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 = timer_container_of(entry, timer, 274 add_timer); 275 struct mptcp_sock *msk = entry->sock; 276 struct sock *sk = (struct sock *)msk; 277 unsigned int timeout; 278 279 pr_debug("msk=%p\n", msk); 280 281 if (!msk) 282 return; 283 284 if (inet_sk_state_load(sk) == TCP_CLOSE) 285 return; 286 287 if (!entry->addr.id) 288 return; 289 290 if (mptcp_pm_should_add_signal_addr(msk)) { 291 sk_reset_timer(sk, timer, jiffies + TCP_RTO_MAX / 8); 292 goto out; 293 } 294 295 timeout = mptcp_get_add_addr_timeout(sock_net(sk)); 296 if (!timeout) 297 goto out; 298 299 spin_lock_bh(&msk->pm.lock); 300 301 if (!mptcp_pm_should_add_signal_addr(msk)) { 302 pr_debug("retransmit ADD_ADDR id=%d\n", entry->addr.id); 303 mptcp_pm_announce_addr(msk, &entry->addr, false); 304 mptcp_pm_add_addr_send_ack(msk); 305 entry->retrans_times++; 306 } 307 308 if (entry->retrans_times < ADD_ADDR_RETRANS_MAX) 309 sk_reset_timer(sk, timer, 310 jiffies + timeout); 311 312 spin_unlock_bh(&msk->pm.lock); 313 314 if (entry->retrans_times == ADD_ADDR_RETRANS_MAX) 315 mptcp_pm_subflow_established(msk); 316 317 out: 318 __sock_put(sk); 319 } 320 321 struct mptcp_pm_add_entry * 322 mptcp_pm_del_add_timer(struct mptcp_sock *msk, 323 const struct mptcp_addr_info *addr, bool check_id) 324 { 325 struct mptcp_pm_add_entry *entry; 326 struct sock *sk = (struct sock *)msk; 327 struct timer_list *add_timer = NULL; 328 329 spin_lock_bh(&msk->pm.lock); 330 entry = mptcp_lookup_anno_list_by_saddr(msk, addr); 331 if (entry && (!check_id || entry->addr.id == addr->id)) { 332 entry->retrans_times = ADD_ADDR_RETRANS_MAX; 333 add_timer = &entry->add_timer; 334 } 335 if (!check_id && entry) 336 list_del(&entry->list); 337 spin_unlock_bh(&msk->pm.lock); 338 339 /* no lock, because sk_stop_timer_sync() is calling timer_delete_sync() */ 340 if (add_timer) 341 sk_stop_timer_sync(sk, add_timer); 342 343 return entry; 344 } 345 346 bool mptcp_pm_alloc_anno_list(struct mptcp_sock *msk, 347 const struct mptcp_addr_info *addr) 348 { 349 struct mptcp_pm_add_entry *add_entry = NULL; 350 struct sock *sk = (struct sock *)msk; 351 struct net *net = sock_net(sk); 352 unsigned int timeout; 353 354 lockdep_assert_held(&msk->pm.lock); 355 356 add_entry = mptcp_lookup_anno_list_by_saddr(msk, addr); 357 358 if (add_entry) { 359 if (WARN_ON_ONCE(mptcp_pm_is_kernel(msk))) 360 return false; 361 362 goto reset_timer; 363 } 364 365 add_entry = kmalloc(sizeof(*add_entry), GFP_ATOMIC); 366 if (!add_entry) 367 return false; 368 369 list_add(&add_entry->list, &msk->pm.anno_list); 370 371 add_entry->addr = *addr; 372 add_entry->sock = msk; 373 add_entry->retrans_times = 0; 374 375 timer_setup(&add_entry->add_timer, mptcp_pm_add_timer, 0); 376 reset_timer: 377 timeout = mptcp_get_add_addr_timeout(net); 378 if (timeout) 379 sk_reset_timer(sk, &add_entry->add_timer, jiffies + timeout); 380 381 return true; 382 } 383 384 static void mptcp_pm_free_anno_list(struct mptcp_sock *msk) 385 { 386 struct mptcp_pm_add_entry *entry, *tmp; 387 struct sock *sk = (struct sock *)msk; 388 LIST_HEAD(free_list); 389 390 pr_debug("msk=%p\n", msk); 391 392 spin_lock_bh(&msk->pm.lock); 393 list_splice_init(&msk->pm.anno_list, &free_list); 394 spin_unlock_bh(&msk->pm.lock); 395 396 list_for_each_entry_safe(entry, tmp, &free_list, list) { 397 sk_stop_timer_sync(sk, &entry->add_timer); 398 kfree(entry); 399 } 400 } 401 402 /* path manager command handlers */ 403 404 int mptcp_pm_announce_addr(struct mptcp_sock *msk, 405 const struct mptcp_addr_info *addr, 406 bool echo) 407 { 408 u8 add_addr = READ_ONCE(msk->pm.addr_signal); 409 410 pr_debug("msk=%p, local_id=%d, echo=%d\n", msk, addr->id, echo); 411 412 lockdep_assert_held(&msk->pm.lock); 413 414 if (add_addr & 415 (echo ? BIT(MPTCP_ADD_ADDR_ECHO) : BIT(MPTCP_ADD_ADDR_SIGNAL))) { 416 MPTCP_INC_STATS(sock_net((struct sock *)msk), 417 echo ? MPTCP_MIB_ECHOADDTXDROP : MPTCP_MIB_ADDADDRTXDROP); 418 return -EINVAL; 419 } 420 421 if (echo) { 422 msk->pm.remote = *addr; 423 add_addr |= BIT(MPTCP_ADD_ADDR_ECHO); 424 } else { 425 msk->pm.local = *addr; 426 add_addr |= BIT(MPTCP_ADD_ADDR_SIGNAL); 427 } 428 WRITE_ONCE(msk->pm.addr_signal, add_addr); 429 return 0; 430 } 431 432 int mptcp_pm_remove_addr(struct mptcp_sock *msk, const struct mptcp_rm_list *rm_list) 433 { 434 u8 rm_addr = READ_ONCE(msk->pm.addr_signal); 435 436 pr_debug("msk=%p, rm_list_nr=%d\n", msk, rm_list->nr); 437 438 if (rm_addr) { 439 MPTCP_ADD_STATS(sock_net((struct sock *)msk), 440 MPTCP_MIB_RMADDRTXDROP, rm_list->nr); 441 return -EINVAL; 442 } 443 444 msk->pm.rm_list_tx = *rm_list; 445 rm_addr |= BIT(MPTCP_RM_ADDR_SIGNAL); 446 WRITE_ONCE(msk->pm.addr_signal, rm_addr); 447 mptcp_pm_addr_send_ack(msk); 448 return 0; 449 } 450 451 /* path manager event handlers */ 452 453 void mptcp_pm_new_connection(struct mptcp_sock *msk, const struct sock *ssk, int server_side) 454 { 455 struct mptcp_pm_data *pm = &msk->pm; 456 457 pr_debug("msk=%p, token=%u side=%d\n", msk, READ_ONCE(msk->token), server_side); 458 459 WRITE_ONCE(pm->server_side, server_side); 460 mptcp_event(MPTCP_EVENT_CREATED, msk, ssk, GFP_ATOMIC); 461 } 462 463 bool mptcp_pm_allow_new_subflow(struct mptcp_sock *msk) 464 { 465 struct mptcp_pm_data *pm = &msk->pm; 466 unsigned int subflows_max; 467 int ret = 0; 468 469 if (mptcp_pm_is_userspace(msk)) { 470 if (mptcp_userspace_pm_active(msk)) { 471 spin_lock_bh(&pm->lock); 472 pm->subflows++; 473 spin_unlock_bh(&pm->lock); 474 return true; 475 } 476 return false; 477 } 478 479 subflows_max = mptcp_pm_get_subflows_max(msk); 480 481 pr_debug("msk=%p subflows=%d max=%d allow=%d\n", msk, pm->subflows, 482 subflows_max, READ_ONCE(pm->accept_subflow)); 483 484 /* try to avoid acquiring the lock below */ 485 if (!READ_ONCE(pm->accept_subflow)) 486 return false; 487 488 spin_lock_bh(&pm->lock); 489 if (READ_ONCE(pm->accept_subflow)) { 490 ret = pm->subflows < subflows_max; 491 if (ret && ++pm->subflows == subflows_max) 492 