1 // SPDX-License-Identifier: GPL-2.0 2 /* Multipath TCP 3 * 4 * Copyright (c) 2020, Red Hat, Inc. 5 */ 6 7 #define pr_fmt(fmt) "MPTCP: " fmt 8 9 #include <linux/inet.h> 10 #include <linux/kernel.h> 11 #include <net/inet_common.h> 12 #include <net/netns/generic.h> 13 #include <net/mptcp.h> 14 15 #include "protocol.h" 16 #include "mib.h" 17 #include "mptcp_pm_gen.h" 18 19 static int pm_nl_pernet_id; 20 21 struct mptcp_pm_add_entry { 22 struct list_head list; 23 struct mptcp_addr_info addr; 24 u8 retrans_times; 25 struct timer_list add_timer; 26 struct mptcp_sock *sock; 27 }; 28 29 struct pm_nl_pernet { 30 /* protects pernet updates */ 31 spinlock_t lock; 32 struct list_head local_addr_list; 33 unsigned int addrs; 34 unsigned int stale_loss_cnt; 35 unsigned int add_addr_signal_max; 36 unsigned int add_addr_accept_max; 37 unsigned int local_addr_max; 38 unsigned int subflows_max; 39 unsigned int next_id; 40 DECLARE_BITMAP(id_bitmap, MPTCP_PM_MAX_ADDR_ID + 1); 41 }; 42 43 #define MPTCP_PM_ADDR_MAX 8 44 #define ADD_ADDR_RETRANS_MAX 3 45 46 static struct pm_nl_pernet *pm_nl_get_pernet(const struct net *net) 47 { 48 return net_generic(net, pm_nl_pernet_id); 49 } 50 51 static struct pm_nl_pernet * 52 pm_nl_get_pernet_from_msk(const struct mptcp_sock *msk) 53 { 54 return pm_nl_get_pernet(sock_net((struct sock *)msk)); 55 } 56 57 bool mptcp_addresses_equal(const struct mptcp_addr_info *a, 58 const struct mptcp_addr_info *b, bool use_port) 59 { 60 bool addr_equals = false; 61 62 if (a->family == b->family) { 63 if (a->family == AF_INET) 64 addr_equals = a->addr.s_addr == b->addr.s_addr; 65 #if IS_ENABLED(CONFIG_MPTCP_IPV6) 66 else 67 addr_equals = !ipv6_addr_cmp(&a->addr6, &b->addr6); 68 } else if (a->family == AF_INET) { 69 if (ipv6_addr_v4mapped(&b->addr6)) 70 addr_equals = a->addr.s_addr == b->addr6.s6_addr32[3]; 71 } else if (b->family == AF_INET) { 72 if (ipv6_addr_v4mapped(&a->addr6)) 73 addr_equals = a->addr6.s6_addr32[3] == b->addr.s_addr; 74 #endif 75 } 76 77 if (!addr_equals) 78 return false; 79 if (!use_port) 80 return true; 81 82 return a->port == b->port; 83 } 84 85 void mptcp_local_address(const struct sock_common *skc, struct mptcp_addr_info *addr) 86 { 87 addr->family = skc->skc_family; 88 addr->port = htons(skc->skc_num); 89 if (addr->family == AF_INET) 90 addr->addr.s_addr = skc->skc_rcv_saddr; 91 #if IS_ENABLED(CONFIG_MPTCP_IPV6) 92 else if (addr->family == AF_INET6) 93 addr->addr6 = skc->skc_v6_rcv_saddr; 94 #endif 95 } 96 97 static void remote_address(const struct sock_common *skc, 98 struct mptcp_addr_info *addr) 99 { 100 addr->family = skc->skc_family; 101 addr->port = skc->skc_dport; 102 if (addr->family == AF_INET) 103 addr->addr.s_addr = skc->skc_daddr; 104 #if IS_ENABLED(CONFIG_MPTCP_IPV6) 105 else if (addr->family == AF_INET6) 106 addr->addr6 = skc->skc_v6_daddr; 107 #endif 108 } 109 110 static bool lookup_subflow_by_saddr(const struct list_head *list, 111 const struct mptcp_addr_info *saddr) 112 { 113 struct mptcp_subflow_context *subflow; 114 struct mptcp_addr_info cur; 115 struct sock_common *skc; 116 117 list_for_each_entry(subflow, list, node) { 118 skc = (struct sock_common *)mptcp_subflow_tcp_sock(subflow); 119 120 mptcp_local_address(skc, &cur); 121 if (mptcp_addresses_equal(&cur, saddr, saddr->port)) 122 return true; 123 } 124 125 return false; 126 } 127 128 static bool lookup_subflow_by_daddr(const struct list_head *list, 129 const struct mptcp_addr_info *daddr) 130 { 131 struct mptcp_subflow_context *subflow; 132 struct mptcp_addr_info cur; 133 134 list_for_each_entry(subflow, list, node) { 135 struct sock *ssk = mptcp_subflow_tcp_sock(subflow); 136 137 if (!((1 << inet_sk_state_load(ssk)) & 138 (TCPF_ESTABLISHED | TCPF_SYN_SENT | TCPF_SYN_RECV))) 139 continue; 140 141 remote_address((struct sock_common *)ssk, &cur); 142 if (mptcp_addresses_equal(&cur, daddr, daddr->port)) 143 return true; 144 } 145 146 return false; 147 } 148 149 static bool 150 select_local_address(const struct pm_nl_pernet *pernet, 151 const struct mptcp_sock *msk, 152 struct mptcp_pm_local *new_local) 153 { 154 struct mptcp_pm_addr_entry *entry; 155 bool found = false; 156 157 msk_owned_by_me(msk); 158 159 rcu_read_lock(); 160 list_for_each_entry_rcu(entry, &pernet->local_addr_list, list) { 161 if (!(entry->flags & MPTCP_PM_ADDR_FLAG_SUBFLOW)) 162 continue; 163 164 if (!test_bit(entry->addr.id, msk->pm.id_avail_bitmap)) 165 continue; 166 167 new_local->addr = entry->addr; 168 new_local->flags = entry->flags; 169 new_local->ifindex = entry->ifindex; 170 found = true; 171 break; 172 } 173 rcu_read_unlock(); 174 175 return found; 176 } 177 178 static bool 179 select_signal_address(struct pm_nl_pernet *pernet, const struct mptcp_sock *msk, 180 struct mptcp_pm_local *new_local) 181 { 182 struct mptcp_pm_addr_entry *entry; 183 bool found = false; 184 185 rcu_read_lock(); 186 /* do not keep any additional per socket state, just signal 187 * the address list in order. 188 * Note: removal from the local address list during the msk life-cycle 189 * can lead to additional addresses not being announced. 190 */ 191 list_for_each_entry_rcu(entry, &pernet->local_addr_list, list) { 192 if (!test_bit(entry->addr.id, msk->pm.id_avail_bitmap)) 193 continue; 194 195 if (!(entry->flags & MPTCP_PM_ADDR_FLAG_SIGNAL)) 196 continue; 197 198 new_local->addr = entry->addr; 199 new_local->flags = entry->flags; 200 new_local->ifindex = entry->ifindex; 201 found = true; 202 break; 203 } 204 rcu_read_unlock(); 205 206 return found; 207 } 208 209 unsigned int mptcp_pm_get_add_addr_signal_max(const struct mptcp_sock *msk) 210 { 211 const struct pm_nl_pernet *pernet = pm_nl_get_pernet_from_msk(msk); 212 213 return READ_ONCE(pernet->add_addr_signal_max); 214 } 215 EXPORT_SYMBOL_GPL(mptcp_pm_get_add_addr_signal_max); 216 217 unsigned int mptcp_pm_get_add_addr_accept_max(const struct mptcp_sock *msk) 218 { 219 struct pm_nl_pernet *pernet = pm_nl_get_pernet_from_msk(msk); 220 221 return READ_ONCE(pernet->add_addr_accept_max); 222 } 223 EXPORT_SYMBOL_GPL(mptcp_pm_get_add_addr_accept_max); 224 225 unsigned int mptcp_pm_get_subflows_max(const struct mptcp_sock *msk) 226 { 227 struct pm_nl_pernet *pernet = pm_nl_get_pernet_from_msk(msk); 228 229 return READ_ONCE(pernet->subflows_max); 230 } 231 EXPORT_SYMBOL_GPL(mptcp_pm_get_subflows_max); 232 233 unsigned int mptcp_pm_get_local_addr_max(const struct mptcp_sock *msk) 234 { 235 struct pm_nl_pernet *pernet = pm_nl_get_pernet_from_msk(msk); 236 237 return READ_ONCE(pernet->local_addr_max); 238 } 239 EXPORT_SYMBOL_GPL(mptcp_pm_get_local_addr_max); 240 241 bool mptcp_pm_nl_check_work_pending(struct mptcp_sock *msk) 242 { 243 struct pm_nl_pernet *pernet = pm_nl_get_pernet_from_msk(msk); 244 245 if (msk->pm.subflows == mptcp_pm_get_subflows_max(msk) || 246 (find_next_and_bit(pernet->id_bitmap, msk->pm.id_avail_bitmap, 247 MPTCP_PM_MAX_ADDR_ID + 1, 0) == MPTCP_PM_MAX_ADDR_ID + 1)) { 248 WRITE_ONCE(msk->pm.work_pending, false); 249 return false; 250 } 251 return true; 252 } 253 254 struct mptcp_pm_add_entry * 255 mptcp_lookup_anno_list_by_saddr(const struct mptcp_sock *msk, 256 const struct mptcp_addr_info *addr) 257 { 258 struct mptcp_pm_add_entry *entry; 259 260 lockdep_assert_held(&msk->pm.lock); 261 262 list_for_each_entry(entry, &msk->pm.anno_list, list) { 263 if (mptcp_addresses_equal(&entry->addr, addr, true)) 264 return entry; 265 } 266 267 return NULL; 268 } 269 270 bool mptcp_pm_sport_in_anno_list(struct mptcp_sock *msk, const struct sock *sk) 271 { 272 struct mptcp_pm_add_entry *entry; 273 struct mptcp_addr_info saddr; 274 bool ret = false; 275 276 mptcp_local_address((struct sock_common *)sk, &saddr); 277 278 spin_lock_bh(&msk->pm.lock); 279 list_for_each_entry(entry, &msk->pm.anno_list, list) { 280 if (mptcp_addresses_equal(&entry->addr, &saddr, true)) { 281 ret = true; 282 goto out; 283 } 284 } 285 286 out: 287 spin_unlock_bh(&msk->pm.lock); 288 return ret; 289 } 290 291 static void mptcp_pm_add_timer(struct timer_list *timer) 292 { 293 struct mptcp_pm_add_entry *entry = from_timer(entry, timer, add_timer); 294 struct mptcp_sock *msk = entry->sock; 295 struct sock *sk = (struct sock *)msk; 296 297 pr_debug("msk=%p\n", msk); 298 299 if (!msk) 300 return; 301 302 if (inet_sk_state_load(sk) == TCP_CLOSE) 303 return; 304 305 if (!entry->addr.id) 306 return; 307 308 if (mptcp_pm_should_add_signal_addr(msk)) { 309 sk_reset_timer(sk, timer, jiffies + TCP_RTO_MAX / 8); 310 goto out; 311 } 312 313 spin_lock_bh(&msk->pm.lock); 314 315 if (!mptcp_pm_should_add_signal_addr(msk)) { 316 pr_debug("retransmit ADD_ADDR id=%d\n", entry->addr.id); 317 mptcp_pm_announce_addr(msk, &entry->addr, false); 318 mptcp_pm_add_addr_send_ack(msk); 319 entry->retrans_times++; 320 } 321 322 if (entry->retrans_times < ADD_ADDR_RETRANS_MAX) 323 sk_reset_timer(sk, timer, 324 jiffies + mptcp_get_add_addr_timeout(sock_net(sk))); 325 326 spin_unlock_bh(&msk->pm.lock); 327 328 if (entry->retrans_times == ADD_ADDR_RETRANS_MAX) 329 mptcp_pm_subflow_established(msk); 330 331 out: 332 __sock_put(sk); 333 } 334 335 struct mptcp_pm_add_entry * 336 mptcp_pm_del_add_timer(struct mptcp_sock *msk, 337 const struct mptcp_addr_info *addr, bool check_id) 338 { 339 struct mptcp_pm_add_entry *entry; 340 struct sock *sk = (struct sock *)msk; 341 struct timer_list *add_timer = NULL; 342 343 spin_lock_bh(&msk->pm.lock); 344 entry = mptcp_lookup_anno_list_by_saddr(msk, addr); 345 if (entry && (!check_id || entry->addr.id == addr->id)) { 346 entry->retrans_times = ADD_ADDR_RETRANS_MAX; 347 add_timer = &entry->add_timer; 348 } 349 if (!check_id && entry) 350 list_del(&entry->list); 351 spin_unlock_bh(&msk->pm.lock); 352 353 /* no lock, because sk_stop_timer_sync() is calling del_timer_sync() */ 354 if (add_timer) 355 sk_stop_timer_sync(sk, add_timer); 356 357 return entry; 358 } 359 360 bool mptcp_pm_alloc_anno_list(struct mptcp_sock *msk, 361 const struct mptcp_addr_info *addr) 362 { 363 struct mptcp_pm_add_entry *add_entry = NULL; 364 struct sock *sk = (struct sock *)msk; 365 struct net *net = sock_net(sk); 366 367 lockdep_assert_held(&msk->pm.lock); 368 369 add_entry = mptcp_lookup_anno_list_by_saddr(msk, addr); 370 371 if (add_entry) { 372 if (WARN_ON_ONCE(mptcp_pm_is_kernel(msk))) 373 return false; 374 375 sk_reset_timer(sk, &add_entry->add_timer, 376 jiffies + mptcp_get_add_addr_timeout(net)); 377 return true; 378 } 379 380 add_entry = kmalloc(sizeof(*add_entry), GFP_ATOMIC); 381 if (!add_entry) 382 return false; 383 384 list_add(&add_entry->list, &msk->pm.anno_list); 385 386 add_entry->addr = *addr; 387 add_entry->sock = msk; 388 add_entry->retrans_times = 0; 389 390 timer_setup(&add_entry->add_timer, mptcp_pm_add_timer, 0); 391 sk_reset_timer(sk, &add_entry->add_timer, 392 jiffies + mptcp_get_add_addr_timeout(net)); 393 394 return true; 395 } 396 397 void mptcp_pm_free_anno_list(struct mptcp_sock *msk) 398 { 399 struct mptcp_pm_add_entry *entry, *tmp; 400 struct sock *sk = (struct sock *)msk; 401 LIST_HEAD(free_list); 402 403 pr_debug("msk=%p\n", msk); 404 405 spin_lock_bh(&msk->pm.lock); 406 list_splice_init(&msk->pm.anno_list, &free_list); 407 spin_unlock_bh(&msk->pm.lock); 408 409 list_for_each_entry_safe(entry, tmp, &free_list, list) { 410 sk_stop_timer_sync(sk, &entry->add_timer); 411 kfree(entry); 412 } 413 } 414 415 /* Fill all the remote addresses into the array addrs[], 416 * and return the array size. 417 */ 418 static unsigned int fill_remote_addresses_vec(struct mptcp_sock *msk, 419 struct mptcp_addr_info *local, 420 bool fullmesh, 421 struct mptcp_addr_info *addrs) 422 { 423 bool deny_id0 = READ_ONCE(msk->pm.remote_deny_join_id0); 424 struct sock *sk = (struct sock *)msk, *ssk; 425 struct mptcp_subflow_context *subflow; 426 struct mptcp_addr_info remote = { 0 }; 427 unsigned int subflows_max; 428 int i = 0; 429 430 subflows_max = mptcp_pm_get_subflows_max(msk); 431 remote_address((struct sock_common *)sk, &remote); 432 433 /* Non-fullmesh endpoint, fill in the single entry 434 * corresponding to the primary MPC subflow remote address 435 */ 436 if (!fullmesh) { 437 if (deny_id0) 438 return 0; 439 440 if (!mptcp_pm_addr_families_match(sk, local, &remote)) 441 return 0; 442 443 msk->pm.subflows++; 444 addrs[i++] = remote; 445 } else { 446 DECLARE_BITMAP(unavail_id, MPTCP_PM_MAX_ADDR_ID + 1); 447 448 /* Forbid creation of new subflows matching existing 449 * ones, possibly already created by incoming ADD_ADDR 450 */ 451 bitmap_zero(unavail_id, MPTCP_PM_MAX_ADDR_ID + 1); 452 mptcp_for_each_subflow(msk, subflow) 453 if (READ_ONCE(subflow->local_id) == local->id) 454 __set_bit(subflow->remote_id, unavail_id); 455 456 mptcp_for_each_subflow(msk, subflow) { 457 ssk = mptcp_subflow_tcp_sock(subflow); 458 remote_address((struct sock_common *)ssk, &addrs[i]); 459 addrs[i].id = READ_ONCE(subflow->remote_id); 460 if (deny_id0 && !addrs[i].id) 461 continue; 462 463 if (test_bit(addrs[i].id, unavail_id)) 464 continue; 465 466 if (!mptcp_pm_addr_families_match(sk, local, &addrs[i])) 467 continue; 468 469 if (msk->pm.subflows < subflows_max) { 470 /* forbid creating multiple address towards 471 * this id 472 */ 473 __set_bit(addrs[i].id, unavail_id); 474 msk->pm.subflows++; 475 i++; 476 } 477 } 478 } 479 480 return i; 481 } 482 483 static void __mptcp_pm_send_ack(struct mptcp_sock *msk, struct mptcp_subflow_context *subflow, 484 bool prio, bool backup) 485 { 486 struct sock *ssk = mptcp_subflow_tcp_sock(subflow); 487 bool slow; 488 489 pr_debug("send ack for %s\n", 490 prio ? "mp_prio" : (mptcp_pm_should_add_signal(msk) ? "add_addr" : "rm_addr")); 491 492 slow = lock_sock_fast(ssk); 493 if (prio) { 494 subflow->send_mp_prio = 1; 495 subflow->request_bkup = backup; 496 } 497 498 __mptcp_subflow_send_ack(ssk); 499 unlock_sock_fast(ssk, slow); 500 } 501 502 static void mptcp_pm_send_ack(struct mptcp_sock *msk, struct mptcp_subflow_context *subflow, 503 bool prio, bool backup) 504 { 505 spin_unlock_bh(&msk->pm.lock); 506 __mptcp_pm_send_ack(msk, subflow, prio, backup); 507 spin_lock_bh(&msk->pm.lock); 508 } 509 510 static struct mptcp_pm_addr_entry * 511 __lookup_addr_by_id(struct pm_nl_pernet *pernet, unsigned int id) 512 { 513 struct mptcp_pm_addr_entry *entry; 514 515 list_for_each_entry(entry, &pernet->local_addr_list, list) { 516 if (entry->addr.id == id) 517 return entry; 518 } 519 return NULL; 520 } 521 522 static struct mptcp_pm_addr_entry * 523 __lookup_addr(struct pm_nl_pernet *pernet, const struct mptcp_addr_info *info) 524 { 525 struct mptcp_pm_addr_entry *entry; 526 527 list_for_each_entry_rcu(entry, &pernet->local_addr_list, list, 528 lockdep_is_held(&pernet->lock)) { 529 if (mptcp_addresses_equal(&entry->addr, info, entry->addr.port)) 530 return entry; 531 } 532 return NULL; 533 } 534 535 static void mptcp_pm_create_subflow_or_signal_addr(struct mptcp_sock *msk) 536 { 537 struct sock *sk = (struct sock *)msk; 538 unsigned int add_addr_signal_max; 539 bool signal_and_subflow = false; 540 unsigned int local_addr_max; 541 struct pm_nl_pernet *pernet; 542 struct mptcp_pm_local local; 543 unsigned int subflows_max; 544 545 pernet = pm_nl_get_pernet(sock_net(sk)); 546 547 add_addr_signal_max = mptcp_pm_get_add_addr_signal_max(msk); 548 local_addr_max = mptcp_pm_get_local_addr_max(msk); 549 subflows_max = mptcp_pm_get_subflows_max(msk); 550 551 /* do lazy endpoint usage accounting for the MPC subflows */ 552 if (unlikely(!(msk->pm.status & BIT(MPTCP_PM_MPC_ENDPOINT_ACCOUNTED))) && msk->first) { 553 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(msk->first); 554 struct mptcp_pm_addr_entry *entry; 555 struct mptcp_addr_info mpc_addr; 556 bool backup = false; 557 558 mptcp_local_address((struct sock_common *)msk->first, &mpc_addr); 559 rcu_read_lock(); 560 entry = __lookup_addr(pernet, &mpc_addr); 561 if (entry) { 562 __clear_bit(entry->addr.id, msk->pm.id_avail_bitmap); 563 msk->mpc_endpoint_id = entry->addr.id; 564 backup = !!(entry->flags & MPTCP_PM_ADDR_FLAG_BACKUP); 565 } 566 rcu_read_unlock(); 567 568 if (backup) 569 mptcp_pm_send_ack(msk, subflow, true, backup); 570 571 msk->pm.status |= BIT(MPTCP_PM_MPC_ENDPOINT_ACCOUNTED); 572 } 573 574 pr_debug("local %d:%d signal %d:%d subflows %d:%d\n", 575 msk->pm.local_addr_used, local_addr_max, 576 msk->pm.add_addr_signaled, add_addr_signal_max, 577 msk->pm.subflows, subflows_max); 578 579 /* check first for announce */ 580 if (msk->pm.add_addr_signaled < add_addr_signal_max) { 581 /* due to racing events on both ends we can reach here while 582 * previous add address is still running: if we invoke now 583 * mptcp_pm_announce_addr(), that will fail and the 584 * corresponding id will be marked as used. 585 * Instead let the PM machinery reschedule us when the 586 * current address announce will be completed. 587 */ 588 if (msk->pm.addr_signal & BIT(MPTCP_ADD_ADDR_SIGNAL)) 589 return; 590 591 if (!select_signal_address(pernet, msk, &local)) 592 goto subflow; 593 594 /* If the alloc fails, we are on memory pressure, not worth 595 * continuing, and trying to create subflows. 596 */ 597 if (!mptcp_pm_alloc_anno_list(msk, &local.addr)) 598 return; 599 600 __clear_bit(local.addr.id, msk->pm.id_avail_bitmap); 601 msk->pm.add_addr_signaled++; 602 603 /* Special case for ID0: set the correct ID */ 604 if (local.addr.id == msk->mpc_endpoint_id) 605 local.addr.id = 0; 606 607 mptcp_pm_announce_addr(msk, &local.addr, false); 608 mptcp_pm_nl_addr_send_ack(msk); 609 610 if (local.flags & MPTCP_PM_ADDR_FLAG_SUBFLOW) 611 signal_and_subflow = true; 612 } 613 614 subflow: 615 /* check if should create a new subflow */ 616 while (msk->pm.local_addr_used < local_addr_max && 617 msk->pm.subflows < subflows_max) { 618 struct mptcp_addr_info addrs[MPTCP_PM_ADDR_MAX]; 619 bool fullmesh; 620 int i, nr; 621 622 if (signal_and_subflow) 623 signal_and_subflow = false; 624 else if (!select_local_address(pernet, msk, &local)) 625 break; 626 627 fullmesh = !!