->addr.s_addr)) 1934 goto nla_put_failure; 1935 break; 1936 #if IS_ENABLED(CONFIG_MPTCP_IPV6) 1937 case AF_INET6: 1938 if (nla_put_in6_addr(skb, MPTCP_ATTR_DADDR6, &info->addr6)) 1939 goto nla_put_failure; 1940 break; 1941 #endif 1942 default: 1943 WARN_ON_ONCE(1); 1944 goto nla_put_failure; 1945 } 1946 1947 genlmsg_end(skb, nlh); 1948 mptcp_nl_mcast_send(net, skb, GFP_ATOMIC); 1949 return; 1950 1951 nla_put_failure: 1952 kfree_skb(skb); 1953 } 1954 1955 void mptcp_event(enum mptcp_event_type type, const struct mptcp_sock *msk, 1956 const struct sock *ssk, gfp_t gfp) 1957 { 1958 struct net *net = sock_net((const struct sock *)msk); 1959 struct nlmsghdr *nlh; 1960 struct sk_buff *skb; 1961 1962 if (!genl_has_listeners(&mptcp_genl_family, net, MPTCP_PM_EV_GRP_OFFSET)) 1963 return; 1964 1965 skb = nlmsg_new(NLMSG_DEFAULT_SIZE, gfp); 1966 if (!skb) 1967 return; 1968 1969 nlh = genlmsg_put(skb, 0, 0, &mptcp_genl_family, 0, type); 1970 if (!nlh) 1971 goto nla_put_failure; 1972 1973 switch (type) { 1974 case MPTCP_EVENT_UNSPEC: 1975 WARN_ON_ONCE(1); 1976 break; 1977 case MPTCP_EVENT_CREATED: 1978 case MPTCP_EVENT_ESTABLISHED: 1979 if (mptcp_event_created(skb, msk, ssk) < 0) 1980 goto nla_put_failure; 1981 break; 1982 case MPTCP_EVENT_CLOSED: 1983 if (nla_put_u32(skb, MPTCP_ATTR_TOKEN, msk->token) < 0) 1984 goto nla_put_failure; 1985 break; 1986 case MPTCP_EVENT_ANNOUNCED: 1987 case MPTCP_EVENT_REMOVED: 1988 /* call mptcp_event_addr_announced()/removed instead */ 1989 WARN_ON_ONCE(1); 1990 break; 1991 case MPTCP_EVENT_SUB_ESTABLISHED: 1992 case MPTCP_EVENT_SUB_PRIORITY: 1993 if (mptcp_event_sub_established(skb, msk, ssk) < 0) 1994 goto nla_put_failure; 1995 break; 1996 case MPTCP_EVENT_SUB_CLOSED: 1997 if (mptcp_event_sub_closed(skb, msk, ssk) < 0) 1998 goto nla_put_failure; 1999 break; 2000 } 2001 2002 genlmsg_end(skb, nlh); 2003 mptcp_nl_mcast_send(net, skb, gfp); 2004 return; 2005 2006 nla_put_failure: 2007 kfree_skb(skb); 2008 } 2009 2010 static const struct genl_small_ops mptcp_pm_ops[] = { 2011 { 2012 .cmd = MPTCP_PM_CMD_ADD_ADDR, 2013 .doit = mptcp_nl_cmd_add_addr, 2014 .flags = GENL_ADMIN_PERM, 2015 }, 2016 { 2017 .cmd = MPTCP_PM_CMD_DEL_ADDR, 2018 .doit = mptcp_nl_cmd_del_addr, 2019 .flags = GENL_ADMIN_PERM, 2020 }, 2021 { 2022 .cmd = MPTCP_PM_CMD_FLUSH_ADDRS, 2023 .doit = mptcp_nl_cmd_flush_addrs, 2024 .flags = GENL_ADMIN_PERM, 2025 }, 2026 { 2027 .cmd = MPTCP_PM_CMD_GET_ADDR, 2028 .doit = mptcp_nl_cmd_get_addr, 2029 .dumpit = mptcp_nl_cmd_dump_addrs, 2030 }, 2031 { 2032 .cmd = MPTCP_PM_CMD_SET_LIMITS, 2033 .doit = mptcp_nl_cmd_set_limits, 2034 .flags = GENL_ADMIN_PERM, 2035 }, 2036 { 2037 .cmd = MPTCP_PM_CMD_GET_LIMITS, 2038 .doit = mptcp_nl_cmd_get_limits, 2039 }, 2040 { 2041 .cmd = MPTCP_PM_CMD_SET_FLAGS, 2042 .doit = mptcp_nl_cmd_set_flags, 2043 .flags = GENL_ADMIN_PERM, 2044 }, 2045 }; 2046 2047 static struct genl_family mptcp_genl_family __ro_after_init = { 2048 .name = MPTCP_PM_NAME, 2049 .version = MPTCP_PM_VER, 2050 .maxattr = MPTCP_PM_ATTR_MAX, 2051 .policy = mptcp_pm_policy, 2052 .netnsok = true, 2053 .module = THIS_MODULE, 2054 .small_ops = mptcp_pm_ops, 2055 .n_small_ops = ARRAY_SIZE(mptcp_pm_ops), 2056 .mcgrps = mptcp_pm_mcgrps, 2057 .n_mcgrps = ARRAY_SIZE(mptcp_pm_mcgrps), 2058 }; 2059 2060 static int __net_init pm_nl_init_net(struct net *net) 2061 { 2062 struct pm_nl_pernet *pernet = net_generic(net, pm_nl_pernet_id); 2063 2064 INIT_LIST_HEAD_RCU(&pernet->local_addr_list); 2065 pernet->next_id = 1; 2066 pernet->stale_loss_cnt = 4; 2067 spin_lock_init(&pernet->lock); 2068 2069 /* No need to initialize other pernet fields, the struct is zeroed at 2070 * allocation time. 2071 */ 2072 2073 return 0; 2074 } 2075 2076 static void __net_exit pm_nl_exit_net(struct list_head *net_list) 2077 { 2078 struct net *net; 2079 2080 list_for_each_entry(net, net_list, exit_list) { 2081 struct pm_nl_pernet *pernet = net_generic(net, pm_nl_pernet_id); 2082 2083 /* net is removed from namespace list, can't race with 2084 * other modifiers, also netns core already waited for a 2085 * RCU grace period. 2086 */ 2087 __flush_addrs(&pernet->local_addr_list); 2088 } 2089 } 2090 2091 static struct pernet_operations mptcp_pm_pernet_ops = { 2092 .init = pm_nl_init_net, 2093 .exit_batch = pm_nl_exit_net, 2094 .id = &pm_nl_pernet_id, 2095 .size = sizeof(struct pm_nl_pernet), 2096 }; 2097 2098 void __init mptcp_pm_nl_init(void) 2099 { 2100 if (register_pernet_subsys(&mptcp_pm_pernet_ops) < 0) 2101 panic("Failed to register MPTCP PM pernet subsystem.\n"); 2102 2103 if (genl_register_family(&mptcp_genl_family)) 2104 panic("Failed to register MPTCP PM netlink family\n"); 2105 } 2106