1 // SPDX-License-Identifier: GPL-2.0 2 /* Multipath TCP 3 * 4 * Copyright (c) 2019, Intel Corporation. 5 */ 6 #define pr_fmt(fmt) "MPTCP: " fmt 7 8 #include <linux/rculist.h> 9 #include <linux/spinlock.h> 10 #include "protocol.h" 11 #include "mib.h" 12 13 #define ADD_ADDR_RETRANS_MAX 3 14 15 struct mptcp_pm_add_entry { 16 struct list_head list; 17 struct mptcp_addr_info addr; 18 u8 retrans_times; 19 bool timer_done; 20 struct timer_list add_timer; 21 struct mptcp_sock *sock; 22 struct rcu_head rcu; 23 }; 24 25 static DEFINE_SPINLOCK(mptcp_pm_list_lock); 26 static LIST_HEAD(mptcp_pm_list); 27 28 /* path manager helpers */ 29 30 /* if sk is ipv4 or ipv6_only allows only same-family local and remote addresses, 31 * otherwise allow any matching local/remote pair 32 */ 33 bool mptcp_pm_addr_families_match(const struct sock *sk, 34 const struct mptcp_addr_info *loc, 35 const struct mptcp_addr_info *rem) 36 { 37 bool mptcp_is_v4 = sk->sk_family == AF_INET; 38 39 #if IS_ENABLED(CONFIG_MPTCP_IPV6) 40 bool loc_is_v4 = loc->family == AF_INET || ipv6_addr_v4mapped(&loc->addr6); 41 bool rem_is_v4 = rem->family == AF_INET || ipv6_addr_v4mapped(&rem->addr6); 42 43 if (mptcp_is_v4) 44 return loc_is_v4 && rem_is_v4; 45 46 if (ipv6_only_sock(sk)) 47 return !loc_is_v4 && !rem_is_v4; 48 49 return loc_is_v4 == rem_is_v4; 50 #else 51 return mptcp_is_v4 && loc->family == AF_INET && rem->family == AF_INET; 52 #endif 53 } 54 55 bool mptcp_addresses_equal(const struct mptcp_addr_info *a, 56 const struct mptcp_addr_info *b, bool use_port) 57 { 58 bool addr_equals = false; 59 60 if (a->family == b->family) { 61 if (a->family == AF_INET) 62 addr_equals = a->addr.s_addr == b->addr.s_addr; 63 #if IS_ENABLED(CONFIG_MPTCP_IPV6) 64 else 65 addr_equals = ipv6_addr_equal(&a->addr6, &b->addr6); 66 } else if (a->family == AF_INET) { 67 if (ipv6_addr_v4mapped(&b->addr6)) 68 addr_equals = a->addr.s_addr == b->addr6.s6_addr32[3]; 69 } else if (b->family == AF_INET) { 70 if (ipv6_addr_v4mapped(&a->addr6)) 71 addr_equals = a->addr6.s6_addr32[3] == b->addr.s_addr; 72 #endif 73 } 74 75 if (!addr_equals) 76 return false; 77 if (!use_port) 78 return true; 79 80 return a->port == b->port; 81 } 82 83 void mptcp_local_address(const struct sock_common *skc, 84 struct mptcp_addr_info *addr) 85 { 86 addr->family = skc->skc_family; 87 addr->port = htons(skc->skc_num); 88 if (addr->family == AF_INET) 89 addr->addr.s_addr = skc->skc_rcv_saddr; 90 #if IS_ENABLED(CONFIG_MPTCP_IPV6) 91 else if (addr->family == AF_INET6) 92 addr->addr6 = skc->skc_v6_rcv_saddr; 93 #endif 94 } 95 96 void mptcp_remote_address(const struct sock_common *skc, 97 struct mptcp_addr_info *addr) 98 { 99 addr->family = skc->skc_family; 100 addr->port = skc->skc_dport; 101 if (addr->family == AF_INET) 102 addr->addr.s_addr = skc->skc_daddr; 103 #if IS_ENABLED(CONFIG_MPTCP_IPV6) 104 else if (addr->family == AF_INET6) 105 addr->addr6 = skc->skc_v6_daddr; 106 #endif 107 } 108 109 static bool mptcp_pm_is_init_remote_addr(struct mptcp_sock *msk, 110 const struct mptcp_addr_info *remote) 111 { 112 struct mptcp_addr_info mpc_remote; 113 114 mptcp_remote_address((struct sock_common *)msk, &mpc_remote); 115 return mptcp_addresses_equal(&mpc_remote, remote, remote->port); 116 } 117 118 bool mptcp_lookup_subflow_by_saddr(const struct list_head *list, 119 const struct mptcp_addr_info *saddr) 120 { 121 struct mptcp_subflow_context *subflow; 122 struct mptcp_addr_info cur; 123 struct sock_common *skc; 124 125 list_for_each_entry(subflow, list, node) { 126 skc = (struct sock_common *)mptcp_subflow_tcp_sock(subflow); 127 128 mptcp_local_address(skc, &cur); 129 if (mptcp_addresses_equal(&cur, saddr, saddr->port)) 130 return true; 131 } 132 133 return false; 134 } 135 136 static struct mptcp_pm_add_entry * 137 mptcp_lookup_anno_list_by_saddr(const struct mptcp_sock *msk, 138 const struct mptcp_addr_info *addr) 139 { 140 struct mptcp_pm_add_entry *entry; 141 142 lockdep_assert_held(&msk->pm.lock); 143 144 list_for_each_entry(entry, &msk->pm.anno_list, list) { 145 if (mptcp_addresses_equal(&entry->addr, addr, true)) 146 return entry; 147 } 148 149 return NULL; 150 } 151 152 bool mptcp_remove_anno_list_by_saddr(struct mptcp_sock *msk, 153 const struct mptcp_addr_info *addr) 154 { 155 struct mptcp_pm_add_entry *entry; 156 bool