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/inet_common.h> 13 #include <net/netns/generic.h> 14 #include <net/mptcp.h> 15 #include <net/genetlink.h> 16 #include <uapi/linux/mptcp.h> 17 18 #include "protocol.h" 19 #include "mib.h" 20 21 /* forward declaration */ 22 static struct genl_family mptcp_genl_family; 23 24 static int pm_nl_pernet_id; 25 26 struct mptcp_pm_add_entry { 27 struct list_head list; 28 struct mptcp_addr_info addr; 29 u8 retrans_times; 30 struct timer_list add_timer; 31 struct mptcp_sock *sock; 32 }; 33 34 struct pm_nl_pernet { 35 /* protects pernet updates */ 36 spinlock_t lock; 37 struct list_head local_addr_list; 38 unsigned int addrs; 39 unsigned int stale_loss_cnt; 40 unsigned int add_addr_signal_max; 41 unsigned int add_addr_accept_max; 42 unsigned int local_addr_max; 43 unsigned int subflows_max; 44 unsigned int next_id; 45 DECLARE_BITMAP(id_bitmap, MPTCP_PM_MAX_ADDR_ID + 1); 46 }; 47 48 #define MPTCP_PM_ADDR_MAX 8 49 #define ADD_ADDR_RETRANS_MAX 3 50 51 static struct pm_nl_pernet *pm_nl_get_pernet(const struct net *net) 52 { 53 return net_generic(net, pm_nl_pernet_id); 54 } 55 56 static struct pm_nl_pernet * 57 pm_nl_get_pernet_from_msk(const struct mptcp_sock *msk) 58 { 59 return pm_nl_get_pernet(sock_net((struct sock *)msk)); 60 } 61 62 bool mptcp_addresses_equal(const struct mptcp_addr_info *a, 63 const struct mptcp_addr_info *b, bool use_port) 64 { 65 bool addr_equals = false; 66 67 if (a->family == b->family) { 68 if (a->family == AF_INET) 69 addr_equals = a->addr.s_addr == b->addr.s_addr; 70 #if IS_ENABLED(CONFIG_MPTCP_IPV6) 71 else 72 addr_equals = !ipv6_addr_cmp(&a->addr6, &b->addr6); 73 } else if (a->family == AF_INET) { 74 if (ipv6_addr_v4mapped(&b->addr6)) 75 addr_equals = a->addr.s_addr == b->addr6.s6_addr32[3]; 76 } else if (b->family == AF_INET) { 77 if (ipv6_addr_v4mapped(&a->addr6)) 78 addr_equals = a->addr6.s6_addr32[3] == b->addr.s_addr; 79 #endif 80 } 81 82 if (!addr_equals) 83 return false; 84 if (!use_port) 85 return true; 86 87 return a->port == b->port; 88 } 89 90 void mptcp_local_address(const struct sock_common *skc, 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 mptcp_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 struct mptcp_pm_addr_entry *entry, *ret = NULL; 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 ret = entry; 168 break; 169 } 170 rcu_read_unlock(); 171 return ret; 172 } 173 174 static struct mptcp_pm_addr_entry * 175 select_signal_address(struct pm_nl_pernet *pernet, const struct mptcp_sock *msk) 176 { 177 struct mptcp_pm_addr_entry *entry, *ret = NULL; 178 179 rcu_read_lock(); 180 /* do not keep any additional per socket state, just signal 181 * the address list in order. 182 * Note: removal from the local address list during the msk life-cycle 183 * can lead to additional addresses not being announced. 184 */ 185 list_for_each_entry_rcu(entry, &pernet->local_addr_list, list) { 186 if (!test_bit(entry->addr.id, msk->pm.id_avail_bitmap)) 187 continue; 188 189 if (!(entry->flags & MPTCP_PM_ADDR_FLAG_SIGNAL)) 190 continue; 191 192 ret = entry; 193 break; 194 } 195 rcu_read_unlock(); 196 return ret; 197 } 198 199 unsigned int mptcp_pm_get_add_addr_signal_max(const struct mptcp_sock *msk) 200 { 201 const struct pm_nl_pernet *pernet = pm_nl_get_pernet_from_msk(msk); 202 203 return READ_ONCE(pernet->add_addr_signal_max); 204 } 205 EXPORT_SYMBOL_GPL(mptcp_pm_get_add_addr_signal_max); 206 207 unsigned int mptcp_pm_get_add_addr_accept_max(const struct mptcp_sock *msk) 208 { 209 struct pm_nl_pernet *pernet = pm_nl_get_pernet_from_msk(msk); 210 211 return READ_ONCE(pernet->add_addr_accept_max); 212 } 213 EXPORT_SYMBOL_GPL(mptcp_pm_get_add_addr_accept_max); 214 215 unsigned int mptcp_pm_get_subflows_max(const struct mptcp_sock *msk) 216 { 217 struct pm_nl_pernet *pernet = pm_nl_get_pernet_from_msk(msk); 218 219 return READ_ONCE(pernet->subflows_max); 220 } 221 EXPORT_SYMBOL_GPL(mptcp_pm_get_subflows_max); 222 223 unsigned int mptcp_pm_get_local_addr_max(const struct mptcp_sock *msk) 224 { 225 struct pm_nl_pernet *pernet = pm_nl_get_pernet_from_msk(msk); 226 227 return READ_ONCE(pernet->local_addr_max); 228 } 229 EXPORT_SYMBOL_GPL(mptcp_pm_get_local_addr_max); 230 231 bool mptcp_pm_nl_check_work_pending(struct mptcp_sock *msk) 232 { 233 struct pm_nl_pernet *pernet = pm_nl_get_pernet_from_msk(msk); 234 235 if (msk->pm.subflows == mptcp_pm_get_subflows_max(msk) || 236 (find_next_and_bit(pernet->id_bitmap, msk->pm.id_avail_bitmap, 237 MPTCP_PM_MAX_ADDR_ID + 1, 0) == MPTCP_PM_MAX_ADDR_ID + 1)) { 238 WRITE_ONCE(msk->pm.work_pending, false); 239 return false; 240 } 241 return true; 242 } 243 244 struct mptcp_pm_add_entry * 245 mptcp_lookup_anno_list_by_saddr(const struct mptcp_sock *msk, 246 const struct mptcp_addr_info *addr) 247 { 248 struct mptcp_pm_add_entry *entry; 249 250 lockdep_assert_held(&msk->pm.lock); 251 252 list_for_each_entry(entry, &msk->pm.anno_list, list) { 253 if (mptcp_addresses_equal(&entry->addr, addr, true)) 254 return entry; 255 } 256 257 return NULL; 258 } 259 260 bool mptcp_pm_sport_in_anno_list(struct mptcp_sock *msk, const struct sock *sk) 261 { 262 struct mptcp_pm_add_entry *entry; 263 struct mptcp_addr_info saddr; 264 bool ret = false; 265 266 mptcp_local_address((struct sock_common *)sk, &saddr); 267 268 spin_lock_bh(&msk->pm.lock); 269 list_for_each_entry(entry, &msk->pm.anno_list, list) { 270 if (mptcp_addresses_equal(&entry->addr, &saddr, true)) { 271 ret = true; 272 goto out; 273 } 274 } 275 276 out: 277 spin_unlock_bh(&msk->pm.lock); 278 return ret; 279 } 280 281 static void mptcp_pm_add_timer(struct timer_list *timer) 282 { 283 struct mptcp_pm_add_entry *entry = from_timer(entry, timer, add_timer); 284 struct mptcp_sock *msk = entry->sock; 285 struct sock *sk = (struct sock *)msk; 286 287 pr_debug("msk=%p", msk); 288 289 if (!msk) 290 return; 291 292 if (inet_sk_state_load(sk) == TCP_CLOSE) 293 return; 294 295 if (!entry->addr.id) 296 return; 297 298 if (mptcp_pm_should_add_signal_addr(msk)) { 299 sk_reset_timer(sk, timer, jiffies + TCP_RTO_MAX / 8); 300 goto out; 301 } 302 303 spin_lock_bh(&msk->pm.lock); 304 305 if (!mptcp_pm_should_add_signal_addr(msk)) { 306 pr_debug("retransmit ADD_ADDR id=%d", entry->addr.id); 307 mptcp_pm_announce_addr(msk, &entry->addr, false); 308 mptcp_pm_add_addr_send_ack(msk); 309 entry->retrans_times++; 310 } 311 312 if (entry->retrans_times < ADD_ADDR_RETRANS_MAX) 313 sk_reset_timer(sk, timer, 314 jiffies + mptcp_get_add_addr_timeout(sock_net(sk))); 315 316 spin_unlock_bh(&msk->pm.lock); 317 318 if (entry->retrans_times == ADD_ADDR_RETRANS_MAX) 319 mptcp_pm_subflow_established(msk); 320 321 out: 322 __sock_put(sk); 323 } 324 325 struct mptcp_pm_add_entry * 326 mptcp_pm_del_add_timer(struct mptcp_sock *msk, 327 const struct mptcp_addr_info *addr, bool check_id) 328 { 329 struct mptcp_pm_add_entry *entry; 330 struct sock *sk = (struct sock *)msk; 331 332 spin_lock_bh(&msk->pm.lock); 333 entry = mptcp_lookup_anno_list_by_saddr(msk, addr); 334 if (entry && (!check_id || entry->addr.id == addr->id)) 335 entry->retrans_times = ADD_ADDR_RETRANS_MAX; 336 spin_unlock_bh(&msk->pm.lock); 337 338 if (entry && (!check_id || entry->addr.id == addr->id)) 339 sk_stop_timer_sync(sk, &entry->add_timer); 340 341 return entry; 342 } 343 344 bool mptcp_pm_alloc_anno_list(struct mptcp_sock *msk, 345 const struct mptcp_addr_info *addr) 346 { 347 struct mptcp_pm_add_entry *add_entry = NULL; 348 struct sock *sk = (struct sock *)msk; 349 struct net *net = sock_net(sk); 350 351 lockdep_assert_held(&msk->pm.lock); 352 353 add_entry = mptcp_lookup_anno_list_by_saddr(msk, addr); 354 355 if (add_entry) { 356 if (mptcp_pm_is_kernel(msk)) 357 return false; 358 359 sk_reset_timer(sk, &add_entry->add_timer, 360 jiffies + mptcp_get_add_addr_timeout(net)); 361 return true; 362 } 363 364 add_entry = kmalloc(sizeof(*add_entry), GFP_ATOMIC); 365 if (!add_entry) 366 return false; 367 368 list_add(&add_entry->list, &msk->pm.anno_list); 369 370 add_entry->addr = *addr; 371 add_entry->sock = msk; 372 add_entry->retrans_times = 0; 373 374 timer_setup(&add_entry->add_timer, mptcp_pm_add_timer, 0); 375 sk_reset_timer(sk, &add_entry->add_timer, 376 jiffies + mptcp_get_add_addr_timeout(net)); 377 378 return true; 379 } 380 381 void mptcp_pm_free_anno_list(struct mptcp_sock *msk) 382 { 383 struct mptcp_pm_add_entry *entry, *tmp; 384 struct sock *sk = (struct sock *)msk; 385 LIST_HEAD(free_list); 386 387 pr_debug("msk=%p", msk); 388 389 spin_lock_bh(&msk->pm.lock); 390 list_splice_init(&msk->pm.anno_list, &free_list); 391 spin_unlock_bh(&msk->pm.lock); 392 393 list_for_each_entry_safe(entry, tmp, &free_list, list) { 394 sk_stop_timer_sync(sk, &entry->add_timer); 395 kfree(entry); 396 } 397 } 398 399 static bool lookup_address_in_vec(const struct mptcp_addr_info *addrs, unsigned int nr, 400 const struct mptcp_addr_info *addr) 401 { 402 int i; 403 404 for (i = 0; i < nr; i++) { 405 if (addrs[i].id == addr->id) 406 return true; 407 } 408 409 return false; 410 } 411 412 /* Fill all the remote addresses into the array addrs[], 413 * and return the array size. 414 */ 415 static unsigned int fill_remote_addresses_vec(struct mptcp_sock *msk, 416 struct mptcp_addr_info *local, 417 bool fullmesh, 418 struct mptcp_addr_info *addrs) 419 { 420 bool deny_id0 = READ_ONCE(msk->pm.remote_deny_join_id0); 421 struct sock *sk = (struct sock *)msk, *ssk; 422 struct mptcp_subflow_context *subflow; 423 struct mptcp_addr_info remote = { 0 }; 424 unsigned int subflows_max; 425 int i = 0; 426 427 subflows_max = mptcp_pm_get_subflows_max(msk); 428 remote_address((struct sock_common *)sk, &remote); 429 430 /* Non-fullmesh endpoint, fill in the single entry 431 * corresponding to the primary MPC subflow remote address 432 */ 433 if (!fullmesh) { 434 if (deny_id0) 435 return 0; 436 437 if (!mptcp_pm_addr_families_match(sk, local, &remote)) 438 return 0; 439 440 msk->pm.subflows++; 441 addrs[i++] = remote; 442 } else { 443 mptcp_for_each_subflow(msk, subflow) { 444 ssk = mptcp_subflow_tcp_sock(subflow); 445 remote_address((struct sock_common *)ssk, &addrs[i]); 446 addrs[i].id = subflow->remote_id; 447 if (deny_id0 && !addrs[i].id) 448 continue; 449 450 if (!mptcp_pm_addr_families_match(sk, local, &addrs[i])) 451 continue; 452 453 if (!lookup_address_in_vec(addrs, i, &addrs[i]) && 454 msk->pm.subflows < subflows_max) { 455 msk->pm.subflows++; 456 i++; 457 } 458 } 459 } 460 461 return i; 462 } 463 464 static void __mptcp_pm_send_ack(struct mptcp_sock *msk, struct mptcp_subflow_context *subflow, 465 bool prio, bool backup) 466 { 467 struct sock *ssk = mptcp_subflow_tcp_sock(subflow); 468 bool slow; 469 470 pr_debug("send ack for %s", 471 prio ? "mp_prio" : (mptcp_pm_should_add_signal(msk) ? "add_addr" : "rm_addr")); 472 473 slow = lock_sock_fast(ssk); 474 if (prio) { 475 if (subflow->backup != backup) 476 msk->last_snd = NULL; 477 478 subflow->send_mp_prio = 1; 479 subflow->backup = backup; 480 subflow->request_bkup = backup; 481 } 482 483 __mptcp_subflow_send_ack(ssk); 484 unlock_sock_fast(ssk, slow); 485 } 486 487 static void mptcp_pm_send_ack(struct mptcp_sock *msk, struct mptcp_subflow_context *subflow, 488 bool prio, bool backup) 489 { 490 spin_unlock_bh(&msk->pm.lock); 491 __mptcp_pm_send_ack(msk, subflow, prio, backup); 492 spin_lock_bh(&msk->pm.lock); 493 } 494 495 static struct mptcp_pm_addr_entry * 496 __lookup_addr_by_id(struct pm_nl_pernet *pernet, unsigned int id) 497 { 498 struct mptcp_pm_addr_entry *entry; 499 500 list_for_each_entry(entry, &pernet->local_addr_list, list) { 501 if (entry->addr.id == id) 502 return entry; 503 } 504 return NULL; 505 } 506 507 static struct mptcp_pm_addr_entry * 508 __lookup_addr(struct pm_nl_pernet *pernet, const struct mptcp_addr_info *info, 509 bool lookup_by_id) 510 { 511 struct mptcp_pm_addr_entry *entry; 512 513 list_for_each_entry(entry, &pernet->local_addr_list, list) { 514 if ((!lookup_by_id && 515 mptcp_addresses_equal(&entry->addr, info, entry->addr.port)) || 516 (lookup_by_id && entry->addr.id == info->id)) 517 return entry; 518 } 519 return NULL; 520 } 521 522 static void mptcp_pm_create_subflow_or_signal_addr(struct mptcp_sock *msk) 523 { 524 struct sock *sk = (struct sock *)msk; 525 struct mptcp_pm_addr_entry *local; 526 unsigned int add_addr_signal_max; 527 unsigned int local_addr_max; 528 struct pm_nl_pernet *pernet; 529 unsigned int subflows_max; 530 531 pernet = pm_nl_get_pernet(sock_net(sk)); 532 533 add_addr_signal_max = mptcp_pm_get_add_addr_signal_max(msk); 534 local_addr_max = mptcp_pm_get_local_addr_max(msk); 535 subflows_max = mptcp_pm_get_subflows_max(msk); 536 537 /* do lazy endpoint usage accounting for the MPC subflows */ 538 if (unlikely(!(msk->pm.status & BIT(MPTCP_PM_MPC_ENDPOINT_ACCOUNTED))) && msk->first) { 539 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(msk->first); 540 struct mptcp_pm_addr_entry *entry; 541 struct mptcp_addr_info mpc_addr; 542 bool backup = false; 543 544 mptcp_local_address((struct sock_common *)msk->first, &mpc_addr); 545 rcu_read_lock(); 546 entry = __lookup_addr(pernet, &mpc_addr, false); 547 if (entry) { 548 __clear_bit(entry->addr.id, msk->pm.id_avail_bitmap); 549 msk->mpc_endpoint_id = entry->addr.id; 550 backup = !!(entry->flags & MPTCP_PM_ADDR_FLAG_BACKUP); 551 } 552 rcu_read_unlock(); 553 554 if (backup) 555 mptcp_pm_send_ack(msk, subflow, true, backup); 556 557 msk->pm.status |= BIT(MPTCP_PM_MPC_ENDPOINT_ACCOUNTED); 558 } 559 560 pr_debug("local %d:%d signal %d:%d subflows %d:%d\n", 561 msk->pm.local_addr_used, local_addr_max, 562 msk->pm.add_addr_signaled, add_addr_signal_max, 563 msk->pm.subflows, subflows_max); 564 565 /* check first for announce */ 566 if (msk->pm.add_addr_signaled < add_addr_signal_max) { 567 local = select_signal_address(pernet, msk); 568 569 /* due to racing events on both ends we can reach here while 570 * previous add address is still running: if we invoke now 571 * mptcp_pm_announce_addr(), that will fail and the 572 * corresponding id will be marked as used. 573 * Instead let the PM machinery reschedule us when the 574 * current address announce will be completed. 575 */ 576 if (msk->pm.addr_signal & BIT(MPTCP_ADD_ADDR_SIGNAL)) 577 return; 578 579 if (local) { 580 if (mptcp_pm_alloc_anno_list(msk, &local->addr)) { 581 __clear_bit(local->addr.id, msk->pm.id_avail_bitmap); 582 msk->pm.add_addr_signaled++; 583 mptcp_pm_announce_addr(msk, &local->addr, false); 584 mptcp_pm_nl_addr_send_ack(msk); 585 } 586 } 587 } 588 589 /* check if should create a new subflow */ 590 while (msk->pm.local_addr_used < local_addr_max && 591 msk->pm.subflows < subflows_max) { 592 struct mptcp_addr_info addrs[MPTCP_PM_ADDR_MAX]; 593 bool fullmesh; 594 int i, nr; 595 596 local = select_local_address(pernet, msk); 597 if (!local) 598 break; 599 600 fullmesh = !!(local->flags & MPTCP_PM_ADDR_FLAG_FULLMESH); 601 602 msk->pm.local_addr_used++; 603 __clear_bit(local->addr.id, msk->pm.id_avail_bitmap); 604 nr = fill_remote_addresses_vec(msk, &local->addr, fullmesh, addrs); 605 if (nr == 0) 606 continue; 607 608 spin_unlock_bh(&msk->pm.lock); 609 for (i = 0; i < nr; i++) 610 __mptcp_subflow_connect(sk, &local->addr, &addrs[i]); 611 spin_lock_bh(&msk->pm.lock); 612 } 613 mptcp_pm_nl_check_work_pending(msk); 614 } 615 616 static void mptcp_pm_nl_fully_established(struct mptcp_sock *msk) 617 { 618 mptcp_pm_create_subflow_or_signal_addr(msk); 619 } 620 621 static void mptcp_pm_nl_subflow_established(struct mptcp_sock *msk) 622 { 623 mptcp_pm_create_subflow_or_signal_addr(msk); 624 } 625 626 /* Fill all the local addresses into the array addrs[], 627 * and return the array size. 628 */ 629 static unsigned int fill_local_addresses_vec(struct mptcp_sock *msk, 630 struct mptcp_addr_info *remote, 631 struct mptcp_addr_info *addrs) 632 { 633 struct sock *sk = (struct sock *)msk; 634 struct mptcp_pm_addr_entry *entry; 635 struct pm_nl_pernet *pernet; 636 unsigned int subflows_max; 637 int i = 0; 638 639 pernet = pm_nl_get_pernet_from_msk(msk); 640 subflows_max = mptcp_pm_get_subflows_max(msk); 641 642 rcu_read_lock(); 643 list_for_each_entry_rcu(entry, &pernet->local_addr_list, list) { 644 if (!