WRITE_ONCE(pm->accept_subflow, false); 493 } 494 spin_unlock_bh(&pm->lock); 495 496 return ret; 497 } 498 499 /* return true if the new status bit is currently cleared, that is, this event 500 * can be server, eventually by an already scheduled work 501 */ 502 static bool mptcp_pm_schedule_work(struct mptcp_sock *msk, 503 enum mptcp_pm_status new_status) 504 { 505 pr_debug("msk=%p status=%x new=%lx\n", msk, msk->pm.status, 506 BIT(new_status)); 507 if (msk->pm.status & BIT(new_status)) 508 return false; 509 510 msk->pm.status |= BIT(new_status); 511 mptcp_schedule_work((struct sock *)msk); 512 return true; 513 } 514 515 void mptcp_pm_fully_established(struct mptcp_sock *msk, const struct sock *ssk) 516 { 517 struct mptcp_pm_data *pm = &msk->pm; 518 bool announce = false; 519 520 pr_debug("msk=%p\n", msk); 521 522 spin_lock_bh(&pm->lock); 523 524 /* mptcp_pm_fully_established() can be invoked by multiple 525 * racing paths - accept() and check_fully_established() 526 * be sure to serve this event only once. 527 */ 528 if (READ_ONCE(pm->work_pending) && 529 !(pm->status & BIT(MPTCP_PM_ALREADY_ESTABLISHED))) 530 mptcp_pm_schedule_work(msk, MPTCP_PM_ESTABLISHED); 531 532 if ((pm->status & BIT(MPTCP_PM_ALREADY_ESTABLISHED)) == 0) 533 announce = true; 534 535 pm->status |= BIT(MPTCP_PM_ALREADY_ESTABLISHED); 536 spin_unlock_bh(&pm->lock); 537 538 if (announce) 539 mptcp_event(MPTCP_EVENT_ESTABLISHED, msk, ssk, GFP_ATOMIC); 540 } 541 542 void mptcp_pm_connection_closed(struct mptcp_sock *msk) 543 { 544 pr_debug("msk=%p\n", msk); 545 546 if (msk->token) 547 mptcp_event(MPTCP_EVENT_CLOSED, msk, NULL, GFP_KERNEL); 548 } 549 550 void mptcp_pm_subflow_established(struct mptcp_sock *msk) 551 { 552 struct mptcp_pm_data *pm = &msk->pm; 553 554 pr_debug("msk=%p\n", msk); 555 556 if (!READ_ONCE(pm->work_pending)) 557 return; 558 559 spin_lock_bh(&pm->lock); 560 561 if (READ_ONCE(pm->work_pending)) 562 mptcp_pm_schedule_work(msk, MPTCP_PM_SUBFLOW_ESTABLISHED); 563 564 spin_unlock_bh(&pm->lock); 565 } 566 567 void mptcp_pm_subflow_check_next(struct mptcp_sock *msk, 568 const struct mptcp_subflow_context *subflow) 569 { 570 struct mptcp_pm_data *pm = &msk->pm; 571 bool update_subflows; 572 573 update_subflows = subflow->request_join || subflow->mp_join; 574 if (mptcp_pm_is_userspace(msk)) { 575 if (update_subflows) { 576 spin_lock_bh(&pm->lock); 577 pm->subflows--; 578 spin_unlock_bh(&pm->lock); 579 } 580 return; 581 } 582 583 if (!READ_ONCE(pm->work_pending) && !update_subflows) 584 return; 585 586 spin_lock_bh(&pm->lock); 587 if (update_subflows) 588 __mptcp_pm_close_subflow(msk); 589 590 /* Even if this subflow is not really established, tell the PM to try 591 * to pick the next ones, if possible. 