ret; 157 158 entry = mptcp_pm_del_add_timer(msk, addr, false); 159 ret = entry; 160 kfree_rcu(entry, rcu); 161 162 return ret; 163 } 164 165 bool mptcp_pm_sport_in_anno_list(struct mptcp_sock *msk, const struct sock *sk) 166 { 167 struct mptcp_pm_add_entry *entry; 168 struct mptcp_addr_info saddr; 169 bool ret = false; 170 171 mptcp_local_address((struct sock_common *)sk, &saddr); 172 173 spin_lock_bh(&msk->pm.lock); 174 list_for_each_entry(entry, &msk->pm.anno_list, list) { 175 if (mptcp_addresses_equal(&entry->addr, &saddr, true)) { 176 ret = true; 177 goto out; 178 } 179 } 180 181 out: 182 spin_unlock_bh(&msk->pm.lock); 183 return ret; 184 } 185 186 static void __mptcp_pm_send_ack(struct mptcp_sock *msk, 187 struct mptcp_subflow_context *subflow, 188 bool prio, bool backup) 189 { 190 struct sock *ssk = mptcp_subflow_tcp_sock(subflow); 191 bool slow; 192 193 pr_debug("send ack for %s\n", 194 prio ? "mp_prio" : 195 (mptcp_pm_should_add_signal(msk) ? "add_addr" : "rm_addr")); 196 197 slow = lock_sock_fast(ssk); 198 if (prio) { 199 subflow->send_mp_prio = 1; 200 subflow->request_bkup = backup; 201 } 202 203 __mptcp_subflow_send_ack(ssk); 204 unlock_sock_fast(ssk, slow); 205 } 206 207 void mptcp_pm_send_ack(struct mptcp_sock *msk, 208 struct mptcp_subflow_context *subflow, 209 bool prio, bool backup) 210 { 211 spin_unlock_bh(&msk->pm.lock); 212 __mptcp_pm_send_ack(msk, subflow, prio, backup); 213 spin_lock_bh(&msk->pm.lock); 214 } 215 216 static bool subflow_in_rm_list(const struct mptcp_subflow_context *subflow, 217 const struct mptcp_rm_list *rm_list) 218 { 219 u8 i, id = subflow_get_local_id(subflow); 220 221 for (i = 0; i < rm_list->nr; i++) { 222 if (rm_list->ids[i] == id) 223 return true; 224 } 225 226 return false; 227 } 228 229 static void 230 mptcp_pm_addr_send_ack_avoid_list(struct mptcp_sock *msk, 231 const struct mptcp_rm_list *rm_list) 232 { 233 struct mptcp_subflow_context *subflow, *stale = NULL, *same_id = NULL; 234 235 msk_owned_by_me(msk); 236 lockdep_assert_held(&msk->pm.lock); 237 238 if (!mptcp_pm_should_add_signal(msk) && 239 !mptcp_pm_should_rm_signal(msk)) 240 return; 241 242 mptcp_for_each_subflow(msk, subflow) { 243 if (!__mptcp_subflow_active(subflow)) 244 continue; 245 246 if (unlikely(subflow->stale)) { 247 if (!stale) 248 stale = subflow; 249 } else if (unlikely(rm_list && 250 subflow_in_rm_list(subflow, rm_list))) { 251 if (!same_id) 252 same_id = subflow; 253 } else { 254 goto send_ack; 255 } 256 } 257 258 if (same_id) 259 subflow = same_id; 260 else if (stale) 261 subflow = stale; 262 else 263 return; 264 265 send_ack: 266 mptcp_pm_send_ack(msk, subflow, false, false); 267 } 268 269 void mptcp_pm_addr_send_ack(struct mptcp_sock *msk) 270 { 271 mptcp_pm_addr_send_ack_avoid_list(msk, NULL); 272 } 273 274 int mptcp_pm_mp_prio_send_ack(struct mptcp_sock *msk, 275 struct mptcp_addr_info *addr, 276 struct mptcp_addr_info *rem, 277 u8 bkup) 278 { 279 struct mptcp_subflow_context *subflow; 280 281 pr_debug("bkup=%d\n", bkup); 282 283 mptcp_for_each_subflow(msk, subflow) { 284 struct sock *ssk = mptcp_subflow_tcp_sock(subflow); 285 struct mptcp_addr_info local, remote; 286 287 if (!__mptcp_subflow_active(subflow)) 288 continue; 289 290 mptcp_local_address((struct sock_common *)ssk, &local); 291 if (!mptcp_addresses_equal(&local, addr, addr->port)) 292 continue; 293 294 if (rem && rem->family != AF_UNSPEC) { 295 mptcp_remote_address((struct sock_common *)ssk, &remote); 296 if (!mptcp_addresses_equal(&remote, rem, rem->port)) 297 continue; 298 } 299 300 __mptcp_pm_send_ack(msk, subflow, true, bkup); 301 return 0; 302 } 303 304 return -EINVAL; 305 } 306 307 static unsigned int mptcp_adjust_add_addr_timeout(struct mptcp_sock *msk) 308 { 309 const struct net *net = sock_net((struct sock *)msk); 310 unsigned int rto = mptcp_get_add_addr_timeout(net); 311 struct mptcp_subflow_context *subflow; 312 unsigned int max = 0, max_stale = 0; 313 314 if (!rto) 315 return 0; 316 317 mptcp_for_each_subflow(msk, subflow) { 318 struct sock *ssk = mptcp_subflow_tcp_sock(subflow); 319 struct inet_connection_sock *icsk = inet_csk(ssk); 320 321 if (!