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 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 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_pm_addr_entry *entry) 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, &entry->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 = entry->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 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)) { 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 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 socket *ssock; 1010 struct sock *newsk; 1011 int backlog = 1024; 1012 int err; 1013 1014 err = sock_create_kern(sock_net(sk), entry->addr.family, 1015 SOCK_STREAM, IPPROTO_MPTCP, &entry->lsk); 1016 if (err) 1017 return err; 1018 1019 newsk = entry->lsk->sk; 1020 if (!newsk) 1021 return -EINVAL; 1022 1023 /* The subflow socket lock is acquired in a nested to the msk one 1024 * in several places, even by the TCP stack, and this msk is a kernel 1025 * socket: lockdep complains. Instead of propagating the _nested 1026 * modifiers in several places, re-init the lock class for the msk 1027 * socket to an mptcp specific one. 1028 */ 1029 sock_lock_init_class_and_name(newsk, 1030 is_ipv6 ? "mlock-AF_INET6" : "mlock-AF_INET", 1031 &mptcp_slock_keys[is_ipv6], 1032 is_ipv6 ? "msk_lock-AF_INET6" : "msk_lock-AF_INET", 1033 &mptcp_keys[is_ipv6]); 1034 1035 lock_sock(newsk); 1036 ssock = __mptcp_nmpc_socket(mptcp_sk(newsk)); 1037 release_sock(newsk); 1038 if (IS_ERR(ssock)) 1039 return PTR_ERR(ssock); 1040 1041 mptcp_info2sockaddr(&entry->addr, &addr, entry->addr.family); 1042 #if IS_ENABLED(CONFIG_MPTCP_IPV6) 1043 if (entry->addr.family == AF_INET6) 1044 addrlen = sizeof(struct sockaddr_in6); 1045 #endif 1046 err = kernel_bind(ssock, (struct sockaddr *)&addr, addrlen); 1047 if (err) 1048 return err; 1049 1050 inet_sk_state_store(newsk, TCP_LISTEN); 1051 err = kernel_listen(ssock, backlog); 1052 if (err) 1053 return err; 1054 1055 mptcp_event_pm_listener(ssock->sk, MPTCP_EVENT_LISTENER_CREATED); 1056 1057 return 0; 1058 } 1059 1060 int mptcp_pm_nl_get_local_id(struct mptcp_sock *msk, struct sock_common *skc) 1061 { 1062 struct mptcp_pm_addr_entry *entry; 1063 struct mptcp_addr_info skc_local; 1064 struct mptcp_addr_info msk_local; 1065 struct pm_nl_pernet *pernet; 1066 int ret = -1; 1067 1068 if (WARN_ON_ONCE(!msk)) 1069 return -1; 1070 1071 /* The 0 ID mapping is defined by the first subflow, copied into the msk 1072 * addr 1073 */ 1074 local_address((struct sock_common *)msk, &msk_local); 1075 local_address((struct sock_common *)skc, &skc_local); 1076 if (mptcp_addresses_equal(&msk_local, &skc_local, false)) 1077 return 0; 1078 1079 if (mptcp_pm_is_userspace(msk)) 1080 return mptcp_userspace_pm_get_local_id(msk, &skc_local); 1081 1082 pernet = pm_nl_get_pernet_from_msk(msk); 1083 1084 rcu_read_lock(); 1085 list_for_each_entry_rcu(entry, &pernet->local_addr_list, list) { 1086 if (mptcp_addresses_equal(&entry->addr, &skc_local, entry->addr.port)) { 1087 ret = entry->addr.id; 1088 break; 1089 } 1090 } 1091 rcu_read_unlock(); 1092 if (ret >= 0) 1093 return ret; 1094 1095 /* address not found, add to local list */ 1096 entry = kmalloc(sizeof(*entry), GFP_ATOMIC); 1097 if (!entry) 1098 return -ENOMEM; 1099 1100 entry->addr = skc_local; 1101 entry->addr.id = 0; 1102 entry->addr.port = 0; 1103 entry->ifindex = 0; 1104 entry->flags = MPTCP_PM_ADDR_FLAG_IMPLICIT; 1105 entry->lsk = NULL; 1106 ret = mptcp_pm_nl_append_new_local_addr(pernet, entry); 1107 if (ret < 0) 1108 kfree(entry); 1109 1110 return ret; 1111 } 1112 1113 #define MPTCP_PM_CMD_GRP_OFFSET 0 1114 #define MPTCP_PM_EV_GRP_OFFSET 1 1115 1116 static const struct genl_multicast_group mptcp_pm_mcgrps[] = { 1117 [MPTCP_PM_CMD_GRP_OFFSET] = { .name = MPTCP_PM_CMD_GRP_NAME, }, 1118 [MPTCP_PM_EV_GRP_OFFSET] = { .name = MPTCP_PM_EV_GRP_NAME, 1119 .flags = GENL_UNS_ADMIN_PERM, 1120 }, 1121 }; 1122 1123 static const struct nla_policy 1124 mptcp_pm_addr_policy[MPTCP_PM_ADDR_ATTR_MAX + 1] = { 1125 [MPTCP_PM_ADDR_ATTR_FAMILY] = { .type = NLA_U16, }, 1126 [MPTCP_PM_ADDR_ATTR_ID] = { .type = NLA_U8, }, 1127 [MPTCP_PM_ADDR_ATTR_ADDR4] = { .type = NLA_U32, }, 1128 [MPTCP_PM_ADDR_ATTR_ADDR6] = 1129 NLA_POLICY_EXACT_LEN(sizeof(struct in6_addr)), 1130 [MPTCP_PM_ADDR_ATTR_PORT] = { .type = NLA_U16 }, 1131 [MPTCP_PM_ADDR_ATTR_FLAGS] = { .type = NLA_U32 }, 1132 [MPTCP_PM_ADDR_ATTR_IF_IDX] = { .type = NLA_S32 }, 1133 }; 1134 1135 static const struct nla_policy mptcp_pm_policy[MPTCP_PM_ATTR_MAX + 1] = { 1136 [MPTCP_PM_ATTR_ADDR] = 1137 NLA_POLICY_NESTED(mptcp_pm_addr_policy), 1138 [MPTCP_PM_ATTR_RCV_ADD_ADDRS] = { .type = NLA_U32, }, 1139 [MPTCP_PM_ATTR_SUBFLOWS] = { .type = NLA_U32, }, 1140 [MPTCP_PM_ATTR_TOKEN] = { .type = NLA_U32, }, 1141 [MPTCP_PM_ATTR_LOC_ID] = { .type = NLA_U8, }, 1142 [MPTCP_PM_ATTR_ADDR_REMOTE] = 1143 NLA_POLICY_NESTED(mptcp_pm_addr_policy), 1144 }; 1145 1146 void mptcp_pm_nl_subflow_chk_stale(const struct mptcp_sock *msk, struct sock *ssk) 1147 { 1148 struct mptcp_subflow_context *iter, *subflow = mptcp_subflow_ctx(ssk); 1149 struct sock *sk = (struct sock *)msk; 1150 unsigned int active_max_loss_cnt; 1151 struct net *net = sock_net(sk); 1152 unsigned int stale_loss_cnt; 1153 bool slow; 1154 1155 stale_loss_cnt = mptcp_stale_loss_cnt(net); 1156 if (subflow->stale || !stale_loss_cnt || subflow->stale_count <= stale_loss_cnt) 1157 return; 1158 1159 /* look for another available subflow not in loss state */ 1160 active_max_loss_cnt = max_t(int, stale_loss_cnt - 1, 1); 1161 mptcp_for_each_subflow(msk, iter) { 1162 if (iter != subflow && mptcp_subflow_active(iter) && 1163 iter->stale_count < active_max_loss_cnt) { 1164 /* we have some alternatives, try to mark this subflow as idle ...