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 unsigned int mptcp_adjust_add_addr_timeout(struct mptcp_sock *msk) 272 { 273 const struct net *net = sock_net((struct sock *)msk); 274 unsigned int rto = mptcp_get_add_addr_timeout(net); 275 struct mptcp_subflow_context *subflow; 276 unsigned int max = 0; 277 278 mptcp_for_each_subflow(msk, subflow) { 279 struct sock *ssk = mptcp_subflow_tcp_sock(subflow); 280 struct inet_connection_sock *icsk = inet_csk(ssk); 281 282 if (icsk->icsk_rto > max) 283 max = icsk->icsk_rto; 284 } 285 286 if (max && max < rto) 287 rto = max; 288 289 return rto; 290 } 291 292 static void mptcp_pm_add_timer(struct timer_list *timer) 293 { 294 struct mptcp_pm_add_entry *entry = timer_container_of(entry, timer, 295 add_timer); 296 struct mptcp_sock *msk = entry->sock; 297 struct sock *sk = (struct sock *)msk; 298 unsigned int timeout; 299 300 pr_debug("msk=%p\n", msk); 301 302 if (!msk) 303 return; 304 305 if (inet_sk_state_load(sk) == TCP_CLOSE) 306 return; 307 308 if (!entry->addr.id) 309 return; 310 311 if (mptcp_pm_should_add_signal_addr(msk)) { 312 sk_reset_timer(sk, timer, jiffies + TCP_RTO_MAX / 8); 313 goto out; 314 } 315 316 timeout = mptcp_adjust_add_addr_timeout(msk); 317 if (!timeout) 318 goto out; 319 320 spin_lock_bh(&msk->pm.lock); 321 322 if (!mptcp_pm_should_add_signal_addr(msk)) { 323 pr_debug("retransmit ADD_ADDR id=%d\n", entry->addr.id); 324 mptcp_pm_announce_addr(msk, &entry->addr, false); 325 mptcp_pm_add_addr_send_ack(msk); 326 entry->retrans_times++; 327 } 328 329 if (entry->retrans_times < ADD_ADDR_RETRANS_MAX) 330 sk_reset_timer(sk, timer, 331 jiffies + (timeout << entry->retrans_times)); 332 333 spin_unlock_bh(&msk->pm.lock); 334 335 if (entry->retrans_times == ADD_ADDR_RETRANS_MAX) 336 mptcp_pm_subflow_established(msk); 337 338 out: 339 __sock_put(sk); 340 } 341 342 struct mptcp_pm_add_entry * 343 mptcp_pm_del_add_timer(struct mptcp_sock *msk, 344 const struct mptcp_addr_info *addr, bool check_id) 345 { 346 struct mptcp_pm_add_entry *entry; 347 struct sock *sk = (struct sock *)msk; 348 struct timer_list *add_timer = NULL; 349 350 spin_lock_bh(&msk->pm.lock); 351 entry = mptcp_lookup_anno_list_by_saddr(msk, addr); 352 if (entry && (!check_id || entry->addr.id == addr->id)) { 353 entry->retrans_times = ADD_ADDR_RETRANS_MAX; 354 add_timer = &entry->add_timer; 355 } 356 if (!check_id && entry) 357 list_del(&entry->list); 358 spin_unlock_bh(&msk->pm.lock); 359 360 /* no lock, because sk_stop_timer_sync() is calling timer_delete_sync() */ 361 if (add_timer) 362 sk_stop_timer_sync(sk, add_timer); 363 364 return entry; 365 } 366 367 bool mptcp_pm_alloc_anno_list(struct mptcp_sock *msk, 368 const struct mptcp_addr_info *addr) 369 { 370 struct mptcp_pm_add_entry *add_entry = NULL; 371 struct sock *sk = (struct sock *)msk; 372 unsigned int timeout; 373 374 lockdep_assert_held(&msk->pm.lock); 375 376 add_entry = mptcp_lookup_anno_list_by_saddr(msk, addr); 377 378 if (add_entry) { 379 if (WARN_ON_ONCE(mptcp_pm_is_kernel(msk))) 380 return false; 381 382 goto reset_timer; 383 } 384 