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/tcp.h> 12 #include <net/netns/generic.h> 13 #include <net/mptcp.h> 14 #include <net/genetlink.h> 15 #include <uapi/linux/mptcp.h> 16 17 #include "protocol.h" 18 #include "mib.h" 19 20 /* forward declaration */ 21 static struct genl_family mptcp_genl_family; 22 23 static int pm_nl_pernet_id; 24 25 struct mptcp_pm_add_entry { 26 struct list_head list; 27 struct mptcp_addr_info addr; 28 struct timer_list add_timer; 29 struct mptcp_sock *sock; 30 u8 retrans_times; 31 }; 32 33 struct pm_nl_pernet { 34 /* protects pernet updates */ 35 spinlock_t lock; 36 struct list_head local_addr_list; 37 unsigned int addrs; 38 unsigned int stale_loss_cnt; 39 unsigned int add_addr_signal_max; 40 unsigned int add_addr_accept_max; 41 unsigned int local_addr_max; 42 unsigned int subflows_max; 43 unsigned int next_id; 44 DECLARE_BITMAP(id_bitmap, MPTCP_PM_MAX_ADDR_ID + 1); 45 }; 46 47 #define MPTCP_PM_ADDR_MAX 8 48 #define ADD_ADDR_RETRANS_MAX 3 49 50 static struct pm_nl_pernet *pm_nl_get_pernet(const struct net *net) 51 { 52 return net_generic(net, pm_nl_pernet_id); 53 } 54 55 static struct pm_nl_pernet * 56 pm_nl_get_pernet_from_msk(const struct mptcp_sock *msk) 57 { 58 return pm_nl_get_pernet(sock_net((struct sock *)msk)); 59 } 60 61 bool mptcp_addresses_equal(const struct mptcp_addr_info *a, 62 const struct mptcp_addr_info *b, bool use_port) 63 { 64 bool addr_equals = false; 65 66 if (a->family == b->family) { 67 if (a->family == AF_INET) 68 addr_equals = a->addr.s_addr == b->addr.s_addr; 69 #if IS_ENABLED(CONFIG_MPTCP_IPV6) 70 else 71 addr_equals = !ipv6_addr_cmp(&a->addr6, &b->addr6); 72 } else if (a->family == AF_INET) { 73 if (ipv6_addr_v4mapped(&b->addr6)) 74 addr_equals = a->addr.s_addr == b->addr6.s6_addr32[3]; 75 } else if (b->family == AF_INET) { 76 if (ipv6_addr_v4mapped(&a->addr6)) 77 addr_equals = a->addr6.s6_addr32[3] == b->addr.s_addr; 78 #endif 79 } 80 81 if (!addr_equals) 82 return false; 83 if (!use_port) 84 return true; 85 86 return a->port == b->port; 87 } 88 89 static void local_address(const struct sock_common *skc, 90 struct mptcp_addr_info *addr) 91 { 92 addr->family = skc->skc_family; 93 addr->port = htons(skc->skc_num); 94 if (addr->family == AF_INET) 95 addr->addr.s_addr = skc->skc_rcv_saddr; 96 #if IS_ENABLED(CONFIG_MPTCP_IPV6) 97 else if (addr->family == AF_INET6) 98 addr->addr6 = skc->skc_v6_rcv_saddr; 99 #endif 100 } 101 102 static void remote_address(const struct sock_common *skc, 103 struct mptcp_addr_info *addr) 104 { 105 addr->family = skc->skc_family; 106 addr->port = skc->skc_dport; 107 if (addr->family == AF_INET) 108 addr->addr.s_addr = skc->skc_daddr; 109 #if IS_ENABLED(CONFIG_MPTCP_IPV6) 110 else if (addr->family == AF_INET6) 111 addr->addr6 = skc->skc_v6_daddr; 112 #endif 113 } 114 115 static bool lookup_subflow_by_saddr(const struct list_head *list, 116 const struct mptcp_addr_info *saddr) 117 { 118 struct mptcp_subflow_context *subflow; 119 struct mptcp_addr_info cur; 120 struct sock_common *skc; 121 122 list_for_each_entry(subflow, list, node) { 123 skc = (struct sock_common *)mptcp_subflow_tcp_sock(subflow); 124 125 local_address(skc, &cur); 126 if (mptcp_addresses_equal(&cur, saddr, saddr->port)) 127 return true; 128 } 129 130 return false; 131 } 132 133 static bool lookup_subflow_by_daddr(const struct list_head *list, 134 const struct mptcp_addr_info *daddr) 135 { 136 struct mptcp_subflow_context *subflow; 137 struct mptcp_addr_info cur; 138 struct sock_common *skc; 139 140 list_for_each_entry(subflow, list, node) { 141 skc = (struct sock_common *)mptcp_subflow_tcp_sock(subflow); 142 143 remote_address(skc, &cur); 144 if (mptcp_addresses_equal(&cur, daddr, daddr->port)) 145 return true; 146 } 147 148 return false; 149 } 150 151 static struct mptcp_pm_addr_entry * 152 select_local_address(const struct pm_nl_pernet *pernet, 153 const struct mptcp_sock *msk) 154 { 155 const struct sock *sk = (const struct sock *)msk; 156 struct mptcp_pm_addr_entry *entry, *ret = NULL; 157 158 msk_owned_by_me(msk); 159 160 rcu_read_lock(); 161 list_for_each_entry_rcu(entry, &pernet->local_addr_list, list) { 162 if (!(entry->flags & MPTCP_PM_ADDR_FLAG_SUBFLOW)) 163 continue; 164 165 if (!test_bit(entry->addr.id, msk->pm.id_avail_bitmap)) 166 continue; 167 168 if (entry->addr.family != sk->sk_family) { 169 #if IS_ENABLED(CONFIG_MPTCP_IPV6) 170 if ((entry->addr.family == AF_INET && 171 !ipv6_addr_v4mapped(&sk->sk_v6_daddr)) || 172 (sk->sk_family == AF_INET && 173 !ipv6_addr_v4mapped(&entry->addr.addr6))) 174 #endif 175 continue; 176 } 177 178 ret = entry; 179 break; 180 } 181 rcu_read_unlock(); 182 return ret; 183 } 184 185 static struct mptcp_pm_addr_entry * 186 select_signal_address(struct pm_nl_pernet *pernet, const struct mptcp_sock *msk) 187 { 188 struct mptcp_pm_addr_entry *entry, *ret = NULL; 189 190 rcu_read_lock(); 191 /* do not keep any additional per socket state, just signal 192 * the address list in order. 193 * Note: removal from the local address list during the msk life-cycle 194 * can lead to additional addresses not being announced. 195 */ 196 list_for_each_entry_rcu(entry, &pernet->local_addr_list, list) { 197 if (!test_bit(entry->addr.id, msk->pm.id_avail_bitmap)) 198 continue; 199 200 if (!(entry->flags & MPTCP_PM_ADDR_FLAG_SIGNAL)) 201 continue; 202 203 ret = entry; 204 break; 205 } 206 rcu_read_unlock(); 207 return ret; 208 } 209 210 unsigned int mptcp_pm_get_add_addr_signal_max(const struct mptcp_sock *msk) 211 { 212 const struct pm_nl_pernet *pernet = pm_nl_get_pernet_from_msk(msk); 213 214 return READ_ONCE(pernet->add_addr_signal_max); 215 } 216 EXPORT_SYMBOL_GPL(mptcp_pm_get_add_addr_signal_max); 217 218 unsigned int mptcp_pm_get_add_addr_accept_max(const struct mptcp_sock *msk) 219 { 220 struct pm_nl_pernet *pernet = pm_nl_get_pernet_from_msk(msk); 221 222 return READ_ONCE(pernet->add_addr_accept_max); 223 } 224 EXPORT_SYMBOL_GPL(mptcp_pm_get_add_addr_accept_max); 225 226 unsigned int mptcp_pm_get_subflows_max(const struct mptcp_sock *msk) 227 { 228 struct pm_nl_pernet *pernet = pm_nl_get_pernet_from_msk(msk); 229 230 return READ_ONCE(pernet->subflows_max); 231 } 232 EXPORT_SYMBOL_GPL(mptcp_pm_get_subflows_max); 233 234 unsigned int mptcp_pm_get_local_addr_max(const struct mptcp_sock *msk) 235 { 236 struct pm_nl_pernet *pernet = pm_nl_get_pernet_from_msk(msk); 237 238 return READ_ONCE(pernet->local_addr_max); 239 } 240 EXPORT_SYMBOL_GPL(mptcp_pm_get_local_addr_max); 241 242 bool mptcp_pm_nl_check_work_pending(struct mptcp_sock *msk) 243 { 244 struct pm_nl_pernet *pernet = pm_nl_get_pernet_from_msk(msk); 245 246 if (msk->pm.subflows == mptcp_pm_get_subflows_max(msk) || 247 (find_next_and_bit(pernet->id_bitmap, msk->pm.id_avail_bitmap, 248 MPTCP_PM_MAX_ADDR_ID + 1, 0) == MPTCP_PM_MAX_ADDR_ID + 1)) { 249 WRITE_ONCE(msk->pm.work_pending, false); 250 return false; 251 } 252 return true; 253 } 254 255 struct mptcp_pm_add_entry * 256 mptcp_lookup_anno_list_by_saddr(const struct mptcp_sock *msk, 257 const struct mptcp_addr_info *addr) 258 { 259 struct mptcp_pm_add_entry *entry; 260 261 lockdep_assert_held(&msk->pm.lock); 262 263 list_for_each_entry(entry, &msk->pm.anno_list, list) { 264 if (mptcp_addresses_equal(&entry->addr, addr, true)) 265 return entry; 266 } 267 268 return NULL; 269 } 270 271 bool mptcp_pm_sport_in_anno_list(struct mptcp_sock *msk, const struct sock *sk) 272 { 273 struct mptcp_pm_add_entry *entry; 274 struct mptcp_addr_info saddr; 275 bool ret = false; 276 277 local_address((struct sock_common *)sk, &saddr); 278 279 spin_lock_bh(&msk->pm.lock); 280 list_for_each_entry(entry, &msk->pm.anno_list, list) { 281 if (mptcp_addresses_equal(&entry->addr, &saddr, true)) { 282 ret = true; 283 goto out; 284 } 285 } 286 287 out: 288 spin_unlock_bh(&msk->pm.lock); 289 return ret; 290 } 291 292 static void mptcp_pm_add_timer(struct timer_list *timer) 293 { 294 struct mptcp_pm_add_entry *entry = from_timer(entry, timer, add_timer); 295 struct mptcp_sock *msk = entry->sock; 296 struct sock *sk = (struct sock *)msk; 297 298 pr_debug("msk=%p", msk); 299 300 if (!msk) 301 return; 302 303 if (inet_sk_state_load(sk) == TCP_CLOSE) 304 return; 305 306 if (!entry->addr.id) 307 return; 308 309 if (mptcp_pm_should_add_signal_addr(msk)) { 