(entry->flags & MPTCP_PM_ADDR_FLAG_FULLMESH)) 645 continue; 646 647 if (!mptcp_pm_addr_families_match(sk, &entry->addr, remote)) 648 continue; 649 650 if (msk->pm.subflows < subflows_max) { 651 msk->pm.subflows++; 652 addrs[i++] = entry->addr; 653 } 654 } 655 rcu_read_unlock(); 656 657 /* If the array is empty, fill in the single 658 * 'IPADDRANY' local address 659 */ 660 if (!i) { 661 struct mptcp_addr_info local; 662 663 memset(&local, 0, sizeof(local)); 664 local.family = 665 #if IS_ENABLED(CONFIG_MPTCP_IPV6) 666 remote->family == AF_INET6 && 667 ipv6_addr_v4mapped(&remote->addr6) ? AF_INET : 668 #endif 669 remote->family; 670 671 if (!mptcp_pm_addr_families_match(sk, &local, remote)) 672 return 0; 673 674 msk->pm.subflows++; 675 addrs[i++] = local; 676 } 677 678 return i; 679 } 680 681 static void mptcp_pm_nl_add_addr_received(struct mptcp_sock *msk) 682 { 683 struct mptcp_addr_info addrs[MPTCP_PM_ADDR_MAX]; 684 struct sock *sk = (struct sock *)msk; 685 unsigned int add_addr_accept_max; 686 struct mptcp_addr_info remote; 687 unsigned int subflows_max; 688 int i, nr; 689 690 add_addr_accept_max = mptcp_pm_get_add_addr_accept_max(msk); 691 subflows_max = mptcp_pm_get_subflows_max(msk); 692 693 pr_debug("accepted %d:%d remote family %d", 694 msk->pm.add_addr_accepted, add_addr_accept_max, 695 msk->pm.remote.family); 696 697 remote = msk->pm.remote; 698 mptcp_pm_announce_addr(msk, &remote, true); 699 mptcp_pm_nl_addr_send_ack(msk); 700 701 if (lookup_subflow_by_daddr(&msk->conn_list, &remote)) 702 return; 703 704 /* pick id 0 port, if none is provided the remote address */ 705 if (!remote.port) 706 remote.port = sk->sk_dport; 707 708 /* connect to the specified remote address, using whatever 709 * local address the routing configuration will pick. 710 */ 711 nr = fill_local_addresses_vec(msk, &remote, addrs); 712 if (nr == 0) 713 return; 714 715 msk->pm.add_addr_accepted++; 716 if (msk->pm.add_addr_accepted >= add_addr_accept_max || 717 msk->pm.subflows >= subflows_max) 718 WRITE_ONCE(msk->pm.accept_addr, false); 719 720 spin_unlock_bh(&msk->pm.lock); 721 for (i = 0; i < nr; i++) 722 __mptcp_subflow_connect(sk, &addrs[i], &remote); 723 spin_lock_bh(&msk->pm.lock); 724 } 725 726 void mptcp_pm_nl_addr_send_ack(struct mptcp_sock *msk) 727 { 728 struct mptcp_subflow_context *subflow; 729 730 msk_owned_by_me(msk); 731 lockdep_assert_held(&msk->pm.lock); 732 733 if (!mptcp_pm_should_add_signal(msk) && 734 !mptcp_pm_should_rm_signal(msk)) 735 return; 736 737 subflow = list_first_entry_or_null(&msk->conn_list, typeof(*subflow), node); 738 if (subflow) 739 mptcp_pm_send_ack(msk, subflow, false, false); 740 } 741 742 int mptcp_pm_nl_mp_prio_send_ack(struct mptcp_sock *msk, 743 struct mptcp_addr_info *addr, 744 struct mptcp_addr_info *rem, 745 u8 bkup) 746 { 747 struct mptcp_subflow_context *subflow; 748 749 pr_debug("bkup=%d", bkup); 750 751 mptcp_for_each_subflow(msk, subflow) { 752 struct sock *ssk = mptcp_subflow_tcp_sock(subflow); 753 struct mptcp_addr_info local, remote; 754 755 mptcp_local_address((struct sock_common *)ssk, &local); 756 if (!mptcp_addresses_equal(&local, addr, addr->port)) 757 continue; 758 759 if (rem && rem->family != AF_UNSPEC) { 760 remote_address((struct sock_common *)ssk, &remote); 761 if (!mptcp_addresses_equal(&remote, rem, rem->port)) 762 continue; 763 } 764 765 __mptcp_pm_send_ack(msk, subflow, true, bkup); 766 return 0; 767 } 768 769 return -EINVAL; 770 } 771 772 static bool mptcp_local_id_match(const struct mptcp_sock *msk, u8 local_id, u8 id) 773 { 774 return local_id == id || (!local_id && msk->mpc_endpoint_id == id); 775 } 776 777 static void mptcp_pm_nl_rm_addr_or_subflow(struct mptcp_sock *msk, 778 const struct mptcp_rm_list *rm_list, 779 enum linux_mptcp_mib_field rm_type) 780 { 781 struct mptcp_subflow_context *subflow, *tmp; 782 struct sock *sk = (struct sock *)msk; 783 u8 i; 784 785 pr_debug("%s rm_list_nr %d", 786 rm_type == MPTCP_MIB_RMADDR ? "address" : "subflow", rm_list->nr); 787 788 msk_owned_by_me(msk); 789 790 if (sk->sk_state == TCP_LISTEN) 791 return; 792 793 if (!rm_list->nr) 794 return; 795 796 if (list_empty(&msk->conn_list)) 797 return; 798 799 for (i = 0; i < rm_list->nr; i++) { 800 u8 rm_id = rm_list->ids[i]; 801 bool removed = false; 802 803 mptcp_for_each_subflow_safe(msk, subflow, tmp) { 804 struct sock *ssk = mptcp_subflow_tcp_sock(subflow); 805 int how = RCV_SHUTDOWN | SEND_SHUTDOWN; 806 u8 id = subflow->local_id; 807 808 if (rm_type == MPTCP_MIB_RMADDR && subflow->remote_id != rm_id) 809 continue; 810 if (rm_type == MPTCP_MIB_RMSUBFLOW && !mptcp_local_id_match(msk, id, rm_id)) 811 continue; 812 813 pr_debug(" -> %s rm_list_ids[%d]=%u local_id=%u remote_id=%u mpc_id=%u", 814 rm_type == MPTCP_MIB_RMADDR ? "address" : "subflow", 815 i, rm_id, subflow->local_id, subflow->remote_id, 816 msk->mpc_endpoint_id); 817 spin_unlock_bh(&msk->pm.lock); 818 mptcp_subflow_shutdown(sk, ssk, how); 819 820 /* the following takes care of updating the subflows counter */ 821 mptcp_close_ssk(sk, ssk, subflow); 822 spin_lock_bh(&msk->pm.lock); 823 824 removed = true; 825 __MPTCP_INC_STATS(sock_net(sk), rm_type); 826 } 827 if (rm_type == MPTCP_MIB_RMSUBFLOW) 828 __set_bit(rm_id ? rm_id : msk->mpc_endpoint_id, msk->pm.id_avail_bitmap); 829 if (!removed) 830 continue; 831 832 if (!mptcp_pm_is_kernel(msk)) 833 continue; 834 835 if (rm_type == MPTCP_MIB_RMADDR) { 836 msk->pm.add_addr_accepted--; 837 WRITE_ONCE(msk->pm.accept_addr, true); 838 } else if (rm_type == MPTCP_MIB_RMSUBFLOW) { 839 msk->pm.local_addr_used--; 840 } 841 } 842 } 843 844 static void mptcp_pm_nl_rm_addr_received(struct mptcp_sock *msk) 845 { 846 mptcp_pm_nl_rm_addr_or_subflow(msk, &msk->pm.rm_list_rx, MPTCP_MIB_RMADDR); 847 } 848 849 void mptcp_pm_nl_rm_subflow_received(struct mptcp_sock *msk, 850 const struct mptcp_rm_list *rm_list) 851 { 852 mptcp_pm_nl_rm_addr_or_subflow(msk, rm_list, MPTCP_MIB_RMSUBFLOW); 853 } 854 855 void mptcp_pm_nl_work(struct mptcp_sock *msk) 856 { 857 struct mptcp_pm_data *pm = &msk->pm; 858 859 msk_owned_by_me(msk); 860 861 if (!(pm->status & MPTCP_PM_WORK_MASK)) 862 return; 863 864 spin_lock_bh(&msk->pm.lock); 865 866 pr_debug("msk=%p status=%x", msk, pm->status); 867 if (pm->status & BIT(MPTCP_PM_ADD_ADDR_RECEIVED)) { 868 pm->status &= ~BIT(MPTCP_PM_ADD_ADDR_RECEIVED); 869 mptcp_pm_nl_add_addr_received(msk); 870 } 871 if (pm->status & BIT(MPTCP_PM_ADD_ADDR_SEND_ACK)) { 872 pm->status &= ~BIT(MPTCP_PM_ADD_ADDR_SEND_ACK); 873 mptcp_pm_nl_addr_send_ack(msk); 874 } 875 if (pm->status & BIT(MPTCP_PM_RM_ADDR_RECEIVED)) { 876 pm->status &= ~BIT(MPTCP_PM_RM_ADDR_RECEIVED); 877 mptcp_pm_nl_rm_addr_received(msk); 878 } 879 if (pm->status & BIT(MPTCP_PM_ESTABLISHED)) { 880 pm->status &= ~BIT(MPTCP_PM_ESTABLISHED); 881 mptcp_pm_nl_fully_established(msk); 882 } 883 if (pm->status & BIT(MPTCP_PM_SUBFLOW_ESTABLISHED)) { 884 pm->status &= ~BIT(MPTCP_PM_SUBFLOW_ESTABLISHED); 885 mptcp_pm_nl_subflow_established(msk); 886 } 887 888 spin_unlock_bh(&msk->pm.lock); 889 } 890 891 static bool address_use_port(struct mptcp_pm_addr_entry *entry) 892 { 893 return (entry->flags & 894 (MPTCP_PM_ADDR_FLAG_SIGNAL | MPTCP_PM_ADDR_FLAG_SUBFLOW)) == 895 MPTCP_PM_ADDR_FLAG_SIGNAL; 896 } 897 898 /* caller must ensure the RCU grace period is already elapsed */ 899 static void __mptcp_pm_release_addr_entry(struct mptcp_pm_addr_entry *entry) 900 { 901 if (entry->lsk) 902 sock_release(entry->lsk); 903 kfree(entry); 904 } 905 906 static int mptcp_pm_nl_append_new_local_addr(struct pm_nl_pernet *pernet, 907 struct mptcp_pm_addr_entry *entry) 908 { 909 struct mptcp_pm_addr_entry *cur, *del_entry = NULL; 910 unsigned int addr_max; 911 int ret = -EINVAL; 912 913 spin_lock_bh(&pernet->lock); 914 /* to keep the code simple, don't do IDR-like allocation for address ID, 915 * just bail when we exceed limits 916 */ 917 if (pernet->next_id == MPTCP_PM_MAX_ADDR_ID) 918 pernet->next_id = 1; 919 if (pernet->addrs >= MPTCP_PM_ADDR_MAX) { 920 ret = -ERANGE; 921 goto out; 922 } 923 if (test_bit(entry->addr.id, pernet->id_bitmap)) { 924 ret = -EBUSY; 925 goto out; 926 } 927 928 /* do not insert duplicate address, differentiate on port only 929 * singled addresses 930 */ 931 if (!address_use_port(entry)) 932 entry->addr.port = 0; 933 list_for_each_entry(cur, &pernet->local_addr_list, list) { 934 if (mptcp_addresses_equal(&cur->addr, &entry->addr, 935 cur->addr.port || entry->addr.port)) { 936 /* allow replacing the exiting endpoint only if such 937 * endpoint is an implicit one and the user-space 938 * did not provide an endpoint id 939 */ 940 if (!