385 add_entry = kmalloc(sizeof(*add_entry), GFP_ATOMIC); 386 if (!add_entry) 387 return false; 388 389 list_add(&add_entry->list, &msk->pm.anno_list); 390 391 add_entry->addr = *addr; 392 add_entry->sock = msk; 393 add_entry->retrans_times = 0; 394 395 timer_setup(&add_entry->add_timer, mptcp_pm_add_timer, 0); 396 reset_timer: 397 timeout = mptcp_adjust_add_addr_timeout(msk); 398 if (timeout) 399 sk_reset_timer(sk, &add_entry->add_timer, jiffies + timeout); 400 401 return true; 402 } 403 404 static void mptcp_pm_free_anno_list(struct mptcp_sock *msk) 405 { 406 struct mptcp_pm_add_entry *entry, *tmp; 407 struct sock *sk = (struct sock *)msk; 408 LIST_HEAD(free_list); 409 410 pr_debug("msk=%p\n", msk); 411 412 spin_lock_bh(&msk->pm.lock); 413 list_splice_init(&msk->pm.anno_list, &free_list); 414 spin_unlock_bh(&msk->pm.lock); 415 416 list_for_each_entry_safe(entry, tmp, &free_list, list) { 417 sk_stop_timer_sync(sk, &entry->add_timer); 418 kfree(entry); 419 } 420 } 421 422 /* path manager command handlers */ 423 424 int mptcp_pm_announce_addr(struct mptcp_sock *msk, 425 const struct mptcp_addr_info *addr, 426 bool echo) 427 { 428 u8 add_addr = READ_ONCE(msk->pm.addr_signal); 429 430 pr_debug("msk=%p, local_id=%d, echo=%d\n", msk, addr->id, echo); 431 432 lockdep_assert_held(&msk->pm.lock); 433 434 if (add_addr & 435 (echo ? BIT(MPTCP_ADD_ADDR_ECHO) : BIT(MPTCP_ADD_ADDR_SIGNAL))) { 436 MPTCP_INC_STATS(sock_net((struct sock *)msk), 437 echo ? MPTCP_MIB_ECHOADDTXDROP : MPTCP_MIB_ADDADDRTXDROP); 438 return -EINVAL; 439 } 440 441 if (echo) { 442 msk->pm.remote = *addr; 443 add_addr |= BIT(MPTCP_ADD_ADDR_ECHO); 444 } else { 445 msk->pm.local = *addr; 446 add_addr |= BIT(MPTCP_ADD_ADDR_SIGNAL); 447 } 448 WRITE_ONCE(msk->pm.addr_signal, add_addr); 449 return 0; 450 } 451 452 int mptcp_pm_remove_addr(struct mptcp_sock *msk, const struct mptcp_rm_list *rm_list) 453 { 454 u8 rm_addr = READ_ONCE(msk->pm.addr_signal); 455 456 pr_debug("msk=%p, rm_list_nr=%d\n", msk, rm_list->nr); 457 458 if (rm_addr) { 459 MPTCP_ADD_STATS(sock_net((struct sock *)msk), 460 MPTCP_MIB_RMADDRTXDROP, rm_list->nr); 461 return -EINVAL; 462 } 463 464 msk->pm.rm_list_tx = *rm_list; 465 rm_addr |= BIT(MPTCP_RM_ADDR_SIGNAL); 466 WRITE_ONCE(msk->pm.addr_signal, rm_addr); 467 mptcp_pm_addr_send_ack(msk); 468 return 0; 469 } 470 471 /* path manager event handlers */ 472 473 void mptcp_pm_new_connection(struct mptcp_sock *msk, const struct sock *ssk, int server_side) 474 { 475 struct mptcp_pm_data *pm = &msk->pm; 476 477 pr_debug("msk=%p, token=%u side=%d\n", msk, READ_ONCE(msk->token), server_side); 478 479 WRITE_ONCE(pm->server_side, server_side); 480 mptcp_event(MPTCP_EVENT_CREATED, msk, ssk, GFP_ATOMIC); 481 } 482 483 bool mptcp_pm_allow_new_subflow(struct mptcp_sock *msk) 484 { 485 struct mptcp_pm_data *pm = &msk->pm; 486 unsigned int subflows_max; 487 int ret = 0; 488 489 if (mptcp_pm_is_userspace(msk)) { 490 if (mptcp_userspace_pm_active(msk)) { 491 spin_lock_bh(&pm->lock); 492 pm->subflows++; 493 spin_unlock_bh(&pm->lock); 494 return true; 495 } 496 return