310 sk_reset_timer(sk, timer, jiffies + TCP_RTO_MAX / 8); 311 goto out; 312 } 313 314 spin_lock_bh(&msk->pm.lock); 315 316 if (!mptcp_pm_should_add_signal_addr(msk)) { 317 pr_debug("retransmit ADD_ADDR id=%d", entry->addr.id); 318 mptcp_pm_announce_addr(msk, &entry->addr, false); 319 mptcp_pm_add_addr_send_ack(msk); 320 entry->retrans_times++; 321 } 322 323 if (entry->retrans_times < ADD_ADDR_RETRANS_MAX) 324 sk_reset_timer(sk, timer, 325 jiffies + mptcp_get_add_addr_timeout(sock_net(sk))); 326 327 spin_unlock_bh(&msk->pm.lock); 328 329 if (entry->retrans_times == ADD_ADDR_RETRANS_MAX) 330 mptcp_pm_subflow_established(msk); 331 332 out: 333 __sock_put(sk); 334 } 335 336 struct mptcp_pm_add_entry * 337 mptcp_pm_del_add_timer(struct mptcp_sock *msk, 338 const struct mptcp_addr_info *addr, bool check_id) 339 { 340 struct mptcp_pm_add_entry *entry; 341 struct sock *sk = (struct sock *)msk; 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 spin_unlock_bh(&msk->pm.lock); 348 349 if (entry && (!check_id || entry->addr.id == addr->id)) 350 sk_stop_timer_sync(sk, &entry->add_timer); 351 352 return entry; 353 } 354 355 bool mptcp_pm_alloc_anno_list(struct mptcp_sock *msk, 356 const struct mptcp_pm_addr_entry *entry) 357 { 358 struct mptcp_pm_add_entry *add_entry = NULL; 359 struct sock *sk = (struct sock *)msk; 360 struct net *net = sock_net(sk); 361 362 lockdep_assert_held(&msk->pm.lock); 363 364 add_entry = mptcp_lookup_anno_list_by_saddr(msk, &entry->addr); 365 366 if (add_entry) { 367 if (mptcp_pm_is_kernel(msk)) 368 return false; 369 370 sk_reset_timer(sk, &add_entry->add_timer, 371 jiffies + mptcp_get_add_addr_timeout(net)); 372 return true; 373 } 374 375 add_entry = kmalloc(sizeof(*add_entry), GFP_ATOMIC); 376 if (!add_entry) 377 return false; 378 379 list_add(&add_entry->list, &msk->pm.anno_list); 380 381 add_entry->addr = entry->addr; 382 add_entry->sock = msk; 383 add_entry->retrans_times = 0; 384 385 timer_setup(&add_entry->add_timer, mptcp_pm_add_timer, 0); 386 sk_reset_timer(sk, &add_entry->add_timer, 387 jiffies + mptcp_get_add_addr_timeout(net)); 388 389 return true; 390 } 391 392 void mptcp_pm_free_anno_list(struct mptcp_sock *msk) 393 { 394 struct mptcp_pm_add_entry *entry, *tmp; 395 struct sock *sk = (struct sock *)msk; 396 LIST_HEAD(free_list); 397 398 pr_debug("msk=%p", msk); 399 400 spin_lock_bh(&msk->pm.lock); 401 list_splice_init(&msk->pm.anno_list, &free_list); 402 spin_unlock_bh(&msk->pm.lock); 403 404 list_for_each_entry_safe(entry, tmp, &free_list, list) { 405 sk_stop_timer_sync(sk, &entry->add_timer); 406 kfree(entry); 407 } 408 } 409 410 static bool lookup_address_in_vec(const struct mptcp_addr_info *addrs, unsigned int nr, 411 const struct mptcp_addr_info *addr) 412 { 413 int i; 414 415 for (i = 0; i < nr; i++) { 416 if (addrs[i].id == addr->id) 417 return true; 418 } 419 420 return false; 421 } 422 423 /* Fill all the remote addresses into the array addrs[], 424 * and return the array size. 425 */ 426 static unsigned int fill_remote_addresses_vec(struct mptcp_sock *msk, bool fullmesh, 427 struct mptcp_addr_info *addrs) 428 { 429 bool deny_id0 = READ_ONCE(msk->pm.remote_deny_join_id0); 430 struct sock *sk = (struct sock *)msk, *ssk; 431 struct mptcp_subflow_context *subflow; 432 struct mptcp_addr_info remote = { 0 }; 433 unsigned int subflows_max; 434 int i = 0; 435 436 subflows_max = mptcp_pm_get_subflows_max(msk); 437 remote_address((struct sock_common *)sk, &remote); 438 439 /* Non-fullmesh endpoint, fill in the single entry 440 * corresponding to the primary MPC subflow remote address 441 */ 442 if (!fullmesh) { 443 if (deny_id0) 444 return 0; 445 446 msk->pm.subflows++; 447 addrs[i++] = remote; 448 } else { 449 mptcp_for_each_subflow(msk, subflow) { 450 ssk = mptcp_subflow_tcp_sock(subflow); 451 remote_address((struct sock_common *)ssk, &addrs[i]); 452 addrs[i].id = subflow->remote_id; 453 if (deny_id0 && !addrs[i].id) 454 continue; 455 456 if (!lookup_address_in_vec(addrs, i, &addrs[i]) && 457 msk->pm.subflows < subflows_max) { 458 msk->pm.subflows++; 459 i++; 460 } 461 } 462 } 463 464 return i; 465 } 466 467 static void __mptcp_pm_send_ack(struct mptcp_sock *msk, struct mptcp_subflow_context *subflow, 468 bool prio, bool backup) 469 { 470 struct sock *ssk = mptcp_subflow_tcp_sock(subflow); 471 bool slow; 472 473 pr_debug("send ack for %s", 474 prio ? "mp_prio" : (mptcp_pm_should_add_signal(msk) ? "add_addr" : "rm_addr")); 475 476 slow = lock_sock_fast(ssk); 477 if (prio) { 478 if (subflow->backup != backup) 479 msk->last_snd = NULL; 480 481 subflow->send_mp_prio = 1; 482 subflow->backup = backup; 483 subflow->request_bkup = backup; 484 } 485 486 __mptcp_subflow_send_ack(ssk); 487 unlock_sock_fast(ssk, slow); 488 } 489 490 static void mptcp_pm_send_ack(struct mptcp_sock *msk, struct mptcp_subflow_context *subflow, 491 bool prio, bool backup) 492 { 493 spin_unlock_bh(&msk->pm.lock); 494 __mptcp_pm_send_ack(msk, subflow, prio, backup); 495 spin_lock_bh(&msk->pm.lock); 496 } 497 498 static struct mptcp_pm_addr_entry * 499 __lookup_addr_by_id(struct pm_nl_pernet *pernet, unsigned int id) 500 { 501 struct mptcp_pm_addr_entry *entry; 502 503 list_for_each_entry(entry, &pernet->local_addr_list, list) { 504 if (entry->addr.id == id) 505 return entry; 506 } 507 return NULL; 508 } 509 510 static struct mptcp_pm_addr_entry * 511 __lookup_addr(struct pm_nl_pernet *pernet, const struct mptcp_addr_info *info, 512 bool lookup_by_id) 513 { 514 struct mptcp_pm_addr_entry *entry; 515 516 list_for_each_entry(entry, &pernet->local_addr_list, list) { 517 if ((!lookup_by_id && 518 mptcp_addresses_equal(&entry->addr, info, entry->addr.port)) || 519 (lookup_by_id && entry->addr.id == info->id)) 520 return entry; 521 } 522 return NULL; 523 } 524 525 static void mptcp_pm_create_subflow_or_signal_addr(struct mptcp_sock *msk) 526 { 527 struct sock *sk = (struct sock *)msk; 528 struct mptcp_pm_addr_entry *local; 529 unsigned int add_addr_signal_max; 530 unsigned int local_addr_max; 531 struct pm_nl_pernet *pernet; 532 unsigned int subflows_max; 533 534 pernet = pm_nl_get_pernet(sock_net(sk)); 535 536 add_addr_signal_max = mptcp_pm_get_add_addr_signal_max(msk); 537 local_addr_max = mptcp_pm_get_local_addr_max(msk); 538 subflows_max = mptcp_pm_get_subflows_max(msk); 539 540 /* do lazy endpoint usage accounting for the MPC subflows */ 541 if (unlikely(!(msk->pm.status & BIT(MPTCP_PM_MPC_ENDPOINT_ACCOUNTED))) && msk->first) { 542 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(msk->first); 543 struct mptcp_pm_addr_entry *entry; 544 struct mptcp_addr_info mpc_addr; 545 bool backup = false; 546 547 local_address((struct sock_common *)msk->first, &mpc_addr); 548 rcu_read_lock(); 549 entry = __lookup_addr(pernet, &mpc_addr, false); 550 if (entry) { 551 __clear_bit(entry->addr.id, msk->pm.id_avail_bitmap); 552 msk->mpc_endpoint_id = entry->addr.id; 553 backup = !!(entry->flags & MPTCP_PM_ADDR_FLAG_BACKUP); 554 } 555 rcu_read_unlock(); 556 557 if (backup) 558 mptcp_pm_send_ack(msk, subflow, true, backup); 559 560 msk->pm.status |= BIT(MPTCP_PM_MPC_ENDPOINT_ACCOUNTED); 561 } 562 563 pr_debug("local %d:%d signal %d:%d subflows %d:%d\n", 564 msk->pm.local_addr_used, local_addr_max, 565 msk->pm.add_addr_signaled, add_addr_signal_max, 566 msk->pm.subflows, subflows_max); 567 568 /* check first for announce */ 569 if (msk->pm.add_addr_signaled < add_addr_signal_max) { 570 local = select_signal_address(pernet, msk); 571 572 /* due to racing events on both ends we can reach here while 573 * previous add address is still running: if we invoke now 574 * mptcp_pm_announce_addr(), that will fail and the 575 * corresponding id will be marked as used. 576 * Instead let the PM machinery reschedule us when the 577 * current address announce will be completed. 578 */ 579 if (msk->pm.addr_signal & BIT(MPTCP_ADD_ADDR_SIGNAL)) 580 return; 581 582 if (local) { 583 if (mptcp_pm_alloc_anno_list(msk, local)) { 584 __clear_bit(local->addr.id, msk->pm.id_avail_bitmap); 585 msk->pm.add_addr_signaled++; 586 mptcp_pm_announce_addr(msk, &local->addr, false); 587 mptcp_pm_nl_addr_send_ack(msk); 588 } 589 } 590 } 591 592 /* check if should create a new subflow */ 593 while (msk->pm.local_addr_used < local_addr_max && 594 msk->pm.subflows < subflows_max) { 595 struct mptcp_addr_info addrs[MPTCP_PM_ADDR_MAX]; 596 bool fullmesh; 597 int i, nr; 598 599 local = select_local_address(pernet, msk); 600 if (!local) 601 break; 602 603 fullmesh = !!