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/inet_common.h> 13 #include <net/netns/generic.h> 14 #include <net/mptcp.h> 15 #include <net/genetlink.h> 16 #include <uapi/linux/mptcp.h> 17 18 #include "protocol.h" 19 #include "mib.h" 20 21 /* forward declaration */ 22 static struct genl_family mptcp_genl_family; 23 24 static int pm_nl_pernet_id; 25 26 struct mptcp_pm_add_entry { 27 struct list_head list; 28 struct mptcp_addr_info addr; 29 u8 retrans_times; 30 struct timer_list add_timer; 31 struct mptcp_sock *sock; 32 }; 33 34 struct pm_nl_pernet { 35 /* protects pernet updates */ 36 spinlock_t lock; 37 struct list_head local_addr_list; 38 unsigned int addrs; 39 unsigned int stale_loss_cnt; 40 unsigned int add_addr_signal_max; 41 unsigned int add_addr_accept_max; 42 unsigned int local_addr_max; 43 unsigned int subflows_max; 44 unsigned int next_id; 45 DECLARE_BITMAP(id_bitmap, MPTCP_PM_MAX_ADDR_ID + 1); 46 }; 47 48 #define MPTCP_PM_ADDR_MAX 8 49 #define ADD_ADDR_RETRANS_MAX 3 50 51 static struct pm_nl_pernet *pm_nl_get_pernet(const struct net *net) 52 { 53 return net_generic(net, pm_nl_pernet_id); 54 } 55 56 static struct pm_nl_pernet * 57 pm_nl_get_pernet_from_msk(const struct mptcp_sock *msk) 58 { 59 return pm_nl_get_pernet(sock_net((struct sock *)msk)); 60 } 61 62 bool mptcp_addresses_equal(const struct mptcp_addr_info *a, 63 const struct mptcp_addr_info *b, bool use_port) 64 { 65 bool addr_equals = false; 66 67 if (a->family == b->family) { 68 if (a->family == AF_INET) 69 addr_equals = a->addr.s_addr == b->addr.s_addr; 70 #if IS_ENABLED(CONFIG_MPTCP_IPV6) 71 else 72 addr_equals = !ipv6_addr_cmp(&a->addr6, &b->addr6); 73 } else if (a->family == AF_INET) { 74 if (ipv6_addr_v4mapped(&b->addr6)) 75 addr_equals = a->addr.s_addr == b->addr6.s6_addr32[3]; 76 } else if (b->family == AF_INET) { 77 if (ipv6_addr_v4mapped(&a->addr6)) 78 addr_equals = a->addr6.s6_addr32[3] == b->addr.s_addr; 79 #endif 80 } 81 82 if (!addr_equals) 83 return false; 84 if (!use_port) 85 return true; 86 87 return a->port == b->port; 88 } 89 90 void mptcp_local_address(const struct sock_common *skc, struct mptcp_addr_info *addr) 91 { 92 addr->family = skc->skc_family; 93 addr->port = htons(skc->skc_num); 94 if (addr->family == AF_INET) 95 addr->addr.s_addr = skc->skc_rcv_saddr; 96 #if IS_ENABLED(CONFIG_MPTCP_IPV6) 97 else if (addr->family == AF_INET6) 98 addr->addr6 = skc->skc_v6_rcv_saddr; 99 #endif 100 } 101 102 static void remote_address(const struct sock_common *skc, 103 struct mptcp_addr_info *addr) 104 { 105 addr->family = skc->skc_family; 106 addr->port = skc->skc_dport; 107 if (addr->family == AF_INET) 108 addr->addr.s_addr = skc->skc_daddr; 109 #if IS_ENABLED(CONFIG_MPTCP_IPV6) 110 else if (addr->family == AF_INET6) 111 addr->addr6 = skc->skc_v6_daddr; 112 #endif 113 } 114 115 static bool lookup_subflow_by_saddr(const struct list_head *list, 116 const struct mptcp_addr_info *saddr) 117 { 118 struct mptcp_subflow_context *subflow; 119 struct mptcp_addr_info cur; 120 struct sock_common *skc; 121 122 list_for_each_entry(subflow, list, node) { 123 skc = (struct sock_common *)mptcp_subflow_tcp_sock(subflow); 124 125 mptcp_local_address(skc, &cur); 126 if (mptcp_addresses_equal(&cur, saddr, saddr->port)) 127 return true; 128 } 129 130 return false; 131 } 132 133 static bool lookup_subflow_by_daddr(const struct list_head *list, 134 const struct mptcp_addr_info *daddr) 135 { 136 struct mptcp_subflow_context *subflow; 137 struct mptcp_addr_info cur; 138 struct sock_common *skc; 139 140 list_for_each_entry(subflow, list, node) { 141 skc = (struct sock_common *)mptcp_subflow_tcp_sock(subflow); 142 143 remote_address(skc, &cur); 144 if (mptcp_addresses_equal(&cur, daddr, daddr->port)) 145 return true; 146 } 147 148 return false; 149 } 150 151 static struct mptcp_pm_addr_entry * 152 select_local_address(const struct pm_nl_pernet *pernet, 153 const struct mptcp_sock *msk) 154 { 155 struct mptcp_pm_addr_entry *entry, *ret = NULL; 156 157 msk_owned_by_me(msk); 158 159 rcu_read_lock(); 160 list_for_each_entry_rcu(entry, &pernet->local_addr_list, list) { 161 if (!(entry->flags & MPTCP_PM_ADDR_FLAG_SUBFLOW)) 162 continue; 163 164 if (!test_bit(entry->addr.id, msk->pm.id_avail_bitmap)) 165 continue; 166 167 ret = entry; 168 break; 169 } 170 rcu_read_unlock(); 171 return ret; 172 } 173 174 static struct mptcp_pm_addr_entry * 175 select_signal_address(struct pm_nl_pernet *pernet, const struct mptcp_sock *msk) 176 { 177 struct mptcp_pm_addr_entry *entry, *ret = NULL; 178 179 rcu_read_lock(); 180 /* do not keep any additional per socket state, just signal 181 * the address list in order. 182 * Note: removal from the local address list during the msk life-cycle 183 * can lead to additional addresses not being announced. 184 */ 185 list_for_each_entry_rcu(entry, &pernet->local_addr_list, list) { 186 if (!test_bit(entry->addr.id, msk->pm.id_avail_bitmap)) 187 continue; 188 189 if (!(entry->flags & MPTCP_PM_ADDR_FLAG_SIGNAL)) 190 continue; 191 192 ret = entry; 193 break; 194 } 195 rcu_read_unlock(); 196 return ret; 197 } 198 199 unsigned int mptcp_pm_get_add_addr_signal_max(const struct mptcp_sock *msk) 200 { 201 const struct pm_nl_pernet *pernet = pm_nl_get_pernet_from_msk(msk); 202 203 return READ_ONCE(pernet->add_addr_signal_max); 204 } 205 EXPORT_SYMBOL_GPL(mptcp_pm_get_add_addr_signal_max); 206 207 unsigned int mptcp_pm_get_add_addr_accept_max(const struct mptcp_sock *msk) 208 { 209 struct pm_nl_pernet *pernet = pm_nl_get_pernet_from_msk(msk); 210 211 return READ_ONCE(pernet->add_addr_accept_max); 212 } 213 EXPORT_SYMBOL_GPL(mptcp_pm_get_add_addr_accept_max); 214 215 unsigned int mptcp_pm_get_subflows_max(const struct mptcp_sock *msk) 216 { 217 struct pm_nl_pernet *pernet = pm_nl_get_pernet_from_msk(msk); 218 219 return READ_ONCE(pernet->subflows_max); 220 } 221 EXPORT_SYMBOL_GPL(mptcp_pm_get_subflows_max); 222 223 unsigned int mptcp_pm_get_local_addr_max(const struct mptcp_sock *msk) 224 { 225 struct pm_nl_pernet *pernet = pm_nl_get_pernet_from_msk(msk); 226 227 return READ_ONCE(pernet->local_addr_max); 228 } 229 EXPORT_SYMBOL_GPL(mptcp_pm_get_local_addr_max); 230 231 bool mptcp_pm_nl_check_work_pending(struct mptcp_sock *msk) 232 { 233 struct pm_nl_pernet *pernet = pm_nl_get_pernet_from_msk(msk); 234 235 if (msk->pm.subflows == mptcp_pm_get_subflows_max(msk) || 236 (find_next_and_bit(pernet->id_bitmap, msk->pm.id_avail_bitmap, 237 MPTCP_PM_MAX_ADDR_ID + 1, 0) == MPTCP_PM_MAX_ADDR_ID + 1)) { 238 WRITE_ONCE(msk->pm.work_pending, false); 239 return false; 240 } 241 return true; 242 } 243 244 struct mptcp_pm_add_entry * 245 mptcp_lookup_anno_list_by_saddr(const struct mptcp_sock *msk, 246 const struct mptcp_addr_info *addr) 247 { 248 struct mptcp_pm_add_entry *entry; 249 250 lockdep_assert_held(&msk->pm.lock); 251 252 list_for_each_entry(entry, &msk->pm.anno_list, list) { 253 if (mptcp_addresses_equal(&entry->addr, addr, true)) 254 return entry; 255 } 256 257 return NULL; 258 } 259 260 bool mptcp_pm_sport_in_anno_list(struct mptcp_sock *msk, const struct sock *sk) 261 { 262 struct mptcp_pm_add_entry *entry; 263 struct mptcp_addr_info saddr; 264 bool ret = false; 265 266 mptcp_local_address((struct sock_common *)sk, &saddr); 267 268 spin_lock_bh(&msk->pm.lock); 269 list_for_each_entry(entry, &msk->pm.anno_list, list) { 270 if (mptcp_addresses_equal(&entry->addr, &saddr, true)) { 271 ret = true; 272 goto out; 273 } 274 } 275 276 out: 277 spin_unlock_bh(&msk->pm.lock); 278 return ret; 279 } 280 281 static void mptcp_pm_add_timer(struct timer_list *timer) 282 { 283 struct mptcp_pm_add_entry *entry = from_timer(entry, timer, add_timer); 284 struct mptcp_sock *msk = entry->sock; 285 struct sock *sk = (struct sock *)msk; 286 287 pr_debug("msk=%p", msk); 288 289 if (!msk) 290 return; 291 292 if (inet_sk_state_load(sk) == TCP_CLOSE) 293 return; 294 295 if (!entry->addr.id) 296 return; 297 298 if (mptcp_pm_should_add_signal_addr(msk)) { 299 sk_reset_timer(sk, timer, jiffies + TCP_RTO_MAX / 8); 300 goto out; 301 } 302 303 spin_lock_bh(&msk->pm.lock); 304 305 if (!mptcp_pm_should_add_signal_addr(msk)) { 306 pr_debug("retransmit ADD_ADDR id=%d", entry->addr.id); 307 mptcp_pm_announce_addr(msk, &entry->addr, false); 308 mptcp_pm_add_addr_send_ack(msk); 309 entry->retrans_times++; 310 } 311 312 if (entry->retrans_times < ADD_ADDR_RETRANS_MAX) 313 sk_reset_timer(sk, timer, 314 jiffies + mptcp_get_add_addr_timeout(sock_net(sk))); 315 316 spin_unlock_bh(&msk->pm.lock); 317 318 if (entry->retrans_times == ADD_ADDR_RETRANS_MAX) 319 mptcp_pm_subflow_established(msk); 