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; 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 mptcp_pm_send_ack(msk, subflow, false, false); 797 break; 798 } 799 } 800 } 801 802 int mptcp_pm_nl_mp_prio_send_ack(struct mptcp_sock *msk, 803 struct mptcp_addr_info *addr, 804 struct mptcp_addr_info *rem, 805 u8 bkup) 806 { 807 struct mptcp_subflow_context *subflow; 808 809 pr_debug("bkup=%d\n", bkup); 810 811 mptcp_for_each_subflow(msk, subflow) { 812 struct sock *ssk = mptcp_subflow_tcp_sock(subflow); 813 struct mptcp_addr_info local, remote; 814 815 mptcp_local_address((struct sock_common *)ssk, &local); 816 if (!mptcp_addresses_equal(&local, addr, addr->port)) 817 continue; 818 819 if (rem && rem->family != AF_UNSPEC) { 820 remote_address((struct sock_common *)ssk, &remote); 821 if (!mptcp_addresses_equal(&remote, rem, rem->port)) 822 continue; 823 } 824 825 __mptcp_pm_send_ack(msk, subflow, true, bkup); 826 return 0; 827 } 828 829 return -EINVAL; 830 } 831 832 static void mptcp_pm_nl_rm_addr_or_subflow(struct mptcp_sock *msk, 833 const struct mptcp_rm_list *rm_list, 834 enum linux_mptcp_mib_field rm_type) 835 { 836 struct mptcp_subflow_context *subflow, *tmp; 837 struct sock *sk = (struct sock *)msk; 838 u8 i; 839 840 pr_debug("%s rm_list_nr %d\n", 841 rm_type == MPTCP_MIB_RMADDR ? "address" : "subflow", rm_list->nr); 842 843 msk_owned_by_me(msk); 844 845 if (sk->sk_state == TCP_LISTEN) 846 return; 847 848 if (!rm_list->nr) 849 return; 850 851 if (list_empty(&msk->conn_list)) 852 return; 853 854 for (i = 0; i < rm_list->nr; i++) { 855 u8 rm_id = rm_list->ids[i]; 856 bool removed = false; 857 858 mptcp_for_each_subflow_safe(msk, subflow, tmp) { 859 struct sock *ssk = mptcp_subflow_tcp_sock(subflow); 860 u8 remote_id = READ_ONCE(subflow->remote_id); 861 int how = RCV_SHUTDOWN | SEND_SHUTDOWN; 862 u8 id = subflow_get_local_id(subflow); 863 864 if ((1 << inet_sk_state_load(ssk)) & 865 (TCPF_FIN_WAIT1 | TCPF_FIN_WAIT2 | TCPF_CLOSING | TCPF_CLOSE)) 866 continue; 867 if (rm_type == MPTCP_MIB_RMADDR && remote_id != rm_id) 868 continue; 869 if (rm_type == MPTCP_MIB_RMSUBFLOW && id != rm_id) 870 continue; 871 872 pr_debug(" -> %s rm_list_ids[%d]=%u local_id=%u remote_id=%u mpc_id=%u\n", 873 rm_type == MPTCP_MIB_RMADDR ? "address" : "subflow", 874 i, rm_id, id, remote_id, msk->mpc_endpoint_id); 875 spin_unlock_bh(&msk->pm.lock); 876 mptcp_subflow_shutdown(sk, ssk, how); 877 removed |= subflow->request_join; 878 879 /* the following takes care of updating the subflows counter */ 880 mptcp_close_ssk(sk, ssk, subflow); 881 spin_lock_bh(&msk->pm.lock); 882 883 if (rm_type == MPTCP_MIB_RMSUBFLOW) 884 __MPTCP_INC_STATS(sock_net(sk), rm_type); 885 } 886 887 if (rm_type == MPTCP_MIB_RMADDR) 888 __MPTCP_INC_STATS(sock_net(sk), rm_type); 889 890 if (!removed) 891 continue; 892 893 if (!mptcp_pm_is_kernel(msk)) 894 continue; 895 896 if (rm_type == MPTCP_MIB_RMADDR && rm_id && 897 !WARN_ON_ONCE(msk->pm.add_addr_accepted == 0)) { 898 /* Note: if the subflow has been closed before, this 899 * add_addr_accepted counter will not be decremented. 900 */ 901 if (--msk->pm.add_addr_accepted < mptcp_pm_get_add_addr_accept_max(msk)) 902 WRITE_ONCE(msk->pm.accept_addr, true); 903 } 904 } 905 } 906 907 static void mptcp_pm_nl_rm_addr_received(struct mptcp_sock *msk) 908 { 909 mptcp_pm_nl_rm_addr_or_subflow(msk, &msk->pm.rm_list_rx, MPTCP_MIB_RMADDR); 910 } 911 912 static void mptcp_pm_nl_rm_subflow_received(struct mptcp_sock *msk, 913 const struct mptcp_rm_list *rm_list) 914 { 915 mptcp_pm_nl_rm_addr_or_subflow(msk, rm_list, MPTCP_MIB_RMSUBFLOW); 916 } 917 918 void mptcp_pm_nl_work(struct mptcp_sock *msk) 919 { 920 struct mptcp_pm_data *pm = &msk->pm; 921 922 msk_owned_by_me(msk); 923 924 if (!(pm->status & MPTCP_PM_WORK_MASK)) 925 return; 926 927 spin_lock_bh(&msk->pm.lock); 928 929 pr_debug("msk=%p status=%x\n", msk, pm->status); 930 if (pm->status & BIT(MPTCP_PM_ADD_ADDR_RECEIVED)) { 931 pm->status &= ~BIT(MPTCP_PM_ADD_ADDR_RECEIVED); 932 mptcp_pm_nl_add_addr_received(msk); 933 } 934 if (pm->status & BIT(MPTCP_PM_ADD_ADDR_SEND_ACK)) { 935 pm->status &= ~BIT(MPTCP_PM_ADD_ADDR_SEND_ACK); 936 mptcp_pm_nl_addr_send_ack(msk); 937 } 938 if (pm->status & BIT(MPTCP_PM_RM_ADDR_RECEIVED)) { 939 pm->status &= ~BIT(MPTCP_PM_RM_ADDR_RECEIVED); 940 mptcp_pm_nl_rm_addr_received(msk); 941 } 942 if (pm->status & BIT(MPTCP_PM_ESTABLISHED)) { 943 pm->status &= ~BIT(MPTCP_PM_ESTABLISHED); 944 mptcp_pm_nl_fully_established(msk); 945 } 946 if (pm->status & BIT(MPTCP_PM_SUBFLOW_ESTABLISHED)) { 947 pm->status &= ~BIT(MPTCP_PM_SUBFLOW_ESTABLISHED); 948 mptcp_pm_nl_subflow_established(msk); 949 } 950 951 spin_unlock_bh(&msk->pm.lock); 952 } 953 954 static bool address_use_port(struct mptcp_pm_addr_entry *entry) 955 { 956 return (entry->flags & 957 (MPTCP_PM_ADDR_FLAG_SIGNAL | MPTCP_PM_ADDR_FLAG_SUBFLOW)) == 958 MPTCP_PM_ADDR_FLAG_SIGNAL; 959 } 960 961 /* caller must ensure the RCU grace period is already elapsed */ 962 static void __mptcp_pm_release_addr_entry(struct mptcp_pm_addr_entry *entry) 963 { 964 if (entry->lsk) 965 sock_release(entry->lsk); 966 kfree(entry); 967 } 968 969 static int mptcp_pm_nl_append_new_local_addr(struct pm_nl_pernet *pernet, 970 struct mptcp_pm_addr_entry *entry, 971 bool needs_id) 972 { 973 struct mptcp_pm_addr_entry *cur, *del_entry = NULL; 974 unsigned int addr_max; 975 int ret = -EINVAL; 976 977 spin_lock_bh(&pernet->lock); 978 /* to keep the code simple, don't do IDR-like allocation for address ID, 979 * just bail when we exceed limits 980 */ 981 if (pernet->next_id == MPTCP_PM_MAX_ADDR_ID) 982 pernet->next_id = 1; 983 if (pernet->addrs >= MPTCP_PM_ADDR_MAX) { 984 ret = -ERANGE; 985 goto out; 986 } 987 if (test_bit(entry->addr.id, pernet->id_bitmap)) { 988 ret = -EBUSY; 989 goto out; 990 } 991 992 /* do not insert duplicate address, differentiate on port only 993 * singled addresses 994 */ 995 if (!address_use_port(entry)) 996 entry->addr.port = 0; 997 list_for_each_entry(cur, &pernet->local_addr_list, list) { 998 if (mptcp_addresses_equal(&cur->addr, &entry->addr, 999 cur->addr.port || entry->addr.port)) { 1000 /* allow replacing the exiting endpoint only if such 1001 * endpoint is an implicit one and the user-space 1002 * did not provide an endpoint id 1003 */ 1004 if (!