(local->flags & MPTCP_PM_ADDR_FLAG_FULLMESH); 604 605 msk->pm.local_addr_used++; 606 nr = fill_remote_addresses_vec(msk, fullmesh, addrs); 607 if (nr) 608 __clear_bit(local->addr.id, msk->pm.id_avail_bitmap); 609 spin_unlock_bh(&msk->pm.lock); 610 for (i = 0; i < nr; i++) 611 __mptcp_subflow_connect(sk, &local->addr, &addrs[i]); 612 spin_lock_bh(&msk->pm.lock); 613 } 614 mptcp_pm_nl_check_work_pending(msk); 615 } 616 617 static void mptcp_pm_nl_fully_established(struct mptcp_sock *msk) 618 { 619 mptcp_pm_create_subflow_or_signal_addr(msk); 620 } 621 622 static void mptcp_pm_nl_subflow_established(struct mptcp_sock *msk) 623 { 624 mptcp_pm_create_subflow_or_signal_addr(msk); 625 } 626 627 /* Fill all the local addresses into the array addrs[], 628 * and return the array size. 629 */ 630 static unsigned int fill_local_addresses_vec(struct mptcp_sock *msk, 631 struct mptcp_addr_info *addrs) 632 { 633 struct sock *sk = (struct sock *)msk; 634 struct mptcp_pm_addr_entry *entry; 635 struct mptcp_addr_info local; 636 struct pm_nl_pernet *pernet; 637 unsigned int subflows_max; 638 int i = 0; 639 640 pernet = pm_nl_get_pernet_from_msk(msk); 641 subflows_max = mptcp_pm_get_subflows_max(msk); 642 643 rcu_read_lock(); 644 list_for_each_entry_rcu(entry, &pernet->local_addr_list, list) { 645 if (!(entry->flags & MPTCP_PM_ADDR_FLAG_FULLMESH)) 646 continue; 647 648 if (entry->addr.family != sk->sk_family) { 649 #if IS_ENABLED(CONFIG_MPTCP_IPV6) 650 if ((entry->addr.family == AF_INET && 651 !ipv6_addr_v4mapped(&sk->sk_v6_daddr)) || 652 (sk->sk_family == AF_INET && 653 !ipv6_addr_v4mapped(&entry->addr.addr6))) 654 #endif 655 continue; 656 } 657 658 if (msk->pm.subflows < subflows_max) { 659 msk->pm.subflows++; 660 addrs[i++] = entry->addr; 661 } 662 } 663 rcu_read_unlock(); 664 665 /* If the array is empty, fill in the single 666 * 'IPADDRANY' local address 667 */ 668 if (!i) { 669 memset(&local, 0, sizeof(local)); 670 local.family = msk->pm.remote.family; 671 672 msk->pm.subflows++; 673 addrs[i++] = local; 674 } 675 676 return i; 677 } 678 679 static void mptcp_pm_nl_add_addr_received(struct mptcp_sock *msk) 680 { 681 struct mptcp_addr_info addrs[MPTCP_PM_ADDR_MAX]; 682 struct sock *sk = (struct sock *)msk; 683 unsigned int add_addr_accept_max; 684 struct mptcp_addr_info remote; 685 unsigned int subflows_max; 686 int i, nr; 687 688 add_addr_accept_max = mptcp_pm_get_add_addr_accept_max(msk); 689 subflows_max = mptcp_pm_get_subflows_max(msk); 690 691 pr_debug("accepted %d:%d remote family %d", 692 msk->pm.add_addr_accepted, add_addr_accept_max, 693 msk->pm.remote.family); 694 695 remote = msk->pm.remote; 696 mptcp_pm_announce_addr(msk, &remote, true); 697 mptcp_pm_nl_addr_send_ack(msk); 698 699 if (lookup_subflow_by_daddr(&msk->conn_list, &remote)) 700 return; 701 702 /* pick id 0 port, if none is provided the remote address */ 703 if (!remote.port) 704 remote.port = sk->sk_dport; 705 706 /* connect to the specified remote address, using whatever 707 * local address the routing configuration will pick. 708 */ 709 nr = fill_local_addresses_vec(msk, addrs); 710 711 msk->pm.add_addr_accepted++; 712 if (msk->pm.add_addr_accepted >= add_addr_accept_max || 713 msk->pm.subflows >= subflows_max) 714 WRITE_ONCE(msk->pm.accept_addr, false); 715 716 spin_unlock_bh(&msk->pm.lock); 717 for (i = 0; i < nr; i++) 718 __mptcp_subflow_connect(sk, &addrs[i], &remote); 719 spin_lock_bh(&msk->pm.lock); 720 } 721 722 void mptcp_pm_nl_addr_send_ack(struct mptcp_sock *msk) 723 { 724 struct mptcp_subflow_context *subflow; 725 726 msk_owned_by_me(msk); 727 lockdep_assert_held(&msk->pm.lock); 728 729 if (!mptcp_pm_should_add_signal(msk) && 730 !mptcp_pm_should_rm_signal(msk)) 731 return; 732 733 subflow = list_first_entry_or_null(&msk->conn_list, typeof(*subflow), node); 734 if (subflow) 735 mptcp_pm_send_ack(msk, subflow, false, false); 736 } 737 738 int mptcp_pm_nl_mp_prio_send_ack(struct mptcp_sock *msk, 739 struct mptcp_addr_info *addr, 740 struct mptcp_addr_info *rem, 741 u8 bkup) 742 { 743 struct mptcp_subflow_context *subflow; 744 745 pr_debug("bkup=%d", bkup); 746 747 mptcp_for_each_subflow(msk, subflow) { 748 struct sock *ssk = mptcp_subflow_tcp_sock(subflow); 749 struct mptcp_addr_info local, remote; 750 751 local_address((struct sock_common *)ssk, &local); 752 if (!mptcp_addresses_equal(&local, addr, addr->port)) 753 continue; 754 755 if (rem && rem->family != AF_UNSPEC) { 756 remote_address((struct sock_common *)ssk, &remote); 757 if (!mptcp_addresses_equal(&remote, rem, rem->port)) 758 continue; 759 } 760 761 __mptcp_pm_send_ack(msk, subflow, true, bkup); 762 return 0; 763 } 764 765 return -EINVAL; 766 } 767 768 static bool mptcp_local_id_match(const struct mptcp_sock *msk, u8 local_id, u8 id) 769 { 770 return local_id == id || (!local_id && msk->mpc_endpoint_id == id); 771 } 772 773 static void mptcp_pm_nl_rm_addr_or_subflow(struct mptcp_sock *msk, 774 const struct mptcp_rm_list *rm_list, 775 enum linux_mptcp_mib_field rm_type) 776 { 777 struct mptcp_subflow_context *subflow, *tmp; 778 struct sock *sk = (struct sock *)msk; 779 u8 i; 780 781 pr_debug("%s rm_list_nr %d", 782 rm_type == MPTCP_MIB_RMADDR ? "address" : "subflow", rm_list->nr); 783 784 msk_owned_by_me(msk); 785 786 if (sk->sk_state == TCP_LISTEN) 787 return; 788 789 if (!rm_list->nr) 790 return; 791 792 if (list_empty(&msk->conn_list)) 793 return; 794 795 for (i = 0; i < rm_list->nr; i++) { 796 u8 rm_id = rm_list->ids[i]; 797 bool removed = false; 798 799 mptcp_for_each_subflow_safe(msk, subflow, tmp) { 800 struct sock *ssk = mptcp_subflow_tcp_sock(subflow); 801 int how = RCV_SHUTDOWN | SEND_SHUTDOWN; 802 u8 id = subflow->local_id; 803 804 if (rm_type == MPTCP_MIB_RMADDR && subflow->remote_id != rm_id) 805 continue; 806 if (rm_type == MPTCP_MIB_RMSUBFLOW && !mptcp_local_id_match(msk, id, rm_id)) 807 continue; 808 809 pr_debug(" -> %s rm_list_ids[%d]=%u local_id=%u remote_id=%u mpc_id=%u", 810 rm_type == MPTCP_MIB_RMADDR ? "address" : "subflow", 811 i, rm_id, subflow->local_id, subflow->remote_id, 812 msk->mpc_endpoint_id); 813 spin_unlock_bh(&msk->pm.lock); 814 mptcp_subflow_shutdown(sk, ssk, how); 815 816 /* the following takes care of updating the subflows counter */ 817 mptcp_close_ssk(sk, ssk, subflow); 818 spin_lock_bh(&msk->pm.lock); 819 820 removed = true; 821 __MPTCP_INC_STATS(sock_net(sk), rm_type); 822 } 823 if (rm_type == MPTCP_MIB_RMSUBFLOW) 824 __set_bit(rm_id ? rm_id : msk->mpc_endpoint_id, msk->pm.id_avail_bitmap); 825 if (!removed) 826 continue; 827 828 if (!mptcp_pm_is_kernel(msk)) 829 continue; 830 831 if (rm_type == MPTCP_MIB_RMADDR) { 832 msk->pm.add_addr_accepted--; 833 WRITE_ONCE(msk->pm.accept_addr, true); 834 } else if (rm_type == MPTCP_MIB_RMSUBFLOW) { 835 msk->pm.local_addr_used--; 836 } 837 } 838 } 839 840 static void mptcp_pm_nl_rm_addr_received(struct mptcp_sock *msk) 841 { 842 mptcp_pm_nl_rm_addr_or_subflow(msk, &msk->pm.rm_list_rx, MPTCP_MIB_RMADDR); 843 } 844 845 void mptcp_pm_nl_rm_subflow_received(struct mptcp_sock *msk, 846 const struct mptcp_rm_list *rm_list) 847 { 848 mptcp_pm_nl_rm_addr_or_subflow(msk, rm_list, MPTCP_MIB_RMSUBFLOW); 849 } 850 851 void mptcp_pm_nl_work(struct mptcp_sock *msk) 852 { 853 struct mptcp_pm_data *pm = &msk->pm; 854 855 msk_owned_by_me(msk); 856 857 if (!