320 321 out: 322 __sock_put(sk); 323 } 324 325 struct mptcp_pm_add_entry * 326 mptcp_pm_del_add_timer(struct mptcp_sock *msk, 327 const struct mptcp_addr_info *addr, bool check_id) 328 { 329 struct mptcp_pm_add_entry *entry; 330 struct sock *sk = (struct sock *)msk; 331 332 spin_lock_bh(&msk->pm.lock); 333 entry = mptcp_lookup_anno_list_by_saddr(msk, addr); 334 if (entry && (!check_id || entry->addr.id == addr->id)) 335 entry->retrans_times = ADD_ADDR_RETRANS_MAX; 336 spin_unlock_bh(&msk->pm.lock); 337 338 if (entry && (!check_id || entry->addr.id == addr->id)) 339 sk_stop_timer_sync(sk, &entry->add_timer); 340 341 return entry; 342 } 343 344 bool mptcp_pm_alloc_anno_list(struct mptcp_sock *msk, 345 const struct mptcp_addr_info *addr) 346 { 347 struct mptcp_pm_add_entry *add_entry = NULL; 348 struct sock *sk = (struct sock *)msk; 349 struct net *net = sock_net(sk); 350 351 lockdep_assert_held(&msk->pm.lock); 352 353 add_entry = mptcp_lookup_anno_list_by_saddr(msk, addr); 354 355 if (add_entry) { 356 if (mptcp_pm_is_kernel(msk)) 357 return false; 358 359 sk_reset_timer(sk, &add_entry->add_timer, 360 jiffies + mptcp_get_add_addr_timeout(net)); 361 return true; 362 } 363 364 add_entry = kmalloc(sizeof(*add_entry), GFP_ATOMIC); 365 if (!add_entry) 366 return false; 367 368 list_add(&add_entry->list, &msk->pm.anno_list); 369 370 add_entry->addr = *addr; 371 add_entry->sock = msk; 372 add_entry->retrans_times = 0; 373 374 timer_setup(&add_entry->add_timer, mptcp_pm_add_timer, 0); 375 sk_reset_timer(sk, &add_entry->add_timer, 376 jiffies + mptcp_get_add_addr_timeout(net)); 377 378 return true; 379 } 380 381 void mptcp_pm_free_anno_list(struct mptcp_sock *msk) 382 { 383 struct mptcp_pm_add_entry *entry, *tmp; 384 struct sock *sk = (struct sock *)msk; 385 LIST_HEAD(free_list); 386 387 pr_debug("msk=%p", msk); 388 389 spin_lock_bh(&msk->pm.lock); 390 list_splice_init(&msk->pm.anno_list, &free_list); 391 spin_unlock_bh(&msk->pm.lock); 392 393 list_for_each_entry_safe(entry, tmp, &free_list, list) { 394 sk_stop_timer_sync(sk, &entry->add_timer); 395 kfree(entry); 396 } 397 } 398 399 static bool lookup_address_in_vec(const struct mptcp_addr_info *addrs, unsigned int nr, 400 const struct mptcp_addr_info *addr) 401 { 402 int i; 403 404 for (i = 0; i < nr; i++) { 405 if (addrs[i].id == addr->id) 406 return true; 407 } 408 409 return false; 410 } 411 412 /* Fill all the remote addresses into the array addrs[], 413 * and return the array size. 414 */ 415 static unsigned int fill_remote_addresses_vec(struct mptcp_sock *msk, 416 struct mptcp_addr_info *local, 417 bool fullmesh, 418 struct mptcp_addr_info *addrs) 419 { 420 bool deny_id0 = READ_ONCE(msk->pm.remote_deny_join_id0); 421 struct sock *sk = (struct sock *)msk, *ssk; 422 struct mptcp_subflow_context *subflow; 423 struct mptcp_addr_info remote = { 0 }; 424 unsigned int subflows_max; 425 int i = 0; 426 427 subflows_max = mptcp_pm_get_subflows_max(msk); 428 remote_address((struct sock_common *)sk, &remote); 429 430 /* Non-fullmesh endpoint, fill in the single entry 431 * corresponding to the primary MPC subflow remote address 432 */ 433 if (!fullmesh) { 434 if (deny_id0) 435 return 0; 436 437 if (!mptcp_pm_addr_families_match(sk, local, &remote)) 438 return 0; 439 440 msk->pm.subflows++; 441 addrs[i++] = remote; 442 } else { 443 mptcp_for_each_subflow(msk, subflow) { 444 ssk = mptcp_subflow_tcp_sock(subflow); 445 remote_address((struct sock_common *)ssk, &addrs[i]); 446 addrs[i].id = subflow->remote_id; 447 if (deny_id0 && !addrs[i].id) 448 continue; 449 450 if (!mptcp_pm_addr_families_match(sk, local, &addrs[i])) 451 continue; 452 453 if (!lookup_address_in_vec(addrs, i, &addrs[i]) && 454 msk->pm.subflows < subflows_max) { 455 msk->pm.subflows++; 456 i++; 457 } 458 } 459 } 460 461 return i; 462 } 463 464 static void __mptcp_pm_send_ack(struct mptcp_sock *msk, struct mptcp_subflow_context *subflow, 465 bool prio, bool backup) 466 { 467 struct sock *ssk = mptcp_subflow_tcp_sock(subflow); 468 bool slow; 469 470 pr_debug("send ack for %s", 471 prio ? "mp_prio" : (mptcp_pm_should_add_signal(msk) ? "add_addr" : "rm_addr")); 472 473 slow = lock_sock_fast(ssk); 474 if (prio) { 475 subflow->send_mp_prio = 1; 476 subflow->backup = backup; 477 subflow->request_bkup = backup; 478 } 479 480 __mptcp_subflow_send_ack(ssk); 481 unlock_sock_fast(ssk, slow); 482 } 483 484 static void mptcp_pm_send_ack(struct mptcp_sock *msk, struct mptcp_subflow_context *subflow, 485 bool prio, bool backup) 486 { 487 spin_unlock_bh(&msk->pm.lock); 488 __mptcp_pm_send_ack(msk, subflow, prio, backup); 489 spin_lock_bh(&msk->pm.lock); 490 } 491 492 static struct mptcp_pm_addr_entry * 493 __lookup_addr_by_id(struct pm_nl_pernet *pernet, unsigned int id) 494 { 495 struct mptcp_pm_addr_entry *entry; 496 497 list_for_each_entry(entry, &pernet->local_addr_list, list) { 498 if (entry->addr.id == id) 499 return entry; 500 } 501 return NULL; 502 } 503 504 static struct mptcp_pm_addr_entry * 505 __lookup_addr(struct pm_nl_pernet *pernet, const struct mptcp_addr_info *info, 506 bool lookup_by_id) 507 { 508 struct mptcp_pm_addr_entry *entry; 509 510 list_for_each_entry(entry, &pernet->local_addr_list, list) { 511 if ((!lookup_by_id && 512 mptcp_addresses_equal(&entry->addr, info, entry->addr.port)) || 513 (lookup_by_id && entry->addr.id == info->id)) 514 return entry; 515 } 516 return NULL; 517 } 518 519 static void mptcp_pm_create_subflow_or_signal_addr(struct mptcp_sock *msk) 520 { 521 struct sock *sk = (struct sock *)msk; 522 struct mptcp_pm_addr_entry *local; 523 unsigned int add_addr_signal_max; 524 unsigned int local_addr_max; 525 struct pm_nl_pernet *pernet; 526 unsigned int subflows_max; 527 528 pernet = pm_nl_get_pernet(sock_net(sk)); 529 530 add_addr_signal_max = mptcp_pm_get_add_addr_signal_max(msk); 531 local_addr_max = mptcp_pm_get_local_addr_max(msk); 532 subflows_max = mptcp_pm_get_subflows_max(msk); 533 534 /* do lazy endpoint usage accounting for the MPC subflows */ 535 if (unlikely(!(msk->pm.status & BIT(MPTCP_PM_MPC_ENDPOINT_ACCOUNTED))) && msk->first) { 536 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(msk->first); 537 struct mptcp_pm_addr_entry *entry; 538 struct mptcp_addr_info mpc_addr; 539 bool backup = false; 540 541 mptcp_local_address((struct sock_common *)msk->first, &mpc_addr); 542 rcu_read_lock(); 543 entry = __lookup_addr(pernet, &mpc_addr, false); 544 if (entry) { 545 __clear_bit(entry->addr.id, msk->pm.id_avail_bitmap); 546 msk->mpc_endpoint_id = entry->addr.id; 547 backup = !!(entry->flags & MPTCP_PM_ADDR_FLAG_BACKUP); 548 } 549 rcu_read_unlock(); 550 551 if (backup) 552 mptcp_pm_send_ack(msk, subflow, true, backup); 553 554 msk->pm.status |= BIT(MPTCP_PM_MPC_ENDPOINT_ACCOUNTED); 555 } 556 557 pr_debug("local %d:%d signal %d:%d subflows %d:%d\n", 558 msk->pm.local_addr_used, local_addr_max, 559 msk->pm.add_addr_signaled, add_addr_signal_max, 560 msk->pm.subflows, subflows_max); 561 562 /* check first for announce */ 563 if (msk->pm.add_addr_signaled < add_addr_signal_max) { 564 local = select_signal_address(pernet, msk); 565 566 /* due to racing events on both ends we can reach here while 567 * previous add address is still running: if we invoke now 568 * mptcp_pm_announce_addr(), that will fail and the 569 * corresponding id will be marked as used. 570 * Instead let the PM machinery reschedule us when the 571 * current address announce will be completed. 572 */ 573 if (msk->pm.addr_signal & BIT(MPTCP_ADD_ADDR_SIGNAL)) 574 return; 575 576 if (local) { 577 if (mptcp_pm_alloc_anno_list(msk, &local->addr)) { 578 __clear_bit(local->addr.id, msk->pm.id_avail_bitmap); 579 msk->pm.add_addr_signaled++; 580 mptcp_pm_announce_addr(msk, &local->addr, false); 581 mptcp_pm_nl_addr_send_ack(msk); 582 } 583 } 584 } 585 586 /* check if should create a new subflow */ 587 while (msk->pm.local_addr_used < local_addr_max && 588 msk->pm.subflows < subflows_max) { 589 struct mptcp_addr_info addrs[MPTCP_PM_ADDR_MAX]; 590 bool fullmesh; 591 int i, nr; 592 593 local = select_local_address(pernet, msk); 594 if (!local) 595 break; 596 597 fullmesh = !!