592 */ 593 if (mptcp_pm_nl_check_work_pending(msk)) 594 mptcp_pm_schedule_work(msk, MPTCP_PM_SUBFLOW_ESTABLISHED); 595 596 spin_unlock_bh(&pm->lock); 597 } 598 599 void mptcp_pm_add_addr_received(const struct sock *ssk, 600 const struct mptcp_addr_info *addr) 601 { 602 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk); 603 struct mptcp_sock *msk = mptcp_sk(subflow->conn); 604 struct mptcp_pm_data *pm = &msk->pm; 605 606 pr_debug("msk=%p remote_id=%d accept=%d\n", msk, addr->id, 607 READ_ONCE(pm->accept_addr)); 608 609 mptcp_event_addr_announced(ssk, addr); 610 611 spin_lock_bh(&pm->lock); 612 613 if (mptcp_pm_is_userspace(msk)) { 614 if (mptcp_userspace_pm_active(msk)) { 615 mptcp_pm_announce_addr(msk, addr, true); 616 mptcp_pm_add_addr_send_ack(msk); 617 } else { 618 __MPTCP_INC_STATS(sock_net((struct sock *)msk), MPTCP_MIB_ADDADDRDROP); 619 } 620 /* id0 should not have a different address */ 621 } else if ((addr->id == 0 && !mptcp_pm_is_init_remote_addr(msk, addr)) || 622 (addr->id > 0 && !READ_ONCE(pm->accept_addr))) { 623 mptcp_pm_announce_addr(msk, addr, true); 624 mptcp_pm_add_addr_send_ack(msk); 625 } else if (mptcp_pm_schedule_work(msk, MPTCP_PM_ADD_ADDR_RECEIVED)) { 626 pm->remote = *addr; 627 } else { 628 __MPTCP_INC_STATS(sock_net((struct sock *)msk), MPTCP_MIB_ADDADDRDROP); 629 } 630 631 spin_unlock_bh(&pm->lock); 632 } 633 634 void mptcp_pm_add_addr_echoed(struct mptcp_sock *msk, 635 const struct mptcp_addr_info *addr) 636 { 637 struct mptcp_pm_data *pm = &msk->pm; 638 639 pr_debug("msk=%p\n", msk); 640 641 if (!READ_ONCE(pm->work_pending)) 642 return; 643 644 spin_lock_bh(&pm->lock); 645 646 if (mptcp_lookup_anno_list_by_saddr(msk, addr) && READ_ONCE(pm->work_pending)) 647 mptcp_pm_schedule_work(msk, MPTCP_PM_SUBFLOW_ESTABLISHED); 648 649 spin_unlock_bh(&pm->lock); 650 } 651 652 void mptcp_pm_add_addr_send_ack(struct mptcp_sock *msk) 653 { 654 if (!mptcp_pm_should_add_signal(msk)) 655 return; 656 657 mptcp_pm_schedule_work(msk, MPTCP_PM_ADD_ADDR_SEND_ACK); 658 } 659 660 static void mptcp_pm_rm_addr_or_subflow(struct mptcp_sock *msk, 661 const struct mptcp_rm_list *rm_list, 662 enum linux_mptcp_mib_field rm_type) 663 { 664 struct mptcp_subflow_context *subflow, *tmp; 665 struct sock *sk = (struct sock *)msk; 666 u8 i; 667 668 pr_debug("%s rm_list_nr %d\n", 669 rm_type == MPTCP_MIB_RMADDR ? "address" : "subflow", rm_list->nr); 670 671 msk_owned_by_me(msk); 672 673 if (sk->sk_state == TCP_LISTEN) 674 return; 675 676 if (!rm_list->nr) 677 return; 678 679 if (list_empty(&msk->conn_list)) 680 return; 681 682 for (i = 0; i < rm_list->nr; i++) { 683 u8 rm_id = rm_list->ids[i]; 684 bool removed = false; 685 686 mptcp_for_each_subflow_safe(msk, subflow, tmp) { 687 struct sock *ssk = mptcp_subflow_tcp_sock(subflow); 688 u8 remote_id = READ_ONCE(subflow->remote_id); 689 int how = RCV_SHUTDOWN | SEND_SHUTDOWN; 690 u8 id = subflow_get_local_id(subflow); 691 692 if ((1 << inet_sk_state_load(ssk)) & 693 (TCPF_FIN_WAIT1 | TCPF_FIN_WAIT2 | TCPF_CLOSING | TCPF_CLOSE)) 694 continue; 695 if (rm_type == MPTCP_MIB_RMADDR && remote_id != rm_id) 696 continue; 697 if (rm_type == MPTCP_MIB_RMSUBFLOW && id != rm_id) 698 continue; 699 700 pr_debug(" -> %s rm_list_ids[%d]=%u local_id=%u remote_id=%u mpc_id=%u\n", 701 