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_addr { 16 struct list_head list; 17 struct mptcp_addr_info addr; 18 u8 retrans_times; 19 bool timer_done; 20 struct timer_list 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_pm_has_subflow_saddr(const struct mptcp_sock *msk, 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 mptcp_for_each_subflow(msk, subflow) { 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_addr * 137 mptcp_pm_announced_lookup(const struct mptcp_sock *msk, 138 const struct mptcp_addr_info *addr) 139 { 140 struct mptcp_pm_add_addr *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 static struct mptcp_pm_add_addr * 153 mptcp_pm_announced_del_timer(struct mptcp_sock *msk, 154 const struct mptcp_addr_info *addr, bool check_id) 155 { 156 struct sock *sk = (struct sock *)msk; 157 struct mptcp_pm_add_addr *entry; 158 bool stop_timer = false; 159 160 rcu_read_lock(); 161 162 spin_lock_bh(&msk->pm.lock); 163 entry = mptcp_pm_announced_lookup(msk, addr); 164 if (entry && (!check_id || entry->addr.id == addr->id)) { 165 entry->retrans_times = ADD_ADDR_RETRANS_MAX; 166 stop_timer = true; 167 } 168 if (!check_id && entry) 169 list_del(&entry->list); 170 spin_unlock_bh(&msk->pm.lock); 171 172 /* Note: entry might have been removed by another thread. 173 * We hold rcu_read_lock() to ensure it is not freed under us. 174 */ 175 if (stop_timer) { 176 if (check_id) 177 sk_stop_timer(sk, &entry->timer); 178 else 179 sk_stop_timer_sync(sk, &entry->timer); 180 } 181 182 rcu_read_unlock(); 183 return entry; 184 } 185 186 bool mptcp_pm_announced_remove(struct mptcp_sock *msk, 187 const struct mptcp_addr_info *addr) 188 { 189 struct mptcp_pm_add_addr *entry; 190 bool ret; 191 192 entry = mptcp_pm_announced_del_timer(msk, addr, false); 193 ret = entry; 194 kfree_rcu(entry, rcu); 195 196 return ret; 197 } 198 199 bool mptcp_pm_announced_has_ssk(struct mptcp_sock *msk, const struct sock *ssk) 200 { 201 struct mptcp_pm_add_addr *entry; 202 struct mptcp_addr_info saddr; 203 bool ret = false; 204 205 mptcp_local_address((struct sock_common *)ssk, &saddr); 206 207 spin_lock_bh(&msk->pm.lock); 208 list_for_each_entry(entry, &msk->pm.anno_list, list) { 209 if (mptcp_addresses_equal(&entry->addr, &saddr, true)) { 210 ret = true; 211 goto out; 212 } 213 } 214 215 out: 216 spin_unlock_bh(&msk->pm.lock); 217 return ret; 218 } 219 220 static void __mptcp_pm_send_ack(struct mptcp_sock *msk, 221 struct mptcp_subflow_context *subflow, 222 bool prio, bool backup) 223 { 224 struct sock *ssk = mptcp_subflow_tcp_sock(subflow); 225 bool slow; 226 227 pr_debug("send ack for %s\n", 228 prio ? "mp_prio" : 229 (mptcp_pm_should_add_signal(msk) ? "add_addr" : "rm_addr")); 230 231 slow = lock_sock_fast(ssk); 232 if (prio) { 233 subflow->send_mp_prio = 1; 234 subflow->request_bkup = backup; 235 } 236 237 __mptcp_subflow_send_ack(ssk); 238 unlock_sock_fast(ssk, slow); 239 } 240 241 void mptcp_pm_send_ack(struct mptcp_sock *msk, 242 struct mptcp_subflow_context *subflow, 243 bool