false; 497 } 498 499 subflows_max = mptcp_pm_get_subflows_max(msk); 500 501 pr_debug("msk=%p subflows=%d max=%d allow=%d\n", msk, pm->subflows, 502 subflows_max, READ_ONCE(pm->accept_subflow)); 503 504 /* try to avoid acquiring the lock below */ 505 if (!READ_ONCE(pm->accept_subflow)) 506 return false; 507 508 spin_lock_bh(&pm->lock); 509 if (READ_ONCE(pm->accept_subflow)) { 510 ret = pm->subflows < subflows_max; 511 if (ret && ++pm->subflows == subflows_max) 512 WRITE_ONCE(pm->accept_subflow, false); 513 } 514 spin_unlock_bh(&pm->lock); 515 516 return ret; 517 } 518 519 /* return true if the new status bit is currently cleared, that is, this event 520 * can be server, eventually by an already scheduled work 521 */ 522 static bool mptcp_pm_schedule_work(struct mptcp_sock *msk, 523 enum mptcp_pm_status new_status) 524 { 525 pr_debug("msk=%p status=%x new=%lx\n", msk, msk->pm.status, 526 BIT(new_status)); 527 if (msk->pm.status & BIT(new_status)) 528 return false; 529 530 msk->pm.status |= BIT(new_status); 531 mptcp_schedule_work((struct sock *)msk); 532 return true; 533 } 534 535 void mptcp_pm_fully_established(struct mptcp_sock *msk, const struct sock *ssk) 536 { 537 struct mptcp_pm_data *pm = &msk->pm; 538 bool announce = false; 539 540 pr_debug("msk=%p\n", msk); 541 542 spin_lock_bh(&pm->lock); 543 544 /* mptcp_pm_fully_established() can be invoked by multiple 545 * racing paths - accept() and check_fully_established() 546 * be sure to serve this event only once. 547 */ 548 if (READ_ONCE(pm->work_pending) && 549 !(pm->status & BIT(MPTCP_PM_ALREADY_ESTABLISHED))) 550 mptcp_pm_schedule_work(msk, MPTCP_PM_ESTABLISHED); 551 552 if ((pm->status & BIT(MPTCP_PM_ALREADY_ESTABLISHED)) == 0) 553 announce = true; 554 555 pm->status |= BIT(MPTCP_PM_ALREADY_ESTABLISHED); 556 spin_unlock_bh(&pm->lock); 557 558 if (announce) 559 mptcp_event(MPTCP_EVENT_ESTABLISHED, msk, ssk, GFP_ATOMIC); 560 } 561 562 void mptcp_pm_connection_closed(struct mptcp_sock *msk) 563 { 564 pr_debug("msk=%p\n", msk); 565 566 if (msk->token) 567 mptcp_event(MPTCP_EVENT_CLOSED, msk, NULL, GFP_KERNEL); 568 } 569 570 void mptcp_pm_subflow_established(struct mptcp_sock *msk) 571 { 572 struct mptcp_pm_data *pm = &msk->pm; 573 574 pr_debug("msk=%p\n", msk); 575 576 if (!READ_ONCE(pm->work_pending)) 577 return; 578 579 spin_lock_bh(&pm->lock); 580 581 if (READ_ONCE(pm->work_pending)) 582 mptcp_pm_schedule_work(msk, MPTCP_PM_SUBFLOW_ESTABLISHED); 583 584 spin_unlock_bh(&pm->lock); 585 } 586 587 void mptcp_pm_subflow_check_next(struct mptcp_sock *msk, 588 const struct mptcp_subflow_context *subflow) 589 { 590 struct mptcp_pm_data *pm = &msk->pm; 591 bool update_subflows; 592 593 update_subflows = subflow->request_join || subflow->mp_join; 594 if (mptcp_pm_is_userspace(msk)) { 595 if (update_subflows) { 596 spin_lock_bh(&pm->lock); 597 pm->subflows--; 598 spin_unlock_bh(&pm->lock); 599 } 600 return; 601 } 602 603 if (!READ_ONCE(pm->work_pending) && !update_subflows) 604 return; 605 606 spin_lock_bh(&pm->lock); 607 if (update_subflows) 608 __mptcp_pm_close_subflow(msk); 609 610 /* Even if this subflow is not really established, tell the PM to try 611 * to pick the next ones, if possible. 