(cur->flags & MPTCP_PM_ADDR_FLAG_IMPLICIT)) { 1005 ret = -EEXIST; 1006 goto out; 1007 } 1008 if (entry->addr.id) 1009 goto out; 1010 1011 pernet->addrs--; 1012 entry->addr.id = cur->addr.id; 1013 list_del_rcu(&cur->list); 1014 del_entry = cur; 1015 break; 1016 } 1017 } 1018 1019 if (!entry->addr.id && needs_id) { 1020 find_next: 1021 entry->addr.id = find_next_zero_bit(pernet->id_bitmap, 1022 MPTCP_PM_MAX_ADDR_ID + 1, 1023 pernet->next_id); 1024 if (!entry->addr.id && pernet->next_id != 1) { 1025 pernet->next_id = 1; 1026 goto find_next; 1027 } 1028 } 1029 1030 if (!entry->addr.id && needs_id) 1031 goto out; 1032 1033 __set_bit(entry->addr.id, pernet->id_bitmap); 1034 if (entry->addr.id > pernet->next_id) 1035 pernet->next_id = entry->addr.id; 1036 1037 if (entry->flags & MPTCP_PM_ADDR_FLAG_SIGNAL) { 1038 addr_max = pernet->add_addr_signal_max; 1039 WRITE_ONCE(pernet->add_addr_signal_max, addr_max + 1); 1040 } 1041 if (entry->flags & MPTCP_PM_ADDR_FLAG_SUBFLOW) { 1042 addr_max = pernet->local_addr_max; 1043 WRITE_ONCE(pernet->local_addr_max, addr_max + 1); 1044 } 1045 1046 pernet->addrs++; 1047 if (!entry->addr.port) 1048 list_add_tail_rcu(&entry->list, &pernet->local_addr_list); 1049 else 1050 list_add_rcu(&entry->list, &pernet->local_addr_list); 1051 ret = entry->addr.id; 1052 1053 out: 1054 spin_unlock_bh(&pernet->lock); 1055 1056 /* just replaced an existing entry, free it */ 1057 if (del_entry) { 1058 synchronize_rcu(); 1059 __mptcp_pm_release_addr_entry(del_entry); 1060 } 1061 return ret; 1062 } 1063 1064 static struct lock_class_key mptcp_slock_keys[2]; 1065 static struct lock_class_key mptcp_keys[2]; 1066 1067 static int mptcp_pm_nl_create_listen_socket(struct sock *sk, 1068 struct mptcp_pm_addr_entry *entry) 1069 { 1070 bool is_ipv6 = sk->sk_family == AF_INET6; 1071 int addrlen = sizeof(struct sockaddr_in); 1072 struct sockaddr_storage addr; 1073 struct sock *newsk, *ssk; 1074 int backlog = 1024; 1075 int err; 1076 1077 err = sock_create_kern(sock_net(sk), entry->addr.family, 1078 SOCK_STREAM, IPPROTO_MPTCP, &entry->lsk); 1079 if (err) 1080 return err; 1081 1082 newsk = entry->lsk->sk; 1083 if (!newsk) 1084 return -EINVAL; 1085 1086 /* The subflow socket lock is acquired in a nested to the msk one 1087 * in several places, even by the TCP stack, and this msk is a kernel 1088 * socket: lockdep complains. Instead of propagating the _nested 1089 * modifiers in several places, re-init the lock class for the msk 1090 * socket to an mptcp specific one. 1091 */ 1092 sock_lock_init_class_and_name(newsk, 1093 is_ipv6 ? "mlock-AF_INET6" : "mlock-AF_INET", 1094 &mptcp_slock_keys[is_ipv6], 1095 is_ipv6 ? "msk_lock-AF_INET6" : "msk_lock-AF_INET", 1096 &mptcp_keys[is_ipv6]); 1097 1098 lock_sock(newsk); 1099 ssk = __mptcp_nmpc_sk(mptcp_sk(newsk)); 1100 release_sock(newsk); 1101 if (IS_ERR(ssk)) 1102 return PTR_ERR(ssk); 1103 1104 mptcp_info2sockaddr(&entry->addr, &addr, entry->addr.family); 1105 #if IS_ENABLED(CONFIG_MPTCP_IPV6) 1106 if (entry->addr.family == AF_INET6) 1107 addrlen = sizeof(struct sockaddr_in6); 1108 #endif 1109 if (ssk->sk_family == AF_INET) 1110 err = inet_bind_sk(ssk, (struct sockaddr *)&addr, addrlen); 1111 #if IS_ENABLED(CONFIG_MPTCP_IPV6) 1112 else if (ssk->sk_family == AF_INET6) 1113 err = inet6_bind_sk(ssk, (struct sockaddr *)&addr, addrlen); 1114 #endif 1115 if (err) 1116 return err; 1117 1118 /* We don't use mptcp_set_state() here because it needs to be called 1119 * under the msk socket lock. For the moment, that will not bring 1120 * anything more than only calling inet_sk_state_store(), because the 1121 * old status is known (TCP_CLOSE). 1122 */ 1123 inet_sk_state_store(newsk, TCP_LISTEN); 1124 lock_sock(ssk); 1125 WRITE_ONCE(mptcp_subflow_ctx(ssk)->pm_listener, true); 1126 err = __inet_listen_sk(ssk, backlog); 1127 if (!err) 1128 mptcp_event_pm_listener(ssk, MPTCP_EVENT_LISTENER_CREATED); 1129 release_sock(ssk); 1130 return err; 1131 } 1132 1133 int mptcp_pm_nl_get_local_id(struct mptcp_sock *msk, struct mptcp_addr_info *skc) 1134 { 1135 struct mptcp_pm_addr_entry *entry; 1136 struct pm_nl_pernet *pernet; 1137 int ret = -1; 1138 1139 pernet = pm_nl_get_pernet_from_msk(msk); 1140 1141 rcu_read_lock(); 1142 list_for_each_entry_rcu(entry, &pernet->local_addr_list, list) { 1143 if (mptcp_addresses_equal(&entry->addr, skc, entry->addr.port)) { 1144 ret = entry->addr.id; 1145 break; 1146 } 1147 } 1148 rcu_read_unlock(); 1149 if (ret >= 0) 1150 return ret; 1151 1152 /* address not found, add to local list */ 1153 entry = kmalloc(sizeof(*entry), GFP_ATOMIC); 1154 if (!entry) 1155 return -ENOMEM; 1156 1157 entry->addr = *skc; 1158 entry->addr.id = 0; 1159 entry->addr.port = 0; 1160 entry->ifindex = 0; 1161 entry->flags = MPTCP_PM_ADDR_FLAG_IMPLICIT; 1162 entry->lsk = NULL; 1163 ret = mptcp_pm_nl_append_new_local_addr(pernet, entry, true); 1164 if (ret < 0) 1165 kfree(entry); 1166 1167 return ret; 1168 } 1169 1170 bool mptcp_pm_nl_is_backup(struct mptcp_sock *msk, struct mptcp_addr_info *skc) 1171 { 1172 struct pm_nl_pernet *pernet = pm_nl_get_pernet_from_msk(msk); 1173 struct mptcp_pm_addr_entry *entry; 1174 bool backup = false; 1175 1176 rcu_read_lock(); 1177 list_for_each_entry_rcu(entry, &pernet->local_addr_list, list) { 1178 if (mptcp_addresses_equal(&entry->addr, skc, entry->addr.port)) { 1179 backup = !!