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 void mptcp_local_address(const struct sock_common *skc, struct mptcp_addr_info *addr) 90 { 91 addr->family = skc->skc_family; 92 addr->port = htons(skc->skc_num); 93 if (addr->family == AF_INET) 94 addr->addr.s_addr = skc->skc_rcv_saddr; 95 #if IS_ENABLED(CONFIG_MPTCP_IPV6) 96 else if (addr->family == AF_INET6) 97 addr->addr6 = skc->skc_v6_rcv_saddr; 98 #endif 99 } 100 101 static void remote_address(const struct sock_common *skc, 102 struct mptcp_addr_info *addr) 103 { 104 addr->family = skc->skc_family; 105 addr->port = skc->skc_dport; 106 if (addr->family == AF_INET) 107 addr->addr.s_addr = skc->skc_daddr; 108 #if IS_ENABLED(CONFIG_MPTCP_IPV6) 109 else if (addr->family == AF_INET6) 110 addr->addr6 = skc->skc_v6_daddr; 111 #endif 112 } 113 114 static bool lookup_subflow_by_saddr(const struct list_head *list, 115 const struct mptcp_addr_info *saddr) 116 { 117 struct mptcp_subflow_context *subflow; 118 struct mptcp_addr_info cur; 119 struct sock_common *skc; 120 121 list_for_each_entry(subflow, list, node) { 122 skc = (struct sock_common *)mptcp_subflow_tcp_sock(subflow); 123 124 mptcp_local_address(skc, &cur); 125 if (mptcp_addresses_equal(&cur, saddr, saddr->port)) 126 return true; 127 } 128 129 return false; 130 } 131 132 static bool lookup_subflow_by_daddr(const struct list_head *list, 133 const struct mptcp_addr_info *daddr) 134 { 135 struct mptcp_subflow_context *subflow; 136 struct mptcp_addr_info cur; 137 struct sock_common *skc; 138 139 list_for_each_entry(subflow, list, node) { 140 skc = (struct sock_common *)mptcp_subflow_tcp_sock(subflow); 141 142 remote_address(skc, &cur); 143 if (mptcp_addresses_equal(&cur, daddr, daddr->port)) 144 return true; 145 } 146 147 return false; 148 } 149 150 static struct mptcp_pm_addr_entry * 151 select_local_address(const struct pm_nl_pernet *pernet, 152 const struct mptcp_sock *msk) 153 { 154 struct mptcp_pm_addr_entry *entry, *ret = NULL; 155 156 msk_owned_by_me(msk); 157 158 rcu_read_lock(); 159 list_for_each_entry_rcu(entry, &pernet->local_addr_list, list) { 160 if (!(entry->flags & MPTCP_PM_ADDR_FLAG_SUBFLOW)) 161 continue; 162 163 if (!test_bit(entry->addr.id, msk->pm.id_avail_bitmap)) 164 continue; 165 166 ret = entry; 167 break; 168 } 169 rcu_read_unlock(); 170 return ret; 171 } 172 173 static struct mptcp_pm_addr_entry * 174 select_signal_address(struct pm_nl_pernet *pernet, const struct mptcp_sock *msk) 175 { 176 struct mptcp_pm_addr_entry *entry, *ret = NULL; 177 178 rcu_read_lock(); 179 /* do not keep any additional per socket state, just signal 180 * the address list in order. 181 * Note: removal from the local address list during the msk life-cycle 182 * can lead to additional addresses not being announced. 183 */ 184 list_for_each_entry_rcu(entry, &pernet->local_addr_list, list) { 185 if (!test_bit(entry->addr.id, msk->pm.id_avail_bitmap)) 186 continue; 187 188 if (!(entry->flags & MPTCP_PM_ADDR_FLAG_SIGNAL)) 189 continue; 190 191 ret = entry; 192 break; 193 } 194 rcu_read_unlock(); 195 return ret; 196 } 197 198 unsigned int mptcp_pm_get_add_addr_signal_max(const struct mptcp_sock *msk) 199 { 200 const struct pm_nl_pernet *pernet = pm_nl_get_pernet_from_msk(msk); 201 202 return READ_ONCE(pernet->add_addr_signal_max); 203 } 204 EXPORT_SYMBOL_GPL(mptcp_pm_get_add_addr_signal_max); 205 206 unsigned int mptcp_pm_get_add_addr_accept_max(const struct mptcp_sock *msk) 207 { 208 struct pm_nl_pernet *pernet = pm_nl_get_pernet_from_msk(msk); 209 210 return READ_ONCE(pernet->add_addr_accept_max); 211 } 212 EXPORT_SYMBOL_GPL(mptcp_pm_get_add_addr_accept_max); 213 214 unsigned int mptcp_pm_get_subflows_max(const struct mptcp_sock *msk) 215 { 216 struct pm_nl_pernet *pernet = pm_nl_get_pernet_from_msk(msk); 217 218 return READ_ONCE(pernet->subflows_max); 219 } 220 EXPORT_SYMBOL_GPL(mptcp_pm_get_subflows_max); 221 222 unsigned int mptcp_pm_get_local_addr_max(const struct mptcp_sock *msk) 223 { 224 struct pm_nl_pernet *pernet = pm_nl_get_pernet_from_msk(msk); 225 226 return READ_ONCE(pernet->local_addr_max); 227 } 228 EXPORT_SYMBOL_GPL(mptcp_pm_get_local_addr_max); 229 230 bool mptcp_pm_nl_check_work_pending(struct mptcp_sock *msk) 231 { 232 struct pm_nl_pernet *pernet = pm_nl_get_pernet_from_msk(msk); 233 234 if (msk->pm.subflows == mptcp_pm_get_subflows_max(msk) || 235 (find_next_and_bit(pernet->id_bitmap, msk->pm.id_avail_bitmap, 236 MPTCP_PM_MAX_ADDR_ID + 1, 0) == MPTCP_PM_MAX_ADDR_ID + 1)) { 237 WRITE_ONCE(msk->pm.work_pending, false); 238 return false; 239 } 240 return true; 241 } 242 243 struct mptcp_pm_add_entry * 244 mptcp_lookup_anno_list_by_saddr(const struct mptcp_sock *msk, 245 const struct mptcp_addr_info *addr) 246 { 247 struct mptcp_pm_add_entry *entry; 248 249 lockdep_assert_held(&msk->pm.lock); 250 251 list_for_each_entry(entry, &msk->pm.anno_list, list) { 252 if (mptcp_addresses_equal(&entry->addr, addr, true)) 253 return entry; 254 } 255 256 return NULL; 257 } 258 259 bool mptcp_pm_sport_in_anno_list(struct mptcp_sock *msk, const struct sock *sk) 260 { 261 struct mptcp_pm_add_entry *entry; 262 struct mptcp_addr_info saddr; 263 bool ret = false; 264 265 mptcp_local_address((struct sock_common *)sk, &saddr); 266 267 spin_lock_bh(&msk->pm.lock); 268 list_for_each_entry(entry, &msk->pm.anno_list, list) { 269 if (mptcp_addresses_equal(&entry->addr, &saddr, true)) { 270 ret = true; 271 goto out; 272 } 273 } 274 275 out: 276 spin_unlock_bh(&msk->pm.lock); 277 return ret; 278 } 279 280 static void mptcp_pm_add_timer(struct timer_list *timer) 281 { 282 struct mptcp_pm_add_entry *entry = from_timer(entry, timer, add_timer); 283 struct mptcp_sock *msk = entry->sock; 284 struct sock *sk = (struct sock *)msk; 285 286 pr_debug("msk=%p", msk); 287 288 if (!msk) 289 return; 290 291 if (inet_sk_state_load(sk) == TCP_CLOSE) 292 return; 293 294 if (!entry->addr.id) 295 return; 296 297 if (mptcp_pm_should_add_signal_addr(msk)) { 298 sk_reset_timer(sk, timer, jiffies + TCP_RTO_MAX / 8); 299 goto out; 300 } 301 302 spin_lock_bh(&msk->pm.lock); 303 304 if (!mptcp_pm_should_add_signal_addr(msk)) { 305 pr_debug("retransmit ADD_ADDR id=%d", entry->addr.id); 306 mptcp_pm_announce_addr(msk, &entry->addr, false); 307 mptcp_pm_add_addr_send_ack(msk); 308 entry->retrans_times++; 309 } 310 311 if (entry->retrans_times < ADD_ADDR_RETRANS_MAX) 312 sk_reset_timer(sk, timer, 313 jiffies + mptcp_get_add_addr_timeout(sock_net(sk))); 314 315 spin_unlock_bh(&msk->pm.lock); 316 317 if (entry->retrans_times == ADD_ADDR_RETRANS_MAX) 318 mptcp_pm_subflow_established(msk); 319 320 out: 321 __sock_put(sk); 322 } 323 324 struct mptcp_pm_add_entry * 325 mptcp_pm_del_add_timer(struct mptcp_sock *msk, 326 const struct mptcp_addr_info *addr, bool check_id) 327 { 328 struct mptcp_pm_add_entry *entry; 329 struct sock *sk = (struct sock *)msk; 330 331 spin_lock_bh(&msk->pm.lock); 332 entry = mptcp_lookup_anno_list_by_saddr(msk, addr); 333 if (entry && (!check_id || entry->addr.id == addr->id)) 334 entry->retrans_times = ADD_ADDR_RETRANS_MAX; 335 spin_unlock_bh(&msk->pm.lock); 336 337 if (entry && (!check_id || entry->addr.id == addr->id)) 338 sk_stop_timer_sync(sk, &entry->add_timer); 339 340 return entry; 341 } 342 343 bool mptcp_pm_alloc_anno_list(struct mptcp_sock *msk, 344 const struct mptcp_pm_addr_entry *entry) 345 { 346 struct mptcp_pm_add_entry *add_entry = NULL; 347 struct sock *sk = (struct sock *)msk; 348 struct net *net = sock_net(sk); 349 350 lockdep_assert_held(&msk->pm.lock); 351 352 add_entry = mptcp_lookup_anno_list_by_saddr(msk, &entry->addr); 353 354 if (add_entry) { 355 if (mptcp_pm_is_kernel(msk)) 356 return false; 357 358 sk_reset_timer(sk, &add_entry->add_timer, 359 jiffies + mptcp_get_add_addr_timeout(net)); 360 return true; 361 } 362 363 add_entry = kmalloc(sizeof(*add_entry), GFP_ATOMIC); 364 if (!add_entry) 365 return false; 366 367 list_add(&add_entry->list, &msk->pm.anno_list); 368 369 add_entry->addr = entry->addr; 370 add_entry->sock = msk; 371 add_entry->retrans_times = 0; 372 373 timer_setup(&add_entry->add_timer, mptcp_pm_add_timer, 0); 374 sk_reset_timer(sk, &add_entry->add_timer, 375 jiffies + mptcp_get_add_addr_timeout(net)); 376 377 return true; 378 } 379 380 void mptcp_pm_free_anno_list(struct mptcp_sock *msk) 381 { 382 struct