rm_type == MPTCP_MIB_RMADDR ? "address" : "subflow", 702 i, rm_id, id, remote_id, msk->mpc_endpoint_id); 703 spin_unlock_bh(&msk->pm.lock); 704 mptcp_subflow_shutdown(sk, ssk, how); 705 removed |= subflow->request_join; 706 707 /* the following takes care of updating the subflows counter */ 708 mptcp_close_ssk(sk, ssk, subflow); 709 spin_lock_bh(&msk->pm.lock); 710 711 if (rm_type == MPTCP_MIB_RMSUBFLOW) 712 __MPTCP_INC_STATS(sock_net(sk), rm_type); 713 } 714 715 if (rm_type == MPTCP_MIB_RMADDR) { 716 __MPTCP_INC_STATS(sock_net(sk), rm_type); 717 if (removed && mptcp_pm_is_kernel(msk)) 718 mptcp_pm_nl_rm_addr(msk, rm_id); 719 } 720 } 721 } 722 723 static void mptcp_pm_rm_addr_recv(struct mptcp_sock *msk) 724 { 725 mptcp_pm_rm_addr_or_subflow(msk, &msk->pm.rm_list_rx, MPTCP_MIB_RMADDR); 726 } 727 728 void mptcp_pm_rm_subflow(struct mptcp_sock *msk, 729 const struct mptcp_rm_list *rm_list) 730 { 731 mptcp_pm_rm_addr_or_subflow(msk, rm_list, MPTCP_MIB_RMSUBFLOW); 732 } 733 734 void mptcp_pm_rm_addr_received(struct mptcp_sock *msk, 735 const struct mptcp_rm_list *rm_list) 736 { 737 struct mptcp_pm_data *pm = &msk->pm; 738 u8 i; 739 740 pr_debug("msk=%p remote_ids_nr=%d\n", msk, rm_list->nr); 741 742 for (i = 0; i < rm_list->nr; i++) 743 mptcp_event_addr_removed(msk, rm_list->ids[i]); 744 745 spin_lock_bh(&pm->lock); 746 if (mptcp_pm_schedule_work(msk, MPTCP_PM_RM_ADDR_RECEIVED)) 747 pm->rm_list_rx = *rm_list; 748 else 749 __MPTCP_INC_STATS(sock_net((struct sock *)msk), MPTCP_MIB_RMADDRDROP); 750 spin_unlock_bh(&pm->lock); 751 } 752 753 void mptcp_pm_mp_prio_received(struct sock *ssk, u8 bkup) 754 { 755 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk); 756 struct sock *sk = subflow->conn; 757 struct mptcp_sock *msk; 758 759 pr_debug("subflow->backup=%d, bkup=%d\n", subflow->backup, bkup); 760 msk = mptcp_sk(sk); 761 if (subflow->backup != bkup) 762 subflow->backup = bkup; 763 764 mptcp_event(MPTCP_EVENT_SUB_PRIORITY, msk, ssk, GFP_ATOMIC); 765 } 766 767 void mptcp_pm_mp_fail_received(struct sock *sk, u64 fail_seq) 768 { 769 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk); 770 struct mptcp_sock *msk = mptcp_sk(subflow->conn); 771 772 pr_debug("fail_seq=%llu\n", fail_seq); 773 774 /* After accepting the fail, we can't create any other subflows */ 775 spin_lock_bh(&msk->fallback_lock); 776 if (!msk->allow_infinite_fallback) { 777 spin_unlock_bh(&msk->fallback_lock); 778 return; 779 } 780 msk->allow_subflows = false; 781 spin_unlock_bh(&msk->fallback_lock); 782 783 if (!subflow->fail_tout) { 784 pr_debug("send MP_FAIL response and infinite map\n"); 785 786 subflow->send_mp_fail = 1; 787 subflow->send_infinite_map = 1; 788 tcp_send_ack(sk); 789 } else { 790 pr_debug("MP_FAIL response received\n"); 791 WRITE_ONCE(subflow->fail_tout, 0); 792 } 793 } 794 795 bool mptcp_pm_add_addr_signal(struct mptcp_sock *msk, const struct sk_buff *skb, 796 unsigned int opt_size, unsigned int remaining, 797 struct mptcp_addr_info *addr, bool *echo, 798 