(entry->flags & MPTCP_PM_ADDR_FLAG_BACKUP); 1180 break; 1181 } 1182 } 1183 rcu_read_unlock(); 1184 1185 return backup; 1186 } 1187 1188 #define MPTCP_PM_CMD_GRP_OFFSET 0 1189 #define MPTCP_PM_EV_GRP_OFFSET 1 1190 1191 static const struct genl_multicast_group mptcp_pm_mcgrps[] = { 1192 [MPTCP_PM_CMD_GRP_OFFSET] = { .name = MPTCP_PM_CMD_GRP_NAME, }, 1193 [MPTCP_PM_EV_GRP_OFFSET] = { .name = MPTCP_PM_EV_GRP_NAME, 1194 .flags = GENL_MCAST_CAP_NET_ADMIN, 1195 }, 1196 }; 1197 1198 void mptcp_pm_nl_subflow_chk_stale(const struct mptcp_sock *msk, struct sock *ssk) 1199 { 1200 struct mptcp_subflow_context *iter, *subflow = mptcp_subflow_ctx(ssk); 1201 struct sock *sk = (struct sock *)msk; 1202 unsigned int active_max_loss_cnt; 1203 struct net *net = sock_net(sk); 1204 unsigned int stale_loss_cnt; 1205 bool slow; 1206 1207 stale_loss_cnt = mptcp_stale_loss_cnt(net); 1208 if (subflow->stale || !stale_loss_cnt || subflow->stale_count <= stale_loss_cnt) 1209 return; 1210 1211 /* look for another available subflow not in loss state */ 1212 active_max_loss_cnt = max_t(int, stale_loss_cnt - 1, 1); 1213 mptcp_for_each_subflow(msk, iter) { 1214 if (iter != subflow && mptcp_subflow_active(iter) && 1215 iter->stale_count < active_max_loss_cnt) { 1216 /* we have some alternatives, try to mark this subflow as idle ...*/ 1217 slow = lock_sock_fast(ssk); 1218 if (!tcp_rtx_and_write_queues_empty(ssk)) { 1219 subflow->stale = 1; 1220 __mptcp_retransmit_pending_data(sk); 1221 MPTCP_INC_STATS(net, MPTCP_MIB_SUBFLOWSTALE); 1222 } 1223 unlock_sock_fast(ssk, slow); 1224 1225 /* always try to push the pending data regardless of re-injections: 1226 * we can possibly use backup subflows now, and subflow selection 1227 * is cheap under the msk socket lock 1228 */ 1229 __mptcp_push_pending(sk, 0); 1230 return; 1231 } 1232 } 1233 } 1234 1235 static int mptcp_pm_family_to_addr(int family) 1236 { 1237 #if IS_ENABLED(CONFIG_MPTCP_IPV6) 1238 if (family == AF_INET6) 1239 return MPTCP_PM_ADDR_ATTR_ADDR6; 1240 #endif 1241 return MPTCP_PM_ADDR_ATTR_ADDR4; 1242 } 1243 1244 static int mptcp_pm_parse_pm_addr_attr(struct nlattr *tb[], 1245 const struct nlattr *attr, 1246 struct genl_info *info, 1247 struct mptcp_addr_info *addr, 1248 bool require_family) 1249 { 1250 int err, addr_addr; 1251 1252 if (!attr) { 1253 GENL_SET_ERR_MSG(info, "missing address info"); 1254 return -EINVAL; 1255 } 1256 1257 /* no validation needed - was already done via nested policy */ 1258 err = nla_parse_nested_deprecated(tb, MPTCP_PM_ADDR_ATTR_MAX, attr, 1259 mptcp_pm_address_nl_policy, info->extack); 1260 if (err) 1261 return err; 1262 1263 if (tb[MPTCP_PM_ADDR_ATTR_ID]) 1264 addr->id = nla_get_u8(tb[MPTCP_PM_ADDR_ATTR_ID]); 1265 1266 if (!tb[MPTCP_PM_ADDR_ATTR_FAMILY]) { 1267 if (!require_family) 1268 return 0; 1269 1270 NL_SET_ERR_MSG_ATTR(info->extack, attr, 1271 "missing family"); 1272 return -EINVAL; 1273 } 1274 1275 addr->family = nla_get_u16(tb[MPTCP_PM_ADDR_ATTR_FAMILY]); 1276 if (addr->family != AF_INET 1277 #if IS_ENABLED(CONFIG_MPTCP_IPV6) 1278 && addr->family != AF_INET6 1279 #endif 1280 ) { 1281 NL_SET_ERR_MSG_ATTR(info->extack, attr, 1282 "unknown address family"); 1283 return -EINVAL; 1284 } 1285 addr_addr = mptcp_pm_family_to_addr(addr->family); 1286 if (!tb[addr_addr]) { 1287 NL_SET_ERR_MSG_ATTR(info->extack, attr, 1288 "missing address data"); 1289 return -EINVAL; 1290 } 1291 1292 #if IS_ENABLED(CONFIG_MPTCP_IPV6) 1293 if (addr->family == AF_INET6) 1294 addr->addr6 = nla_get_in6_addr(tb[addr_addr]); 1295 else 1296 #endif 1297 addr->addr.s_addr = nla_get_in_addr(tb[addr_addr]); 1298 1299 if (tb[MPTCP_PM_ADDR_ATTR_PORT]) 1300 addr->port = htons(nla_get_u16(tb[MPTCP_PM_ADDR_ATTR_PORT])); 1301 1302 return 0; 1303 } 1304 1305 int mptcp_pm_parse_addr(struct nlattr *attr, struct genl_info *info, 1306 struct mptcp_addr_info *addr) 1307 { 1308 struct nlattr *tb[MPTCP_PM_ADDR_ATTR_MAX + 1]; 1309 1310 memset(addr, 0, sizeof(*addr)); 1311 1312 return mptcp_pm_parse_pm_addr_attr(tb, attr, info, addr, true); 1313 } 1314 1315 int mptcp_pm_parse_entry(struct nlattr *attr, struct genl_info *info, 1316 bool require_family, 1317 struct mptcp_pm_addr_entry *entry) 1318 { 1319 struct nlattr *tb[MPTCP_PM_ADDR_ATTR_MAX + 1]; 1320 int err; 1321 1322 memset(entry, 0, sizeof(*entry)); 1323 1324 err = mptcp_pm_parse_pm_addr_attr(tb, attr, info, &entry->addr, require_family); 1325 if (err) 1326 return err; 1327 1328 if (tb[MPTCP_PM_ADDR_ATTR_IF_IDX]) { 1329 u32 val = nla_get_s32(tb[MPTCP_PM_ADDR_ATTR_IF_IDX]); 1330 1331 entry->ifindex = val; 1332 } 1333 1334 if (tb[MPTCP_PM_ADDR_ATTR_FLAGS]) 1335 entry->flags = nla_get_u32(tb[MPTCP_PM_ADDR_ATTR_FLAGS]); 1336 1337 if (tb[MPTCP_PM_ADDR_ATTR_PORT]) 1338 entry->addr.port = htons(nla_get_u16(tb[MPTCP_PM_ADDR_ATTR_PORT])); 1339 1340 return 0; 1341 } 1342 1343 static struct pm_nl_pernet *genl_info_pm_nl(struct genl_info *info) 1344 { 1345 return pm_nl_get_pernet(genl_info_net(info)); 1346 } 1347 1348 static int mptcp_nl_add_subflow_or_signal_addr(struct net *net, 1349 struct mptcp_addr_info *addr) 1350 { 1351 struct mptcp_sock *msk; 1352 long s_slot = 0, s_num = 0; 1353 1354 while ((msk = mptcp_token_iter_next(net, &s_slot, &s_num)) != NULL) { 1355 struct sock *sk = (struct sock *)msk; 1356 struct mptcp_addr_info mpc_addr; 1357 1358 if (!READ_ONCE(msk->fully_established) || 1359 mptcp_pm_is_userspace(msk)) 1360 goto next; 1361 1362 /* if the endp linked to the init sf is re-added with a != ID */ 1363 mptcp_local_address((struct sock_common *)msk, &mpc_addr); 1364 1365 lock_sock(sk); 1366 spin_lock_bh(&msk->pm.lock); 1367 if (mptcp_addresses_equal(addr, &mpc_addr, addr->port)) 1368 msk->mpc_endpoint_id = addr->id; 1369 mptcp_pm_create_subflow_or_signal_addr(msk); 1370 spin_unlock_bh(&msk->pm.lock); 1371 release_sock(sk); 1372 1373 next: 1374 sock_put(sk); 1375 cond_resched(); 1376 } 1377 1378 return 0; 1379 } 1380 1381 static bool mptcp_pm_has_addr_attr_id(const struct nlattr *attr, 1382 struct genl_info *info) 1383 { 1384 struct nlattr *tb[MPTCP_PM_ADDR_ATTR_MAX + 1]; 1385 1386 if (!nla_parse_nested_deprecated(tb, MPTCP_PM_ADDR_ATTR_MAX, attr, 1387 mptcp_pm_address_nl_policy, info->extack) && 1388 tb[MPTCP_PM_ADDR_ATTR_ID]) 1389 return true; 1390 return false; 1391 } 1392 1393 int mptcp_pm_nl_add_addr_doit(struct sk_buff *skb, struct genl_info *info) 1394 { 1395 struct nlattr *attr = info->attrs[MPTCP_PM_ENDPOINT_ADDR]; 1396 struct pm_nl_pernet *pernet = genl_info_pm_nl(info); 1397 struct mptcp_pm_addr_entry addr, *entry; 1398 int ret; 1399 1400 ret = mptcp_pm_parse_entry(attr, info, true, &addr); 1401 if (ret < 0) 1402 return ret; 1403 1404 if (addr.addr.port && !address_use_port(&addr)) { 1405 GENL_SET_ERR_MSG(info, "flags