(local->flags & MPTCP_PM_ADDR_FLAG_FULLMESH); 598 599 msk->pm.local_addr_used++; 600 __clear_bit(local->addr.id, msk->pm.id_avail_bitmap); 601 nr = fill_remote_addresses_vec(msk, &local->addr, fullmesh, addrs); 602 if (nr == 0) 603 continue; 604 605 spin_unlock_bh(&msk->pm.lock); 606 for (i = 0; i < nr; i++) 607 __mptcp_subflow_connect(sk, &local->addr, &addrs[i]); 608 spin_lock_bh(&msk->pm.lock); 609 } 610 mptcp_pm_nl_check_work_pending(msk); 611 } 612 613 static void mptcp_pm_nl_fully_established(struct mptcp_sock *msk) 614 { 615 mptcp_pm_create_subflow_or_signal_addr(msk); 616 } 617 618 static void mptcp_pm_nl_subflow_established(struct mptcp_sock *msk) 619 { 620 mptcp_pm_create_subflow_or_signal_addr(msk); 621 } 622 623 /* Fill all the local addresses into the array addrs[], 624 * and return the array size. 625 */ 626 static unsigned int fill_local_addresses_vec(struct mptcp_sock *msk, 627 struct mptcp_addr_info *remote, 628 struct mptcp_addr_info *addrs) 629 { 630 struct sock *sk = (struct sock *)msk; 631 struct mptcp_pm_addr_entry *entry; 632 struct pm_nl_pernet *pernet; 633 unsigned int subflows_max; 634 int i = 0; 635 636 pernet = pm_nl_get_pernet_from_msk(msk); 637 subflows_max = mptcp_pm_get_subflows_max(msk); 638 639 rcu_read_lock(); 640 list_for_each_entry_rcu(entry, &pernet->local_addr_list, list) { 641 if (!(entry->flags & MPTCP_PM_ADDR_FLAG_FULLMESH)) 642 continue; 643 644 if (!mptcp_pm_addr_families_match(sk, &entry->addr, remote)) 645 continue; 646 647 if (msk->pm.subflows < subflows_max) { 648 msk->pm.subflows++; 649 addrs[i++] = entry->addr; 650 } 651 } 652 rcu_read_unlock(); 653 654 /* If the array is empty, fill in the single 655 * 'IPADDRANY' local address 656 */ 657 if (!i) { 658 struct mptcp_addr_info local; 659 660 memset(&local, 0, sizeof(local)); 661 local.family = 662 #if IS_ENABLED(CONFIG_MPTCP_IPV6) 663 remote->family == AF_INET6 && 664 ipv6_addr_v4mapped(&remote->addr6) ? AF_INET : 665 #endif 666 remote->family; 667 668 if (!mptcp_pm_addr_families_match(sk, &local, remote)) 669 return 0; 670 671 msk->pm.subflows++; 672 addrs[i++] = local; 673 } 674 675 return i; 676 } 677 678 static void mptcp_pm_nl_add_addr_received(struct mptcp_sock *msk) 679 { 680 struct mptcp_addr_info addrs[MPTCP_PM_ADDR_MAX]; 681 struct sock *sk = (struct sock *)msk; 682 unsigned int add_addr_accept_max; 683 struct mptcp_addr_info remote; 684 unsigned int subflows_max; 685 int i, nr; 686 687 add_addr_accept_max = mptcp_pm_get_add_addr_accept_max(msk); 688 subflows_max = mptcp_pm_get_subflows_max(msk); 689 690 pr_debug("accepted %d:%d remote family %d", 691 msk->pm.add_addr_accepted, add_addr_accept_max, 692 msk->pm.remote.family); 693 694 remote = msk->pm.remote; 695 mptcp_pm_announce_addr(msk, &remote, true); 696 mptcp_pm_nl_addr_send_ack(msk); 697 698 if (lookup_subflow_by_daddr(&msk->conn_list, &remote)) 699 return; 700 701 /* pick id 0 port, if none is provided the remote address */ 702 if (!remote.port) 703 remote.port = sk->sk_dport; 704 705 /* connect to the specified remote address, using whatever 706 * local address the routing configuration will pick. 707 */ 708 nr = fill_local_addresses_vec(msk, &remote, addrs); 709 if (nr == 0) 710 return; 711 712 msk->pm.add_addr_accepted++; 713 if (msk->pm.add_addr_accepted >= add_addr_accept_max || 714 msk->pm.subflows >= subflows_max) 715 WRITE_ONCE(msk->pm.accept_addr, false); 716 717 spin_unlock_bh(&msk->pm.lock); 718 for (i = 0; i < nr; i++) 719 __mptcp_subflow_connect(sk, &addrs[i], &remote); 720 spin_lock_bh(&msk->pm.lock); 721 } 722 723 void mptcp_pm_nl_addr_send_ack(struct mptcp_sock *msk) 724 { 725 struct mptcp_subflow_context *subflow; 726 727 msk_owned_by_me(msk); 728 lockdep_assert_held(&msk->pm.lock); 729 730 if (!mptcp_pm_should_add_signal(msk) && 731 !mptcp_pm_should_rm_signal(msk)) 732 return; 733 734 subflow = list_first_entry_or_null(&msk->conn_list, typeof(*subflow), node); 735 if (subflow) 736 mptcp_pm_send_ack(msk, subflow, false, false); 737 } 738 739 int mptcp_pm_nl_mp_prio_send_ack(struct mptcp_sock *msk, 740 struct mptcp_addr_info *addr, 741 struct mptcp_addr_info *rem, 742 u8 bkup) 743 { 744 struct mptcp_subflow_context *subflow; 745 746 pr_debug("bkup=%d", bkup); 747 748 mptcp_for_each_subflow(msk, subflow) { 749 struct sock *ssk = mptcp_subflow_tcp_sock(subflow); 750 struct mptcp_addr_info local, remote; 751 752 mptcp_local_address((struct sock_common *)ssk, &local); 753 if (!mptcp_addresses_equal(&local, addr, addr->port)) 754 continue; 755 756 if (rem && rem->family != AF_UNSPEC) { 757 remote_address((struct sock_common *)ssk, &remote); 758 if (!mptcp_addresses_equal(&remote, rem, rem->port)) 759 continue; 760 } 761 762 __mptcp_pm_send_ack(msk, subflow, true, bkup); 763 return 0; 764 } 765 766 return -EINVAL; 767 } 768 769 static bool mptcp_local_id_match(const struct mptcp_sock *msk, u8 local_id, u8 id) 770 { 771 return local_id == id || (!local_id && msk->mpc_endpoint_id == id); 772 } 773 774 static void mptcp_pm_nl_rm_addr_or_subflow(struct mptcp_sock *msk, 775 const struct mptcp_rm_list *rm_list, 776 enum linux_mptcp_mib_field rm_type) 777 { 778 struct mptcp_subflow_context *subflow, *tmp; 779 struct sock *sk = (struct sock *)msk; 780 u8 i; 781 782 pr_debug("%s rm_list_nr %d", 783 rm_type == MPTCP_MIB_RMADDR ? "address" : "subflow", rm_list->nr); 784 785 msk_owned_by_me(msk); 786 787 if (sk->sk_state == TCP_LISTEN) 788 return; 789 790 if (!rm_list->nr) 791 return; 792 793 if (list_empty(&msk->conn_list)) 794 return; 795 796 for (i = 0; i < rm_list->nr; i++) { 797 u8 rm_id = rm_list->ids[i]; 798 bool removed = false; 799 800 mptcp_for_each_subflow_safe(msk, subflow, tmp) { 801 struct sock *ssk = mptcp_subflow_tcp_sock(subflow); 802 int how = RCV_SHUTDOWN | SEND_SHUTDOWN; 803 u8 id = subflow->local_id; 804 805 if (rm_type == MPTCP_MIB_RMADDR && subflow->remote_id != rm_id) 806 continue; 807 if (rm_type == MPTCP_MIB_RMSUBFLOW && !mptcp_local_id_match(msk, id, rm_id)) 808 continue; 809 810 pr_debug(" -> %s rm_list_ids[%d]=%u local_id=%u remote_id=%u mpc_id=%u", 811 rm_type == MPTCP_MIB_RMADDR ? "address" : "subflow", 812 i, rm_id, subflow->local_id, subflow->remote_id, 813 msk->mpc_endpoint_id); 814 spin_unlock_bh(&msk->pm.lock); 815 mptcp_subflow_shutdown(sk, ssk, how); 816 817 /* the following takes care of updating the subflows counter */ 818 mptcp_close_ssk(sk, ssk, subflow); 819 spin_lock_bh(&msk->pm.lock); 820 821 removed = true; 822 __MPTCP_INC_STATS(sock_net(sk), rm_type); 823 } 824 if (rm_type == MPTCP_MIB_RMSUBFLOW) 825 __set_bit(rm_id ? rm_id : msk->mpc_endpoint_id, msk->pm.id_avail_bitmap); 826 if (!removed) 827 continue; 828 829 if (!mptcp_pm_is_kernel(msk)) 830 continue; 831 832 if (rm_type == MPTCP_MIB_RMADDR) { 833 msk->pm.add_addr_accepted--; 834 WRITE_ONCE(msk->pm.accept_addr, true); 835 } else if (rm_type == MPTCP_MIB_RMSUBFLOW) { 836 msk->pm.local_addr_used--; 837 } 838 } 839 } 840 841 static void mptcp_pm_nl_rm_addr_received(struct mptcp_sock *msk) 842 { 843 mptcp_pm_nl_rm_addr_or_subflow(msk, &msk->pm.rm_list_rx, MPTCP_MIB_RMADDR); 844 } 845 846 void mptcp_pm_nl_rm_subflow_received(struct mptcp_sock *msk, 847 const struct mptcp_rm_list *rm_list) 848 { 849 mptcp_pm_nl_rm_addr_or_subflow(msk, rm_list, MPTCP_MIB_RMSUBFLOW); 850 } 851 852 void mptcp_pm_nl_work(struct mptcp_sock *msk) 853 { 854 struct mptcp_pm_data *pm = &msk->pm; 855 856 msk_owned_by_me(msk); 857 858 if (!(pm->status & MPTCP_PM_WORK_MASK)) 859 return; 860 861 spin_lock_bh(&msk->pm.lock); 862 863 pr_debug("msk=%p status=%x", msk, pm->status); 864 if (pm->status & BIT(MPTCP_PM_ADD_ADDR_RECEIVED)) { 865 pm->status &= ~BIT(MPTCP_PM_ADD_ADDR_RECEIVED); 866 mptcp_pm_nl_add_addr_received(msk); 867 } 868 if (pm->status & BIT(MPTCP_PM_ADD_ADDR_SEND_ACK)) { 869 pm->status &= ~BIT(MPTCP_PM_ADD_ADDR_SEND_ACK); 870 mptcp_pm_nl_addr_send_ack(msk); 871 } 872 if (pm->status & BIT(MPTCP_PM_RM_ADDR_RECEIVED)) { 873 pm->status &= ~BIT(MPTCP_PM_RM_ADDR_RECEIVED); 874 mptcp_pm_nl_rm_addr_received(msk); 875 } 876 if (pm->status & BIT(MPTCP_PM_ESTABLISHED)) { 877 pm->status &= ~BIT(MPTCP_PM_ESTABLISHED); 878 mptcp_pm_nl_fully_established(msk); 879 } 880 if (pm->status & BIT(MPTCP_PM_SUBFLOW_ESTABLISHED)) { 881 pm->status &= ~BIT(MPTCP_PM_SUBFLOW_ESTABLISHED); 882 mptcp_pm_nl_subflow_established(msk); 883 } 884 885 spin_unlock_bh(&msk->pm.lock); 886 } 887 888 static bool address_use_port(struct mptcp_pm_addr_entry *entry) 889 { 890 return (entry->flags & 891 (MPTCP_PM_ADDR_FLAG_SIGNAL | MPTCP_PM_ADDR_FLAG_SUBFLOW)) == 892 MPTCP_PM_ADDR_FLAG_SIGNAL; 893 } 894 895 /* caller must ensure the RCU grace period is already elapsed */ 896 static void __mptcp_pm_release_addr_entry(struct mptcp_pm_addr_entry *entry) 897 { 898 if (entry->lsk) 899 sock_release(entry->lsk); 900 kfree(entry); 901 } 902 903 static int mptcp_pm_nl_append_new_local_addr(struct pm_nl_pernet *pernet, 904 struct mptcp_pm_addr_entry *entry) 905 { 906 struct mptcp_pm_addr_entry *cur, *del_entry = NULL; 907 unsigned int addr_max; 908 int ret = -EINVAL; 909 910 spin_lock_bh(&pernet->lock); 911 /* to keep the code simple, don't do IDR-like allocation for address ID, 912 * just bail when we exceed limits 913 */ 914 if (pernet->next_id == MPTCP_PM_MAX_ADDR_ID) 915 pernet->next_id = 1; 916 if (pernet->addrs >= MPTCP_PM_ADDR_MAX) { 917 ret = -ERANGE; 918 goto out; 919 } 920 if (test_bit(entry->addr.id, pernet->id_bitmap)) { 921 ret = -EBUSY; 922 goto out; 923 } 924 925 /* do not insert duplicate address, differentiate on port only 926 * singled addresses 927 */ 928 if (!address_use_port(entry)) 929 entry->addr.port = 0; 930 list_for_each_entry(cur, &pernet->local_addr_list, list) { 931 if (mptcp_addresses_equal(&cur->addr, &entry->addr, 932 cur->addr.port || entry->addr.port)) { 933 /* allow replacing the exiting endpoint only if such 934 * endpoint is an implicit one and the user-space 935 * did not provide an endpoint id 936 */ 937 if (!