(local.flags & MPTCP_PM_ADDR_FLAG_FULLMESH); 628 629 __clear_bit(local.addr.id, msk->pm.id_avail_bitmap); 630 631 /* Special case for ID0: set the correct ID */ 632 if (local.addr.id == msk->mpc_endpoint_id) 633 local.addr.id = 0; 634 else /* local_addr_used is not decr for ID 0 */ 635 msk->pm.local_addr_used++; 636 637 nr = fill_remote_addresses_vec(msk, &local.addr, fullmesh, addrs); 638 if (nr == 0) 639 continue; 640 641 spin_unlock_bh(&msk->pm.lock); 642 for (i = 0; i < nr; i++) 643 __mptcp_subflow_connect(sk, &local, &addrs[i]); 644 spin_lock_bh(&msk->pm.lock); 645 } 646 mptcp_pm_nl_check_work_pending(msk); 647 } 648 649 static void mptcp_pm_nl_fully_established(struct mptcp_sock *msk) 650 { 651 mptcp_pm_create_subflow_or_signal_addr(msk); 652 } 653 654 static void mptcp_pm_nl_subflow_established(struct mptcp_sock *msk) 655 { 656 mptcp_pm_create_subflow_or_signal_addr(msk); 657 } 658 659 /* Fill all the local addresses into the array addrs[], 660 * and return the array size. 661 */ 662 static unsigned int fill_local_addresses_vec(struct mptcp_sock *msk, 663 struct mptcp_addr_info *remote, 664 struct mptcp_pm_local *locals) 665 { 666 struct sock *sk = (struct sock *)msk; 667 struct mptcp_pm_addr_entry *entry; 668 struct mptcp_addr_info mpc_addr; 669 struct pm_nl_pernet *pernet; 670 unsigned int subflows_max; 671 int i = 0; 672 673 pernet = pm_nl_get_pernet_from_msk(msk); 674 subflows_max = mptcp_pm_get_subflows_max(msk); 675 676 mptcp_local_address((struct sock_common *)msk, &mpc_addr); 677 678 rcu_read_lock(); 679 list_for_each_entry_rcu(entry, &pernet->local_addr_list, list) { 680 if (!(entry->flags & MPTCP_PM_ADDR_FLAG_FULLMESH)) 681 continue; 682 683 if (!mptcp_pm_addr_families_match(sk, &entry->addr, remote)) 684 continue; 685 686 if (msk->pm.subflows < subflows_max) { 687 locals[i].addr = entry->addr; 688 locals[i].flags = entry->flags; 689 locals[i].ifindex = entry->ifindex; 690 691 /* Special case for ID0: set the correct ID */ 692 if (mptcp_addresses_equal(&locals[i].addr, &mpc_addr, locals[i].addr.port)) 693 locals[i].addr.id = 0; 694 695 msk->pm.subflows++; 696 i++; 697 } 698 } 699 rcu_read_unlock(); 700 701 /* If the array is empty, fill in the single 702 * 'IPADDRANY' local address 703 */ 704 if (!i) { 705 memset(&locals[i], 0, sizeof(locals[i])); 706 locals[i].addr.family = 707 #if IS_ENABLED(CONFIG_MPTCP_IPV6) 708 remote->family == AF_INET6 && 709 ipv6_addr_v4mapped(&remote->addr6) ? AF_INET : 710 #endif 711 remote->family; 712 713 if (!mptcp_pm_addr_families_match(sk, &locals[i].addr, remote)) 714 return 0; 715 716 msk->pm.subflows++; 717 i++; 718 } 719 720 return i; 721 } 722 723 static void mptcp_pm_nl_add_addr_received(struct mptcp_sock *msk) 724 { 725 struct mptcp_pm_local locals[MPTCP_PM_ADDR_MAX]; 726 struct sock *sk = (struct sock *)msk; 727 unsigned int add_addr_accept_max; 728 struct mptcp_addr_info remote; 729 unsigned int subflows_max; 730 bool sf_created = false; 731 int i, nr; 732 733 add_addr_accept_max = mptcp_pm_get_add_addr_accept_max(msk); 734 subflows_max = mptcp_pm_get_subflows_max(msk); 735 736 pr_debug("accepted %d:%d remote family %d\n", 737 msk->pm.add_addr_accepted, add_addr_accept_max, 738 msk->pm.remote.family); 739 740 remote = msk->pm.remote; 741 mptcp_pm_announce_addr(msk, &remote, true); 742 mptcp_pm_nl_addr_send_ack(msk); 743 744 if (lookup_subflow_by_daddr(&msk->conn_list, &remote)) 745 return; 746 747 /* pick id 0 port, if none is provided the remote address */ 748 if (!remote.port) 749 remote.port = sk->sk_dport; 750 751 /* connect to the specified remote address, using whatever 752 * local address the routing configuration will pick. 753 */ 754 nr = fill_local_addresses_vec(msk, &remote, locals); 755 if (nr == 0) 756 return; 757 758 spin_unlock_bh(&msk->pm.lock); 759 for (i = 0; i < nr; i++) 760 if (__mptcp_subflow_connect(sk, &locals[i], &remote) == 0) 761 sf_created = true; 762 spin_lock_bh(&msk->pm.lock); 763 764 if (sf_created) { 765 /* add_addr_accepted is not decr for ID 0 */ 766 if (remote.id) 767 msk->pm.add_addr_accepted++; 768 if (msk->pm.add_addr_accepted >= add_addr_accept_max || 769 msk->pm.subflows >= subflows_max) 770 WRITE_ONCE(msk->pm.accept_addr, false); 771 } 772 } 773 774 bool mptcp_pm_nl_is_init_remote_addr(struct mptcp_sock *msk, 775 const struct mptcp_addr_info *remote) 776 { 777 struct mptcp_addr_info mpc_remote; 778 779 remote_address((struct sock_common *)msk, &mpc_remote); 780 return mptcp_addresses_equal(&mpc_remote, remote, remote->port); 781 } 782 783 void mptcp_pm_nl_addr_send_ack(struct mptcp_sock *msk) 784 { 785 struct mptcp_subflow_context *subflow, *alt = NULL; 786 787 msk_owned_by_me(msk); 788 lockdep_assert_held(&msk->pm.lock); 789 790 if (!mptcp_pm_should_add_signal(msk) && 791 !mptcp_pm_should_rm_signal(msk)) 792 return; 793 794 mptcp_for_each_subflow(msk, subflow) { 795 if (__mptcp_subflow_active(subflow)) { 796 if (!subflow->stale) { 797 mptcp_pm_send_ack(msk, subflow, false, false); 798 return; 799 } 800 801 if (!alt) 802 alt = subflow; 803 } 804 } 805 806 if (alt) 807 mptcp_pm_send_ack(msk, alt, false, false); 808 } 809 810 int mptcp_pm_nl_mp_prio_send_ack(struct mptcp_sock *msk, 811 struct mptcp_addr_info *addr, 812 struct mptcp_addr_info *rem, 813 u8 bkup) 814 { 815 struct mptcp_subflow_context *subflow; 816 817 pr_debug("bkup=%d\n", bkup); 818 819 mptcp_for_each_subflow(msk, subflow) { 820 struct sock *ssk = mptcp_subflow_tcp_sock(subflow); 821 struct mptcp_addr_info local, remote; 822 823 mptcp_local_address((struct sock_common *)ssk, &local); 824 if (!mptcp_addresses_equal(&local, addr, addr->port)) 825 continue; 826 827 if (rem && rem->family != AF_UNSPEC) { 828 remote_address((struct sock_common *)ssk, &remote); 829 if (!mptcp_addresses_equal(&remote, rem, rem->port)) 830 continue; 831 } 832 833 __mptcp_pm_send_ack(msk, subflow, true, bkup); 834 return 0; 835 } 836 837 return -EINVAL; 838 } 839 840 static void mptcp_pm_nl_rm_addr_or_subflow(struct mptcp_sock *msk, 841 const struct mptcp_rm_list *rm_list, 842 enum linux_mptcp_mib_field rm_type) 843 { 844 struct mptcp_subflow_context *subflow, *tmp; 845 struct sock *sk = (struct sock *)msk; 846 u8 i; 847 848 pr_debug("%s rm_list_nr %d\n", 849 rm_type == MPTCP_MIB_RMADDR ? "address" : "subflow", rm_list->nr); 850 851 msk_owned_by_me(msk); 852 853 if (sk->sk_state == TCP_LISTEN) 854 return; 855 856 if (!rm_list->nr) 857 return; 858 859 if (list_empty(&msk->conn_list)) 860 return; 861 862 for (i = 0; i < rm_list->nr; i++) { 863 u8 rm_id = rm_list->ids[i]; 864 bool removed = false; 865 866 mptcp_for_each_subflow_safe(msk, subflow, tmp) { 867 struct sock *ssk = mptcp_subflow_tcp_sock(subflow); 868 u8 remote_id = READ_ONCE(subflow->remote_id); 869 int how = RCV_SHUTDOWN | SEND_SHUTDOWN; 870 u8 id = subflow_get_local_id(subflow); 871 872 if ((1 << inet_sk_state_load(ssk)) & 873 (TCPF_FIN_WAIT1 | TCPF_FIN_WAIT2 | TCPF_CLOSING | TCPF_CLOSE)) 874 continue; 875 if (rm_type == MPTCP_MIB_RMADDR && remote_id != rm_id) 876 continue; 877 if (rm_type == MPTCP_MIB_RMSUBFLOW && id != rm_id) 878 continue; 879 880 pr_debug(" -> %s rm_list_ids[%d]=%u local_id=%u remote_id=%u mpc_id=%u\n", 881 rm_type == MPTCP_MIB_RMADDR ? "address" : "subflow", 882 i, rm_id, id, remote_id, msk->mpc_endpoint_id); 883 spin_unlock_bh(&msk->pm.lock); 884 mptcp_subflow_shutdown(sk, ssk, how); 885 removed |= subflow->request_join; 886 887 /* the following takes care of updating the subflows counter */ 888 mptcp_close_ssk(sk, ssk, subflow); 889 spin_lock_bh(&msk->pm.lock); 890 891 if (rm_type == MPTCP_MIB_RMSUBFLOW) 892 __MPTCP_INC_STATS(sock_net(sk), rm_type); 893 } 894 895 if (rm_type == MPTCP_MIB_RMADDR) 896 __MPTCP_INC_STATS(sock_net(sk), rm_type); 897 898 if (!removed) 899 continue; 900 901 if (!mptcp_pm_is_kernel(msk)) 902 continue; 903 904 if (rm_type == MPTCP_MIB_RMADDR && rm_id && 905 !WARN_ON_ONCE(msk->pm.add_addr_accepted == 0)) { 906 /* Note: if the subflow has been closed before, this 907 * add_addr_accepted counter will not be decremented. 908 */ 909 if (--msk->pm.add_addr_accepted < mptcp_pm_get_add_addr_accept_max(msk)) 910 WRITE_ONCE(msk->pm.accept_addr, true); 911 } 912 } 913 } 914 915 static void mptcp_pm_nl_rm_addr_received(struct mptcp_sock *msk) 916 { 917 mptcp_pm_nl_rm_addr_or_subflow(msk, &msk->pm.rm_list_rx, MPTCP_MIB_RMADDR); 918 } 919 920 static void mptcp_pm_nl_rm_subflow_received(struct mptcp_sock *msk, 921 const struct mptcp_rm_list *rm_list) 922 { 923 mptcp_pm_nl_rm_addr_or_subflow(msk, rm_list, MPTCP_MIB_RMSUBFLOW); 924 } 925 926 void mptcp_pm_nl_work(struct mptcp_sock *msk) 927 { 928 struct mptcp_pm_data *pm = &msk->pm; 929 930 msk_owned_by_me(msk); 931 932 if (!