(cur->flags & MPTCP_PM_ADDR_FLAG_IMPLICIT)) { 941 ret = -EEXIST; 942 goto out; 943 } 944 if (entry->addr.id) 945 goto out; 946 947 pernet->addrs--; 948 entry->addr.id = cur->addr.id; 949 list_del_rcu(&cur->list); 950 del_entry = cur; 951 break; 952 } 953 } 954 955 if (!entry->addr.id) { 956 find_next: 957 entry->addr.id = find_next_zero_bit(pernet->id_bitmap, 958 MPTCP_PM_MAX_ADDR_ID + 1, 959 pernet->next_id); 960 if (!entry->addr.id && pernet->next_id != 1) { 961 pernet->next_id = 1; 962 goto find_next; 963 } 964 } 965 966 if (!entry->addr.id) 967 goto out; 968 969 __set_bit(entry->addr.id, pernet->id_bitmap); 970 if (entry->addr.id > pernet->next_id) 971 pernet->next_id = entry->addr.id; 972 973 if (entry->flags & MPTCP_PM_ADDR_FLAG_SIGNAL) { 974 addr_max = pernet->add_addr_signal_max; 975 WRITE_ONCE(pernet->add_addr_signal_max, addr_max + 1); 976 } 977 if (entry->flags & MPTCP_PM_ADDR_FLAG_SUBFLOW) { 978 addr_max = pernet->local_addr_max; 979 WRITE_ONCE(pernet->local_addr_max, addr_max + 1); 980 } 981 982 pernet->addrs++; 983 if (!entry->addr.port) 984 list_add_tail_rcu(&entry->list, &pernet->local_addr_list); 985 else 986 list_add_rcu(&entry->list, &pernet->local_addr_list); 987 ret = entry->addr.id; 988 989 out: 990 spin_unlock_bh(&pernet->lock); 991 992 /* just replaced an existing entry, free it */ 993 if (del_entry) { 994 synchronize_rcu(); 995 __mptcp_pm_release_addr_entry(del_entry); 996 } 997 return ret; 998 } 999 1000 static struct lock_class_key mptcp_slock_keys[2]; 1001 static struct lock_class_key mptcp_keys[2]; 1002 1003 static int mptcp_pm_nl_create_listen_socket(struct sock *sk, 1004 struct mptcp_pm_addr_entry *entry) 1005 { 1006 bool is_ipv6 = sk->sk_family == AF_INET6; 1007 int addrlen = sizeof(struct sockaddr_in); 1008 struct sockaddr_storage addr; 1009 struct sock *newsk, *ssk; 1010 int backlog = 1024; 1011 int err; 1012 1013 err = sock_create_kern(sock_net(sk), entry->addr.family, 1014 SOCK_STREAM, IPPROTO_MPTCP, &entry->lsk); 1015 if (err) 1016 return err; 1017 1018 newsk = entry->lsk->sk; 1019 if (!newsk) 1020 return -EINVAL; 1021 1022 /* The subflow socket lock is acquired in a nested to the msk one 1023 * in several places, even by the TCP stack, and this msk is a kernel 1024 * socket: lockdep complains. Instead of propagating the _nested 1025 * modifiers in several places, re-init the lock class for the msk 1026 * socket to an mptcp specific one. 1027 */ 1028 sock_lock_init_class_and_name(newsk, 1029 is_ipv6 ? "mlock-AF_INET6" : "mlock-AF_INET", 1030 &mptcp_slock_keys[is_ipv6], 1031 is_ipv6 ? "msk_lock-AF_INET6" : "msk_lock-AF_INET", 1032 &mptcp_keys[is_ipv6]); 1033 1034 lock_sock(newsk); 1035 ssk = __mptcp_nmpc_sk(mptcp_sk(newsk)); 1036 release_sock(newsk); 1037 if (IS_ERR(ssk)) 1038 return PTR_ERR(ssk); 1039 1040 mptcp_info2sockaddr(&entry->addr, &addr, entry->addr.family); 1041 #if IS_ENABLED(CONFIG_MPTCP_IPV6) 1042 if (entry->addr.family == AF_INET6) 1043 addrlen = sizeof(struct sockaddr_in6); 1044 #endif 1045 if (ssk->sk_family == AF_INET) 1046 err = inet_bind_sk(ssk, (struct sockaddr *)&addr, addrlen); 1047 #if IS_ENABLED(CONFIG_MPTCP_IPV6) 1048 else if (ssk->sk_family == AF_INET6) 1049 err = inet6_bind_sk(ssk, (struct sockaddr *)&addr, addrlen); 1050 #endif 1051 if (err) 1052 return err; 1053 1054 inet_sk_state_store(newsk, TCP_LISTEN); 1055 lock_sock(ssk); 1056 err = __inet_listen_sk(ssk, backlog); 1057 if (!err) 1058 mptcp_event_pm_listener(ssk, MPTCP_EVENT_LISTENER_CREATED); 1059 release_sock(ssk); 1060 return err; 1061 } 1062 1063 int mptcp_pm_nl_get_local_id(struct mptcp_sock *msk, struct mptcp_addr_info *skc) 1064 { 1065 struct mptcp_pm_addr_entry *entry; 1066 struct pm_nl_pernet *pernet; 1067 int ret = -1; 1068 1069 pernet = pm_nl_get_pernet_from_msk(msk); 1070 1071 rcu_read_lock(); 1072 list_for_each_entry_rcu(entry, &pernet->local_addr_list, list) { 1073 if (mptcp_addresses_equal(&entry->addr, skc, entry->addr.port)) { 1074 ret = entry->addr.id; 1075 break; 1076 } 1077 } 1078 rcu_read_unlock(); 1079 if (ret >= 0) 1080 return ret; 1081 1082 /* address not found, add to local list */ 1083 entry = kmalloc(sizeof(*entry), GFP_ATOMIC); 1084 if (!entry) 1085 return -ENOMEM; 1086 1087 entry->addr = *skc; 1088 entry->addr.id = 0; 1089 entry->addr.port = 0; 1090 entry->ifindex = 0; 1091 entry->flags = MPTCP_PM_ADDR_FLAG_IMPLICIT; 1092 entry->lsk = NULL; 1093 ret = mptcp_pm_nl_append_new_local_addr(pernet, entry); 1094 if (ret < 0) 1095 kfree(entry); 1096 1097 return ret; 1098 } 1099 1100 #define MPTCP_PM_CMD_GRP_OFFSET 0 1101 #define MPTCP_PM_EV_GRP_OFFSET 1 1102 1103 static const struct genl_multicast_group mptcp_pm_mcgrps[] = { 1104 [MPTCP_PM_CMD_GRP_OFFSET] = { .name = MPTCP_PM_CMD_GRP_NAME, }, 1105 [MPTCP_PM_EV_GRP_OFFSET] = { .name = MPTCP_PM_EV_GRP_NAME, 1106 .flags = GENL_UNS_ADMIN_PERM, 1107 }, 1108 }; 1109 1110 static const struct nla_policy 1111 mptcp_pm_addr_policy[MPTCP_PM_ADDR_ATTR_MAX + 1] = { 1112 [MPTCP_PM_ADDR_ATTR_FAMILY] = { .type = NLA_U16, }, 1113 [MPTCP_PM_ADDR_ATTR_ID] = { .type = NLA_U8, }, 1114 [MPTCP_PM_ADDR_ATTR_ADDR4] = { .type = NLA_U32, }, 1115 [MPTCP_PM_ADDR_ATTR_ADDR6] = 1116 NLA_POLICY_EXACT_LEN(sizeof(struct in6_addr)), 1117 [MPTCP_PM_ADDR_ATTR_PORT] = { .type = NLA_U16 }, 1118 [MPTCP_PM_ADDR_ATTR_FLAGS] = { .type = NLA_U32 }, 1119 [MPTCP_PM_ADDR_ATTR_IF_IDX] = { .type = NLA_S32 }, 1120 }; 1121 1122 static const struct nla_policy mptcp_pm_policy[MPTCP_PM_ATTR_MAX + 1] = { 1123 [MPTCP_PM_ATTR_ADDR] = 1124 NLA_POLICY_NESTED(mptcp_pm_addr_policy), 1125 [MPTCP_PM_ATTR_RCV_ADD_ADDRS] = { .type = NLA_U32, }, 1126 [MPTCP_PM_ATTR_SUBFLOWS] = { .type = NLA_U32, }, 1127 [MPTCP_PM_ATTR_TOKEN] = { .type = NLA_U32, }, 1128 [MPTCP_PM_ATTR_LOC_ID] = { .type = NLA_U8, }, 1129 [MPTCP_PM_ATTR_ADDR_REMOTE] = 1130 NLA_POLICY_NESTED(mptcp_pm_addr_policy), 1131 }; 1132 1133 void mptcp_pm_nl_subflow_chk_stale(const struct mptcp_sock *msk, struct sock *ssk) 1134 { 1135 struct mptcp_subflow_context *iter, *subflow = mptcp_subflow_ctx(ssk); 1136 struct sock *sk = (struct sock *)msk; 1137 unsigned int active_max_loss_cnt; 1138 struct net *net = sock_net(sk); 1139 unsigned int stale_loss_cnt; 1140 bool slow; 1141 1142 stale_loss_cnt = mptcp_stale_loss_cnt(net); 1143 if (subflow->stale || !stale_loss_cnt || subflow->stale_count <= stale_loss_cnt) 1144 return; 1145 1146 /* look for another available subflow not in loss state */ 1147 active_max_loss_cnt = max_t(int, stale_loss_cnt - 1, 1); 1148 mptcp_for_each_subflow(msk, iter) { 1149 if (iter != subflow && mptcp_subflow_active(iter) && 1150 iter->stale_count < active_max_loss_cnt) { 1151 /* we have some alternatives, try to mark this subflow as idle ...*/ 1152 slow = lock_sock_fast(ssk); 1153 if (!tcp_rtx_and_write_queues_empty(ssk)) { 1154 subflow->stale = 1; 1155 __mptcp_retransmit_pending_data(sk); 1156 MPTCP_INC_STATS(net, MPTCP_MIB_SUBFLOWSTALE); 1157 } 1158 unlock_sock_fast(ssk, slow); 1159 1160 /* always try to push the pending data regardless of re-injections: 1161 * we can possibly use backup subflows now, and subflow selection 1162 * is cheap under the msk socket lock 1163 */ 1164 __mptcp_push_pending(sk, 0); 1165 return; 1166 } 1167 } 1168 } 1169 1170 static int mptcp_pm_family_to_addr(int family) 1171 { 1172 #if IS_ENABLED(CONFIG_MPTCP_IPV6) 1173 if (family == AF_INET6) 1174 return MPTCP_PM_ADDR_ATTR_ADDR6; 1175 #endif 1176 return MPTCP_PM_ADDR_ATTR_ADDR4; 1177 } 1178 1179 static int mptcp_pm_parse_pm_addr_attr(struct nlattr *tb[], 1180 const struct nlattr *attr, 1181 struct genl_info *info, 1182 struct mptcp_addr_info *addr, 1183 bool require_family) 1184 { 1185 int err, addr_addr; 1186 1187 if (!attr) { 1188 GENL_SET_ERR_MSG(info, "missing address info"); 1189 return -EINVAL; 1190 } 1191 1192 /* no validation needed - was already done via nested policy */ 1193 err = nla_parse_nested_deprecated(tb, MPTCP_PM_ADDR_ATTR_MAX, attr, 1194 mptcp_pm_addr_policy, info->extack); 1195 if (err) 1196 return err; 1197 1198 if (tb[MPTCP_PM_ADDR_ATTR_ID]) 1199 addr->id = nla_get_u8(tb[MPTCP_PM_ADDR_ATTR_ID]); 1200 1201 if (!tb[MPTCP_PM_ADDR_ATTR_FAMILY]) { 1202 if (!require_family) 1203 return 0; 1204 1205 NL_SET_ERR_MSG_ATTR(info->extack, attr, 1206 "missing family"); 1207 return -EINVAL; 1208 } 1209 1210 addr->family = nla_get_u16(tb[MPTCP_PM_ADDR_ATTR_FAMILY]); 1211 if (addr->family != AF_INET 1212 #if IS_ENABLED(CONFIG_MPTCP_IPV6) 1213 && addr->family != AF_INET6 1214 #endif 1215 ) { 1216 NL_SET_ERR_MSG_ATTR(info->extack, attr, 1217 "unknown address family"); 1218 return -EINVAL; 1219 } 1220 addr_addr = mptcp_pm_family_to_addr(addr->family); 1221 if (!tb[addr_addr]) { 1222 NL_SET_ERR_MSG_ATTR(info->extack, attr, 1223 "missing address data"); 1224 return -EINVAL; 1225 } 1226 1227 #if IS_ENABLED(CONFIG_MPTCP_IPV6) 1228 if (addr->family == AF_INET6) 1229 addr->addr6 = nla_get_in6_addr(tb[addr_addr]); 1230 else 1231 #endif 1232 addr->addr.s_addr = nla_get_in_addr(tb[addr_addr]); 1233 1234 if (tb[MPTCP_PM_ADDR_ATTR_PORT]) 1235 addr->port = htons(nla_get_u16(tb[MPTCP_PM_ADDR_ATTR_PORT])); 1236 1237 return 0; 1238 } 1239 1240 int mptcp_pm_parse_addr(struct nlattr *attr, struct genl_info *info, 1241 struct mptcp_addr_info *addr) 1242 { 1243 struct nlattr *tb[MPTCP_PM_ADDR_ATTR_MAX + 1]; 1244 1245 memset(addr, 0, sizeof(*addr)); 1246 1247 return mptcp_pm_parse_pm_addr_attr(tb, attr, info, addr, true); 1248 } 1249 1250 int mptcp_pm_parse_entry(struct nlattr *attr, struct genl_info *info, 1251 bool require_family, 1252 struct mptcp_pm_addr_entry *entry) 1253 { 1254 struct nlattr *tb[MPTCP_PM_ADDR_ATTR_MAX + 1]; 1255 int err; 1256 1257 memset(entry, 0, sizeof(*entry)); 1258 1259 err = mptcp_pm_parse_pm_addr_attr(tb, attr, info, &entry->addr, require_family); 1260 if (err) 1261 return err; 1262 1263 if (tb[MPTCP_PM_ADDR_ATTR_IF_IDX]) { 1264 u32 val = nla_get_s32(tb[MPTCP_PM_ADDR_ATTR_IF_IDX]); 1265 1266 entry->ifindex = val; 1267 } 1268 1269 if (tb[MPTCP_PM_ADDR_ATTR_FLAGS]) 1270 entry->flags = nla_get_u32(tb[MPTCP_PM_ADDR_ATTR_FLAGS]); 1271 1272 if (tb[MPTCP_PM_ADDR_ATTR_PORT]) 1273 entry->addr.port = htons(nla_get_u16(tb[MPTCP_PM_ADDR_ATTR_PORT])); 1274 1275 return 0; 1276 } 1277 1278 static struct pm_nl_pernet *genl_info_pm_nl(struct genl_info *info) 1279 { 1280 return pm_nl_get_pernet(genl_info_net(info)); 1281 } 1282 1283 static int mptcp_nl_add_subflow_or_signal_addr(struct net *net) 1284 { 1285 struct mptcp_sock *msk; 1286 long s_slot = 0, s_num = 0; 1287 1288 while ((msk = mptcp_token_iter_next(net, &s_slot, &s_num)) != NULL) { 1289 struct sock *sk = (struct sock *)msk; 1290 1291 if (!READ_ONCE(msk->fully_established) || 1292 mptcp_pm_is_userspace(msk)) 1293 goto next; 1294 1295 lock_sock(sk); 1296 spin_lock_bh(&msk->pm.lock); 1297 mptcp_pm_create_subflow_or_signal_addr(msk); 1298 spin_unlock_bh(&msk->pm.lock); 1299 release_sock(sk); 1300 1301 next: 1302 sock_put(sk); 1303 cond_resched(); 1304 } 1305 1306 return 0; 1307 } 1308 1309 static int mptcp_nl_cmd_add_addr(struct sk_buff *skb, struct genl_info *info) 1310 { 1311 struct nlattr *attr = info->attrs[MPTCP_PM_ATTR_ADDR]; 1312 struct pm_nl_pernet *pernet = genl_info_pm_nl(info); 1313 struct mptcp_pm_addr_entry addr, *entry; 1314 int ret; 1315 1316 ret = mptcp_pm_parse_entry(attr, info, true, &addr); 1317 if (ret < 0) 1318 return ret; 1319 1320 if (addr.addr.port && !(addr.flags & MPTCP_PM_ADDR_FLAG_SIGNAL)) { 1321 GENL_SET_ERR_MSG(info, "flags must have signal when using port"); 1322 return -EINVAL; 1323 } 1324 1325 if (addr.flags & MPTCP_PM_ADDR_FLAG_SIGNAL && 1326 addr.flags & MPTCP_PM_ADDR_FLAG_FULLMESH) { 1327 GENL_SET_ERR_MSG(info, "flags mustn't have both signal and fullmesh"); 1328 return -EINVAL; 1329 } 1330 1331 if (addr.flags & MPTCP_PM_ADDR_FLAG_IMPLICIT) { 1332 GENL_SET_ERR_MSG(info, "can't create IMPLICIT endpoint"); 1333 return -EINVAL; 1334 } 1335 1336 entry = kzalloc(sizeof(*entry), GFP_KERNEL_ACCOUNT); 1337 if (!entry) { 1338 GENL_SET_ERR_MSG(info, "can't allocate addr"); 1339 return -ENOMEM; 1340 } 1341 1342 *entry = addr; 1343 if (entry->addr.port) { 1344 ret = mptcp_pm_nl_create_listen_socket(skb->sk, entry); 1345 if (ret) { 1346 GENL_SET_ERR_MSG_FMT(info, "create listen socket error: %d", ret); 1347 goto out_free; 1348 } 1349 } 1350 ret = mptcp_pm_nl_append_new_local_addr(pernet, entry); 1351 if (ret < 0) { 1352 GENL_SET_ERR_MSG_FMT(info, "too many addresses or duplicate one: %d", ret); 1353 goto out_free; 1354 } 1355 1356 mptcp_nl_add_subflow_or_signal_addr(sock_net(skb->sk)); 1357 return 0; 1358 1359 out_free: 1360 __mptcp_pm_release_addr_entry(entry); 1361 return ret; 1362 } 1363 1364 int mptcp_pm_nl_get_flags_and_ifindex_by_id(struct mptcp_sock *msk, unsigned int id, 1365 u8 *flags, int *ifindex) 1366 { 1367 struct mptcp_pm_addr_entry *entry; 1368 struct sock *sk = (struct sock *)msk; 1369 struct net *net = sock_net(sk); 1370 1371 rcu_read_lock(); 1372 entry = __lookup_addr_by_id(pm_nl_get_pernet(net), id); 1373 if (entry) { 1374 *flags = entry->flags; 1375 *ifindex = entry->ifindex; 1376 } 1377 rcu_read_unlock(); 1378 1379 return 0; 1380 } 1381 1382 static bool remove_anno_list_by_saddr(struct mptcp_sock *msk, 1383 const struct mptcp_addr_info *addr) 1384 { 1385 struct mptcp_pm_add_entry *entry; 1386 1387 entry = mptcp_pm_del_add_timer(msk, addr, false); 1388 if (entry) { 1389 list_del(&entry->list); 1390 kfree(entry); 1391 return true; 1392 } 1393 1394 return false; 1395 } 1396 1397 static bool mptcp_pm_remove_anno_addr(struct mptcp_sock *msk, 1398 const struct mptcp_addr_info *addr, 1399 bool force) 1400 { 1401 struct mptcp_rm_list list = { .nr = 0 }; 1402 bool ret; 1403 1404 list.ids[list.nr++] = addr->id; 1405 1406 ret = remove_anno_list_by_saddr(msk, addr); 1407 if (ret || force) { 1408 spin_lock_bh(&msk->pm.lock); 1409 mptcp_pm_remove_addr(msk, &list); 1410 spin_unlock_bh(&msk->pm.lock); 1411 } 1412 return ret; 1413 } 1414 1415 static int mptcp_nl_remove_subflow_and_signal_addr(struct net *net, 1416 const struct mptcp_pm_addr_entry *entry) 1417 { 1418 const struct mptcp_addr_info *addr = &entry->addr; 1419 struct mptcp_rm_list list = { .nr = 0 }; 1420 long s_slot = 0, s_num = 0; 1421 struct mptcp_sock *msk; 1422 1423 pr_debug("remove_id=%d", addr->id); 1424 1425 list.ids[list.nr++] = addr->id; 1426 1427 while ((msk = mptcp_token_iter_next(net, &s_slot, &s_num)) != NULL) { 1428 struct sock *sk = (struct sock *)msk; 1429 bool remove_subflow; 1430 1431 if (mptcp_pm_is_userspace(msk)) 1432 goto next; 1433 1434 if (list_empty(&msk->conn_list)) { 1435 mptcp_pm_remove_anno_addr(msk, addr, false); 1436 goto next; 1437 } 1438 1439 lock_sock(sk); 1440 remove_subflow = lookup_subflow_by_saddr(&msk->conn_list, addr); 1441 mptcp_pm_remove_anno_addr(msk, addr, remove_subflow && 1442 !(entry->flags & MPTCP_PM_ADDR_FLAG_IMPLICIT)); 1443 if (remove_subflow) 1444 mptcp_pm_remove_subflow(msk, &list); 1445 release_sock(sk); 1446 1447 next: 1448 sock_put(sk); 1449 cond_resched(); 1450 } 1451 1452 return 0; 1453 } 1454 1455 static int mptcp_nl_remove_id_zero_address(struct net *net, 1456 struct mptcp_addr_info *addr) 1457 { 1458 struct mptcp_rm_list list = { .nr = 0 }; 1459 long s_slot = 0, s_num = 0; 1460 struct mptcp_sock *msk; 1461 1462 list.ids[list.nr++] = 0; 1463 1464 while ((msk = mptcp_token_iter_next(net, &s_slot, &s_num)) != NULL) { 1465 struct sock *sk = (struct sock *)msk; 1466 struct mptcp_addr_info msk_local; 1467 1468 if (list_empty(&msk->conn_list) || mptcp_pm_is_userspace(msk)) 1469 goto next; 1470 1471 mptcp_local_address((struct sock_common *)msk, &msk_local); 1472 if (!mptcp_addresses_equal(&msk_local, addr, addr->port)) 1473 goto next; 1474 1475 lock_sock(sk); 1476 spin_lock_bh(&msk->pm.lock); 1477 mptcp_pm_remove_addr(msk, &list); 1478 mptcp_pm_nl_rm_subflow_received(msk, &list); 1479 spin_unlock_bh(&msk->pm.lock); 1480 release_sock(sk); 1481 1482 next: 1483 sock_put(sk); 1484 cond_resched(); 1485 } 1486 1487 return 0; 1488 } 1489 1490 static int mptcp_nl_cmd_del_addr(struct sk_buff *skb, struct genl_info *info) 1491 { 1492 struct nlattr *attr = info->attrs[MPTCP_PM_ATTR_ADDR]; 1493 struct pm_nl_pernet *pernet = genl_info_pm_nl(info); 1494 struct mptcp_pm_addr_entry addr, *entry; 1495 unsigned int addr_max; 1496 int ret; 1497 1498 ret = mptcp_pm_parse_entry(attr, info, false, &addr); 1499 if (ret < 0) 1500 return ret; 1501 1502 /* the zero id address is special: the first address used by the msk 1503 * always gets such an id, so different subflows can have different zero 1504 * id addresses. Additionally zero id is not accounted for in id_bitmap. 