(pm->status & MPTCP_PM_WORK_MASK)) 858 return; 859 860 spin_lock_bh(&msk->pm.lock); 861 862 pr_debug("msk=%p status=%x", msk, pm->status); 863 if (pm->status & BIT(MPTCP_PM_ADD_ADDR_RECEIVED)) { 864 pm->status &= ~BIT(MPTCP_PM_ADD_ADDR_RECEIVED); 865 mptcp_pm_nl_add_addr_received(msk); 866 } 867 if (pm->status & BIT(MPTCP_PM_ADD_ADDR_SEND_ACK)) { 868 pm->status &= ~BIT(MPTCP_PM_ADD_ADDR_SEND_ACK); 869 mptcp_pm_nl_addr_send_ack(msk); 870 } 871 if (pm->status & BIT(MPTCP_PM_RM_ADDR_RECEIVED)) { 872 pm->status &= ~BIT(MPTCP_PM_RM_ADDR_RECEIVED); 873 mptcp_pm_nl_rm_addr_received(msk); 874 } 875 if (pm->status & BIT(MPTCP_PM_ESTABLISHED)) { 876 pm->status &= ~BIT(MPTCP_PM_ESTABLISHED); 877 mptcp_pm_nl_fully_established(msk); 878 } 879 if (pm->status & BIT(MPTCP_PM_SUBFLOW_ESTABLISHED)) { 880 pm->status &= ~BIT(MPTCP_PM_SUBFLOW_ESTABLISHED); 881 mptcp_pm_nl_subflow_established(msk); 882 } 883 884 spin_unlock_bh(&msk->pm.lock); 885 } 886 887 static bool address_use_port(struct mptcp_pm_addr_entry *entry) 888 { 889 return (entry->flags & 890 (MPTCP_PM_ADDR_FLAG_SIGNAL | MPTCP_PM_ADDR_FLAG_SUBFLOW)) == 891 MPTCP_PM_ADDR_FLAG_SIGNAL; 892 } 893 894 /* caller must ensure the RCU grace period is already elapsed */ 895 static void __mptcp_pm_release_addr_entry(struct mptcp_pm_addr_entry *entry) 896 { 897 if (entry->lsk) 898 sock_release(entry->lsk); 899 kfree(entry); 900 } 901 902 static int mptcp_pm_nl_append_new_local_addr(struct pm_nl_pernet *pernet, 903 struct mptcp_pm_addr_entry *entry) 904 { 905 struct mptcp_pm_addr_entry *cur, *del_entry = NULL; 906 unsigned int addr_max; 907 int ret = -EINVAL; 908 909 spin_lock_bh(&pernet->lock); 910 /* to keep the code simple, don't do IDR-like allocation for address ID, 911 * just bail when we exceed limits 912 */ 913 if (pernet->next_id == MPTCP_PM_MAX_ADDR_ID) 914 pernet->next_id = 1; 915 if (pernet->addrs >= MPTCP_PM_ADDR_MAX) { 916 ret = -ERANGE; 917 goto out; 918 } 919 if (test_bit(entry->addr.id, pernet->id_bitmap)) { 920 ret = -EBUSY; 921 goto out; 922 } 923 924 /* do not insert duplicate address, differentiate on port only 925 * singled addresses 926 */ 927 if (!address_use_port(entry)) 928 entry->addr.port = 0; 929 list_for_each_entry(cur, &pernet->local_addr_list, list) { 930 if (mptcp_addresses_equal(&cur->addr, &entry->addr, 931 cur->addr.port || entry->addr.port)) { 932 /* allow replacing the exiting endpoint only if such 933 * endpoint is an implicit one and the user-space 934 * did not provide an endpoint id 935 */ 936 if (!(cur->flags & MPTCP_PM_ADDR_FLAG_IMPLICIT)) { 937 ret = -EEXIST; 938 goto out; 939 } 940 if (entry->addr.id) 941 goto out; 942 943 pernet->addrs--; 944 entry->addr.id = cur->addr.id; 945 list_del_rcu(&cur->list); 946 del_entry = cur; 947 break; 948 } 949 } 950 951 if (!entry->addr.id) { 952 find_next: 953 entry->addr.id = find_next_zero_bit(pernet->id_bitmap, 954 MPTCP_PM_MAX_ADDR_ID + 1, 955 pernet->next_id); 956 if (!entry->addr.id && pernet->next_id != 1) { 957 pernet->next_id = 1; 958 goto find_next; 959 } 960 } 961 962 if (!entry->addr.id) 963 goto out; 964 965 __set_bit(entry->addr.id, pernet->id_bitmap); 966 if (entry->addr.id > pernet->next_id) 967 pernet->next_id = entry->addr.id; 968 969 if (entry->flags & MPTCP_PM_ADDR_FLAG_SIGNAL) { 970 addr_max = pernet->add_addr_signal_max; 971 WRITE_ONCE(pernet->add_addr_signal_max, addr_max + 1); 972 } 973 if (entry->flags & MPTCP_PM_ADDR_FLAG_SUBFLOW) { 974 addr_max = pernet->local_addr_max; 975 WRITE_ONCE(pernet->local_addr_max, addr_max + 1); 976 } 977 978 pernet->addrs++; 979 if (!entry->addr.port) 980 list_add_tail_rcu(&entry->list, &pernet->local_addr_list); 981 else 982 list_add_rcu(&entry->list, &pernet->local_addr_list); 983 ret = entry->addr.id; 984 985 out: 986 spin_unlock_bh(&pernet->lock); 987 988 /* just replaced an existing entry, free it */ 989 if (del_entry) { 990 synchronize_rcu(); 991 __mptcp_pm_release_addr_entry(del_entry); 992 } 993 return ret; 994 } 995 996 static int mptcp_pm_nl_create_listen_socket(struct sock *sk, 997 struct mptcp_pm_addr_entry *entry) 998 { 999 int addrlen = sizeof(struct sockaddr_in); 1000 struct sockaddr_storage addr; 1001 struct mptcp_sock *msk; 1002 struct socket *ssock; 1003 int backlog = 1024; 1004 int err; 1005 1006 err = sock_create_kern(sock_net(sk), entry->addr.family, 1007 SOCK_STREAM, IPPROTO_MPTCP, &entry->lsk); 1008 if (err) 1009 return err; 1010 1011 msk = mptcp_sk(entry->lsk->sk); 1012 if (!msk) 1013 return -EINVAL; 1014 1015 ssock = __mptcp_nmpc_socket(msk); 1016 if (!ssock) 1017 return -EINVAL; 1018 1019 mptcp_info2sockaddr(&entry->addr, &addr, entry->addr.family); 1020 #if IS_ENABLED(CONFIG_MPTCP_IPV6) 1021 if (entry->addr.family == AF_INET6) 1022 addrlen = sizeof(struct sockaddr_in6); 1023 #endif 1024 err = kernel_bind(ssock, (struct sockaddr *)&addr, addrlen); 1025 if (err) 1026 return err; 1027 1028 err = kernel_listen(ssock, backlog); 1029 if (err) 1030 return err; 1031 1032 return 0; 1033 } 1034 1035 int mptcp_pm_nl_get_local_id(struct mptcp_sock *msk, struct sock_common *skc) 1036 { 1037 struct mptcp_pm_addr_entry *entry; 1038 struct mptcp_addr_info skc_local; 1039 struct mptcp_addr_info msk_local; 1040 struct pm_nl_pernet *pernet; 1041 int ret = -1; 1042 1043 if (WARN_ON_ONCE(!msk)) 1044 return -1; 1045 1046 /* The 0 ID mapping is defined by the first subflow, copied into the msk 1047 * addr 1048 */ 1049 local_address((struct sock_common *)msk, &msk_local); 1050 local_address((struct sock_common *)skc, &skc_local); 1051 if (mptcp_addresses_equal(&msk_local, &skc_local, false)) 1052 return 0; 1053 1054 if (mptcp_pm_is_userspace(msk)) 1055 return mptcp_userspace_pm_get_local_id(msk, &skc_local); 1056 1057 pernet = pm_nl_get_pernet_from_msk(msk); 1058 1059 rcu_read_lock(); 1060 list_for_each_entry_rcu(entry, &pernet->local_addr_list, list) { 1061 if (mptcp_addresses_equal(&entry->addr, &skc_local, entry->addr.port)) { 1062 ret = entry->addr.id; 1063 break; 1064 } 1065 } 1066 rcu_read_unlock(); 1067 if (ret >= 0) 1068 return ret; 1069 1070 /* address not found, add to local list */ 1071 entry = kmalloc(sizeof(*entry), GFP_ATOMIC); 1072 if (!entry) 1073 return -ENOMEM; 1074 1075 entry->addr = skc_local; 1076 entry->addr.id = 0; 1077 entry->addr.port = 0; 1078 entry->ifindex = 0; 1079 entry->flags = MPTCP_PM_ADDR_FLAG_IMPLICIT; 1080 entry->lsk = NULL; 1081 ret = mptcp_pm_nl_append_new_local_addr(pernet, entry); 1082 if (ret < 0) 1083 kfree(entry); 1084 1085 return ret; 1086 } 1087 1088 #define MPTCP_PM_CMD_GRP_OFFSET 0 1089 #define MPTCP_PM_EV_GRP_OFFSET 1 1090 1091 static const struct genl_multicast_group mptcp_pm_mcgrps[] = { 1092 [MPTCP_PM_CMD_GRP_OFFSET] = { .name = MPTCP_PM_CMD_GRP_NAME, }, 1093 [MPTCP_PM_EV_GRP_OFFSET] = { .name = MPTCP_PM_EV_GRP_NAME, 1094 .flags = GENL_UNS_ADMIN_PERM, 1095 }, 1096 }; 1097 1098 static const struct nla_policy 1099 mptcp_pm_addr_policy[MPTCP_PM_ADDR_ATTR_MAX + 1] = { 1100 [MPTCP_PM_ADDR_ATTR_FAMILY] = { .type = NLA_U16, }, 1101 [MPTCP_PM_ADDR_ATTR_ID] = { .type = NLA_U8, }, 1102 [MPTCP_PM_ADDR_ATTR_ADDR4] = { .type = NLA_U32, }, 1103 [MPTCP_PM_ADDR_ATTR_ADDR6] = 1104 NLA_POLICY_EXACT_LEN(sizeof(struct in6_addr)), 1105 [MPTCP_PM_ADDR_ATTR_PORT] = { .type = NLA_U16 }, 1106 [MPTCP_PM_ADDR_ATTR_FLAGS] = { .type = NLA_U32 }, 1107 [MPTCP_PM_ADDR_ATTR_IF_IDX] = { .type = NLA_S32 }, 1108 }; 1109 1110 static const struct nla_policy mptcp_pm_policy[MPTCP_PM_ATTR_MAX + 1] = { 1111 [MPTCP_PM_ATTR_ADDR] = 1112 NLA_POLICY_NESTED(mptcp_pm_addr_policy), 1113 [MPTCP_PM_ATTR_RCV_ADD_ADDRS] = { .type = NLA_U32, }, 1114 [MPTCP_PM_ATTR_SUBFLOWS] = { .type = NLA_U32, }, 1115 [MPTCP_PM_ATTR_TOKEN] = { .type = NLA_U32, }, 1116 [MPTCP_PM_ATTR_LOC_ID] = { .type = NLA_U8, }, 1117 [MPTCP_PM_ATTR_ADDR_REMOTE] = 1118 NLA_POLICY_NESTED(mptcp_pm_addr_policy), 1119 }; 1120 1121 void mptcp_pm_nl_subflow_chk_stale(const struct mptcp_sock *msk, struct sock *ssk) 1122 { 1123 struct mptcp_subflow_context *iter, *subflow = mptcp_subflow_ctx(ssk); 1124 struct sock *sk = (struct sock *)msk; 1125 unsigned int active_max_loss_cnt; 1126 struct net *net = sock_net(sk); 1127 unsigned int stale_loss_cnt; 1128 bool slow; 1129 1130 stale_loss_cnt = mptcp_stale_loss_cnt(net); 1131 if (subflow->stale || !stale_loss_cnt || subflow->stale_count <= stale_loss_cnt) 1132 return; 1133 1134 /* look for another available subflow not in loss state */ 1135 active_max_loss_cnt = max_t(int, stale_loss_cnt - 1, 1); 1136 mptcp_for_each_subflow(msk, iter) { 1137 if (iter != subflow && mptcp_subflow_active(iter) && 1138 iter->stale_count < active_max_loss_cnt) { 1139 /* we have some alternatives, try to mark this subflow as idle ...