612 */ 613 if (mptcp_pm_nl_check_work_pending(msk)) 614 mptcp_pm_schedule_work(msk, MPTCP_PM_SUBFLOW_ESTABLISHED); 615 616 spin_unlock_bh(&pm->lock); 617 } 618 619 void mptcp_pm_add_addr_received(const struct sock *ssk, 620 const struct mptcp_addr_info *addr) 621 { 622 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk); 623 struct mptcp_sock *msk = mptcp_sk(subflow->conn); 624 struct mptcp_pm_data *pm = &msk->pm; 625 626 pr_debug("msk=%p remote_id=%d accept=%d\n", msk, addr->id, 627 READ_ONCE(pm->accept_addr)); 628 629 mptcp_event_addr_announced(ssk, addr); 630 631 spin_lock_bh(&pm->lock); 632 633 if (mptcp_pm_is_userspace(msk)) { 634 if (mptcp_userspace_pm_active(msk)) { 635 mptcp_pm_announce_addr(msk, addr, true); 636 mptcp_pm_add_addr_send_ack(msk); 637 } else { 638 __MPTCP_INC_STATS(sock_net((struct sock *)msk), MPTCP_MIB_ADDADDRDROP); 639 } 640 /* id0 should not have a different address */ 641 } else if ((addr->id == 0 && !mptcp_pm_is_init_remote_addr(msk, addr)) || 642 (addr->id > 0 && !READ_ONCE(pm->accept_addr))) { 643 mptcp_pm_announce_addr(msk, addr, true); 644 mptcp_pm_add_addr_send_ack(msk); 645 } else if (mptcp_pm_schedule_work(msk, MPTCP_PM_ADD_ADDR_RECEIVED)) { 646 pm->remote = *addr; 647 } else { 648 __MPTCP_INC_STATS(sock_net((struct sock *)msk), MPTCP_MIB_ADDADDRDROP); 649 } 650 651 spin_unlock_bh(&pm->lock); 652 } 653 654 void mptcp_pm_add_addr_echoed(struct mptcp_sock *msk, 655 const struct mptcp_addr_info *addr) 656 { 657 struct mptcp_pm_data *pm = &msk->pm; 658 659 pr_debug("msk=%p\n", msk); 660 661 if (!READ_ONCE(pm->work_pending)) 662 return; 663 664 spin_lock_bh(&pm->lock); 665 666 if (mptcp_lookup_anno_list_by_saddr(msk, addr) && READ_ONCE(pm->work_pending)) 667 mptcp_pm_schedule_work(msk, MPTCP_PM_SUBFLOW_ESTABLISHED); 668 669 spin_unlock_bh(&pm->lock); 670 } 671 672 void mptcp_pm_add_addr_send_ack(struct mptcp_sock *msk) 673 { 674 if (!mptcp_pm_should_add_signal(msk)) 675 return; 676 677 mptcp_pm_schedule_work(msk, MPTCP_PM_ADD_ADDR_SEND_ACK); 678 } 679 680 static void mptcp_pm_rm_addr_or_subflow(struct mptcp_sock *msk, 681 const struct mptcp_rm_list *rm_list, 682 enum linux_mptcp_mib_field rm_type) 683 { 684 struct mptcp_subflow_context *subflow, *tmp; 685 struct sock *sk = (struct sock *)msk; 686 u8 i; 687 688 pr_debug("%s rm_list_nr %d\n", 689 rm_type == MPTCP_MIB_RMADDR ? "address" : "subflow", rm_list->nr); 690 691 msk_owned_by_me(msk); 692 693 if (sk->sk_state == TCP_LISTEN) 694 return; 695 696 if (!rm_list->nr) 697 return; 698 699 if (list_empty(&msk->conn_list)) 700 return; 701 702 for (i = 0; i < rm_list->nr; i++) { 703 u8 rm_id = rm_list->ids[i]; 704 bool removed = false; 705 706 mptcp_for_each_subflow_safe(msk, subflow, tmp) { 707 struct sock *ssk = mptcp_subflow_tcp_sock(subflow); 708 u8 remote_id = READ_ONCE(subflow->remote_id); 709 int how = RCV_SHUTDOWN | SEND_SHUTDOWN; 710 u8 id = subflow_get_local_id(subflow); 711 712 if ((1 << inet_sk_state_load(ssk)) & 