prio, bool backup) 244 { 245 spin_unlock_bh(&msk->pm.lock); 246 __mptcp_pm_send_ack(msk, subflow, prio, backup); 247 spin_lock_bh(&msk->pm.lock); 248 } 249 250 static bool subflow_in_rm_list(const struct mptcp_subflow_context *subflow, 251 const struct mptcp_rm_list *rm_list) 252 { 253 u8 i, id = subflow_get_local_id(subflow); 254 255 for (i = 0; i < rm_list->nr; i++) { 256 if (rm_list->ids[i] == id) 257 return true; 258 } 259 260 return false; 261 } 262 263 static void mptcp_pm_add_addr_send_ack(struct mptcp_sock *msk); 264 static void 265 mptcp_pm_addr_send_ack_avoid_list(struct mptcp_sock *msk, 266 const struct mptcp_rm_list *rm_list) 267 { 268 struct mptcp_subflow_context *subflow, *stale = NULL, *same_id = NULL; 269 270 msk_owned_by_me(msk); 271 lockdep_assert_held(&msk->pm.lock); 272 273 if (!mptcp_pm_should_add_signal(msk) && 274 !mptcp_pm_should_rm_signal(msk)) 275 return; 276 277 mptcp_for_each_subflow(msk, subflow) { 278 if (!__mptcp_subflow_active(subflow)) 279 continue; 280 281 if (unlikely(subflow->stale)) { 282 if (!stale) 283 stale = subflow; 284 } else if (unlikely(rm_list && 285 subflow_in_rm_list(subflow, rm_list))) { 286 if (!same_id) 287 same_id = subflow; 288 } else { 289 goto send_ack; 290 } 291 } 292 293 if (same_id) 294 subflow = same_id; 295 else if (stale) 296 subflow = stale; 297 else 298 return; 299 300 send_ack: 301 mptcp_pm_send_ack(msk, subflow, false, false); 302 } 303 304 void mptcp_pm_addr_send_ack(struct mptcp_sock *msk) 305 { 306 mptcp_pm_addr_send_ack_avoid_list(msk, NULL); 307 } 308 309 int mptcp_pm_mp_prio_send_ack(struct mptcp_sock *msk, 310 struct mptcp_addr_info *addr, 311 struct mptcp_addr_info *rem, 312 u8 bkup) 313 { 314 struct mptcp_subflow_context *subflow; 315 316 pr_debug("bkup=%d\n", bkup); 317 318 mptcp_for_each_subflow(msk, subflow) { 319 struct sock *ssk = mptcp_subflow_tcp_sock(subflow); 320 struct mptcp_addr_info local, remote; 321 322 if (!__mptcp_subflow_active(subflow)) 323 continue; 324 325 mptcp_local_address((struct sock_common *)ssk, &local); 326 if (!mptcp_addresses_equal(&local, addr, addr->port)) 327 continue; 328 329 if (rem && rem->family != AF_UNSPEC) { 330 mptcp_remote_address((struct sock_common *)ssk, &remote); 331 if (!mptcp_addresses_equal(&remote, rem, rem->port)) 332 continue; 333 } 334 335 __mptcp_pm_send_ack(msk, subflow, true, bkup); 336 return 0; 337 } 338 339 return -EINVAL; 340 } 341 342 static unsigned int mptcp_adjust_add_addr_timeout(struct mptcp_sock *msk) 343 { 344 const struct net *net = sock_net((struct sock *)msk); 345 unsigned int rto = mptcp_get_add_addr_timeout(net); 346 struct mptcp_subflow_context *subflow; 347 unsigned int max = 0, max_stale = 0; 348 349 if (!rto) 350 return 0; 351 352 mptcp_for_each_subflow(msk, subflow) { 353 struct sock *ssk = mptcp_subflow_tcp_sock(subflow); 354 struct inet_connection_sock *icsk = inet_csk(ssk); 355 356 if (!