mptcp_pm_add_entry *entry, *tmp; 383 struct sock *sk = (struct sock *)msk; 384 LIST_HEAD(free_list); 385 386 pr_debug("msk=%p", msk); 387 388 spin_lock_bh(&msk->pm.lock); 389 list_splice_init(&msk->pm.anno_list, &free_list); 390 spin_unlock_bh(&msk->pm.lock); 391 392 list_for_each_entry_safe(entry, tmp, &free_list, list) { 393 sk_stop_timer_sync(sk, &entry->add_timer); 394 kfree(entry); 395 } 396 } 397 398 static bool lookup_address_in_vec(const struct mptcp_addr_info *addrs, unsigned int nr, 399 const struct mptcp_addr_info *addr) 400 { 401 int i; 402 403 for (i = 0; i < nr; i++) { 404 if (addrs[i].id == addr->id) 405 return true; 406 } 407 408 return false; 409 } 410 411 /* Fill all the remote addresses into the array addrs[], 412 * and return the array size. 413 */ 414 static unsigned int fill_remote_addresses_vec(struct mptcp_sock *msk, 415 struct mptcp_addr_info *local, 416 bool fullmesh, 417 struct mptcp_addr_info *addrs) 418 { 419 bool deny_id0 = READ_ONCE(msk->pm.remote_deny_join_id0); 420 struct sock *sk = (struct sock *)msk, *ssk; 421 struct mptcp_subflow_context *subflow; 422 struct mptcp_addr_info remote = { 0 }; 423 unsigned int subflows_max; 424 int i = 0; 425 426 subflows_max = mptcp_pm_get_subflows_max(msk); 427 remote_address((struct sock_common *)sk, &remote); 428 429 /* Non-fullmesh endpoint, fill in the single entry 430 * corresponding to the primary MPC subflow remote address 431 */ 432 if (!fullmesh) { 433 if (deny_id0) 434 return 0; 435 436 if (!mptcp_pm_addr_families_match(sk, local, &remote)) 437 return 0; 438 439 msk->pm.subflows++; 440 addrs[i++] = remote; 441 } else { 442 mptcp_for_each_subflow(msk, subflow) { 443 ssk = mptcp_subflow_tcp_sock(subflow); 444 remote_address((struct sock_common *)ssk, &addrs[i]); 445 addrs[i].id = subflow->remote_id; 446 if (deny_id0 && !addrs[i].id) 447 continue; 448 449 if (!mptcp_pm_addr_families_match(sk, local, &addrs[i])) 450 continue; 451 452 if (!lookup_address_in_vec(addrs, i, &addrs[i]) && 453 msk->pm.subflows < subflows_max) { 454 msk->pm.subflows++; 455 i++; 456 } 457 } 458 } 459 460 return i; 461 } 462 463 static void __mptcp_pm_send_ack(struct mptcp_sock *msk, struct mptcp_subflow_context *subflow, 464 bool prio, bool backup) 465 { 466 struct sock *ssk = mptcp_subflow_tcp_sock(subflow); 467 bool slow; 468 469 pr_debug("send ack for %s", 470 prio ? "mp_prio" : (mptcp_pm_should_add_signal(msk) ? "add_addr" : "rm_addr")); 471 472 slow = lock_sock_fast(ssk); 473 if (prio) { 474 if (subflow->backup != backup) 475 msk->last_snd = NULL; 476 477 subflow->send_mp_prio = 1; 478 subflow->backup = backup; 479 subflow->request_bkup = backup; 480 } 481 482 __mptcp_subflow_send_ack(ssk); 483 unlock_sock_fast(ssk, slow); 484 } 485 486 static void mptcp_pm_send_ack(struct mptcp_sock *msk, struct mptcp_subflow_context *subflow, 487 bool prio, bool backup) 488 { 489 spin_unlock_bh(&msk->pm.lock); 490 __mptcp_pm_send_ack(msk, subflow, prio, backup); 491 spin_lock_bh(&msk->pm.lock); 492 } 493 494 static struct mptcp_pm_addr_entry * 495 __lookup_addr_by_id(struct pm_nl_pernet *pernet, unsigned int id) 496 { 497 struct mptcp_pm_addr_entry *entry; 498 499 list_for_each_entry(entry, &pernet->local_addr_list, list) { 500 if (entry->addr.id == id) 501 return entry; 502 } 503 return NULL; 504 } 505 506 static struct mptcp_pm_addr_entry * 507 __lookup_addr(struct pm_nl_pernet *pernet, const struct mptcp_addr_info *info, 508 bool lookup_by_id) 509 { 510 struct mptcp_pm_addr_entry *entry; 511 512 list_for_each_entry(entry, &pernet->local_addr_list, list) { 513 if ((!lookup_by_id && 514 mptcp_addresses_equal(&entry->addr, info, entry->addr.port)) || 515 (lookup_by_id && entry->addr.id == info->id)) 516 return entry; 517 } 518 return NULL; 519 } 520 521 static void mptcp_pm_create_subflow_or_signal_addr(struct mptcp_sock *msk) 522 { 523 struct sock *sk = (struct sock *)msk; 524 struct mptcp_pm_addr_entry *local; 525 unsigned int add_addr_signal_max; 526 unsigned int local_addr_max; 527 struct pm_nl_pernet *pernet; 528 unsigned int subflows_max; 529 530 pernet = pm_nl_get_pernet(sock_net(sk)); 531 532 add_addr_signal_max = mptcp_pm_get_add_addr_signal_max(msk); 533 local_addr_max = mptcp_pm_get_local_addr_max(msk); 534 subflows_max = mptcp_pm_get_subflows_max(msk); 535 536 /* do lazy endpoint usage accounting for the MPC subflows */ 537 if (unlikely(!(msk->pm.status & BIT(MPTCP_PM_MPC_ENDPOINT_ACCOUNTED))) && msk->first) { 538 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(msk->first); 539 struct mptcp_pm_addr_entry *entry; 540 struct mptcp_addr_info mpc_addr; 541 bool backup = false; 542 543 mptcp_local_address((struct sock_common *)msk->first, &mpc_addr); 544 rcu_read_lock(); 545 entry = __lookup_addr(pernet, &mpc_addr, false); 546 if (entry) { 547 __clear_bit(entry->addr.id, msk->pm.id_avail_bitmap); 548 msk->mpc_endpoint_id = entry->addr.id; 549 backup = !!(entry->flags & MPTCP_PM_ADDR_FLAG_BACKUP); 550 } 551 rcu_read_unlock(); 552 553 if (backup) 554 mptcp_pm_send_ack(msk, subflow, true, backup); 555 556 msk->pm.status |= BIT(MPTCP_PM_MPC_ENDPOINT_ACCOUNTED); 557 } 558 559 pr_debug("local %d:%d signal %d:%d subflows %d:%d\n", 560 msk->pm.local_addr_used, local_addr_max, 561 msk->pm.add_addr_signaled, add_addr_signal_max, 562 msk->pm.subflows, subflows_max); 563 564 /* check first for announce */ 565 if (msk->pm.add_addr_signaled < add_addr_signal_max) { 566 local = select_signal_address(pernet, msk); 567 568 /* due to racing events on both ends we can reach here while 569 * previous add address is still running: if we invoke now 570 * mptcp_pm_announce_addr(), that will fail and the 571 * corresponding id will be marked as used. 572 * Instead let the PM machinery reschedule us when the 573 * current address announce will be completed. 574 */ 575 if (msk->pm.addr_signal & BIT(MPTCP_ADD_ADDR_SIGNAL)) 576 return; 577 578 if (local) { 579 if (mptcp_pm_alloc_anno_list(msk, local)) { 580 __clear_bit(local->addr.id, msk->pm.id_avail_bitmap); 581 msk->pm.add_addr_signaled++; 582 mptcp_pm_announce_addr(msk, &local->addr, false); 583 mptcp_pm_nl_addr_send_ack(msk); 584 } 585 } 586 } 587 588 /* check if should create a new subflow */ 589 while (msk->pm.local_addr_used < local_addr_max && 590 msk->pm.subflows < subflows_max) { 591 struct mptcp_addr_info addrs[MPTCP_PM_ADDR_MAX]; 592 bool fullmesh; 593 int i, nr; 594 595 local = select_local_address(pernet, msk); 596 if (!local) 597 break; 598 599 fullmesh = !!(local->flags & MPTCP_PM_ADDR_FLAG_FULLMESH); 600 601 msk->pm.local_addr_used++; 602 __clear_bit(local->addr.id, msk->pm.id_avail_bitmap); 603 nr = fill_remote_addresses_vec(msk, &local->addr, fullmesh, addrs); 604 if (nr == 0) 605 continue; 606 607 spin_unlock_bh(&msk->pm.lock); 608 for (i = 0; i < nr; i++) 609 __mptcp_subflow_connect(sk, &local->addr, &addrs[i]); 610 spin_lock_bh(&msk->pm.lock); 611 } 612 mptcp_pm_nl_check_work_pending(msk); 613 } 614 615 static void mptcp_pm_nl_fully_established(struct mptcp_sock *msk) 616 { 617 mptcp_pm_create_subflow_or_signal_addr(msk); 618 } 619 620 static void mptcp_pm_nl_subflow_established(struct mptcp_sock *msk) 621 { 622 mptcp_pm_create_subflow_or_signal_addr(msk); 623 } 624 625 /* Fill all the local addresses into the array addrs[], 626 * and return the array size. 627 */ 628 static unsigned int fill_local_addresses_vec(struct mptcp_sock *msk, 629 struct mptcp_addr_info *remote, 630 struct mptcp_addr_info *addrs) 631 { 632 struct sock *sk = (struct sock *)msk; 633 struct mptcp_pm_addr_entry *entry; 634 struct pm_nl_pernet *pernet; 635 unsigned int subflows_max; 636 int i = 0; 637 638 pernet = pm_nl_get_pernet_from_msk(msk); 639 subflows_max = mptcp_pm_get_subflows_max(msk); 640 641 rcu_read_lock(); 642 list_for_each_entry_rcu(entry, &pernet->local_addr_list, list) { 643 if (!