__mptcp_subflow_active(subflow)) 322 continue; 323 324 if (unlikely(subflow->stale)) { 325 if (icsk->icsk_rto > max_stale) 326 max_stale = icsk->icsk_rto; 327 } else if (icsk->icsk_rto > max) { 328 max = icsk->icsk_rto; 329 } 330 } 331 332 if (max) 333 return min(max, rto); 334 335 if (max_stale) 336 return min(max_stale, rto); 337 338 return rto; 339 } 340 341 static void mptcp_pm_add_timer(struct timer_list *timer) 342 { 343 struct mptcp_pm_add_entry *entry = timer_container_of(entry, timer, 344 add_timer); 345 struct mptcp_sock *msk = entry->sock; 346 struct sock *sk = (struct sock *)msk; 347 unsigned int timeout = 0; 348 349 pr_debug("msk=%p\n", msk); 350 351 bh_lock_sock(sk); 352 if (unlikely(inet_sk_state_load(sk) == TCP_CLOSE)) 353 goto out; 354 355 if (sock_owned_by_user(sk)) { 356 /* Try again later. */ 357 timeout = HZ / 20; 358 goto out; 359 } 360 361 timeout = mptcp_adjust_add_addr_timeout(msk); 362 if (!timeout || mptcp_pm_should_add_signal_addr(msk)) 363 goto out; 364 365 spin_lock_bh(&msk->pm.lock); 366 367 /* The cancel path (mptcp_pm_del_add_timer()) can race with this 368 * callback. Once cancel updates retrans_times to MAX, suppress further 369 * retransmissions here. If this callback acquires pm.lock first, one 370 * final transmit attempt is still possible. 371 */ 372 if (entry->retrans_times < ADD_ADDR_RETRANS_MAX && 373 !mptcp_pm_should_add_signal_addr(msk)) { 374 pr_debug("retransmit ADD_ADDR id=%d\n", entry->addr.id); 375 mptcp_pm_announce_addr(msk, &entry->addr, false); 376 mptcp_pm_add_addr_send_ack(msk); 377 entry->retrans_times++; 378 } 379 380 if (entry->retrans_times < ADD_ADDR_RETRANS_MAX) 381 timeout <<= entry->retrans_times; 382 else 383 timeout = 0; 384 385 spin_unlock_bh(&msk->pm.lock); 386 387 if (entry->retrans_times == ADD_ADDR_RETRANS_MAX) 388 mptcp_pm_subflow_established(msk); 389 390 out: 391 if (timeout) 392 sk_reset_timer(sk, timer, jiffies + timeout); 393 else 394 /* if sock_put calls sk_free: avoid waiting for this timer */ 395 entry->timer_done = true; 396 bh_unlock_sock(sk); 397 sock_put(sk); 398 } 399 400 struct mptcp_pm_add_entry * 401 mptcp_pm_del_add_timer(struct mptcp_sock *msk, 402 const struct mptcp_addr_info *addr, bool check_id) 403 { 404 struct mptcp_pm_add_entry *entry; 405 struct sock *sk = (struct sock *)msk; 406 bool stop_timer = false; 407 408 rcu_read_lock(); 409 410 spin_lock_bh(&msk->pm.lock); 411 entry = mptcp_lookup_anno_list_by_saddr(msk, addr); 412 if (entry && (!check_id || entry->addr.id == addr->id)) { 413 entry->retrans_times = ADD_ADDR_RETRANS_MAX; 414 stop_timer = true; 415 } 416 if (!check_id && entry) 417 list_del(&entry->list); 418 spin_unlock_bh(&msk->pm.lock); 419 420 /* Note: entry might have been removed by another thread. 421 * We hold rcu_read_lock() to ensure it is not freed under us. 422 */ 423 if (stop_timer) { 424 if (check_id) 425 sk_stop_timer(sk, &entry->add_timer); 426 else 427 sk_stop_timer_sync(sk, &entry->add_timer); 428 } 429 430 rcu_read_unlock(); 431 return entry; 432 } 433 434 bool mptcp_pm_alloc_anno_list(struct mptcp_sock *msk, 435 const struct mptcp_addr_info *addr) 436 { 437 struct mptcp_pm_add_entry *add_entry = NULL; 438 struct sock *sk = (struct sock *)msk; 439 unsigned int timeout; 440 441 lockdep_assert_held(&msk->pm.lock); 442 443 add_entry = mptcp_lookup_anno_list_by_saddr(msk, addr); 444 445 if (add_entry) { 446 if (WARN_ON_ONCE(mptcp_pm_is_kernel(msk))) 447 return false; 448 449 goto reset_timer; 450 } 451 452 add_entry = kmalloc_obj(*add_entry, GFP_ATOMIC); 453 if (!add_entry) 454 return false; 455 456 list_add(&add_entry->list, &msk->pm.anno_list); 457 458 add_entry->addr = *addr; 459 add_entry->sock = msk; 460 add_entry->retrans_times = 0; 461 462 timer_setup(&add_entry->add_timer, mptcp_pm_add_timer, 0); 463 reset_timer: 464 add_entry->timer_done = false; 465 timeout = mptcp_adjust_add_addr_timeout(msk); 466 if (timeout) 467 sk_reset_timer(sk, &add_entry->add_timer, jiffies + timeout); 468 469 return true; 470 } 471 472 static void mptcp_pm_free_anno_list(struct mptcp_sock *msk) 473 { 474 struct mptcp_pm_add_entry *entry, *tmp; 475 struct sock *sk = (struct sock *)msk; 476 LIST_HEAD(free_list); 477 478 pr_debug("msk=%p\n", msk); 479 480 spin_lock_bh(&msk->pm.lock); 481 list_splice_init(&msk->pm.anno_list, &free_list); 482 spin_unlock_bh(&msk->pm.lock); 483 484 list_for_each_entry_safe(entry, tmp, &free_list, list) { 485 if (!entry->timer_done) 486 sk_stop_timer_sync(sk, &entry->add_timer); 