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 int mptcp_pm_nl_create_listen_socket(struct sock *sk, 1001 struct mptcp_pm_addr_entry *entry) 1002 { 1003 int addrlen = sizeof(struct sockaddr_in); 1004 struct sockaddr_storage addr; 1005 struct mptcp_sock *msk; 1006 struct socket *ssock; 1007 int backlog = 1024; 1008 int err; 1009 1010 err = sock_create_kern(sock_net(sk), entry->addr.family, 1011 SOCK_STREAM, IPPROTO_MPTCP, &entry->lsk); 1012 if (err) 1013 return err; 1014 1015 msk = mptcp_sk(entry->lsk->sk); 1016 if (!msk) 1017 return -EINVAL; 1018 1019 ssock = __mptcp_nmpc_socket(msk); 1020 if (!ssock) 1021 return -EINVAL; 1022 1023 mptcp_info2sockaddr(&entry->addr, &addr, entry->addr.family); 1024 #if IS_ENABLED(CONFIG_MPTCP_IPV6) 1025 if (entry->addr.family == AF_INET6) 1026 addrlen = sizeof(struct sockaddr_in6); 1027 #endif 1028 err = kernel_bind(ssock, (struct sockaddr *)&addr, addrlen); 1029 if (err) 1030 return err; 1031 1032 err = kernel_listen(ssock, backlog); 1033 if (err) 1034 return err; 1035 1036 mptcp_event_pm_listener(ssock->sk, MPTCP_EVENT_LISTENER_CREATED); 1037 1038 return 0; 1039 } 1040 1041 int mptcp_pm_nl_get_local_id(struct mptcp_sock *msk, struct sock_common *skc) 1042 { 1043 struct mptcp_pm_addr_entry *entry; 1044 struct mptcp_addr_info skc_local; 1045 struct mptcp_addr_info msk_local; 1046 struct pm_nl_pernet *pernet; 1047 int ret = -1; 1048 1049 if (WARN_ON_ONCE(!msk)) 1050 return -1; 1051 1052 /* The 0 ID mapping is defined by the first subflow, copied into the msk 1053 * addr 1054 */ 1055 local_address((struct sock_common *)msk, &msk_local); 1056 local_address((struct sock_common *)skc, &skc_local); 1057 if (mptcp_addresses_equal(&msk_local, &skc_local, false)) 1058 return 0; 1059 1060 if (mptcp_pm_is_userspace(msk)) 1061 return mptcp_userspace_pm_get_local_id(msk, &skc_local); 1062 1063 pernet = pm_nl_get_pernet_from_msk(msk); 1064 1065 rcu_read_lock(); 1066 list_for_each_entry_rcu(entry, &pernet->local_addr_list, list) { 1067 if (mptcp_addresses_equal(&entry->addr, &skc_local, entry->addr.port)) { 1068 ret = entry->addr.id; 1069 break; 1070 } 1071 } 1072 rcu_read_unlock(); 1073 if (ret >= 0) 1074 return ret; 1075 1076 /* address not found, add to local list */ 1077 entry = kmalloc(sizeof(*entry), GFP_ATOMIC); 1078 if (!entry) 1079 return -ENOMEM; 1080 1081 entry->addr = skc_local; 1082 entry->addr.id = 0; 1083 entry->addr.port = 0; 1084 entry->ifindex = 0; 1085 entry->flags = MPTCP_PM_ADDR_FLAG_IMPLICIT; 1086 entry->lsk = NULL; 1087 ret = mptcp_pm_nl_append_new_local_addr(pernet, entry); 1088 if (ret < 0) 1089 kfree(entry); 1090 1091 return ret; 1092 } 1093 1094 #define MPTCP_PM_CMD_GRP_OFFSET 0 1095 #define