__mptcp_subflow_active(subflow)) 357 continue; 358 359 if (unlikely(subflow->stale)) { 360 if (icsk->icsk_rto > max_stale) 361 max_stale = icsk->icsk_rto; 362 } else if (icsk->icsk_rto > max) { 363 max = icsk->icsk_rto; 364 } 365 } 366 367 if (max) 368 return min(max, rto); 369 370 if (max_stale) 371 return min(max_stale, rto); 372 373 return rto; 374 } 375 376 static void mptcp_pm_add_addr_timer(struct timer_list *timer) 377 { 378 struct mptcp_pm_add_addr *entry = timer_container_of(entry, timer, 379 timer); 380 struct mptcp_sock *msk = entry->sock; 381 struct sock *sk = (struct sock *)msk; 382 unsigned int timeout = 0; 383 384 pr_debug("msk=%p\n", msk); 385 386 bh_lock_sock(sk); 387 if (unlikely(inet_sk_state_load(sk) == TCP_CLOSE)) 388 goto out; 389 390 if (sock_owned_by_user(sk)) { 391 /* Try again later. */ 392 timeout = HZ / 20; 393 goto out; 394 } 395 396 timeout = mptcp_adjust_add_addr_timeout(msk); 397 if (!timeout || mptcp_pm_should_add_signal_addr(msk)) 398 goto out; 399 400 spin_lock_bh(&msk->pm.lock); 401 402 /* The cancel path (mptcp_pm_announced_del_timer()) can race with this 403 * callback. Once cancel updates retrans_times to MAX, suppress further 404 * retransmissions here. If this callback acquires pm.lock first, one 405 * final transmit attempt is still possible. 406 */ 407 if (entry->retrans_times < ADD_ADDR_RETRANS_MAX && 408 !mptcp_pm_should_add_signal_addr(msk)) { 409 pr_debug("retransmit ADD_ADDR id=%d\n", entry->addr.id); 410 mptcp_pm_announce_addr(msk, &entry->addr, false); 411 mptcp_pm_add_addr_send_ack(msk); 412 entry->retrans_times++; 413 } 414 415 if (entry->retrans_times < ADD_ADDR_RETRANS_MAX) 416 timeout <<= entry->retrans_times; 417 else 418 timeout = 0; 419 420 spin_unlock_bh(&msk->pm.lock); 421 422 if (entry->retrans_times == ADD_ADDR_RETRANS_MAX) 423 mptcp_pm_subflow_established(msk); 424 425 out: 426 if (timeout) 427 sk_reset_timer(sk, timer, jiffies + timeout); 428 else 429 /* if sock_put calls sk_free: avoid waiting for this timer */ 430 entry->timer_done = true; 431 bh_unlock_sock(sk); 432 sock_put(sk); 433 } 434 435 bool mptcp_pm_announced_alloc(struct mptcp_sock *msk, 436 const struct mptcp_addr_info *addr) 437 { 438 struct mptcp_pm_add_addr *add_entry = NULL; 439 struct sock *sk = (struct sock *)msk; 440 unsigned int timeout; 441 442 lockdep_assert_held(&msk->pm.lock); 443 444 add_entry = mptcp_pm_announced_lookup(msk, addr); 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->timer, mptcp_pm_add_addr_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->timer, jiffies + timeout); 468 469 return true; 470 } 471 472 static void mptcp_pm_free_announced_list(struct mptcp_sock *msk) 473 { 474 struct mptcp_pm_add_addr *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->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 struct mptcp_pm_add_addr *entry; 734 735 pr_debug("msk=%p\n", msk); 736 737 entry = mptcp_pm_announced_del_timer(msk, addr, true); 738 739 if (!entry || !READ_ONCE(pm->work_pending)) 740 return; 741 742 spin_lock_bh(&pm->lock); 743 744 if (READ_ONCE(pm->work_pending)) 745 mptcp_pm_schedule_work(msk, MPTCP_PM_SUBFLOW_ESTABLISHED); 