(entry->flags & MPTCP_PM_ADDR_FLAG_FULLMESH)) 644 continue; 645 646 if (!mptcp_pm_addr_families_match(sk, &entry->addr, remote)) 647 continue; 648 649 if (msk->pm.subflows < subflows_max) { 650 msk->pm.subflows++; 651 addrs[i++] = entry->addr; 652 } 653 } 654 rcu_read_unlock(); 655 656 /* If the array is empty, fill in the single 657 * 'IPADDRANY' local address 658 */ 659 if (!i) { 660 struct mptcp_addr_info local; 661 662 memset(&local, 0, sizeof(local)); 663 local.family = 664 #if IS_ENABLED(CONFIG_MPTCP_IPV6) 665 remote->family == AF_INET6 && 666 ipv6_addr_v4mapped(&remote->addr6) ? AF_INET : 667 #endif 668 remote->family; 669 670 if (!mptcp_pm_addr_families_match(sk, &local, remote)) 671 return 0; 672 673 msk->pm.subflows++; 674 addrs[i++] = local; 675 } 676 677 return i; 678 } 679 680 static void mptcp_pm_nl_add_addr_received(struct mptcp_sock *msk) 681 { 682 struct mptcp_addr_info addrs[MPTCP_PM_ADDR_MAX]; 683 struct sock *sk = (struct sock *)msk; 684 unsigned int add_addr_accept_max; 685 struct mptcp_addr_info remote; 686 unsigned int subflows_max; 687 int i, nr; 688 689 add_addr_accept_max = mptcp_pm_get_add_addr_accept_max(msk); 690 subflows_max = mptcp_pm_get_subflows_max(msk); 691 692 pr_debug("accepted %d:%d remote family %d", 693 msk->pm.add_addr_accepted, add_addr_accept_max, 694 msk->pm.remote.family); 695 696 remote = msk->pm.remote; 697 mptcp_pm_announce_addr(msk, &remote, true); 698 mptcp_pm_nl_addr_send_ack(msk); 699 700 if (lookup_subflow_by_daddr(&msk->conn_list, &remote)) 701 return; 702 703 /* pick id 0 port, if none is provided the remote address */ 704 if (!remote.port) 705 remote.port = sk->sk_dport; 706 707 /* connect to the specified remote address, using whatever 708 * local address the routing configuration will pick. 709 */ 710 nr = fill_local_addresses_vec(msk, &remote, addrs); 711 if (nr == 0) 712 return; 713 714 msk->pm.add_addr_accepted++; 715 if (msk->pm.add_addr_accepted >= add_addr_accept_max || 716 msk->pm.subflows >= subflows_max) 717 WRITE_ONCE(msk->pm.accept_addr, false); 718 719 spin_unlock_bh(&msk->pm.lock); 720 for (i = 0; i < nr; i++) 721 __mptcp_subflow_connect(sk, &addrs[i], &remote); 722 spin_lock_bh(&msk->pm.lock); 723 } 724 725 void mptcp_pm_nl_addr_send_ack(struct mptcp_sock *msk) 726 { 727 struct mptcp_subflow_context *subflow; 728 729 msk_owned_by_me(msk); 730 lockdep_assert_held(&msk->pm.lock); 731 732 if (!mptcp_pm_should_add_signal(msk) && 733 !mptcp_pm_should_rm_signal(msk)) 734 return; 735 736 subflow = list_first_entry_or_null(&msk->conn_list, typeof(*subflow), node); 737 if (subflow) 738 mptcp_pm_send_ack(msk, subflow, false, false); 739 } 740 741 int mptcp_pm_nl_mp_prio_send_ack(struct mptcp_sock *msk, 742 struct mptcp_addr_info *addr, 743 struct mptcp_addr_info *rem, 744 u8 bkup) 745 { 746 struct mptcp_subflow_context *subflow; 747 748 pr_debug("bkup=%d", bkup); 749 750 mptcp_for_each_subflow(msk, subflow) { 751 struct sock *ssk = mptcp_subflow_tcp_sock(subflow); 752 struct mptcp_addr_info local, remote; 753 754 mptcp_local_address((struct sock_common *)ssk, &local); 755 if (!mptcp_addresses_equal(&local, addr, addr->port)) 756 continue; 757 758 if (rem && rem->family != AF_UNSPEC) { 759 remote_address((struct sock_common *)ssk, &remote); 760 if (!mptcp_addresses_equal(&remote, rem, rem->port)) 761 continue; 762 } 763 764 __mptcp_pm_send_ack(msk, subflow, true, bkup); 765 return 0; 766 } 767 768 return -EINVAL; 769 } 770 771 static bool mptcp_local_id_match(const struct mptcp_sock *msk, u8 local_id, u8 id) 772 { 773 return local_id == id || (!local_id && msk->mpc_endpoint_id == id); 774 } 775 776 static void mptcp_pm_nl_rm_addr_or_subflow(struct mptcp_sock *msk, 777 const struct mptcp_rm_list *rm_list, 778 enum linux_mptcp_mib_field rm_type) 779 { 780 struct mptcp_subflow_context *subflow, *tmp; 781 struct sock *sk = (struct sock *)msk; 782 u8 i; 783 784 pr_debug("%s rm_list_nr %d", 785 rm_type == MPTCP_MIB_RMADDR ? "address" : "subflow", rm_list->nr); 786 787 msk_owned_by_me(msk); 788 789 if (sk->sk_state == TCP_LISTEN) 790 return; 791 792 if (!rm_list->nr) 793 return; 794 795 if (list_empty(&msk->conn_list)) 796 return; 797 798 for (i = 0; i < rm_list->nr; i++) { 799 u8 rm_id = rm_list->ids[i]; 800 bool removed = false; 801 802 mptcp_for_each_subflow_safe(msk, subflow, tmp) { 803 struct sock *ssk = mptcp_subflow_tcp_sock(subflow); 804 int how = RCV_SHUTDOWN | SEND_SHUTDOWN; 805 u8 id = subflow->local_id; 806 807 if (rm_type == MPTCP_MIB_RMADDR && subflow->remote_id != rm_id) 808 continue; 809 if (rm_type == MPTCP_MIB_RMSUBFLOW && !mptcp_local_id_match(msk, id, rm_id)) 810 continue; 811 812 pr_debug(" -> %s rm_list_ids[%d]=%u local_id=%u remote_id=%u mpc_id=%u", 813 rm_type == MPTCP_MIB_RMADDR ? "address" : "subflow", 814 i, rm_id, subflow->local_id, subflow->remote_id, 815 msk->mpc_endpoint_id); 816 spin_unlock_bh(&msk->pm.lock); 817 mptcp_subflow_shutdown(sk, ssk, how); 818 819 /* the following takes care of updating the subflows counter */ 820 mptcp_close_ssk(sk, ssk, subflow); 821 spin_lock_bh(&msk->pm.lock); 822 823 removed = true; 824 __MPTCP_INC_STATS(sock_net(sk), rm_type); 825 } 826 if (rm_type == MPTCP_MIB_RMSUBFLOW) 827 __set_bit(rm_id ? rm_id : msk->mpc_endpoint_id, msk->pm.id_avail_bitmap); 828 if (!removed) 829 continue; 830 831 if (!mptcp_pm_is_kernel(msk)) 832 continue; 833 834 if (rm_type == MPTCP_MIB_RMADDR) { 835 msk->pm.add_addr_accepted--; 836 WRITE_ONCE(msk->pm.accept_addr, true); 837 } else if (rm_type == MPTCP_MIB_RMSUBFLOW) { 838 msk->pm.local_addr_used--; 839 } 840 } 841 } 842 843 static void mptcp_pm_nl_rm_addr_received(struct mptcp_sock *msk) 844 { 845 mptcp_pm_nl_rm_addr_or_subflow(msk, &msk->pm.rm_list_rx, MPTCP_MIB_RMADDR); 846 } 847 848 void mptcp_pm_nl_rm_subflow_received(struct mptcp_sock *msk, 849 const struct mptcp_rm_list *rm_list) 850 { 851 mptcp_pm_nl_rm_addr_or_subflow(msk, rm_list, MPTCP_MIB_RMSUBFLOW); 852 } 853 854 void mptcp_pm_nl_work(struct mptcp_sock *msk) 855 { 856 struct mptcp_pm_data *pm = &msk->pm; 857 858 msk_owned_by_me(msk); 859 860 if (!(pm->status & MPTCP_PM_WORK_MASK)) 861 return; 862 863 spin_lock_bh(&msk->pm.lock); 864 865 pr_debug("msk=%p status=%x", msk, pm->status); 866 if (pm->status & BIT(MPTCP_PM_ADD_ADDR_RECEIVED)) { 867 pm->status &= ~BIT(MPTCP_PM_ADD_ADDR_RECEIVED); 868 mptcp_pm_nl_add_addr_received(msk); 869 } 870 if (pm->status & BIT(MPTCP_PM_ADD_ADDR_SEND_ACK)) { 871 pm->status &= ~BIT(MPTCP_PM_ADD_ADDR_SEND_ACK); 872 mptcp_pm_nl_addr_send_ack(msk); 873 } 874 if (pm->status & BIT(MPTCP_PM_RM_ADDR_RECEIVED)) { 875 pm->status &= ~BIT(MPTCP_PM_RM_ADDR_RECEIVED); 876 mptcp_pm_nl_rm_addr_received(msk); 877 } 878 if (pm->status & BIT(MPTCP_PM_ESTABLISHED)) { 879 pm->status &= ~BIT(MPTCP_PM_ESTABLISHED); 880 mptcp_pm_nl_fully_established(msk); 881 } 882 if (pm->status & BIT(MPTCP_PM_SUBFLOW_ESTABLISHED)) { 883 pm->status &= ~BIT(MPTCP_PM_SUBFLOW_ESTABLISHED); 884 mptcp_pm_nl_subflow_established(msk); 885 } 886 887 spin_unlock_bh(&msk->pm.lock); 888 } 889 890 static bool address_use_port(struct mptcp_pm_addr_entry *entry) 891 { 892 return (entry->flags & 893 (MPTCP_PM_ADDR_FLAG_SIGNAL | MPTCP_PM_ADDR_FLAG_SUBFLOW)) == 894 MPTCP_PM_ADDR_FLAG_SIGNAL; 895 } 896 897 /* caller must ensure the RCU grace period is already elapsed */ 898 static void __mptcp_pm_release_addr_entry(struct mptcp_pm_addr_entry *entry) 899 { 900 if (entry->lsk) 901 sock_release(entry->lsk); 902 kfree(entry); 903 } 904 905 static int mptcp_pm_nl_append_new_local_addr(struct pm_nl_pernet *pernet, 906 struct mptcp_pm_addr_entry *entry) 907 { 908 struct mptcp_pm_addr_entry *cur, *del_entry = NULL; 909 unsigned int addr_max; 910 int ret = -EINVAL; 911 912 spin_lock_bh(&pernet->lock); 913 /* to keep the code simple, don't do IDR-like allocation for address ID, 914 * just bail when we exceed limits 915 */ 916 if (pernet->next_id == MPTCP_PM_MAX_ADDR_ID) 917 pernet->next_id = 1; 918 if (pernet->addrs >= MPTCP_PM_ADDR_MAX) { 919 ret = -ERANGE; 920 goto out; 921 } 922 if (test_bit(entry->addr.id, pernet->id_bitmap)) { 923 ret = -EBUSY; 924 goto out; 925 } 926 927 /* do not insert duplicate address, differentiate on port only 928 * singled addresses 929 */ 930 if (!address_use_port(entry)) 931 entry->addr.port = 0; 932 list_for_each_entry(cur, &pernet->local_addr_list, list) { 933 if (mptcp_addresses_equal(&cur->addr, &entry->addr, 934 cur->addr.port || entry->addr.port)) { 935 /* allow replacing the exiting endpoint only if such 936 * endpoint is an implicit one and the user-space 937 * did not provide an endpoint id 938 */ 939 if (!