MPTCP_PM_EV_GRP_OFFSET 1 1096 1097 static const struct genl_multicast_group mptcp_pm_mcgrps[] = { 1098 [MPTCP_PM_CMD_GRP_OFFSET] = { .name = MPTCP_PM_CMD_GRP_NAME, }, 1099 [MPTCP_PM_EV_GRP_OFFSET] = { .name = MPTCP_PM_EV_GRP_NAME, 1100 .flags = GENL_UNS_ADMIN_PERM, 1101 }, 1102 }; 1103 1104 static const struct nla_policy 1105 mptcp_pm_addr_policy[MPTCP_PM_ADDR_ATTR_MAX + 1] = { 1106 [MPTCP_PM_ADDR_ATTR_FAMILY] = { .type = NLA_U16, }, 1107 [MPTCP_PM_ADDR_ATTR_ID] = { .type = NLA_U8, }, 1108 [MPTCP_PM_ADDR_ATTR_ADDR4] = { .type = NLA_U32, }, 1109 [MPTCP_PM_ADDR_ATTR_ADDR6] = 1110 NLA_POLICY_EXACT_LEN(sizeof(struct in6_addr)), 1111 [MPTCP_PM_ADDR_ATTR_PORT] = { .type = NLA_U16 }, 1112 [MPTCP_PM_ADDR_ATTR_FLAGS] = { .type = NLA_U32 }, 1113 [MPTCP_PM_ADDR_ATTR_IF_IDX] = { .type = NLA_S32 }, 1114 }; 1115 1116 static const struct nla_policy mptcp_pm_policy[MPTCP_PM_ATTR_MAX + 1] = { 1117 [MPTCP_PM_ATTR_ADDR] = 1118 NLA_POLICY_NESTED(mptcp_pm_addr_policy), 1119 [MPTCP_PM_ATTR_RCV_ADD_ADDRS] = { .type = NLA_U32, }, 1120 [MPTCP_PM_ATTR_SUBFLOWS] = { .type = NLA_U32, }, 1121 [MPTCP_PM_ATTR_TOKEN] = { .type = NLA_U32, }, 1122 [MPTCP_PM_ATTR_LOC_ID] = { .type = NLA_U8, }, 1123 [MPTCP_PM_ATTR_ADDR_REMOTE] = 1124 NLA_POLICY_NESTED(mptcp_pm_addr_policy), 1125 }; 1126 1127 void mptcp_pm_nl_subflow_chk_stale(const struct mptcp_sock *msk, struct sock *ssk) 1128 { 1129 struct mptcp_subflow_context *iter, *subflow = mptcp_subflow_ctx(ssk); 1130 struct sock *sk = (struct sock *)msk; 1131 unsigned int active_max_loss_cnt; 1132 struct net *net = sock_net(sk); 1133 unsigned int stale_loss_cnt; 1134 bool slow; 1135 1136 stale_loss_cnt = mptcp_stale_loss_cnt(net); 1137 if (subflow->stale || !stale_loss_cnt || subflow->stale_count <= stale_loss_cnt) 1138 return; 1139 1140 /* look for another available subflow not in loss state */ 1141 active_max_loss_cnt = max_t(int, stale_loss_cnt - 1, 1); 1142 mptcp_for_each_subflow(msk, iter) { 1143 if (iter != subflow && mptcp_subflow_active(iter) && 1144 iter->stale_count < active_max_loss_cnt) { 1145 /* we have some alternatives, try to mark this subflow as idle ...*/ 1146 slow = lock_sock_fast(ssk); 1147 if (!tcp_rtx_and_write_queues_empty(ssk)) { 1148 subflow->stale = 1; 1149 __mptcp_retransmit_pending_data(sk); 1150 MPTCP_INC_STATS(net, MPTCP_MIB_SUBFLOWSTALE); 1151 } 1152 unlock_sock_fast(ssk, slow); 1153 1154 /* always try to push the pending data regardless of re-injections: 1155 * we can possibly use backup subflows now, and subflow selection 1156 * is cheap under the msk socket lock 1157 */ 1158 __mptcp_push_pending(sk, 0); 1159 return; 1160 } 1161 } 1162 } 1163 1164 static int mptcp_pm_family_to_addr(int family) 1165 { 1166 #if