746 747 spin_unlock_bh(&pm->lock); 748 } 749 750 /* To be called while pm->lock is held */ 751 static void mptcp_pm_add_addr_send_ack(struct mptcp_sock *msk) 752 { 753 if (!mptcp_pm_should_add_signal(msk)) 754 return; 755 756 mptcp_pm_schedule_work(msk, MPTCP_PM_ADD_ADDR_SEND_ACK); 757 } 758 759 static void mptcp_pm_rm_addr_or_subflow(struct mptcp_sock *msk, 760 const struct mptcp_rm_list *rm_list, 761 enum linux_mptcp_mib_field rm_type) 762 { 763 struct mptcp_subflow_context *subflow, *tmp; 764 struct sock *sk = (struct sock *)msk; 765 u8 i; 766 767 pr_debug("%s rm_list_nr %d\n", 768 rm_type == MPTCP_MIB_RMADDR ? "address" : "subflow", rm_list->nr); 769 770 msk_owned_by_me(msk); 771 772 if (sk->sk_state == TCP_LISTEN) 773 return; 774 775 if (!rm_list->nr) 776 return; 777 778 if (list_empty(&msk->conn_list)) 779 return; 780 781 for (i = 0; i < rm_list->nr; i++) { 782 u8 rm_id = rm_list->ids[i]; 783 bool removed = false; 784 785 mptcp_for_each_subflow_safe(msk, subflow, tmp) { 786 struct sock *ssk = mptcp_subflow_tcp_sock(subflow); 787 u8 remote_id = READ_ONCE(subflow->remote_id); 788 int how = RCV_SHUTDOWN | SEND_SHUTDOWN; 789 u8 id = subflow_get_local_id(subflow); 790 791 if ((1 << inet_sk_state_load(ssk)) & 792 (TCPF_FIN_WAIT1 | TCPF_FIN_WAIT2 | TCPF_CLOSING | TCPF_CLOSE)) 793 continue; 794 if (rm_type == MPTCP_MIB_RMADDR && remote_id != rm_id) 795 continue; 796 if (rm_type == MPTCP_MIB_RMSUBFLOW && id != rm_id) 797 continue; 798 799 pr_debug(" -> %s rm_list_ids[%d]=%u local_id=%u remote_id=%u mpc_id=%u\n", 800 rm_type == MPTCP_MIB_RMADDR ? "address" : "subflow", 801 i, rm_id, id, remote_id, msk->mpc_endpoint_id); 802 spin_unlock_bh(&msk->pm.lock); 803 mptcp_subflow_shutdown(sk, ssk, how); 804 removed |= subflow->request_join; 805 806 /* the following takes care of updating the subflows counter */ 807 mptcp_close_ssk(sk, ssk, subflow); 808 spin_lock_bh(&msk->pm.lock); 809 810 if (rm_type == MPTCP_MIB_RMSUBFLOW) 811 __MPTCP_INC_STATS(sock_net(sk), rm_type); 812 } 813 814 if (rm_type == MPTCP_MIB_RMADDR) { 815 __MPTCP_INC_STATS(sock_net(sk), rm_type); 816 if (removed && mptcp_pm_is_kernel(msk)) 817 mptcp_pm_nl_rm_addr(msk, rm_id); 818 } 819 } 820 } 821 822 static void mptcp_pm_rm_addr_recv(struct mptcp_sock *msk) 823 { 824 mptcp_pm_rm_addr_or_subflow(msk, &msk->pm.rm_list_rx, MPTCP_MIB_RMADDR); 825 } 826 827 void mptcp_pm_rm_subflow(struct mptcp_sock *msk, 828 const struct mptcp_rm_list *rm_list) 829 { 830 mptcp_pm_rm_addr_or_subflow(msk, rm_list, MPTCP_MIB_RMSUBFLOW); 831 } 832 833 void mptcp_pm_rm_addr_received(struct mptcp_sock *msk, 834 const struct mptcp_rm_list *rm_list) 835 { 836 struct mptcp_pm_data *pm = &msk->pm; 837 u8 i; 838 839 pr_debug("msk=%p remote_ids_nr=%d\n", msk, rm_list->nr); 840 841 for (i = 0; i < rm_list->nr; i++) 842 mptcp_event_addr_removed(msk, rm_list->ids[i]); 843 844 spin_lock_bh(&pm->lock); 845 if (mptcp_pm_schedule_work(msk, MPTCP_PM_RM_ADDR_RECEIVED)) 846 pm->rm_list_rx = *rm_list; 847 else 848 __MPTCP_INC_STATS(sock_net((struct