(cur->flags & MPTCP_PM_ADDR_FLAG_IMPLICIT)) { 940 ret = -EEXIST; 941 goto out; 942 } 943 if (entry->addr.id) 944 goto out; 945 946 pernet->addrs--; 947 entry->addr.id = cur->addr.id; 948 list_del_rcu(&cur->list); 949 del_entry = cur; 950 break; 951 } 952 } 953 954 if (!entry->addr.id) { 955 find_next: 956 entry->addr.id = find_next_zero_bit(pernet->id_bitmap, 957 MPTCP_PM_MAX_ADDR_ID + 1, 958 pernet->next_id); 959 if (!entry->addr.id && pernet->next_id != 1) { 960 pernet->next_id = 1; 961 goto find_next; 962 } 963 } 964 965 if (!entry->addr.id) 966 goto out; 967 968 __set_bit(entry->addr.id, pernet->id_bitmap); 969 if (entry->addr.id > pernet->next_id) 970 pernet->next_id = entry->addr.id; 971 972 if (entry->flags & MPTCP_PM_ADDR_FLAG_SIGNAL) { 973 addr_max = pernet->add_addr_signal_max; 974 WRITE_ONCE(pernet->add_addr_signal_max, addr_max + 1); 975 } 976 if (entry->flags & MPTCP_PM_ADDR_FLAG_SUBFLOW) { 977 addr_max = pernet->local_addr_max; 978 WRITE_ONCE(pernet->local_addr_max, addr_max + 1); 979 } 980 981 pernet->addrs++; 982 if (!entry->addr.port) 983 list_add_tail_rcu(&entry->list, &pernet->local_addr_list); 984 else 985 list_add_rcu(&entry->list, &pernet->local_addr_list); 986 ret = entry->addr.id; 987 988 out: 989 spin_unlock_bh(&pernet->lock); 990 991 /* just replaced an existing entry, free it */ 992 if (del_entry) { 993 synchronize_rcu(); 994 __mptcp_pm_release_addr_entry(del_entry); 995 } 996 return ret; 997 } 998 999 static struct lock_class_key mptcp_slock_keys[2]; 1000 static struct lock_class_key mptcp_keys[2]; 1001 1002 static int mptcp_pm_nl_create_listen_socket(struct sock *sk, 1003 struct mptcp_pm_addr_entry *entry) 1004 { 1005 bool is_ipv6 = sk->sk_family == AF_INET6; 1006 int addrlen = sizeof(struct sockaddr_in); 1007 struct sockaddr_storage addr; 1008 struct socket *ssock; 1009 struct sock *newsk; 1010 int backlog = 1024; 1011 int err; 1012 1013 err = sock_create_kern(sock_net(sk), entry->addr.family, 1014 SOCK_STREAM, IPPROTO_MPTCP, &entry->lsk); 1015 if (err) 1016 return err; 1017 1018 newsk = entry->lsk->sk; 1019 if (!newsk) 1020 return -EINVAL; 1021 1022 /* The subflow socket lock is acquired in a nested to the msk one 1023 * in several places, even by the TCP stack, and this msk is a kernel 1024 * socket: lockdep complains. Instead of propagating the _nested 1025 * modifiers in several places, re-init the lock class for the msk 1026 * socket to an mptcp specific one. 1027 */ 1028 sock_lock_init_class_and_name(newsk, 1029 is_ipv6 ? "mlock-AF_INET6" : "mlock-AF_INET", 1030 &mptcp_slock_keys[is_ipv6], 1031 is_ipv6 ? "msk_lock-AF_INET6" : "msk_lock-AF_INET", 1032 &mptcp_keys[is_ipv6]); 1033 1034 lock_sock(newsk); 1035 ssock = __mptcp_nmpc_socket(mptcp_sk(newsk)); 1036 release_sock(newsk); 1037 if (IS_ERR(ssock)) 1038 return PTR_ERR(ssock); 1039 1040 mptcp_info2sockaddr(&entry->addr, &addr, entry->addr.family); 1041 #if IS_ENABLED(CONFIG_MPTCP_IPV6) 1042 if (entry->addr.family == AF_INET6) 1043 addrlen = sizeof(struct sockaddr_in6); 1044 #endif 1045 err = kernel_bind(ssock, (struct sockaddr *)&addr, addrlen); 1046 if (err) 1047 return err; 1048 1049 err = kernel_listen(ssock, backlog); 1050 if (err) 1051 return err; 1052 1053 mptcp_event_pm_listener(ssock->sk, MPTCP_EVENT_LISTENER_CREATED); 1054 1055 return 0; 1056 } 1057 1058 int mptcp_pm_nl_get_local_id(struct mptcp_sock *msk, struct mptcp_addr_info *skc) 1059 { 1060 struct mptcp_pm_addr_entry *entry; 1061 struct pm_nl_pernet *pernet; 1062 int ret = -1; 1063 1064 pernet = pm_nl_get_pernet_from_msk(msk); 1065 1066 rcu_read_lock(); 1067 list_for_each_entry_rcu(entry, &pernet->local_addr_list, list) { 1068 if (mptcp_addresses_equal(&entry->addr, skc, entry->addr.port)) { 1069 ret = entry->addr.id; 1070 break; 1071 } 1072 } 1073 rcu_read_unlock(); 1074 if (ret >= 0) 1075 return ret; 1076 1077 /* address not found, add to local list */ 1078 entry = kmalloc(sizeof(*entry), GFP_ATOMIC); 1079 if (!entry) 1080 return -ENOMEM; 1081 1082 entry->addr = *skc; 1083 entry->addr.id = 0; 1084 entry->addr.port = 0; 1085 entry->ifindex = 0; 1086 entry->flags = MPTCP_PM_ADDR_FLAG_IMPLICIT; 1087 entry->lsk = NULL; 1088 ret = mptcp_pm_nl_append_new_local_addr(pernet, entry); 1089 if (ret < 0) 1090 kfree(entry); 1091 1092 return ret; 1093 } 1094 1095 #define MPTCP_PM_CMD_GRP_OFFSET 0 1096 #define MPTCP_PM_EV_GRP_OFFSET 1 1097 1098 static const struct genl_multicast_group mptcp_pm_mcgrps[] = { 1099 [MPTCP_PM_CMD_GRP_OFFSET] = { .name = MPTCP_PM_CMD_GRP_NAME, }, 1100 [MPTCP_PM_EV_GRP_OFFSET] = { .name = MPTCP_PM_EV_GRP_NAME, 1101 .flags = GENL_UNS_ADMIN_PERM, 1102 }, 1103 }; 1104 1105 static const struct nla_policy 1106 mptcp_pm_addr_policy[MPTCP_PM_ADDR_ATTR_MAX + 1] = { 1107 [MPTCP_PM_ADDR_ATTR_FAMILY] = { .type = NLA_U16, }, 1108 [MPTCP_PM_ADDR_ATTR_ID] = { .type = NLA_U8, }, 1109 [MPTCP_PM_ADDR_ATTR_ADDR4] = { .type = NLA_U32, }, 1110 [MPTCP_PM_ADDR_ATTR_ADDR6] = 1111 NLA_POLICY_EXACT_LEN(sizeof(struct in6_addr)), 1112 [MPTCP_PM_ADDR_ATTR_PORT] = { .type = NLA_U16 }, 1113 [MPTCP_PM_ADDR_ATTR_FLAGS] = { .type = NLA_U32 }, 1114 [MPTCP_PM_ADDR_ATTR_IF_IDX] = { .type = NLA_S32 }, 1115 }; 1116 1117 static const struct nla_policy mptcp_pm_policy[MPTCP_PM_ATTR_MAX + 1] = { 1118 [MPTCP_PM_ATTR_ADDR] = 1119 NLA_POLICY_NESTED(mptcp_pm_addr_policy), 1120 [MPTCP_PM_ATTR_RCV_ADD_ADDRS] = { .type = NLA_U32, }, 1121 [MPTCP_PM_ATTR_SUBFLOWS] = { .type = NLA_U32, }, 1122 [MPTCP_PM_ATTR_TOKEN] = { .type = NLA_U32, }, 1123 [MPTCP_PM_ATTR_LOC_ID] = { .type = NLA_U8, }, 1124 [MPTCP_PM_ATTR_ADDR_REMOTE] = 1125 NLA_POLICY_NESTED(mptcp_pm_addr_policy), 1126 }; 1127 1128 void mptcp_pm_nl_subflow_chk_stale(const struct mptcp_sock *msk, struct sock *ssk) 1129 { 1130 struct mptcp_subflow_context *iter, *subflow = mptcp_subflow_ctx(ssk); 1131 struct sock *sk = (struct sock *)msk; 1132 unsigned int active_max_loss_cnt; 1133 struct net *net = sock_net(sk); 1134 unsigned int stale_loss_cnt; 1135 bool slow; 1136 1137 stale_loss_cnt = mptcp_stale_loss_cnt(net); 1138 if (subflow->stale || !stale_loss_cnt || subflow->stale_count <= stale_loss_cnt) 1139 return; 1140 1141 /* look for another available subflow not in loss state */ 1142 active_max_loss_cnt = max_t(int, stale_loss_cnt - 1, 1); 1143 mptcp_for_each_subflow(msk, iter) { 1144 if (iter != subflow && mptcp_subflow_active(iter) && 1145 iter->stale_count < active_max_loss_cnt) { 1146 /* we have some alternatives, try to mark this subflow as idle ...*/ 1147 slow = lock_sock_fast(ssk); 1148 if (!tcp_rtx_and_write_queues_empty(ssk)) { 1149 subflow->stale = 1; 1150 __mptcp_retransmit_pending_data(sk); 1151 MPTCP_INC_STATS(net, MPTCP_MIB_SUBFLOWSTALE); 1152 } 1153 unlock_sock_fast(ssk, slow); 1154 1155 /* always try to push the pending data regardless of re-injections: 1156 * we can possibly use backup subflows now, and subflow selection 1157 * is cheap under the msk socket lock 1158 */ 1159 __mptcp_push_pending(sk, 0); 1160 return; 1161 } 1162 } 1163 } 1164 1165 static int mptcp_pm_family_to_addr(int family) 1166 { 1167 #if IS_ENABLED(CONFIG_MPTCP_IPV6) 1168 if (family == AF_INET6) 1169 return MPTCP_PM_ADDR_ATTR_ADDR6; 1170 #endif 1171 return MPTCP_PM_ADDR_ATTR_ADDR4; 1172 } 1173 1174 static int mptcp_pm_parse_pm_addr_attr(struct nlattr *tb[], 1175 const struct nlattr *attr, 1176 struct genl_info *info, 1177 struct mptcp_addr_info *addr, 1178 bool require_family) 1179 { 1180 int err, addr_addr; 1181 1182 if (!attr) { 1183 GENL_SET_ERR_MSG(info, "missing address info"); 1184 return -EINVAL; 1185 } 1186 1187 /* no validation needed - was already done via nested policy */ 1188 err = nla_parse_nested_deprecated(tb, MPTCP_PM_ADDR_ATTR_MAX, attr, 1189 mptcp_pm_addr_policy, info->extack); 1190 if (err) 1191 return err; 1192 1193 if (tb[MPTCP_PM_ADDR_ATTR_ID]) 1194 addr->id = nla_get_u8(tb[MPTCP_PM_ADDR_ATTR_ID]); 1195 1196 if (!tb[MPTCP_PM_ADDR_ATTR_FAMILY]) { 1197 if (!require_family) 1198 return 0; 1199 1200 NL_SET_ERR_MSG_ATTR(info->extack, attr, 1201 "missing family"); 1202 return -EINVAL; 1203 } 1204 1205 addr->family = nla_get_u16(tb[MPTCP_PM_ADDR_ATTR_FAMILY]); 1206 if (addr->family != AF_INET 1207 #if IS_ENABLED(CONFIG_MPTCP_IPV6) 1208 && addr->family != AF_INET6 1209 #endif 1210 ) { 1211 NL_SET_ERR_MSG_ATTR(info->extack, attr, 1212 "unknown address family"); 1213 return -EINVAL; 1214 } 1215 addr_addr = mptcp_pm_family_to_addr(addr->family); 1216 if (!tb[addr_addr]) { 1217 NL_SET_ERR_MSG_ATTR(info->extack, attr, 1218 "missing address data"); 1219 