1505 * Let's use an 'mptcp_rm_list' instead of the common remove code. 1506 */ 1507 if (addr.addr.id == 0) 1508 return mptcp_nl_remove_id_zero_address(sock_net(skb->sk), &addr.addr); 1509 1510 spin_lock_bh(&pernet->lock); 1511 entry = __lookup_addr_by_id(pernet, addr.addr.id); 1512 if (!entry) { 1513 GENL_SET_ERR_MSG(info, "address not found"); 1514 spin_unlock_bh(&pernet->lock); 1515 return -EINVAL; 1516 } 1517 if (entry->flags & MPTCP_PM_ADDR_FLAG_SIGNAL) { 1518 addr_max = pernet->add_addr_signal_max; 1519 WRITE_ONCE(pernet->add_addr_signal_max, addr_max - 1); 1520 } 1521 if (entry->flags & MPTCP_PM_ADDR_FLAG_SUBFLOW) { 1522 addr_max = pernet->local_addr_max; 1523 WRITE_ONCE(pernet->local_addr_max, addr_max - 1); 1524 } 1525 1526 pernet->addrs--; 1527 list_del_rcu(&entry->list); 1528 __clear_bit(entry->addr.id, pernet->id_bitmap); 1529 spin_unlock_bh(&pernet->lock); 1530 1531 mptcp_nl_remove_subflow_and_signal_addr(sock_net(skb->sk), entry); 1532 synchronize_rcu(); 1533 __mptcp_pm_release_addr_entry(entry); 1534 1535 return ret; 1536 } 1537 1538 void mptcp_pm_remove_addrs(struct mptcp_sock *msk, struct list_head *rm_list) 1539 { 1540 struct mptcp_rm_list alist = { .nr = 0 }; 1541 struct mptcp_pm_addr_entry *entry; 1542 1543 list_for_each_entry(entry, rm_list, list) { 1544 remove_anno_list_by_saddr(msk, &entry->addr); 1545 if (alist.nr < MPTCP_RM_IDS_MAX) 1546 alist.ids[alist.nr++] = entry->addr.id; 1547 } 1548 1549 if (alist.nr) { 1550 spin_lock_bh(&msk->pm.lock); 1551 mptcp_pm_remove_addr(msk, &alist); 1552 spin_unlock_bh(&msk->pm.lock); 1553 } 1554 } 1555 1556 void mptcp_pm_remove_addrs_and_subflows(struct mptcp_sock *msk, 1557 struct list_head *rm_list) 1558 { 1559 struct mptcp_rm_list alist = { .nr = 0 }, slist = { .nr = 0 }; 1560 struct mptcp_pm_addr_entry *entry; 1561 1562 list_for_each_entry(entry, rm_list, list) { 1563 if (lookup_subflow_by_saddr(&msk->conn_list, &entry->addr) && 1564 slist.nr < MPTCP_RM_IDS_MAX) 1565 slist.ids[slist.nr++] = entry->addr.id; 1566 1567 if (remove_anno_list_by_saddr(msk, &entry->addr) && 1568 alist.nr < MPTCP_RM_IDS_MAX) 1569 alist.ids[alist.nr++] = entry->addr.id; 1570 } 1571 1572 if (alist.nr) { 1573 spin_lock_bh(&msk->pm.lock); 1574 mptcp_pm_remove_addr(msk, &alist); 1575 spin_unlock_bh(&msk->pm.lock); 1576 } 1577 if (slist.nr) 1578 mptcp_pm_remove_subflow(msk, &slist); 1579 } 1580 1581 static void mptcp_nl_remove_addrs_list(struct net *net, 1582 struct list_head *rm_list) 1583 { 1584 long s_slot = 0, s_num = 0; 1585 struct mptcp_sock *msk; 1586 1587 if (list_empty(rm_list)) 1588 return; 1589 1590 while ((msk = mptcp_token_iter_next(net, &s_slot, &s_num)) != NULL) { 1591 struct sock *sk = (struct sock *)msk; 1592 1593 if (!mptcp_pm_is_userspace(msk)) { 1594 lock_sock(sk); 1595 mptcp_pm_remove_addrs_and_subflows(msk, rm_list); 1596 release_sock(sk); 1597 } 1598 1599 sock_put(sk); 1600 cond_resched(); 1601 } 1602 } 1603 1604 /* caller must ensure the RCU grace period is already elapsed */ 1605 static void __flush_addrs(struct list_head *list) 1606 { 1607 while (!list_empty(list)) { 1608 struct mptcp_pm_addr_entry *cur; 1609 1610 cur = list_entry(list->next, 1611 struct mptcp_pm_addr_entry, list); 1612 list_del_rcu(&cur->list); 1613 __mptcp_pm_release_addr_entry(cur); 1614 } 1615 } 1616 1617 static void __reset_counters(struct pm_nl_pernet *pernet) 1618 { 1619 WRITE_ONCE(pernet->add_addr_signal_max, 0); 1620 WRITE_ONCE(pernet->add_addr_accept_max, 0); 1621 WRITE_ONCE(pernet->local_addr_max, 0); 1622 pernet->addrs = 0; 1623 } 1624 1625 static int mptcp_nl_cmd_flush_addrs(struct sk_buff *skb, struct genl_info *info) 1626 { 1627 struct pm_nl_pernet *pernet = genl_info_pm_nl(info); 1628 LIST_HEAD(free_list); 1629 1630 spin_lock_bh(&pernet->lock); 1631 list_splice_init(&pernet->local_addr_list, &free_list); 1632 __reset_counters(pernet); 1633 pernet->next_id = 1; 1634 bitmap_zero(pernet->id_bitmap, MPTCP_PM_MAX_ADDR_ID + 1); 1635 spin_unlock_bh(&pernet->lock); 1636 mptcp_nl_remove_addrs_list(sock_net(skb->sk), &free_list); 1637 synchronize_rcu(); 1638 __flush_addrs(&free_list); 1639 return 0; 1640 } 1641 1642 static int mptcp_nl_fill_addr(struct sk_buff *skb, 1643 struct mptcp_pm_addr_entry *entry) 1644 { 1645 struct mptcp_addr_info *addr = &entry->addr; 1646 struct nlattr *attr; 1647 1648 attr = nla_nest_start(skb, MPTCP_PM_ATTR_ADDR); 1649 if (!attr) 1650 return -EMSGSIZE; 1651 1652 if (nla_put_u16(skb, MPTCP_PM_ADDR_ATTR_FAMILY, addr->family)) 1653 goto nla_put_failure; 1654 if (nla_put_u16(skb, MPTCP_PM_ADDR_ATTR_PORT, ntohs(addr->port))) 1655 goto nla_put_failure; 1656 if (nla_put_u8(skb, MPTCP_PM_ADDR_ATTR_ID, addr->id)) 1657 goto nla_put_failure; 1658 if (nla_put_u32(skb, MPTCP_PM_ADDR_ATTR_FLAGS, entry->flags)) 1659 goto nla_put_failure; 1660 if (entry->ifindex && 1661 nla_put_s32(skb, MPTCP_PM_ADDR_ATTR_IF_IDX, entry->ifindex)) 1662 goto nla_put_failure; 1663 1664 if (addr->family == AF_INET && 1665 nla_put_in_addr(skb, MPTCP_PM_ADDR_ATTR_ADDR4, 1666 addr->addr.s_addr)) 1667 goto nla_put_failure; 1668 #if IS_ENABLED(CONFIG_MPTCP_IPV6) 1669 else if (addr->family == AF_INET6 && 1670 nla_put_in6_addr(skb, MPTCP_PM_ADDR_ATTR_ADDR6, &addr->addr6)) 1671 goto nla_put_failure; 1672 #endif 1673 nla_nest_end(skb, attr); 1674 return 0; 1675 1676 nla_put_failure: 1677 nla_nest_cancel(skb, attr); 1678 return -EMSGSIZE; 1679 } 1680 1681 static int mptcp_nl_cmd_get_addr(struct sk_buff *skb, struct genl_info *info) 1682 { 1683 struct nlattr *attr = info->attrs[MPTCP_PM_ATTR_ADDR]; 1684 struct pm_nl_pernet *pernet = genl_info_pm_nl(info); 1685 struct mptcp_pm_addr_entry addr, *entry; 1686 struct sk_buff *msg; 1687 void *reply; 1688 int ret; 1689 1690 ret = mptcp_pm_parse_entry(attr, info, false, &addr); 1691 if (ret < 0) 1692 return ret; 1693 1694 msg = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL); 1695 if (!msg) 1696 return -ENOMEM; 1697 1698 reply = genlmsg_put_reply(msg, info, &mptcp_genl_family, 0, 1699 info->genlhdr->cmd); 1700 if (!reply) { 1701 GENL_SET_ERR_MSG(info, "not enough space in Netlink message"); 1702 ret = -EMSGSIZE; 1703 goto fail; 1704 } 1705 1706 spin_lock_bh(&pernet->lock); 1707 entry = __lookup_addr_by_id(pernet, addr.addr.id); 1708 if (!entry) { 1709 GENL_SET_ERR_MSG(info, "address not found"); 1710 ret = -EINVAL; 1711 goto unlock_fail; 1712 } 1713 1714 ret = mptcp_nl_fill_addr(msg, entry); 1715 if (ret) 1716 goto unlock_fail; 1717 1718 genlmsg_end(msg, reply); 1719 ret = genlmsg_reply(msg, info); 1720 spin_unlock_bh(&pernet->lock); 1721 return ret; 1722 1723 unlock_fail: 1724 spin_unlock_bh(&pernet->lock); 1725 1726 fail: 1727 nlmsg_free(msg); 1728 return ret; 1729 } 1730 1731 static int mptcp_nl_cmd_dump_addrs(struct sk_buff *msg, 1732 struct netlink_callback *cb) 1733 { 1734 struct net *net = sock_net(msg->sk); 1735 struct mptcp_pm_addr_entry *entry; 1736 struct pm_nl_pernet *pernet; 1737 int id = cb->args[0]; 1738 void *hdr; 1739 int i; 1740 1741 pernet = pm_nl_get_pernet(net); 1742 1743 spin_lock_bh(&pernet->lock); 1744 for (i = id; i < MPTCP_PM_MAX_ADDR_ID + 1; i++) { 1745 if (test_bit(i, pernet->id_bitmap)) { 1746 entry = __lookup_addr_by_id(pernet, i); 1747 if (!entry) 1748 break; 1749 1750 if (entry->addr.id <= id) 1751 continue; 1752 1753 hdr = genlmsg_put(msg, NETLINK_CB(cb->skb).portid, 1754 cb->nlh->nlmsg_seq, &mptcp_genl_family, 1755 NLM_F_MULTI, MPTCP_PM_CMD_GET_ADDR); 1756 if (!hdr) 1757 break; 1758 1759 if (mptcp_nl_fill_addr(msg, entry) < 0) { 1760 genlmsg_cancel(msg, hdr); 1761 break; 1762 } 1763 1764 id = entry->addr.id; 1765 genlmsg_end(msg, hdr); 1766 } 1767 } 1768 spin_unlock_bh(&pernet->lock); 1769 1770 cb->args[0] = id; 1771 return msg->len; 1772 } 1773 1774 static int parse_limit(struct genl_info *info, int id, unsigned int *limit) 1775 { 1776 struct nlattr *attr = info->attrs[id]; 1777 1778 if (!attr) 1779 return 0; 1780 1781 *limit = nla_get_u32(attr); 1782 if (*limit > MPTCP_PM_ADDR_MAX) { 1783 GENL_SET_ERR_MSG(info, "limit greater than maximum"); 1784 return -EINVAL; 1785 } 1786 return 0; 1787 } 1788 1789 static int 1790 mptcp_nl_cmd_set_limits(struct sk_buff *skb, struct genl_info *info) 1791 { 1792 struct pm_nl_pernet *pernet = genl_info_pm_nl(info); 1793 unsigned int rcv_addrs, subflows; 1794 int ret; 1795 1796 spin_lock_bh(&pernet->lock); 1797 rcv_addrs = pernet->add_addr_accept_max; 1798 ret = parse_limit(info, MPTCP_PM_ATTR_RCV_ADD_ADDRS, &rcv_addrs); 1799 if (ret) 1800 goto unlock; 1801 1802 subflows = pernet->subflows_max; 1803 ret = parse_limit(info, MPTCP_PM_ATTR_SUBFLOWS, &subflows); 1804 if (ret) 1805 goto unlock; 1806 1807 WRITE_ONCE(pernet->add_addr_accept_max, rcv_addrs); 1808 WRITE_ONCE(pernet->subflows_max, subflows); 