IS_ENABLED(CONFIG_MPTCP_IPV6) 1167 if (family == AF_INET6) 1168 return MPTCP_PM_ADDR_ATTR_ADDR6; 1169 #endif 1170 return MPTCP_PM_ADDR_ATTR_ADDR4; 1171 } 1172 1173 static int mptcp_pm_parse_pm_addr_attr(struct nlattr *tb[], 1174 const struct nlattr *attr, 1175 struct genl_info *info, 1176 struct mptcp_addr_info *addr, 1177 bool require_family) 1178 { 1179 int err, addr_addr; 1180 1181 if (!attr) { 1182 GENL_SET_ERR_MSG(info, "missing address info"); 1183 return -EINVAL; 1184 } 1185 1186 /* no validation needed - was already done via nested policy */ 1187 err = nla_parse_nested_deprecated(tb, MPTCP_PM_ADDR_ATTR_MAX, attr, 1188 mptcp_pm_addr_policy, info->extack); 1189 if (err) 1190 return err; 1191 1192 if (tb[MPTCP_PM_ADDR_ATTR_ID]) 1193 addr->id = nla_get_u8(tb[MPTCP_PM_ADDR_ATTR_ID]); 1194 1195 if (!tb[MPTCP_PM_ADDR_ATTR_FAMILY]) { 1196 if (!require_family) 1197 return 0; 1198 1199 NL_SET_ERR_MSG_ATTR(info->extack, attr, 1200 "missing family"); 1201 return -EINVAL; 1202 } 1203 1204 addr->family = nla_get_u16(tb[MPTCP_PM_ADDR_ATTR_FAMILY]); 1205 if (addr->family != AF_INET 1206 #if IS_ENABLED(CONFIG_MPTCP_IPV6) 1207 && addr->family != AF_INET6 1208 #endif 1209 ) { 1210 NL_SET_ERR_MSG_ATTR(info->extack, attr, 1211 "unknown address family"); 1212 return -EINVAL; 1213 } 1214 addr_addr = mptcp_pm_family_to_addr(addr->family); 1215 if (!tb[addr_addr]) { 1216 NL_SET_ERR_MSG_ATTR(info->extack, attr, 1217 "missing address data"); 1218 return -EINVAL; 1219 } 1220 1221 #if IS_ENABLED(CONFIG_MPTCP_IPV6) 1222 if (addr->family == AF_INET6) 1223 addr->addr6 = nla_get_in6_addr(tb[addr_addr]); 1224 else 1225 #endif 1226 addr->addr.s_addr = nla_get_in_addr(tb[addr_addr]); 1227 1228 if (tb[MPTCP_PM_ADDR_ATTR_PORT]) 1229 addr->port = htons(nla_get_u16(tb[MPTCP_PM_ADDR_ATTR_PORT])); 1230 1231 return 0; 1232 } 1233 1234 int mptcp_pm_parse_addr(struct nlattr *attr, struct genl_info *info, 1235 struct mptcp_addr_info *addr) 1236 { 1237 struct nlattr *tb[MPTCP_PM_ADDR_ATTR_MAX + 1]; 1238 1239 memset(addr, 0, sizeof(*addr)); 1240 1241 return mptcp_pm_parse_pm_addr_attr(tb, attr, info, addr, true); 1242 } 1243 1244 int mptcp_pm_parse_entry(struct nlattr *attr, struct genl_info *info, 1245 bool require_family, 1246 struct mptcp_pm_addr_entry *entry) 1247 { 1248 struct nlattr *tb[MPTCP_PM_ADDR_ATTR_MAX + 1]; 1249 int err; 1250 1251 memset(entry, 0, sizeof(*entry)); 1252 1253 err = mptcp_pm_parse_pm_addr_attr(tb, attr, info, &entry->addr, require_family); 1254 if (err) 1255 return err; 1256 1257 if (tb[MPTCP_PM_ADDR_ATTR_IF_IDX]) { 1258 u32 val = nla_get_s32(tb[MPTCP_PM_ADDR_ATTR_IF_IDX]); 1259 1260 entry->ifindex = val; 1261 } 1262 1263 if (tb[MPTCP_PM_ADDR_ATTR_FLAGS]) 1264 entry->flags = nla_get_u32(tb[MPTCP_PM_ADDR_ATTR_FLAGS]); 1265 1266 if (tb[MPTCP_PM_ADDR_ATTR_PORT]) 1267 entry->addr.port = htons(nla_get_u16(tb[MPTCP_PM_ADDR_ATTR_PORT])); 1268 1269 return 0; 1270 } 1271 1272 static struct pm_nl_pernet *genl_info_pm_nl(struct genl_info *info) 1273 { 1274 return pm_nl_get_pernet(genl_info_net(info)); 1275 } 1276 1277 static int mptcp_nl_add_subflow_or_signal_addr(struct net *net) 1278 { 1279 struct mptcp_sock *msk; 1280 long s_slot = 0, s_num = 0; 1281 1282 while ((msk = mptcp_token_iter_next(net, &s_slot, &s_num)) != NULL) { 1283 struct sock *sk = (struct sock *)msk; 1284 1285 if (!READ_ONCE(msk->fully_established) || 1286 mptcp_pm_is_userspace(msk)) 1287 goto next; 1288 1289 lock_sock(sk); 1290 spin_lock_bh(&msk->pm.lock); 1291 mptcp_pm_create_subflow_or_signal_addr(msk); 1292 spin_unlock_bh(&msk->pm.lock); 1293 release_sock(sk); 1294 1295 next: 1296 sock_put(sk); 1297 cond_resched(); 1298 } 1299 1300 return 0; 1301 } 1302 1303 static int mptcp_nl_cmd_add_addr(struct sk_buff *skb, struct genl_info *info) 1304 { 1305 struct nlattr *attr = info->attrs[MPTCP_PM_ATTR_ADDR]; 1306 struct pm_nl_pernet *pernet = genl_info_pm_nl(info); 1307 struct mptcp_pm_addr_entry addr, *entry; 1308 int ret; 1309 1310 ret = mptcp_pm_parse_entry(attr, info, true, &addr); 1311 if (ret < 0) 1312 return ret; 1313 1314 if (addr.addr.port && !(addr.flags & MPTCP_PM_ADDR_FLAG_SIGNAL)) { 1315 GENL_SET_ERR_MSG(info, "flags must have signal when using port"); 1316 return -EINVAL; 1317 } 1318 1319 if (addr.flags & MPTCP_PM_ADDR_FLAG_SIGNAL && 1320 addr.flags & MPTCP_PM_ADDR_FLAG_FULLMESH) { 1321 GENL_SET_ERR_MSG(info, "flags mustn't have both signal and fullmesh"); 1322 return -EINVAL; 1323 } 1324 1325 if (addr.flags & MPTCP_PM_ADDR_FLAG_IMPLICIT) { 1326 GENL_SET_ERR_MSG(info, "can't create IMPLICIT endpoint"); 1327 return -EINVAL; 1328 } 1329 1330 entry = kzalloc(sizeof(*entry), GFP_KERNEL_ACCOUNT); 1331 if (!entry) { 1332 GENL_SET_ERR_MSG(info, "can't allocate addr"); 1333 return -ENOMEM; 1334 } 1335 1336 *entry = addr; 1337 if (entry->addr.port) { 1338 ret = mptcp_pm_nl_create_listen_socket(skb->sk, entry); 1339 if (ret) { 1340 GENL_SET_ERR_MSG_FMT(info, "create listen socket error: %d", ret); 1341 goto out_free; 1342 } 1343 } 1344 ret = mptcp_pm_nl_append_new_local_addr(pernet, entry); 1345 if (ret < 0) { 1346 GENL_SET_ERR_MSG_FMT(info, "too many addresses or duplicate one: %d", ret); 1347 goto out_free; 1348 } 1349 1350 mptcp_nl_add_subflow_or_signal_addr(sock_net(skb->sk)); 1351 return 0; 1352 1353 out_free: 1354 __mptcp_pm_release_addr_entry(entry); 1355 return ret; 1356 } 1357 1358 int mptcp_pm_get_flags_and_ifindex_by_id(struct mptcp_sock *msk, unsigned int