return -EINVAL; 1220 } 1221 1222 #if IS_ENABLED(CONFIG_MPTCP_IPV6) 1223 if (addr->family == AF_INET6) 1224 addr->addr6 = nla_get_in6_addr(tb[addr_addr]); 1225 else 1226 #endif 1227 addr->addr.s_addr = nla_get_in_addr(tb[addr_addr]); 1228 1229 if (tb[MPTCP_PM_ADDR_ATTR_PORT]) 1230 addr->port = htons(nla_get_u16(tb[MPTCP_PM_ADDR_ATTR_PORT])); 1231 1232 return 0; 1233 } 1234 1235 int mptcp_pm_parse_addr(struct nlattr *attr, struct genl_info *info, 1236 struct mptcp_addr_info *addr) 1237 { 1238 struct nlattr *tb[MPTCP_PM_ADDR_ATTR_MAX + 1]; 1239 1240 memset(addr, 0, sizeof(*addr)); 1241 1242 return mptcp_pm_parse_pm_addr_attr(tb, attr, info, addr, true); 1243 } 1244 1245 int mptcp_pm_parse_entry(struct nlattr *attr, struct genl_info *info, 1246 bool require_family, 1247 struct mptcp_pm_addr_entry *entry) 1248 { 1249 struct nlattr *tb[MPTCP_PM_ADDR_ATTR_MAX + 1]; 1250 int err; 1251 1252 memset(entry, 0, sizeof(*entry)); 1253 1254 err = mptcp_pm_parse_pm_addr_attr(tb, attr, info, &entry->addr, require_family); 1255 if (err) 1256 return err; 1257 1258 if (tb[MPTCP_PM_ADDR_ATTR_IF_IDX]) { 1259 u32 val = nla_get_s32(tb[MPTCP_PM_ADDR_ATTR_IF_IDX]); 1260 1261 entry->ifindex = val; 1262 } 1263 1264 if (tb[MPTCP_PM_ADDR_ATTR_FLAGS]) 1265 entry->flags = nla_get_u32(tb[MPTCP_PM_ADDR_ATTR_FLAGS]); 1266 1267 if (tb[MPTCP_PM_ADDR_ATTR_PORT]) 1268 entry->addr.port = htons(nla_get_u16(tb[MPTCP_PM_ADDR_ATTR_PORT])); 1269 1270 return 0; 1271 } 1272 1273 static struct pm_nl_pernet *genl_info_pm_nl(struct genl_info *info) 1274 { 1275 return pm_nl_get_pernet(genl_info_net(info)); 1276 } 1277 1278 static int mptcp_nl_add_subflow_or_signal_addr(struct net *net) 1279 { 1280 struct mptcp_sock *msk; 1281 long s_slot = 0, s_num = 0; 1282 1283 while ((msk = mptcp_token_iter_next(net, &s_slot, &s_num)) != NULL) { 1284 struct sock *sk = (struct sock *)msk; 1285 1286 if (!READ_ONCE(msk->fully_established) || 1287 mptcp_pm_is_userspace(msk)) 1288 goto next; 1289 1290 lock_sock(sk); 1291 spin_lock_bh(&msk->pm.lock); 1292 mptcp_pm_create_subflow_or_signal_addr(msk); 1293 spin_unlock_bh(&msk->pm.lock); 1294 release_sock(sk); 1295 1296 next: 1297 sock_put(sk); 1298 cond_resched(); 1299 } 1300 1301 return 0; 1302 } 1303 1304 static int mptcp_nl_cmd_add_addr(struct sk_buff *skb, struct genl_info *info) 1305 { 1306 struct nlattr *attr = info->attrs[MPTCP_PM_ATTR_ADDR]; 1307 struct pm_nl_pernet *pernet = genl_info_pm_nl(info); 1308 struct mptcp_pm_addr_entry addr, *entry; 1309 int ret; 1310 1311 ret = mptcp_pm_parse_entry(attr, info, true, &addr); 1312 if (ret < 0) 1313 return ret; 1314 1315 if (addr.addr.port && !(addr.flags & MPTCP_PM_ADDR_FLAG_SIGNAL)) { 1316 GENL_SET_ERR_MSG(info, "flags must have signal when using port"); 1317 return -EINVAL; 1318 } 1319 1320 if (addr.flags & MPTCP_PM_ADDR_FLAG_SIGNAL && 1321 addr.flags & MPTCP_PM_ADDR_FLAG_FULLMESH) { 1322 GENL_SET_ERR_MSG(info, "flags mustn't have both signal and fullmesh"); 1323 return -EINVAL; 1324 } 1325 1326 if (addr.flags & MPTCP_PM_ADDR_FLAG_IMPLICIT) { 1327 GENL_SET_ERR_MSG(info, "can't create IMPLICIT endpoint"); 1328 return -EINVAL; 1329 } 1330 1331 entry = kzalloc(sizeof(*entry), GFP_KERNEL_ACCOUNT); 1332 if (!entry) { 1333 GENL_SET_ERR_MSG(info, "can't allocate addr"); 1334 return -ENOMEM; 1335 } 1336 1337 *entry = addr; 1338 if (entry->addr.port) { 1339 ret = mptcp_pm_nl_create_listen_socket(skb->sk, entry); 1340 if (ret) { 1341 GENL_SET_ERR_MSG_FMT(info, "create listen socket error: %d", ret); 1342 goto out_free; 1343 } 1344 } 1345 ret = mptcp_pm_nl_append_new_local_addr(pernet, entry); 1346 if (ret < 0) { 1347 GENL_SET_ERR_MSG_FMT(info, "too many addresses or duplicate one: %d", ret); 1348 goto out_free; 1349 } 1350 1351 mptcp_nl_add_subflow_or_signal_addr(sock_net(skb->sk)); 1352 return 0; 1353 1354 out_free: 1355 __mptcp_pm_release_addr_entry(entry); 1356 return ret; 1357 } 1358 1359 int mptcp_pm_nl_get_flags_and_ifindex_by_id(struct mptcp_sock *msk, unsigned int id, 1360 u8 *flags, int *ifindex) 1361 { 1362 struct mptcp_pm_addr_entry *entry; 1363 struct sock *sk = (struct sock *)msk; 1364 struct net *net = sock_net(sk); 1365 1366 rcu_read_lock(); 1367 entry = __lookup_addr_by_id(pm_nl_get_pernet(net), id); 1368 if (entry) { 1369 *flags = entry->flags; 1370 *ifindex = entry->ifindex; 1371 } 1372 rcu_read_unlock(); 1373 1374 return 0; 1375 } 1376 1377 static bool remove_anno_list_by_saddr(struct mptcp_sock *msk, 1378 const struct mptcp_addr_info *addr) 1379 { 1380 struct mptcp_pm_add_entry *entry; 1381 1382 entry = mptcp_pm_del_add_timer(msk, addr, false); 1383 if (entry) { 1384 list_del(&entry->list); 1385 kfree(entry); 1386 return true; 1387 } 1388 1389 return false; 1390 } 1391 1392 static bool mptcp_pm_remove_anno_addr(struct mptcp_sock *msk, 1393 const struct mptcp_addr_info *addr, 1394 bool force) 1395 { 1396 struct mptcp_rm_list list = { .nr = 0 }; 1397 bool ret; 1398 1399 list.ids[list.nr++] = addr->id; 1400 1401 ret = remove_anno_list_by_saddr(msk, addr); 1402 if (ret || force) { 1403 spin_lock_bh(&msk->pm.lock); 1404 mptcp_pm_remove_addr(msk, &list); 1405 spin_unlock_bh(&msk->pm.lock); 1406 } 1407 return ret; 1408 } 1409 1410 static int mptcp_nl_remove_subflow_and_signal_addr(struct net *net, 1411 const struct mptcp_pm_addr_entry *entry) 1412 { 1413 const struct mptcp_addr_info *addr = &entry->addr; 1414 struct mptcp_rm_list list = { .nr = 0 }; 1415 long s_slot = 0, s_num = 0; 1416 struct mptcp_sock *msk; 1417 1418 pr_debug("remove_id=%d", addr->id); 1419 1420 list.ids[list.nr++] = addr->id; 1421 1422 while ((msk = mptcp_token_iter_next(net, &s_slot, &s_num)) != NULL) { 1423 struct sock *sk = (struct sock *)msk; 1424 bool remove_subflow; 1425 1426 if (mptcp_pm_is_userspace(msk)) 1427 goto next; 1428 1429 if (list_empty(&msk->conn_list)) { 1430 mptcp_pm_remove_anno_addr(msk, addr, false); 1431 goto next; 1432 } 1433 1434 lock_sock(sk); 1435 remove_subflow = lookup_subflow_by_saddr(&msk->conn_list, addr); 1436 mptcp_pm_remove_anno_addr(msk, addr, remove_subflow && 1437 !(entry->flags & MPTCP_PM_ADDR_FLAG_IMPLICIT)); 1438 if (remove_subflow) 1439 mptcp_pm_remove_subflow(msk, &list); 1440 release_sock(sk); 1441 1442 next: 1443 sock_put(sk); 1444 cond_resched(); 1445 } 1446 1447 return 0; 1448 } 1449 1450 static int mptcp_nl_remove_id_zero_address(struct net *net, 1451 struct mptcp_addr_info *addr) 1452 { 1453 struct mptcp_rm_list list = { .nr = 0 }; 1454 long s_slot = 0, s_num = 0; 1455 struct mptcp_sock *msk; 1456 1457 list.ids[list.nr++] = 0; 1458 1459 while ((msk = mptcp_token_iter_next(net, &s_slot, &s_num)) != NULL) { 1460 struct sock *sk = (struct sock *)msk; 1461 struct mptcp_addr_info msk_local; 1462 1463 if (list_empty(&msk->conn_list) || mptcp_pm_is_userspace(msk)) 1464 goto next; 1465 1466 mptcp_local_address((struct sock_common *)msk, &msk_local); 1467 if (!mptcp_addresses_equal(&msk_local, addr, addr->port)) 1468 goto next; 1469 1470 lock_sock(sk); 1471 spin_lock_bh(&msk->pm.lock); 1472 mptcp_pm_remove_addr(msk, &list); 1473 mptcp_pm_nl_rm_subflow_received(msk, &list); 1474 spin_unlock_bh(&msk->pm.lock); 1475 release_sock(sk); 1476 1477 next: 1478 sock_put(sk); 1479 cond_resched(); 1480 } 1481 1482 return 0; 1483 } 1484 1485 static int mptcp_nl_cmd_del_addr(struct sk_buff *skb, struct genl_info *info) 1486 { 1487 struct nlattr *attr = info->attrs[MPTCP_PM_ATTR_ADDR]; 1488 struct pm_nl_pernet *pernet = genl_info_pm_nl(info); 1489 struct mptcp_pm_addr_entry addr, *entry; 1490 unsigned int addr_max; 1491 int ret; 1492 1493 ret = mptcp_pm_parse_entry(attr, info, false, &addr); 1494 if (ret < 0) 1495 return ret; 1496 1497 /* the zero id address is special: the first address used by the msk 1498 * always gets such an id, so different subflows can have different zero 1499 * id addresses. Additionally zero id is not accounted for in id_bitmap. 