*/ 1165 slow = lock_sock_fast(ssk); 1166 if (!tcp_rtx_and_write_queues_empty(ssk)) { 1167 subflow->stale = 1; 1168 __mptcp_retransmit_pending_data(sk); 1169 MPTCP_INC_STATS(net, MPTCP_MIB_SUBFLOWSTALE); 1170 } 1171 unlock_sock_fast(ssk, slow); 1172 1173 /* always try to push the pending data regardless of re-injections: 1174 * we can possibly use backup subflows now, and subflow selection 1175 * is cheap under the msk socket lock 1176 */ 1177 __mptcp_push_pending(sk, 0); 1178 return; 1179 } 1180 } 1181 } 1182 1183 static int mptcp_pm_family_to_addr(int family) 1184 { 1185 #if IS_ENABLED(CONFIG_MPTCP_IPV6) 1186 if (family == AF_INET6) 1187 return MPTCP_PM_ADDR_ATTR_ADDR6; 1188 #endif 1189 return MPTCP_PM_ADDR_ATTR_ADDR4; 1190 } 1191 1192 static int mptcp_pm_parse_pm_addr_attr(struct nlattr *tb[], 1193 const struct nlattr *attr, 1194 struct genl_info *info, 1195 struct mptcp_addr_info *addr, 1196 bool require_family) 1197 { 1198 int err, addr_addr; 1199 1200 if (!attr) { 1201 GENL_SET_ERR_MSG(info, "missing address info"); 1202 return -EINVAL; 1203 } 1204 1205 /* no validation needed - was already done via nested policy */ 1206 err = nla_parse_nested_deprecated(tb, MPTCP_PM_ADDR_ATTR_MAX, attr, 1207 mptcp_pm_addr_policy, info->extack); 1208 if (err) 1209 return err; 1210 1211 if (tb[MPTCP_PM_ADDR_ATTR_ID]) 1212 addr->id = nla_get_u8(tb[MPTCP_PM_ADDR_ATTR_ID]); 1213 1214 if (!tb[MPTCP_PM_ADDR_ATTR_FAMILY]) { 1215 if (!require_family) 1216 return 0; 1217 1218 NL_SET_ERR_MSG_ATTR(info->extack, attr, 1219 "missing family"); 1220 return -EINVAL; 1221 } 1222 1223 addr->family = nla_get_u16(tb[MPTCP_PM_ADDR_ATTR_FAMILY]); 1224 if (addr->family != AF_INET 1225 #if IS_ENABLED(CONFIG_MPTCP_IPV6) 1226 && addr->family != AF_INET6 1227 #endif 1228 ) { 1229 NL_SET_ERR_MSG_ATTR(info->extack, attr, 1230 "unknown address family"); 1231 return -EINVAL; 1232 } 1233 addr_addr = mptcp_pm_family_to_addr(addr->family); 1234 if (!tb[addr_addr]) { 1235 NL_SET_ERR_MSG_ATTR(info->extack, attr, 1236 "missing address data"); 1237 return -EINVAL; 1238 } 1239 1240 #if IS_ENABLED(CONFIG_MPTCP_IPV6) 1241 if (addr->family == AF_INET6) 1242 addr->addr6 = nla_get_in6_addr(tb[addr_addr]); 1243 else 1244 #endif 1245 addr->addr.s_addr = nla_get_in_addr(tb[addr_addr]); 1246 1247 if (tb[MPTCP_PM_ADDR_ATTR_PORT]) 1248 addr->port = htons(nla_get_u16(tb[MPTCP_PM_ADDR_ATTR_PORT])); 1249 1250 return 0; 1251 } 1252 1253 int mptcp_pm_parse_addr(struct nlattr *attr, struct genl_info *info, 1254 struct mptcp_addr_info *addr) 1255 { 1256 struct nlattr *tb[MPTCP_PM_ADDR_ATTR_MAX + 1]; 1257 1258 memset(addr, 0, sizeof(*addr)); 1259 1260 return mptcp_pm_parse_pm_addr_attr(tb, attr, info, addr, true); 1261 } 1262 1263 int mptcp_pm_parse_entry(struct nlattr *attr, struct genl_info *info, 1264 bool require_family, 1265 struct mptcp_pm_addr_entry *entry) 1266 { 1267 struct nlattr *tb[MPTCP_PM_ADDR_ATTR_MAX + 1]; 1268 int err; 1269 1270 memset(entry, 0, sizeof(*entry)); 1271 1272 err = mptcp_pm_parse_pm_addr_attr(tb, attr, info, &entry->addr, require_family); 1273 if (err) 1274 return err; 1275 1276 if (tb[MPTCP_PM_ADDR_ATTR_IF_IDX]) { 1277 u32 val = nla_get_s32(tb[MPTCP_PM_ADDR_ATTR_IF_IDX]); 1278 1279 entry->ifindex = val; 1280 } 1281 1282 if (tb[MPTCP_PM_ADDR_ATTR_FLAGS]) 1283 entry->flags = nla_get_u32(tb[MPTCP_PM_ADDR_ATTR_FLAGS]); 1284 1285 if (tb[MPTCP_PM_ADDR_ATTR_PORT]) 1286 entry->addr.port = htons(nla_get_u16(tb[MPTCP_PM_ADDR_ATTR_PORT])); 1287 1288 return 0; 1289 } 1290 1291 static struct pm_nl_pernet *genl_info_pm_nl(struct genl_info *info) 1292 { 1293 return pm_nl_get_pernet(genl_info_net(info)); 1294 } 1295 1296 static int mptcp_nl_add_subflow_or_signal_addr(struct net *net) 1297 { 1298 struct mptcp_sock *msk; 1299 long s_slot = 0, s_num = 0; 1300 1301 while ((msk = mptcp_token_iter_next(net, &s_slot, &s_num)) != NULL) { 1302 struct sock *sk = (struct sock *)msk; 1303 1304 if (!READ_ONCE(msk->fully_established) || 1305 mptcp_pm_is_userspace(msk)) 1306 goto next; 1307 1308 lock_sock(sk); 1309 spin_lock_bh(&msk->pm.lock); 1310 mptcp_pm_create_subflow_or_signal_addr(msk); 1311 spin_unlock_bh(&msk->pm.lock); 1312 release_sock(sk); 1313 1314 next: 1315 sock_put(sk); 1316 cond_resched(); 1317 } 1318 1319 return 0; 1320 } 1321 1322 static int mptcp_nl_cmd_add_addr(struct sk_buff *skb, struct genl_info *info) 1323 { 1324 struct nlattr *attr = info->attrs[MPTCP_PM_ATTR_ADDR]; 1325 struct pm_nl_pernet *pernet = genl_info_pm_nl(info); 1326 struct mptcp_pm_addr_entry addr, *entry; 1327 int ret; 1328 1329 ret = mptcp_pm_parse_entry(attr, info, true, &addr); 1330 if (ret < 0) 1331 return ret; 1332 1333 if (addr.addr.port && !