(cur->flags & MPTCP_PM_ADDR_FLAG_IMPLICIT)) { 938 ret = -EEXIST; 939 goto out; 940 } 941 if (entry->addr.id) 942 goto out; 943 944 pernet->addrs--; 945 entry->addr.id = cur->addr.id; 946 list_del_rcu(&cur->list); 947 del_entry = cur; 948 break; 949 } 950 } 951 952 if (!entry->addr.id) { 953 find_next: 954 entry->addr.id = find_next_zero_bit(pernet->id_bitmap, 955 MPTCP_PM_MAX_ADDR_ID + 1, 956 pernet->next_id); 957 if (!entry->addr.id && pernet->next_id != 1) { 958 pernet->next_id = 1; 959 goto find_next; 960 } 961 } 962 963 if (!entry->addr.id) 964 goto out; 965 966 __set_bit(entry->addr.id, pernet->id_bitmap); 967 if (entry->addr.id > pernet->next_id) 968 pernet->next_id = entry->addr.id; 969 970 if (entry->flags & MPTCP_PM_ADDR_FLAG_SIGNAL) { 971 addr_max = pernet->add_addr_signal_max; 972 WRITE_ONCE(pernet->add_addr_signal_max, addr_max + 1); 973 } 974 if (entry->flags & MPTCP_PM_ADDR_FLAG_SUBFLOW) { 975 addr_max = pernet->local_addr_max; 976 WRITE_ONCE(pernet->local_addr_max, addr_max + 1); 977 } 978 979 pernet->addrs++; 980 if (!entry->addr.port) 981 list_add_tail_rcu(&entry->list, &pernet->local_addr_list); 982 else 983 list_add_rcu(&entry->list, &pernet->local_addr_list); 984 ret = entry->addr.id; 985 986 out: 987 spin_unlock_bh(&pernet->lock); 988 989 /* just replaced an existing entry, free it */ 990 if (del_entry) { 991 synchronize_rcu(); 992 __mptcp_pm_release_addr_entry(del_entry); 993 } 994 return ret; 995 } 996 997 static struct lock_class_key mptcp_slock_keys[2]; 998 static struct lock_class_key mptcp_keys[2]; 999 1000 static int mptcp_pm_nl_create_listen_socket(struct sock *sk, 1001 struct mptcp_pm_addr_entry *entry) 1002 { 1003 bool is_ipv6 = sk->sk_family == AF_INET6; 1004 int addrlen = sizeof(struct sockaddr_in); 1005 struct sockaddr_storage addr; 1006 struct sock *newsk, *ssk; 1007 int backlog = 1024; 1008 int err; 1009 1010 err = sock_create_kern(sock_net(sk), entry->addr.family, 1011 SOCK_STREAM, IPPROTO_MPTCP, &entry->lsk); 1012 if (err) 1013 return err; 1014 1015 newsk = entry->lsk->sk; 1016 if (!newsk) 1017 return -EINVAL; 1018 1019 /* The subflow socket lock is acquired in a nested to the msk one 1020 * in several places, even by the TCP stack, and this msk is a kernel 1021 * socket: lockdep complains. Instead of propagating the _nested 1022 * modifiers in several places, re-init the lock class for the msk 1023 * socket to an mptcp specific one. 1024 */ 1025 sock_lock_init_class_and_name(newsk, 1026 is_ipv6 ? "mlock-AF_INET6" : "mlock-AF_INET", 1027 &mptcp_slock_keys[is_ipv6], 1028 is_ipv6 ? "msk_lock-AF_INET6" : "msk_lock-AF_INET", 1029 &mptcp_keys[is_ipv6]); 1030 1031 lock_sock(newsk); 1032 ssk = __mptcp_nmpc_sk(mptcp_sk(newsk)); 1033 release_sock(newsk); 1034 if (IS_ERR(ssk)) 1035 return PTR_ERR(ssk); 1036 1037 mptcp_info2sockaddr(&entry->addr, &addr, entry->addr.family); 1038 #if IS_ENABLED(CONFIG_MPTCP_IPV6) 1039 if (entry->addr.family == AF_INET6) 1040 addrlen = sizeof(struct sockaddr_in6); 1041 #endif 1042 if (ssk->sk_family == AF_INET) 1043 err = inet_bind_sk(ssk, (struct sockaddr *)&addr, addrlen); 1044 #if IS_ENABLED(CONFIG_MPTCP_IPV6) 1045 else if (ssk->sk_family == AF_INET6) 1046 err = inet6_bind_sk(ssk, (struct sockaddr *)&addr, addrlen); 1047 #endif 1048 if (err) 1049 return err; 1050 1051 inet_sk_state_store(newsk, TCP_LISTEN); 1052 lock_sock(ssk); 1053 err = __inet_listen_sk(ssk, backlog); 1054 if (!err) 1055 mptcp_event_pm_listener(ssk, MPTCP_EVENT_LISTENER_CREATED); 1056 release_sock(ssk); 1057 return err; 1058 } 1059 1060 int mptcp_pm_nl_get_local_id(struct mptcp_sock *msk, struct mptcp_addr_info *skc) 1061 { 1062 struct mptcp_pm_addr_entry *entry; 1063 struct pm_nl_pernet *pernet; 1064 int ret = -1; 1065 1066 pernet = pm_nl_get_pernet_from_msk(msk); 1067 1068 rcu_read_lock(); 1069 list_for_each_entry_rcu(entry, &pernet->local_addr_list, list) { 1070 if (mptcp_addresses_equal(&entry->addr, skc, entry->addr.port)) { 1071 ret = entry->addr.id; 1072 break; 1073 } 1074 } 1075 rcu_read_unlock(); 1076 if (ret >= 0) 1077 return ret; 1078 1079 /* address not found, add to local list */ 1080 entry = kmalloc(sizeof(*entry), GFP_ATOMIC); 1081 if (!entry) 1082 return -ENOMEM; 1083 1084 entry->addr = *skc; 1085 entry->addr.id = 0; 1086 entry->addr.port = 0; 1087 entry->ifindex = 0; 1088 entry->flags = MPTCP_PM_ADDR_FLAG_IMPLICIT; 1089 entry->lsk = NULL; 1090 ret = mptcp_pm_nl_append_new_local_addr(pernet, entry); 1091 if (ret < 0) 1092 kfree(entry); 1093 1094 return ret; 1095 } 1096 1097 #define MPTCP_PM_CMD_GRP_OFFSET 0 1098 #define MPTCP_PM_EV_GRP_OFFSET 1 1099 1100 static const struct genl_multicast_group mptcp_pm_mcgrps[] = { 1101 [MPTCP_PM_CMD_GRP_OFFSET] = { .name = MPTCP_PM_CMD_GRP_NAME, }, 1102 [MPTCP_PM_EV_GRP_OFFSET] = { .name = MPTCP_PM_EV_GRP_NAME, 1103 .flags = GENL_UNS_ADMIN_PERM, 1104 }, 1105 }; 1106 1107 static const struct nla_policy 1108 mptcp_pm_addr_policy[MPTCP_PM_ADDR_ATTR_MAX + 1] = { 1109 [MPTCP_PM_ADDR_ATTR_FAMILY] = { .type = NLA_U16, }, 1110 [MPTCP_PM_ADDR_ATTR_ID] = { .type = NLA_U8, }, 1111 [MPTCP_PM_ADDR_ATTR_ADDR4] = { .type = NLA_U32, }, 1112 [MPTCP_PM_ADDR_ATTR_ADDR6] = 1113 NLA_POLICY_EXACT_LEN(sizeof(struct in6_addr)), 1114 [MPTCP_PM_ADDR_ATTR_PORT] = { .type = NLA_U16 }, 1115 [MPTCP_PM_ADDR_ATTR_FLAGS] = { .type = NLA_U32 }, 1116 [MPTCP_PM_ADDR_ATTR_IF_IDX] = { .type = NLA_S32 }, 1117 }; 1118 1119 static const struct nla_policy mptcp_pm_policy[MPTCP_PM_ATTR_MAX + 1] = { 1120 [MPTCP_PM_ATTR_ADDR] = 1121 NLA_POLICY_NESTED(mptcp_pm_addr_policy), 1122 [MPTCP_PM_ATTR_RCV_ADD_ADDRS] = { .type = NLA_U32, }, 1123 [MPTCP_PM_ATTR_SUBFLOWS] = { .type = NLA_U32, }, 1124 [MPTCP_PM_ATTR_TOKEN] = { .type = NLA_U32, }, 1125 [MPTCP_PM_ATTR_LOC_ID] = { .type = NLA_U8, }, 1126 [MPTCP_PM_ATTR_ADDR_REMOTE] = 1127 NLA_POLICY_NESTED(mptcp_pm_addr_policy), 1128 }; 1129 1130 void mptcp_pm_nl_subflow_chk_stale(const struct mptcp_sock *msk, struct sock *ssk) 1131 { 1132 struct mptcp_subflow_context *iter, *subflow = mptcp_subflow_ctx(ssk); 1133 struct sock *sk = (struct sock *)msk; 1134 unsigned int active_max_loss_cnt; 1135 struct net *net = sock_net(sk); 1136 unsigned int stale_loss_cnt; 1137 bool slow; 1138 1139 stale_loss_cnt = mptcp_stale_loss_cnt(net); 1140 if (subflow->stale || !stale_loss_cnt || subflow->stale_count <= stale_loss_cnt) 1141 return; 1142 1143 /* look for another available subflow not in loss state */ 1144 active_max_loss_cnt = max_t(int, stale_loss_cnt - 1, 1); 1145 mptcp_for_each_subflow(msk, iter) { 1146 if (iter != subflow && mptcp_subflow_active(iter) && 1147 iter->stale_count < active_max_loss_cnt) { 1148 /* we have some alternatives, try to mark this subflow as idle ...