(pm->status & MPTCP_PM_WORK_MASK)) 933 return; 934 935 spin_lock_bh(&msk->pm.lock); 936 937 pr_debug("msk=%p status=%x\n", msk, pm->status); 938 if (pm->status & BIT(MPTCP_PM_ADD_ADDR_RECEIVED)) { 939 pm->status &= ~BIT(MPTCP_PM_ADD_ADDR_RECEIVED); 940 mptcp_pm_nl_add_addr_received(msk); 941 } 942 if (pm->status & BIT(MPTCP_PM_ADD_ADDR_SEND_ACK)) { 943 pm->status &= ~BIT(MPTCP_PM_ADD_ADDR_SEND_ACK); 944 mptcp_pm_nl_addr_send_ack(msk); 945 } 946 if (pm->status & BIT(MPTCP_PM_RM_ADDR_RECEIVED)) { 947 pm->status &= ~BIT(MPTCP_PM_RM_ADDR_RECEIVED); 948 mptcp_pm_nl_rm_addr_received(msk); 949 } 950 if (pm->status & BIT(MPTCP_PM_ESTABLISHED)) { 951 pm->status &= ~BIT(MPTCP_PM_ESTABLISHED); 952 mptcp_pm_nl_fully_established(msk); 953 } 954 if (pm->status & BIT(MPTCP_PM_SUBFLOW_ESTABLISHED)) { 955 pm->status &= ~BIT(MPTCP_PM_SUBFLOW_ESTABLISHED); 956 mptcp_pm_nl_subflow_established(msk); 957 } 958 959 spin_unlock_bh(&msk->pm.lock); 960 } 961 962 static bool address_use_port(struct mptcp_pm_addr_entry *entry) 963 { 964 return (entry->flags & 965 (MPTCP_PM_ADDR_FLAG_SIGNAL | MPTCP_PM_ADDR_FLAG_SUBFLOW)) == 966 MPTCP_PM_ADDR_FLAG_SIGNAL; 967 } 968 969 /* caller must ensure the RCU grace period is already elapsed */ 970 static void __mptcp_pm_release_addr_entry(struct mptcp_pm_addr_entry *entry) 971 { 972 if (entry->lsk) 973 sock_release(entry->lsk); 974 kfree(entry); 975 } 976 977 static int mptcp_pm_nl_append_new_local_addr(struct pm_nl_pernet *pernet, 978 struct mptcp_pm_addr_entry *entry, 979 bool needs_id) 980 { 981 struct mptcp_pm_addr_entry *cur, *del_entry = NULL; 982 unsigned int addr_max; 983 int ret = -EINVAL; 984 985 spin_lock_bh(&pernet->lock); 986 /* to keep the code simple, don't do IDR-like allocation for address ID, 987 * just bail when we exceed limits 988 */ 989 if (pernet->next_id == MPTCP_PM_MAX_ADDR_ID) 990 pernet->next_id = 1; 991 if (pernet->addrs >= MPTCP_PM_ADDR_MAX) { 992 ret = -ERANGE; 993 goto out; 994 } 995 if (test_bit(entry->addr.id, pernet->id_bitmap)) { 996 ret = -EBUSY; 997 goto out; 998 } 999 1000 /* do not insert duplicate address, differentiate on port only 1001 * singled addresses 1002 */ 1003 if (!address_use_port(entry)) 1004 entry->addr.port = 0; 1005 list_for_each_entry(cur, &pernet->local_addr_list, list) { 1006 if (mptcp_addresses_equal(&cur->addr, &entry->addr, 1007 cur->addr.port || entry->addr.port)) { 1008 /* allow replacing the exiting endpoint only if such 1009 * endpoint is an implicit one and the user-space 1010 * did not provide an endpoint id 1011 */ 1012 if (!(cur->flags & MPTCP_PM_ADDR_FLAG_IMPLICIT)) { 1013 ret = -EEXIST; 1014 goto out; 1015 } 1016 if (entry->addr.id) 1017 goto out; 1018 1019 pernet->addrs--; 1020 entry->addr.id = cur->addr.id; 1021 list_del_rcu(&cur->list); 1022 del_entry = cur; 1023 break; 1024 } 1025 } 1026 1027 if (!entry->addr.id && needs_id) { 1028 find_next: 1029 entry->addr.id = find_next_zero_bit(pernet->id_bitmap, 1030 MPTCP_PM_MAX_ADDR_ID + 1, 1031 pernet->next_id); 1032 if (!entry->addr.id && pernet->next_id != 1) { 1033 pernet->next_id = 1; 1034 goto find_next; 1035 } 1036 } 1037 1038 if (!entry->addr.id && needs_id) 1039 goto out; 1040 1041 __set_bit(entry->addr.id, pernet->id_bitmap); 1042 if (entry->addr.id > pernet->next_id) 1043 pernet->next_id = entry->addr.id; 1044 1045 if (entry->flags & MPTCP_PM_ADDR_FLAG_SIGNAL) { 1046 addr_max = pernet->add_addr_signal_max; 1047 WRITE_ONCE(pernet->add_addr_signal_max, addr_max + 1); 1048 } 1049 if (entry->flags & MPTCP_PM_ADDR_FLAG_SUBFLOW) { 1050 addr_max = pernet->local_addr_max; 1051 WRITE_ONCE(pernet->local_addr_max, addr_max + 1); 1052 } 1053 1054 pernet->addrs++; 1055 if (!entry->addr.port) 1056 list_add_tail_rcu(&entry->list, &pernet->local_addr_list); 1057 else 1058 list_add_rcu(&entry->list, &pernet->local_addr_list); 1059 ret = entry->addr.id; 1060 1061 out: 1062 spin_unlock_bh(&pernet->lock); 1063 1064 /* just replaced an existing entry, free it */ 1065 if (del_entry) { 1066 synchronize_rcu(); 1067 __mptcp_pm_release_addr_entry(del_entry); 1068 } 1069 return ret; 1070 } 1071 1072 static struct lock_class_key mptcp_slock_keys[2]; 1073 static struct lock_class_key mptcp_keys[2]; 1074 1075 static int mptcp_pm_nl_create_listen_socket(struct sock *sk, 1076 struct mptcp_pm_addr_entry *entry) 1077 { 1078 bool is_ipv6 = sk->sk_family == AF_INET6; 1079 int addrlen = sizeof(struct sockaddr_in); 1080 struct sockaddr_storage addr; 1081 struct sock *newsk, *ssk; 1082 int backlog = 1024; 1083 int err; 1084 1085 err = sock_create_kern(sock_net(sk), entry->addr.family, 1086 SOCK_STREAM, IPPROTO_MPTCP, &entry->lsk); 1087 if (err) 1088 return err; 1089 1090 newsk = entry->lsk->sk; 1091 if (!newsk) 1092 return -EINVAL; 1093 1094 /* The subflow socket lock is acquired in a nested to the msk one 1095 * in several places, even by the TCP stack, and this msk is a kernel 1096 * socket: lockdep complains. Instead of propagating the _nested 1097 * modifiers in several places, re-init the lock class for the msk 1098 * socket to an mptcp specific one. 1099 */ 1100 sock_lock_init_class_and_name(newsk, 1101 is_ipv6 ? "mlock-AF_INET6" : "mlock-AF_INET", 1102 &mptcp_slock_keys[is_ipv6], 1103 is_ipv6 ? "msk_lock-AF_INET6" : "msk_lock-AF_INET", 1104 &mptcp_keys[is_ipv6]); 1105 1106 lock_sock(newsk); 1107 ssk = __mptcp_nmpc_sk(mptcp_sk(newsk)); 1108 release_sock(newsk); 1109 if (IS_ERR(ssk)) 1110 return PTR_ERR(ssk); 1111 1112 mptcp_info2sockaddr(&entry->addr, &addr, entry->addr.family); 1113 #if IS_ENABLED(CONFIG_MPTCP_IPV6) 1114 if (entry->addr.family == AF_INET6) 1115 addrlen = sizeof(struct sockaddr_in6); 1116 #endif 1117 if (ssk->sk_family == AF_INET) 1118 err = inet_bind_sk(ssk, (struct sockaddr *)&addr, addrlen); 1119 #if IS_ENABLED(CONFIG_MPTCP_IPV6) 1120 else if (ssk->sk_family == AF_INET6) 1121 err = inet6_bind_sk(ssk, (struct sockaddr *)&addr, addrlen); 1122 #endif 1123 if (err) 1124 return err; 1125 1126 /* We don't use mptcp_set_state() here because it needs to be called 1127 * under the msk socket lock. For the moment, that will not bring 1128 * anything more than only calling inet_sk_state_store(), because the 1129 * old status is known (TCP_CLOSE). 1130 */ 1131 inet_sk_state_store(newsk, TCP_LISTEN); 1132 lock_sock(ssk); 1133 WRITE_ONCE(mptcp_subflow_ctx(ssk)->pm_listener, true); 1134 err = __inet_listen_sk(ssk, backlog); 1135 if (!err) 1136 mptcp_event_pm_listener(ssk, MPTCP_EVENT_LISTENER_CREATED); 1137 release_sock(ssk); 1138 return err; 1139 } 1140 1141 int mptcp_pm_nl_get_local_id(struct mptcp_sock *msk, struct mptcp_addr_info *skc) 1142 { 1143 struct mptcp_pm_addr_entry *entry; 1144 struct pm_nl_pernet *pernet; 1145 int ret = -1; 1146 1147 pernet = pm_nl_get_pernet_from_msk(msk); 1148 1149 rcu_read_lock(); 1150 list_for_each_entry_rcu(entry, &pernet->local_addr_list, list) { 1151 if (mptcp_addresses_equal(&entry->addr, skc, entry->addr.port)) { 1152 ret = entry->addr.id; 1153 break; 1154 } 1155 } 1156 rcu_read_unlock(); 1157 if (ret >= 0) 1158 return ret; 1159 1160 /* address not found, add to local list */ 1161 entry = kmalloc(sizeof(*entry), GFP_ATOMIC); 1162 if (!entry) 1163 return -ENOMEM; 1164 1165 entry->addr = *skc; 1166 entry->addr.id = 0; 1167 entry->addr.port = 0; 1168 entry->ifindex = 0; 1169 entry->flags = MPTCP_PM_ADDR_FLAG_IMPLICIT; 1170 entry->lsk = NULL; 1171 ret = mptcp_pm_nl_append_new_local_addr(pernet, entry, true); 1172 if (ret < 0) 1173 kfree(entry); 1174 1175 return ret; 1176 } 1177 1178 bool mptcp_pm_nl_is_backup(struct mptcp_sock *msk, struct mptcp_addr_info *skc) 1179 { 1180 struct pm_nl_pernet *pernet = pm_nl_get_pernet_from_msk(msk); 1181 struct mptcp_pm_addr_entry *entry; 1182 bool backup = false; 1183 1184 rcu_read_lock(); 1185 list_for_each_entry_rcu(entry, &pernet->local_addr_list, list) { 1186 if (mptcp_addresses_equal(&entry->addr, skc, entry->addr.port)) { 1187 backup = !!(entry->flags & MPTCP_PM_ADDR_FLAG_BACKUP); 1188 break; 1189 } 1190 } 1191 rcu_read_unlock(); 1192 1193 return backup; 1194 } 1195 1196 #define MPTCP_PM_CMD_GRP_OFFSET 0 1197 #define MPTCP_PM_EV_GRP_OFFSET 1 1198 1199 static const struct genl_multicast_group mptcp_pm_mcgrps[] = { 1200 [MPTCP_PM_CMD_GRP_OFFSET] = { .name = MPTCP_PM_CMD_GRP_NAME, }, 1201 [MPTCP_PM_EV_GRP_OFFSET] = { .name = MPTCP_PM_EV_GRP_NAME, 1202 .flags = GENL_MCAST_CAP_NET_ADMIN, 1203 }, 1204 }; 1205 1206 void mptcp_pm_nl_subflow_chk_stale(const struct mptcp_sock *msk, struct sock *ssk) 1207 { 1208 struct mptcp_subflow_context *iter, *subflow = mptcp_subflow_ctx(ssk); 1209 struct sock *sk = (struct sock *)msk; 1210 unsigned int active_max_loss_cnt; 1211 struct net *net = sock_net(sk); 1212 unsigned int stale_loss_cnt; 1213 bool slow; 1214 1215 stale_loss_cnt = mptcp_stale_loss_cnt(net); 1216 if (subflow->stale || !stale_loss_cnt || subflow->stale_count <= stale_loss_cnt) 1217 return; 1218 1219 /* look for another available subflow not in loss state */ 1220 active_max_loss_cnt = max_t(int, stale_loss_cnt - 1, 1); 1221 mptcp_for_each_subflow(msk, iter) { 1222 if (iter != subflow && mptcp_subflow_active(iter) && 1223 iter->stale_count < active_max_loss_cnt) { 1224 /* we have some alternatives, try to mark this subflow as idle ...