bool *drop_other_suboptions) 799 { 800 int ret = false; 801 u8 add_addr; 802 u8 family; 803 bool port; 804 805 spin_lock_bh(&msk->pm.lock); 806 807 /* double check after the lock is acquired */ 808 if (!mptcp_pm_should_add_signal(msk)) 809 goto out_unlock; 810 811 /* always drop every other options for pure ack ADD_ADDR; this is a 812 * plain dup-ack from TCP perspective. The other MPTCP-relevant info, 813 * if any, will be carried by the 'original' TCP ack 814 */ 815 if (skb && skb_is_tcp_pure_ack(skb)) { 816 remaining += opt_size; 817 *drop_other_suboptions = true; 818 } 819 820 *echo = mptcp_pm_should_add_signal_echo(msk); 821 port = !!(*echo ? msk->pm.remote.port : msk->pm.local.port); 822 823 family = *echo ? msk->pm.remote.family : msk->pm.local.family; 824 if (remaining < mptcp_add_addr_len(family, *echo, port)) 825 goto out_unlock; 826 827 if (*echo) { 828 *addr = msk->pm.remote; 829 add_addr = msk->pm.addr_signal & ~BIT(MPTCP_ADD_ADDR_ECHO); 830 } else { 831 *addr = msk->pm.local; 832 add_addr = msk->pm.addr_signal & ~BIT(MPTCP_ADD_ADDR_SIGNAL); 833 } 834 WRITE_ONCE(msk->pm.addr_signal, add_addr); 835 ret = true; 836 837 out_unlock: 838 spin_unlock_bh(&msk->pm.lock); 839 return ret; 840 } 841 842 bool mptcp_pm_rm_addr_signal(struct mptcp_sock *msk, unsigned int remaining, 843 struct mptcp_rm_list *rm_list) 844 { 845 int ret = false, len; 846 u8 rm_addr; 847 848 spin_lock_bh(&msk->pm.lock); 849 850 /* double check after the lock is acquired */ 851 if (!mptcp_pm_should_rm_signal(msk)) 852 goto out_unlock; 853 854 rm_addr = msk->pm.addr_signal & ~BIT(MPTCP_RM_ADDR_SIGNAL); 855 len = mptcp_rm_addr_len(&msk->pm.rm_list_tx); 856 if (len < 0) { 857 WRITE_ONCE(msk->pm.addr_signal, rm_addr); 858 goto out_unlock; 859 } 860 if (remaining < len) 861 goto out_unlock; 862 863 *rm_list = msk->pm.rm_list_tx; 864 WRITE_ONCE(msk->pm.addr_signal, rm_addr); 865 ret = true; 866 867 out_unlock: 868 spin_unlock_bh(&msk->pm.lock); 869 return ret; 870 } 871 872 int mptcp_pm_get_local_id(struct mptcp_sock *msk, struct sock_common *skc) 873 { 874 struct mptcp_pm_addr_entry skc_local = { 0 }; 875 struct mptcp_addr_info msk_local; 876 877 if (WARN_ON_ONCE(!msk)) 878 return -1; 879 880 /* The 0 ID mapping is defined by the first subflow, copied into the msk 881 * addr 882 */ 883 mptcp_local_address((struct sock_common *)msk, &msk_local); 884 mptcp_local_address((struct sock_common *)skc, &skc_local.addr); 885 if (mptcp_addresses_equal(&msk_local, &skc_local.addr, false)) 886 return 0; 887 888 skc_local.addr.id = 0; 889 skc_local.flags = MPTCP_PM_ADDR_FLAG_IMPLICIT; 890 891 if (mptcp_pm_is_userspace(msk)) 892 return mptcp_userspace_pm_get_local_id(msk, &skc_local); 893 return mptcp_pm_nl_get_local_id(msk, &skc_local); 894 } 895 896 bool mptcp_pm_is_backup(struct mptcp_sock *msk, struct sock_common *skc) 897 { 898 struct mptcp_addr_info skc_local; 899 900 mptcp_local_address((struct sock_common *)skc, &skc_local); 901 902 if (mptcp_pm_is_userspace(msk)) 903 return