*/ 1149 slow = lock_sock_fast(ssk); 1150 if (!tcp_rtx_and_write_queues_empty(ssk)) { 1151 subflow->stale = 1; 1152 __mptcp_retransmit_pending_data(sk); 1153 MPTCP_INC_STATS(net, MPTCP_MIB_SUBFLOWSTALE); 1154 } 1155 unlock_sock_fast(ssk, slow); 1156 1157 /* always try to push the pending data regardless of re-injections: 1158 * we can possibly use backup subflows now, and subflow selection 1159 * is cheap under the msk socket lock 1160 */ 1161 __mptcp_push_pending(sk, 0); 1162 return; 1163 } 1164 } 1165 } 1166 1167 static int mptcp_pm_family_to_addr(int family) 1168 { 1169 #if IS_ENABLED(CONFIG_MPTCP_IPV6) 1170 if (family == AF_INET6) 1171 return MPTCP_PM_ADDR_ATTR_ADDR6; 1172 #endif 1173 return MPTCP_PM_ADDR_ATTR_ADDR4; 1174 } 1175 1176 static int mptcp_pm_parse_pm_addr_attr(struct nlattr *tb[], 1177 const struct nlattr *attr, 1178 struct genl_info *info, 1179 struct mptcp_addr_info *addr, 1180 bool require_family) 1181 { 1182 int err, addr_addr; 1183 1184 if (!attr) { 1185 GENL_SET_ERR_MSG(info, "missing address info"); 1186 return -EINVAL; 1187 } 1188 1189 /* no validation needed - was already done via nested policy */ 1190 err = nla_parse_nested_deprecated(tb, MPTCP_PM_ADDR_ATTR_MAX, attr, 1191 mptcp_pm_addr_policy, info->extack); 1192 if (err) 1193 return err; 1194 1195 if (tb[MPTCP_PM_ADDR_ATTR_ID]) 1196 addr->id = nla_get_u8(tb[MPTCP_PM_ADDR_ATTR_ID]); 1197 1198 if (!tb[MPTCP_PM_ADDR_ATTR_FAMILY]) { 1199 if (!require_family) 1200 return 0; 1201 1202 NL_SET_ERR_MSG_ATTR(info->extack, attr, 1203 "missing family"); 1204 return -EINVAL; 1205 } 1206 1207 addr->family = nla_get_u16(tb[MPTCP_PM_ADDR_ATTR_FAMILY]); 1208 if (addr->family != AF_INET 1209 #if IS_ENABLED(CONFIG_MPTCP_IPV6) 1210 && addr->family != AF_INET6 1211 #endif 1212 ) { 1213 NL_SET_ERR_MSG_ATTR(info->extack, attr, 1214 "unknown address family"); 1215 return -EINVAL; 1216 } 1217 addr_addr = mptcp_pm_family_to_addr(addr->family); 1218 if (!tb[addr_addr]) { 1219 NL_SET_ERR_MSG_ATTR(info->extack, attr, 1220 "missing address data"); 1221 return -EINVAL; 1222 } 1223 1224 #if IS_ENABLED(CONFIG_MPTCP_IPV6) 1225 if (addr->family == AF_INET6) 1226 addr->addr6 = nla_get_in6_addr(tb[addr_addr]); 1227 else 1228 #endif 1229 addr->addr.s_addr = nla_get_in_addr(tb[addr_addr]); 1230 1231 if (tb[MPTCP_PM_ADDR_ATTR_PORT]) 1232 addr->port = htons(nla_get_u16(tb[MPTCP_PM_ADDR_ATTR_PORT])); 1233 1234 return 0; 1235 } 1236 1237 int mptcp_pm_parse_addr(struct nlattr *attr, struct genl_info *info, 1238 struct mptcp_addr_info *addr) 1239 { 1240 struct nlattr *tb[MPTCP_PM_ADDR_ATTR_MAX + 1]; 1241 1242 memset(addr, 0, sizeof(*addr)); 1243 1244 return mptcp_pm_parse_pm_addr_attr(tb, attr, info, addr, true); 1245 } 1246 1247 int mptcp_pm_parse_entry(struct nlattr *attr, struct genl_info *info, 1248 bool require_family, 1249 struct mptcp_pm_addr_entry *entry) 1250 { 1251 struct nlattr *tb[MPTCP_PM_ADDR_ATTR_MAX + 1]; 1252 int err; 1253 1254 memset(entry, 0, sizeof(*entry)); 1255 1256 err = mptcp_pm_parse_pm_addr_attr(tb, attr, info, &entry->addr, require_family); 1257 if (err) 1258 return err; 1259 1260 if (tb[MPTCP_PM_ADDR_ATTR_IF_IDX]) { 1261 u32 val = nla_get_s32(tb[MPTCP_PM_ADDR_ATTR_IF_IDX]); 1262 1263 entry->ifindex = val; 1264 } 1265 1266 if (tb[MPTCP_PM_ADDR_ATTR_FLAGS]) 1267 entry->flags = nla_get_u32(tb[MPTCP_PM_ADDR_ATTR_FLAGS]); 1268 1269 if (tb[MPTCP_PM_ADDR_ATTR_PORT]) 1270 entry->addr.port = htons(nla_get_u16(tb[MPTCP_PM_ADDR_ATTR_PORT])); 1271 1272 return 0; 1273 } 1274 1275 static struct pm_nl_pernet *genl_info_pm_nl(struct genl_info *info) 1276 { 1277 return pm_nl_get_pernet(genl_info_net(info)); 1278 } 1279 1280 static int mptcp_nl_add_subflow_or_signal_addr(struct net *net) 1281 { 1282 struct mptcp_sock *msk; 1283 long s_slot = 0, s_num = 0; 1284 1285 while ((msk = mptcp_token_iter_next(net, &s_slot, &s_num)) != NULL) { 1286 struct sock *sk = (struct sock *)msk; 1287 1288 if (!READ_ONCE(msk->fully_established) || 1289 mptcp_pm_is_userspace(msk)) 1290 goto next; 1291 1292 lock_sock(sk); 1293 spin_lock_bh(&msk->pm.lock); 1294 mptcp_pm_create_subflow_or_signal_addr(msk); 1295 spin_unlock_bh(&msk->pm.lock); 1296 release_sock(sk); 1297 1298 next: 1299 sock_put(sk); 1300 cond_resched(); 1301 } 1302 1303 return 0; 1304 } 1305 1306 static int mptcp_nl_cmd_add_addr(struct sk_buff *skb, struct genl_info *info) 1307 { 1308 struct nlattr *attr = info->attrs[MPTCP_PM_ATTR_ADDR]; 1309 struct pm_nl_pernet *pernet = genl_info_pm_nl(info); 1310 struct mptcp_pm_addr_entry addr, *entry; 1311 int ret; 1312 1313 ret = mptcp_pm_parse_entry(attr, info, true, &addr); 1314 if (ret < 0) 1315 return ret; 1316 1317 if (addr.addr.port && !(addr.flags & MPTCP_PM_ADDR_FLAG_SIGNAL)) { 1318 GENL_SET_ERR_MSG(info, "flags must have signal when using port"); 1319 return -EINVAL; 1320 } 1321 1322 if (addr.flags & MPTCP_PM_ADDR_FLAG_SIGNAL && 1323 addr.flags & MPTCP_PM_ADDR_FLAG_FULLMESH) { 1324 GENL_SET_ERR_MSG(info, "flags mustn't have both signal and fullmesh"); 1325 return -EINVAL; 1326 } 1327 1328 if (addr.flags & MPTCP_PM_ADDR_FLAG_IMPLICIT) { 1329 GENL_SET_ERR_MSG(info, "can't create IMPLICIT endpoint"); 1330 return -EINVAL; 1331 } 1332 1333 entry = kzalloc(sizeof(*entry), GFP_KERNEL_ACCOUNT); 1334 if (!entry) { 1335 GENL_SET_ERR_MSG(info, "can't allocate addr"); 1336 return -ENOMEM; 1337 } 1338 1339 *entry = addr; 1340 if (entry->addr.port) { 1341 ret = mptcp_pm_nl_create_listen_socket(skb->sk, entry); 1342 if (ret) { 1343 GENL_SET_ERR_MSG_FMT(info, "create listen socket error: %d", ret); 1344 goto out_free; 1345 } 1346 } 1347 ret = mptcp_pm_nl_append_new_local_addr(pernet, entry); 1348 if (ret < 0) { 1349 GENL_SET_ERR_MSG_FMT(info, "too many addresses or duplicate one: %d", ret); 1350 goto out_free; 1351 } 1352 1353 mptcp_nl_add_subflow_or_signal_addr(sock_net(skb->sk)); 1354 return 0; 1355 1356 out_free: 1357 __mptcp_pm_release_addr_entry(entry); 1358 return ret; 1359 } 1360 1361 int mptcp_pm_nl_get_flags_and_ifindex_by_id(struct mptcp_sock *msk, unsigned int id, 1362 u8 *flags, int *ifindex) 1363 { 1364 struct mptcp_pm_addr_entry *entry; 1365 struct sock *sk = (struct sock *)msk; 1366 struct net *net = sock_net(sk); 1367 1368 rcu_read_lock(); 1369 entry = __lookup_addr_by_id(pm_nl_get_pernet(net), id); 1370 if (entry) { 1371 *flags = entry->flags; 1372 *ifindex = entry->ifindex; 1373 } 1374 rcu_read_unlock(); 1375 1376 return 0; 1377 } 1378 1379 static bool remove_anno_list_by_saddr(struct mptcp_sock *msk, 1380 const struct mptcp_addr_info *addr) 1381 { 1382 struct mptcp_pm_add_entry *entry; 1383 1384 entry = mptcp_pm_del_add_timer(msk, addr, false); 1385 if (entry) { 1386 list_del(&entry->list); 1387 kfree(entry); 1388 return true; 1389 } 1390 1391 return false; 1392 } 1393 1394 static bool mptcp_pm_remove_anno_addr(struct mptcp_sock *msk, 1395 const struct mptcp_addr_info *addr, 1396 bool force) 1397 { 1398 struct mptcp_rm_list list = { .nr = 0 }; 1399 bool ret; 1400 1401 list.ids[list.nr++] = addr->id; 1402 1403 ret = remove_anno_list_by_saddr(msk, addr); 1404 if (ret || force) { 1405 spin_lock_bh(&msk->pm.lock); 1406 mptcp_pm_remove_addr(msk, &list); 1407 spin_unlock_bh(&msk->pm.lock); 1408 } 1409 return ret; 1410 } 1411 1412 static int mptcp_nl_remove_subflow_and_signal_addr(struct net *net, 1413 const struct mptcp_pm_addr_entry *entry) 1414 { 1415 const struct mptcp_addr_info *addr = &entry->addr; 1416 struct mptcp_rm_list list = { .nr = 0 }; 1417 long s_slot = 0, s_num = 0; 1418 struct mptcp_sock *msk; 1419 1420 pr_debug("remove_id=%d", addr->id); 1421 1422 list.ids[list.nr++] = addr->id; 1423 1424 while ((msk = mptcp_token_iter_next(net, &s_slot, &s_num)) != NULL) { 1425 struct sock *sk = (struct sock *)msk; 1426 bool remove_subflow; 1427 1428 if (mptcp_pm_is_userspace(msk)) 1429 goto next; 1430 1431 if (list_empty(&msk->conn_list)) { 1432 mptcp_pm_remove_anno_addr(msk, addr, false); 1433 goto next; 1434 } 1435 1436 lock_sock(sk); 1437 remove_subflow = lookup_subflow_by_saddr(&msk->conn_list, addr); 1438 mptcp_pm_remove_anno_addr(msk, addr, remove_subflow && 1439 !(entry->flags & MPTCP_PM_ADDR_FLAG_IMPLICIT)); 1440 if (remove_subflow) 1441 mptcp_pm_remove_subflow(msk, &list); 1442 release_sock(sk); 1443 1444 next: 1445 sock_put(sk); 1446 cond_resched(); 1447 } 1448 1449 return 0; 1450 } 1451 1452 static int mptcp_nl_remove_id_zero_address(struct net *net, 1453 struct mptcp_addr_info *addr) 1454 { 1455 struct mptcp_rm_list list = { .nr = 0 }; 1456 long s_slot = 0, s_num = 0; 1457 struct mptcp_sock *msk; 1458 1459 list.ids[list.nr++] = 0; 1460 1461 while ((msk = mptcp_token_iter_next(net, &s_slot, &s_num)) != NULL) { 1462 struct sock *sk = (struct sock *)msk; 1463 struct mptcp_addr_info msk_local; 1464 1465 if (list_empty(&msk->conn_list) || mptcp_pm_is_userspace(msk)) 1466 goto next; 1467 1468 mptcp_local_address((struct sock_common *)msk, &msk_local); 1469 if (!mptcp_addresses_equal(&msk_local, addr, addr->port)) 1470 goto next; 1471 1472 lock_sock(sk); 1473 spin_lock_bh(&msk->pm.lock); 1474 mptcp_pm_remove_addr(msk, &list); 1475 mptcp_pm_nl_rm_subflow_received(msk, &list); 1476 spin_unlock_bh(&msk->pm.lock); 1477 release_sock(sk); 1478 1479 next: 1480 sock_put(sk); 1481 cond_resched(); 1482 } 1483 1484 return 0; 1485 } 1486 1487 static int mptcp_nl_cmd_del_addr(struct sk_buff *skb, struct genl_info *info) 1488 { 1489 struct nlattr *attr = info->attrs[MPTCP_PM_ATTR_ADDR]; 1490 struct pm_nl_pernet *pernet = genl_info_pm_nl(info); 1491 struct mptcp_pm_addr_entry addr, *entry; 1492 unsigned int addr_max; 1493 int ret; 1494 1495 ret = mptcp_pm_parse_entry(attr, info, false, &addr); 1496 if (ret < 0) 1497 return ret; 1498 1499 /* the zero id address is special: the first address used by the msk 1500 * always gets such an id, so different subflows can have different zero 1501 * id addresses. Additionally zero id is not accounted for in id_bitmap. 