1500 * Let's use an 'mptcp_rm_list' instead of the common remove code. 1501 */ 1502 if (addr.addr.id == 0) 1503 return mptcp_nl_remove_id_zero_address(sock_net(skb->sk), &addr.addr); 1504 1505 spin_lock_bh(&pernet->lock); 1506 entry = __lookup_addr_by_id(pernet, addr.addr.id); 1507 if (!entry) { 1508 GENL_SET_ERR_MSG(info, "address not found"); 1509 spin_unlock_bh(&pernet->lock); 1510 return -EINVAL; 1511 } 1512 if (entry->flags & MPTCP_PM_ADDR_FLAG_SIGNAL) { 1513 addr_max = pernet->add_addr_signal_max; 1514 WRITE_ONCE(pernet->add_addr_signal_max, addr_max - 1); 1515 } 1516 if (entry->flags & MPTCP_PM_ADDR_FLAG_SUBFLOW) { 1517 addr_max = pernet->local_addr_max; 1518 WRITE_ONCE(pernet->local_addr_max, addr_max - 1); 1519 } 1520 1521 pernet->addrs--; 1522 list_del_rcu(&entry->list); 1523 __clear_bit(entry->addr.id, pernet->id_bitmap); 1524 spin_unlock_bh(&pernet->lock); 1525 1526 mptcp_nl_remove_subflow_and_signal_addr(sock_net(skb->sk), entry); 1527 synchronize_rcu(); 1528 __mptcp_pm_release_addr_entry(entry); 1529 1530 return ret; 1531 } 1532 1533 void mptcp_pm_remove_addrs(struct mptcp_sock *msk, struct list_head *rm_list) 1534 { 1535 struct mptcp_rm_list alist = { .nr = 0 }; 1536 struct mptcp_pm_addr_entry *entry; 1537 1538 list_for_each_entry(entry, rm_list, list) { 1539 remove_anno_list_by_saddr(msk, &entry->addr); 1540 if (alist.nr < MPTCP_RM_IDS_MAX) 1541 alist.ids[alist.nr++] = entry->addr.id; 1542 } 1543 1544 if (alist.nr) { 1545 spin_lock_bh(&msk->pm.lock); 1546 mptcp_pm_remove_addr(msk, &alist); 1547 spin_unlock_bh(&msk->pm.lock); 1548 } 1549 } 1550 1551 void mptcp_pm_remove_addrs_and_subflows(struct mptcp_sock *msk, 1552 struct list_head *rm_list) 1553 { 1554 struct mptcp_rm_list alist = { .nr = 0 }, slist = { .nr = 0 }; 1555 struct mptcp_pm_addr_entry *entry; 1556 1557 list_for_each_entry(entry, rm_list, list) { 1558 if (lookup_subflow_by_saddr(&msk->conn_list, &entry->addr) && 1559 slist.nr < MPTCP_RM_IDS_MAX) 1560 slist.ids[slist.nr++] = entry->addr.id; 1561 1562 if (remove_anno_list_by_saddr(msk, &entry->addr) && 1563 alist.nr < MPTCP_RM_IDS_MAX) 1564 alist.ids[alist.nr++] = entry->addr.id; 1565 } 1566 1567 if (alist.nr) { 1568 spin_lock_bh(&msk->pm.lock); 1569 mptcp_pm_remove_addr(msk, &alist); 1570 spin_unlock_bh(&msk->pm.lock); 1571 } 1572 if (slist.nr) 1573 mptcp_pm_remove_subflow(msk, &slist); 1574 } 1575 1576 static void mptcp_nl_remove_addrs_list(struct net *net, 1577 struct list_head *rm_list) 1578 { 1579 long s_slot = 0, s_num = 0; 1580 struct mptcp_sock *msk; 1581 1582 if (list_empty(rm_list)) 1583 return; 1584 1585 while ((msk = mptcp_token_iter_next(net, &s_slot, &s_num)) != NULL) { 1586 struct sock *sk = (struct sock *)msk; 1587 1588 if (!mptcp_pm_is_userspace(msk)) { 1589 lock_sock(sk); 1590 mptcp_pm_remove_addrs_and_subflows(msk, rm_list); 1591 release_sock(sk); 1592 } 1593 1594 sock_put(sk); 1595 cond_resched(); 1596 } 1597 } 1598 1599 /* caller must ensure the RCU grace period is already elapsed */ 1600 static void __flush_addrs(struct list_head *list) 1601 { 1602 while (!list_empty(list)) { 1603 struct mptcp_pm_addr_entry *cur; 1604 1605 cur = list_entry(list->next, 1606 struct mptcp_pm_addr_entry, list); 1607 list_del_rcu(&cur->list); 1608 __mptcp_pm_release_addr_entry(cur); 1609 } 1610 } 1611 1612 static void __reset_counters(struct pm_nl_pernet *pernet) 1613 { 1614 WRITE_ONCE(pernet->add_addr_signal_max, 0); 1615 WRITE_ONCE(pernet->add_addr_accept_max, 0); 1616 WRITE_ONCE(pernet->local_addr_max, 0); 1617 pernet->addrs = 0; 1618 } 1619 1620 static int mptcp_nl_cmd_flush_addrs(struct sk_buff *skb, struct genl_info *info) 1621 { 1622 struct pm_nl_pernet *pernet = genl_info_pm_nl(info); 1623 LIST_HEAD(free_list); 1624 1625 spin_lock_bh(&pernet->lock); 1626 list_splice_init(&pernet->local_addr_list, &free_list); 1627 __reset_counters(pernet); 1628 pernet->next_id = 1; 1629 bitmap_zero(pernet->id_bitmap, MPTCP_PM_MAX_ADDR_ID + 1); 1630 spin_unlock_bh(&pernet->lock); 1631 mptcp_nl_remove_addrs_list(sock_net(skb->sk), &free_list); 1632 synchronize_rcu(); 1633 __flush_addrs(&free_list); 1634 return 0; 1635 } 1636 1637 static int mptcp_nl_fill_addr(struct sk_buff *skb, 1638 struct mptcp_pm_addr_entry *entry) 1639 { 1640 struct mptcp_addr_info *addr = &entry->addr; 1641 struct nlattr *attr; 1642 1643 attr = nla_nest_start(skb, MPTCP_PM_ATTR_ADDR); 1644 if (!attr) 1645 return -EMSGSIZE; 1646 1647 if (nla_put_u16(skb, MPTCP_PM_ADDR_ATTR_FAMILY, addr->family)) 1648 goto nla_put_failure; 1649 if (nla_put_u16(skb, MPTCP_PM_ADDR_ATTR_PORT, ntohs(addr->port))) 1650 goto nla_put_failure; 1651 if (nla_put_u8(skb, MPTCP_PM_ADDR_ATTR_ID, addr->id)) 1652 goto nla_put_failure; 1653 if (nla_put_u32(skb, MPTCP_PM_ADDR_ATTR_FLAGS, entry->flags)) 1654 goto nla_put_failure; 1655 if (entry->ifindex && 1656 nla_put_s32(skb, MPTCP_PM_ADDR_ATTR_IF_IDX, entry->ifindex)) 1657 goto nla_put_failure; 1658 1659 if (addr->family == AF_INET && 1660 nla_put_in_addr(skb, MPTCP_PM_ADDR_ATTR_ADDR4, 1661 addr->addr.s_addr)) 1662 goto nla_put_failure; 1663 #if IS_ENABLED(CONFIG_MPTCP_IPV6) 1664 else if (addr->family == AF_INET6 && 1665 nla_put_in6_addr(skb, MPTCP_PM_ADDR_ATTR_ADDR6, &addr->addr6)) 1666 goto nla_put_failure; 1667 #endif 1668 nla_nest_end(skb, attr); 1669 return 0; 1670 1671 nla_put_failure: 1672 nla_nest_cancel(skb, attr); 1673 return -EMSGSIZE; 1674 } 1675 1676 static int mptcp_nl_cmd_get_addr(struct sk_buff *skb, struct genl_info *info) 1677 { 1678 struct nlattr *attr = info->attrs[MPTCP_PM_ATTR_ADDR]; 1679 struct pm_nl_pernet *pernet = genl_info_pm_nl(info); 1680 struct mptcp_pm_addr_entry addr, *entry; 1681 struct sk_buff *msg; 1682 void *reply; 1683 int ret; 1684 1685 ret = mptcp_pm_parse_entry(attr, info, false, &addr); 1686 if (ret < 0) 1687 return ret; 1688 1689 msg = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL); 1690 if (!msg) 1691 return -ENOMEM; 1692 1693 reply = genlmsg_put_reply(msg, info, &mptcp_genl_family, 0, 1694 info->genlhdr->cmd); 1695 if (!reply) { 1696 GENL_SET_ERR_MSG(info, "not enough space in Netlink message"); 1697 ret = -EMSGSIZE; 1698 goto fail; 1699 } 1700 1701 spin_lock_bh(&pernet->lock); 1702 entry = __lookup_addr_by_id(pernet, addr.addr.id); 1703 if (!entry) { 1704 GENL_SET_ERR_MSG(info, "address not found"); 1705 ret = -EINVAL; 1706 goto unlock_fail; 1707 } 1708 1709 ret = mptcp_nl_fill_addr(msg, entry); 1710 if (ret) 1711 goto unlock_fail; 1712 1713 genlmsg_end(msg, reply); 1714 ret = genlmsg_reply(msg, info); 1715 spin_unlock_bh(&pernet->lock); 1716 return ret; 1717 1718 unlock_fail: 1719 spin_unlock_bh(&pernet->lock); 1720 1721 fail: 1722 nlmsg_free(msg); 1723 return ret; 1724 } 1725 1726 static int mptcp_nl_cmd_dump_addrs(struct sk_buff *msg, 1727 struct netlink_callback *cb) 1728 { 1729 struct net *net = sock_net(msg->sk); 1730 struct mptcp_pm_addr_entry *entry; 1731 struct pm_nl_pernet *pernet; 1732 int id = cb->args[0]; 1733 void *hdr; 1734 int i; 1735 1736 pernet = pm_nl_get_pernet(net); 1737 1738 spin_lock_bh(&pernet->lock); 1739 for (i = id; i < MPTCP_PM_MAX_ADDR_ID + 1; i++) { 1740 if (test_bit(i, pernet->id_bitmap)) { 1741 entry = __lookup_addr_by_id(pernet, i); 1742 if (!entry) 1743 break; 1744 1745 if (entry->addr.id <= id) 1746 continue; 1747 1748 hdr = genlmsg_put(msg, NETLINK_CB(cb->skb).portid, 1749 cb->nlh->nlmsg_seq, &mptcp_genl_family, 1750 NLM_F_MULTI, MPTCP_PM_CMD_GET_ADDR); 1751 if (!hdr) 1752 break; 1753 1754 if (mptcp_nl_fill_addr(msg, entry) < 0) { 1755 genlmsg_cancel(msg, hdr); 1756 break; 1757 } 1758 1759 id = entry->addr.id; 1760 genlmsg_end(msg, hdr); 1761 } 1762 } 1763 spin_unlock_bh(&pernet->lock); 1764 1765 cb->args[0] = id; 1766 return msg->len; 1767 } 1768 1769 static int parse_limit(struct genl_info *info, int id, unsigned int *limit) 1770 { 1771 struct nlattr *attr = info->attrs[id]; 1772 1773 if (!attr) 1774 return 0; 1775 1776 *limit = nla_get_u32(attr); 1777 if (*limit > MPTCP_PM_ADDR_MAX) { 1778 GENL_SET_ERR_MSG(info, "limit greater than maximum"); 1779 return -EINVAL; 1780 } 1781 return 0; 1782 } 1783 1784 static int 1785 mptcp_nl_cmd_set_limits(struct sk_buff *skb, struct genl_info *info) 1786 { 1787 struct pm_nl_pernet *pernet = genl_info_pm_nl(info); 1788 unsigned int rcv_addrs, subflows; 1789 int ret; 1790 1791 spin_lock_bh(&pernet->lock); 1792 rcv_addrs = pernet->add_addr_accept_max; 1793 ret = parse_limit(info, MPTCP_PM_ATTR_RCV_ADD_ADDRS, &rcv_addrs); 1794 if (ret) 1795 goto unlock; 1796 1797 subflows = pernet->subflows_max; 1798 ret = parse_limit(info, MPTCP_PM_ATTR_SUBFLOWS, &subflows); 1799 if (ret) 1800 goto unlock; 1801 1802 WRITE_ONCE(pernet->add_addr_accept_max, rcv_addrs); 1803 WRITE_ONCE(pernet->subflows_max, subflows); 1804 1805 unlock: 1806 spin_unlock_bh(&pernet->lock); 