must have signal and not subflow when using port"); 1406 return -EINVAL; 1407 } 1408 1409 if (addr.flags & MPTCP_PM_ADDR_FLAG_SIGNAL && 1410 addr.flags & MPTCP_PM_ADDR_FLAG_FULLMESH) { 1411 GENL_SET_ERR_MSG(info, "flags mustn't have both signal and fullmesh"); 1412 return -EINVAL; 1413 } 1414 1415 if (addr.flags & MPTCP_PM_ADDR_FLAG_IMPLICIT) { 1416 GENL_SET_ERR_MSG(info, "can't create IMPLICIT endpoint"); 1417 return -EINVAL; 1418 } 1419 1420 entry = kzalloc(sizeof(*entry), GFP_KERNEL_ACCOUNT); 1421 if (!entry) { 1422 GENL_SET_ERR_MSG(info, "can't allocate addr"); 1423 return -ENOMEM; 1424 } 1425 1426 *entry = addr; 1427 if (entry->addr.port) { 1428 ret = mptcp_pm_nl_create_listen_socket(skb->sk, entry); 1429 if (ret) { 1430 GENL_SET_ERR_MSG_FMT(info, "create listen socket error: %d", ret); 1431 goto out_free; 1432 } 1433 } 1434 ret = mptcp_pm_nl_append_new_local_addr(pernet, entry, 1435 !mptcp_pm_has_addr_attr_id(attr, info)); 1436 if (ret < 0) { 1437 GENL_SET_ERR_MSG_FMT(info, "too many addresses or duplicate one: %d", ret); 1438 goto out_free; 1439 } 1440 1441 mptcp_nl_add_subflow_or_signal_addr(sock_net(skb->sk), &entry->addr); 1442 return 0; 1443 1444 out_free: 1445 __mptcp_pm_release_addr_entry(entry); 1446 return ret; 1447 } 1448 1449 static bool remove_anno_list_by_saddr(struct mptcp_sock *msk, 1450 const struct mptcp_addr_info *addr) 1451 { 1452 struct mptcp_pm_add_entry *entry; 1453 1454 entry = mptcp_pm_del_add_timer(msk, addr, false); 1455 if (entry) { 1456 kfree(entry); 1457 return true; 1458 } 1459 1460 return false; 1461 } 1462 1463 static u8 mptcp_endp_get_local_id(struct mptcp_sock *msk, 1464 const struct mptcp_addr_info *addr) 1465 { 1466 return msk->mpc_endpoint_id == addr->id ? 0 : addr->id; 1467 } 1468 1469 static bool mptcp_pm_remove_anno_addr(struct mptcp_sock *msk, 1470 const struct mptcp_addr_info *addr, 1471 bool force) 1472 { 1473 struct mptcp_rm_list list = { .nr = 0 }; 1474 bool ret; 1475 1476 list.ids[list.nr++] = mptcp_endp_get_local_id(msk, addr); 1477 1478 ret = remove_anno_list_by_saddr(msk, addr); 1479 if (ret || force) { 1480 spin_lock_bh(&msk->pm.lock); 1481 if (ret) { 1482 __set_bit(addr->id, msk->pm.id_avail_bitmap); 1483 msk->pm.add_addr_signaled--; 1484 } 1485 mptcp_pm_remove_addr(msk, &list); 1486 spin_unlock_bh(&msk->pm.lock); 1487 } 1488 return ret; 1489 } 1490 1491 static void __mark_subflow_endp_available(struct mptcp_sock *msk, u8 id) 1492 { 1493 /* If it was marked as used, and not ID 0, decrement local_addr_used */ 1494 if (!__test_and_set_bit(id ? : msk->mpc_endpoint_id, msk->pm.id_avail_bitmap) && 1495 id && !WARN_ON_ONCE(msk->pm.local_addr_used == 0)) 1496 msk->pm.local_addr_used--; 1497 } 1498 1499 static int mptcp_nl_remove_subflow_and_signal_addr(struct net *net, 1500 const struct mptcp_pm_addr_entry *entry) 1501 { 1502 const struct mptcp_addr_info *addr = &entry->addr; 1503 struct mptcp_rm_list list = { .nr = 1 }; 1504 long s_slot = 0, s_num = 0; 1505 struct mptcp_sock *msk; 1506 1507 pr_debug("remove_id=%d\n", addr->id); 1508 1509 while ((msk = mptcp_token_iter_next(net, &s_slot, &s_num)) != NULL) { 1510 struct sock *sk = (struct sock *)msk; 1511 bool remove_subflow; 1512 1513 if (mptcp_pm_is_userspace(msk)) 1514 goto next; 1515 1516 if (list_empty(&msk->conn_list)) { 1517 mptcp_pm_remove_anno_addr(msk, addr, false); 1518 goto next; 1519 } 1520 1521 lock_sock(sk); 1522 remove_subflow = lookup_subflow_by_saddr(&msk->conn_list, addr); 1523 mptcp_pm_remove_anno_addr(msk, addr, remove_subflow && 1524 !(entry->flags & MPTCP_PM_ADDR_FLAG_IMPLICIT)); 1525 1526 list.ids[0] = mptcp_endp_get_local_id(msk, addr); 1527 if (remove_subflow) { 1528 spin_lock_bh(&msk->pm.lock); 1529 mptcp_pm_nl_rm_subflow_received(msk, &list); 1530 spin_unlock_bh(&msk->pm.lock); 1531 } 1532 1533 if (entry->flags & MPTCP_PM_ADDR_FLAG_SUBFLOW) { 1534 spin_lock_bh(&msk->pm.lock); 1535 __mark_subflow_endp_available(msk, list.ids[0]); 1536 spin_unlock_bh(&msk->pm.lock); 1537 } 1538 1539 if (msk->mpc_endpoint_id == entry->addr.id) 1540 msk->mpc_endpoint_id = 0; 1541 release_sock(sk); 1542 1543 next: 1544 sock_put(sk); 1545 cond_resched(); 1546 } 1547 1548 return 0; 1549 } 1550 1551 static int mptcp_nl_remove_id_zero_address(struct net *net, 1552 struct mptcp_addr_info *addr) 1553 { 1554 struct mptcp_rm_list list = { .nr = 0 }; 1555 long s_slot = 0, s_num = 0; 1556 struct mptcp_sock *msk; 1557 1558 list.ids[list.nr++] = 0; 1559 1560 while ((msk = mptcp_token_iter_next(net, &s_slot, &s_num)) != NULL) { 1561 struct sock *sk = (struct sock *)msk; 1562 struct mptcp_addr_info msk_local; 1563 1564 if (list_empty(&msk->conn_list) || mptcp_pm_is_userspace(msk)) 1565 goto next; 1566 1567 mptcp_local_address((struct sock_common *)msk, &msk_local); 1568 if (!mptcp_addresses_equal(&msk_local, addr, addr->port)) 1569 goto next; 1570 1571 lock_sock(sk); 1572 spin_lock_bh(&msk->pm.lock); 1573 mptcp_pm_remove_addr(msk, &list); 1574 mptcp_pm_nl_rm_subflow_received(msk, &list); 1575 __mark_subflow_endp_available(msk, 0); 1576 spin_unlock_bh(&msk->pm.lock); 1577 release_sock(sk); 1578 1579 next: 1580 sock_put(sk); 1581 cond_resched(); 1582 } 1583 1584 return 0; 1585 } 1586 1587 int mptcp_pm_nl_del_addr_doit(struct sk_buff *skb, struct genl_info *info) 1588 { 1589 struct nlattr *attr = info->attrs[MPTCP_PM_ENDPOINT_ADDR]; 1590 struct pm_nl_pernet *pernet = genl_info_pm_nl(info); 1591 struct mptcp_pm_addr_entry addr, *entry; 1592 unsigned int addr_max; 1593 int ret; 1594 1595 ret = mptcp_pm_parse_entry(attr, info, false, &addr); 1596 if (ret < 0) 1597 return ret; 1598 1599 /* the zero id address is special: the first address used by the msk 1600 * always gets such an id, so different subflows can have different zero 1601 * id addresses. Additionally zero id is not accounted for in id_bitmap. 