713 (TCPF_FIN_WAIT1 | TCPF_FIN_WAIT2 | TCPF_CLOSING | TCPF_CLOSE)) 714 continue; 715 if (rm_type == MPTCP_MIB_RMADDR && remote_id != rm_id) 716 continue; 717 if (rm_type == MPTCP_MIB_RMSUBFLOW && id != rm_id) 718 continue; 719 720 pr_debug(" -> %s rm_list_ids[%d]=%u local_id=%u remote_id=%u mpc_id=%u\n", 721 rm_type == MPTCP_MIB_RMADDR ? "address" : "subflow", 722 i, rm_id, id, remote_id, msk->mpc_endpoint_id); 723 spin_unlock_bh(&msk->pm.lock); 724 mptcp_subflow_shutdown(sk, ssk, how); 725 removed |= subflow->request_join; 726 727 /* the following takes care of updating the subflows counter */ 728 mptcp_close_ssk(sk, ssk, subflow); 729 spin_lock_bh(&msk->pm.lock); 730 731 if (rm_type == MPTCP_MIB_RMSUBFLOW) 732 __MPTCP_INC_STATS(sock_net(sk), rm_type); 733 } 734 735 if (rm_type == MPTCP_MIB_RMADDR) { 736 __MPTCP_INC_STATS(sock_net(sk), rm_type); 737 if (removed && mptcp_pm_is_kernel(msk)) 738 mptcp_pm_nl_rm_addr(msk, rm_id); 739 } 740 } 741 } 742 743 static void mptcp_pm_rm_addr_recv(struct mptcp_sock *msk) 744 { 745 mptcp_pm_rm_addr_or_subflow(msk, &msk->pm.rm_list_rx, MPTCP_MIB_RMADDR); 746 } 747 748 void mptcp_pm_rm_subflow(struct mptcp_sock *msk, 749 const struct mptcp_rm_list *rm_list) 750 { 751 mptcp_pm_rm_addr_or_subflow(msk, rm_list, MPTCP_MIB_RMSUBFLOW); 752 } 753 754 void mptcp_pm_rm_addr_received(struct mptcp_sock *msk, 755 const struct mptcp_rm_list *rm_list) 756 { 757 struct mptcp_pm_data *pm = &msk->pm; 758 u8 i; 759 760 pr_debug("msk=%p remote_ids_nr=%d\n", msk, rm_list->nr); 761 762 for (i = 0; i < rm_list->nr; i++) 763 mptcp_event_addr_removed(msk, rm_list->ids[i]); 764 765 spin_lock_bh(&pm->lock); 766 if (mptcp_pm_schedule_work(msk, MPTCP_PM_RM_ADDR_RECEIVED)) 767 pm->rm_list_rx = *rm_list; 768 else 769 __MPTCP_INC_STATS(sock_net((struct sock *)msk), MPTCP_MIB_RMADDRDROP); 770 spin_unlock_bh(&pm->lock); 771 } 772 773 void mptcp_pm_mp_prio_received(struct sock *ssk, u8 bkup) 774 { 775 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk); 776 struct sock *sk = subflow->conn; 777 struct mptcp_sock *msk; 778 779 pr_debug("subflow->backup=%d, bkup=%d\n", subflow->backup, bkup); 780 msk = mptcp_sk(sk); 781 if (subflow->backup != bkup) 782 subflow->backup = bkup; 783 784 mptcp_event(MPTCP_EVENT_SUB_PRIORITY, msk, ssk, GFP_ATOMIC); 785 } 786 787 void mptcp_pm_mp_fail_received(struct sock *sk, u64 fail_seq) 788 { 789 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk); 790 struct mptcp_sock *msk = mptcp_sk(subflow->conn); 791 792 pr_debug("fail_seq=%llu\n", fail_seq); 793 794 /* After accepting the fail, we can't create any other subflows */ 795 spin_lock_bh(&msk->fallback_lock); 796 if (!msk->allow_infinite_fallback) { 797 spin_unlock_bh(&msk->fallback_lock); 798 return; 799 } 800 msk->allow_subflows = false; 801 spin_unlock_bh(&msk->fallback_lock); 802 803 if (!subflow->fail_tout) { 804 pr_debug("send MP_FAIL response and infinite map\n"); 805 806 subflow->send_mp_fail = 1; 807 subflow->send_infinite_map = 1; 808 tcp_send_ack(sk); 809 } else { 810 pr_debug("MP_FAIL response received\n"); 811 