1807 return ret; 1808 } 1809 1810 static int 1811 mptcp_nl_cmd_get_limits(struct sk_buff *skb, struct genl_info *info) 1812 { 1813 struct pm_nl_pernet *pernet = genl_info_pm_nl(info); 1814 struct sk_buff *msg; 1815 void *reply; 1816 1817 msg = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL); 1818 if (!msg) 1819 return -ENOMEM; 1820 1821 reply = genlmsg_put_reply(msg, info, &mptcp_genl_family, 0, 1822 MPTCP_PM_CMD_GET_LIMITS); 1823 if (!reply) 1824 goto fail; 1825 1826 if (nla_put_u32(msg, MPTCP_PM_ATTR_RCV_ADD_ADDRS, 1827 READ_ONCE(pernet->add_addr_accept_max))) 1828 goto fail; 1829 1830 if (nla_put_u32(msg, MPTCP_PM_ATTR_SUBFLOWS, 1831 READ_ONCE(pernet->subflows_max))) 1832 goto fail; 1833 1834 genlmsg_end(msg, reply); 1835 return genlmsg_reply(msg, info); 1836 1837 fail: 1838 GENL_SET_ERR_MSG(info, "not enough space in Netlink message"); 1839 nlmsg_free(msg); 1840 return -EMSGSIZE; 1841 } 1842 1843 static void mptcp_pm_nl_fullmesh(struct mptcp_sock *msk, 1844 struct mptcp_addr_info *addr) 1845 { 1846 struct mptcp_rm_list list = { .nr = 0 }; 1847 1848 list.ids[list.nr++] = addr->id; 1849 1850 spin_lock_bh(&msk->pm.lock); 1851 mptcp_pm_nl_rm_subflow_received(msk, &list); 1852 mptcp_pm_create_subflow_or_signal_addr(msk); 1853 spin_unlock_bh(&msk->pm.lock); 1854 } 1855 1856 static int mptcp_nl_set_flags(struct net *net, 1857 struct mptcp_addr_info *addr, 1858 u8 bkup, u8 changed) 1859 { 1860 long s_slot = 0, s_num = 0; 1861 struct mptcp_sock *msk; 1862 int ret = -EINVAL; 1863 1864 while ((msk = mptcp_token_iter_next(net, &s_slot, &s_num)) != NULL) { 1865 struct sock *sk = (struct sock *)msk; 1866 1867 if (list_empty(&msk->conn_list) || mptcp_pm_is_userspace(msk)) 1868 goto next; 1869 1870 lock_sock(sk); 1871 if (changed & MPTCP_PM_ADDR_FLAG_BACKUP) 1872 ret = mptcp_pm_nl_mp_prio_send_ack(msk, addr, NULL, bkup); 1873 if (changed & MPTCP_PM_ADDR_FLAG_FULLMESH) 1874 mptcp_pm_nl_fullmesh(msk, addr); 1875 release_sock(sk); 1876 1877 next: 1878 sock_put(sk); 1879 cond_resched(); 1880 } 1881 1882 return ret; 1883 } 1884 1885 int mptcp_pm_nl_set_flags(struct net *net, struct mptcp_pm_addr_entry *addr, u8 bkup) 1886 { 1887 struct pm_nl_pernet *pernet = pm_nl_get_pernet(net); 1888 u8 changed, mask = MPTCP_PM_ADDR_FLAG_BACKUP | 1889 MPTCP_PM_ADDR_FLAG_FULLMESH; 1890 struct mptcp_pm_addr_entry *entry; 1891 u8 lookup_by_id = 0; 1892 1893 if (addr->addr.family == AF_UNSPEC) { 1894 lookup_by_id = 1; 1895 if (!addr->addr.id) 1896 return -EOPNOTSUPP; 1897 } 1898 1899 spin_lock_bh(&pernet->lock); 1900 entry = __lookup_addr(pernet, &addr->addr, lookup_by_id); 1901 if (!entry) { 1902 spin_unlock_bh(&pernet->lock); 1903 return -EINVAL; 1904 } 1905 if ((addr->flags & MPTCP_PM_ADDR_FLAG_FULLMESH) && 1906 (entry->flags & MPTCP_PM_ADDR_FLAG_SIGNAL)) { 1907 spin_unlock_bh(&pernet->lock); 1908 return -EINVAL; 1909 } 1910 1911 changed = (addr->flags ^ entry->flags) & mask; 1912 entry->flags = (entry->flags & ~mask) | (addr->flags & mask); 1913 *addr = *entry; 1914 spin_unlock_bh(&pernet->lock); 1915 1916 mptcp_nl_set_flags(net, &addr->addr, bkup, changed); 1917 return 0; 1918 } 1919 1920 static int mptcp_nl_cmd_set_flags(struct sk_buff *skb, struct genl_info *info) 1921 { 1922 struct mptcp_pm_addr_entry remote = { .addr = { .family = AF_UNSPEC }, }; 1923 struct mptcp_pm_addr_entry addr = { .addr = { .family = AF_UNSPEC }, }; 1924 struct nlattr *attr_rem = info->attrs[MPTCP_PM_ATTR_ADDR_REMOTE]; 1925 struct nlattr *token = info->attrs[MPTCP_PM_ATTR_TOKEN]; 1926 struct nlattr *attr = info->attrs[MPTCP_PM_ATTR_ADDR]; 1927 struct net *net = sock_net(skb->sk); 1928 u8 bkup = 0; 1929 int ret; 1930 1931 ret = mptcp_pm_parse_entry(attr, info, false, &addr); 1932 if (ret < 0) 1933 return ret; 1934 1935 if (attr_rem) { 1936 ret = mptcp_pm_parse_entry(attr_rem, info, false, &remote); 1937 if (ret < 0) 1938 return ret; 1939 } 1940 1941 if (addr.flags & MPTCP_PM_ADDR_FLAG_BACKUP) 1942 bkup = 1; 1943 1944 return mptcp_pm_set_flags(net, token, &addr, &remote, bkup); 1945 } 1946 1947 static void mptcp_nl_mcast_send(struct net *net, struct sk_buff *nlskb, gfp_t gfp) 1948 { 1949 genlmsg_multicast_netns(&mptcp_genl_family, net, 1950 nlskb, 0, MPTCP_PM_EV_GRP_OFFSET, gfp); 1951 } 1952 1953 bool mptcp_userspace_pm_active(const struct mptcp_sock *msk) 1954 { 1955 return genl_has_listeners(&mptcp_genl_family, 1956 sock_net((const struct sock *)msk), 1957 MPTCP_PM_EV_GRP_OFFSET); 1958 } 1959 1960 static int mptcp_event_add_subflow(struct sk_buff *skb, const struct sock *ssk) 1961 { 1962 const struct inet_sock *issk = inet_sk(ssk); 1963 const struct mptcp_subflow_context *sf; 1964 1965 if (nla_put_u16(skb, MPTCP_ATTR_FAMILY, ssk->sk_family)) 1966 return -EMSGSIZE; 1967 1968 switch (ssk->sk_family) { 1969 case AF_INET: 1970 if (nla_put_in_addr(skb, MPTCP_ATTR_SADDR4, issk->inet_saddr)) 1971 return -EMSGSIZE; 1972 if (nla_put_in_addr(skb, MPTCP_ATTR_DADDR4, issk->inet_daddr)) 1973 return -EMSGSIZE; 1974 break; 1975 #if IS_ENABLED(CONFIG_MPTCP_IPV6) 1976 case AF_INET6: { 1977 const struct ipv6_pinfo *np = inet6_sk(ssk); 1978 1979 if (nla_put_in6_addr(skb, MPTCP_ATTR_SADDR6, &np->saddr)) 1980 return -EMSGSIZE; 1981 if (nla_put_in6_addr(skb, MPTCP_ATTR_DADDR6, &ssk->sk_v6_daddr)) 1982 return -EMSGSIZE; 1983 break; 1984 } 1985 #endif 1986 default: 1987 WARN_ON_ONCE(1); 1988 return -EMSGSIZE; 1989 } 1990 1991 if (nla_put_be16(skb, MPTCP_ATTR_SPORT, issk->inet_sport)) 1992 return -EMSGSIZE; 1993 if (nla_put_be16(skb, MPTCP_ATTR_DPORT, issk->inet_dport)) 1994 return -EMSGSIZE; 1995 1996 sf = mptcp_subflow_ctx(ssk); 1997 if (WARN_ON_ONCE(!sf)) 1998 return -EINVAL; 1999 2000 if (nla_put_u8(skb, MPTCP_ATTR_LOC_ID, sf->local_id)) 2001 return -EMSGSIZE; 2002 2003 if (nla_put_u8(skb, MPTCP_ATTR_REM_ID, sf->remote_id)) 2004 return -EMSGSIZE; 2005 2006 return 0; 2007 } 2008 2009 static int mptcp_event_put_token_and_ssk(struct sk_buff *skb, 2010 const struct mptcp_sock *msk, 2011 const struct sock *ssk) 2012 { 2013 const struct sock *sk = (const struct sock *)msk; 2014 const struct mptcp_subflow_context *sf; 2015 u8 sk_err; 2016 2017 if (nla_put_u32(skb, MPTCP_ATTR_TOKEN, msk->token)) 2018 return -EMSGSIZE; 2019 2020 if (mptcp_event_add_subflow(skb, ssk)) 2021 return -EMSGSIZE; 2022 2023 sf = mptcp_subflow_ctx(ssk); 2024 if (WARN_ON_ONCE(!sf)) 2025 return -EINVAL; 2026 2027 if (nla_put_u8(skb, MPTCP_ATTR_BACKUP, sf->backup)) 2028 return -EMSGSIZE; 2029 2030 if (ssk->sk_bound_dev_if && 2031 nla_put_s32(skb, MPTCP_ATTR_IF_IDX, ssk->sk_bound_dev_if)) 2032 return -EMSGSIZE; 2033 2034 sk_err = READ_ONCE(ssk->sk_err); 2035 if (sk_err && sk->sk_state == TCP_ESTABLISHED && 2036 nla_put_u8(skb, MPTCP_ATTR_ERROR, sk_err)) 2037 return -EMSGSIZE; 2038 2039 return 0; 2040 } 2041 2042 static int mptcp_event_sub_established(struct sk_buff *skb, 2043 const struct mptcp_sock *msk, 2044 const struct sock *ssk) 2045 { 2046 return mptcp_event_put_token_and_ssk(skb, msk, ssk); 2047 } 2048 2049 static int mptcp_event_sub_closed(struct sk_buff *skb, 2050 const struct mptcp_sock *msk, 2051 const struct sock *ssk) 2052 { 2053 const struct mptcp_subflow_context *sf; 2054 2055 if (mptcp_event_put_token_and_ssk(skb, msk, ssk)) 2056 return -EMSGSIZE; 2057 2058 sf = mptcp_subflow_ctx(ssk); 2059 if (!sf->reset_seen) 2060 return 0; 2061 2062 if (nla_put_u32(skb, MPTCP_ATTR_RESET_REASON, sf->reset_reason)) 2063 return -EMSGSIZE; 2064 2065 if (nla_put_u32(skb, MPTCP_ATTR_RESET_FLAGS, sf->reset_transient)) 2066 return -EMSGSIZE; 2067 2068 return 0; 2069 } 2070 2071 static int mptcp_event_created(struct sk_buff *skb, 2072 const struct mptcp_sock *msk, 2073 const struct sock *ssk) 2074 { 2075 int err = nla_put_u32(skb, MPTCP_ATTR_TOKEN, msk->token); 2076 2077 if (err) 2078 return err; 2079 2080 if (nla_put_u8(skb, MPTCP_ATTR_SERVER_SIDE, READ_ONCE(msk->pm.server_side))) 2081 return -EMSGSIZE; 2082 2083 return mptcp_event_add_subflow(skb, ssk); 2084 } 2085 2086 void mptcp_event_addr_removed(const struct mptcp_sock *msk, uint8_t id) 2087 { 2088 struct net *net = sock_net((const struct sock *)msk); 2089 struct nlmsghdr *nlh; 2090 struct sk_buff *skb; 2091 2092 if (!genl_has_listeners(&mptcp_genl_family, net, MPTCP_PM_EV_GRP_OFFSET)) 2093 return; 2094 2095 skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_ATOMIC); 2096 if (!skb) 2097 