sock *)msk), MPTCP_MIB_RMADDRDROP); 849 spin_unlock_bh(&pm->lock); 850 } 851 852 void mptcp_pm_mp_prio_received(struct sock *ssk, u8 bkup) 853 { 854 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk); 855 struct sock *sk = subflow->conn; 856 struct mptcp_sock *msk; 857 858 pr_debug("subflow->backup=%d, bkup=%d\n", subflow->backup, bkup); 859 msk = mptcp_sk(sk); 860 if (subflow->backup != bkup) 861 subflow->backup = bkup; 862 863 mptcp_event(MPTCP_EVENT_SUB_PRIORITY, msk, ssk, GFP_ATOMIC); 864 } 865 866 void mptcp_pm_mp_fail_received(struct sock *sk, u64 fail_seq) 867 { 868 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk); 869 struct mptcp_sock *msk = mptcp_sk(subflow->conn); 870 871 pr_debug("fail_seq=%llu\n", fail_seq); 872 873 /* After accepting the fail, we can't create any other subflows */ 874 spin_lock_bh(&msk->fallback_lock); 875 if (!msk->allow_infinite_fallback) { 876 spin_unlock_bh(&msk->fallback_lock); 877 return; 878 } 879 msk->allow_subflows = false; 880 spin_unlock_bh(&msk->fallback_lock); 881 882 if (!subflow->fail_tout) { 883 pr_debug("send MP_FAIL response and infinite map\n"); 884 885 subflow->send_mp_fail = 1; 886 subflow->send_infinite_map = 1; 887 tcp_send_ack(sk); 888 } else { 889 pr_debug("MP_FAIL response received\n"); 890 WRITE_ONCE(subflow->fail_tout, 0); 891 } 892 } 893 894 static int mptcp_add_addr_len(int family, bool echo, bool port) 895 { 896 int len = TCPOLEN_MPTCP_ADD_ADDR_BASE; 897 898 if (family == AF_INET6) 899 len = TCPOLEN_MPTCP_ADD_ADDR6_BASE; 900 if (!echo) 901 len += MPTCPOPT_THMAC_LEN; 902 /* account for 2 trailing 'nop' options */ 903 if (port) 904 len += TCPOLEN_MPTCP_PORT_LEN + TCPOLEN_MPTCP_PORT_ALIGN; 905 906 return len; 907 } 908 909 bool mptcp_pm_add_addr_signal(struct mptcp_sock *msk, int *size, int remaining, 910 struct mptcp_addr_info *addr, bool *echo, 911 bool *drop_ts) 912 { 913 bool skip_add_addr = false; 914 bool ret = false; 915 u8 add_addr; 916 int len = 0; 917 u8 family; 918 bool port; 919 920 spin_lock_bh(&msk->pm.lock); 921 922 /* double check after the lock is acquired */ 923 if (!mptcp_pm_should_add_signal(msk)) 924 goto out_unlock; 925 926 /* always drop every other options for pure ack ADD_ADDR; this is a 927 * plain dup-ack from TCP perspective. The other MPTCP-relevant info, 928 * if any, will be carried by the 'original' TCP ack 929 */ 930 len -= *size; 931 932 *echo = mptcp_pm_should_add_signal_echo(msk); 933 if (*echo) { 934 *addr = msk->pm.remote; 935 add_addr = msk->pm.addr_signal & ~BIT(MPTCP_ADD_ADDR_ECHO); 936 port = !!msk->pm.remote.port; 937 family = msk->pm.remote.family; 938 } else { 939 *addr = msk->pm.local; 940 add_addr = msk->pm.addr_signal & ~BIT(MPTCP_ADD_ADDR_SIGNAL); 941 port = !!msk->pm.local.port; 942 family = msk->pm.local.family; 943 } 944 945 len += mptcp_add_addr_len(family, *echo, port); 946 if (len > remaining) { 947 struct net *net = sock_net((struct sock *)msk); 948 949 if (*drop_ts && mptcp_add_addr_v6_port_drop_ts(net)) { 950 /* OK without TCP