1502 * Let's use an 'mptcp_rm_list' instead of the common remove code. 1503 */ 1504 if (addr.addr.id == 0) 1505 return mptcp_nl_remove_id_zero_address(sock_net(skb->sk), &addr.addr); 1506 1507 spin_lock_bh(&pernet->lock); 1508 entry = __lookup_addr_by_id(pernet, addr.addr.id); 1509 if (!entry) { 1510 GENL_SET_ERR_MSG(info, "address not found"); 1511 spin_unlock_bh(&pernet->lock); 1512 return -EINVAL; 1513 } 1514 if (entry->flags & MPTCP_PM_ADDR_FLAG_SIGNAL) { 1515 addr_max = pernet->add_addr_signal_max; 1516 WRITE_ONCE(pernet->add_addr_signal_max, addr_max - 1); 1517 } 1518 if (entry->flags & MPTCP_PM_ADDR_FLAG_SUBFLOW) { 1519 addr_max = pernet->local_addr_max; 1520 WRITE_ONCE(pernet->local_addr_max, addr_max - 1); 1521 } 1522 1523 pernet->addrs--; 1524 list_del_rcu(&entry->list); 1525 __clear_bit(entry->addr.id, pernet->id_bitmap); 1526 spin_unlock_bh(&pernet->lock); 1527 1528 mptcp_nl_remove_subflow_and_signal_addr(sock_net(skb->sk), entry); 1529 synchronize_rcu(); 1530 __mptcp_pm_release_addr_entry(entry); 1531 1532 return ret; 1533 } 1534 1535 void mptcp_pm_remove_addrs(struct mptcp_sock *msk, struct list_head *rm_list) 1536 { 1537 struct mptcp_rm_list alist = { .nr = 0 }; 1538 struct mptcp_pm_addr_entry *entry; 1539 1540 list_for_each_entry(entry, rm_list, list) { 1541 if ((remove_anno_list_by_saddr(msk, &entry->addr) || 1542 lookup_subflow_by_saddr(&msk->conn_list, &entry->addr)) && 1543 alist.nr < MPTCP_RM_IDS_MAX) 1544 alist.ids[alist.nr++] = entry->addr.id; 1545 } 1546 1547 if (alist.nr) { 1548 spin_lock_bh(&msk->pm.lock); 1549 mptcp_pm_remove_addr(msk, &alist); 1550 spin_unlock_bh(&msk->pm.lock); 1551 } 1552 } 1553 1554 void mptcp_pm_remove_addrs_and_subflows(struct mptcp_sock *msk, 1555 struct list_head *rm_list) 1556 { 1557 struct mptcp_rm_list alist = { .nr = 0 }, slist = { .nr = 0 }; 1558 struct mptcp_pm_addr_entry *entry; 1559 1560 list_for_each_entry(entry, rm_list, list) { 1561 if (lookup_subflow_by_saddr(&msk->conn_list, &entry->addr) && 1562 slist.nr < MPTCP_RM_IDS_MAX) 1563 slist.ids[slist.nr++] = entry->addr.id; 1564 1565 if (remove_anno_list_by_saddr(msk, &entry->addr) && 1566 alist.nr < MPTCP_RM_IDS_MAX) 1567 alist.ids[alist.nr++] = entry->addr.id; 1568 } 1569 1570 if (alist.nr) { 1571 spin_lock_bh(&msk->pm.lock); 1572 mptcp_pm_remove_addr(msk, &alist); 1573 spin_unlock_bh(&msk->pm.lock); 1574 } 1575 if (slist.nr) 1576 mptcp_pm_remove_subflow(msk, &slist); 1577 } 1578 1579 static void mptcp_nl_remove_addrs_list(struct net *net, 1580 struct list_head *rm_list) 1581 { 1582 long s_slot = 0, s_num = 0; 1583 struct mptcp_sock *msk; 1584 1585 if (list_empty(rm_list)) 1586 return; 1587 1588 while ((msk = mptcp_token_iter_next(net, &s_slot, &s_num)) != NULL) { 1589 struct sock *sk = (struct sock *)msk; 1590 1591 if (!mptcp_pm_is_userspace(msk)) { 1592 lock_sock(sk); 1593 mptcp_pm_remove_addrs_and_subflows(msk, rm_list); 1594 release_sock(sk); 1595 } 1596 1597 sock_put(sk); 1598 cond_resched(); 1599 } 1600 } 1601 1602 /* caller must ensure the RCU grace period is already elapsed */ 1603 static void __flush_addrs(struct list_head *list) 1604 { 1605 while (!list_empty(list)) { 1606 struct mptcp_pm_addr_entry *cur; 1607 1608 cur = list_entry(list->next, 1609 struct mptcp_pm_addr_entry, list); 1610 list_del_rcu(&cur->list); 1611 __mptcp_pm_release_addr_entry(cur); 1612 } 1613 } 1614 1615 static void __reset_counters(struct pm_nl_pernet *pernet) 1616 { 1617 WRITE_ONCE(pernet->add_addr_signal_max, 0); 1618 WRITE_ONCE(pernet->add_addr_accept_max, 0); 1619 WRITE_ONCE(pernet->local_addr_max, 0); 1620 pernet->addrs = 0; 1621 } 1622 1623 static int mptcp_nl_cmd_flush_addrs(struct sk_buff *skb, struct genl_info *info) 1624 { 1625 struct pm_nl_pernet *pernet = genl_info_pm_nl(info); 1626 LIST_HEAD(free_list); 1627 1628 spin_lock_bh(&pernet->lock); 1629 list_splice_init(&pernet->local_addr_list, &free_list); 1630 __reset_counters(pernet); 1631 pernet->next_id = 1; 1632 bitmap_zero(pernet->id_bitmap, MPTCP_PM_MAX_ADDR_ID + 1); 1633 spin_unlock_bh(&pernet->lock); 1634 mptcp_nl_remove_addrs_list(sock_net(skb->sk), &free_list); 1635 synchronize_rcu(); 1636 __flush_addrs(&free_list); 1637 return 0; 1638 } 1639 1640 static int mptcp_nl_fill_addr(struct sk_buff *skb, 1641 struct mptcp_pm_addr_entry *entry) 1642 { 1643 struct mptcp_addr_info *addr = &entry->addr; 1644 struct nlattr *attr; 1645 1646 attr = nla_nest_start(skb, MPTCP_PM_ATTR_ADDR); 1647 if (!attr) 1648 return -EMSGSIZE; 1649 1650 if (nla_put_u16(skb, MPTCP_PM_ADDR_ATTR_FAMILY, addr->family)) 1651 goto nla_put_failure; 1652 if (nla_put_u16(skb, MPTCP_PM_ADDR_ATTR_PORT, ntohs(addr->port))) 1653 goto nla_put_failure; 1654 if (nla_put_u8(skb, MPTCP_PM_ADDR_ATTR_ID, addr->id)) 1655 goto nla_put_failure; 1656 if (nla_put_u32(skb, MPTCP_PM_ADDR_ATTR_FLAGS, entry->flags)) 1657 goto nla_put_failure; 1658 if (entry->ifindex && 1659 nla_put_s32(skb, MPTCP_PM_ADDR_ATTR_IF_IDX, entry->ifindex)) 1660 goto nla_put_failure; 1661 1662 if (addr->family == AF_INET && 1663 nla_put_in_addr(skb, MPTCP_PM_ADDR_ATTR_ADDR4, 1664 addr->addr.s_addr)) 1665 goto nla_put_failure; 1666 #if IS_ENABLED(CONFIG_MPTCP_IPV6) 1667 else if (addr->family == AF_INET6 && 1668 nla_put_in6_addr(skb, MPTCP_PM_ADDR_ATTR_ADDR6, &addr->addr6)) 1669 goto nla_put_failure; 1670 #endif 1671 nla_nest_end(skb, attr); 1672 return 0; 1673 1674 nla_put_failure: 1675 nla_nest_cancel(skb, attr); 1676 return -EMSGSIZE; 1677 } 1678 1679 static int mptcp_nl_cmd_get_addr(struct sk_buff *skb, struct genl_info *info) 1680 { 1681 struct nlattr *attr = info->attrs[MPTCP_PM_ATTR_ADDR]; 1682 struct pm_nl_pernet *pernet = genl_info_pm_nl(info); 1683 struct mptcp_pm_addr_entry addr, *entry; 1684 struct sk_buff *msg; 1685 void *reply; 1686 int ret; 1687 1688 ret = mptcp_pm_parse_entry(attr, info, false, &addr); 1689 if (ret < 0) 1690 return ret; 1691 1692 msg = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL); 1693 if (!msg) 1694 return -ENOMEM; 1695 1696 reply = genlmsg_put_reply(msg, info, &mptcp_genl_family, 0, 1697 info->genlhdr->cmd); 1698 if (!reply) { 1699 GENL_SET_ERR_MSG(info, "not enough space in Netlink message"); 1700 ret = -EMSGSIZE; 1701 goto fail; 1702 } 1703 1704 spin_lock_bh(&pernet->lock); 1705 entry = __lookup_addr_by_id(pernet, addr.addr.id); 1706 if (!entry) { 1707 GENL_SET_ERR_MSG(info, "address not found"); 1708 ret = -EINVAL; 1709 goto unlock_fail; 1710 } 1711 1712 ret = mptcp_nl_fill_addr(msg, entry); 1713 if (ret) 1714 goto unlock_fail; 1715 1716 genlmsg_end(msg, reply); 1717 ret = genlmsg_reply(msg, info); 1718 spin_unlock_bh(&pernet->lock); 1719 return ret; 1720 1721 unlock_fail: 1722 spin_unlock_bh(&pernet->lock); 1723 1724 fail: 1725 nlmsg_free(msg); 1726 return ret; 1727 } 1728 1729 static int mptcp_nl_cmd_dump_addrs(struct sk_buff *msg, 1730 struct netlink_callback *cb) 1731 { 1732 struct net *net = sock_net(msg->sk); 1733 struct mptcp_pm_addr_entry *entry; 1734 struct pm_nl_pernet *pernet; 1735 int id = cb->args[0]; 1736 void *hdr; 1737 int i; 1738 1739 pernet = pm_nl_get_pernet(net); 1740 1741 spin_lock_bh(&pernet->lock); 1742 for (i = id; i < MPTCP_PM_MAX_ADDR_ID + 1; i++) { 1743 if (test_bit(i, pernet->id_bitmap)) { 1744 entry = __lookup_addr_by_id(pernet, i); 1745 if (!entry) 1746 break; 1747 1748 if (entry->addr.id <= id) 1749 continue; 1750 1751 hdr = genlmsg_put(msg, NETLINK_CB(cb->skb).portid, 1752 cb->nlh->nlmsg_seq, &mptcp_genl_family, 1753 NLM_F_MULTI, MPTCP_PM_CMD_GET_ADDR); 1754 if (!hdr) 1755 break; 1756 1757 if (mptcp_nl_fill_addr(msg, entry) < 0) { 1758 genlmsg_cancel(msg, hdr); 1759 break; 1760 } 1761 1762 id = entry->addr.id; 1763 genlmsg_end(msg, hdr); 1764 } 1765 } 1766 spin_unlock_bh(&pernet->lock); 1767 1768 cb->args[0] = id; 1769 return msg->len; 1770 } 1771 1772 static int parse_limit(struct genl_info *info, int id, unsigned int *limit) 1773 { 1774 struct nlattr *attr = info->attrs[id]; 1775 1776 if (!attr) 1777 return 0; 1778 1779 *limit = nla_get_u32(attr); 1780 if (*limit > MPTCP_PM_ADDR_MAX) { 1781 GENL_SET_ERR_MSG(info, "limit greater than maximum"); 1782 return -EINVAL; 1783 } 1784 return 0; 1785 } 1786 1787 static int 1788 mptcp_nl_cmd_set_limits(struct sk_buff *skb, struct genl_info *info) 1789 { 1790 struct pm_nl_pernet *pernet = genl_info_pm_nl(info); 