1602 * Let's use an 'mptcp_rm_list' instead of the common remove code. 1603 */ 1604 if (addr.addr.id == 0) 1605 return mptcp_nl_remove_id_zero_address(sock_net(skb->sk), &addr.addr); 1606 1607 spin_lock_bh(&pernet->lock); 1608 entry = __lookup_addr_by_id(pernet, addr.addr.id); 1609 if (!entry) { 1610 GENL_SET_ERR_MSG(info, "address not found"); 1611 spin_unlock_bh(&pernet->lock); 1612 return -EINVAL; 1613 } 1614 if (entry->flags & MPTCP_PM_ADDR_FLAG_SIGNAL) { 1615 addr_max = pernet->add_addr_signal_max; 1616 WRITE_ONCE(pernet->add_addr_signal_max, addr_max - 1); 1617 } 1618 if (entry->flags & MPTCP_PM_ADDR_FLAG_SUBFLOW) { 1619 addr_max = pernet->local_addr_max; 1620 WRITE_ONCE(pernet->local_addr_max, addr_max - 1); 1621 } 1622 1623 pernet->addrs--; 1624 list_del_rcu(&entry->list); 1625 __clear_bit(entry->addr.id, pernet->id_bitmap); 1626 spin_unlock_bh(&pernet->lock); 1627 1628 mptcp_nl_remove_subflow_and_signal_addr(sock_net(skb->sk), entry); 1629 synchronize_rcu(); 1630 __mptcp_pm_release_addr_entry(entry); 1631 1632 return ret; 1633 } 1634 1635 /* Called from the userspace PM only */ 1636 void mptcp_pm_remove_addrs(struct mptcp_sock *msk, struct list_head *rm_list) 1637 { 1638 struct mptcp_rm_list alist = { .nr = 0 }; 1639 struct mptcp_pm_addr_entry *entry; 1640 int anno_nr = 0; 1641 1642 list_for_each_entry(entry, rm_list, list) { 1643 if (alist.nr >= MPTCP_RM_IDS_MAX) 1644 break; 1645 1646 /* only delete if either announced or matching a subflow */ 1647 if (remove_anno_list_by_saddr(msk, &entry->addr)) 1648 anno_nr++; 1649 else if (!lookup_subflow_by_saddr(&msk->conn_list, 1650 &entry->addr)) 1651 continue; 1652 1653 alist.ids[alist.nr++] = entry->addr.id; 1654 } 1655 1656 if (alist.nr) { 1657 spin_lock_bh(&msk->pm.lock); 1658 msk->pm.add_addr_signaled -= anno_nr; 1659 mptcp_pm_remove_addr(msk, &alist); 1660 spin_unlock_bh(&msk->pm.lock); 1661 } 1662 } 1663 1664 /* Called from the in-kernel PM only */ 1665 static void mptcp_pm_flush_addrs_and_subflows(struct mptcp_sock *msk, 1666 struct list_head *rm_list) 1667 { 1668 struct mptcp_rm_list alist = { .nr = 0 }, slist = { .nr = 0 }; 1669 struct mptcp_pm_addr_entry *entry; 1670 1671 list_for_each_entry(entry, rm_list, list) { 1672 if (slist.nr < MPTCP_RM_IDS_MAX && 1673 lookup_subflow_by_saddr(&msk->conn_list, &entry->addr)) 1674 slist.ids[slist.nr++] = mptcp_endp_get_local_id(msk, &entry->addr); 1675 1676 if (alist.nr < MPTCP_RM_IDS_MAX && 1677 remove_anno_list_by_saddr(msk, &entry->addr)) 1678 alist.ids[alist.nr++] = mptcp_endp_get_local_id(msk, &entry->addr); 1679 } 1680 1681 spin_lock_bh(&msk->pm.lock); 1682 if (alist.nr) { 1683 msk->pm.add_addr_signaled -= alist.nr; 1684 mptcp_pm_remove_addr(msk, &alist); 1685 } 1686 if (slist.nr) 1687 mptcp_pm_nl_rm_subflow_received(msk, &slist); 1688 /* Reset counters: maybe some subflows have been removed before */ 1689 bitmap_fill(msk->pm.id_avail_bitmap, MPTCP_PM_MAX_ADDR_ID + 1); 1690 msk->pm.local_addr_used = 0; 1691 spin_unlock_bh(&msk->pm.lock); 1692 } 1693 1694 static void mptcp_nl_flush_addrs_list(struct net *net, 1695 struct list_head *rm_list) 1696 { 1697 long s_slot = 0, s_num = 0; 1698 struct mptcp_sock *msk; 1699 1700 if (list_empty(rm_list)) 1701 return; 1702 1703 while ((msk = mptcp_token_iter_next(net, &s_slot, &s_num)) != NULL) { 1704 struct sock *sk = (struct sock *)msk; 1705 1706 if (!mptcp_pm_is_userspace(msk)) { 1707 lock_sock(sk); 1708 mptcp_pm_flush_addrs_and_subflows(msk, rm_list); 1709 release_sock(sk); 1710 } 1711 1712 sock_put(sk); 1713 cond_resched(); 1714 } 1715 } 1716 1717 /* caller must ensure the RCU grace period is already elapsed */ 1718 static void __flush_addrs(struct list_head *list) 1719 { 1720 while (!list_empty(list)) { 1721 struct mptcp_pm_addr_entry *cur; 1722 1723 cur = list_entry(list->next, 1724 struct mptcp_pm_addr_entry, list); 1725 list_del_rcu(&cur->list); 1726 __mptcp_pm_release_addr_entry(cur); 1727 } 1728 } 1729 1730 static void __reset_counters(struct pm_nl_pernet *pernet) 1731 { 1732 WRITE_ONCE(pernet->add_addr_signal_max, 0); 1733 WRITE_ONCE(pernet->add_addr_accept_max, 0); 1734 WRITE_ONCE(pernet->local_addr_max, 0); 1735 pernet->addrs = 0; 1736 } 1737 1738 int mptcp_pm_nl_flush_addrs_doit(struct sk_buff *skb, struct genl_info *info) 1739 { 1740 struct pm_nl_pernet *pernet = genl_info_pm_nl(info); 1741 LIST_HEAD(free_list); 1742 1743 spin_lock_bh(&pernet->lock); 1744 list_splice_init(&pernet->local_addr_list, &free_list); 1745 __reset_counters(pernet); 1746 pernet->next_id = 1; 1747 bitmap_zero(pernet->id_bitmap, MPTCP_PM_MAX_ADDR_ID + 1); 1748 spin_unlock_bh(&pernet->lock); 1749 mptcp_nl_flush_addrs_list(sock_net(skb->sk), &free_list); 1750 synchronize_rcu(); 1751 __flush_addrs(&free_list); 1752 return 0; 1753 } 1754 1755 int mptcp_nl_fill_addr(struct sk_buff *skb, 1756 struct mptcp_pm_addr_entry *entry) 1757 { 1758 struct mptcp_addr_info *addr = &entry->addr; 1759 struct nlattr *attr; 1760 1761 attr = nla_nest_start(skb, MPTCP_PM_ATTR_ADDR); 1762 if (!attr) 1763 return -EMSGSIZE; 1764 1765 if (nla_put_u16(skb, MPTCP_PM_ADDR_ATTR_FAMILY, addr->family)) 1766 goto nla_put_failure; 1767 if (nla_put_u16(skb, MPTCP_PM_ADDR_ATTR_PORT, ntohs(addr->port))) 1768 goto nla_put_failure; 1769 if (nla_put_u8(skb, MPTCP_PM_ADDR_ATTR_ID, addr->id)) 1770 goto nla_put_failure; 1771 if (nla_put_u32(skb, MPTCP_PM_ADDR_ATTR_FLAGS, entry->flags)) 1772 goto nla_put_failure; 1773 if (entry->ifindex && 1774 nla_put_s32(skb, MPTCP_PM_ADDR_ATTR_IF_IDX, entry->ifindex)) 1775 goto nla_put_failure; 1776 1777 if (addr->family == AF_INET && 1778 nla_put_in_addr(skb, MPTCP_PM_ADDR_ATTR_ADDR4, 1779 addr->addr.s_addr)) 1780 goto nla_put_failure; 1781 #if IS_ENABLED(CONFIG_MPTCP_IPV6) 1782 else if (addr->family == AF_INET6 && 1783 nla_put_in6_addr(skb, MPTCP_PM_ADDR_ATTR_ADDR6, &addr->addr6)) 1784 goto nla_put_failure; 1785 #endif 1786 nla_nest_end(skb, attr); 1787 return 0; 1788 1789 nla_put_failure: 1790 nla_nest_cancel(skb, attr); 1791 return -EMSGSIZE; 1792 } 1793 1794 int mptcp_pm_nl_get_addr(struct sk_buff *skb, struct genl_info *info) 1795 { 1796 struct nlattr *attr = info->attrs[MPTCP_PM_ENDPOINT_ADDR]; 1797 struct pm_nl_pernet *pernet = genl_info_pm_nl(info); 1798 struct mptcp_pm_addr_entry addr, *entry; 1799 struct sk_buff *msg; 1800 void *reply; 1801 int ret; 1802 1803 ret = mptcp_pm_parse_entry(attr, info, false, &addr); 1804 if (ret < 0) 1805 return ret; 1806 