WRITE_ONCE(subflow->fail_tout, 0); 812 } 813 } 814 815 bool mptcp_pm_add_addr_signal(struct mptcp_sock *msk, const struct sk_buff *skb, 816 unsigned int opt_size, unsigned int remaining, 817 struct mptcp_addr_info *addr, bool *echo, 818 bool *drop_other_suboptions) 819 { 820 int ret = false; 821 u8 add_addr; 822 u8 family; 823 bool port; 824 825 spin_lock_bh(&msk->pm.lock); 826 827 /* double check after the lock is acquired */ 828 if (!mptcp_pm_should_add_signal(msk)) 829 goto out_unlock; 830 831 /* always drop every other options for pure ack ADD_ADDR; this is a 832 * plain dup-ack from TCP perspective. The other MPTCP-relevant info, 833 * if any, will be carried by the 'original' TCP ack 834 */ 835 if (skb && skb_is_tcp_pure_ack(skb)) { 836 remaining += opt_size; 837 *drop_other_suboptions = true; 838 } 839 840 *echo = mptcp_pm_should_add_signal_echo(msk); 841 port = !!(*echo ? msk->pm.remote.port : msk->pm.local.port); 842 843 family = *echo ? msk->pm.remote.family : msk->pm.local.family; 844 if (remaining < mptcp_add_addr_len(family, *echo, port)) 845 goto out_unlock; 846 847 if (*echo) { 848 *addr = msk->pm.remote; 849 add_addr = msk->pm.addr_signal & ~BIT(MPTCP_ADD_ADDR_ECHO); 850 } else { 851 *addr = msk->pm.local; 852 add_addr = msk->pm.addr_signal & ~BIT(MPTCP_ADD_ADDR_SIGNAL); 853 } 854 WRITE_ONCE(msk->pm.addr_signal, add_addr); 855 ret = true; 856 857 out_unlock: 858 spin_unlock_bh(&msk->pm.lock); 859 return ret; 860 } 861 862 bool mptcp_pm_rm_addr_signal(struct mptcp_sock *msk, unsigned int remaining, 863 struct mptcp_rm_list *rm_list) 864 { 865 int ret = false, len; 866 u8 rm_addr; 867 868 spin_lock_bh(&msk->pm.lock); 869 870 /* double check after the lock is acquired */ 871 if (!mptcp_pm_should_rm_signal(msk)) 872 goto out_unlock; 873 874 rm_addr = msk->pm.addr_signal & ~BIT(MPTCP_RM_ADDR_SIGNAL); 875 len = mptcp_rm_addr_len(&msk->pm.rm_list_tx); 876 if (len < 0) { 877 WRITE_ONCE(msk->pm.addr_signal, rm_addr); 878 goto out_unlock; 879 } 880 if (remaining < len) 881 goto out_unlock; 882 883 *rm_list = msk->pm.rm_list_tx; 884 WRITE_ONCE(msk->pm.addr_signal, rm_addr); 885 ret = true; 886 887 out_unlock: 888 spin_unlock_bh(&msk->pm.lock); 889 return ret; 890 } 891 892 int mptcp_pm_get_local_id(struct mptcp_sock *msk, struct sock_common *skc) 893 { 894 struct mptcp_pm_addr_entry skc_local = { 0 }; 895 struct mptcp_addr_info msk_local; 896 897 if (WARN_ON_ONCE(!msk)) 898 return -1; 899 900 /* The 0 ID mapping is defined by the first subflow, copied into the msk 901 * addr 902 */ 903 mptcp_local_address((struct sock_common *)msk, &msk_local); 904 mptcp_local_address((struct sock_common *)skc, &skc_local.addr); 905 if (mptcp_addresses_equal(&msk_local, &skc_local.addr, false)) 906 return 0; 907 908 skc_local.addr.id = 0; 909 skc_local.flags = MPTCP_PM_ADDR_FLAG_IMPLICIT; 910 911 if (mptcp_pm_is_userspace(msk)) 912 return mptcp_userspace_pm_get_local_id(msk, &skc_local); 913 return mptcp_pm_nl_get_local_id(msk, &skc_local); 914 } 915 916 bool mptcp_pm_is_backup(struct mptcp_sock *msk, struct sock_common *skc) 917 { 918 struct