Timestamps? */ 951 len -= TCPOLEN_TSTAMP_ALIGNED; 952 if (len <= remaining) 953 goto enough_space; 954 } 955 956 if (*echo) { 957 MPTCP_INC_STATS(net, MPTCP_MIB_ECHOADDTXDROP); 958 } else { 959 skip_add_addr = true; 960 MPTCP_INC_STATS(net, MPTCP_MIB_ADDADDRTXDROP); 961 } 962 goto drop_signal_mark; 963 } 964 965 *drop_ts = false; 966 967 enough_space: 968 ret = true; 969 *size = len; 970 971 drop_signal_mark: 972 WRITE_ONCE(msk->pm.addr_signal, add_addr); 973 974 out_unlock: 975 spin_unlock_bh(&msk->pm.lock); 976 977 /* On pure-ACK option-space exhaustion, stop retrying this ADD_ADDR: 978 * clear the signal bit, cancel the matching retransmission timer, and 979 * let the PM state machine progress. 980 */ 981 if (skip_add_addr) { 982 mptcp_pm_announced_del_timer(msk, addr, true); 983 mptcp_pm_subflow_established(msk); 984 } 985 return ret; 986 } 987 988 static int mptcp_rm_addr_len(const struct mptcp_rm_list *rm_list) 989 { 990 if (rm_list->nr == 0 || rm_list->nr > MPTCP_RM_IDS_MAX) 991 return -EINVAL; 992 993 return TCPOLEN_MPTCP_RM_ADDR_BASE + roundup(rm_list->nr - 1, 4) + 1; 994 } 995 996 bool mptcp_pm_rm_addr_signal(struct mptcp_sock *msk, unsigned int remaining, 997 struct mptcp_rm_list *rm_list, int *size) 998 { 999 int ret = false, len; 1000 u8 rm_addr; 1001 1002 spin_lock_bh(&msk->pm.lock); 1003 1004 /* double check after the lock is acquired */ 1005 if (!mptcp_pm_should_rm_signal(msk)) 1006 goto out_unlock; 1007 1008 rm_addr = msk->pm.addr_signal & ~BIT(MPTCP_RM_ADDR_SIGNAL); 1009 len = mptcp_rm_addr_len(&msk->pm.rm_list_tx); 1010 if (len < 0) { 1011 WRITE_ONCE(msk->pm.addr_signal, rm_addr); 1012 goto out_unlock; 1013 } 1014 if (remaining < len) 1015 goto out_unlock; 1016 1017 *size = len; 1018 *rm_list = msk->pm.rm_list_tx; 1019 WRITE_ONCE(msk->pm.addr_signal, rm_addr); 1020 ret = true; 1021 1022 out_unlock: 1023 spin_unlock_bh(&msk->pm.lock); 1024 return ret; 1025 } 1026 1027 int mptcp_pm_get_local_id(struct mptcp_sock *msk, struct sock_common *skc) 1028 { 1029 struct mptcp_pm_addr_entry skc_local = { 0 }; 1030 struct mptcp_addr_info msk_local; 1031 1032 if (WARN_ON_ONCE(!msk)) 1033 return -1; 1034 1035 /* The 0 ID mapping is defined by the first subflow, copied into the msk 1036 * addr 1037 */ 1038 mptcp_local_address((struct sock_common *)msk, &msk_local); 1039 mptcp_local_address((struct sock_common *)skc, &skc_local.addr); 1040 if (mptcp_addresses_equal(&msk_local, &skc_local.addr, false)) 1041 return 0; 1042 1043 skc_local.addr.id = 0; 1044 skc_local.flags = MPTCP_PM_ADDR_FLAG_IMPLICIT; 1045 1046 if (mptcp_pm_is_userspace(msk)) 1047 return mptcp_userspace_pm_get_local_id(msk, &skc_local); 1048 return mptcp_pm_nl_get_local_id(msk, &skc_local); 1049 } 1050 1051 bool mptcp_pm_is_backup(struct mptcp_sock *msk, struct sock_common *skc) 1052 { 1053 struct mptcp_addr_info skc_local; 1054 1055 mptcp_local_address((struct sock_common *)skc, &skc_local); 1056 1057 if (mptcp_pm_is_userspace(msk)) 1058 return mptcp_userspace_pm_is_backup(msk, &skc_local); 1059 1060 