1807 msg = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL); 1808 if (!msg) 1809 return -ENOMEM; 1810 1811 reply = genlmsg_put_reply(msg, info, &mptcp_genl_family, 0, 1812 info->genlhdr->cmd); 1813 if (!reply) { 1814 GENL_SET_ERR_MSG(info, "not enough space in Netlink message"); 1815 ret = -EMSGSIZE; 1816 goto fail; 1817 } 1818 1819 spin_lock_bh(&pernet->lock); 1820 entry = __lookup_addr_by_id(pernet, addr.addr.id); 1821 if (!entry) { 1822 GENL_SET_ERR_MSG(info, "address not found"); 1823 ret = -EINVAL; 1824 goto unlock_fail; 1825 } 1826 1827 ret = mptcp_nl_fill_addr(msg, entry); 1828 if (ret) 1829 goto unlock_fail; 1830 1831 genlmsg_end(msg, reply); 1832 ret = genlmsg_reply(msg, info); 1833 spin_unlock_bh(&pernet->lock); 1834 return ret; 1835 1836 unlock_fail: 1837 spin_unlock_bh(&pernet->lock); 1838 1839 fail: 1840 nlmsg_free(msg); 1841 return ret; 1842 } 1843 1844 int mptcp_pm_nl_get_addr_doit(struct sk_buff *skb, struct genl_info *info) 1845 { 1846 return mptcp_pm_get_addr(skb, info); 1847 } 1848 1849 int mptcp_pm_nl_dump_addr(struct sk_buff *msg, 1850 struct netlink_callback *cb) 1851 { 1852 struct net *net = sock_net(msg->sk); 1853 struct mptcp_pm_addr_entry *entry; 1854 struct pm_nl_pernet *pernet; 1855 int id = cb->args[0]; 1856 void *hdr; 1857 int i; 1858 1859 pernet = pm_nl_get_pernet(net); 1860 1861 spin_lock_bh(&pernet->lock); 1862 for (i = id; i < MPTCP_PM_MAX_ADDR_ID + 1; i++) { 1863 if (test_bit(i, pernet->id_bitmap)) { 1864 entry = __lookup_addr_by_id(pernet, i); 1865 if (!entry) 1866 break; 1867 1868 if (entry->addr.id <= id) 1869 continue; 1870 1871 hdr = genlmsg_put(msg, NETLINK_CB(cb->skb).portid, 1872 cb->nlh->nlmsg_seq, &mptcp_genl_family, 1873 NLM_F_MULTI, MPTCP_PM_CMD_GET_ADDR); 1874 if (!hdr) 1875 break; 1876 1877 if (mptcp_nl_fill_addr(msg, entry) < 0) { 1878 genlmsg_cancel(msg, hdr); 1879 break; 1880 } 1881 1882 id = entry->addr.id; 1883 genlmsg_end(msg, hdr); 1884 } 1885 } 1886 spin_unlock_bh(&pernet->lock); 1887 1888 cb->args[0] = id; 1889 return msg->len; 1890 } 1891 1892 int mptcp_pm_nl_get_addr_dumpit(struct sk_buff *msg, 1893 struct netlink_callback *cb) 1894 { 1895 return mptcp_pm_dump_addr(msg, cb); 1896 } 1897 1898 static int parse_limit(struct genl_info *info, int id, unsigned int *limit) 1899 { 1900 struct nlattr *attr = info->attrs[id]; 1901 1902 if (!attr) 1903 return 0; 1904 1905 *limit = nla_get_u32(attr); 1906 if (*limit > MPTCP_PM_ADDR_MAX) { 1907 GENL_SET_ERR_MSG(info, "limit greater than maximum"); 1908 return -EINVAL; 1909 } 1910 return 0; 1911 } 1912 1913 int mptcp_pm_nl_set_limits_doit(struct sk_buff *skb, struct genl_info *info) 1914 { 1915 struct pm_nl_pernet *pernet = genl_info_pm_nl(info); 1916 unsigned int rcv_addrs, subflows; 1917 int ret; 1918 1919 spin_lock_bh(&pernet->lock); 1920 rcv_addrs = pernet->add_addr_accept_max; 1921 ret = parse_limit(info, MPTCP_PM_ATTR_RCV_ADD_ADDRS, &rcv_addrs); 1922 if (ret) 1923 goto unlock; 1924 1925 subflows = pernet->subflows_max; 1926 ret = parse_limit(info, MPTCP_PM_ATTR_SUBFLOWS, &subflows); 1927 if (ret) 1928 goto unlock; 1929 1930 WRITE_ONCE(pernet->add_addr_accept_max, rcv_addrs); 1931 WRITE_ONCE(pernet->subflows_max, subflows); 1932 1933 unlock: 1934 spin_unlock_bh(&pernet->lock); 1935 return ret; 1936 } 1937 1938 int mptcp_pm_nl_get_limits_doit(struct sk_buff *skb, struct genl_info *info) 1939 { 1940 struct pm_nl_pernet *pernet = genl_info_pm_nl(info); 1941 struct sk_buff *msg; 1942 void *reply; 1943 1944 msg = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL); 1945 if (!msg) 1946 return -ENOMEM; 1947 1948 reply = genlmsg_put_reply(msg, info, &mptcp_genl_family, 0, 1949 MPTCP_PM_CMD_GET_LIMITS); 1950 if (!reply) 1951 goto fail; 1952 1953 if (nla_put_u32(msg, MPTCP_PM_ATTR_RCV_ADD_ADDRS, 1954 READ_ONCE(pernet->add_addr_accept_max))) 1955 goto fail; 1956 1957 if (nla_put_u32(msg, MPTCP_PM_ATTR_SUBFLOWS, 1958 READ_ONCE(pernet->subflows_max))) 1959 goto fail; 1960 1961 genlmsg_end(msg, reply); 1962 return genlmsg_reply(msg, info); 1963 1964 fail: 1965 GENL_SET_ERR_MSG(info, "not enough space in Netlink message"); 1966 nlmsg_free(msg); 1967 return -EMSGSIZE; 1968 } 1969 1970 static void mptcp_pm_nl_fullmesh(struct mptcp_sock *msk, 1971 struct mptcp_addr_info *addr) 1972 { 1973 struct mptcp_rm_list list = { .nr = 0 }; 1974 1975 list.ids[list.nr++] = mptcp_endp_get_local_id(msk, addr); 1976 1977 spin_lock_bh(&msk->pm.lock); 1978 mptcp_pm_nl_rm_subflow_received(msk, &list); 1979 __mark_subflow_endp_available(msk, list.ids[0]); 1980 mptcp_pm_create_subflow_or_signal_addr(msk); 1981 spin_unlock_bh(&msk->pm.lock); 1982 } 1983 1984 static int mptcp_nl_set_flags(struct net *net, 1985 struct mptcp_addr_info *addr, 1986 u8 bkup, u8 changed) 1987 { 1988 long s_slot = 0, s_num = 0; 1989 struct mptcp_sock *msk; 1990 int ret = -EINVAL; 1991 1992 while ((msk = mptcp_token_iter_next(net, &s_slot, &s_num)) != NULL) { 1993 struct sock *sk = (struct sock *)msk; 1994 1995 if (list_empty(&msk->conn_list) || mptcp_pm_is_userspace(msk)) 1996 goto next; 1997 1998 lock_sock(sk); 1999 if (changed & MPTCP_PM_ADDR_FLAG_BACKUP) 2000 ret = mptcp_pm_nl_mp_prio_send_ack(msk, addr, NULL, bkup); 2001 if (changed & MPTCP_PM_ADDR_FLAG_FULLMESH) 2002 mptcp_pm_nl_fullmesh(msk, addr); 2003 release_sock(sk); 2004 2005 next: 2006 sock_put(sk); 2007 cond_resched(); 2008 } 2009 2010 return ret; 2011 } 2012 2013 int mptcp_pm_nl_set_flags(struct sk_buff *skb, struct genl_info *info) 2014 { 2015 struct mptcp_pm_addr_entry addr = { .addr = { .family = AF_UNSPEC }, }; 2016 struct nlattr *attr = info->attrs[MPTCP_PM_ATTR_ADDR]; 2017 u8 changed, mask = MPTCP_PM_ADDR_FLAG_BACKUP | 2018 MPTCP_PM_ADDR_FLAG_FULLMESH; 2019 struct net *net = sock_net(skb->sk); 2020 struct mptcp_pm_addr_entry *entry; 2021 struct pm_nl_pernet *pernet; 2022 u8 lookup_by_id = 0; 2023 u8 bkup = 0; 2024 int ret; 2025 2026 pernet = pm_nl_get_pernet(net); 2027 2028 ret = mptcp_pm_parse_entry(attr, info, false, &addr); 2029 if (ret < 0) 2030 return ret; 2031 2032 if (addr.addr.family == AF_UNSPEC) { 2033 lookup_by_id = 1; 2034 if (!addr.addr.id) { 2035 