(addr.flags & MPTCP_PM_ADDR_FLAG_SIGNAL)) { 1334 GENL_SET_ERR_MSG(info, "flags must have signal when using port"); 1335 return -EINVAL; 1336 } 1337 1338 if (addr.flags & MPTCP_PM_ADDR_FLAG_SIGNAL && 1339 addr.flags & MPTCP_PM_ADDR_FLAG_FULLMESH) { 1340 GENL_SET_ERR_MSG(info, "flags mustn't have both signal and fullmesh"); 1341 return -EINVAL; 1342 } 1343 1344 if (addr.flags & MPTCP_PM_ADDR_FLAG_IMPLICIT) { 1345 GENL_SET_ERR_MSG(info, "can't create IMPLICIT endpoint"); 1346 return -EINVAL; 1347 } 1348 1349 entry = kzalloc(sizeof(*entry), GFP_KERNEL_ACCOUNT); 1350 if (!entry) { 1351 GENL_SET_ERR_MSG(info, "can't allocate addr"); 1352 return -ENOMEM; 1353 } 1354 1355 *entry = addr; 1356 if (entry->addr.port) { 1357 ret = mptcp_pm_nl_create_listen_socket(skb->sk, entry); 1358 if (ret) { 1359 GENL_SET_ERR_MSG_FMT(info, "create listen socket error: %d", ret); 1360 goto out_free; 1361 } 1362 } 1363 ret = mptcp_pm_nl_append_new_local_addr(pernet, entry); 1364 if (ret < 0) { 1365 GENL_SET_ERR_MSG_FMT(info, "too many addresses or duplicate one: %d", ret); 1366 goto out_free; 1367 } 1368 1369 mptcp_nl_add_subflow_or_signal_addr(sock_net(skb->sk)); 1370 return 0; 1371 1372 out_free: 1373 __mptcp_pm_release_addr_entry(entry); 1374 return ret; 1375 } 1376 1377 int mptcp_pm_get_flags_and_ifindex_by_id(struct mptcp_sock *msk, unsigned int id, 1378 u8 *flags, int *ifindex) 1379 { 1380 struct mptcp_pm_addr_entry *entry; 1381 struct sock *sk = (struct sock *)msk; 1382 struct net *net = sock_net(sk); 1383 1384 *flags = 0; 1385 *ifindex = 0; 1386 1387 if (id) { 1388 if (mptcp_pm_is_userspace(msk)) 1389 return mptcp_userspace_pm_get_flags_and_ifindex_by_id(msk, 1390 id, 1391 flags, 1392 ifindex); 1393 1394 rcu_read_lock(); 1395 entry = __lookup_addr_by_id(pm_nl_get_pernet(net), id); 1396 if (entry) { 1397 *flags = entry->flags; 1398 *ifindex = entry->ifindex; 1399 } 1400 rcu_read_unlock(); 1401 } 1402 1403 return 0; 1404 } 1405 1406 static bool remove_anno_list_by_saddr(struct mptcp_sock *msk, 1407 const struct mptcp_addr_info *addr) 1408 { 1409 struct mptcp_pm_add_entry *entry; 1410 1411 entry = mptcp_pm_del_add_timer(msk, addr, false); 1412 if (entry) { 1413 list_del(&entry->list); 1414 kfree(entry); 1415 return true; 1416 } 1417 1418 return false; 1419 } 1420 1421 static bool mptcp_pm_remove_anno_addr(struct mptcp_sock *msk, 1422 const struct mptcp_addr_info *addr, 1423 bool force) 1424 { 1425 struct mptcp_rm_list list = { .nr = 0 }; 1426 bool ret; 1427 1428 list.ids[list.nr++] = addr->id; 1429 1430 ret = remove_anno_list_by_saddr(msk, addr); 1431 if (ret || force) { 1432 spin_lock_bh(&msk->pm.lock); 1433 mptcp_pm_remove_addr(msk, &list); 1434 spin_unlock_bh(&msk->pm.lock); 1435 } 1436 return ret; 1437 } 1438 1439 static int mptcp_nl_remove_subflow_and_signal_addr(struct net *net, 1440 const struct mptcp_pm_addr_entry *entry) 1441 { 1442 const struct mptcp_addr_info *addr = &entry->addr; 1443 struct mptcp_rm_list list = { .nr = 0 }; 1444 long s_slot = 0, s_num = 0; 1445 struct mptcp_sock *msk; 1446 1447 pr_debug("remove_id=%d", addr->id); 1448 1449 list.ids[list.nr++] = addr->id; 1450 1451 while ((msk = mptcp_token_iter_next(net, &s_slot, &s_num)) != NULL) { 1452 struct sock *sk = (struct sock *)msk; 1453 bool remove_subflow; 1454 1455 if (mptcp_pm_is_userspace(msk)) 1456 goto next; 1457 1458 if (list_empty(&msk->conn_list)) { 1459 mptcp_pm_remove_anno_addr(msk, addr, false); 1460 goto next; 1461 } 1462 1463 lock_sock(sk); 1464 remove_subflow = lookup_subflow_by_saddr(&msk->conn_list, addr); 1465 mptcp_pm_remove_anno_addr(msk, addr, remove_subflow && 1466 !(entry->flags & MPTCP_PM_ADDR_FLAG_IMPLICIT)); 1467 if (remove_subflow) 1468 mptcp_pm_remove_subflow(msk, &list); 1469 release_sock(sk); 1470 1471 next: 1472 sock_put(sk); 1473 cond_resched(); 1474 } 1475 1476 return 0; 1477 } 1478 1479 static int mptcp_nl_remove_id_zero_address(struct net *net, 1480 struct mptcp_addr_info *addr) 1481 { 1482 struct mptcp_rm_list list = { .nr = 0 }; 1483 long s_slot = 0, s_num = 0; 1484 struct mptcp_sock *msk; 1485 1486 list.ids[list.nr++] = 0; 1487 1488 while ((msk = mptcp_token_iter_next(net, &s_slot, &s_num)) != NULL) { 1489 struct sock *sk = (struct sock *)msk; 1490 struct mptcp_addr_info msk_local; 1491 1492 if (list_empty(&msk->conn_list) || mptcp_pm_is_userspace(msk)) 1493 goto next; 1494 1495 local_address((struct sock_common *)msk, &msk_local); 1496 if (!mptcp_addresses_equal(&msk_local, addr, addr->port)) 1497 goto next; 1498 1499 lock_sock(sk); 1500 spin_lock_bh(&msk->pm.lock); 1501 mptcp_pm_remove_addr(msk, &list); 1502 mptcp_pm_nl_rm_subflow_received(msk, &list); 1503 spin_unlock_bh(&msk->pm.lock); 1504 release_sock(sk); 1505 1506 next: 1507 sock_put(sk); 1508 cond_resched(); 1509 } 1510 1511 return 0; 1512 } 1513 1514 static int mptcp_nl_cmd_del_addr(struct sk_buff *skb, struct genl_info *info) 1515 { 1516 struct nlattr *attr = info->attrs[MPTCP_PM_ATTR_ADDR]; 1517 struct pm_nl_pernet *pernet = genl_info_pm_nl(info); 1518 struct mptcp_pm_addr_entry addr, *entry; 1519 unsigned int addr_max; 1520 int ret; 1521 1522 ret = mptcp_pm_parse_entry(attr, info, false, &addr); 1523 if (ret < 0) 1524 return ret; 1525 1526 /* the zero id address is special: the first address used by the msk 1527 * always gets such an id, so different subflows can have different zero 1528 * id addresses. Additionally zero id is not accounted for in id_bitmap. 