return mptcp_pm_nl_is_backup(msk, &skc_local); 1061 } 1062 1063 static void mptcp_pm_subflows_chk_stale(const struct mptcp_sock *msk, struct sock *ssk) 1064 { 1065 struct mptcp_subflow_context *iter, *subflow = mptcp_subflow_ctx(ssk); 1066 struct sock *sk = (struct sock *)msk; 1067 unsigned int active_max_loss_cnt; 1068 struct net *net = sock_net(sk); 1069 unsigned int stale_loss_cnt; 1070 bool slow; 1071 1072 stale_loss_cnt = mptcp_stale_loss_cnt(net); 1073 if (subflow->stale || !stale_loss_cnt || subflow->stale_count <= stale_loss_cnt) 1074 return; 1075 1076 /* look for another available subflow not in loss state */ 1077 active_max_loss_cnt = max_t(int, stale_loss_cnt - 1, 1); 1078 mptcp_for_each_subflow(msk, iter) { 1079 if (iter != subflow && mptcp_subflow_active(iter) && 1080 iter->stale_count < active_max_loss_cnt) { 1081 /* we have some alternatives, try to mark this subflow as idle ...*/ 1082 slow = lock_sock_fast(ssk); 1083 if (!tcp_rtx_and_write_queues_empty(ssk)) { 1084 subflow->stale = 1; 1085 __mptcp_retransmit_pending_data(sk); 1086 MPTCP_INC_STATS(net, MPTCP_MIB_SUBFLOWSTALE); 1087 } 1088 unlock_sock_fast(ssk, slow); 1089 1090 /* always try to push the pending data regardless of re-injections: 1091 * we can possibly use backup subflows now, and subflow selection 1092 * is cheap under the msk socket lock 1093 */ 1094 __mptcp_push_pending(sk, 0); 1095 return; 1096 } 1097 } 1098 } 1099 1100 void mptcp_pm_subflow_chk_stale(const struct mptcp_sock *msk, struct sock *ssk) 1101 { 1102 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk); 1103 u32 rcv_tstamp = READ_ONCE(tcp_sk(ssk)->rcv_tstamp); 1104 1105 /* keep track of rtx periods with no progress */ 1106 if (!subflow->stale_count) { 1107 subflow->stale_rcv_tstamp = rcv_tstamp; 1108 subflow->stale_count++; 1109 } else if (subflow->stale_rcv_tstamp == rcv_tstamp) { 1110 if (subflow->stale_count < U8_MAX) 1111 subflow->stale_count++; 1112 mptcp_pm_subflows_chk_stale(msk, ssk); 1113 } else { 1114 subflow->stale_count = 0; 1115 mptcp_subflow_set_active(subflow); 1116 } 1117 } 1118 1119 void mptcp_pm_worker(struct mptcp_sock *msk) 1120 { 1121 struct mptcp_pm_data *pm = &msk->pm; 1122 1123 msk_owned_by_me(msk); 1124 1125 if (!(pm->status & MPTCP_PM_WORK_MASK)) 1126 return; 1127 1128 spin_lock_bh(&msk->pm.lock); 1129 1130 pr_debug("msk=%p status=%x\n", msk, pm->status); 1131 if (pm->status & BIT(MPTCP_PM_ADD_ADDR_SEND_ACK)) { 1132 pm->status &= ~BIT(MPTCP_PM_ADD_ADDR_SEND_ACK); 1133 mptcp_pm_addr_send_ack(msk); 1134 } 1135 if (pm->status & BIT(MPTCP_PM_RM_ADDR_RECEIVED)) { 1136 pm->status &= ~BIT(MPTCP_PM_RM_ADDR_RECEIVED); 1137 mptcp_pm_rm_addr_recv(msk); 1138 } 1139 __mptcp_pm_kernel_worker(msk); 1140 1141 spin_unlock_bh(&msk->pm.lock); 1142 } 1143 1144 void mptcp_pm_destroy(struct mptcp_sock *msk) 1145 { 1146 mptcp_pm_free_announced_list(msk); 1147 1148 if (mptcp_pm_is_userspace(msk)) 1149 mptcp_userspace_pm_free_local_addr_list(msk); 1150 } 1151 1152 void mptcp_pm_data_reset(struct mptcp_sock *msk) 1153 { 