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