return; 2098 2099 nlh = genlmsg_put(skb, 0, 0, &mptcp_genl_family, 0, MPTCP_EVENT_REMOVED); 2100 if (!nlh) 2101 goto nla_put_failure; 2102 2103 if (nla_put_u32(skb, MPTCP_ATTR_TOKEN, msk->token)) 2104 goto nla_put_failure; 2105 2106 if (nla_put_u8(skb, MPTCP_ATTR_REM_ID, id)) 2107 goto nla_put_failure; 2108 2109 genlmsg_end(skb, nlh); 2110 mptcp_nl_mcast_send(net, skb, GFP_ATOMIC); 2111 return; 2112 2113 nla_put_failure: 2114 nlmsg_free(skb); 2115 } 2116 2117 void mptcp_event_addr_announced(const struct sock *ssk, 2118 const struct mptcp_addr_info *info) 2119 { 2120 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk); 2121 struct mptcp_sock *msk = mptcp_sk(subflow->conn); 2122 struct net *net = sock_net(ssk); 2123 struct nlmsghdr *nlh; 2124 struct sk_buff *skb; 2125 2126 if (!genl_has_listeners(&mptcp_genl_family, net, MPTCP_PM_EV_GRP_OFFSET)) 2127 return; 2128 2129 skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_ATOMIC); 2130 if (!skb) 2131 return; 2132 2133 nlh = genlmsg_put(skb, 0, 0, &mptcp_genl_family, 0, 2134 MPTCP_EVENT_ANNOUNCED); 2135 if (!nlh) 2136 goto nla_put_failure; 2137 2138 if (nla_put_u32(skb, MPTCP_ATTR_TOKEN, msk->token)) 2139 goto nla_put_failure; 2140 2141 if (nla_put_u8(skb, MPTCP_ATTR_REM_ID, info->id)) 2142 goto nla_put_failure; 2143 2144 if (nla_put_be16(skb, MPTCP_ATTR_DPORT, 2145 info->port == 0 ? 2146 inet_sk(ssk)->inet_dport : 2147 info->port)) 2148 goto nla_put_failure; 2149 2150 switch (info->family) { 2151 case AF_INET: 2152 if (nla_put_in_addr(skb, MPTCP_ATTR_DADDR4, info->addr.s_addr)) 2153 goto nla_put_failure; 2154 break; 2155 #if IS_ENABLED(CONFIG_MPTCP_IPV6) 2156 case AF_INET6: 2157 if (nla_put_in6_addr(skb, MPTCP_ATTR_DADDR6, &info->addr6)) 2158 goto nla_put_failure; 2159 break; 2160 #endif 2161 default: 2162 WARN_ON_ONCE(1); 2163 goto nla_put_failure; 2164 } 2165 2166 genlmsg_end(skb, nlh); 2167 mptcp_nl_mcast_send(net, skb, GFP_ATOMIC); 2168 return; 2169 2170 nla_put_failure: 2171 nlmsg_free(skb); 2172 } 2173 2174 void mptcp_event_pm_listener(const struct sock *ssk, 2175 enum mptcp_event_type event) 2176 { 2177 const struct inet_sock *issk = inet_sk(ssk); 2178 struct net *net = sock_net(ssk); 2179 struct nlmsghdr *nlh; 2180 struct sk_buff *skb; 2181 2182 if (!genl_has_listeners(&mptcp_genl_family, net, MPTCP_PM_EV_GRP_OFFSET)) 2183 return; 2184 2185 skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL); 2186 if (!skb) 2187 return; 2188 2189 nlh = genlmsg_put(skb, 0, 0, &mptcp_genl_family, 0, event); 2190 if (!nlh) 2191 goto nla_put_failure; 2192 2193 if (nla_put_u16(skb, MPTCP_ATTR_FAMILY, ssk->sk_family)) 2194 goto nla_put_failure; 2195 2196 if (nla_put_be16(skb, MPTCP_ATTR_SPORT, issk->inet_sport)) 2197 goto nla_put_failure; 2198 2199 switch (ssk->sk_family) { 2200 case AF_INET: 2201 if (nla_put_in_addr(skb, MPTCP_ATTR_SADDR4, issk->inet_saddr)) 2202 goto nla_put_failure; 2203 break; 2204 #if IS_ENABLED(CONFIG_MPTCP_IPV6) 2205 case AF_INET6: { 2206 const struct ipv6_pinfo *np = inet6_sk(ssk); 2207 2208 if (nla_put_in6_addr(skb, MPTCP_ATTR_SADDR6, &np->saddr)) 2209 goto nla_put_failure; 2210 break; 2211 } 2212 #endif 2213 default: 2214 WARN_ON_ONCE(1); 2215 goto nla_put_failure; 2216 } 2217 2218 genlmsg_end(skb, nlh); 2219 mptcp_nl_mcast_send(net, skb, GFP_KERNEL); 2220 return; 2221 2222 nla_put_failure: 2223 nlmsg_free(skb); 2224 } 2225 2226 void mptcp_event(enum mptcp_event_type type, const struct mptcp_sock *msk, 2227 const struct sock *ssk, gfp_t gfp) 2228 { 2229 struct net *net = sock_net((const struct sock *)msk); 2230 struct nlmsghdr *nlh; 2231 struct sk_buff *skb; 2232 2233 if (!genl_has_listeners(&mptcp_genl_family, net, MPTCP_PM_EV_GRP_OFFSET)) 2234 return; 2235 2236 skb = nlmsg_new(NLMSG_DEFAULT_SIZE, gfp); 2237 if (!skb) 2238 return; 2239 2240 nlh = genlmsg_put(skb, 0, 0, &mptcp_genl_family, 0, type); 2241 if (!nlh) 2242 goto nla_put_failure; 2243 2244 switch (type) { 2245 case MPTCP_EVENT_UNSPEC: 2246 WARN_ON_ONCE(1); 2247 break; 2248 case MPTCP_EVENT_CREATED: 2249 case MPTCP_EVENT_ESTABLISHED: 2250 if (mptcp_event_created(skb, msk, ssk) < 0) 2251 goto nla_put_failure; 2252 break; 2253 case MPTCP_EVENT_CLOSED: 2254 if (nla_put_u32(skb, MPTCP_ATTR_TOKEN, msk->token) < 0) 2255 goto nla_put_failure; 2256 break; 2257 case MPTCP_EVENT_ANNOUNCED: 2258 case MPTCP_EVENT_REMOVED: 2259 /* call mptcp_event_addr_announced()/removed instead */ 2260 WARN_ON_ONCE(1); 2261 break; 2262 case MPTCP_EVENT_SUB_ESTABLISHED: 2263 case MPTCP_EVENT_SUB_PRIORITY: 2264 if (mptcp_event_sub_established(skb, msk, ssk) < 0) 2265 goto nla_put_failure; 2266 break; 2267 case MPTCP_EVENT_SUB_CLOSED: 2268 if (mptcp_event_sub_closed(skb, msk, ssk) < 0) 2269 goto nla_put_failure; 2270 break; 2271 case MPTCP_EVENT_LISTENER_CREATED: 2272 case MPTCP_EVENT_LISTENER_CLOSED: 2273 break; 2274 } 2275 2276 genlmsg_end(skb, nlh); 2277 mptcp_nl_mcast_send(net, skb, gfp); 2278 return; 2279 2280 nla_put_failure: 2281 nlmsg_free(skb); 2282 } 2283 2284 static const struct genl_small_ops mptcp_pm_ops[] = { 2285 { 2286 .cmd = MPTCP_PM_CMD_ADD_ADDR, 2287 .doit = mptcp_nl_cmd_add_addr, 2288 .flags = GENL_UNS_ADMIN_PERM, 2289 }, 2290 { 2291 .cmd = MPTCP_PM_CMD_DEL_ADDR, 2292 .doit = mptcp_nl_cmd_del_addr, 2293 .flags = GENL_UNS_ADMIN_PERM, 2294 }, 2295 { 2296 .cmd = MPTCP_PM_CMD_FLUSH_ADDRS, 2297 .doit = mptcp_nl_cmd_flush_addrs, 2298 .flags = GENL_UNS_ADMIN_PERM, 2299 }, 2300 { 2301 .cmd = MPTCP_PM_CMD_GET_ADDR, 2302 .doit = mptcp_nl_cmd_get_addr, 2303 .dumpit = mptcp_nl_cmd_dump_addrs, 2304 }, 2305 { 2306 .cmd = MPTCP_PM_CMD_SET_LIMITS, 2307 .doit = mptcp_nl_cmd_set_limits, 2308 .flags = GENL_UNS_ADMIN_PERM, 2309 }, 2310 { 2311 .cmd = MPTCP_PM_CMD_GET_LIMITS, 2312 .doit = mptcp_nl_cmd_get_limits, 2313 }, 2314 { 2315 .cmd = MPTCP_PM_CMD_SET_FLAGS, 2316 .doit = mptcp_nl_cmd_set_flags, 2317 .flags = GENL_UNS_ADMIN_PERM, 2318 }, 2319 { 2320 .cmd = MPTCP_PM_CMD_ANNOUNCE, 2321 .doit = mptcp_nl_cmd_announce, 2322 .flags = GENL_UNS_ADMIN_PERM, 2323 }, 2324 { 2325 .cmd = MPTCP_PM_CMD_REMOVE, 2326 .doit = mptcp_nl_cmd_remove, 2327 .flags = GENL_UNS_ADMIN_PERM, 2328 }, 2329 { 2330 .cmd = MPTCP_PM_CMD_SUBFLOW_CREATE, 2331 .doit = mptcp_nl_cmd_sf_create, 2332 .flags = GENL_UNS_ADMIN_PERM, 2333 }, 2334 { 2335 .cmd = MPTCP_PM_CMD_SUBFLOW_DESTROY, 2336 .doit = mptcp_nl_cmd_sf_destroy, 2337 .flags = GENL_UNS_ADMIN_PERM, 2338 }, 2339 }; 2340 2341 static struct genl_family mptcp_genl_family __ro_after_init = { 2342 .name = MPTCP_PM_NAME, 2343 .version = MPTCP_PM_VER, 2344 .maxattr = MPTCP_PM_ATTR_MAX, 2345 .policy = mptcp_pm_policy, 2346 .netnsok = true, 2347 .module = THIS_MODULE, 2348 .small_ops = mptcp_pm_ops, 2349 .n_small_ops = ARRAY_SIZE(mptcp_pm_ops), 2350 .resv_start_op = MPTCP_PM_CMD_SUBFLOW_DESTROY + 1, 2351 .mcgrps = mptcp_pm_mcgrps, 2352 .n_mcgrps = ARRAY_SIZE(mptcp_pm_mcgrps), 2353 }; 2354 2355 static int __net_init pm_nl_init_net(struct net *net) 2356 { 2357 struct pm_nl_pernet *pernet = pm_nl_get_pernet(net); 2358 2359 INIT_LIST_HEAD_RCU(&pernet->local_addr_list); 2360 2361 /* Cit. 2 subflows ought to be enough for anybody. */ 2362 pernet->subflows_max = 2; 2363 pernet->next_id = 1; 2364 pernet->stale_loss_cnt = 4; 2365 spin_lock_init(&pernet->lock); 2366 2367 /* No need to initialize other pernet fields, the struct is zeroed at 2368 * allocation time. 2369 */ 2370 2371 return 0; 2372 } 2373 2374 static void __net_exit pm_nl_exit_net(struct list_head *net_list) 2375 { 2376 struct net *net; 2377 2378 list_for_each_entry(net, net_list, exit_list) { 2379 struct pm_nl_pernet *pernet = pm_nl_get_pernet(net); 2380 2381 /* net is removed from namespace list, can't race with 2382 * other modifiers, also netns core already waited for a 2383 * RCU grace period. 2384 */ 2385 __flush_addrs(&pernet->local_addr_list); 2386 } 2387 } 2388 2389 static struct pernet_operations mptcp_pm_pernet_ops = { 2390 .init = pm_nl_init_net, 2391 .exit_batch = pm_nl_exit_net, 2392 .id = &pm_nl_pernet_id, 2393 .size = sizeof(struct pm_nl_pernet), 2394 }; 2395 2396 void __init mptcp_pm_nl_init(void) 2397 { 2398 if (register_pernet_subsys(&mptcp_pm_pernet_ops) < 0) 2399 panic("Failed to register MPTCP PM pernet subsystem.\n"); 2400 2401 if (genl_register_family(&mptcp_genl_family)) 2402 panic("Failed to register MPTCP PM netlink family\n"); 2403 } 2404