1791 unsigned int rcv_addrs, subflows; 1792 int ret; 1793 1794 spin_lock_bh(&pernet->lock); 1795 rcv_addrs = pernet->add_addr_accept_max; 1796 ret = parse_limit(info, MPTCP_PM_ATTR_RCV_ADD_ADDRS, &rcv_addrs); 1797 if (ret) 1798 goto unlock; 1799 1800 subflows = pernet->subflows_max; 1801 ret = parse_limit(info, MPTCP_PM_ATTR_SUBFLOWS, &subflows); 1802 if (ret) 1803 goto unlock; 1804 1805 WRITE_ONCE(pernet->add_addr_accept_max, rcv_addrs); 1806 WRITE_ONCE(pernet->subflows_max, subflows); 1807 1808 unlock: 1809 spin_unlock_bh(&pernet->lock); 1810 return ret; 1811 } 1812 1813 static int 1814 mptcp_nl_cmd_get_limits(struct sk_buff *skb, struct genl_info *info) 1815 { 1816 struct pm_nl_pernet *pernet = genl_info_pm_nl(info); 1817 struct sk_buff *msg; 1818 void *reply; 1819 1820 msg = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL); 1821 if (!msg) 1822 return -ENOMEM; 1823 1824 reply = genlmsg_put_reply(msg, info, &mptcp_genl_family, 0, 1825 MPTCP_PM_CMD_GET_LIMITS); 1826 if (!reply) 1827 goto fail; 1828 1829 if (nla_put_u32(msg, MPTCP_PM_ATTR_RCV_ADD_ADDRS, 1830 READ_ONCE(pernet->add_addr_accept_max))) 1831 goto fail; 1832 1833 if (nla_put_u32(msg, MPTCP_PM_ATTR_SUBFLOWS, 1834 READ_ONCE(pernet->subflows_max))) 1835 goto fail; 1836 1837 genlmsg_end(msg, reply); 1838 return genlmsg_reply(msg, info); 1839 1840 fail: 1841 GENL_SET_ERR_MSG(info, "not enough space in Netlink message"); 1842 nlmsg_free(msg); 1843 return -EMSGSIZE; 1844 } 1845 1846 static void mptcp_pm_nl_fullmesh(struct mptcp_sock *msk, 1847 struct mptcp_addr_info *addr) 1848 { 1849 struct mptcp_rm_list list = { .nr = 0 }; 1850 1851 list.ids[list.nr++] = addr->id; 1852 1853 spin_lock_bh(&msk->pm.lock); 1854 mptcp_pm_nl_rm_subflow_received(msk, &list); 1855 mptcp_pm_create_subflow_or_signal_addr(msk); 1856 spin_unlock_bh(&msk->pm.lock); 1857 } 1858 1859 static int mptcp_nl_set_flags(struct net *net, 1860 struct mptcp_addr_info *addr, 1861 u8 bkup, u8 changed) 1862 { 1863 long s_slot = 0, s_num = 0; 1864 struct mptcp_sock *msk; 1865 int ret = -EINVAL; 1866 1867 while ((msk = mptcp_token_iter_next(net, &s_slot, &s_num)) != NULL) { 1868 struct sock *sk = (struct sock *)msk; 1869 1870 if (list_empty(&msk->conn_list) || mptcp_pm_is_userspace(msk)) 1871 goto next; 1872 1873 lock_sock(sk); 1874 if (changed & MPTCP_PM_ADDR_FLAG_BACKUP) 1875 ret = mptcp_pm_nl_mp_prio_send_ack(msk, addr, NULL, bkup); 1876 if (changed & MPTCP_PM_ADDR_FLAG_FULLMESH) 1877 mptcp_pm_nl_fullmesh(msk, addr); 1878 release_sock(sk); 1879 1880 next: 1881 sock_put(sk); 1882 cond_resched(); 1883 } 1884 1885 return ret; 1886 } 1887 1888 int mptcp_pm_nl_set_flags(struct net *net, struct mptcp_pm_addr_entry *addr, u8 bkup) 1889 { 1890 struct pm_nl_pernet *pernet = pm_nl_get_pernet(net); 1891 u8 changed, mask = MPTCP_PM_ADDR_FLAG_BACKUP | 1892 MPTCP_PM_ADDR_FLAG_FULLMESH; 1893 struct mptcp_pm_addr_entry *entry; 1894 u8 lookup_by_id = 0; 1895 1896 if (addr->addr.family == AF_UNSPEC) { 1897 lookup_by_id = 1; 1898 if (!addr->addr.id) 1899 return -EOPNOTSUPP; 1900 } 1901 1902 spin_lock_bh(&pernet->lock); 1903 entry = __lookup_addr(pernet, &addr->addr, lookup_by_id); 1904 if (!entry) { 1905 spin_unlock_bh(&pernet->lock); 1906 return -EINVAL; 1907 } 1908 if ((addr->flags & MPTCP_PM_ADDR_FLAG_FULLMESH) && 1909 (entry->flags & MPTCP_PM_ADDR_FLAG_SIGNAL)) { 1910 spin_unlock_bh(&pernet->lock); 1911 return -EINVAL; 1912 } 1913 1914 changed = (addr->flags ^ entry->flags) & mask; 1915 entry->flags = (entry->flags & ~mask) | (addr->flags & mask); 1916 *addr = *entry; 1917 spin_unlock_bh(&pernet->lock); 1918 1919 mptcp_nl_set_flags(net, &addr->addr, bkup, changed); 1920 return 0; 1921 } 1922 1923 static int mptcp_nl_cmd_set_flags(struct sk_buff *skb, struct genl_info *info) 1924 { 1925 struct mptcp_pm_addr_entry remote = { .addr = { .family = AF_UNSPEC }, }; 1926 struct mptcp_pm_addr_entry addr = { .addr = { .family = AF_UNSPEC }, }; 1927 struct nlattr *attr_rem = info->attrs[MPTCP_PM_ATTR_ADDR_REMOTE]; 1928 struct nlattr *token = info->attrs[MPTCP_PM_ATTR_TOKEN]; 1929 struct nlattr *attr = info->attrs[MPTCP_PM_ATTR_ADDR]; 1930 struct net *net = sock_net(skb->sk); 1931 u8 bkup = 0; 1932 int ret; 1933 1934 ret = mptcp_pm_parse_entry(attr, info, false, &addr); 1935 if (ret < 0) 1936 return ret; 1937 1938 if (attr_rem) { 1939 ret = mptcp_pm_parse_entry(attr_rem, info, false, &remote); 1940 if (ret < 0) 1941 return ret; 1942 } 1943 1944 if (addr.flags & MPTCP_PM_ADDR_FLAG_BACKUP) 1945 bkup = 1; 1946 1947 return mptcp_pm_set_flags(net, token, &addr, &remote, bkup); 1948 } 1949 1950 static void mptcp_nl_mcast_send(struct net *net, struct sk_buff *nlskb, gfp_t gfp) 1951 { 1952 genlmsg_multicast_netns(&mptcp_genl_family, net, 1953 nlskb, 0, MPTCP_PM_EV_GRP_OFFSET, gfp); 1954 } 1955 1956 bool mptcp_userspace_pm_active(const struct mptcp_sock *msk) 1957 { 1958 return genl_has_listeners(&mptcp_genl_family, 1959 sock_net((const struct sock *)msk), 1960 MPTCP_PM_EV_GRP_OFFSET); 1961 } 1962 1963 static int mptcp_event_add_subflow(struct sk_buff *skb, const struct sock *ssk) 1964 { 1965 const struct inet_sock *issk = inet_sk(ssk); 1966 const struct mptcp_subflow_context *sf; 1967 1968 if (nla_put_u16(skb, MPTCP_ATTR_FAMILY, ssk->sk_family)) 1969 return -EMSGSIZE; 1970 1971 switch (ssk->sk_family) { 1972 case AF_INET: 1973 if (nla_put_in_addr(skb, MPTCP_ATTR_SADDR4, issk->inet_saddr)) 1974 return -EMSGSIZE; 1975 if (nla_put_in_addr(skb, MPTCP_ATTR_DADDR4, issk->inet_daddr)) 1976 return -EMSGSIZE; 1977 break; 1978 #if IS_ENABLED(CONFIG_MPTCP_IPV6) 1979 case AF_INET6: { 1980 const struct ipv6_pinfo *np = inet6_sk(ssk); 1981 1982 if (nla_put_in6_addr(skb, MPTCP_ATTR_SADDR6, &np->saddr)) 1983 return -EMSGSIZE; 1984 if (nla_put_in6_addr(skb, MPTCP_ATTR_DADDR6, &ssk->sk_v6_daddr)) 1985 return -EMSGSIZE; 1986 break; 1987 } 1988 #endif 1989 default: 1990 WARN_ON_ONCE(1); 1991 return -EMSGSIZE; 1992 } 1993 1994 if (nla_put_be16(skb, MPTCP_ATTR_SPORT, issk->inet_sport)) 1995 return -EMSGSIZE; 1996 if (nla_put_be16(skb, MPTCP_ATTR_DPORT, issk->inet_dport)) 1997 return -EMSGSIZE; 1998 1999 sf = mptcp_subflow_ctx(ssk); 2000 if (WARN_ON_ONCE(!sf)) 2001 return -EINVAL; 2002 2003 if (nla_put_u8(skb, MPTCP_ATTR_LOC_ID, sf->local_id)) 2004 return -EMSGSIZE; 2005 2006 if (nla_put_u8(skb, MPTCP_ATTR_REM_ID, sf->remote_id)) 2007 return -EMSGSIZE; 2008 2009 return 0; 2010 } 2011 2012 static int mptcp_event_put_token_and_ssk(struct sk_buff *skb, 2013 const struct mptcp_sock *msk, 2014 const struct sock *ssk) 2015 { 2016 const struct sock *sk = (const struct sock *)msk; 2017 const struct mptcp_subflow_context *sf; 2018 u8 sk_err; 2019 2020 if (nla_put_u32(skb, MPTCP_ATTR_TOKEN, msk->token)) 2021 return -EMSGSIZE; 2022 2023 if (mptcp_event_add_subflow(skb, ssk)) 2024 return -EMSGSIZE; 2025 2026 sf = mptcp_subflow_ctx(ssk); 2027 if (WARN_ON_ONCE(!sf)) 2028 return -EINVAL; 2029 2030 if (nla_put_u8(skb, MPTCP_ATTR_BACKUP, sf->backup)) 2031 return -EMSGSIZE; 2032 2033 if (ssk->sk_bound_dev_if && 2034 nla_put_s32(skb, MPTCP_ATTR_IF_IDX, ssk->sk_bound_dev_if)) 2035 return -EMSGSIZE; 2036 2037 sk_err = READ_ONCE(ssk->sk_err); 2038 if (sk_err && sk->sk_state == TCP_ESTABLISHED && 2039 nla_put_u8(skb, MPTCP_ATTR_ERROR, sk_err)) 2040 return -EMSGSIZE; 2041 2042 return 0; 2043 } 2044 2045 static int mptcp_event_sub_established(struct sk_buff *skb, 2046 const struct mptcp_sock *msk, 2047 const struct sock *ssk) 2048 { 2049 return mptcp_event_put_token_and_ssk(skb, msk, ssk); 2050 } 2051 2052 static int mptcp_event_sub_closed(struct sk_buff *skb, 2053 const struct mptcp_sock *msk, 2054 const struct sock *ssk) 2055 { 2056 const struct mptcp_subflow_context *sf; 2057 2058 if (mptcp_event_put_token_and_ssk(skb, msk, ssk)) 2059 return -EMSGSIZE; 2060 2061 sf = mptcp_subflow_ctx(ssk); 2062 if (!sf->reset_seen) 2063 return 0; 2064 2065 if (nla_put_u32(skb, MPTCP_ATTR_RESET_REASON, sf->reset_reason)) 2066 return -EMSGSIZE; 2067 2068 if (nla_put_u32(skb, MPTCP_ATTR_RESET_FLAGS, sf->reset_transient)) 2069 return -EMSGSIZE; 2070 2071 return 0; 2072 } 2073 2074 static int mptcp_event_created(struct sk_buff *skb, 2075 const struct mptcp_sock *msk, 2076 const struct sock *ssk) 2077 { 2078 int err = nla_put_u32(skb, MPTCP_ATTR_TOKEN, msk->token); 2079 2080 if (err) 2081 return err; 2082 2083 if (nla_put_u8(skb, MPTCP_ATTR_SERVER_SIDE, READ_ONCE(msk->pm.server_side))) 2084 return -EMSGSIZE; 2085 2086 return mptcp_event_add_subflow(skb, ssk); 2087 } 2088 2089 void mptcp_event_addr_removed(const struct mptcp_sock *msk, uint8_t id) 2090 { 2091 struct net *net = sock_net((const struct sock *)msk); 2092 struct nlmsghdr *nlh; 2093 struct sk_buff *skb; 2094 2095 if (!genl_has_listeners(&mptcp_genl_family, net, MPTCP_PM_EV_GRP_OFFSET)) 2096 return; 2097 2098 skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_ATOMIC); 2099 if (!skb) 2100 return; 2101 2102 nlh = genlmsg_put(skb, 0, 0, &mptcp_genl_family, 0, MPTCP_EVENT_REMOVED); 2103 if (!nlh) 2104 goto nla_put_failure; 2105 2106 if (nla_put_u32(skb, MPTCP_ATTR_TOKEN, msk->token)) 2107 goto nla_put_failure; 2108 2109 if (nla_put_u8(skb, MPTCP_ATTR_REM_ID, id)) 2110 goto nla_put_failure; 2111 2112 genlmsg_end(skb, nlh); 2113 mptcp_nl_mcast_send(net, skb, GFP_ATOMIC); 2114 return; 2115 2116 nla_put_failure: 2117 nlmsg_free(skb); 2118 } 2119 2120 void mptcp_event_addr_announced(const struct sock *ssk, 2121 const struct mptcp_addr_info *info) 2122 { 2123 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk); 2124 struct mptcp_sock *msk = mptcp_sk(subflow->conn); 2125 struct net *net = sock_net(ssk); 2126 struct nlmsghdr *nlh; 2127 struct sk_buff *skb; 2128 2129 if (!genl_has_listeners(&mptcp_genl_family, net, MPTCP_PM_EV_GRP_OFFSET)) 2130 return; 2131 2132 skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_ATOMIC); 2133 if (!skb) 2134 return; 2135 2136 nlh = genlmsg_put(skb, 0, 0, &mptcp_genl_family, 0, 2137 MPTCP_EVENT_ANNOUNCED); 2138 if (!nlh) 2139 goto nla_put_failure; 2140 2141 if (nla_put_u32(skb, MPTCP_ATTR_TOKEN, msk->token)) 2142 goto nla_put_failure; 2143 2144 if (nla_put_u8(skb, MPTCP_ATTR_REM_ID, info->id)) 2145 goto nla_put_failure; 2146 2147 if (nla_put_be16(skb, MPTCP_ATTR_DPORT, 2148 info->port == 0 ? 2149 inet_sk(ssk)->inet_dport : 2150 info->port)) 2151 goto nla_put_failure; 2152 2153 switch (info->family) { 2154 case AF_INET: 2155 if (nla_put_in_addr(skb, MPTCP_ATTR_DADDR4, info->addr.s_addr)) 2156 goto nla_put_failure; 2157 break; 2158 #if IS_ENABLED(CONFIG_MPTCP_IPV6) 2159 case AF_INET6: 2160 if (nla_put_in6_addr(skb, MPTCP_ATTR_DADDR6, &info->addr6)) 2161 goto nla_put_failure; 2162 break; 2163 #endif 2164 default: 2165 WARN_ON_ONCE(1); 2166 goto nla_put_failure; 2167 } 2168 2169 genlmsg_end(skb, nlh); 2170 mptcp_nl_mcast_send(net, skb, GFP_ATOMIC); 2171 return; 2172 2173 nla_put_failure: 2174 nlmsg_free(skb); 2175 } 2176 2177 void mptcp_event_pm_listener(const struct sock *ssk, 2178 enum mptcp_event_type event) 2179 { 2180 const struct inet_sock *issk = inet_sk(ssk); 2181 struct net *net = sock_net(ssk); 2182 struct nlmsghdr *nlh; 2183 struct sk_buff *skb; 2184 2185 if (!genl_has_listeners(&mptcp_genl_family, net, MPTCP_PM_EV_GRP_OFFSET)) 2186 return; 2187 2188 skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL); 2189 if (!skb) 2190 return; 2191 2192 nlh = genlmsg_put(skb, 0, 0, &mptcp_genl_family, 0, event); 2193 if (!nlh) 2194 goto nla_put_failure; 2195 2196 if (nla_put_u16(skb, MPTCP_ATTR_FAMILY, ssk->sk_family)) 2197 goto nla_put_failure; 2198 2199 if (nla_put_be16(skb, MPTCP_ATTR_SPORT, issk->inet_sport)) 2200 goto nla_put_failure; 2201 2202 switch (ssk->sk_family) { 2203 case AF_INET: 2204 if (nla_put_in_addr(skb, MPTCP_ATTR_SADDR4, issk->inet_saddr)) 2205 goto nla_put_failure; 2206 break; 2207 #if IS_ENABLED(CONFIG_MPTCP_IPV6) 2208 case AF_INET6: { 2209 const struct ipv6_pinfo *np = inet6_sk(ssk); 2210 2211 if (nla_put_in6_addr(skb, MPTCP_ATTR_SADDR6, &np->saddr)) 2212 goto nla_put_failure; 2213 break; 2214 } 2215 #endif 2216 default: 2217 WARN_ON_ONCE(1); 2218 goto nla_put_failure; 2219 } 2220 2221 genlmsg_end(skb, nlh); 2222 mptcp_nl_mcast_send(net, skb, GFP_KERNEL); 2223 return; 2224 2225 nla_put_failure: 2226 nlmsg_free(skb); 2227 } 2228 2229 void mptcp_event(enum mptcp_event_type type, const struct mptcp_sock *msk, 2230 const struct sock *ssk, gfp_t gfp) 2231 { 2232 struct net *net = sock_net((const struct sock *)msk); 2233 struct nlmsghdr *nlh; 2234 struct sk_buff *skb; 2235 2236 if (!genl_has_listeners(&mptcp_genl_family, net, MPTCP_PM_EV_GRP_OFFSET)) 2237 return; 2238 2239 skb = nlmsg_new(NLMSG_DEFAULT_SIZE, gfp); 2240 if (!skb) 2241 return; 2242 2243 nlh = genlmsg_put(skb, 0, 0, &mptcp_genl_family, 0, type); 2244 if (!nlh) 2245 goto nla_put_failure; 2246 2247 switch (type) { 2248 case MPTCP_EVENT_UNSPEC: 2249 WARN_ON_ONCE(1); 2250 break; 2251 case MPTCP_EVENT_CREATED: 2252 case MPTCP_EVENT_ESTABLISHED: 2253 if (mptcp_event_created(skb, msk, ssk) < 0) 2254 goto nla_put_failure; 2255 break; 2256 case MPTCP_EVENT_CLOSED: 2257 if (nla_put_u32(skb, MPTCP_ATTR_TOKEN, msk->token) < 0) 2258 goto nla_put_failure; 2259 break; 2260 case MPTCP_EVENT_ANNOUNCED: 2261 case MPTCP_EVENT_REMOVED: 2262 /* call mptcp_event_addr_announced()/removed instead */ 2263 WARN_ON_ONCE(1); 2264 break; 2265 case MPTCP_EVENT_SUB_ESTABLISHED: 2266 case MPTCP_EVENT_SUB_PRIORITY: 2267 if (mptcp_event_sub_established(skb, msk, ssk) < 0) 2268 goto nla_put_failure; 2269 break; 2270 case MPTCP_EVENT_SUB_CLOSED: 2271 if (mptcp_event_sub_closed(skb, msk, ssk) < 0) 2272 goto nla_put_failure; 2273 break; 2274 case MPTCP_EVENT_LISTENER_CREATED: 2275 case MPTCP_EVENT_LISTENER_CLOSED: 2276 break; 2277 } 2278 2279 genlmsg_end(skb, nlh); 2280 mptcp_nl_mcast_send(net, skb, gfp); 2281 return; 2282 2283 nla_put_failure: 2284 nlmsg_free(skb); 2285 } 2286 2287 static const struct genl_small_ops mptcp_pm_ops[] = { 2288 { 2289 .cmd = MPTCP_PM_CMD_ADD_ADDR, 2290 .doit = mptcp_nl_cmd_add_addr, 2291 .flags = GENL_UNS_ADMIN_PERM, 2292 }, 2293 { 2294 .cmd = MPTCP_PM_CMD_DEL_ADDR, 2295 .doit = mptcp_nl_cmd_del_addr, 2296 .flags = GENL_UNS_ADMIN_PERM, 2297 }, 2298 { 2299 .cmd = MPTCP_PM_CMD_FLUSH_ADDRS, 2300 .doit = mptcp_nl_cmd_flush_addrs, 2301 .flags = GENL_UNS_ADMIN_PERM, 2302 }, 2303 { 2304 .cmd = MPTCP_PM_CMD_GET_ADDR, 2305 .doit = mptcp_nl_cmd_get_addr, 2306 .dumpit = mptcp_nl_cmd_dump_addrs, 2307 }, 2308 { 2309 .cmd = MPTCP_PM_CMD_SET_LIMITS, 2310 .doit = mptcp_nl_cmd_set_limits, 2311 .flags = GENL_UNS_ADMIN_PERM, 2312 }, 2313 { 2314 .cmd = MPTCP_PM_CMD_GET_LIMITS, 2315 .doit = mptcp_nl_cmd_get_limits, 2316 }, 2317 { 2318 .cmd = MPTCP_PM_CMD_SET_FLAGS, 2319 .doit = mptcp_nl_cmd_set_flags, 2320 .flags = GENL_UNS_ADMIN_PERM, 2321 }, 2322 { 2323 .cmd = MPTCP_PM_CMD_ANNOUNCE, 2324 .doit = mptcp_nl_cmd_announce, 2325 .flags = GENL_UNS_ADMIN_PERM, 2326 }, 2327 { 2328 .cmd = MPTCP_PM_CMD_REMOVE, 2329 .doit = mptcp_nl_cmd_remove, 2330 .flags = GENL_UNS_ADMIN_PERM, 2331 }, 2332 { 2333 .cmd = MPTCP_PM_CMD_SUBFLOW_CREATE, 2334 .doit = mptcp_nl_cmd_sf_create, 2335 .flags = GENL_UNS_ADMIN_PERM, 2336 }, 2337 { 2338 .cmd = MPTCP_PM_CMD_SUBFLOW_DESTROY, 2339 .doit = mptcp_nl_cmd_sf_destroy, 2340 .flags = GENL_UNS_ADMIN_PERM, 2341 }, 2342 }; 2343 2344 static struct genl_family mptcp_genl_family __ro_after_init = { 2345 .name = MPTCP_PM_NAME, 2346 .version = MPTCP_PM_VER, 2347 .maxattr = MPTCP_PM_ATTR_MAX, 2348 .policy = mptcp_pm_policy, 2349 .netnsok = true, 2350 .module = THIS_MODULE, 2351 .small_ops = mptcp_pm_ops, 2352 .n_small_ops = ARRAY_SIZE(mptcp_pm_ops), 2353 .resv_start_op = MPTCP_PM_CMD_SUBFLOW_DESTROY + 1, 2354 .mcgrps = mptcp_pm_mcgrps, 2355 .n_mcgrps = ARRAY_SIZE(mptcp_pm_mcgrps), 2356 }; 2357 2358 static int __net_init pm_nl_init_net(struct net *net) 2359 { 2360 struct pm_nl_pernet *pernet = pm_nl_get_pernet(net); 2361 2362 INIT_LIST_HEAD_RCU(&pernet->local_addr_list); 2363 2364 /* Cit. 2 subflows ought to be enough for anybody. */ 2365 pernet->subflows_max = 2; 2366 pernet->next_id = 1; 2367 pernet->stale_loss_cnt = 4; 2368 spin_lock_init(&pernet->lock); 2369 2370 /* No need to initialize other pernet fields, the struct is zeroed at 2371 * allocation time. 2372 */ 2373 2374 return 0; 2375 } 2376 2377 static void __net_exit pm_nl_exit_net(struct list_head *net_list) 2378 { 2379 struct net *net; 2380 2381 list_for_each_entry(net, net_list, exit_list) { 2382 struct pm_nl_pernet *pernet = pm_nl_get_pernet(net); 2383 2384 /* net is removed from namespace list, can't race with 2385 * other modifiers, also netns core already waited for a 2386 * RCU grace period. 2387 */ 2388 __flush_addrs(&pernet->local_addr_list); 2389 } 2390 } 2391 2392 static struct pernet_operations mptcp_pm_pernet_ops = { 2393 .init = pm_nl_init_net, 2394 .exit_batch = pm_nl_exit_net, 2395 .id = &pm_nl_pernet_id, 2396 .size = sizeof(struct pm_nl_pernet), 2397 }; 2398 2399 void __init mptcp_pm_nl_init(void) 2400 { 2401 if (register_pernet_subsys(&mptcp_pm_pernet_ops) < 0) 2402 panic("Failed to register MPTCP PM pernet subsystem.\n"); 2403 2404 if (genl_register_family(&mptcp_genl_family)) 2405 panic("Failed to register MPTCP PM netlink family\n"); 2406 } 2407