GENL_SET_ERR_MSG(info, "missing required inputs"); 2036 return -EOPNOTSUPP; 2037 } 2038 } 2039 2040 if (addr.flags & MPTCP_PM_ADDR_FLAG_BACKUP) 2041 bkup = 1; 2042 2043 spin_lock_bh(&pernet->lock); 2044 entry = lookup_by_id ? __lookup_addr_by_id(pernet, addr.addr.id) : 2045 __lookup_addr(pernet, &addr.addr); 2046 if (!entry) { 2047 spin_unlock_bh(&pernet->lock); 2048 GENL_SET_ERR_MSG(info, "address not found"); 2049 return -EINVAL; 2050 } 2051 if ((addr.flags & MPTCP_PM_ADDR_FLAG_FULLMESH) && 2052 (entry->flags & MPTCP_PM_ADDR_FLAG_SIGNAL)) { 2053 spin_unlock_bh(&pernet->lock); 2054 GENL_SET_ERR_MSG(info, "invalid addr flags"); 2055 return -EINVAL; 2056 } 2057 2058 changed = (addr.flags ^ entry->flags) & mask; 2059 entry->flags = (entry->flags & ~mask) | (addr.flags & mask); 2060 addr = *entry; 2061 spin_unlock_bh(&pernet->lock); 2062 2063 mptcp_nl_set_flags(net, &addr.addr, bkup, changed); 2064 return 0; 2065 } 2066 2067 int mptcp_pm_nl_set_flags_doit(struct sk_buff *skb, struct genl_info *info) 2068 { 2069 return mptcp_pm_set_flags(skb, info); 2070 } 2071 2072 static void mptcp_nl_mcast_send(struct net *net, struct sk_buff *nlskb, gfp_t gfp) 2073 { 2074 genlmsg_multicast_netns(&mptcp_genl_family, net, 2075 nlskb, 0, MPTCP_PM_EV_GRP_OFFSET, gfp); 2076 } 2077 2078 bool mptcp_userspace_pm_active(const struct mptcp_sock *msk) 2079 { 2080 return genl_has_listeners(&mptcp_genl_family, 2081 sock_net((const struct sock *)msk), 2082 MPTCP_PM_EV_GRP_OFFSET); 2083 } 2084 2085 static int mptcp_event_add_subflow(struct sk_buff *skb, const struct sock *ssk) 2086 { 2087 const struct inet_sock *issk = inet_sk(ssk); 2088 const struct mptcp_subflow_context *sf; 2089 2090 if (nla_put_u16(skb, MPTCP_ATTR_FAMILY, ssk->sk_family)) 2091 return -EMSGSIZE; 2092 2093 switch (ssk->sk_family) { 2094 case AF_INET: 2095 if (nla_put_in_addr(skb, MPTCP_ATTR_SADDR4, issk->inet_saddr)) 2096 return -EMSGSIZE; 2097 if (nla_put_in_addr(skb, MPTCP_ATTR_DADDR4, issk->inet_daddr)) 2098 return -EMSGSIZE; 2099 break; 2100 #if IS_ENABLED(CONFIG_MPTCP_IPV6) 2101 case AF_INET6: { 2102 const struct ipv6_pinfo *np = inet6_sk(ssk); 2103 2104 if (nla_put_in6_addr(skb, MPTCP_ATTR_SADDR6, &np->saddr)) 2105 return -EMSGSIZE; 2106 if (nla_put_in6_addr(skb, MPTCP_ATTR_DADDR6, &ssk->sk_v6_daddr)) 2107 return -EMSGSIZE; 2108 break; 2109 } 2110 #endif 2111 default: 2112 WARN_ON_ONCE(1); 2113 return -EMSGSIZE; 2114 } 2115 2116 if (nla_put_be16(skb, MPTCP_ATTR_SPORT, issk->inet_sport)) 2117 return -EMSGSIZE; 2118 if (nla_put_be16(skb, MPTCP_ATTR_DPORT, issk->inet_dport)) 2119 return -EMSGSIZE; 2120 2121 sf = mptcp_subflow_ctx(ssk); 2122 if (WARN_ON_ONCE(!sf)) 2123 return -EINVAL; 2124 2125 if (nla_put_u8(skb, MPTCP_ATTR_LOC_ID, subflow_get_local_id(sf))) 2126 return -EMSGSIZE; 2127 2128 if (nla_put_u8(skb, MPTCP_ATTR_REM_ID, sf->remote_id)) 2129 return -EMSGSIZE; 2130 2131 return 0; 2132 } 2133 2134 static int mptcp_event_put_token_and_ssk(struct sk_buff *skb, 2135 const struct mptcp_sock *msk, 2136 const struct sock *ssk) 2137 { 2138 const struct sock *sk = (const struct sock *)msk; 2139 const struct mptcp_subflow_context *sf; 2140 u8 sk_err; 2141 2142 if (nla_put_u32(skb, MPTCP_ATTR_TOKEN, READ_ONCE(msk->token))) 2143 return -EMSGSIZE; 2144 2145 if (mptcp_event_add_subflow(skb, ssk)) 2146 return -EMSGSIZE; 2147 2148 sf = mptcp_subflow_ctx(ssk); 2149 if (WARN_ON_ONCE(!sf)) 2150 return -EINVAL; 2151 2152 if (nla_put_u8(skb, MPTCP_ATTR_BACKUP, sf->backup)) 2153 return -EMSGSIZE; 2154 2155 if (ssk->sk_bound_dev_if && 2156 nla_put_s32(skb, MPTCP_ATTR_IF_IDX, ssk->sk_bound_dev_if)) 2157 return -EMSGSIZE; 2158 2159 sk_err = READ_ONCE(ssk->sk_err); 2160 if (sk_err && sk->sk_state == TCP_ESTABLISHED && 2161 nla_put_u8(skb, MPTCP_ATTR_ERROR, sk_err)) 2162 return -EMSGSIZE; 2163 2164 return 0; 2165 } 2166 2167 static int mptcp_event_sub_established(struct sk_buff *skb, 2168 const struct mptcp_sock *msk, 2169 const struct sock *ssk) 2170 { 2171 return mptcp_event_put_token_and_ssk(skb, msk, ssk); 2172 } 2173 2174 static int mptcp_event_sub_closed(struct sk_buff *skb, 2175 const struct mptcp_sock *msk, 2176 const struct sock *ssk) 2177 { 2178 const struct mptcp_subflow_context *sf; 2179 2180 if (mptcp_event_put_token_and_ssk(skb, msk, ssk)) 2181 return -EMSGSIZE; 2182 2183 sf = mptcp_subflow_ctx(ssk); 2184 if (!sf->reset_seen) 2185 return 0; 2186 2187 if (nla_put_u32(skb, MPTCP_ATTR_RESET_REASON, sf->reset_reason)) 2188 return -EMSGSIZE; 2189 2190 if (nla_put_u32(skb, MPTCP_ATTR_RESET_FLAGS, sf->reset_transient)) 2191 return -EMSGSIZE; 2192 2193 return 0; 2194 } 2195 2196 static int mptcp_event_created(struct sk_buff *skb, 2197 const struct mptcp_sock *msk, 2198 const struct sock *ssk) 2199 { 2200 int err = nla_put_u32(skb, MPTCP_ATTR_TOKEN, READ_ONCE(msk->token)); 2201 2202 if (err) 2203 return err; 2204 2205 if (nla_put_u8(skb, MPTCP_ATTR_SERVER_SIDE, READ_ONCE(msk->pm.server_side))) 2206 return -EMSGSIZE; 2207 2208 return mptcp_event_add_subflow(skb, ssk); 2209 } 2210 2211 void mptcp_event_addr_removed(const struct mptcp_sock *msk, uint8_t id) 2212 { 2213 struct net *net = sock_net((const struct sock *)msk); 2214 struct nlmsghdr *nlh; 2215 struct sk_buff *skb; 2216 2217 if (!genl_has_listeners(&mptcp_genl_family, net, MPTCP_PM_EV_GRP_OFFSET)) 2218 return; 2219 2220 skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_ATOMIC); 2221 if (!skb) 2222 return; 2223 2224 nlh = genlmsg_put(skb, 0, 0, &mptcp_genl_family, 0, MPTCP_EVENT_REMOVED); 2225 if (!nlh) 2226 goto nla_put_failure; 2227 2228 if (nla_put_u32(skb, MPTCP_ATTR_TOKEN, READ_ONCE(msk->token))) 2229 goto nla_put_failure; 2230 2231 if (nla_put_u8(skb, MPTCP_ATTR_REM_ID, id)) 2232 goto nla_put_failure; 2233 2234 genlmsg_end(skb, nlh); 2235 mptcp_nl_mcast_send(net, skb, GFP_ATOMIC); 2236 return; 2237 2238 nla_put_failure: 2239 nlmsg_free(skb); 2240 } 2241 2242 void mptcp_event_addr_announced(const struct sock *ssk, 2243 const struct mptcp_addr_info *info) 2244 { 2245 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk); 2246 struct mptcp_sock *msk = mptcp_sk(subflow->conn); 2247 struct net *net = sock_net(ssk); 2248 struct nlmsghdr *nlh; 2249 struct sk_buff *skb; 2250 2251 if (!genl_has_listeners(&mptcp_genl_family, net, MPTCP_PM_EV_GRP_OFFSET)) 2252 return; 2253 2254 skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_ATOMIC); 2255 if (!skb) 2256 return; 2257 2258 nlh = genlmsg_put(skb, 0, 0, &mptcp_genl_family, 0, 2259 MPTCP_EVENT_ANNOUNCED); 2260 if (!nlh) 2261 goto nla_put_failure; 2262 2263 if (nla_put_u32(skb, MPTCP_ATTR_TOKEN, READ_ONCE(msk->token))) 2264 goto nla_put_failure; 2265 2266 if (nla_put_u8(skb, MPTCP_ATTR_REM_ID, info->id)) 2267 goto nla_put_failure; 2268 2269 if (nla_put_be16(skb, MPTCP_ATTR_DPORT, 2270 info->port == 0 ? 