487 kfree_rcu(entry, rcu); 488 } 489 } 490 491 /* path manager command handlers */ 492 493 int mptcp_pm_announce_addr(struct mptcp_sock *msk, 494 const struct mptcp_addr_info *addr, 495 bool echo) 496 { 497 u8 add_addr = READ_ONCE(msk->pm.addr_signal); 498 499 pr_debug("msk=%p, local_id=%d, echo=%d\n", msk, addr->id, echo); 500 501 lockdep_assert_held(&msk->pm.lock); 502 503 if (add_addr & 504 (echo ? BIT(MPTCP_ADD_ADDR_ECHO) : BIT(MPTCP_ADD_ADDR_SIGNAL))) { 505 MPTCP_INC_STATS(sock_net((struct sock *)msk), 506 echo ? MPTCP_MIB_ECHOADDTXDROP : MPTCP_MIB_ADDADDRTXDROP); 507 return -EINVAL; 508 } 509 510 if (echo) { 511 msk->pm.remote = *addr; 512 add_addr |= BIT(MPTCP_ADD_ADDR_ECHO); 513 } else { 514 msk->pm.local = *addr; 515 add_addr |= BIT(MPTCP_ADD_ADDR_SIGNAL); 516 } 517 WRITE_ONCE(msk->pm.addr_signal, add_addr); 518 return 0; 519 } 520 521 int mptcp_pm_remove_addr(struct mptcp_sock *msk, const struct mptcp_rm_list *rm_list) 522 { 523 u8 rm_addr = READ_ONCE(msk->pm.addr_signal); 524 525 pr_debug("msk=%p, rm_list_nr=%d\n", msk, rm_list->nr); 526 527 if (rm_addr) { 528 MPTCP_ADD_STATS(sock_net((struct sock *)msk), 529 MPTCP_MIB_RMADDRTXDROP, rm_list->nr); 530 return -EINVAL; 531 } 532 533 msk->pm.rm_list_tx = *rm_list; 534 rm_addr |= BIT(MPTCP_RM_ADDR_SIGNAL); 535 WRITE_ONCE(msk->pm.addr_signal, rm_addr); 536 mptcp_pm_addr_send_ack_avoid_list(msk, rm_list); 537 return 0; 538 } 539 540 /* path manager event handlers */ 541 542 void mptcp_pm_new_connection(struct mptcp_sock *msk, const struct sock *ssk, int server_side) 543 { 544 struct mptcp_pm_data *pm = &msk->pm; 545 546 pr_debug("msk=%p, token=%u side=%d\n", msk, READ_ONCE(msk->token), server_side); 547 548 WRITE_ONCE(pm->server_side, server_side); 549 mptcp_event(MPTCP_EVENT_CREATED, msk, ssk, GFP_ATOMIC); 550 } 551 552 bool mptcp_pm_allow_new_subflow(struct mptcp_sock *msk) 553 { 554 struct mptcp_pm_data *pm = &msk->pm; 555 unsigned int limit_extra_subflows; 556 int ret = 0; 557 558 if (mptcp_pm_is_userspace(msk)) { 559 if (mptcp_userspace_pm_active(msk)) { 560 spin_lock_bh(&pm->lock); 561 pm->extra_subflows++; 562 spin_unlock_bh(&pm->lock); 563 return true; 564 } 565 return false; 566 } 567 568 limit_extra_subflows = mptcp_pm_get_limit_extra_subflows(msk); 569 570 pr_debug("msk=%p subflows=%d max=%d allow=%d\n", msk, 571 pm->extra_subflows, limit_extra_subflows, 572 READ_ONCE(pm->accept_subflow)); 573 574 /* try to avoid acquiring the lock below */ 575 if (!READ_ONCE(pm->accept_subflow)) 576 return false; 577 578 spin_lock_bh(&pm->lock); 579 if (READ_ONCE(pm->accept_subflow)) { 580 ret = pm->extra_subflows < limit_extra_subflows; 581 if (ret && ++pm->extra_subflows == limit_extra_subflows) 582 WRITE_ONCE(pm->accept_subflow, false); 583 } 584 spin_unlock_bh(&pm->lock); 585 586 return ret; 587 } 588 589 /* return true if the new status bit is currently cleared, that is, this event 590 * can be server, eventually by an already scheduled work 591 */ 592 static bool mptcp_pm_schedule_work(struct mptcp_sock *msk, 593 enum mptcp_pm_status new_status) 594 { 595 pr_debug("msk=%p status=%x new=%lx\n", msk, msk->pm.status, 596 BIT(new_status)); 597 if (msk->pm.status & BIT(new_status)) 598 return false; 599 600 msk->pm.status |= BIT(new_status); 601 mptcp_schedule_work((struct sock *)msk); 602 return true; 603 } 604 605 void mptcp_pm_fully_established(struct mptcp_sock *msk, const struct sock *ssk) 606 { 607 struct mptcp_pm_data *pm = &msk->pm; 608 bool announce = false; 609 610 pr_debug("msk=%p\n", msk); 611 612 spin_lock_bh(&pm->lock); 613 614 /* mptcp_pm_fully_established() can be invoked by multiple 615 * racing paths - accept() and check_fully_established() 616 * be sure to serve this event only once. 617 */ 618 if (READ_ONCE(pm->work_pending) && 619 !