1529 * Let's use an 'mptcp_rm_list' instead of the common remove code. 1530 */ 1531 if (addr.addr.id == 0) 1532 return mptcp_nl_remove_id_zero_address(sock_net(skb->sk), &addr.addr); 1533 1534 spin_lock_bh(&pernet->lock); 1535 entry = __lookup_addr_by_id(pernet, addr.addr.id); 1536 if (!entry) { 1537 GENL_SET_ERR_MSG(info, "address not found"); 1538 spin_unlock_bh(&pernet->lock); 1539 return -EINVAL; 1540 } 1541 if (entry->flags & MPTCP_PM_ADDR_FLAG_SIGNAL) { 1542 addr_max = pernet->add_addr_signal_max; 1543 WRITE_ONCE(pernet->add_addr_signal_max, addr_max - 1); 1544 } 1545 if (entry->flags & MPTCP_PM_ADDR_FLAG_SUBFLOW) { 1546 addr_max = pernet->local_addr_max; 1547 WRITE_ONCE(pernet->local_addr_max, addr_max - 1); 1548 } 1549 1550 pernet->addrs--; 1551 list_del_rcu(&entry->list); 1552 __clear_bit(entry->addr.id, pernet->id_bitmap); 1553 spin_unlock_bh(&pernet->lock); 1554 1555 mptcp_nl_remove_subflow_and_signal_addr(sock_net(skb->sk), entry); 1556 synchronize_rcu(); 1557 __mptcp_pm_release_addr_entry(entry); 1558 1559 return ret; 1560 } 1561 1562 void mptcp_pm_remove_addrs(struct mptcp_sock *msk, struct list_head *rm_list) 1563 { 1564 struct mptcp_rm_list alist = { .nr = 0 }; 1565 struct mptcp_pm_addr_entry *entry; 1566 1567 list_for_each_entry(entry, rm_list, list) { 1568 remove_anno_list_by_saddr(msk, &entry->addr); 1569 if (alist.nr < MPTCP_RM_IDS_MAX) 1570 alist.ids[alist.nr++] = entry->addr.id; 1571 } 1572 1573 if (alist.nr) { 1574 spin_lock_bh(&msk->pm.lock); 1575 mptcp_pm_remove_addr(msk, &alist); 1576 spin_unlock_bh(&msk->pm.lock); 1577 } 1578 } 1579 1580 void mptcp_pm_remove_addrs_and_subflows(struct mptcp_sock *msk, 1581 struct list_head *rm_list) 1582 { 1583 struct mptcp_rm_list alist = { .nr = 0 }, slist = { .nr = 0 }; 1584 struct mptcp_pm_addr_entry *entry; 1585 1586 list_for_each_entry(entry, rm_list, list) { 1587 if (lookup_subflow_by_saddr(&msk->conn_list, &entry->addr) && 1588 slist.nr < MPTCP_RM_IDS_MAX) 1589 slist.ids[slist.nr++] = entry->addr.id; 1590 1591 if (remove_anno_list_by_saddr(msk, &entry->addr) && 1592 alist.nr < MPTCP_RM_IDS_MAX) 1593 alist.ids[alist.nr++] = entry->addr.id; 1594 } 1595 1596 if (alist.nr) { 1597 spin_lock_bh(&msk->pm.lock); 1598 mptcp_pm_remove_addr(msk, &alist); 1599 spin_unlock_bh(&msk->pm.lock); 1600 } 1601 if (slist.nr) 1602 mptcp_pm_remove_subflow(msk, &slist); 1603 } 1604 1605 static void mptcp_nl_remove_addrs_list(struct net *net, 1606 struct list_head *rm_list) 1607 { 1608 long s_slot = 0, s_num = 0; 1609 struct mptcp_sock *msk; 1610 1611 if (list_empty(rm_list)) 1612 return; 1613 1614 while ((msk = mptcp_token_iter_next(net, &s_slot, &s_num)) != NULL) { 1615 struct sock *sk = (struct sock *)msk; 1616 1617 if (!mptcp_pm_is_userspace(msk)) { 1618 lock_sock(sk); 1619 mptcp_pm_remove_addrs_and_subflows(msk, rm_list); 1620 release_sock(sk); 1621 } 1622 1623 sock_put(sk); 1624 cond_resched(); 1625 } 1626 } 1627 1628 /* caller must ensure the RCU grace period is already elapsed */ 1629 static void __flush_addrs(struct list_head *list) 1630 { 1631 while (!list_empty(list)) { 1632 struct mptcp_pm_addr_entry *cur; 1633 1634 cur = list_entry(list->next, 1635 struct mptcp_pm_addr_entry, list); 1636 list_del_rcu(&cur->list); 1637 __mptcp_pm_release_addr_entry(cur); 1638 } 1639 } 1640 1641 static void __reset_counters(struct pm_nl_pernet *pernet) 1642 { 1643 WRITE_ONCE(pernet->add_addr_signal_max, 0); 1644 WRITE_ONCE(pernet->add_addr_accept_max, 0); 1645 WRITE_ONCE(pernet->local_addr_max, 0); 1646 pernet->addrs = 0; 1647 } 1648 1649 static int mptcp_nl_cmd_flush_addrs(struct sk_buff *skb, struct genl_info *info) 1650 { 1651 struct pm_nl_pernet *pernet = genl_info_pm_nl(info); 1652 LIST_HEAD(free_list); 1653 1654 spin_lock_bh(&pernet->lock); 1655 list_splice_init(&pernet->local_addr_list, &free_list); 1656 __reset_counters(pernet); 1657 pernet->next_id = 1; 1658 bitmap_zero(pernet->id_bitmap, MPTCP_PM_MAX_ADDR_ID + 1); 1659 spin_unlock_bh(&pernet->lock); 1660 mptcp_nl_remove_addrs_list(sock_net(skb->sk), &free_list); 1661 synchronize_rcu(); 1662 __flush_addrs(&free_list); 1663 return 0; 1664 } 1665 1666 static int mptcp_nl_fill_addr(struct sk_buff *skb, 1667 struct mptcp_pm_addr_entry *entry) 1668 { 1669 struct mptcp_addr_info *addr = &entry->addr; 1670 struct nlattr *attr; 1671 1672 attr = nla_nest_start(skb, MPTCP_PM_ATTR_ADDR); 1673 if (!attr) 1674 return -EMSGSIZE; 1675 1676 if (nla_put_u16(skb, MPTCP_PM_ADDR_ATTR_FAMILY, addr->family)) 1677 goto nla_put_failure; 1678 if (nla_put_u16(skb, MPTCP_PM_ADDR_ATTR_PORT, ntohs(addr->port))) 1679 goto nla_put_failure; 1680 if (nla_put_u8(skb, MPTCP_PM_ADDR_ATTR_ID, addr->id)) 1681 goto nla_put_failure; 1682 if (nla_put_u32(skb, MPTCP_PM_ADDR_ATTR_FLAGS, entry->flags)) 1683 goto nla_put_failure; 1684 if (entry->ifindex && 1685 nla_put_s32(skb, MPTCP_PM_ADDR_ATTR_IF_IDX, entry->ifindex)) 1686 goto nla_put_failure; 1687 1688 if (addr->family == AF_INET && 1689 nla_put_in_addr(skb, MPTCP_PM_ADDR_ATTR_ADDR4, 1690 addr->addr.s_addr)) 1691 goto nla_put_failure; 1692 #if IS_ENABLED(CONFIG_MPTCP_IPV6) 1693 else if (addr->family == AF_INET6 && 1694 nla_put_in6_addr(skb, MPTCP_PM_ADDR_ATTR_ADDR6, &addr->addr6)) 1695 goto nla_put_failure; 1696 #endif 1697 nla_nest_end(skb, attr); 1698 return 0; 1699 1700 nla_put_failure: 1701 nla_nest_cancel(skb, attr); 1702 return -EMSGSIZE; 1703 } 1704 1705 static int mptcp_nl_cmd_get_addr(struct sk_buff *skb, struct genl_info *info) 1706 { 1707 struct nlattr *attr = info->attrs[MPTCP_PM_ATTR_ADDR]; 1708 struct pm_nl_pernet *pernet = genl_info_pm_nl(info); 1709 struct mptcp_pm_addr_entry addr, *entry; 1710 struct sk_buff *msg; 1711 void *reply; 1712 int ret; 1713 1714 ret = mptcp_pm_parse_entry(attr, info, false, &addr); 1715 if (ret < 0) 1716 return ret; 1717 1718 msg = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL); 1719 if (!msg) 1720 return -ENOMEM; 1721 1722 reply = genlmsg_put_reply(msg, info, &mptcp_genl_family, 0, 1723 info->genlhdr->cmd); 1724 if (!reply) { 1725 GENL_SET_ERR_MSG(info, "not enough space in Netlink message"); 1726 ret = -EMSGSIZE; 1727 goto fail; 1728 } 1729 1730 spin_lock_bh(&pernet->lock); 1731 entry = __lookup_addr_by_id(pernet, addr.addr.id); 1732 if (!entry) { 1733 GENL_SET_ERR_MSG(info, "address not found"); 1734 ret = -EINVAL; 1735 goto unlock_fail; 1736 } 1737 1738 ret = mptcp_nl_fill_addr(msg, entry); 1739 if (ret) 1740 goto unlock_fail; 1741 1742 genlmsg_end(msg, reply); 1743 ret = genlmsg_reply(msg, info); 1744 spin_unlock_bh(&pernet->lock); 1745 return ret; 1746 1747 unlock_fail: 1748 spin_unlock_bh(&pernet->lock); 1749 1750 fail: 1751 nlmsg_free(msg); 1752 return ret; 1753 } 1754 1755 static int mptcp_nl_cmd_dump_addrs(struct sk_buff *msg, 1756 struct netlink_callback *cb) 1757 { 1758 struct net *net = sock_net(msg->sk); 1759 struct mptcp_pm_addr_entry *entry; 1760 struct pm_nl_pernet *pernet; 1761 int id = cb->args[0]; 1762 void *hdr; 1763 int i; 1764 1765 pernet = pm_nl_get_pernet(net); 1766 1767 spin_lock_bh(&pernet->lock); 1768 for (i = id; i < MPTCP_PM_MAX_ADDR_ID + 1; i++) { 1769 if (test_bit(i, pernet->id_bitmap)) { 1770 entry = __lookup_addr_by_id(pernet, i); 1771 if (!entry) 1772 break; 1773 1774 if (entry->addr.id <= id) 1775 continue; 1776 1777 hdr = genlmsg_put(msg, NETLINK_CB(cb->skb).portid, 1778 cb->nlh->nlmsg_seq, &mptcp_genl_family, 1779 NLM_F_MULTI, MPTCP_PM_CMD_GET_ADDR); 1780 if (!hdr) 1781 break; 1782 1783 if (mptcp_nl_fill_addr(msg, entry) < 0) { 1784 genlmsg_cancel(msg, hdr); 1785 break; 1786 } 1787 1788 id = entry->addr.id; 1789 genlmsg_end(msg, hdr); 1790 } 1791 } 1792 spin_unlock_bh(&pernet->lock); 1793 1794 cb->args[0] = id; 1795 return msg->len; 1796 } 1797 1798 static int parse_limit(struct genl_info *info, int id, unsigned int *limit) 1799 { 1800 struct nlattr *attr = info->attrs[id]; 1801 1802 if (!attr) 1803 return 0; 1804 1805 *limit = nla_get_u32(attr); 1806 if (*limit > MPTCP_PM_ADDR_MAX) { 1807 GENL_SET_ERR_MSG(info, "limit greater than maximum"); 1808 return -EINVAL; 1809 } 1810 return 0; 1811 } 1812 1813 static int 1814 mptcp_nl_cmd_set_limits(struct sk_buff *skb, struct genl_info *info) 1815 { 1816 struct pm_nl_pernet *pernet = genl_info_pm_nl(info); 1817 unsigned int rcv_addrs, subflows; 1818 int ret; 1819 1820 spin_lock_bh(&pernet->lock); 1821 rcv_addrs = pernet->add_addr_accept_max; 1822 ret = parse_limit(info, MPTCP_PM_ATTR_RCV_ADD_ADDRS, &rcv_addrs); 1823 if (ret) 1824 goto unlock; 1825 1826 subflows = pernet->subflows_max; 1827 ret = parse_limit(info, MPTCP_PM_ATTR_SUBFLOWS, &subflows); 1828 if (ret) 1829 goto unlock; 1830 1831 WRITE_ONCE(pernet->add_addr_accept_max, rcv_addrs); 1832 WRITE_ONCE(pernet->subflows_max, subflows); 1833 1834 unlock: 1835 spin_unlock_bh(&pernet->lock); 1836 return ret; 1837 } 1838 1839 static int 1840 mptcp_nl_cmd_get_limits(struct sk_buff *skb, struct genl_info *info) 1841 { 1842 struct pm_nl_pernet *pernet = genl_info_pm_nl(info); 1843 struct sk_buff *msg; 1844 void *reply; 1845 1846 msg = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL); 1847 if (!msg) 1848 return -ENOMEM; 1849 1850 reply = genlmsg_put_reply(msg, info, &mptcp_genl_family, 0, 1851 MPTCP_PM_CMD_GET_LIMITS); 1852 if (!reply) 1853 goto fail; 1854 1855 if (nla_put_u32(msg, MPTCP_PM_ATTR_RCV_ADD_ADDRS, 1856 READ_ONCE(pernet->add_addr_accept_max))) 1857 goto fail; 1858 1859 if (nla_put_u32(msg, MPTCP_PM_ATTR_SUBFLOWS, 1860 READ_ONCE(pernet->subflows_max))) 1861 goto fail; 1862 1863 genlmsg_end(msg, reply); 1864 return genlmsg_reply(msg, info); 1865 1866 fail: 1867 GENL_SET_ERR_MSG(info, "not enough space in Netlink message"); 1868 nlmsg_free(msg); 1869 return -EMSGSIZE; 1870 } 1871 1872 static void mptcp_pm_nl_fullmesh(struct mptcp_sock *msk, 1873 struct mptcp_addr_info *addr) 1874 { 1875 struct mptcp_rm_list list = { .nr = 0 }; 1876 