mptcp_userspace_pm_is_backup(msk, &skc_local); 904 905 return mptcp_pm_nl_is_backup(msk, &skc_local); 906 } 907 908 static void mptcp_pm_subflows_chk_stale(const struct mptcp_sock *msk, struct sock *ssk) 909 { 910 struct mptcp_subflow_context *iter, *subflow = mptcp_subflow_ctx(ssk); 911 struct sock *sk = (struct sock *)msk; 912 unsigned int active_max_loss_cnt; 913 struct net *net = sock_net(sk); 914 unsigned int stale_loss_cnt; 915 bool slow; 916 917 stale_loss_cnt = mptcp_stale_loss_cnt(net); 918 if (subflow->stale || !stale_loss_cnt || subflow->stale_count <= stale_loss_cnt) 919 return; 920 921 /* look for another available subflow not in loss state */ 922 active_max_loss_cnt = max_t(int, stale_loss_cnt - 1, 1); 923 mptcp_for_each_subflow(msk, iter) { 924 if (iter != subflow && mptcp_subflow_active(iter) && 925 iter->stale_count < active_max_loss_cnt) { 926 /* we have some alternatives, try to mark this subflow as idle ...*/ 927 slow = lock_sock_fast(ssk); 928 if (!tcp_rtx_and_write_queues_empty(ssk)) { 929 subflow->stale = 1; 930 __mptcp_retransmit_pending_data(sk); 931 MPTCP_INC_STATS(net, MPTCP_MIB_SUBFLOWSTALE); 932 } 933 unlock_sock_fast(ssk, slow); 934 935 /* always try to push the pending data regardless of re-injections: 936 * we can possibly use backup subflows now, and subflow selection 937 * is cheap under the msk socket lock 938 */ 939 __mptcp_push_pending(sk, 0); 940 return; 941 } 942 } 943 } 944 945 void mptcp_pm_subflow_chk_stale(const struct mptcp_sock *msk, struct sock *ssk) 946 { 947 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk); 948 u32 rcv_tstamp = READ_ONCE(tcp_sk(ssk)->rcv_tstamp); 949 950 /* keep track of rtx periods with no progress */ 951 if (!subflow->stale_count) { 952 subflow->stale_rcv_tstamp = rcv_tstamp; 953 subflow->stale_count++; 954 } else if (subflow->stale_rcv_tstamp == rcv_tstamp) { 955 if (subflow->stale_count < U8_MAX) 956 subflow->stale_count++; 957 mptcp_pm_subflows_chk_stale(msk, ssk); 958 } else { 959 subflow->stale_count = 0; 960 mptcp_subflow_set_active(subflow); 961 } 962 } 963 964 void mptcp_pm_worker(struct mptcp_sock *msk) 965 { 966 struct mptcp_pm_data *pm = &msk->pm; 967 968 msk_owned_by_me(msk); 969 970 if (!(pm->status & MPTCP_PM_WORK_MASK)) 971 return; 972 973 spin_lock_bh(&msk->pm.lock); 974 975 pr_debug("msk=%p status=%x\n", msk, pm->status); 976 if (pm->status & BIT(MPTCP_PM_ADD_ADDR_SEND_ACK)) { 977 pm->status &= ~BIT(MPTCP_PM_ADD_ADDR_SEND_ACK); 978 mptcp_pm_addr_send_ack(msk); 979 } 980 if (pm->status & BIT(MPTCP_PM_RM_ADDR_RECEIVED)) { 981 pm->status &= ~BIT(MPTCP_PM_RM_ADDR_RECEIVED); 982 mptcp_pm_rm_addr_recv(msk); 983 } 984 __mptcp_pm_kernel_worker(msk); 985 986 spin_unlock_bh(&msk->pm.lock); 987 } 988 989 void mptcp_pm_destroy(struct mptcp_sock *msk) 990 { 991 mptcp_pm_free_anno_list(msk); 992 993 if (mptcp_pm_is_userspace(msk)) 994 mptcp_userspace_pm_free_local_addr_list(msk); 995 } 996 997 void mptcp_pm_data_reset(struct mptcp_sock *msk) 998 { 999 u8 pm_type = mptcp_get_pm_type(sock_net((struct sock *)msk)); 1000 struct mptcp_pm_data *pm = &msk->pm; 1001 1002 memset(&pm->reset, 0, sizeof(pm->reset)); 1003 pm->rm_list_tx.nr = 0; 1004 pm->rm_list_rx.nr = 0; 1005 WRITE_ONCE(pm->pm_type, pm_type); 1006 1007 if (pm_type == MPTCP_PM_TYPE_KERNEL) { 1008 bool subflows_allowed = !!mptcp_pm_get_subflows_max(msk); 1009 1010 /* pm->work_pending must be only be set to 'true' when 1011 * pm->pm_type is set to MPTCP_PM_TYPE_KERNEL 1012 */ 1013 WRITE_ONCE(pm->work_pending, 1014 (!!mptcp_pm_get_local_addr_max(msk) && 1015 subflows_allowed) || 1016 !!mptcp_pm_get_add_addr_signal_max(msk)); 1017 WRITE_ONCE(pm->accept_addr, 1018 !!mptcp_pm_get_add_addr_accept_max(msk) && 1019 subflows_allowed); 1020 WRITE_ONCE(pm->accept_subflow, subflows_allowed); 1021 1022 bitmap_fill(pm->id_avail_bitmap, MPTCP_PM_MAX_ADDR_ID + 1); 1023 } 1024 } 1025 1026 void mptcp_pm_data_init(struct mptcp_sock *msk) 1027 { 1028 spin_lock_init(&msk->pm.lock); 1029 INIT_LIST_HEAD(&msk->pm.anno_list); 1030 INIT_LIST_HEAD(&msk->pm.userspace_pm_local_addr_list); 1031 mptcp_pm_data_reset(msk); 1032 } 1033 1034 void __init mptcp_pm_init(void) 1035 { 1036 mptcp_pm_kernel_register(); 1037 mptcp_pm_userspace_register(); 1038 mptcp_pm_nl_init(); 1039 } 1040 1041 /* Must be called with rcu read lock held */ 1042 struct mptcp_pm_ops *mptcp_pm_find(const char *name) 1043 { 1044 struct mptcp_pm_ops *pm_ops; 1045 1046 list_for_each_entry_rcu(pm_ops, &mptcp_pm_list, list) { 1047 if (!strcmp(pm_ops->name, name)) 1048 return pm_ops; 1049 } 1050 1051 return NULL; 1052 } 1053 1054 int mptcp_pm_validate(struct mptcp_pm_ops *pm_ops) 1055 { 1056 return 0; 1057 } 1058 1059 int mptcp_pm_register(struct mptcp_pm_ops *pm_ops) 1060 { 1061 int ret; 1062 1063 ret = mptcp_pm_validate(pm_ops); 1064 if (ret) 1065 return ret; 1066 1067 spin_lock(&mptcp_pm_list_lock); 1068 if (mptcp_pm_find(pm_ops->name)) { 1069 spin_unlock(&mptcp_pm_list_lock); 1070 return -EEXIST; 1071 } 1072 list_add_tail_rcu(&pm_ops->list, &mptcp_pm_list); 1073 spin_unlock(&mptcp_pm_list_lock); 1074 1075 pr_debug("%s registered\n", pm_ops->name); 1076 return 0; 1077 } 1078 1079 void mptcp_pm_unregister(struct mptcp_pm_ops *pm_ops) 1080 { 1081 /* skip unregistering the default path manager */ 1082 if (WARN_ON_ONCE(pm_ops == &mptcp_pm_kernel)) 1083 return; 1084 1085 spin_lock(&mptcp_pm_list_lock); 1086 list_del_rcu(&pm_ops->list); 1087 spin_unlock(&mptcp_pm_list_lock); 1088 } 1089 1090 /* Build string with list of available path manager values. 1091 * Similar to tcp_get_available_congestion_control() 1092 */ 1093 void mptcp_pm_get_available(char *buf, size_t maxlen) 1094 { 1095 struct mptcp_pm_ops *pm_ops; 1096 size_t offs = 0; 1097 1098 rcu_read_lock(); 1099 list_for_each_entry_rcu(pm_ops, &mptcp_pm_list, list) { 1100 offs += snprintf(buf + offs, maxlen - offs, "%s%s", 1101 offs == 0 ? "" : " ", pm_ops->name); 1102 1103 if (WARN_ON_ONCE(offs >= maxlen)) 1104 break; 1105 } 1106 rcu_read_unlock(); 1107 } 1108