(pm->status & BIT(MPTCP_PM_ALREADY_ESTABLISHED))) 620 mptcp_pm_schedule_work(msk, MPTCP_PM_ESTABLISHED); 621 622 if ((pm->status & BIT(MPTCP_PM_ALREADY_ESTABLISHED)) == 0) 623 announce = true; 624 625 pm->status |= BIT(MPTCP_PM_ALREADY_ESTABLISHED); 626 spin_unlock_bh(&pm->lock); 627 628 if (announce) 629 mptcp_event(MPTCP_EVENT_ESTABLISHED, msk, ssk, GFP_ATOMIC); 630 } 631 632 void mptcp_pm_connection_closed(struct mptcp_sock *msk) 633 { 634 pr_debug("msk=%p\n", msk); 635 636 if (msk->token) 637 mptcp_event(MPTCP_EVENT_CLOSED, msk, NULL, GFP_KERNEL); 638 } 639 640 void mptcp_pm_subflow_established(struct mptcp_sock *msk) 641 { 642 struct mptcp_pm_data *pm = &msk->pm; 643 644 pr_debug("msk=%p\n", msk); 645 646 if (!READ_ONCE(pm->work_pending)) 647 return; 648 649 spin_lock_bh(&pm->lock); 650 651 if (READ_ONCE(pm->work_pending)) 652 mptcp_pm_schedule_work(msk, MPTCP_PM_SUBFLOW_ESTABLISHED); 653 654 spin_unlock_bh(&pm->lock); 655 } 656 657 void mptcp_pm_subflow_check_next(struct mptcp_sock *msk, 658 const struct mptcp_subflow_context *subflow) 659 { 660 struct sock *sk = (struct sock *)msk; 661 struct mptcp_pm_data *pm = &msk->pm; 662 bool update_subflows; 663 664 update_subflows = subflow->request_join || subflow->mp_join; 665 if (mptcp_pm_is_userspace(msk)) { 666 if (update_subflows) { 667 spin_lock_bh(&pm->lock); 668 pm->extra_subflows--; 669 spin_unlock_bh(&pm->lock); 670 } 671 return; 672 } 673 674 if (!READ_ONCE(pm->work_pending) && !update_subflows) 675 return; 676 677 spin_lock_bh(&pm->lock); 678 if (update_subflows) 679 __mptcp_pm_close_subflow(msk); 680 681 /* Even if this subflow is not really established, tell the PM to try 682 * to pick the next ones, if possible. 683 */ 684 if (mptcp_is_fully_established(sk) && 685 mptcp_pm_nl_check_work_pending(msk)) 686 mptcp_pm_schedule_work(msk, MPTCP_PM_SUBFLOW_ESTABLISHED); 687 688 spin_unlock_bh(&pm->lock); 689 } 690 691 void mptcp_pm_add_addr_received(const struct sock *ssk, 692 const struct mptcp_addr_info *addr) 693 { 694 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk); 695 struct mptcp_sock *msk = mptcp_sk(subflow->conn); 696 struct mptcp_pm_data *pm = &msk->pm; 697 698 pr_debug("msk=%p remote_id=%d accept=%d\n", msk, addr->id, 699 READ_ONCE(pm->accept_addr)); 700 701 mptcp_event_addr_announced(ssk, addr); 702 703 spin_lock_bh(&pm->lock); 704 705 if (mptcp_pm_is_userspace(msk)) { 706 if (mptcp_userspace_pm_active(msk)) { 707 mptcp_pm_announce_addr(msk, addr, true); 708 mptcp_pm_add_addr_send_ack(msk); 709 } else { 710 __MPTCP_INC_STATS(sock_net((struct sock *)msk), MPTCP_MIB_ADDADDRDROP); 711 } 712 /* - id0 should not have a different address 713 * - special case for C-flag: linked to fill_local_addresses_vec() 714 */ 715 } else if ((addr->id == 0 && !mptcp_pm_is_init_remote_addr(msk, addr)) || 716 (addr->id > 0 && !READ_ONCE(pm->accept_addr) && 717 !mptcp_pm_add_addr_c_flag_case(msk))) { 718 mptcp_pm_announce_addr(msk, addr, true); 719 mptcp_pm_add_addr_send_ack(msk); 720 } else if (mptcp_pm_schedule_work(msk, MPTCP_PM_ADD_ADDR_RECEIVED)) { 721 pm->remote = *addr; 722 } else { 723 __MPTCP_INC_STATS(sock_net((struct sock *)msk), MPTCP_MIB_ADDADDRDROP); 724 } 725 726 spin_unlock_bh(&pm->lock); 727 } 728 729 void mptcp_pm_add_addr_echoed(struct mptcp_sock *msk, 730 const struct mptcp_addr_info *addr) 731 { 732 struct mptcp_pm_data *pm = &msk->pm; 733 734 pr_debug("msk=%p\n", msk); 735 736 if (!READ_ONCE(pm->work_pending)) 737 return; 738 739 spin_lock_bh(&pm->lock); 740 741 if (mptcp_lookup_anno_list_by_saddr(msk, addr) && READ_ONCE(pm->work_pending)) 742 mptcp_pm_schedule_work(msk, MPTCP_PM_SUBFLOW_ESTABLISHED); 743 744 spin_unlock_bh(&pm->lock); 745 } 746 747 void mptcp_pm_add_addr_send_ack(struct mptcp_sock *msk) 748 { 749 if (!mptcp_pm_should_add_signal(msk)) 750 return; 751 752 mptcp_pm_schedule_work(msk, MPTCP_PM_ADD_ADDR_SEND_ACK); 753 } 754 755 static void mptcp_pm_rm_addr_or_subflow(struct mptcp_sock *msk, 756 const struct mptcp_rm_list *rm_list, 757 enum linux_mptcp_mib_field rm_type) 758 { 759 struct mptcp_subflow_context *subflow, *tmp; 760 struct sock *sk = (struct sock *)msk; 761 u8 i; 762 763 pr_debug("%s rm_list_nr %d\n", 764 rm_type == MPTCP_MIB_RMADDR ? "address" : "subflow", rm_list->nr); 765 766 msk_owned_by_me(msk); 767 768 if (sk->sk_state == TCP_LISTEN) 769 return; 770 771 if (!rm_list->nr) 772 return; 773 774 if (list_empty(&msk->conn_list)) 775 return; 776 777 for (i = 0; i < rm_list->nr; i++) { 778 u8 rm_id = rm_list->ids[i]; 779 bool removed = false; 780 781 mptcp_for_each_subflow_safe(msk, subflow, tmp) { 782 struct sock *ssk = mptcp_subflow_tcp_sock(subflow); 783 u8 remote_id = READ_ONCE(subflow->remote_id); 784 int how = RCV_SHUTDOWN | SEND_SHUTDOWN; 785 u8 id = subflow_get_local_id(subflow); 786 787 if ((1 << inet_sk_state_load(ssk)) & 788 (TCPF_FIN_WAIT1 | TCPF_FIN_WAIT2 | TCPF_CLOSING | TCPF_CLOSE)) 789 continue; 790 if (rm_type == MPTCP_MIB_RMADDR && remote_id != rm_id) 791 continue; 792 if (rm_type == MPTCP_MIB_RMSUBFLOW && id != rm_id) 793 continue; 794 795 pr_debug(" -> %s rm_list_ids[%d]=%u local_id=%u remote_id=%u mpc_id=%u\n", 796 rm_type == MPTCP_MIB_RMADDR ? "address" : "subflow", 797 i, rm_id, id, remote_id, msk->mpc_endpoint_id); 798 spin_unlock_bh(&msk->pm.lock); 799 mptcp_subflow_shutdown(sk, ssk, how); 800 removed |= subflow->request_join; 801 802 /* the following takes care of updating the subflows counter */ 803 mptcp_close_ssk(sk, ssk, subflow); 804 spin_lock_bh(&msk->pm.lock); 805 806 if (rm_type == MPTCP_MIB_RMSUBFLOW) 807 __MPTCP_INC_STATS(sock_net(sk), rm_type); 808 } 809 810 if (rm_type == MPTCP_MIB_RMADDR) { 811 __MPTCP_INC_STATS(sock_net(sk), rm_type); 812 if (removed && mptcp_pm_is_kernel(msk)) 813 mptcp_pm_nl_rm_addr(msk, rm_id); 814 } 815 } 816 } 817 818 static void mptcp_pm_rm_addr_recv(struct mptcp_sock *msk) 819 { 820 mptcp_pm_rm_addr_or_subflow(msk, &msk->pm.rm_list_rx, MPTCP_MIB_RMADDR); 821 } 822 823 void mptcp_pm_rm_subflow(struct mptcp_sock *msk, 824 const struct mptcp_rm_list *rm_list) 825 { 826 mptcp_pm_rm_addr_or_subflow(msk, rm_list, MPTCP_MIB_RMSUBFLOW); 827 } 828 829 void mptcp_pm_rm_addr_received(struct mptcp_sock *msk, 830 const struct mptcp_rm_list *rm_list) 831 { 832 struct mptcp_pm_data *pm = &msk->pm; 833 u8 i; 834 835 pr_debug("msk=%p remote_ids_nr=%d\n", msk, rm_list->nr); 836 837 for (i = 0; i < rm_list->nr; i++) 838 mptcp_event_addr_removed(msk, rm_list->ids[i]); 839 840 spin_lock_bh(&pm->lock); 841 if (mptcp_pm_schedule_work(msk, MPTCP_PM_RM_ADDR_RECEIVED)) 842 pm->rm_list_rx = *rm_list; 843 else 844 __MPTCP_INC_STATS(sock_net((struct sock *)msk), MPTCP_MIB_RMADDRDROP); 845 spin_unlock_bh(&pm->lock); 846 } 847 848 void mptcp_pm_mp_prio_received(struct sock *ssk, u8 bkup) 849 { 850 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk); 851 struct sock *sk = subflow->conn; 852 struct mptcp_sock *msk; 853 854 pr_debug("subflow->backup=%d, bkup=%d\n", subflow->backup, bkup); 855 msk = mptcp_sk(sk); 856 if (subflow->backup != bkup) 857 subflow->backup = bkup; 858 859 mptcp_event(MPTCP_EVENT_SUB_PRIORITY, msk, ssk, GFP_ATOMIC); 860 } 861 862 void mptcp_pm_mp_fail_received(struct sock *sk, u64 fail_seq) 863 { 864 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk); 865 struct mptcp_sock *msk = mptcp_sk(subflow->conn); 866 867 pr_debug("fail_seq=%llu\n", fail_seq); 868 869 /* After accepting the fail, we can't create any other subflows */ 870 spin_lock_bh(&msk->fallback_lock); 871 if (!msk->allow_infinite_fallback) { 872 spin_unlock_bh(&msk->fallback_lock); 873 return; 874 } 875 msk->allow_subflows = false; 876 spin_unlock_bh(&msk->fallback_lock); 877 878 if (!subflow->fail_tout) { 879 pr_debug("send MP_FAIL response and infinite map\n"); 880 881 subflow->send_mp_fail = 1; 882 subflow->send_infinite_map = 1; 883 tcp_send_ack(sk); 884 } else { 885 pr_debug("MP_FAIL response received\n"); 886 WRITE_ONCE(subflow->fail_tout, 0); 887 } 888 } 889 890 bool mptcp_pm_add_addr_signal(struct mptcp_sock *msk, unsigned int opt_size, 891 unsigned int remaining, 892 struct mptcp_addr_info *addr, bool *echo) 893 { 894 bool skip_add_addr = false; 895 int ret = false; 896 u8 add_addr; 897 u8 family; 898 bool port; 899 900 spin_lock_bh(&msk->pm.lock); 901 902 /* double check after the lock is acquired */ 903 if (!mptcp_pm_should_add_signal(msk)) 904 goto out_unlock; 905 906 /* always drop every other options for pure ack ADD_ADDR; this is a 907 * plain dup-ack from TCP perspective. The other MPTCP-relevant info, 908 * if any, will be carried by the 'original' TCP ack 909 */ 910 remaining += opt_size; 911 912 *echo = mptcp_pm_should_add_signal_echo(msk); 913 if (*echo) { 914 *addr = msk->pm.remote; 915 add_addr = msk->pm.addr_signal & ~BIT(MPTCP_ADD_ADDR_ECHO); 916 port = !!msk->pm.remote.port; 917 family = msk->pm.remote.family; 918 } else { 919 *addr = msk->pm.local; 920 add_addr = msk->pm.addr_signal & ~BIT(MPTCP_ADD_ADDR_SIGNAL); 921 port = !!msk->pm.local.port; 922 family = msk->pm.local.family; 923 } 924 925 if (remaining < mptcp_add_addr_len(family, *echo, port)) { 926 struct net *net = sock_net((struct sock *)msk); 927 928 if (*echo) { 929 MPTCP_INC_STATS(net, MPTCP_MIB_ECHOADDTXDROP); 930 } else { 931 skip_add_addr = true; 932 MPTCP_INC_STATS(net, MPTCP_MIB_ADDADDRTXDROP); 933 } 934 goto drop_signal_mark; 935 } 936 937 ret = true; 938 939 drop_signal_mark: 940 WRITE_ONCE(msk->pm.addr_signal, add_addr); 941 942 out_unlock: 943 spin_unlock_bh(&msk->pm.lock); 944 945 /* On pure-ACK option-space exhaustion, stop retrying this ADD_ADDR: 946 * clear the signal bit, cancel the matching retransmission timer, and 947 * let the PM state machine progress. 