mptcp_addr_info skc_local; 919 920 mptcp_local_address((struct sock_common *)skc, &skc_local); 921 922 if (mptcp_pm_is_userspace(msk)) 923 return mptcp_userspace_pm_is_backup(msk, &skc_local); 924 925 return mptcp_pm_nl_is_backup(msk, &skc_local); 926 } 927 928 static void mptcp_pm_subflows_chk_stale(const struct mptcp_sock *msk, struct sock *ssk) 929 { 930 struct mptcp_subflow_context *iter, *subflow = mptcp_subflow_ctx(ssk); 931 struct sock *sk = (struct sock *)msk; 932 unsigned int active_max_loss_cnt; 933 struct net *net = sock_net(sk); 934 unsigned int stale_loss_cnt; 935 bool slow; 936 937 stale_loss_cnt = mptcp_stale_loss_cnt(net); 938 if (subflow->stale || !stale_loss_cnt || subflow->stale_count <= stale_loss_cnt) 939 return; 940 941 /* look for another available subflow not in loss state */ 942 active_max_loss_cnt = max_t(int, stale_loss_cnt - 1, 1); 943 mptcp_for_each_subflow(msk, iter) { 944 if (iter != subflow && mptcp_subflow_active(iter) && 945 iter->stale_count < active_max_loss_cnt) { 946 /* we have some alternatives, try to mark this subflow as idle ...*/ 947 slow = lock_sock_fast(ssk); 948 if (!tcp_rtx_and_write_queues_empty(ssk)) { 949 subflow->stale = 1; 950 __mptcp_retransmit_pending_data(sk); 951 MPTCP_INC_STATS(net, MPTCP_MIB_SUBFLOWSTALE); 952 } 953 unlock_sock_fast(ssk, slow); 954 955 /* always try to push the pending data regardless of re-injections: 956 * we can possibly use backup subflows now, and subflow selection 957 * is cheap under the msk socket lock 958 */ 959 __mptcp_push_pending(sk, 0); 960 return; 961 } 962 } 963 } 964 965 void mptcp_pm_subflow_chk_stale(const struct mptcp_sock *msk, struct sock *ssk) 966 { 967 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk); 968 u32 rcv_tstamp = READ_ONCE(tcp_sk(ssk)->rcv_tstamp); 969 970 /* keep track of rtx periods with no progress */ 971 if (!subflow->stale_count) { 972 subflow->stale_rcv_tstamp = rcv_tstamp; 973 subflow->stale_count++; 974 } else if (subflow->stale_rcv_tstamp == rcv_tstamp) { 975 if (subflow->stale_count < U8_MAX) 976 subflow->stale_count++; 977 mptcp_pm_subflows_chk_stale(msk, ssk); 978 } else { 979 subflow->stale_count = 0; 980 mptcp_subflow_set_active(subflow); 981 } 982 } 983 984 void mptcp_pm_worker(struct mptcp_sock *msk) 985 { 986 struct mptcp_pm_data *pm = &msk->pm; 987 988 msk_owned_by_me(msk); 989 990 if (!(pm->status & MPTCP_PM_WORK_MASK)) 991 return; 992 993 spin_lock_bh(&msk->pm.lock); 994 995 pr_debug("msk=%p status=%x\n", msk, pm->status); 996 if (pm->status & BIT(MPTCP_PM_ADD_ADDR_SEND_ACK)) { 997 pm->status &= ~BIT(MPTCP_PM_ADD_ADDR_SEND_ACK); 998 mptcp_pm_addr_send_ack(msk); 999 } 1000 if (pm->status & BIT(MPTCP_PM_RM_ADDR_RECEIVED)) { 1001 pm->status &= ~BIT(MPTCP_PM_RM_ADDR_RECEIVED); 1002 mptcp_pm_rm_addr_recv(msk); 1003 } 1004 __mptcp_pm_kernel_worker(msk); 1005 1006 spin_unlock_bh(&msk->pm.lock); 1007 } 1008 1009 void mptcp_pm_destroy(struct mptcp_sock *msk) 1010 { 1011 mptcp_pm_free_anno_list(msk); 1012 1013 if (mptcp_pm_is_userspace(msk)) 1014 mptcp_userspace_pm_free_local_addr_list(msk); 1015 } 1016 1017 void mptcp_pm_data_reset(struct mptcp_sock *msk) 1018 { 1019 u8 pm_type = mptcp_get_pm_type(sock_net((struct sock *)msk)); 1020 struct mptcp_pm_data *pm = &msk->pm; 1021 1022 memset(&pm->reset, 0, sizeof(pm->reset)); 1023 pm->rm_list_tx.nr = 0; 1024 pm->rm_list_rx.nr = 0; 1025 WRITE_ONCE(pm->pm_type, pm_type); 1026 1027 if (pm_type == MPTCP_PM_TYPE_KERNEL) { 1028 bool subflows_allowed = !!mptcp_pm_get_subflows_max(msk); 1029 1030 /* pm->work_pending must be only be set to 'true' when 1031 * pm->pm_type is set to MPTCP_PM_TYPE_KERNEL 1032 */ 1033 WRITE_ONCE(pm->work_pending, 1034 (!!mptcp_pm_get_local_addr_max(msk) && 1035 subflows_allowed) || 1036 !!mptcp_pm_get_add_addr_signal_max(msk)); 1037 WRITE_ONCE(pm->accept_addr, 1038 !!mptcp_pm_get_add_addr_accept_max(msk) && 1039 subflows_allowed); 1040 WRITE_ONCE(pm->accept_subflow, subflows_allowed); 1041 1042 bitmap_fill(pm->id_avail_bitmap, MPTCP_PM_MAX_ADDR_ID + 1); 1043 } 1044 } 1045 1046 void mptcp_pm_data_init(struct mptcp_sock *msk) 1047 { 1048 spin_lock_init(&msk->pm.lock); 1049 INIT_LIST_HEAD(&msk->pm.anno_list); 1050 INIT_LIST_HEAD(&msk->pm.userspace_pm_local_addr_list); 1051 mptcp_pm_data_reset(msk); 1052 } 1053 1054 void __init mptcp_pm_init(void) 1055 { 1056 mptcp_pm_kernel_register(); 1057 mptcp_pm_userspace_register(); 1058 mptcp_pm_nl_init(); 1059 } 1060 1061 /* Must be called with rcu read lock held */ 1062 struct mptcp_pm_ops *mptcp_pm_find(const char *name) 1063 { 1064 struct mptcp_pm_ops *pm_ops; 1065 1066 list_for_each_entry_rcu(pm_ops, &mptcp_pm_list, list) { 1067 if (!strcmp(pm_ops->name, name)) 1068 return pm_ops; 1069 } 1070 1071 return NULL; 1072 } 1073 1074 int mptcp_pm_validate(struct mptcp_pm_ops *pm_ops) 1075 { 1076 return 0; 1077 } 1078 1079 int mptcp_pm_register(struct mptcp_pm_ops *pm_ops) 1080 { 1081 int ret; 1082 1083 ret = mptcp_pm_validate(pm_ops); 1084 if (ret) 1085 return ret; 1086 1087 spin_lock(&mptcp_pm_list_lock); 1088 if (mptcp_pm_find(pm_ops->name)) { 1089 spin_unlock(&mptcp_pm_list_lock); 1090 return -EEXIST; 1091 } 1092 list_add_tail_rcu(&pm_ops->list, &mptcp_pm_list); 1093 spin_unlock(&mptcp_pm_list_lock); 1094 1095 pr_debug("%s registered\n", pm_ops->name); 1096 return 0; 1097 } 1098 1099 void mptcp_pm_unregister(struct mptcp_pm_ops *pm_ops) 1100 { 1101 /* skip unregistering the default path manager */ 1102 if (WARN_ON_ONCE(pm_ops == &mptcp_pm_kernel)) 1103 return; 1104 1105 spin_lock(&mptcp_pm_list_lock); 1106 list_del_rcu(&pm_ops->list); 1107 spin_unlock(&mptcp_pm_list_lock); 1108 } 1109 1110 /* Build string with list of available path manager values. 1111 * Similar to tcp_get_available_congestion_control() 1112 */ 1113 void mptcp_pm_get_available(char *buf, size_t maxlen) 1114 { 1115 struct mptcp_pm_ops *pm_ops; 1116 size_t offs = 0; 1117 1118 rcu_read_lock(); 1119 list_for_each_entry_rcu(pm_ops, &mptcp_pm_list, list) { 1120 offs += snprintf(buf + offs, maxlen - offs, "%s%s", 1121 offs == 0 ? "" : " ", pm_ops->name); 1122 1123 if (WARN_ON_ONCE(offs >= maxlen)) 1124 break; 1125 } 1126 rcu_read_unlock(); 1127 } 1128