*/ 1225 slow = lock_sock_fast(ssk); 1226 if (!tcp_rtx_and_write_queues_empty(ssk)) { 1227 subflow->stale = 1; 1228 __mptcp_retransmit_pending_data(sk); 1229 MPTCP_INC_STATS(net, MPTCP_MIB_SUBFLOWSTALE); 1230 } 1231 unlock_sock_fast(ssk, slow); 1232 1233 /* always try to push the pending data regardless of re-injections: 1234 * we can possibly use backup subflows now, and subflow selection 1235 * is cheap under the msk socket lock 1236 */ 1237 __mptcp_push_pending(sk, 0); 1238 return; 1239 } 1240 } 1241 } 1242 1243 static int mptcp_pm_family_to_addr(int family) 1244 { 1245 #if IS_ENABLED(CONFIG_MPTCP_IPV6) 1246 if (family == AF_INET6) 1247 return MPTCP_PM_ADDR_ATTR_ADDR6; 1248 #endif 1249 return MPTCP_PM_ADDR_ATTR_ADDR4; 1250 } 1251 1252 static int mptcp_pm_parse_pm_addr_attr(struct nlattr *tb[], 1253 const struct nlattr *attr, 1254 struct genl_info *info, 1255 struct mptcp_addr_info *addr, 1256 bool require_family) 1257 { 1258 int err, addr_addr; 1259 1260 if (!attr) { 1261 GENL_SET_ERR_MSG(info, "missing address info"); 1262 return -EINVAL; 1263 } 1264 1265 /* no validation needed - was already done via nested policy */ 1266 err = nla_parse_nested_deprecated(tb, MPTCP_PM_ADDR_ATTR_MAX, attr, 1267 mptcp_pm_address_nl_policy, info->extack); 1268 if (err) 1269 return err; 1270 1271 if (tb[MPTCP_PM_ADDR_ATTR_ID]) 1272 addr->id = nla_get_u8(tb[MPTCP_PM_ADDR_ATTR_ID]); 1273 1274 if (!tb[MPTCP_PM_ADDR_ATTR_FAMILY]) { 1275 if (!require_family) 1276 return 0; 1277 1278 NL_SET_ERR_MSG_ATTR(info->extack, attr, 1279 "missing family"); 1280 return -EINVAL; 1281 } 1282 1283 addr->family = nla_get_u16(tb[MPTCP_PM_ADDR_ATTR_FAMILY]); 1284 if (addr->family != AF_INET 1285 #if IS_ENABLED(CONFIG_MPTCP_IPV6) 1286 && addr->family != AF_INET6 1287 #endif 1288 ) { 1289 NL_SET_ERR_MSG_ATTR(info->extack, attr, 1290 "unknown address family"); 1291 return -EINVAL; 1292 } 1293 addr_addr = mptcp_pm_family_to_addr(addr->family); 1294 if (!tb[addr_addr]) { 1295 NL_SET_ERR_MSG_ATTR(info->extack, attr, 1296 "missing address data"); 1297 return -EINVAL; 1298 } 1299 1300 #if IS_ENABLED(CONFIG_MPTCP_IPV6) 1301 if (addr->family == AF_INET6) 1302 addr->addr6 = nla_get_in6_addr(tb[addr_addr]); 1303 else 1304 #endif 1305 addr->addr.s_addr = nla_get_in_addr(tb[addr_addr]); 1306 1307 if (tb[MPTCP_PM_ADDR_ATTR_PORT]) 1308 addr->port = htons(nla_get_u16(tb[MPTCP_PM_ADDR_ATTR_PORT])); 1309 1310 return 0; 1311 } 1312 1313 int mptcp_pm_parse_addr(struct nlattr *attr, struct genl_info *info, 1314 struct mptcp_addr_info *addr) 1315 { 1316 struct nlattr *tb[MPTCP_PM_ADDR_ATTR_MAX + 1]; 1317 1318 memset(addr, 0, sizeof(*addr)); 1319 1320 return mptcp_pm_parse_pm_addr_attr(tb, attr, info, addr, true); 1321 } 1322 1323 int mptcp_pm_parse_entry(struct nlattr *attr, struct genl_info *info, 1324 bool require_family, 1325 struct mptcp_pm_addr_entry *entry) 1326 { 1327 struct nlattr *tb[MPTCP_PM_ADDR_ATTR_MAX + 1]; 1328 int err; 1329 1330 memset(entry, 0, sizeof(*entry)); 1331 1332 err = mptcp_pm_parse_pm_addr_attr(tb, attr, info, &entry->addr, require_family); 1333 if (err) 1334 return err; 1335 1336 if (tb[MPTCP_PM_ADDR_ATTR_IF_IDX]) { 1337 u32 val = nla_get_s32(tb[MPTCP_PM_ADDR_ATTR_IF_IDX]); 1338 1339 entry->ifindex = val; 1340 } 1341 1342 if (tb[MPTCP_PM_ADDR_ATTR_FLAGS]) 1343 entry->flags = nla_get_u32(tb[MPTCP_PM_ADDR_ATTR_FLAGS]); 1344 1345 if (tb[MPTCP_PM_ADDR_ATTR_PORT]) 1346 entry->addr.port = htons(nla_get_u16(tb[MPTCP_PM_ADDR_ATTR_PORT])); 1347 1348 return 0; 1349 } 1350 1351 static struct pm_nl_pernet *genl_info_pm_nl(struct genl_info *info) 1352 { 1353 return pm_nl_get_pernet(genl_info_net(info)); 1354 } 1355 1356 static int mptcp_nl_add_subflow_or_signal_addr(struct net *net, 1357 struct mptcp_addr_info *addr) 1358 { 1359 struct mptcp_sock *msk; 1360 long s_slot = 0, s_num = 0; 1361 1362 while ((msk = mptcp_token_iter_next(net, &s_slot, &s_num)) != NULL) { 1363 struct sock *sk = (struct sock *)msk; 1364 struct mptcp_addr_info mpc_addr; 1365 1366 if (!READ_ONCE(msk->fully_established) || 1367 mptcp_pm_is_userspace(msk)) 1368 goto next; 1369 1370 /* if the endp linked to the init sf is re-added with a != ID */ 1371 mptcp_local_address((struct sock_common *)msk, &mpc_addr); 1372 1373 lock_sock(sk); 1374 spin_lock_bh(&msk->pm.lock); 1375 if (mptcp_addresses_equal(addr, &mpc_addr, addr->port)) 1376 msk->mpc_endpoint_id = addr->id; 1377 mptcp_pm_create_subflow_or_signal_addr(msk); 1378 spin_unlock_bh(&msk->pm.lock); 1379 release_sock(sk); 1380 1381 next: 1382 sock_put(sk); 1383 cond_resched(); 1384 } 1385 1386 return 0; 1387 } 1388 1389 static bool mptcp_pm_has_addr_attr_id(const struct nlattr *attr, 1390 struct genl_info *info) 1391 { 1392 struct nlattr *tb[MPTCP_PM_ADDR_ATTR_MAX + 1]; 1393 1394 if (!nla_parse_nested_deprecated(tb, MPTCP_PM_ADDR_ATTR_MAX, attr, 1395 mptcp_pm_address_nl_policy, info->extack) && 1396 tb[MPTCP_PM_ADDR_ATTR_ID]) 1397 return true; 1398 return false; 1399 } 1400 1401 int mptcp_pm_nl_add_addr_doit(struct sk_buff *skb, struct genl_info *info) 1402 { 1403 struct nlattr *attr = info->attrs[MPTCP_PM_ENDPOINT_ADDR]; 1404 struct pm_nl_pernet *pernet = genl_info_pm_nl(info); 1405 struct mptcp_pm_addr_entry addr, *entry; 1406 int ret; 1407 1408 ret = mptcp_pm_parse_entry(attr, info, true, &addr); 1409 if (ret < 0) 1410 return ret; 1411 1412 if (addr.addr.port && !address_use_port(&addr)) { 1413 GENL_SET_ERR_MSG(info, "flags must have signal and not subflow when using port"); 1414 return -EINVAL; 1415 } 1416 1417 if (addr.flags & MPTCP_PM_ADDR_FLAG_SIGNAL && 1418 addr.flags & MPTCP_PM_ADDR_FLAG_FULLMESH) { 1419 GENL_SET_ERR_MSG(info, "flags mustn't have both signal and fullmesh"); 1420 return -EINVAL; 1421 } 1422 1423 if (addr.flags & MPTCP_PM_ADDR_FLAG_IMPLICIT) { 1424 GENL_SET_ERR_MSG(info, "can't create IMPLICIT endpoint"); 1425 return -EINVAL; 1426 } 1427 1428 entry = kzalloc(sizeof(*entry), GFP_KERNEL_ACCOUNT); 1429 if (!entry) { 1430 GENL_SET_ERR_MSG(info, "can't allocate addr"); 1431 return -ENOMEM; 1432 } 1433 1434 *entry = addr; 1435 if (entry->addr.port) { 1436 ret = mptcp_pm_nl_create_listen_socket(skb->sk, entry); 1437 if (ret) { 1438 GENL_SET_ERR_MSG_FMT(info, "create listen socket error: %d", ret); 1439 goto out_free; 1440 } 1441 } 1442 ret = mptcp_pm_nl_append_new_local_addr(pernet, entry, 1443 !mptcp_pm_has_addr_attr_id(attr, info)); 1444 if (ret < 0) { 1445 GENL_SET_ERR_MSG_FMT(info, "too many addresses or duplicate one: %d", ret); 1446 goto out_free; 1447 } 1448 1449 mptcp_nl_add_subflow_or_signal_addr(sock_net(skb->sk), &entry->addr); 1450 return 0; 1451 1452 out_free: 1453 __mptcp_pm_release_addr_entry(entry); 1454 return ret; 1455 } 1456 1457 static bool remove_anno_list_by_saddr(struct mptcp_sock *msk, 1458 const struct mptcp_addr_info *addr) 1459 { 1460 struct mptcp_pm_add_entry *entry; 1461 1462 entry = mptcp_pm_del_add_timer(msk, addr, false); 1463 if (entry) { 1464 kfree(entry); 1465 return true; 1466 } 1467 1468 return false; 1469 } 1470 1471 static u8 mptcp_endp_get_local_id(struct mptcp_sock *msk, 1472 const struct mptcp_addr_info *addr) 1473 { 1474 return msk->mpc_endpoint_id == addr->id ? 0 : addr->id; 1475 } 1476 1477 static bool mptcp_pm_remove_anno_addr(struct mptcp_sock *msk, 1478 const struct mptcp_addr_info *addr, 1479 bool force) 1480 { 1481 struct mptcp_rm_list list = { .nr = 0 }; 1482 bool ret; 1483 1484 list.ids[list.nr++] = mptcp_endp_get_local_id(msk, addr); 1485 1486 ret = remove_anno_list_by_saddr(msk, addr); 1487 if (ret || force) { 1488 spin_lock_bh(&msk->pm.lock); 1489 if (ret) { 1490 __set_bit(addr->id, msk->pm.id_avail_bitmap); 1491 msk->pm.add_addr_signaled--; 1492 } 1493 mptcp_pm_remove_addr(msk, &list); 1494 spin_unlock_bh(&msk->pm.lock); 1495 } 1496 return ret; 1497 } 1498 1499 static void __mark_subflow_endp_available(struct mptcp_sock *msk, u8 id) 1500 { 1501 /* If it was marked as used, and not ID 0, decrement local_addr_used */ 1502 if (!__test_and_set_bit(id ? : msk->mpc_endpoint_id, msk->pm.id_avail_bitmap) && 1503 id && !WARN_ON_ONCE(msk->pm.local_addr_used == 0)) 1504 msk->pm.local_addr_used--; 1505 } 1506 1507 static int mptcp_nl_remove_subflow_and_signal_addr(struct net *net, 1508 const struct mptcp_pm_addr_entry *entry) 1509 { 1510 const struct mptcp_addr_info *addr = &entry->addr; 1511 struct mptcp_rm_list list = { .nr = 1 }; 1512 long s_slot = 0, s_num = 0; 1513 struct mptcp_sock *msk; 1514 1515 pr_debug("remove_id=%d\n", addr->id); 1516 1517 while ((msk = mptcp_token_iter_next(net, &s_slot, &s_num)) != NULL) { 1518 struct sock *sk = (struct sock *)msk; 1519 bool remove_subflow; 1520 1521 if (mptcp_pm_is_userspace(msk)) 1522 goto next; 1523 1524 if (list_empty(&msk->conn_list)) { 1525 mptcp_pm_remove_anno_addr(msk, addr, false); 1526 goto next; 1527 } 1528 1529 lock_sock(sk); 1530 remove_subflow = lookup_subflow_by_saddr(&msk->conn_list, addr); 1531 mptcp_pm_remove_anno_addr(msk, addr, remove_subflow && 1532 !