948 */ 949 if (skip_add_addr) { 950 mptcp_pm_del_add_timer(msk, addr, true); 951 mptcp_pm_subflow_established(msk); 952 } 953 return ret; 954 } 955 956 bool mptcp_pm_rm_addr_signal(struct mptcp_sock *msk, unsigned int remaining, 957 struct mptcp_rm_list *rm_list) 958 { 959 int ret = false, len; 960 u8 rm_addr; 961 962 spin_lock_bh(&msk->pm.lock); 963 964 /* double check after the lock is acquired */ 965 if (!mptcp_pm_should_rm_signal(msk)) 966 goto out_unlock; 967 968 rm_addr = msk->pm.addr_signal & ~BIT(MPTCP_RM_ADDR_SIGNAL); 969 len = mptcp_rm_addr_len(&msk->pm.rm_list_tx); 970 if (len < 0) { 971 WRITE_ONCE(msk->pm.addr_signal, rm_addr); 972 goto out_unlock; 973 } 974 if (remaining < len) 975 goto out_unlock; 976 977 *rm_list = msk->pm.rm_list_tx; 978 WRITE_ONCE(msk->pm.addr_signal, rm_addr); 979 ret = true; 980 981 out_unlock: 982 spin_unlock_bh(&msk->pm.lock); 983 return ret; 984 } 985 986 int mptcp_pm_get_local_id(struct mptcp_sock *msk, struct sock_common *skc) 987 { 988 struct mptcp_pm_addr_entry skc_local = { 0 }; 989 struct mptcp_addr_info msk_local; 990 991 if (WARN_ON_ONCE(!msk)) 992 return -1; 993 994 /* The 0 ID mapping is defined by the first subflow, copied into the msk 995 * addr 996 */ 997 mptcp_local_address((struct sock_common *)msk, &msk_local); 998 mptcp_local_address((struct sock_common *)skc, &skc_local.addr); 999 if (mptcp_addresses_equal(&msk_local, &skc_local.addr, false)) 1000 return 0; 1001 1002 skc_local.addr.id = 0; 1003 skc_local.flags = MPTCP_PM_ADDR_FLAG_IMPLICIT; 1004 1005 if (mptcp_pm_is_userspace(msk)) 1006 return mptcp_userspace_pm_get_local_id(msk, &skc_local); 1007 return mptcp_pm_nl_get_local_id(msk, &skc_local); 1008 } 1009 1010 bool mptcp_pm_is_backup(struct mptcp_sock *msk, struct sock_common *skc) 1011 { 1012 struct mptcp_addr_info skc_local; 1013 1014 mptcp_local_address((struct sock_common *)skc, &skc_local); 1015 1016 if (mptcp_pm_is_userspace(msk)) 1017 return mptcp_userspace_pm_is_backup(msk, &skc_local); 1018 1019 return mptcp_pm_nl_is_backup(msk, &skc_local); 1020 } 1021 1022 static void mptcp_pm_subflows_chk_stale(const struct mptcp_sock *msk, struct sock *ssk) 1023 { 1024 struct mptcp_subflow_context *iter, *subflow = mptcp_subflow_ctx(ssk); 1025 struct sock *sk = (struct sock *)msk; 1026 unsigned int active_max_loss_cnt; 1027 struct net *net = sock_net(sk); 1028 unsigned int stale_loss_cnt; 1029 bool slow; 1030 1031 stale_loss_cnt = mptcp_stale_loss_cnt(net); 1032 if (subflow->stale || !stale_loss_cnt || subflow->stale_count <= stale_loss_cnt) 1033 return; 1034 1035 /* look for another available subflow not in loss state */ 1036 active_max_loss_cnt = max_t(int, stale_loss_cnt - 1, 1); 1037 mptcp_for_each_subflow(msk, iter) { 1038 if (iter != subflow && mptcp_subflow_active(iter) && 1039 iter->stale_count < active_max_loss_cnt) { 1040 /* we have some alternatives, try to mark this subflow as idle ...*/ 1041 slow = lock_sock_fast(ssk); 1042 if (!tcp_rtx_and_write_queues_empty(ssk)) { 1043 subflow->stale = 1; 1044 __mptcp_retransmit_pending_data(sk); 1045 MPTCP_INC_STATS(net, MPTCP_MIB_SUBFLOWSTALE); 1046 } 1047 unlock_sock_fast(ssk, slow); 1048 1049 /* always try to push the pending data regardless of re-injections: 1050 * we can possibly use backup subflows now, and subflow selection 1051 * is cheap under the msk socket lock 1052 */ 1053 __mptcp_push_pending(sk, 0); 1054 return; 1055 } 1056 } 1057 } 1058 1059 void mptcp_pm_subflow_chk_stale(const struct mptcp_sock *msk, struct sock *ssk) 1060 { 1061 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk); 1062 u32 rcv_tstamp = READ_ONCE(tcp_sk(ssk)->rcv_tstamp); 1063 1064 /* keep track of rtx periods with no progress */ 1065 if (!subflow->stale_count) { 1066 subflow->stale_rcv_tstamp = rcv_tstamp; 1067 subflow->stale_count++; 1068 } else if (subflow->stale_rcv_tstamp == rcv_tstamp) { 1069 if (subflow->stale_count < U8_MAX) 1070 subflow->stale_count++; 1071 mptcp_pm_subflows_chk_stale(msk, ssk); 1072 } else { 1073 subflow->stale_count = 0; 1074 mptcp_subflow_set_active(subflow); 1075 } 1076 } 1077 1078 void mptcp_pm_worker(struct mptcp_sock *msk) 1079 { 1080 struct mptcp_pm_data *pm = &msk->pm; 1081 1082 msk_owned_by_me(msk); 1083 1084 if (!