*/ 1140 slow = lock_sock_fast(ssk); 1141 if (!tcp_rtx_and_write_queues_empty(ssk)) { 1142 subflow->stale = 1; 1143 __mptcp_retransmit_pending_data(sk); 1144 MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_SUBFLOWSTALE); 1145 } 1146 unlock_sock_fast(ssk, slow); 1147 1148 /* always try to push the pending data regardless of re-injections: 1149 * we can possibly use backup subflows now, and subflow selection 1150 * is cheap under the msk socket lock 1151 */ 1152 __mptcp_push_pending(sk, 0); 1153 return; 1154 } 1155 } 1156 } 1157 1158 static int mptcp_pm_family_to_addr(int family) 1159 { 1160 #if IS_ENABLED(CONFIG_MPTCP_IPV6) 1161 if (family == AF_INET6) 1162 return MPTCP_PM_ADDR_ATTR_ADDR6; 1163 #endif 1164 return MPTCP_PM_ADDR_ATTR_ADDR4; 1165 } 1166 1167 static int mptcp_pm_parse_pm_addr_attr(struct nlattr *tb[], 1168 const struct nlattr *attr, 1169 struct genl_info *info, 1170 struct mptcp_addr_info *addr, 1171 bool require_family) 1172 { 1173 int err, addr_addr; 1174 1175 if (!attr) { 1176 GENL_SET_ERR_MSG(info, "missing address info"); 1177 return -EINVAL; 1178 } 1179 1180 /* no validation needed - was already done via nested policy */ 1181 err = nla_parse_nested_deprecated(tb, MPTCP_PM_ADDR_ATTR_MAX, attr, 1182 mptcp_pm_addr_policy, info->extack); 1183 if (err) 1184 return err; 1185 1186 if (tb[MPTCP_PM_ADDR_ATTR_ID]) 1187 addr->id = nla_get_u8(tb[MPTCP_PM_ADDR_ATTR_ID]); 1188 1189 if (!tb[MPTCP_PM_ADDR_ATTR_FAMILY]) { 1190 if (!require_family) 1191 return err; 1192 1193 NL_SET_ERR_MSG_ATTR(info->extack, attr, 1194 "missing family"); 1195 return -EINVAL; 1196 } 1197 1198 addr->family = nla_get_u16(tb[MPTCP_PM_ADDR_ATTR_FAMILY]); 1199 if (addr->family != AF_INET 1200 #if IS_ENABLED(CONFIG_MPTCP_IPV6) 1201 && addr->family != AF_INET6 1202 #endif 1203 ) { 1204 NL_SET_ERR_MSG_ATTR(info->extack, attr, 1205 "unknown address family"); 1206 return -EINVAL; 1207 } 1208 addr_addr = mptcp_pm_family_to_addr(addr->family); 1209 if (!tb[addr_addr]) { 1210 NL_SET_ERR_MSG_ATTR(info->extack, attr, 1211 "missing address data"); 1212 return -EINVAL; 1213 } 1214 1215 #if IS_ENABLED(CONFIG_MPTCP_IPV6) 1216 if (addr->family == AF_INET6) 1217 addr->addr6 = nla_get_in6_addr(tb[addr_addr]); 1218 else 1219 #endif 1220 addr->addr.s_addr = nla_get_in_addr(tb[addr_addr]); 1221 1222 if (tb[MPTCP_PM_ADDR_ATTR_PORT]) 1223 addr->port = htons(nla_get_u16(tb[MPTCP_PM_ADDR_ATTR_PORT])); 1224 1225 return err; 1226 } 1227 1228 int mptcp_pm_parse_addr(struct nlattr *attr, struct genl_info *info, 1229 struct mptcp_addr_info *addr) 1230 { 1231 struct nlattr *tb[MPTCP_PM_ADDR_ATTR_MAX + 1]; 1232 1233 memset(addr, 0, sizeof(*addr)); 1234 1235 return mptcp_pm_parse_pm_addr_attr(tb, attr, info, addr, true); 1236 } 1237 1238 int mptcp_pm_parse_entry(struct nlattr *attr, struct genl_info *info, 1239 bool require_family, 1240 struct mptcp_pm_addr_entry *entry) 1241 { 1242 struct nlattr *tb[MPTCP_PM_ADDR_ATTR_MAX + 1]; 1243 int err; 1244 1245 memset(entry, 0, sizeof(*entry)); 1246 1247 err = mptcp_pm_parse_pm_addr_attr(tb, attr, info, &entry->addr, require_family); 1248 if (err) 1249 return err; 1250 1251 if (tb[MPTCP_PM_ADDR_ATTR_IF_IDX]) { 1252 u32 val = nla_get_s32(tb[MPTCP_PM_ADDR_ATTR_IF_IDX]); 1253 1254 entry->ifindex = val; 1255 } 1256 1257 if (tb[MPTCP_PM_ADDR_ATTR_FLAGS]) 1258 entry->flags = nla_get_u32(tb[MPTCP_PM_ADDR_ATTR_FLAGS]); 1259 1260 if (tb[MPTCP_PM_ADDR_ATTR_PORT]) 1261 entry->addr.port = htons(nla_get_u16(tb[MPTCP_PM_ADDR_ATTR_PORT])); 1262 1263 return 0; 1264 } 1265 1266 static struct pm_nl_pernet *genl_info_pm_nl(struct genl_info *info) 1267 { 1268 return pm_nl_get_pernet(genl_info_net(info)); 1269 } 1270 1271 static int mptcp_nl_add_subflow_or_signal_addr(struct net *net) 1272 { 1273 struct mptcp_sock *msk; 1274 long s_slot = 0, s_num = 0; 1275 1276 while ((msk = mptcp_token_iter_next(net, &s_slot, &s_num)) != NULL) { 1277 struct sock *sk = (struct sock *)msk; 1278 1279 if (!READ_ONCE(msk->fully_established) || 1280 mptcp_pm_is_userspace(msk)) 1281 goto next; 1282 1283 lock_sock(sk); 1284 spin_lock_bh(&msk->pm.lock); 1285 mptcp_pm_create_subflow_or_signal_addr(msk); 1286 spin_unlock_bh(&msk->pm.lock); 1287 release_sock(sk); 1288 1289 next: 1290 sock_put(sk); 1291 cond_resched(); 1292 } 1293 1294 return 0; 1295 } 1296 1297 static int mptcp_nl_cmd_add_addr(struct sk_buff *skb, struct genl_info *info) 1298 { 1299 struct nlattr *attr = info->attrs[MPTCP_PM_ATTR_ADDR]; 1300 struct pm_nl_pernet *pernet = genl_info_pm_nl(info); 1301 struct mptcp_pm_addr_entry addr, *entry; 1302 int ret; 1303 1304 ret = mptcp_pm_parse_entry(attr, info, true, &addr); 1305 if (ret < 0) 1306 return ret; 1307 1308 if (addr.addr.port && !(addr.flags & MPTCP_PM_ADDR_FLAG_SIGNAL)) { 1309 GENL_SET_ERR_MSG(info, "flags must have signal when using port"); 1310 return -EINVAL; 1311 } 1312 1313 if (addr.flags & MPTCP_PM_ADDR_FLAG_SIGNAL && 1314 addr.flags & MPTCP_PM_ADDR_FLAG_FULLMESH) { 1315 GENL_SET_ERR_MSG(info, "flags mustn't have both signal and fullmesh"); 1316 return -EINVAL; 1317 } 1318 1319 if (addr.flags & MPTCP_PM_ADDR_FLAG_IMPLICIT) { 1320 GENL_SET_ERR_MSG(info, "can't create IMPLICIT endpoint"); 1321 return -EINVAL; 1322 } 1323 1324 entry = kzalloc(sizeof(*entry), GFP_KERNEL_ACCOUNT); 1325 if (!entry) { 1326 GENL_SET_ERR_MSG(info, "can't allocate addr"); 1327 return -ENOMEM; 1328 } 1329 1330 *entry = addr; 1331 if (entry->addr.port) { 1332 ret = mptcp_pm_nl_create_listen_socket(skb->sk, entry); 1333 if (ret) { 1334 GENL_SET_ERR_MSG_FMT(info, "create listen socket error: %d", ret); 1335 goto out_free; 1336 } 1337 } 1338 ret = mptcp_pm_nl_append_new_local_addr(pernet, entry); 1339 if (ret < 0) { 1340 GENL_SET_ERR_MSG_FMT(info, "too many addresses or duplicate one: %d", ret); 1341 goto out_free; 1342 } 1343 1344 mptcp_nl_add_subflow_or_signal_addr(sock_net(skb->sk)); 1345 return 0; 1346 1347 out_free: 1348 __mptcp_pm_release_addr_entry(entry); 1349 return ret; 1350 } 1351 1352 int mptcp_pm_get_flags_and_ifindex_by_id(struct mptcp_sock *msk, unsigned int id, 1353 u8 *flags, int *ifindex) 1354 { 1355 struct mptcp_pm_addr_entry *entry; 1356 struct sock *sk = (struct sock *)msk; 1357 struct net *net = sock_net(sk); 1358 1359 *flags = 0; 1360 *ifindex = 0; 1361 1362 if (id) { 1363 if (mptcp_pm_is_userspace(msk)) 1364 return mptcp_userspace_pm_get_flags_and_ifindex_by_id(msk, 1365 id, 1366 flags, 1367 ifindex); 1368 1369 rcu_read_lock(); 1370 entry = __lookup_addr_by_id(pm_nl_get_pernet(net), id); 1371 if (entry) { 1372 *flags = entry->flags; 1373 *ifindex = entry->ifindex; 1374 } 1375 rcu_read_unlock(); 1376 } 1377 1378 return 0; 1379 } 1380 1381 static bool remove_anno_list_by_saddr(struct mptcp_sock *msk, 1382 const struct mptcp_addr_info *addr) 1383 { 1384 struct mptcp_pm_add_entry *entry; 1385 1386 entry = mptcp_pm_del_add_timer(msk, addr, false); 1387 if (entry) { 1388 list_del(&entry->list); 1389 kfree(entry); 1390 return true; 1391 } 1392 1393 return false; 1394 } 1395 1396 static bool mptcp_pm_remove_anno_addr(struct mptcp_sock *msk, 1397 const struct mptcp_addr_info *addr, 1398 bool force) 1399 { 1400 struct mptcp_rm_list list = { .nr = 0 }; 1401 bool ret; 1402 1403 list.ids[list.nr++] = addr->id; 1404 1405 ret = remove_anno_list_by_saddr(msk, addr); 1406 if (ret || force) { 1407 spin_lock_bh(&msk->pm.lock); 1408 mptcp_pm_remove_addr(msk, &list); 1409 spin_unlock_bh(&msk->pm.lock); 1410 } 1411 return ret; 1412 } 1413 1414 static int mptcp_nl_remove_subflow_and_signal_addr(struct net *net, 1415 const struct mptcp_pm_addr_entry *entry) 1416 { 1417 const struct mptcp_addr_info *addr = &entry->addr; 1418 struct mptcp_rm_list list = { .nr = 0 }; 1419 long s_slot = 0, s_num = 0; 1420 struct mptcp_sock *msk; 1421 1422 pr_debug("remove_id=%d", addr->id); 1423 1424 list.ids[list.nr++] = addr->id; 1425 1426 while ((msk = mptcp_token_iter_next(net, &s_slot, &s_num)) != NULL) { 1427 struct sock *sk = (struct sock *)msk; 1428 bool remove_subflow; 1429 1430 if (mptcp_pm_is_userspace(msk)) 1431 goto next; 1432 1433 if (list_empty(&msk->conn_list)) { 1434 mptcp_pm_remove_anno_addr(msk, addr, false); 1435 goto next; 1436 } 1437 1438 lock_sock(sk); 1439 remove_subflow = lookup_subflow_by_saddr(&msk->conn_list, addr); 1440 mptcp_pm_remove_anno_addr(msk, addr, remove_subflow && 1441 !