1809 1810 unlock: 1811 spin_unlock_bh(&pernet->lock); 1812 return ret; 1813 } 1814 1815 static int 1816 mptcp_nl_cmd_get_limits(struct sk_buff *skb, struct genl_info *info) 1817 { 1818 struct pm_nl_pernet *pernet = genl_info_pm_nl(info); 1819 struct sk_buff *msg; 1820 void *reply; 1821 1822 msg = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL); 1823 if (!msg) 1824 return -ENOMEM; 1825 1826 reply = genlmsg_put_reply(msg, info, &mptcp_genl_family, 0, 1827 MPTCP_PM_CMD_GET_LIMITS); 1828 if (!reply) 1829 goto fail; 1830 1831 if (nla_put_u32(msg, MPTCP_PM_ATTR_RCV_ADD_ADDRS, 1832 READ_ONCE(pernet->add_addr_accept_max))) 1833 goto fail; 1834 1835 if (nla_put_u32(msg, MPTCP_PM_ATTR_SUBFLOWS, 1836 READ_ONCE(pernet->subflows_max))) 1837 goto fail; 1838 1839 genlmsg_end(msg, reply); 1840 return genlmsg_reply(msg, info); 1841 1842 fail: 1843 GENL_SET_ERR_MSG(info, "not enough space in Netlink message"); 1844 nlmsg_free(msg); 1845 return -EMSGSIZE; 1846 } 1847 1848 static void mptcp_pm_nl_fullmesh(struct mptcp_sock *msk, 1849 struct mptcp_addr_info *addr) 1850 { 1851 struct mptcp_rm_list list = { .nr = 0 }; 1852 1853 list.ids[list.nr++] = addr->id; 1854 1855 spin_lock_bh(&msk->pm.lock); 1856 mptcp_pm_nl_rm_subflow_received(msk, &list); 1857 mptcp_pm_create_subflow_or_signal_addr(msk); 1858 spin_unlock_bh(&msk->pm.lock); 1859 } 1860 1861 static int mptcp_nl_set_flags(struct net *net, 1862 struct mptcp_addr_info *addr, 1863 u8 bkup, u8 changed) 1864 { 1865 long s_slot = 0, s_num = 0; 1866 struct mptcp_sock *msk; 1867 int ret = -EINVAL; 1868 1869 while ((msk = mptcp_token_iter_next(net, &s_slot, &s_num)) != NULL) { 1870 struct sock *sk = (struct sock *)msk; 1871 1872 if (list_empty(&msk->conn_list) || mptcp_pm_is_userspace(msk)) 1873 goto next; 1874 1875 lock_sock(sk); 1876 if (changed & MPTCP_PM_ADDR_FLAG_BACKUP) 1877 ret = mptcp_pm_nl_mp_prio_send_ack(msk, addr, NULL, bkup); 1878 if (changed & MPTCP_PM_ADDR_FLAG_FULLMESH) 1879 mptcp_pm_nl_fullmesh(msk, addr); 1880 release_sock(sk); 1881 1882 next: 1883 sock_put(sk); 1884 cond_resched(); 1885 } 1886 1887 return ret; 1888 } 1889 1890 int mptcp_pm_nl_set_flags(struct net *net, struct mptcp_pm_addr_entry *addr, u8 bkup) 1891 { 1892 struct pm_nl_pernet *pernet = pm_nl_get_pernet(net); 1893 u8 changed, mask = MPTCP_PM_ADDR_FLAG_BACKUP | 1894 MPTCP_PM_ADDR_FLAG_FULLMESH; 1895 struct mptcp_pm_addr_entry *entry; 1896 u8 lookup_by_id = 0; 1897 1898 if (addr->addr.family == AF_UNSPEC) { 1899 lookup_by_id = 1; 1900 if (!addr->addr.id) 1901 return -EOPNOTSUPP; 1902 } 1903 1904 spin_lock_bh(&pernet->lock); 1905 entry = __lookup_addr(pernet, &addr->addr, lookup_by_id); 1906 if (!entry) { 1907 spin_unlock_bh(&pernet->lock); 1908 return -EINVAL; 1909 } 1910 if ((addr->flags & MPTCP_PM_ADDR_FLAG_FULLMESH) && 1911 (entry->flags & MPTCP_PM_ADDR_FLAG_SIGNAL)) { 1912 spin_unlock_bh(&pernet->lock); 1913 return -EINVAL; 1914 } 1915 1916 changed = (addr->flags ^ entry->flags) & mask; 1917 entry->flags = (entry->flags & ~mask) | (addr->flags & mask); 1918 *addr = *entry; 1919 spin_unlock_bh(&pernet->lock); 1920 1921 mptcp_nl_set_flags(net, &addr->addr, bkup, changed); 1922 return 0; 1923 } 1924 1925 static int mptcp_nl_cmd_set_flags(struct sk_buff *skb, struct genl_info *info) 1926 { 1927 struct mptcp_pm_addr_entry remote = { .addr = { .family = AF_UNSPEC }, }; 1928 struct mptcp_pm_addr_entry addr = { .addr = { .family = AF_UNSPEC }, }; 1929 struct nlattr *attr_rem = info->attrs[MPTCP_PM_ATTR_ADDR_REMOTE]; 1930 struct nlattr *token = info->attrs[MPTCP_PM_ATTR_TOKEN]; 1931 struct nlattr *attr = info->attrs[MPTCP_PM_ATTR_ADDR]; 1932 struct net *net = sock_net(skb->sk); 1933 u8 bkup = 0; 1934 int ret; 1935 1936 ret = mptcp_pm_parse_entry(attr, info, false, &addr); 1937 if (ret < 0) 1938 return ret; 1939 1940 if (attr_rem) { 1941 ret = mptcp_pm_parse_entry(attr_rem, info, false, &remote); 1942 if (ret < 0) 1943 return ret; 1944 } 1945 1946 if (addr.flags & MPTCP_PM_ADDR_FLAG_BACKUP) 1947 bkup = 1; 1948 1949 return mptcp_pm_set_flags(net, token, &addr, &remote, bkup); 1950 } 1951 1952 static void mptcp_nl_mcast_send(struct net *net, struct sk_buff *nlskb, gfp_t gfp) 1953 { 1954 genlmsg_multicast_netns(&mptcp_genl_family, net, 1955 nlskb, 0, MPTCP_PM_EV_GRP_OFFSET, gfp); 1956 } 1957 1958 bool mptcp_userspace_pm_active(const struct mptcp_sock *msk) 1959 { 1960 return genl_has_listeners(&mptcp_genl_family, 1961 sock_net((const struct sock *)msk), 1962 MPTCP_PM_EV_GRP_OFFSET); 1963 } 1964 1965 static int mptcp_event_add_subflow(struct sk_buff *skb, const struct sock *ssk) 1966 { 1967 const struct inet_sock *issk = inet_sk(ssk); 1968 const struct mptcp_subflow_context *sf; 1969 1970 if (nla_put_u16(skb, MPTCP_ATTR_FAMILY, ssk->sk_family)) 1971 return -EMSGSIZE; 1972 1973 switch (ssk->sk_family) { 1974 case AF_INET: 1975 if (nla_put_in_addr(skb, MPTCP_ATTR_SADDR4, issk->inet_saddr)) 1976 return -EMSGSIZE; 1977 if (nla_put_in_addr(skb, MPTCP_ATTR_DADDR4, issk->inet_daddr)) 1978 return -EMSGSIZE; 1979 break; 1980 #if IS_ENABLED(CONFIG_MPTCP_IPV6) 1981 case AF_INET6: { 1982 const struct ipv6_pinfo *np = inet6_sk(ssk); 1983 1984 if (nla_put_in6_addr(skb, MPTCP_ATTR_SADDR6, &np->saddr)) 1985 return -EMSGSIZE; 1986 if (nla_put_in6_addr(skb, MPTCP_ATTR_DADDR6, &ssk->sk_v6_daddr)) 1987 return -EMSGSIZE; 1988 break; 1989 } 1990 #endif 1991 default: 1992 WARN_ON_ONCE(1); 1993 return -EMSGSIZE; 1994 } 1995 1996 if (nla_put_be16(skb, MPTCP_ATTR_SPORT, issk->inet_sport)) 1997 return -EMSGSIZE; 1998 if (nla_put_be16(skb, MPTCP_ATTR_DPORT, issk->inet_dport)) 1999 return -EMSGSIZE; 2000 2001 sf = mptcp_subflow_ctx(ssk); 2002 if (WARN_ON_ONCE(!sf)) 2003 return -EINVAL; 2004 2005 if (nla_put_u8(skb, MPTCP_ATTR_LOC_ID, sf->local_id)) 2006 return -EMSGSIZE; 2007 2008 if (nla_put_u8(skb, MPTCP_ATTR_REM_ID, sf->remote_id)) 2009 return -EMSGSIZE; 2010 2011 return 0; 2012 } 2013 2014 static int mptcp_event_put_token_and_ssk(struct sk_buff *skb, 2015 const struct mptcp_sock *msk, 2016 const struct sock *ssk) 2017 { 2018 const struct sock *sk = (const struct sock *)msk; 2019 const struct mptcp_subflow_context *sf; 2020 u8 sk_err; 2021 2022 if (nla_put_u32(skb, MPTCP_ATTR_TOKEN, msk->token)) 2023 return -EMSGSIZE; 2024 2025 if (mptcp_event_add_subflow(skb, ssk)) 2026 return -EMSGSIZE; 2027 2028 sf = mptcp_subflow_ctx(ssk); 2029 if (WARN_ON_ONCE(!sf)) 2030 return -EINVAL; 2031 2032 if (nla_put_u8(skb, MPTCP_ATTR_BACKUP, sf->backup)) 2033 return -EMSGSIZE; 2034 2035 if (ssk->sk_bound_dev_if && 2036 nla_put_s32(skb, MPTCP_ATTR_IF_IDX, ssk->sk_bound_dev_if)) 2037 return -EMSGSIZE; 2038 2039 sk_err = READ_ONCE(ssk->sk_err); 2040 if (sk_err && sk->sk_state == TCP_ESTABLISHED && 2041 nla_put_u8(skb, MPTCP_ATTR_ERROR, sk_err)) 2042 return -EMSGSIZE; 2043 2044 return 0; 2045 } 2046 2047 static int mptcp_event_sub_established(struct sk_buff *skb, 2048 const struct mptcp_sock *msk, 2049 const struct sock *ssk) 2050 { 2051 return mptcp_event_put_token_and_ssk(skb, msk, ssk); 2052 } 2053 2054 static int mptcp_event_sub_closed(struct sk_buff *skb, 2055 const struct mptcp_sock *msk, 2056 const struct sock *ssk) 2057 { 2058 const struct mptcp_subflow_context *sf; 2059 2060 if (mptcp_event_put_token_and_ssk(skb, msk, ssk)) 2061 return -EMSGSIZE; 2062 2063 sf = mptcp_subflow_ctx(ssk); 2064 if (!sf->reset_seen) 2065 return 0; 2066 2067 if (nla_put_u32(skb, MPTCP_ATTR_RESET_REASON, sf->reset_reason)) 2068 return -EMSGSIZE; 2069 2070 if (nla_put_u32(skb, MPTCP_ATTR_RESET_FLAGS, sf->reset_transient)) 2071 return -EMSGSIZE; 2072 2073 return 0; 2074 } 2075 2076 static int mptcp_event_created(struct sk_buff *skb, 2077 const struct mptcp_sock *msk, 2078 const struct sock *ssk) 2079 { 2080 int err = nla_put_u32(skb, MPTCP_ATTR_TOKEN, msk->token); 2081 2082 if (err) 2083 return err; 2084 2085 if (nla_put_u8(skb, MPTCP_ATTR_SERVER_SIDE, READ_ONCE(msk->pm.server_side))) 2086 return -EMSGSIZE; 2087 2088 return mptcp_event_add_subflow(skb, ssk); 2089 } 2090 2091 void mptcp_event_addr_removed(const struct mptcp_sock *msk, uint8_t id) 2092 { 2093 struct net *net = sock_net((const struct sock *)msk); 2094 struct nlmsghdr *nlh; 2095 struct sk_buff *skb; 2096 2097 if (!genl_has_listeners(&mptcp_genl_family, net, MPTCP_PM_EV_GRP_OFFSET)) 2098 return; 2099 2100 skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_ATOMIC); 