1154 u8 pm_type = mptcp_get_pm_type(sock_net((struct sock *)msk)); 1155 struct mptcp_pm_data *pm = &msk->pm; 1156 1157 memset(&pm->reset, 0, sizeof(pm->reset)); 1158 pm->rm_list_tx.nr = 0; 1159 pm->rm_list_rx.nr = 0; 1160 WRITE_ONCE(pm->pm_type, pm_type); 1161 1162 if (pm_type == MPTCP_PM_TYPE_KERNEL) { 1163 bool subflows_allowed = !!mptcp_pm_get_limit_extra_subflows(msk); 1164 1165 /* pm->work_pending must be only be set to 'true' when 1166 * pm->pm_type is set to MPTCP_PM_TYPE_KERNEL 1167 */ 1168 WRITE_ONCE(pm->work_pending, 1169 (!!mptcp_pm_get_endp_subflow_max(msk) && 1170 subflows_allowed) || 1171 !!mptcp_pm_get_endp_signal_max(msk)); 1172 WRITE_ONCE(pm->accept_addr, 1173 !!mptcp_pm_get_limit_add_addr_accepted(msk) && 1174 subflows_allowed); 1175 WRITE_ONCE(pm->accept_subflow, subflows_allowed); 1176 1177 bitmap_fill(pm->id_avail_bitmap, MPTCP_PM_MAX_ADDR_ID + 1); 1178 } 1179 } 1180 1181 void mptcp_pm_data_init(struct mptcp_sock *msk) 1182 { 1183 spin_lock_init(&msk->pm.lock); 1184 INIT_LIST_HEAD(&msk->pm.anno_list); 1185 INIT_LIST_HEAD(&msk->pm.userspace_pm_local_addr_list); 1186 mptcp_pm_data_reset(msk); 1187 } 1188 1189 void __init mptcp_pm_init(void) 1190 { 1191 mptcp_pm_kernel_register(); 1192 mptcp_pm_userspace_register(); 1193 mptcp_pm_nl_init(); 1194 } 1195 1196 /* Must be called with rcu read lock held */ 1197 struct mptcp_pm_ops *mptcp_pm_find(const char *name) 1198 { 1199 struct mptcp_pm_ops *pm_ops; 1200 1201 list_for_each_entry_rcu(pm_ops, &mptcp_pm_list, list) { 1202 if (!strcmp(pm_ops->name, name)) 1203 return pm_ops; 1204 } 1205 1206 return NULL; 1207 } 1208 1209 int mptcp_pm_validate(struct mptcp_pm_ops *pm_ops) 1210 { 1211 return 0; 1212 } 1213 1214 int mptcp_pm_register(struct mptcp_pm_ops *pm_ops) 1215 { 1216 int ret; 1217 1218 ret = mptcp_pm_validate(pm_ops); 1219 if (ret) 1220 return ret; 1221 1222 spin_lock(&mptcp_pm_list_lock); 1223 if (mptcp_pm_find(pm_ops->name)) { 1224 spin_unlock(&mptcp_pm_list_lock); 1225 return -EEXIST; 1226 } 1227 list_add_tail_rcu(&pm_ops->list, &mptcp_pm_list); 1228 spin_unlock(&mptcp_pm_list_lock); 1229 1230 pr_debug("%s registered\n", pm_ops->name); 1231 return 0; 1232 } 1233 1234 void mptcp_pm_unregister(struct mptcp_pm_ops *pm_ops) 1235 { 1236 /* skip unregistering the default path manager */ 1237 if (WARN_ON_ONCE(pm_ops == &mptcp_pm_kernel)) 1238 return; 1239 1240 spin_lock(&mptcp_pm_list_lock); 1241 list_del_rcu(&pm_ops->list); 1242 spin_unlock(&mptcp_pm_list_lock); 1243 } 1244 1245 /* Build string with list of available path manager values. 1246 * Similar to tcp_get_available_congestion_control() 1247 */ 1248 void mptcp_pm_get_available(char *buf, size_t maxlen) 1249 { 1250 struct mptcp_pm_ops *pm_ops; 1251 size_t offs = 0; 1252 1253 rcu_read_lock(); 1254 list_for_each_entry_rcu(pm_ops, &mptcp_pm_list, list) { 1255 offs += snprintf(buf + offs, maxlen - offs, "%s%s", 1256 offs == 0 ? "" : " ", pm_ops->name); 1257 1258 if (WARN_ON_ONCE(offs >= maxlen)) 1259 break; 1260 } 1261 rcu_read_unlock(); 1262 } 1263