2271 inet_sk(ssk)->inet_dport : 2272 info->port)) 2273 goto nla_put_failure; 2274 2275 switch (info->family) { 2276 case AF_INET: 2277 if (nla_put_in_addr(skb, MPTCP_ATTR_DADDR4, info->addr.s_addr)) 2278 goto nla_put_failure; 2279 break; 2280 #if IS_ENABLED(CONFIG_MPTCP_IPV6) 2281 case AF_INET6: 2282 if (nla_put_in6_addr(skb, MPTCP_ATTR_DADDR6, &info->addr6)) 2283 goto nla_put_failure; 2284 break; 2285 #endif 2286 default: 2287 WARN_ON_ONCE(1); 2288 goto nla_put_failure; 2289 } 2290 2291 genlmsg_end(skb, nlh); 2292 mptcp_nl_mcast_send(net, skb, GFP_ATOMIC); 2293 return; 2294 2295 nla_put_failure: 2296 nlmsg_free(skb); 2297 } 2298 2299 void mptcp_event_pm_listener(const struct sock *ssk, 2300 enum mptcp_event_type event) 2301 { 2302 const struct inet_sock *issk = inet_sk(ssk); 2303 struct net *net = sock_net(ssk); 2304 struct nlmsghdr *nlh; 2305 struct sk_buff *skb; 2306 2307 if (!genl_has_listeners(&mptcp_genl_family, net, MPTCP_PM_EV_GRP_OFFSET)) 2308 return; 2309 2310 skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL); 2311 if (!skb) 2312 return; 2313 2314 nlh = genlmsg_put(skb, 0, 0, &mptcp_genl_family, 0, event); 2315 if (!nlh) 2316 goto nla_put_failure; 2317 2318 if (nla_put_u16(skb, MPTCP_ATTR_FAMILY, ssk->sk_family)) 2319 goto nla_put_failure; 2320 2321 if (nla_put_be16(skb, MPTCP_ATTR_SPORT, issk->inet_sport)) 2322 goto nla_put_failure; 2323 2324 switch (ssk->sk_family) { 2325 case AF_INET: 2326 if (nla_put_in_addr(skb, MPTCP_ATTR_SADDR4, issk->inet_saddr)) 2327 goto nla_put_failure; 2328 break; 2329 #if IS_ENABLED(CONFIG_MPTCP_IPV6) 2330 case AF_INET6: { 2331 const struct ipv6_pinfo *np = inet6_sk(ssk); 2332 2333 if (nla_put_in6_addr(skb, MPTCP_ATTR_SADDR6, &np->saddr)) 2334 goto nla_put_failure; 2335 break; 2336 } 2337 #endif 2338 default: 2339 WARN_ON_ONCE(1); 2340 goto nla_put_failure; 2341 } 2342 2343 genlmsg_end(skb, nlh); 2344 mptcp_nl_mcast_send(net, skb, GFP_KERNEL); 2345 return; 2346 2347 nla_put_failure: 2348 nlmsg_free(skb); 2349 } 2350 2351 void mptcp_event(enum mptcp_event_type type, const struct mptcp_sock *msk, 2352 const struct sock *ssk, gfp_t gfp) 2353 { 2354 struct net *net = sock_net((const struct sock *)msk); 2355 struct nlmsghdr *nlh; 2356 struct sk_buff *skb; 2357 2358 if (!genl_has_listeners(&mptcp_genl_family, net, MPTCP_PM_EV_GRP_OFFSET)) 2359 return; 2360 2361 skb = nlmsg_new(NLMSG_DEFAULT_SIZE, gfp); 2362 if (!skb) 2363 return; 2364 2365 nlh = genlmsg_put(skb, 0, 0, &mptcp_genl_family, 0, type); 2366 if (!nlh) 2367 goto nla_put_failure; 2368 2369 switch (type) { 2370 case MPTCP_EVENT_UNSPEC: 2371 WARN_ON_ONCE(1); 2372 break; 2373 case MPTCP_EVENT_CREATED: 2374 case MPTCP_EVENT_ESTABLISHED: 2375 if (mptcp_event_created(skb, msk, ssk) < 0) 2376 goto nla_put_failure; 2377 break; 2378 case MPTCP_EVENT_CLOSED: 2379 if (nla_put_u32(skb, MPTCP_ATTR_TOKEN, READ_ONCE(msk->token)) < 0) 2380 goto nla_put_failure; 2381 break; 2382 case MPTCP_EVENT_ANNOUNCED: 2383 case MPTCP_EVENT_REMOVED: 2384 /* call mptcp_event_addr_announced()/removed instead */ 2385 WARN_ON_ONCE(1); 2386 break; 2387 case MPTCP_EVENT_SUB_ESTABLISHED: 2388 case MPTCP_EVENT_SUB_PRIORITY: 2389 if (mptcp_event_sub_established(skb, msk, ssk) < 0) 2390 goto nla_put_failure; 2391 break; 2392 case MPTCP_EVENT_SUB_CLOSED: 2393 if (mptcp_event_sub_closed(skb, msk, ssk) < 0) 2394 goto nla_put_failure; 2395 break; 2396 case MPTCP_EVENT_LISTENER_CREATED: 2397 case MPTCP_EVENT_LISTENER_CLOSED: 2398 break; 2399 } 2400 2401 genlmsg_end(skb, nlh); 2402 mptcp_nl_mcast_send(net, skb, gfp); 2403 return; 2404 2405 nla_put_failure: 2406 nlmsg_free(skb); 2407 } 2408 2409 struct genl_family mptcp_genl_family __ro_after_init = { 2410 .name = MPTCP_PM_NAME, 2411 .version = MPTCP_PM_VER, 2412 .netnsok = true, 2413 .module = THIS_MODULE, 2414 .ops = mptcp_pm_nl_ops, 2415 .n_ops = ARRAY_SIZE(mptcp_pm_nl_ops), 2416 .resv_start_op = MPTCP_PM_CMD_SUBFLOW_DESTROY + 1, 2417 .mcgrps = mptcp_pm_mcgrps, 2418 .n_mcgrps = ARRAY_SIZE(mptcp_pm_mcgrps), 2419 }; 2420 2421 static int __net_init pm_nl_init_net(struct net *net) 2422 { 2423 struct pm_nl_pernet *pernet = pm_nl_get_pernet(net); 2424 2425 INIT_LIST_HEAD_RCU(&pernet->local_addr_list); 2426 2427 /* Cit. 2 subflows ought to be enough for anybody. */ 2428 pernet->subflows_max = 2; 2429 pernet->next_id = 1; 2430 pernet->stale_loss_cnt = 4; 2431 spin_lock_init(&pernet->lock); 2432 2433 /* No need to initialize other pernet fields, the struct is zeroed at 2434 * allocation time. 2435 */ 2436 2437 return 0; 2438 } 2439 2440 static void __net_exit pm_nl_exit_net(struct list_head *net_list) 2441 { 2442 struct net *net; 2443 2444 list_for_each_entry(net, net_list, exit_list) { 2445 struct pm_nl_pernet *pernet = pm_nl_get_pernet(net); 2446 2447 /* net is removed from namespace list, can't race with 2448 * other modifiers, also netns core already waited for a 2449 * RCU grace period. 2450 */ 2451 __flush_addrs(&pernet->local_addr_list); 2452 } 2453 } 2454 2455 static struct pernet_operations mptcp_pm_pernet_ops = { 2456 .init = pm_nl_init_net, 2457 .exit_batch = pm_nl_exit_net, 2458 .id = &pm_nl_pernet_id, 2459 .size = sizeof(struct pm_nl_pernet), 2460 }; 2461 2462 void __init mptcp_pm_nl_init(void) 2463 { 2464 if (register_pernet_subsys(&mptcp_pm_pernet_ops) < 0) 2465 panic("Failed to register MPTCP PM pernet subsystem.\n"); 2466 2467 if (genl_register_family(&mptcp_genl_family)) 2468 panic("Failed to register MPTCP PM netlink family\n"); 2469 } 2470