1877 list.ids[list.nr++] = addr->id; 1878 1879 spin_lock_bh(&msk->pm.lock); 1880 mptcp_pm_nl_rm_subflow_received(msk, &list); 1881 mptcp_pm_create_subflow_or_signal_addr(msk); 1882 spin_unlock_bh(&msk->pm.lock); 1883 } 1884 1885 static int mptcp_nl_set_flags(struct net *net, 1886 struct mptcp_addr_info *addr, 1887 u8 bkup, u8 changed) 1888 { 1889 long s_slot = 0, s_num = 0; 1890 struct mptcp_sock *msk; 1891 int ret = -EINVAL; 1892 1893 while ((msk = mptcp_token_iter_next(net, &s_slot, &s_num)) != NULL) { 1894 struct sock *sk = (struct sock *)msk; 1895 1896 if (list_empty(&msk->conn_list) || mptcp_pm_is_userspace(msk)) 1897 goto next; 1898 1899 lock_sock(sk); 1900 if (changed & MPTCP_PM_ADDR_FLAG_BACKUP) 1901 ret = mptcp_pm_nl_mp_prio_send_ack(msk, addr, NULL, bkup); 1902 if (changed & MPTCP_PM_ADDR_FLAG_FULLMESH) 1903 mptcp_pm_nl_fullmesh(msk, addr); 1904 release_sock(sk); 1905 1906 next: 1907 sock_put(sk); 1908 cond_resched(); 1909 } 1910 1911 return ret; 1912 } 1913 1914 static int mptcp_nl_cmd_set_flags(struct sk_buff *skb, struct genl_info *info) 1915 { 1916 struct mptcp_pm_addr_entry addr = { .addr = { .family = AF_UNSPEC }, }, *entry; 1917 struct mptcp_pm_addr_entry remote = { .addr = { .family = AF_UNSPEC }, }; 1918 struct nlattr *attr_rem = info->attrs[MPTCP_PM_ATTR_ADDR_REMOTE]; 1919 struct nlattr *token = info->attrs[MPTCP_PM_ATTR_TOKEN]; 1920 struct nlattr *attr = info->attrs[MPTCP_PM_ATTR_ADDR]; 1921 struct pm_nl_pernet *pernet = genl_info_pm_nl(info); 1922 u8 changed, mask = MPTCP_PM_ADDR_FLAG_BACKUP | 1923 MPTCP_PM_ADDR_FLAG_FULLMESH; 1924 struct net *net = sock_net(skb->sk); 1925 u8 bkup = 0, lookup_by_id = 0; 1926 int ret; 1927 1928 ret = mptcp_pm_parse_entry(attr, info, false, &addr); 1929 if (ret < 0) 1930 return ret; 1931 1932 if (attr_rem) { 1933 ret = mptcp_pm_parse_entry(attr_rem, info, false, &remote); 1934 if (ret < 0) 1935 return ret; 1936 } 1937 1938 if (addr.flags & MPTCP_PM_ADDR_FLAG_BACKUP) 1939 bkup = 1; 1940 if (addr.addr.family == AF_UNSPEC) { 1941 lookup_by_id = 1; 1942 if (!addr.addr.id) 1943 return -EOPNOTSUPP; 1944 } 1945 1946 if (token) 1947 return mptcp_userspace_pm_set_flags(net, token, &addr, &remote, bkup); 1948 1949 spin_lock_bh(&pernet->lock); 1950 entry = __lookup_addr(pernet, &addr.addr, lookup_by_id); 1951 if (!entry) { 1952 spin_unlock_bh(&pernet->lock); 1953 return -EINVAL; 1954 } 1955 if ((addr.flags & MPTCP_PM_ADDR_FLAG_FULLMESH) && 1956 (entry->flags & MPTCP_PM_ADDR_FLAG_SIGNAL)) { 1957 spin_unlock_bh(&pernet->lock); 1958 return -EINVAL; 1959 } 1960 1961 changed = (addr.flags ^ entry->flags) & mask; 1962 entry->flags = (entry->flags & ~mask) | (addr.flags & mask); 1963 addr = *entry; 1964 spin_unlock_bh(&pernet->lock); 1965 1966 mptcp_nl_set_flags(net, &addr.addr, bkup, changed); 1967 return 0; 1968 } 1969 1970 static void mptcp_nl_mcast_send(struct net *net, struct sk_buff *nlskb, gfp_t gfp) 1971 { 1972 genlmsg_multicast_netns(&mptcp_genl_family, net, 1973 nlskb, 0, MPTCP_PM_EV_GRP_OFFSET, gfp); 1974 } 1975 1976 bool mptcp_userspace_pm_active(const struct mptcp_sock *msk) 1977 { 1978 return genl_has_listeners(&mptcp_genl_family, 1979 sock_net((const struct sock *)msk), 1980 MPTCP_PM_EV_GRP_OFFSET); 1981 } 1982 1983 static int mptcp_event_add_subflow(struct sk_buff *skb, const struct sock *ssk) 1984 { 1985 const struct inet_sock *issk = inet_sk(ssk); 1986 const struct mptcp_subflow_context *sf; 1987 1988 if (nla_put_u16(skb, MPTCP_ATTR_FAMILY, ssk->sk_family)) 1989 return -EMSGSIZE; 1990 1991 switch (ssk->sk_family) { 1992 case AF_INET: 1993 if (nla_put_in_addr(skb, MPTCP_ATTR_SADDR4, issk->inet_saddr)) 1994 return -EMSGSIZE; 1995 if (nla_put_in_addr(skb, MPTCP_ATTR_DADDR4, issk->inet_daddr)) 1996 return -EMSGSIZE; 1997 break; 1998 #if IS_ENABLED(CONFIG_MPTCP_IPV6) 1999 case AF_INET6: { 2000 const struct ipv6_pinfo *np = inet6_sk(ssk); 2001 2002 if (nla_put_in6_addr(skb, MPTCP_ATTR_SADDR6, &np->saddr)) 2003 return -EMSGSIZE; 2004 if (nla_put_in6_addr(skb, MPTCP_ATTR_DADDR6, &ssk->sk_v6_daddr)) 2005 return -EMSGSIZE; 2006 break; 2007 } 2008 #endif 2009 default: 2010 WARN_ON_ONCE(1); 2011 return -EMSGSIZE; 2012 } 2013 2014 if (nla_put_be16(skb, MPTCP_ATTR_SPORT, issk->inet_sport)) 2015 return -EMSGSIZE; 2016 if (nla_put_be16(skb, MPTCP_ATTR_DPORT, issk->inet_dport)) 2017 return -EMSGSIZE; 2018 2019 sf = mptcp_subflow_ctx(ssk); 2020 if (WARN_ON_ONCE(!sf)) 2021 return -EINVAL; 2022 2023 if (nla_put_u8(skb, MPTCP_ATTR_LOC_ID, sf->local_id)) 2024 return -EMSGSIZE; 2025 2026 if (nla_put_u8(skb, MPTCP_ATTR_REM_ID, sf->remote_id)) 2027 return -EMSGSIZE; 2028 2029 return 0; 2030 } 2031 2032 static int mptcp_event_put_token_and_ssk(struct sk_buff *skb, 2033 const struct mptcp_sock *msk, 2034 const struct sock *ssk) 2035 { 2036 const struct sock *sk = (const struct sock *)msk; 2037 const struct mptcp_subflow_context *sf; 2038 u8 sk_err; 2039 2040 if (nla_put_u32(skb, MPTCP_ATTR_TOKEN, msk->token)) 2041 return -EMSGSIZE; 2042 2043 if (mptcp_event_add_subflow(skb, ssk)) 2044 return -EMSGSIZE; 2045 2046 sf = mptcp_subflow_ctx(ssk); 2047 if (WARN_ON_ONCE(!sf)) 2048 return -EINVAL; 2049 2050 if (nla_put_u8(skb, MPTCP_ATTR_BACKUP, sf->backup)) 2051 return -EMSGSIZE; 2052 2053 if (ssk->sk_bound_dev_if && 2054 nla_put_s32(skb, MPTCP_ATTR_IF_IDX, ssk->sk_bound_dev_if)) 2055 return -EMSGSIZE; 2056 2057 sk_err = READ_ONCE(ssk->sk_err); 2058 if (sk_err && sk->sk_state == TCP_ESTABLISHED && 2059 nla_put_u8(skb, MPTCP_ATTR_ERROR, sk_err)) 2060 return -EMSGSIZE; 2061 2062 return 0; 2063 } 2064 2065 static int mptcp_event_sub_established(struct sk_buff *skb, 2066 const struct mptcp_sock *msk, 2067 const struct sock *ssk) 2068 { 2069 return mptcp_event_put_token_and_ssk(skb, msk, ssk); 2070 } 2071 2072 static int mptcp_event_sub_closed(struct sk_buff *skb, 2073 const struct mptcp_sock *msk, 2074 const struct sock *ssk) 2075 { 2076 const struct mptcp_subflow_context *sf; 2077 2078 if (mptcp_event_put_token_and_ssk(skb, msk, ssk)) 2079 return -EMSGSIZE; 2080 2081 sf = mptcp_subflow_ctx(ssk); 2082 if (!sf->reset_seen) 2083 return 0; 2084 2085 if (nla_put_u32(skb, MPTCP_ATTR_RESET_REASON, sf->reset_reason)) 2086 return -EMSGSIZE; 2087 2088 if (nla_put_u32(skb, MPTCP_ATTR_RESET_FLAGS, sf->reset_transient)) 2089 return -EMSGSIZE; 2090 2091 return 0; 2092 } 2093 2094 static int mptcp_event_created(struct sk_buff *skb, 2095 const struct mptcp_sock *msk, 2096 const struct sock *ssk) 2097 { 2098 int err = nla_put_u32(skb, MPTCP_ATTR_TOKEN, msk->token); 2099 2100 if (err) 2101 return err; 2102 2103 if (nla_put_u8(skb, MPTCP_ATTR_SERVER_SIDE, READ_ONCE(msk->pm.server_side))) 2104 return -EMSGSIZE; 2105 2106 return mptcp_event_add_subflow(skb, ssk); 2107 } 2108 2109 void mptcp_event_addr_removed(const struct mptcp_sock *msk, uint8_t id) 2110 { 2111 struct net *net = sock_net((const struct sock *)msk); 2112 struct nlmsghdr *nlh; 2113 struct sk_buff *skb; 2114 2115 if (!genl_has_listeners(&mptcp_genl_family, net, MPTCP_PM_EV_GRP_OFFSET)) 2116 return; 2117 2118 skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_ATOMIC); 2119 if (!skb) 2120 return; 2121 2122 nlh = genlmsg_put(skb, 0, 0, &mptcp_genl_family, 0, MPTCP_EVENT_REMOVED); 2123 if (!nlh) 2124 goto nla_put_failure; 2125 2126 if (nla_put_u32(skb, MPTCP_ATTR_TOKEN, msk->token)) 2127 goto nla_put_failure; 2128 2129 if (nla_put_u8(skb, MPTCP_ATTR_REM_ID, id)) 2130 goto nla_put_failure; 2131 2132 genlmsg_end(skb, nlh); 2133 mptcp_nl_mcast_send(net, skb, GFP_ATOMIC); 2134 return; 2135 2136 nla_put_failure: 2137 nlmsg_free(skb); 2138 } 2139 2140 void mptcp_event_addr_announced(const struct sock *ssk, 2141 const struct mptcp_addr_info *info) 2142 { 2143 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk); 2144 struct mptcp_sock *msk = mptcp_sk(subflow->conn); 2145 struct net *net = sock_net(ssk); 2146 struct nlmsghdr *nlh; 2147 struct sk_buff *skb; 2148 2149 if (!genl_has_listeners(&mptcp_genl_family, net, MPTCP_PM_EV_GRP_OFFSET)) 2150 return; 2151 2152 skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_ATOMIC); 2153 if (!skb) 2154 return; 2155 2156 nlh = genlmsg_put(skb, 0, 0, &mptcp_genl_family, 0, 2157 MPTCP_EVENT_ANNOUNCED); 2158 if (!nlh) 2159 goto nla_put_failure; 2160 2161 if (nla_put_u32(skb, MPTCP_ATTR_TOKEN, msk->token)) 2162 goto nla_put_failure; 2163 2164 if (nla_put_u8(skb, MPTCP_ATTR_REM_ID, info->id)) 2165 goto nla_put_failure; 2166 2167 if (nla_put_be16(skb, MPTCP_ATTR_DPORT, 2168 info->port == 0 ? 2169 inet_sk(ssk)->inet_dport : 2170 info->port)) 2171 goto nla_put_failure; 2172 2173 switch (info->family) { 2174 case AF_INET: 2175 if (nla_put_in_addr(skb, MPTCP_ATTR_DADDR4, info->addr.s_addr)) 2176 goto nla_put_failure; 2177 break; 2178 #if IS_ENABLED(CONFIG_MPTCP_IPV6) 2179 case AF_INET6: 2180 if (nla_put_in6_addr(skb, MPTCP_ATTR_DADDR6, &info->addr6)) 2181 goto nla_put_failure; 2182 break; 2183 #endif 2184 default: 2185 WARN_ON_ONCE(1); 2186 goto nla_put_failure; 2187 } 2188 2189 genlmsg_end(skb, nlh); 2190 mptcp_nl_mcast_send(net, skb, GFP_ATOMIC); 2191 return; 2192 2193 nla_put_failure: 2194 nlmsg_free(skb); 2195 } 2196 2197 void mptcp_event_pm_listener(const struct sock *ssk, 2198 enum mptcp_event_type event) 2199 { 2200 const struct inet_sock *issk = inet_sk(ssk); 2201 struct net *net = sock_net(ssk); 2202 struct nlmsghdr *nlh; 2203 struct sk_buff *skb; 2204 2205 if (!genl_has_listeners(&mptcp_genl_family, net, MPTCP_PM_EV_GRP_OFFSET)) 2206 return; 2207 2208 skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL); 2209 if (!skb) 2210 return; 2211 2212 nlh = genlmsg_put(skb, 0, 0, &mptcp_genl_family, 0, event); 2213 if (!nlh) 2214 goto nla_put_failure; 2215 2216 if (nla_put_u16(skb, MPTCP_ATTR_FAMILY, ssk->sk_family)) 2217 goto nla_put_failure; 2218 2219 if (nla_put_be16(skb, MPTCP_ATTR_SPORT, issk->inet_sport)) 2220 goto nla_put_failure; 2221 2222 switch (ssk->sk_family) { 2223 case AF_INET: 2224 if (nla_put_in_addr(skb, MPTCP_ATTR_SADDR4, issk->inet_saddr)) 2225 goto nla_put_failure; 2226 break; 2227 #if IS_ENABLED(CONFIG_MPTCP_IPV6) 2228 case AF_INET6: { 2229 const struct ipv6_pinfo *np = inet6_sk(ssk); 2230 2231 if (nla_put_in6_addr(skb, MPTCP_ATTR_SADDR6, &np->saddr)) 2232 goto nla_put_failure; 2233 break; 2234 } 2235 #endif 2236 default: 2237 WARN_ON_ONCE(1); 