2101 if (!skb) 2102 return; 2103 2104 nlh = genlmsg_put(skb, 0, 0, &mptcp_genl_family, 0, MPTCP_EVENT_REMOVED); 2105 if (!nlh) 2106 goto nla_put_failure; 2107 2108 if (nla_put_u32(skb, MPTCP_ATTR_TOKEN, msk->token)) 2109 goto nla_put_failure; 2110 2111 if (nla_put_u8(skb, MPTCP_ATTR_REM_ID, id)) 2112 goto nla_put_failure; 2113 2114 genlmsg_end(skb, nlh); 2115 mptcp_nl_mcast_send(net, skb, GFP_ATOMIC); 2116 return; 2117 2118 nla_put_failure: 2119 nlmsg_free(skb); 2120 } 2121 2122 void mptcp_event_addr_announced(const struct sock *ssk, 2123 const struct mptcp_addr_info *info) 2124 { 2125 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk); 2126 struct mptcp_sock *msk = mptcp_sk(subflow->conn); 2127 struct net *net = sock_net(ssk); 2128 struct nlmsghdr *nlh; 2129 struct sk_buff *skb; 2130 2131 if (!genl_has_listeners(&mptcp_genl_family, net, MPTCP_PM_EV_GRP_OFFSET)) 2132 return; 2133 2134 skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_ATOMIC); 2135 if (!skb) 2136 return; 2137 2138 nlh = genlmsg_put(skb, 0, 0, &mptcp_genl_family, 0, 2139 MPTCP_EVENT_ANNOUNCED); 2140 if (!nlh) 2141 goto nla_put_failure; 2142 2143 if (nla_put_u32(skb, MPTCP_ATTR_TOKEN, msk->token)) 2144 goto nla_put_failure; 2145 2146 if (nla_put_u8(skb, MPTCP_ATTR_REM_ID, info->id)) 2147 goto nla_put_failure; 2148 2149 if (nla_put_be16(skb, MPTCP_ATTR_DPORT, 2150 info->port == 0 ? 2151 inet_sk(ssk)->inet_dport : 2152 info->port)) 2153 goto nla_put_failure; 2154 2155 switch (info->family) { 2156 case AF_INET: 2157 if (nla_put_in_addr(skb, MPTCP_ATTR_DADDR4, info->addr.s_addr)) 2158 goto nla_put_failure; 2159 break; 2160 #if IS_ENABLED(CONFIG_MPTCP_IPV6) 2161 case AF_INET6: 2162 if (nla_put_in6_addr(skb, MPTCP_ATTR_DADDR6, &info->addr6)) 2163 goto nla_put_failure; 2164 break; 2165 #endif 2166 default: 2167 WARN_ON_ONCE(1); 2168 goto nla_put_failure; 2169 } 2170 2171 genlmsg_end(skb, nlh); 2172 mptcp_nl_mcast_send(net, skb, GFP_ATOMIC); 2173 return; 2174 2175 nla_put_failure: 2176 nlmsg_free(skb); 2177 } 2178 2179 void mptcp_event_pm_listener(const struct sock *ssk, 2180 enum mptcp_event_type event) 2181 { 2182 const struct inet_sock *issk = inet_sk(ssk); 2183 struct net *net = sock_net(ssk); 2184 struct nlmsghdr *nlh; 2185 struct sk_buff *skb; 2186 2187 if (!genl_has_listeners(&mptcp_genl_family, net, MPTCP_PM_EV_GRP_OFFSET)) 2188 return; 2189 2190 skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL); 2191 if (!skb) 2192 return; 2193 2194 nlh = genlmsg_put(skb, 0, 0, &mptcp_genl_family, 0, event); 2195 if (!nlh) 2196 goto nla_put_failure; 2197 2198 if (nla_put_u16(skb, MPTCP_ATTR_FAMILY, ssk->sk_family)) 2199 goto nla_put_failure; 2200 2201 if (nla_put_be16(skb, MPTCP_ATTR_SPORT, issk->inet_sport)) 2202 goto nla_put_failure; 2203 2204 switch (ssk->sk_family) { 2205 case AF_INET: 2206 if (nla_put_in_addr(skb, MPTCP_ATTR_SADDR4, issk->inet_saddr)) 2207 goto nla_put_failure; 2208 break; 2209 #if IS_ENABLED(CONFIG_MPTCP_IPV6) 2210 case AF_INET6: { 2211 const struct ipv6_pinfo *np = inet6_sk(ssk); 2212 2213 if (nla_put_in6_addr(skb, MPTCP_ATTR_SADDR6, &np->saddr)) 2214 goto nla_put_failure; 2215 break; 2216 } 2217 #endif 2218 default: 2219 WARN_ON_ONCE(1); 2220 goto nla_put_failure; 2221 } 2222 2223 genlmsg_end(skb, nlh); 2224 mptcp_nl_mcast_send(net, skb, GFP_KERNEL); 2225 return; 2226 2227 nla_put_failure: 2228 nlmsg_free(skb); 2229 } 2230 2231 void mptcp_event(enum mptcp_event_type type, const struct mptcp_sock *msk, 2232 const struct sock *ssk, gfp_t gfp) 2233 { 2234 struct net *net = sock_net((const struct sock *)msk); 2235 struct nlmsghdr *nlh; 2236 struct sk_buff *skb; 2237 2238 if (!genl_has_listeners(&mptcp_genl_family, net, MPTCP_PM_EV_GRP_OFFSET)) 2239 return; 2240 2241 skb = nlmsg_new(NLMSG_DEFAULT_SIZE, gfp); 2242 if (!skb) 2243 return; 2244 2245 nlh = genlmsg_put(skb, 0, 0, &mptcp_genl_family, 0, type); 2246 if (!nlh) 2247 goto nla_put_failure; 2248 2249 switch (type) { 2250 case MPTCP_EVENT_UNSPEC: 2251 WARN_ON_ONCE(1); 2252 break; 2253 case MPTCP_EVENT_CREATED: 2254 case MPTCP_EVENT_ESTABLISHED: 2255 if (mptcp_event_created(skb, msk, ssk) < 0) 2256 goto nla_put_failure; 2257 break; 2258 case MPTCP_EVENT_CLOSED: 2259 if (nla_put_u32(skb, MPTCP_ATTR_TOKEN, msk->token) < 0) 2260 goto nla_put_failure; 2261 break; 2262 case MPTCP_EVENT_ANNOUNCED: 2263 case MPTCP_EVENT_REMOVED: 2264 /* call mptcp_event_addr_announced()/removed instead */ 2265 WARN_ON_ONCE(1); 2266 break; 2267 case MPTCP_EVENT_SUB_ESTABLISHED: 2268 case MPTCP_EVENT_SUB_PRIORITY: 2269 if (mptcp_event_sub_established(skb, msk, ssk) < 0) 2270 goto nla_put_failure; 2271 break; 2272 case MPTCP_EVENT_SUB_CLOSED: 2273 if (mptcp_event_sub_closed(skb, msk, ssk) < 0) 2274 goto nla_put_failure; 2275 break; 2276 case MPTCP_EVENT_LISTENER_CREATED: 2277 case MPTCP_EVENT_LISTENER_CLOSED: 2278 break; 2279 } 2280 2281 genlmsg_end(skb, nlh); 2282 mptcp_nl_mcast_send(net, skb, gfp); 2283 return; 2284 2285 nla_put_failure: 2286 nlmsg_free(skb); 2287 } 2288 2289 static const struct genl_small_ops mptcp_pm_ops[] = { 2290 { 2291 .cmd = MPTCP_PM_CMD_ADD_ADDR, 2292 .doit = mptcp_nl_cmd_add_addr, 2293 .flags = GENL_UNS_ADMIN_PERM, 2294 }, 2295 { 2296 .cmd = MPTCP_PM_CMD_DEL_ADDR, 2297 .doit = mptcp_nl_cmd_del_addr, 2298 .flags = GENL_UNS_ADMIN_PERM, 2299 }, 2300 { 2301 .cmd = MPTCP_PM_CMD_FLUSH_ADDRS, 2302 .doit = mptcp_nl_cmd_flush_addrs, 2303 .flags = GENL_UNS_ADMIN_PERM, 2304 }, 2305 { 2306 .cmd = MPTCP_PM_CMD_GET_ADDR, 2307 .doit = mptcp_nl_cmd_get_addr, 2308 .dumpit = mptcp_nl_cmd_dump_addrs, 2309 }, 2310 { 2311 .cmd = MPTCP_PM_CMD_SET_LIMITS, 2312 .doit = mptcp_nl_cmd_set_limits, 2313 .flags = GENL_UNS_ADMIN_PERM, 2314 }, 2315 { 2316 .cmd = MPTCP_PM_CMD_GET_LIMITS, 2317 .doit = mptcp_nl_cmd_get_limits, 2318 }, 2319 { 2320 .cmd = MPTCP_PM_CMD_SET_FLAGS, 2321 .doit = mptcp_nl_cmd_set_flags, 2322 .flags = GENL_UNS_ADMIN_PERM, 2323 }, 2324 { 2325 .cmd = MPTCP_PM_CMD_ANNOUNCE, 2326 .doit = mptcp_nl_cmd_announce, 2327 .flags = GENL_UNS_ADMIN_PERM, 2328 }, 2329 { 2330 .cmd = MPTCP_PM_CMD_REMOVE, 2331 .doit = mptcp_nl_cmd_remove, 2332 .flags = GENL_UNS_ADMIN_PERM, 2333 }, 2334 { 2335 .cmd = MPTCP_PM_CMD_SUBFLOW_CREATE, 2336 .doit = mptcp_nl_cmd_sf_create, 2337 .flags = GENL_UNS_ADMIN_PERM, 2338 }, 2339 { 2340 .cmd = MPTCP_PM_CMD_SUBFLOW_DESTROY, 2341 .doit = mptcp_nl_cmd_sf_destroy, 2342 .flags = GENL_UNS_ADMIN_PERM, 2343 }, 2344 }; 2345 2346 static struct genl_family mptcp_genl_family __ro_after_init = { 2347 .name = MPTCP_PM_NAME, 2348 .version = MPTCP_PM_VER, 2349 .maxattr = MPTCP_PM_ATTR_MAX, 2350 .policy = mptcp_pm_policy, 2351 .netnsok = true, 2352 .module = THIS_MODULE, 2353 .small_ops = mptcp_pm_ops, 2354 .n_small_ops = ARRAY_SIZE(mptcp_pm_ops), 2355 .resv_start_op = MPTCP_PM_CMD_SUBFLOW_DESTROY + 1, 2356 .mcgrps = mptcp_pm_mcgrps, 2357 .n_mcgrps = ARRAY_SIZE(mptcp_pm_mcgrps), 2358 }; 2359 2360 static int __net_init pm_nl_init_net(struct net *net) 2361 { 2362 struct pm_nl_pernet *pernet = pm_nl_get_pernet(net); 2363 2364 INIT_LIST_HEAD_RCU(&pernet->local_addr_list); 2365 2366 /* Cit. 2 subflows ought to be enough for anybody. */ 2367 pernet->subflows_max = 2; 2368 pernet->next_id = 1; 2369 pernet->stale_loss_cnt = 4; 2370 spin_lock_init(&pernet->lock); 2371 2372 /* No need to initialize other pernet fields, the struct is zeroed at 2373 * allocation time. 2374 */ 2375 2376 return 0; 2377 } 2378 2379 static void __net_exit pm_nl_exit_net(struct list_head *net_list) 2380 { 2381 struct net *net; 2382 2383 list_for_each_entry(net, net_list, exit_list) { 2384 struct pm_nl_pernet *pernet = pm_nl_get_pernet(net); 2385 2386 /* net is removed from namespace list, can't race with 2387 * other modifiers, also netns core already waited for a 2388 * RCU grace period. 2389 */ 2390 __flush_addrs(&pernet->local_addr_list); 2391 } 2392 } 2393 2394 static struct pernet_operations mptcp_pm_pernet_ops = { 2395 .init = pm_nl_init_net, 2396 .exit_batch = pm_nl_exit_net, 2397 .id = &pm_nl_pernet_id, 2398 .size = sizeof(struct pm_nl_pernet), 2399 }; 2400 2401 void __init mptcp_pm_nl_init(void) 2402 { 2403 if (register_pernet_subsys(&mptcp_pm_pernet_ops) < 0) 2404 panic("Failed to register MPTCP PM pernet subsystem.\n"); 2405 2406 if (genl_register_family(&mptcp_genl_family)) 2407 panic("Failed to register MPTCP PM netlink family\n"); 2408 } 2409