(entry->flags & MPTCP_PM_ADDR_FLAG_IMPLICIT)); 1442 if (remove_subflow) 1443 mptcp_pm_remove_subflow(msk, &list); 1444 release_sock(sk); 1445 1446 next: 1447 sock_put(sk); 1448 cond_resched(); 1449 } 1450 1451 return 0; 1452 } 1453 1454 static int mptcp_nl_remove_id_zero_address(struct net *net, 1455 struct mptcp_addr_info *addr) 1456 { 1457 struct mptcp_rm_list list = { .nr = 0 }; 1458 long s_slot = 0, s_num = 0; 1459 struct mptcp_sock *msk; 1460 1461 list.ids[list.nr++] = 0; 1462 1463 while ((msk = mptcp_token_iter_next(net, &s_slot, &s_num)) != NULL) { 1464 struct sock *sk = (struct sock *)msk; 1465 struct mptcp_addr_info msk_local; 1466 1467 if (list_empty(&msk->conn_list) || mptcp_pm_is_userspace(msk)) 1468 goto next; 1469 1470 local_address((struct sock_common *)msk, &msk_local); 1471 if (!mptcp_addresses_equal(&msk_local, addr, addr->port)) 1472 goto next; 1473 1474 lock_sock(sk); 1475 spin_lock_bh(&msk->pm.lock); 1476 mptcp_pm_remove_addr(msk, &list); 1477 mptcp_pm_nl_rm_subflow_received(msk, &list); 1478 spin_unlock_bh(&msk->pm.lock); 1479 release_sock(sk); 1480 1481 next: 1482 sock_put(sk); 1483 cond_resched(); 1484 } 1485 1486 return 0; 1487 } 1488 1489 static int mptcp_nl_cmd_del_addr(struct sk_buff *skb, struct genl_info *info) 1490 { 1491 struct nlattr *attr = info->attrs[MPTCP_PM_ATTR_ADDR]; 1492 struct pm_nl_pernet *pernet = genl_info_pm_nl(info); 1493 struct mptcp_pm_addr_entry addr, *entry; 1494 unsigned int addr_max; 1495 int ret; 1496 1497 ret = mptcp_pm_parse_entry(attr, info, false, &addr); 1498 if (ret < 0) 1499 return ret; 1500 1501 /* the zero id address is special: the first address used by the msk 1502 * always gets such an id, so different subflows can have different zero 1503 * id addresses. Additionally zero id is not accounted for in id_bitmap. 1504 * Let's use an 'mptcp_rm_list' instead of the common remove code. 1505 */ 1506 if (addr.addr.id == 0) 1507 return mptcp_nl_remove_id_zero_address(sock_net(skb->sk), &addr.addr); 1508 1509 spin_lock_bh(&pernet->lock); 1510 entry = __lookup_addr_by_id(pernet, addr.addr.id); 1511 if (!entry) { 1512 GENL_SET_ERR_MSG(info, "address not found"); 1513 spin_unlock_bh(&pernet->lock); 1514 return -EINVAL; 1515 } 1516 if (entry->flags & MPTCP_PM_ADDR_FLAG_SIGNAL) { 1517 addr_max = pernet->add_addr_signal_max; 1518 WRITE_ONCE(pernet->add_addr_signal_max, addr_max - 1); 1519 } 1520 if (entry->flags & MPTCP_PM_ADDR_FLAG_SUBFLOW) { 1521 addr_max = pernet->local_addr_max; 1522 WRITE_ONCE(pernet->local_addr_max, addr_max - 1); 1523 } 1524 1525 pernet->addrs--; 1526 list_del_rcu(&entry->list); 1527 __clear_bit(entry->addr.id, pernet->id_bitmap); 1528 spin_unlock_bh(&pernet->lock); 1529 1530 mptcp_nl_remove_subflow_and_signal_addr(sock_net(skb->sk), entry); 1531 synchronize_rcu(); 1532 __mptcp_pm_release_addr_entry(entry); 1533 1534 return ret; 1535 } 1536 1537 void mptcp_pm_remove_addrs_and_subflows(struct mptcp_sock *msk, 1538 struct list_head *rm_list) 1539 { 1540 struct mptcp_rm_list alist = { .nr = 0 }, slist = { .nr = 0 }; 1541 struct mptcp_pm_addr_entry *entry; 1542 1543 list_for_each_entry(entry, rm_list, list) { 1544 if (lookup_subflow_by_saddr(&msk->conn_list, &entry->addr) && 1545 slist.nr < MPTCP_RM_IDS_MAX) 1546 slist.ids[slist.nr++] = entry->addr.id; 1547 1548 if (remove_anno_list_by_saddr(msk, &entry->addr) && 1549 alist.nr < MPTCP_RM_IDS_MAX) 1550 alist.ids[alist.nr++] = entry->addr.id; 1551 } 1552 1553 if (alist.nr) { 1554 spin_lock_bh(&msk->pm.lock); 1555 mptcp_pm_remove_addr(msk, &alist); 1556 spin_unlock_bh(&msk->pm.lock); 1557 } 1558 if (slist.nr) 1559 mptcp_pm_remove_subflow(msk, &slist); 1560 } 1561 1562 static void mptcp_nl_remove_addrs_list(struct net *net, 1563 struct list_head *rm_list) 1564 { 1565 long s_slot = 0, s_num = 0; 1566 struct mptcp_sock *msk; 1567 1568 if (list_empty(rm_list)) 1569 return; 1570 1571 while ((msk = mptcp_token_iter_next(net, &s_slot, &s_num)) != NULL) { 1572 struct sock *sk = (struct sock *)msk; 1573 1574 if (!mptcp_pm_is_userspace(msk)) { 1575 lock_sock(sk); 1576 mptcp_pm_remove_addrs_and_subflows(msk, rm_list); 1577 release_sock(sk); 1578 } 1579 1580 sock_put(sk); 1581 cond_resched(); 1582 } 1583 } 1584 1585 /* caller must ensure the RCU grace period is already elapsed */ 1586 static void __flush_addrs(struct list_head *list) 1587 { 1588 while (!list_empty(list)) { 1589 struct mptcp_pm_addr_entry *cur; 1590 1591 cur = list_entry(list->next, 1592 struct mptcp_pm_addr_entry, list); 1593 list_del_rcu(&cur->list); 1594 __mptcp_pm_release_addr_entry(cur); 1595 } 1596 } 1597 1598 static void __reset_counters(struct pm_nl_pernet *pernet) 1599 { 1600 WRITE_ONCE(pernet->add_addr_signal_max, 0); 1601 WRITE_ONCE(pernet->add_addr_accept_max, 0); 1602 WRITE_ONCE(pernet->local_addr_max, 0); 1603 pernet->addrs = 0; 1604 } 1605 1606 static int mptcp_nl_cmd_flush_addrs(struct sk_buff *skb, struct genl_info *info) 1607 { 1608 struct pm_nl_pernet *pernet = genl_info_pm_nl(info); 1609 LIST_HEAD(free_list); 1610 1611 spin_lock_bh(&pernet->lock); 1612 list_splice_init(&pernet->local_addr_list, &free_list); 1613 __reset_counters(pernet); 1614 pernet->next_id = 1; 1615 bitmap_zero(pernet->id_bitmap, MPTCP_PM_MAX_ADDR_ID + 1); 1616 spin_unlock_bh(&pernet->lock); 1617 mptcp_nl_remove_addrs_list(sock_net(skb->sk), &free_list); 1618 synchronize_rcu(); 1619 __flush_addrs(&free_list); 1620 return 0; 1621 } 1622 1623 static int mptcp_nl_fill_addr(struct sk_buff *skb, 1624 struct mptcp_pm_addr_entry *entry) 1625 { 1626 struct mptcp_addr_info *addr = &entry->addr; 1627 struct nlattr *attr; 1628 1629 attr = nla_nest_start(skb, MPTCP_PM_ATTR_ADDR); 1630 if (!attr) 1631 return -EMSGSIZE; 1632 1633 if (nla_put_u16(skb, MPTCP_PM_ADDR_ATTR_FAMILY, addr->family)) 1634 goto nla_put_failure; 1635 if (nla_put_u16(skb, MPTCP_PM_ADDR_ATTR_PORT, ntohs(addr->port))) 1636 goto nla_put_failure; 1637 if (nla_put_u8(skb, MPTCP_PM_ADDR_ATTR_ID, addr->id)) 1638 goto nla_put_failure; 1639 if (nla_put_u32(skb, MPTCP_PM_ADDR_ATTR_FLAGS, entry->flags)) 1640 goto nla_put_failure; 1641 if (entry->ifindex && 1642 nla_put_s32(skb, MPTCP_PM_ADDR_ATTR_IF_IDX, entry->ifindex)) 1643 goto nla_put_failure; 1644 1645 if (addr->family == AF_INET && 1646 nla_put_in_addr(skb, MPTCP_PM_ADDR_ATTR_ADDR4, 1647 addr->addr.s_addr)) 1648 goto nla_put_failure; 1649 #if IS_ENABLED(CONFIG_MPTCP_IPV6) 1650 else if (addr->family == AF_INET6 && 1651 nla_put_in6_addr(skb, MPTCP_PM_ADDR_ATTR_ADDR6, &addr->addr6)) 1652 goto nla_put_failure; 1653 #endif 1654 nla_nest_end(skb, attr); 1655 return 0; 1656 1657 nla_put_failure: 1658 nla_nest_cancel(skb, attr); 1659 return -EMSGSIZE; 1660 } 1661 1662 static int mptcp_nl_cmd_get_addr(struct sk_buff *skb, struct genl_info *info) 1663 { 1664 struct nlattr *attr = info->attrs[MPTCP_PM_ATTR_ADDR]; 1665 struct pm_nl_pernet *pernet = genl_info_pm_nl(info); 1666 struct mptcp_pm_addr_entry addr, *entry; 1667 struct sk_buff *msg; 1668 void *reply; 1669 int ret; 1670 1671 ret = mptcp_pm_parse_entry(attr, info, false, &addr); 1672 if (ret < 0) 1673 return ret; 1674 1675 msg = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL); 1676 if (!msg) 1677 return -ENOMEM; 1678 1679 reply = genlmsg_put_reply(msg, info, &mptcp_genl_family, 0, 1680 info->genlhdr->cmd); 1681 if (!reply) { 1682 GENL_SET_ERR_MSG(info, "not enough space in Netlink message"); 1683 ret = -EMSGSIZE; 1684 goto fail; 1685 } 1686 1687 spin_lock_bh(&pernet->lock); 1688 entry = __lookup_addr_by_id(pernet, addr.addr.id); 1689 if (!entry) { 1690 GENL_SET_ERR_MSG(info, "address not found"); 1691 ret = -EINVAL; 1692 goto unlock_fail; 1693 } 1694 1695 ret = mptcp_nl_fill_addr(msg, entry); 1696 if (ret) 1697 goto unlock_fail; 1698 1699 genlmsg_end(msg, reply); 1700 ret = genlmsg_reply(msg, info); 1701 spin_unlock_bh(&pernet->lock); 1702 return ret; 1703 1704 unlock_fail: 1705 spin_unlock_bh(&pernet->lock); 1706 1707 fail: 1708 nlmsg_free(msg); 1709 return ret; 1710 } 1711 1712 static int mptcp_nl_cmd_dump_addrs(struct sk_buff *msg, 1713 struct netlink_callback *cb) 1714 { 1715 struct net *net = sock_net(msg->sk); 1716 struct mptcp_pm_addr_entry *entry; 1717 struct pm_nl_pernet *pernet; 1718 int id = cb->args[0]; 1719 void *hdr; 1720 int i; 1721 1722 pernet = pm_nl_get_pernet(net); 1723 1724 spin_lock_bh(&pernet->lock); 1725 for (i = id; i < MPTCP_PM_MAX_ADDR_ID + 1; i++) { 1726 if (test_bit(i, pernet->id_bitmap)) { 1727 entry = __lookup_addr_by_id(pernet, i); 1728 if (!entry) 1729 break; 1730 1731 if (entry->addr.id <= id) 1732 continue; 1733 1734 hdr = genlmsg_put(msg, NETLINK_CB(cb->skb).portid, 1735 cb->nlh->nlmsg_seq, &mptcp_genl_family, 1736 NLM_F_MULTI, MPTCP_PM_CMD_GET_ADDR); 1737 if (!hdr) 1738 break; 1739 1740 if (mptcp_nl_fill_addr(msg, entry) < 0) { 1741 genlmsg_cancel(msg, hdr); 1742 break; 1743 } 1744 1745 id = entry->addr.id; 1746 genlmsg_end(msg, hdr); 1747 } 1748 } 1749 spin_unlock_bh(&pernet->lock); 1750 1751 cb->args[0] = id; 1752 return msg->len; 1753 } 1754 1755 static int parse_limit(struct genl_info *info, int id, unsigned int *limit) 1756 { 1757 struct nlattr *attr = info->attrs[id]; 1758 1759 if (!attr) 1760 return 0; 1761 1762 *limit = nla_get_u32(attr); 1763 if (*limit > MPTCP_PM_ADDR_MAX) { 1764 GENL_SET_ERR_MSG(info, "limit greater than maximum"); 1765 return -EINVAL; 1766 } 1767 return 0; 1768 } 1769 1770 static int 1771 mptcp_nl_cmd_set_limits(struct sk_buff *skb, struct genl_info *info) 1772 { 1773 struct pm_nl_pernet *pernet = genl_info_pm_nl(info); 1774 unsigned int rcv_addrs, subflows; 1775 int ret; 1776 1777 spin_lock_bh(&pernet->lock); 1778 rcv_addrs = pernet->add_addr_accept_max; 1779 ret = parse_limit(info, MPTCP_PM_ATTR_RCV_ADD_ADDRS, &rcv_addrs); 1780 if (ret) 1781 goto unlock; 1782 1783 subflows = pernet->subflows_max; 1784 ret = parse_limit(info, MPTCP_PM_ATTR_SUBFLOWS, &subflows); 1785 if (ret) 1786 goto unlock; 1787 1788 WRITE_ONCE(pernet->add_addr_accept_max, rcv_addrs); 1789 WRITE_ONCE(pernet->subflows_max, subflows); 1790 1791 unlock: 1792 spin_unlock_bh(&pernet->lock); 1793 return ret; 1794 } 1795 1796 static int 1797 mptcp_nl_cmd_get_limits(struct sk_buff *skb, struct genl_info *info) 1798 { 1799 struct pm_nl_pernet *pernet = genl_info_pm_nl(info); 1800 struct sk_buff *msg; 1801 void *reply; 1802 1803 msg = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL); 1804 if (!msg) 1805 return -ENOMEM; 1806 1807 reply = genlmsg_put_reply(msg, info, &mptcp_genl_family, 0, 1808 MPTCP_PM_CMD_GET_LIMITS); 1809 if (!reply) 1810 goto fail; 1811 1812 if (nla_put_u32(msg, MPTCP_PM_ATTR_RCV_ADD_ADDRS, 1813 READ_ONCE(pernet->add_addr_accept_max))) 1814 goto fail; 1815 1816 if (nla_put_u32(msg, MPTCP_PM_ATTR_SUBFLOWS, 1817 READ_ONCE(pernet->subflows_max))) 1818 goto fail; 1819 1820 genlmsg_end(msg, reply); 1821 return genlmsg_reply(msg, info); 1822 1823 fail: 1824 GENL_SET_ERR_MSG(info, "not enough space in Netlink message"); 1825 nlmsg_free(msg); 1826 return -EMSGSIZE; 1827 } 1828 1829 static void mptcp_pm_nl_fullmesh(struct mptcp_sock *msk, 1830 struct mptcp_addr_info *addr) 1831 { 1832 struct mptcp_rm_list list = { .nr = 0 }; 1833 1834 list.ids[list.nr++] = addr->id; 1835 1836 spin_lock_bh(&msk->pm.lock); 1837 mptcp_pm_nl_rm_subflow_received(msk, &list); 1838 mptcp_pm_create_subflow_or_signal_addr(msk); 1839 spin_unlock_bh(&msk->pm.lock); 1840 } 1841 1842 static int mptcp_nl_set_flags(struct net *net, 1843 struct mptcp_addr_info *addr, 1844 u8 bkup, u8 changed) 1845 { 1846 long s_slot = 0, s_num = 0; 1847 struct mptcp_sock *msk; 1848 int ret = -EINVAL; 1849 1850 while ((msk = mptcp_token_iter_next(net, &s_slot, &s_num)) != NULL) { 1851 struct sock *sk = (struct sock *)msk; 1852 1853 if (list_empty(&msk->conn_list) || mptcp_pm_is_userspace(msk)) 1854 goto next; 1855 1856 lock_sock(sk); 1857 if (changed & MPTCP_PM_ADDR_FLAG_BACKUP) 1858 ret = mptcp_pm_nl_mp_prio_send_ack(msk, addr, NULL, bkup); 1859 if (changed & MPTCP_PM_ADDR_FLAG_FULLMESH) 1860 mptcp_pm_nl_fullmesh(msk, addr); 1861 release_sock(sk); 1862 1863 next: 1864 sock_put(sk); 1865 cond_resched(); 1866 } 1867 1868 return ret; 1869 } 1870 1871 static int mptcp_nl_cmd_set_flags(struct sk_buff *skb, struct genl_info *info) 1872 { 1873 struct mptcp_pm_addr_entry addr = { .addr = { .family = AF_UNSPEC }, }, *entry; 1874 struct mptcp_pm_addr_entry remote = { .addr = { .family = AF_UNSPEC }, }; 1875 struct nlattr *attr_rem = info->attrs[MPTCP_PM_ATTR_ADDR_REMOTE]; 1876 struct nlattr *token = info->attrs[MPTCP_PM_ATTR_TOKEN]; 1877 struct nlattr *attr = info->attrs[MPTCP_PM_ATTR_ADDR]; 1878 struct pm_nl_pernet *pernet = genl_info_pm_nl(info); 1879 u8 changed, mask = MPTCP_PM_ADDR_FLAG_BACKUP | 1880 MPTCP_PM_ADDR_FLAG_FULLMESH; 1881 struct net *net = sock_net(skb->sk); 1882 u8 bkup = 0, lookup_by_id = 0; 1883 int ret; 1884 1885 ret = mptcp_pm_parse_entry(attr, info, false, &addr); 1886 if (ret < 0) 1887 return ret; 1888 1889 if (attr_rem) { 1890 ret = mptcp_pm_parse_entry(attr_rem, info, false, &remote); 1891 if (ret < 0) 1892 return ret; 1893 } 1894 1895 if (addr.flags & MPTCP_PM_ADDR_FLAG_BACKUP) 1896 bkup = 1; 1897 if (addr.addr.family == AF_UNSPEC) { 1898 lookup_by_id = 1; 1899 if (!addr.addr.id) 1900 return -EOPNOTSUPP; 1901 } 1902 1903 if (token) 1904 return mptcp_userspace_pm_set_flags(sock_net(skb->sk), 1905 token, &addr, &remote, bkup); 1906 1907 spin_lock_bh(&pernet->lock); 1908 entry = __lookup_addr(pernet, &addr.addr, lookup_by_id); 1909 if (!entry) { 1910 spin_unlock_bh(&pernet->lock); 1911 return -EINVAL; 1912 } 1913 if ((addr.flags & MPTCP_PM_ADDR_FLAG_FULLMESH) && 1914 (entry->flags & MPTCP_PM_ADDR_FLAG_SIGNAL)) { 1915 spin_unlock_bh(&pernet->lock); 1916 return -EINVAL; 1917 } 1918 1919 changed = (addr.flags ^ entry->flags) & mask; 1920 entry->flags = (entry->flags & ~mask) | (addr.flags & mask); 1921 addr = *entry; 1922 spin_unlock_bh(&pernet->lock); 1923 1924 mptcp_nl_set_flags(net, &addr.addr, bkup, changed); 1925 return 0; 1926 } 1927 1928 static void mptcp_nl_mcast_send(struct net *net, struct sk_buff *nlskb, gfp_t gfp) 1929 { 1930 genlmsg_multicast_netns(&mptcp_genl_family, net, 1931 nlskb, 0, MPTCP_PM_EV_GRP_OFFSET, gfp); 1932 } 1933 1934 bool mptcp_userspace_pm_active(const struct mptcp_sock *msk) 1935 { 1936 return genl_has_listeners(&mptcp_genl_family, 1937 sock_net((const struct sock *)msk), 1938 MPTCP_PM_EV_GRP_OFFSET); 1939 } 1940 1941 static int mptcp_event_add_subflow(struct sk_buff *skb, const struct sock *ssk) 1942 { 1943 const struct inet_sock *issk = inet_sk(ssk); 1944 const struct mptcp_subflow_context *sf; 1945 1946 if (nla_put_u16(skb, MPTCP_ATTR_FAMILY, ssk->sk_family)) 1947 return -EMSGSIZE; 1948 1949 switch (ssk->sk_family) { 1950 case AF_INET: 1951 if (nla_put_in_addr(skb, MPTCP_ATTR_SADDR4, issk->inet_saddr)) 1952 return -EMSGSIZE; 1953 if (nla_put_in_addr(skb, MPTCP_ATTR_DADDR4, issk->inet_daddr)) 1954 return -EMSGSIZE; 1955 break; 1956 #if IS_ENABLED(CONFIG_MPTCP_IPV6) 1957 case AF_INET6: { 1958 const struct ipv6_pinfo *np = inet6_sk(ssk); 1959 1960 if (nla_put_in6_addr(skb, MPTCP_ATTR_SADDR6, &np->saddr)) 1961 return -EMSGSIZE; 1962 if (nla_put_in6_addr(skb, MPTCP_ATTR_DADDR6, &ssk->sk_v6_daddr)) 1963 return -EMSGSIZE; 1964 break; 1965 } 1966 #endif 1967 default: 1968 WARN_ON_ONCE(1); 1969 return -EMSGSIZE; 1970 } 1971 1972 if (nla_put_be16(skb, MPTCP_ATTR_SPORT, issk->inet_sport)) 1973 return -EMSGSIZE; 1974 if (nla_put_be16(skb, MPTCP_ATTR_DPORT, issk->inet_dport)) 1975 return -EMSGSIZE; 1976 1977 sf = mptcp_subflow_ctx(ssk); 1978 if (WARN_ON_ONCE(!sf)) 1979 return -EINVAL; 1980 1981 if (nla_put_u8(skb, MPTCP_ATTR_LOC_ID, sf->local_id)) 1982 return -EMSGSIZE; 1983 1984 if (nla_put_u8(skb, MPTCP_ATTR_REM_ID, sf->remote_id)) 1985 return -EMSGSIZE; 1986 1987 return 0; 1988 } 1989 1990 static int mptcp_event_put_token_and_ssk(struct sk_buff *skb, 1991 const struct mptcp_sock *msk, 1992 const struct sock *ssk) 1993 { 1994 const struct sock *sk = (const struct sock *)msk; 1995 const struct mptcp_subflow_context *sf; 1996 u8 sk_err; 1997 1998 if (nla_put_u32(skb, MPTCP_ATTR_TOKEN, msk->token)) 1999 return -EMSGSIZE; 2000 2001 if (mptcp_event_add_subflow(skb, ssk)) 2002 return -EMSGSIZE; 2003 2004 sf = mptcp_subflow_ctx(ssk); 2005 if (WARN_ON_ONCE(!sf)) 2006 return -EINVAL; 2007 2008 if (nla_put_u8(skb, MPTCP_ATTR_BACKUP, sf->backup)) 2009 return -EMSGSIZE; 2010 2011 if (ssk->sk_bound_dev_if && 2012 nla_put_s32(skb, MPTCP_ATTR_IF_IDX, ssk->sk_bound_dev_if)) 2013 return -EMSGSIZE; 2014 2015 sk_err = ssk->sk_err; 2016 if (sk_err && sk->sk_state == TCP_ESTABLISHED && 2017 nla_put_u8(skb, MPTCP_ATTR_ERROR, sk_err)) 2018 return -EMSGSIZE; 2019 2020 return 0; 2021 } 2022 2023 static int mptcp_event_sub_established(struct sk_buff *skb, 2024 const struct mptcp_sock *msk, 2025 const struct sock *ssk) 2026 { 2027 return