(entry->flags & MPTCP_PM_ADDR_FLAG_IMPLICIT)); 1533 1534 list.ids[0] = mptcp_endp_get_local_id(msk, addr); 1535 if (remove_subflow) { 1536 spin_lock_bh(&msk->pm.lock); 1537 mptcp_pm_nl_rm_subflow_received(msk, &list); 1538 spin_unlock_bh(&msk->pm.lock); 1539 } 1540 1541 if (entry->flags & MPTCP_PM_ADDR_FLAG_SUBFLOW) { 1542 spin_lock_bh(&msk->pm.lock); 1543 __mark_subflow_endp_available(msk, list.ids[0]); 1544 spin_unlock_bh(&msk->pm.lock); 1545 } 1546 1547 if (msk->mpc_endpoint_id == entry->addr.id) 1548 msk->mpc_endpoint_id = 0; 1549 release_sock(sk); 1550 1551 next: 1552 sock_put(sk); 1553 cond_resched(); 1554 } 1555 1556 return 0; 1557 } 1558 1559 static int mptcp_nl_remove_id_zero_address(struct net *net, 1560 struct mptcp_addr_info *addr) 1561 { 1562 struct mptcp_rm_list list = { .nr = 0 }; 1563 long s_slot = 0, s_num = 0; 1564 struct mptcp_sock *msk; 1565 1566 list.ids[list.nr++] = 0; 1567 1568 while ((msk = mptcp_token_iter_next(net, &s_slot, &s_num)) != NULL) { 1569 struct sock *sk = (struct sock *)msk; 1570 struct mptcp_addr_info msk_local; 1571 1572 if (list_empty(&msk->conn_list) || mptcp_pm_is_userspace(msk)) 1573 goto next; 1574 1575 mptcp_local_address((struct sock_common *)msk, &msk_local); 1576 if (!mptcp_addresses_equal(&msk_local, addr, addr->port)) 1577 goto next; 1578 1579 lock_sock(sk); 1580 spin_lock_bh(&msk->pm.lock); 1581 mptcp_pm_remove_addr(msk, &list); 1582 mptcp_pm_nl_rm_subflow_received(msk, &list); 1583 __mark_subflow_endp_available(msk, 0); 1584 spin_unlock_bh(&msk->pm.lock); 1585 release_sock(sk); 1586 1587 next: 1588 sock_put(sk); 1589 cond_resched(); 1590 } 1591 1592 return 0; 1593 } 1594 1595 int mptcp_pm_nl_del_addr_doit(struct sk_buff *skb, struct genl_info *info) 1596 { 1597 struct nlattr *attr = info->attrs[MPTCP_PM_ENDPOINT_ADDR]; 1598 struct pm_nl_pernet *pernet = genl_info_pm_nl(info); 1599 struct mptcp_pm_addr_entry addr, *entry; 1600 unsigned int addr_max; 1601 int ret; 1602 1603 ret = mptcp_pm_parse_entry(attr, info, false, &addr); 1604 if (ret < 0) 1605 return ret; 1606 1607 /* the zero id address is special: the first address used by the msk 1608 * always gets such an id, so different subflows can have different zero 1609 * id addresses. Additionally zero id is not accounted for in id_bitmap. 1610 * Let's use an 'mptcp_rm_list' instead of the common remove code. 1611 */ 1612 if (addr.addr.id == 0) 1613 return mptcp_nl_remove_id_zero_address(sock_net(skb->sk), &addr.addr); 1614 1615 spin_lock_bh(&pernet->lock); 1616 entry = __lookup_addr_by_id(pernet, addr.addr.id); 1617 if (!entry) { 1618 GENL_SET_ERR_MSG(info, "address not found"); 1619 spin_unlock_bh(&pernet->lock); 1620 return -EINVAL; 1621 } 1622 if (entry->flags & MPTCP_PM_ADDR_FLAG_SIGNAL) { 1623 addr_max = pernet->add_addr_signal_max; 1624 WRITE_ONCE(pernet->add_addr_signal_max, addr_max - 1); 1625 } 1626 if (entry->flags & MPTCP_PM_ADDR_FLAG_SUBFLOW) { 1627 addr_max = pernet->local_addr_max; 1628 WRITE_ONCE(pernet->local_addr_max, addr_max - 1); 1629 } 1630 1631 pernet->addrs--; 1632 list_del_rcu(&entry->list); 1633 __clear_bit(entry->addr.id, pernet->id_bitmap); 1634 spin_unlock_bh(&pernet->lock); 1635 1636 mptcp_nl_remove_subflow_and_signal_addr(sock_net(skb->sk), entry); 1637 synchronize_rcu(); 1638 __mptcp_pm_release_addr_entry(entry); 1639 1640 return ret; 1641 } 1642 1643 /* Called from the userspace PM only */ 1644 void mptcp_pm_remove_addrs(struct mptcp_sock *msk, struct list_head *rm_list) 1645 { 1646 struct mptcp_rm_list alist = { .nr = 0 }; 1647 struct mptcp_pm_addr_entry *entry; 1648 int anno_nr = 0; 1649 1650 list_for_each_entry(entry, rm_list, list) { 1651 if (alist.nr >= MPTCP_RM_IDS_MAX) 1652 break; 1653 1654 /* only delete if either announced or matching a subflow */ 1655 if (remove_anno_list_by_saddr(msk, &entry->addr)) 1656 anno_nr++; 1657 else if (!lookup_subflow_by_saddr(&msk->conn_list, 1658 &entry->addr)) 1659 continue; 1660 1661 alist.ids[alist.nr++] = entry->addr.id; 1662 } 1663 1664 if (alist.nr) { 1665 spin_lock_bh(&msk->pm.lock); 1666 msk->pm.add_addr_signaled -= anno_nr; 1667 mptcp_pm_remove_addr(msk, &alist); 1668 spin_unlock_bh(&msk->pm.lock); 1669 } 1670 } 1671 1672 /* Called from the in-kernel PM only */ 1673 static void mptcp_pm_flush_addrs_and_subflows(struct mptcp_sock *msk, 1674 struct list_head *rm_list) 1675 { 1676 struct mptcp_rm_list alist = { .nr = 0 }, slist = { .nr = 0 }; 1677 struct mptcp_pm_addr_entry *entry; 1678 1679 list_for_each_entry(entry, rm_list, list) { 1680 if (slist.nr < MPTCP_RM_IDS_MAX && 1681 lookup_subflow_by_saddr(&msk->conn_list, &entry->addr)) 1682 slist.ids[slist.nr++] = mptcp_endp_get_local_id(msk, &entry->addr); 1683 1684 if (alist.nr < MPTCP_RM_IDS_MAX && 1685 remove_anno_list_by_saddr(msk, &entry->addr)) 1686 alist.ids[alist.nr++] = mptcp_endp_get_local_id(msk, &entry->addr); 1687 } 1688 1689 spin_lock_bh(&msk->pm.lock); 1690 if (alist.nr) { 1691 msk->pm.add_addr_signaled -= alist.nr; 1692 mptcp_pm_remove_addr(msk, &alist); 1693 } 1694 if (slist.nr) 1695 mptcp_pm_nl_rm_subflow_received(msk, &slist); 1696 /* Reset counters: maybe some subflows have been removed before */ 1697 bitmap_fill(msk->pm.id_avail_bitmap, MPTCP_PM_MAX_ADDR_ID + 1); 1698 msk->pm.local_addr_used = 0; 1699 spin_unlock_bh(&msk->pm.lock); 1700 } 1701 1702 static void mptcp_nl_flush_addrs_list(struct net *net, 1703 struct list_head *rm_list) 1704 { 1705 long s_slot = 0, s_num = 0; 1706 struct mptcp_sock *msk; 1707 1708 if (list_empty(rm_list)) 1709 return; 1710 1711 while ((msk = mptcp_token_iter_next(net, &s_slot, &s_num)) != NULL) { 1712 struct sock *sk = (struct sock *)msk; 1713 1714 if (!mptcp_pm_is_userspace(msk)) { 1715 lock_sock(sk); 1716 mptcp_pm_flush_addrs_and_subflows(msk, rm_list); 1717 release_sock(sk); 1718 } 1719 1720 sock_put(sk); 1721 cond_resched(); 1722 } 1723 } 1724 1725 /* caller must ensure the RCU grace period is already elapsed */ 1726 static void __flush_addrs(struct list_head *list) 1727 { 1728 while (!list_empty(list)) { 1729 struct mptcp_pm_addr_entry *cur; 1730 1731 cur = list_entry(list->next, 1732 struct mptcp_pm_addr_entry, list); 1733 list_del_rcu(&cur->list); 1734 __mptcp_pm_release_addr_entry(cur); 1735 } 1736 } 1737 1738 static void __reset_counters(struct pm_nl_pernet *pernet) 1739 { 1740 WRITE_ONCE(pernet->add_addr_signal_max, 0); 1741 WRITE_ONCE(pernet->add_addr_accept_max, 0); 1742 WRITE_ONCE(pernet->local_addr_max, 0); 1743 pernet->addrs = 0; 1744 } 1745 1746 int mptcp_pm_nl_flush_addrs_doit(struct sk_buff *skb, struct genl_info *info) 1747 { 1748 struct pm_nl_pernet *pernet = genl_info_pm_nl(info); 1749 LIST_HEAD(free_list); 1750 1751 spin_lock_bh(&pernet->lock); 1752 list_splice_init(&pernet->local_addr_list, &free_list); 1753 __reset_counters(pernet); 1754 pernet->next_id = 1; 1755 bitmap_zero(pernet->id_bitmap, MPTCP_PM_MAX_ADDR_ID + 1); 1756 spin_unlock_bh(&pernet->lock); 1757 mptcp_nl_flush_addrs_list(sock_net(skb->sk), &free_list); 1758 synchronize_rcu(); 1759 __flush_addrs(&free_list); 1760 return 0; 1761 } 1762 1763 int mptcp_nl_fill_addr(struct sk_buff *skb, 1764 struct mptcp_pm_addr_entry *entry) 1765 { 1766 struct mptcp_addr_info *addr = &entry->addr; 1767 struct nlattr *attr; 1768 1769 attr = nla_nest_start(skb, MPTCP_PM_ATTR_ADDR); 1770 if (!attr) 1771 return -EMSGSIZE; 1772 1773 if (nla_put_u16(skb, MPTCP_PM_ADDR_ATTR_FAMILY, addr->family)) 1774 goto nla_put_failure; 1775 if (nla_put_u16(skb, MPTCP_PM_ADDR_ATTR_PORT, ntohs(addr->port))) 1776 goto nla_put_failure; 1777 if (nla_put_u8(skb, MPTCP_PM_ADDR_ATTR_ID, addr->id)) 1778 goto nla_put_failure; 1779 if (nla_put_u32(skb, MPTCP_PM_ADDR_ATTR_FLAGS, entry->flags)) 1780 goto nla_put_failure; 1781 if (entry->ifindex && 1782 nla_put_s32(skb, MPTCP_PM_ADDR_ATTR_IF_IDX, entry->ifindex)) 1783 goto nla_put_failure; 1784 1785 if (addr->family == AF_INET && 1786 nla_put_in_addr(skb, MPTCP_PM_ADDR_ATTR_ADDR4, 1787 addr->addr.s_addr)) 1788 goto nla_put_failure; 1789 #if IS_ENABLED(CONFIG_MPTCP_IPV6) 1790 else if (addr->family == AF_INET6 && 1791 nla_put_in6_addr(skb, MPTCP_PM_ADDR_ATTR_ADDR6, &addr->addr6)) 1792 goto nla_put_failure; 1793 #endif 1794 nla_nest_end(skb, attr); 1795 return 0; 1796 1797 nla_put_failure: 1798 nla_nest_cancel(skb, attr); 1799 return -EMSGSIZE; 1800 } 1801 1802 int mptcp_pm_nl_get_addr(struct sk_buff *skb, struct genl_info *info) 1803 { 1804 struct nlattr *attr = info->attrs[MPTCP_PM_ENDPOINT_ADDR]; 1805 struct pm_nl_pernet *pernet = genl_info_pm_nl(info); 1806 struct mptcp_pm_addr_entry addr, *entry; 1807 struct sk_buff *msg; 1808 void *reply; 1809 int ret; 1810 1811 ret = mptcp_pm_parse_entry(attr, info, false, &addr); 1812 if (ret < 0) 1813 return ret; 1814 1815 msg = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL); 1816 if (!msg) 1817 return -ENOMEM; 1818 1819 reply = genlmsg_put_reply(msg, info, &mptcp_genl_family, 0, 1820 info->genlhdr->cmd); 1821 if (!reply) { 1822 GENL_SET_ERR_MSG(info, "not enough space in Netlink message"); 1823 ret = -EMSGSIZE; 1824 goto fail; 1825 } 1826 1827 spin_lock_bh(&pernet->lock); 1828 entry = __lookup_addr_by_id(pernet, addr.addr.id); 1829 if (!entry) { 1830 GENL_SET_ERR_MSG(info, "address not found"); 1831 ret = -EINVAL; 1832 goto unlock_fail; 1833 } 1834 1835 ret = mptcp_nl_fill_addr(msg, entry); 1836 if (ret) 1837 goto unlock_fail; 1838 1839 genlmsg_end(msg, reply); 1840 ret = genlmsg_reply(msg, info); 1841 spin_unlock_bh(&pernet->lock); 1842 return ret; 1843 1844 unlock_fail: 1845 spin_unlock_bh(&pernet->lock); 1846 1847 fail: 1848 nlmsg_free(msg); 1849 return ret; 1850 } 1851 1852 int mptcp_pm_nl_get_addr_doit(struct sk_buff *skb, struct genl_info *info) 1853 { 1854 return mptcp_pm_get_addr(skb, info); 1855 } 1856 1857 int mptcp_pm_nl_dump_addr(struct sk_buff *msg, 1858 struct netlink_callback *cb) 1859 { 1860 struct net *net = sock_net(msg->sk); 1861 struct mptcp_pm_addr_entry *entry; 1862 struct pm_nl_pernet *pernet; 1863 int id = cb->args[0]; 1864 void *hdr; 1865 int i; 1866 1867 pernet = pm_nl_get_pernet(net); 1868 1869 spin_lock_bh(&pernet->lock); 1870 for (i = id; i < MPTCP_PM_MAX_ADDR_ID + 1; i++) { 1871 if (test_bit(i, pernet->id_bitmap)) { 1872 entry = __lookup_addr_by_id(pernet, i); 1873 if (!entry) 1874 break; 1875 1876 if (entry->addr.id <= id) 1877 continue; 1878 1879 hdr = genlmsg_put(msg, NETLINK_CB(cb->skb).portid, 1880 cb->nlh->nlmsg_seq, &mptcp_genl_family, 1881 NLM_F_MULTI, MPTCP_PM_CMD_GET_ADDR); 1882 if (!hdr) 1883 break; 1884 1885 if (mptcp_nl_fill_addr(msg, entry) < 0) { 1886 genlmsg_cancel(msg, hdr); 1887 break; 1888 } 1889 1890 id = entry->addr.id; 1891 genlmsg_end(msg, hdr); 1892 } 1893 } 1894 spin_unlock_bh(&pernet->lock); 1895 1896 cb->args[0] = id; 1897 return msg->len; 1898 } 1899 1900 int mptcp_pm_nl_get_addr_dumpit(struct sk_buff *msg, 1901 struct netlink_callback *cb) 1902 { 1903 return mptcp_pm_dump_addr(msg, cb); 1904 } 1905 1906 static int parse_limit(struct genl_info *info, int id, unsigned int *limit) 1907 { 1908 struct nlattr *attr = info->attrs[id]; 1909 1910 if (!attr) 1911 return 0; 1912 1913 *limit = nla_get_u32(attr); 1914 if (*limit > MPTCP_PM_ADDR_MAX) { 1915 GENL_SET_ERR_MSG(info, "limit greater than maximum"); 1916 return -EINVAL; 1917 } 1918 return 0; 1919 } 1920 1921 int mptcp_pm_nl_set_limits_doit(struct sk_buff *skb, struct genl_info *info) 1922 { 1923 struct pm_nl_pernet *pernet = genl_info_pm_nl(info); 1924 unsigned int rcv_addrs, subflows; 1925 int ret; 1926 1927 spin_lock_bh(&pernet->lock); 1928 rcv_addrs = pernet->add_addr_accept_max; 1929 ret = parse_limit(info, MPTCP_PM_ATTR_RCV_ADD_ADDRS, &rcv_addrs); 1930 if (ret) 1931 goto unlock; 1932 1933 subflows = pernet->subflows_max; 1934 ret = parse_limit(info, MPTCP_PM_ATTR_SUBFLOWS, &subflows); 1935 if (ret) 1936 goto unlock; 1937 1938 WRITE_ONCE(pernet->add_addr_accept_max, rcv_addrs); 1939 WRITE_ONCE(pernet->subflows_max, subflows); 1940 1941 unlock: 1942 spin_unlock_bh(&pernet->lock); 1943 return ret; 1944 } 1945 1946 int mptcp_pm_nl_get_limits_doit(struct sk_buff *skb, struct genl_info *info) 1947 { 1948 struct pm_nl_pernet *pernet = genl_info_pm_nl(info); 1949 struct sk_buff *msg; 1950 void *reply; 1951 1952 msg = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL); 1953 if (!msg) 1954 return -ENOMEM; 1955 1956 reply = genlmsg_put_reply(msg, info, &mptcp_genl_family, 0, 1957 MPTCP_PM_CMD_GET_LIMITS); 1958 if (!reply) 1959 goto fail; 1960 1961 if (nla_put_u32(msg, MPTCP_PM_ATTR_RCV_ADD_ADDRS, 1962 READ_ONCE(pernet->add_addr_accept_max))) 1963 goto fail; 1964 1965 if (nla_put_u32(msg, MPTCP_PM_ATTR_SUBFLOWS, 1966 READ_ONCE(pernet->subflows_max))) 1967 goto fail; 1968 1969 genlmsg_end(msg, reply); 1970 return genlmsg_reply(msg, info); 1971 1972 fail: 1973 GENL_SET_ERR_MSG(info, "not enough space in Netlink message"); 1974 nlmsg_free(msg); 1975 return -EMSGSIZE; 1976 } 1977 1978 static void mptcp_pm_nl_fullmesh(struct mptcp_sock *msk, 1979 struct mptcp_addr_info *addr) 1980 { 1981 struct mptcp_rm_list list = { .nr = 0 }; 1982 1983 list.ids[list.nr++] = mptcp_endp_get_local_id(msk, addr); 1984 1985 spin_lock_bh(&msk->pm.lock); 1986 mptcp_pm_nl_rm_subflow_received(msk, &list); 1987 __mark_subflow_endp_available(msk, list.ids[0]); 1988 mptcp_pm_create_subflow_or_signal_addr(msk); 1989 spin_unlock_bh(&msk->pm.lock); 1990 } 1991 1992 static int mptcp_nl_set_flags(struct net *net, 1993 struct mptcp_addr_info *addr, 1994 u8 bkup, u8 changed) 1995 { 1996 long s_slot = 0, s_num = 0; 1997 struct mptcp_sock *msk; 1998 int ret = -EINVAL; 1999 2000 while ((msk = mptcp_token_iter_next(net, &s_slot, &s_num)) != NULL) { 2001 struct sock *sk = (struct sock *)msk; 2002 2003 if (list_empty(&msk->conn_list) || mptcp_pm_is_userspace(msk)) 2004 goto next; 2005 2006 lock_sock(sk); 2007 if (changed & MPTCP_PM_ADDR_FLAG_BACKUP) 2008 ret = mptcp_pm_nl_mp_prio_send_ack(msk, addr, NULL, bkup); 2009 if (changed & MPTCP_PM_ADDR_FLAG_FULLMESH) 2010 mptcp_pm_nl_fullmesh(msk, addr); 2011 release_sock(sk); 2012 2013 next: 2014 sock_put(sk); 2015 cond_resched(); 2016 } 2017 2018 return ret; 2019 } 2020 2021 int mptcp_pm_nl_set_flags(struct sk_buff *skb, struct genl_info *info) 2022 { 2023 struct mptcp_pm_addr_entry addr = { .addr = { .family = AF_UNSPEC }, }; 2024 struct nlattr *attr = info->attrs[MPTCP_PM_ATTR_ADDR]; 2025 u8 changed, mask = MPTCP_PM_ADDR_FLAG_BACKUP | 2026 MPTCP_PM_ADDR_FLAG_FULLMESH; 2027 struct net *net = sock_net(skb->sk); 2028 struct mptcp_pm_addr_entry *entry; 2029 struct pm_nl_pernet *pernet; 2030 u8 lookup_by_id = 0; 2031 u8 bkup = 0; 2032 int ret; 2033 2034 pernet = pm_nl_get_pernet(net); 2035 2036 ret = mptcp_pm_parse_entry(attr, info, false, &addr); 2037 if (ret < 0) 2038 return ret; 2039 2040 if (addr.addr.family == AF_UNSPEC) { 2041 lookup_by_id = 1; 2042 if (!addr.addr.id) { 2043 GENL_SET_ERR_MSG(info, "missing required inputs"); 2044 return -EOPNOTSUPP; 2045 } 2046 } 2047 2048 if (addr.flags & MPTCP_PM_ADDR_FLAG_BACKUP) 2049 bkup = 1; 2050 2051 spin_lock_bh(&pernet->lock); 2052 entry = lookup_by_id ? __lookup_addr_by_id(pernet, addr.addr.id) : 2053 __lookup_addr(pernet, &addr.addr); 2054 if (!entry) { 2055 spin_unlock_bh(&pernet->lock); 2056 GENL_SET_ERR_MSG(info, "address not found"); 2057 return -EINVAL; 2058 } 2059 if ((addr.flags & MPTCP_PM_ADDR_FLAG_FULLMESH) && 2060 (entry->flags & MPTCP_PM_ADDR_FLAG_SIGNAL)) { 2061 spin_unlock_bh(&pernet->lock); 2062 GENL_SET_ERR_MSG(info, "invalid addr flags"); 2063 return -EINVAL; 2064 } 2065 2066 changed = (addr.flags ^ entry->flags) & mask; 2067 entry->flags = (entry->flags & ~mask) | (addr.flags & mask); 2068 addr = *entry; 2069 spin_unlock_bh(&pernet->lock); 2070 2071 mptcp_nl_set_flags(net, &addr.addr, bkup, changed); 2072 return 0; 2073 } 2074 2075 int mptcp_pm_nl_set_flags_doit(struct sk_buff *skb, struct genl_info *info) 2076 { 2077 return mptcp_pm_set_flags(skb, info); 2078 } 2079 2080 static void mptcp_nl_mcast_send(struct net *net, struct sk_buff *nlskb, gfp_t gfp) 2081 { 2082 genlmsg_multicast_netns(&mptcp_genl_family, net, 2083 nlskb, 0, MPTCP_PM_EV_GRP_OFFSET, gfp); 2084 } 2085 2086 bool mptcp_userspace_pm_active(const struct mptcp_sock *msk) 2087 { 2088 return genl_has_listeners(&mptcp_genl_family, 2089 sock_net((const struct sock *)msk), 2090 MPTCP_PM_EV_GRP_OFFSET); 2091 } 2092 2093 static int mptcp_event_add_subflow(struct sk_buff *skb, const struct sock *ssk) 2094 { 2095 const struct inet_sock *issk = inet_sk(ssk); 2096 const struct mptcp_subflow_context *sf; 2097 2098 if (nla_put_u16(skb, MPTCP_ATTR_FAMILY, ssk->sk_family)) 2099 return -EMSGSIZE; 2100 2101 switch (ssk->sk_family) { 2102 case AF_INET: 2103 if (nla_put_in_addr(skb, MPTCP_ATTR_SADDR4, issk->inet_saddr)) 2104 return -EMSGSIZE; 2105 if (nla_put_in_addr(skb, MPTCP_ATTR_DADDR4, issk->inet_daddr)) 2106 return -EMSGSIZE; 2107 break; 2108 #if IS_ENABLED(CONFIG_MPTCP_IPV6) 2109 case AF_INET6: { 2110 const struct ipv6_pinfo *np = inet6_sk(ssk); 2111 2112 if (nla_put_in6_addr(skb, MPTCP_ATTR_SADDR6, &np->saddr)) 2113 return -EMSGSIZE; 2114 if (nla_put_in6_addr(skb, MPTCP_ATTR_DADDR6, &ssk->sk_v6_daddr)) 2115 return -EMSGSIZE; 2116 break; 2117 } 2118 #endif 2119 default: 2120 WARN_ON_ONCE(1); 2121 return -EMSGSIZE; 2122 } 2123 2124 if (nla_put_be16(skb, MPTCP_ATTR_SPORT, issk->inet_sport)) 2125 return -EMSGSIZE; 2126 if (nla_put_be16(skb, MPTCP_ATTR_DPORT, issk->inet_dport)) 2127 return -EMSGSIZE; 2128 2129 sf = mptcp_subflow_ctx(ssk); 2130 if (WARN_ON_ONCE(!sf)) 2131 return -EINVAL; 2132 2133 if (nla_put_u8(skb, MPTCP_ATTR_LOC_ID, subflow_get_local_id(sf))) 2134 return -EMSGSIZE; 2135 2136 if (nla_put_u8(skb, MPTCP_ATTR_REM_ID, sf->remote_id)) 2137 return -EMSGSIZE; 2138 2139 return 0; 2140 } 2141 2142 static int mptcp_event_put_token_and_ssk(struct sk_buff *skb, 2143 const struct mptcp_sock *msk, 2144 const struct sock *ssk) 2145 { 2146 const struct sock *sk = (const struct sock *)msk; 2147 const struct mptcp_subflow_context *sf; 2148 u8 sk_err; 2149 2150 if (nla_put_u32(skb, MPTCP_ATTR_TOKEN, READ_ONCE(msk->token))) 2151 return -EMSGSIZE; 2152 2153 if (mptcp_event_add_subflow(skb, ssk)) 2154 return -EMSGSIZE; 2155 2156 sf = mptcp_subflow_ctx(ssk); 2157 if (WARN_ON_ONCE(!sf)) 2158 return -EINVAL; 2159 2160 if (nla_put_u8(skb, MPTCP_ATTR_BACKUP, sf->backup)) 2161 return -EMSGSIZE; 2162 2163 if (ssk->sk_bound_dev_if && 2164 nla_put_s32(skb, MPTCP_ATTR_IF_IDX, ssk->sk_bound_dev_if)) 2165 return -EMSGSIZE; 2166 2167 sk_err = READ_ONCE(ssk->sk_err); 2168 if (sk_err && sk->sk_state == TCP_ESTABLISHED && 2169 nla_put_u8(skb, MPTCP_ATTR_ERROR, sk_err)) 2170 return -EMSGSIZE; 2171 2172 return 0; 2173 } 2174 2175 static int mptcp_event_sub_established(struct sk_buff *skb, 2176 const struct mptcp_sock *msk, 2177 const struct sock *ssk) 2178 { 2179 return mptcp_event_put_token_and_ssk(skb, msk, ssk); 2180 } 2181 2182 static int mptcp_event_sub_closed(struct sk_buff *skb, 2183 const struct mptcp_sock *msk, 2184 const struct sock *ssk) 2185 { 2186 const struct mptcp_subflow_context *sf; 2187 2188 if (mptcp_event_put_token_and_ssk(skb, msk, ssk)) 2189 return -EMSGSIZE; 2190 2191 sf = mptcp_subflow_ctx(ssk); 2192 if (!sf->reset_seen) 2193 return 0; 2194 2195 if (nla_put_u32(skb, MPTCP_ATTR_RESET_REASON, sf->reset_reason)) 2196 return -EMSGSIZE; 2197 2198 if (nla_put_u32(skb, MPTCP_ATTR_RESET_FLAGS, sf->reset_transient)) 2199 return -EMSGSIZE; 2200 2201 return 0; 2202 } 2203 2204 static int mptcp_event_created(struct sk_buff *skb, 2205 const struct mptcp_sock *msk, 2206 const struct sock *ssk) 2207 { 2208 int err = nla_put_u32(skb, MPTCP_ATTR_TOKEN, READ_ONCE(msk->token)); 2209 2210 if (err) 2211 return err; 2212 2213 if (nla_put_u8(skb, MPTCP_ATTR_SERVER_SIDE, READ_ONCE(msk->pm.server_side))) 2214 return -EMSGSIZE; 2215 2216 return mptcp_event_add_subflow(skb, ssk); 2217 } 2218 2219 void mptcp_event_addr_removed(const struct mptcp_sock *msk, uint8_t id) 2220 { 2221 struct net *net = sock_net((const struct sock *)msk); 2222 struct nlmsghdr *nlh; 2223 struct sk_buff *skb; 2224 2225 if (!genl_has_listeners(&mptcp_genl_family, net, MPTCP_PM_EV_GRP_OFFSET)) 2226 return; 2227 2228 skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_ATOMIC); 2229 if (!skb) 2230 return; 2231 2232 nlh = genlmsg_put(skb, 0, 0, &mptcp_genl_family, 0, MPTCP_EVENT_REMOVED); 2233 if (!nlh) 2234 goto nla_put_failure; 2235 2236 if (nla_put_u32(skb, MPTCP_ATTR_TOKEN, READ_ONCE(msk->token))) 2237 goto nla_put_failure; 2238 2239 if (nla_put_u8(skb, MPTCP_ATTR_REM_ID, id)) 2240 goto nla_put_failure; 2241 2242 genlmsg_end(skb, nlh); 2243 mptcp_nl_mcast_send(net, skb, GFP_ATOMIC); 2244 return; 2245 2246 nla_put_failure: 2247 nlmsg_free(skb); 2248 } 2249 2250 void mptcp_event_addr_announced(const struct sock *ssk, 2251 const struct mptcp_addr_info *info) 2252 { 2253 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk); 2254 struct mptcp_sock *msk = mptcp_sk(subflow->conn); 2255 struct net *net = sock_net(ssk); 2256 struct nlmsghdr *nlh; 2257 struct sk_buff *skb; 2258 2259 if (!genl_has_listeners(&mptcp_genl_family, net, MPTCP_PM_EV_GRP_OFFSET)) 2260 return; 2261 2262 skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_ATOMIC); 2263 if (!skb) 2264 return; 2265 2266 nlh = genlmsg_put(skb, 0, 0, &mptcp_genl_family, 0, 2267 MPTCP_EVENT_ANNOUNCED); 2268 if (!nlh) 2269 goto nla_put_failure; 2270 2271 if (nla_put_u32(skb, MPTCP_ATTR_TOKEN, READ_ONCE(msk->token))) 2272 goto nla_put_failure; 2273 2274 if (nla_put_u8(skb, MPTCP_ATTR_REM_ID, info->id)) 2275 goto nla_put_failure; 2276 2277 if (nla_put_be16(skb, MPTCP_ATTR_DPORT, 2278 info->port == 0 ? 2279 inet_sk(ssk)->inet_dport : 2280 info->port)) 2281 goto nla_put_failure; 2282 2283 switch (info->family) { 2284 case AF_INET: 2285 if (nla_put_in_addr(skb, MPTCP_ATTR_DADDR4, info->addr.s_addr)) 2286 goto nla_put_failure; 2287 break; 2288 #if IS_ENABLED(CONFIG_MPTCP_IPV6) 2289 case AF_INET6: 2290 if (nla_put_in6_addr(skb, MPTCP_ATTR_DADDR6, &info->addr6)) 2291 goto nla_put_failure; 2292 break; 2293 #endif 2294 default: 2295 WARN_ON_ONCE(1); 2296 goto nla_put_failure; 2297 } 2298 2299 genlmsg_end(skb, nlh); 2300 mptcp_nl_mcast_send(net, skb, GFP_ATOMIC); 2301 return; 2302 2303 nla_put_failure: 2304 nlmsg_free(skb); 2305 } 2306 2307 void mptcp_event_pm_listener(const struct sock *ssk, 2308 enum mptcp_event_type event) 2309 { 2310 const struct inet_sock *issk = inet_sk(ssk); 2311 struct net *net = sock_net(ssk); 2312 struct nlmsghdr *nlh; 2313 struct sk_buff *skb; 2314 2315 if (!genl_has_listeners(&mptcp_genl_family, net, MPTCP_PM_EV_GRP_OFFSET)) 2316 return; 2317 2318 skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL); 2319 if (!skb) 2320 return; 2321 2322 nlh = genlmsg_put(skb, 0, 0, &mptcp_genl_family, 0, event); 2323 if (!nlh) 2324 goto nla_put_failure; 2325 2326 if (nla_put_u16(skb, MPTCP_ATTR_FAMILY, ssk->sk_family)) 2327 goto nla_put_failure; 2328 2329 if (nla_put_be16(skb, MPTCP_ATTR_SPORT, issk->inet_sport)) 2330 goto nla_put_failure; 2331 2332 switch (ssk->sk_family) { 2333 case AF_INET: 2334 if (nla_put_in_addr(skb, MPTCP_ATTR_SADDR4, issk->inet_saddr)) 2335 goto nla_put_failure; 2336 break; 2337 #if IS_ENABLED(CONFIG_MPTCP_IPV6) 2338 case AF_INET6: { 2339 const struct ipv6_pinfo *np = inet6_sk(ssk); 2340 2341 if (nla_put_in6_addr(skb, MPTCP_ATTR_SADDR6, &np->saddr)) 2342 goto nla_put_failure; 2343 break; 2344 } 2345 #endif 2346 default: 2347 WARN_ON_ONCE(1); 2348 goto nla_put_failure; 2349 } 2350 2351 genlmsg_end(skb, nlh); 2352 mptcp_nl_mcast_send(net, skb, GFP_KERNEL); 2353 return; 2354 2355 nla_put_failure: 2356 nlmsg_free(skb); 2357 } 2358 2359 void mptcp_event(enum mptcp_event_type type, const struct mptcp_sock *msk, 2360 const struct sock *ssk, gfp_t gfp) 2361 { 2362 struct net *net = sock_net((const struct sock *)msk); 2363 struct nlmsghdr *nlh; 2364 struct sk_buff *skb; 2365 2366 if (!genl_has_listeners(&mptcp_genl_family, net, MPTCP_PM_EV_GRP_OFFSET)) 2367 return; 2368 2369 skb = nlmsg_new(NLMSG_DEFAULT_SIZE, gfp); 2370 if (!skb) 2371 return; 2372 2373 nlh = genlmsg_put(skb, 0, 0, &mptcp_genl_family, 0, type); 2374 if (!nlh) 2375 goto nla_put_failure; 2376 2377 switch (type) { 2378 case MPTCP_EVENT_UNSPEC: 2379 WARN_ON_ONCE(1); 2380 break; 2381 case MPTCP_EVENT_CREATED: 2382 case MPTCP_EVENT_ESTABLISHED: 2383 if (mptcp_event_created(skb, msk, ssk) < 0) 2384 goto nla_put_failure; 2385 break; 2386 case MPTCP_EVENT_CLOSED: 2387 if (nla_put_u32(skb, MPTCP_ATTR_TOKEN, READ_ONCE(msk->token)) < 0) 2388 goto nla_put_failure; 2389 break; 2390 case MPTCP_EVENT_ANNOUNCED: 2391 case MPTCP_EVENT_REMOVED: 2392 /* call mptcp_event_addr_announced()/removed instead */ 2393 WARN_ON_ONCE(1); 2394 break; 2395 case MPTCP_EVENT_SUB_ESTABLISHED: 2396 case MPTCP_EVENT_SUB_PRIORITY: 2397 if (mptcp_event_sub_established(skb, msk, ssk) < 0) 2398 goto nla_put_failure; 2399 break; 2400 case MPTCP_EVENT_SUB_CLOSED: 2401 if (mptcp_event_sub_closed(skb, msk, ssk) < 0) 2402 goto nla_put_failure; 2403 break; 2404 case MPTCP_EVENT_LISTENER_CREATED: 2405 case MPTCP_EVENT_LISTENER_CLOSED: 2406 break; 2407 } 2408 2409 genlmsg_end(skb, nlh); 2410 mptcp_nl_mcast_send(net, skb, gfp); 2411 return; 2412 2413 nla_put_failure: 2414 nlmsg_free(skb); 2415 } 2416 2417 struct genl_family mptcp_genl_family __ro_after_init = { 2418 .name = MPTCP_PM_NAME, 2419 .version = MPTCP_PM_VER, 2420 .netnsok = true, 2421 .module = THIS_MODULE, 2422 .ops = mptcp_pm_nl_ops, 2423 .n_ops = ARRAY_SIZE(mptcp_pm_nl_ops), 2424 .resv_start_op = MPTCP_PM_CMD_SUBFLOW_DESTROY + 1, 2425 .mcgrps = mptcp_pm_mcgrps, 2426 .n_mcgrps = ARRAY_SIZE(mptcp_pm_mcgrps), 2427 }; 2428 2429 static int __net_init pm_nl_init_net(struct net *net) 2430 { 2431 struct pm_nl_pernet *pernet = pm_nl_get_pernet(net); 2432 2433 INIT_LIST_HEAD_RCU(&pernet->local_addr_list); 2434 2435 /* Cit. 2 subflows ought to be enough for anybody. */ 2436 pernet->subflows_max = 2; 2437 pernet->next_id = 1; 2438 pernet->stale_loss_cnt = 4; 2439 spin_lock_init(&pernet->lock); 2440 2441 /* No need to initialize other pernet fields, the struct is zeroed at 2442 * allocation time. 2443 */ 2444 2445 return 0; 2446 } 2447 2448 static void __net_exit pm_nl_exit_net(struct list_head *net_list) 2449 { 2450 struct net *net; 2451 2452 list_for_each_entry(net, net_list, exit_list) { 2453 struct pm_nl_pernet *pernet = pm_nl_get_pernet(net); 2454 2455 /* net is removed from namespace list, can't race with 2456 * other modifiers, also netns core already waited for a 2457 * RCU grace period. 2458 */ 2459 __flush_addrs(&pernet->local_addr_list); 2460 } 2461 } 2462 2463 static struct pernet_operations mptcp_pm_pernet_ops = { 2464 .init = pm_nl_init_net, 2465 .exit_batch = pm_nl_exit_net, 2466 .id = &pm_nl_pernet_id, 2467 .size = sizeof(struct pm_nl_pernet), 2468 }; 2469 2470 void __init mptcp_pm_nl_init(void) 2471 { 2472 if (register_pernet_subsys(&mptcp_pm_pernet_ops) < 0) 2473 panic("Failed to register MPTCP PM pernet subsystem.\n"); 2474 2475 if (genl_register_family(&mptcp_genl_family)) 2476 panic("Failed to register MPTCP PM netlink family\n"); 2477 } 2478