(pm->status & MPTCP_PM_WORK_MASK)) 1085 return; 1086 1087 spin_lock_bh(&msk->pm.lock); 1088 1089 pr_debug("msk=%p status=%x\n", msk, pm->status); 1090 if (pm->status & BIT(MPTCP_PM_ADD_ADDR_SEND_ACK)) { 1091 pm->status &= ~BIT(MPTCP_PM_ADD_ADDR_SEND_ACK); 1092 mptcp_pm_addr_send_ack(msk); 1093 } 1094 if (pm->status & BIT(MPTCP_PM_RM_ADDR_RECEIVED)) { 1095 pm->status &= ~BIT(MPTCP_PM_RM_ADDR_RECEIVED); 1096 mptcp_pm_rm_addr_recv(msk); 1097 } 1098 __mptcp_pm_kernel_worker(msk); 1099 1100 spin_unlock_bh(&msk->pm.lock); 1101 } 1102 1103 void mptcp_pm_destroy(struct mptcp_sock *msk) 1104 { 1105 mptcp_pm_free_anno_list(msk); 1106 1107 if (mptcp_pm_is_userspace(msk)) 1108 mptcp_userspace_pm_free_local_addr_list(msk); 1109 } 1110 1111 void mptcp_pm_data_reset(struct mptcp_sock *msk) 1112 { 1113 u8 pm_type = mptcp_get_pm_type(sock_net((struct sock *)msk)); 1114 struct mptcp_pm_data *pm = &msk->pm; 1115 1116 memset(&pm->reset, 0, sizeof(pm->reset)); 1117 pm->rm_list_tx.nr = 0; 1118 pm->rm_list_rx.nr = 0; 1119 WRITE_ONCE(pm->pm_type, pm_type); 1120 1121 if (pm_type == MPTCP_PM_TYPE_KERNEL) { 1122 bool subflows_allowed = !!mptcp_pm_get_limit_extra_subflows(msk); 1123 1124 /* pm->work_pending must be only be set to 'true' when 1125 * pm->pm_type is set to MPTCP_PM_TYPE_KERNEL 1126 */ 1127 WRITE_ONCE(pm->work_pending, 1128 (!!mptcp_pm_get_endp_subflow_max(msk) && 1129 subflows_allowed) || 1130 !!mptcp_pm_get_endp_signal_max(msk)); 1131 WRITE_ONCE(pm->accept_addr, 1132 !!mptcp_pm_get_limit_add_addr_accepted(msk) && 1133 subflows_allowed); 1134 WRITE_ONCE(pm->accept_subflow, subflows_allowed); 1135 1136 bitmap_fill(pm->id_avail_bitmap, MPTCP_PM_MAX_ADDR_ID + 1); 1137 } 1138 } 1139 1140 void mptcp_pm_data_init(struct mptcp_sock *msk) 1141 { 1142 spin_lock_init(&msk->pm.lock); 1143 INIT_LIST_HEAD(&msk->pm.anno_list); 1144 INIT_LIST_HEAD(&msk->pm.userspace_pm_local_addr_list); 1145 mptcp_pm_data_reset(msk); 1146 } 1147 1148 void __init mptcp_pm_init(void) 1149 { 1150 mptcp_pm_kernel_register(); 1151 mptcp_pm_userspace_register(); 1152 mptcp_pm_nl_init(); 1153 } 1154 1155 /* Must be called with rcu read lock held */ 1156 struct mptcp_pm_ops *mptcp_pm_find(const char *name) 1157 { 1158 struct mptcp_pm_ops *pm_ops; 1159 1160 list_for_each_entry_rcu(pm_ops, &mptcp_pm_list, list) { 1161 if (!strcmp(pm_ops->name, name)) 1162 return pm_ops; 1163 } 1164 1165 return NULL; 1166 } 1167 1168 int mptcp_pm_validate(struct mptcp_pm_ops *pm_ops) 1169 { 1170 return 0; 1171 } 1172 1173 int mptcp_pm_register(struct mptcp_pm_ops *pm_ops) 1174 { 1175 int ret; 1176 1177 ret = mptcp_pm_validate(pm_ops); 1178 if (ret) 1179 return ret; 1180 1181 spin_lock(&mptcp_pm_list_lock); 1182 if (mptcp_pm_find(pm_ops->name)) { 1183 spin_unlock(&mptcp_pm_list_lock); 1184 return -EEXIST; 1185 } 1186 list_add_tail_rcu(&pm_ops->list, &mptcp_pm_list); 1187 spin_unlock(&mptcp_pm_list_lock); 1188 1189 pr_debug("%s registered\n", pm_ops->name); 1190 return 0; 1191 } 1192 1193 void mptcp_pm_unregister(struct mptcp_pm_ops *pm_ops) 1194 { 1195 /* skip unregistering the default path manager */ 1196 if (WARN_ON_ONCE(pm_ops == &mptcp_pm_kernel)) 1197 return; 1198 1199 spin_lock(&mptcp_pm_list_lock); 1200 list_del_rcu(&pm_ops->list); 1201 spin_unlock(&mptcp_pm_list_lock); 1202 } 1203 1204 /* Build string with list of available path manager values. 1205 * Similar to tcp_get_available_congestion_control() 1206 */ 1207 void mptcp_pm_get_available(char *buf, size_t maxlen) 1208 { 1209 struct mptcp_pm_ops *pm_ops; 1210 size_t offs = 0; 1211 1212 rcu_read_lock(); 1213 list_for_each_entry_rcu(pm_ops, &mptcp_pm_list, list) { 1214 offs += snprintf(buf + offs, maxlen - offs, "%s%s", 1215 offs == 0 ? "" : " ", pm_ops->name); 1216 1217 if (WARN_ON_ONCE(offs >= maxlen)) 1218 break; 1219 } 1220 rcu_read_unlock(); 1221 } 1222