2238 goto nla_put_failure; 2239 } 2240 2241 genlmsg_end(skb, nlh); 2242 mptcp_nl_mcast_send(net, skb, GFP_KERNEL); 2243 return; 2244 2245 nla_put_failure: 2246 nlmsg_free(skb); 2247 } 2248 2249 void mptcp_event(enum mptcp_event_type type, const struct mptcp_sock *msk, 2250 const struct sock *ssk, gfp_t gfp) 2251 { 2252 struct net *net = sock_net((const struct sock *)msk); 2253 struct nlmsghdr *nlh; 2254 struct sk_buff *skb; 2255 2256 if (!genl_has_listeners(&mptcp_genl_family, net, MPTCP_PM_EV_GRP_OFFSET)) 2257 return; 2258 2259 skb = nlmsg_new(NLMSG_DEFAULT_SIZE, gfp); 2260 if (!skb) 2261 return; 2262 2263 nlh = genlmsg_put(skb, 0, 0, &mptcp_genl_family, 0, type); 2264 if (!nlh) 2265 goto nla_put_failure; 2266 2267 switch (type) { 2268 case MPTCP_EVENT_UNSPEC: 2269 WARN_ON_ONCE(1); 2270 break; 2271 case MPTCP_EVENT_CREATED: 2272 case MPTCP_EVENT_ESTABLISHED: 2273 if (mptcp_event_created(skb, msk, ssk) < 0) 2274 goto nla_put_failure; 2275 break; 2276 case MPTCP_EVENT_CLOSED: 2277 if (nla_put_u32(skb, MPTCP_ATTR_TOKEN, msk->token) < 0) 2278 goto nla_put_failure; 2279 break; 2280 case MPTCP_EVENT_ANNOUNCED: 2281 case MPTCP_EVENT_REMOVED: 2282 /* call mptcp_event_addr_announced()/removed instead */ 2283 WARN_ON_ONCE(1); 2284 break; 2285 case MPTCP_EVENT_SUB_ESTABLISHED: 2286 case MPTCP_EVENT_SUB_PRIORITY: 2287 if (mptcp_event_sub_established(skb, msk, ssk) < 0) 2288 goto nla_put_failure; 2289 break; 2290 case MPTCP_EVENT_SUB_CLOSED: 2291 if (mptcp_event_sub_closed(skb, msk, ssk) < 0) 2292 goto nla_put_failure; 2293 break; 2294 case MPTCP_EVENT_LISTENER_CREATED: 2295 case MPTCP_EVENT_LISTENER_CLOSED: 2296 break; 2297 } 2298 2299 genlmsg_end(skb, nlh); 2300 mptcp_nl_mcast_send(net, skb, gfp); 2301 return; 2302 2303 nla_put_failure: 2304 nlmsg_free(skb); 2305 } 2306 2307 static const struct genl_small_ops mptcp_pm_ops[] = { 2308 { 2309 .cmd = MPTCP_PM_CMD_ADD_ADDR, 2310 .doit = mptcp_nl_cmd_add_addr, 2311 .flags = GENL_UNS_ADMIN_PERM, 2312 }, 2313 { 2314 .cmd = MPTCP_PM_CMD_DEL_ADDR, 2315 .doit = mptcp_nl_cmd_del_addr, 2316 .flags = GENL_UNS_ADMIN_PERM, 2317 }, 2318 { 2319 .cmd = MPTCP_PM_CMD_FLUSH_ADDRS, 2320 .doit = mptcp_nl_cmd_flush_addrs, 2321 .flags = GENL_UNS_ADMIN_PERM, 2322 }, 2323 { 2324 .cmd = MPTCP_PM_CMD_GET_ADDR, 2325 .doit = mptcp_nl_cmd_get_addr, 2326 .dumpit = mptcp_nl_cmd_dump_addrs, 2327 }, 2328 { 2329 .cmd = MPTCP_PM_CMD_SET_LIMITS, 2330 .doit = mptcp_nl_cmd_set_limits, 2331 .flags = GENL_UNS_ADMIN_PERM, 2332 }, 2333 { 2334 .cmd = MPTCP_PM_CMD_GET_LIMITS, 2335 .doit = mptcp_nl_cmd_get_limits, 2336 }, 2337 { 2338 .cmd = MPTCP_PM_CMD_SET_FLAGS, 2339 .doit = mptcp_nl_cmd_set_flags, 2340 .flags = GENL_UNS_ADMIN_PERM, 2341 }, 2342 { 2343 .cmd = MPTCP_PM_CMD_ANNOUNCE, 2344 .doit = mptcp_nl_cmd_announce, 2345 .flags = GENL_UNS_ADMIN_PERM, 2346 }, 2347 { 2348 .cmd = MPTCP_PM_CMD_REMOVE, 2349 .doit = mptcp_nl_cmd_remove, 2350 .flags = GENL_UNS_ADMIN_PERM, 2351 }, 2352 { 2353 .cmd = MPTCP_PM_CMD_SUBFLOW_CREATE, 2354 .doit = mptcp_nl_cmd_sf_create, 2355 .flags = GENL_UNS_ADMIN_PERM, 2356 }, 2357 { 2358 .cmd = MPTCP_PM_CMD_SUBFLOW_DESTROY, 2359 .doit = mptcp_nl_cmd_sf_destroy, 2360 .flags = GENL_UNS_ADMIN_PERM, 2361 }, 2362 }; 2363 2364 static struct genl_family mptcp_genl_family __ro_after_init = { 2365 .name = MPTCP_PM_NAME, 2366 .version = MPTCP_PM_VER, 2367 .maxattr = MPTCP_PM_ATTR_MAX, 2368 .policy = mptcp_pm_policy, 2369 .netnsok = true, 2370 .module = THIS_MODULE, 2371 .small_ops = mptcp_pm_ops, 2372 .n_small_ops = ARRAY_SIZE(mptcp_pm_ops), 2373 .resv_start_op = MPTCP_PM_CMD_SUBFLOW_DESTROY + 1, 2374 .mcgrps = mptcp_pm_mcgrps, 2375 .n_mcgrps = ARRAY_SIZE(mptcp_pm_mcgrps), 2376 }; 2377 2378 static int __net_init pm_nl_init_net(struct net *net) 2379 { 2380 struct pm_nl_pernet *pernet = pm_nl_get_pernet(net); 2381 2382 INIT_LIST_HEAD_RCU(&pernet->local_addr_list); 2383 2384 /* Cit. 2 subflows ought to be enough for anybody. */ 2385 pernet->subflows_max = 2; 2386 pernet->next_id = 1; 2387 pernet->stale_loss_cnt = 4; 2388 spin_lock_init(&pernet->lock); 2389 2390 /* No need to initialize other pernet fields, the struct is zeroed at 2391 * allocation time. 2392 */ 2393 2394 return 0; 2395 } 2396 2397 static void __net_exit pm_nl_exit_net(struct list_head *net_list) 2398 { 2399 struct net *net; 2400 2401 list_for_each_entry(net, net_list, exit_list) { 2402 struct pm_nl_pernet *pernet = pm_nl_get_pernet(net); 2403 2404 /* net is removed from namespace list, can't race with 2405 * other modifiers, also netns core already waited for a 2406 * RCU grace period. 2407 */ 2408 __flush_addrs(&pernet->local_addr_list); 2409 } 2410 } 2411 2412 static struct pernet_operations mptcp_pm_pernet_ops = { 2413 .init = pm_nl_init_net, 2414 .exit_batch = pm_nl_exit_net, 2415 .id = &pm_nl_pernet_id, 2416 .size = sizeof(struct pm_nl_pernet), 2417 }; 2418 2419 void __init mptcp_pm_nl_init(void) 2420 { 2421 if (register_pernet_subsys(&mptcp_pm_pernet_ops) < 0) 2422 panic("Failed to register MPTCP PM pernet subsystem.\n"); 2423 2424 if (genl_register_family(&mptcp_genl_family)) 2425 panic("Failed to register MPTCP PM netlink family\n"); 2426 } 2427