mptcp_event_put_token_and_ssk(skb, msk, ssk); 2028 } 2029 2030 static int mptcp_event_sub_closed(struct sk_buff *skb, 2031 const struct mptcp_sock *msk, 2032 const struct sock *ssk) 2033 { 2034 const struct mptcp_subflow_context *sf; 2035 2036 if (mptcp_event_put_token_and_ssk(skb, msk, ssk)) 2037 return -EMSGSIZE; 2038 2039 sf = mptcp_subflow_ctx(ssk); 2040 if (!sf->reset_seen) 2041 return 0; 2042 2043 if (nla_put_u32(skb, MPTCP_ATTR_RESET_REASON, sf->reset_reason)) 2044 return -EMSGSIZE; 2045 2046 if (nla_put_u32(skb, MPTCP_ATTR_RESET_FLAGS, sf->reset_transient)) 2047 return -EMSGSIZE; 2048 2049 return 0; 2050 } 2051 2052 static int mptcp_event_created(struct sk_buff *skb, 2053 const struct mptcp_sock *msk, 2054 const struct sock *ssk) 2055 { 2056 int err = nla_put_u32(skb, MPTCP_ATTR_TOKEN, msk->token); 2057 2058 if (err) 2059 return err; 2060 2061 if (nla_put_u8(skb, MPTCP_ATTR_SERVER_SIDE, READ_ONCE(msk->pm.server_side))) 2062 return -EMSGSIZE; 2063 2064 return mptcp_event_add_subflow(skb, ssk); 2065 } 2066 2067 void mptcp_event_addr_removed(const struct mptcp_sock *msk, uint8_t id) 2068 { 2069 struct net *net = sock_net((const struct sock *)msk); 2070 struct nlmsghdr *nlh; 2071 struct sk_buff *skb; 2072 2073 if (!genl_has_listeners(&mptcp_genl_family, net, MPTCP_PM_EV_GRP_OFFSET)) 2074 return; 2075 2076 skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_ATOMIC); 2077 if (!skb) 2078 return; 2079 2080 nlh = genlmsg_put(skb, 0, 0, &mptcp_genl_family, 0, MPTCP_EVENT_REMOVED); 2081 if (!nlh) 2082 goto nla_put_failure; 2083 2084 if (nla_put_u32(skb, MPTCP_ATTR_TOKEN, msk->token)) 2085 goto nla_put_failure; 2086 2087 if (nla_put_u8(skb, MPTCP_ATTR_REM_ID, id)) 2088 goto nla_put_failure; 2089 2090 genlmsg_end(skb, nlh); 2091 mptcp_nl_mcast_send(net, skb, GFP_ATOMIC); 2092 return; 2093 2094 nla_put_failure: 2095 kfree_skb(skb); 2096 } 2097 2098 void mptcp_event_addr_announced(const struct sock *ssk, 2099 const struct mptcp_addr_info *info) 2100 { 2101 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk); 2102 struct mptcp_sock *msk = mptcp_sk(subflow->conn); 2103 struct net *net = sock_net(ssk); 2104 struct nlmsghdr *nlh; 2105 struct sk_buff *skb; 2106 2107 if (!genl_has_listeners(&mptcp_genl_family, net, MPTCP_PM_EV_GRP_OFFSET)) 2108 return; 2109 2110 skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_ATOMIC); 2111 if (!skb) 2112 return; 2113 2114 nlh = genlmsg_put(skb, 0, 0, &mptcp_genl_family, 0, 2115 MPTCP_EVENT_ANNOUNCED); 2116 if (!nlh) 2117 goto nla_put_failure; 2118 2119 if (nla_put_u32(skb, MPTCP_ATTR_TOKEN, msk->token)) 2120 goto nla_put_failure; 2121 2122 if (nla_put_u8(skb, MPTCP_ATTR_REM_ID, info->id)) 2123 goto nla_put_failure; 2124 2125 if (nla_put_be16(skb, MPTCP_ATTR_DPORT, 2126 info->port == 0 ? 2127 inet_sk(ssk)->inet_dport : 2128 info->port)) 2129 goto nla_put_failure; 2130 2131 switch (info->family) { 2132 case AF_INET: 2133 if (nla_put_in_addr(skb, MPTCP_ATTR_DADDR4, info->addr.s_addr)) 2134 goto nla_put_failure; 2135 break; 2136 #if IS_ENABLED(CONFIG_MPTCP_IPV6) 2137 case AF_INET6: 2138 if (nla_put_in6_addr(skb, MPTCP_ATTR_DADDR6, &info->addr6)) 2139 goto nla_put_failure; 2140 break; 2141 #endif 2142 default: 2143 WARN_ON_ONCE(1); 2144 goto nla_put_failure; 2145 } 2146 2147 genlmsg_end(skb, nlh); 2148 mptcp_nl_mcast_send(net, skb, GFP_ATOMIC); 2149 return; 2150 2151 nla_put_failure: 2152 kfree_skb(skb); 2153 } 2154 2155 void mptcp_event(enum mptcp_event_type type, const struct mptcp_sock *msk, 2156 const struct sock *ssk, gfp_t gfp) 2157 { 2158 struct net *net = sock_net((const struct sock *)msk); 2159 struct nlmsghdr *nlh; 2160 struct sk_buff *skb; 2161 2162 if (!genl_has_listeners(&mptcp_genl_family, net, MPTCP_PM_EV_GRP_OFFSET)) 2163 return; 2164 2165 skb = nlmsg_new(NLMSG_DEFAULT_SIZE, gfp); 2166 if (!skb) 2167 return; 2168 2169 nlh = genlmsg_put(skb, 0, 0, &mptcp_genl_family, 0, type); 2170 if (!nlh) 2171 goto nla_put_failure; 2172 2173 switch (type) { 2174 case MPTCP_EVENT_UNSPEC: 2175 WARN_ON_ONCE(1); 2176 break; 2177 case MPTCP_EVENT_CREATED: 2178 case MPTCP_EVENT_ESTABLISHED: 2179 if (mptcp_event_created(skb, msk, ssk) < 0) 2180 goto nla_put_failure; 2181 break; 2182 case MPTCP_EVENT_CLOSED: 2183 if (nla_put_u32(skb, MPTCP_ATTR_TOKEN, msk->token) < 0) 2184 goto nla_put_failure; 2185 break; 2186 case MPTCP_EVENT_ANNOUNCED: 2187 case MPTCP_EVENT_REMOVED: 2188 /* call mptcp_event_addr_announced()/removed instead */ 2189 WARN_ON_ONCE(1); 2190 break; 2191 case MPTCP_EVENT_SUB_ESTABLISHED: 2192 case MPTCP_EVENT_SUB_PRIORITY: 2193 if (mptcp_event_sub_established(skb, msk, ssk) < 0) 2194 goto nla_put_failure; 2195 break; 2196 case MPTCP_EVENT_SUB_CLOSED: 2197 if (mptcp_event_sub_closed(skb, msk, ssk) < 0) 2198 goto nla_put_failure; 2199 break; 2200 } 2201 2202 genlmsg_end(skb, nlh); 2203 mptcp_nl_mcast_send(net, skb, gfp); 2204 return; 2205 2206 nla_put_failure: 2207 kfree_skb(skb); 2208 } 2209 2210 static const struct genl_small_ops mptcp_pm_ops[] = { 2211 { 2212 .cmd = MPTCP_PM_CMD_ADD_ADDR, 2213 .doit = mptcp_nl_cmd_add_addr, 2214 .flags = GENL_UNS_ADMIN_PERM, 2215 }, 2216 { 2217 .cmd = MPTCP_PM_CMD_DEL_ADDR, 2218 .doit = mptcp_nl_cmd_del_addr, 2219 .flags = GENL_UNS_ADMIN_PERM, 2220 }, 2221 { 2222 .cmd = MPTCP_PM_CMD_FLUSH_ADDRS, 2223 .doit = mptcp_nl_cmd_flush_addrs, 2224 .flags = GENL_UNS_ADMIN_PERM, 2225 }, 2226 { 2227 .cmd = MPTCP_PM_CMD_GET_ADDR, 2228 .doit = mptcp_nl_cmd_get_addr, 2229 .dumpit = mptcp_nl_cmd_dump_addrs, 2230 }, 2231 { 2232 .cmd = MPTCP_PM_CMD_SET_LIMITS, 2233 .doit = mptcp_nl_cmd_set_limits, 2234 .flags = GENL_UNS_ADMIN_PERM, 2235 }, 2236 { 2237 .cmd = MPTCP_PM_CMD_GET_LIMITS, 2238 .doit = mptcp_nl_cmd_get_limits, 2239 }, 2240 { 2241 .cmd = MPTCP_PM_CMD_SET_FLAGS, 2242 .doit = mptcp_nl_cmd_set_flags, 2243 .flags = GENL_UNS_ADMIN_PERM, 2244 }, 2245 { 2246 .cmd = MPTCP_PM_CMD_ANNOUNCE, 2247 .doit = mptcp_nl_cmd_announce, 2248 .flags = GENL_UNS_ADMIN_PERM, 2249 }, 2250 { 2251 .cmd = MPTCP_PM_CMD_REMOVE, 2252 .doit = mptcp_nl_cmd_remove, 2253 .flags = GENL_UNS_ADMIN_PERM, 2254 }, 2255 { 2256 .cmd = MPTCP_PM_CMD_SUBFLOW_CREATE, 2257 .doit = mptcp_nl_cmd_sf_create, 2258 .flags = GENL_UNS_ADMIN_PERM, 2259 }, 2260 { 2261 .cmd = MPTCP_PM_CMD_SUBFLOW_DESTROY, 2262 .doit = mptcp_nl_cmd_sf_destroy, 2263 .flags = GENL_UNS_ADMIN_PERM, 2264 }, 2265 }; 2266 2267 static struct genl_family mptcp_genl_family __ro_after_init = { 2268 .name = MPTCP_PM_NAME, 2269 .version = MPTCP_PM_VER, 2270 .maxattr = MPTCP_PM_ATTR_MAX, 2271 .policy = mptcp_pm_policy, 2272 .netnsok = true, 2273 .module = THIS_MODULE, 2274 .small_ops = mptcp_pm_ops, 2275 .n_small_ops = ARRAY_SIZE(mptcp_pm_ops), 2276 .resv_start_op = MPTCP_PM_CMD_SUBFLOW_DESTROY + 1, 2277 .mcgrps = mptcp_pm_mcgrps, 2278 .n_mcgrps = ARRAY_SIZE(mptcp_pm_mcgrps), 2279 }; 2280 2281 static int __net_init pm_nl_init_net(struct net *net) 2282 { 2283 struct pm_nl_pernet *pernet = pm_nl_get_pernet(net); 2284 2285 INIT_LIST_HEAD_RCU(&pernet->local_addr_list); 2286 2287 /* Cit. 2 subflows ought to be enough for anybody. */ 2288 pernet->subflows_max = 2; 2289 pernet->next_id = 1; 2290 pernet->stale_loss_cnt = 4; 2291 spin_lock_init(&pernet->lock); 2292 2293 /* No need to initialize other pernet fields, the struct is zeroed at 2294 * allocation time. 2295 */ 2296 2297 return 0; 2298 } 2299 2300 static void __net_exit pm_nl_exit_net(struct list_head *net_list) 2301 { 2302 struct net *net; 2303 2304 list_for_each_entry(net, net_list, exit_list) { 2305 struct pm_nl_pernet *pernet = pm_nl_get_pernet(net); 2306 2307 /* net is removed from namespace list, can't race with 2308 * other modifiers, also netns core already waited for a 2309 * RCU grace period. 2310 */ 2311 __flush_addrs(&pernet->local_addr_list); 2312 } 2313 } 2314 2315 static struct pernet_operations mptcp_pm_pernet_ops = { 2316 .init = pm_nl_init_net, 2317 .exit_batch = pm_nl_exit_net, 2318 .id = &pm_nl_pernet_id, 2319 .size = sizeof(struct pm_nl_pernet), 2320 }; 2321 2322 void __init mptcp_pm_nl_init(void) 2323 { 2324 if (register_pernet_subsys(&mptcp_pm_pernet_ops) < 0) 2325 panic("Failed to register MPTCP PM pernet subsystem.\n"); 2326 2327 if (genl_register_family(&mptcp_genl_family)) 2328 panic("Failed to register MPTCP PM netlink family\n"); 2329 } 2330