id, 1359 u8 *flags, int *ifindex) 1360 { 1361 struct mptcp_pm_addr_entry *entry; 1362 struct sock *sk = (struct sock *)msk; 1363 struct net *net = sock_net(sk); 1364 1365 *flags = 0; 1366 *ifindex = 0; 1367 1368 if (id) { 1369 if (mptcp_pm_is_userspace(msk)) 1370 return mptcp_userspace_pm_get_flags_and_ifindex_by_id(msk, 1371 id, 1372 flags, 1373 ifindex); 1374 1375 rcu_read_lock(); 1376 entry = __lookup_addr_by_id(pm_nl_get_pernet(net), id); 1377 if (entry) { 1378 *flags = entry->flags; 1379 *ifindex = entry->ifindex; 1380 } 1381 rcu_read_unlock(); 1382 } 1383 1384 return 0; 1385 } 1386 1387 static bool remove_anno_list_by_saddr(struct mptcp_sock *msk, 1388 const struct mptcp_addr_info *addr) 1389 { 1390 struct mptcp_pm_add_entry *entry; 1391 1392 entry = mptcp_pm_del_add_timer(msk, addr, false); 1393 if (entry) { 1394 list_del(&entry->list); 1395 kfree(entry); 1396 return true; 1397 } 1398 1399 return false; 1400 } 1401 1402 static bool mptcp_pm_remove_anno_addr(struct mptcp_sock *msk, 1403 const struct mptcp_addr_info *addr, 1404 bool force) 1405 { 1406 struct mptcp_rm_list list = { .nr = 0 }; 1407 bool ret; 1408 1409 list.ids[list.nr++] = addr->id; 1410 1411 ret = remove_anno_list_by_saddr(msk, addr); 1412 if (ret || force) { 1413 spin_lock_bh(&msk->pm.lock); 1414 mptcp_pm_remove_addr(msk, &list); 1415 spin_unlock_bh(&msk->pm.lock); 1416 } 1417 return ret; 1418 } 1419 1420 static int mptcp_nl_remove_subflow_and_signal_addr(struct net *net, 1421 const struct mptcp_pm_addr_entry *entry) 1422 { 1423 const struct mptcp_addr_info *addr = &entry->addr; 1424 struct mptcp_rm_list list = { .nr = 0 }; 1425 long s_slot = 0, s_num = 0; 1426 struct mptcp_sock *msk; 1427 1428 pr_debug("remove_id=%d", addr->id); 1429 1430 list.ids[list.nr++] = addr->id; 1431 1432 while ((msk = mptcp_token_iter_next(net, &s_slot, &s_num)) != NULL) { 1433 struct sock *sk = (struct sock *)msk; 1434 bool remove_subflow; 1435 1436 if (mptcp_pm_is_userspace(msk)) 1437 goto next; 1438 1439 if (list_empty(&msk->conn_list)) { 1440 mptcp_pm_remove_anno_addr(msk, addr, false); 1441 goto next; 1442 } 1443 1444 lock_sock(sk); 1445 remove_subflow = lookup_subflow_by_saddr(&msk->conn_list, addr); 1446 mptcp_pm_remove_anno_addr(msk, addr, remove_subflow && 1447 !(entry->flags & MPTCP_PM_ADDR_FLAG_IMPLICIT)); 1448 if (remove_subflow) 1449 mptcp_pm_remove_subflow(msk, &list); 1450 release_sock(sk); 1451 1452 next: 1453 sock_put(sk); 1454 cond_resched(); 1455 } 1456 1457 return 0; 1458 } 1459 1460 static int mptcp_nl_remove_id_zero_address(struct net *net, 1461 struct mptcp_addr_info *addr) 1462 { 1463 struct mptcp_rm_list list = { .nr = 0 }; 1464 long s_slot = 0, s_num = 0; 1465 struct mptcp_sock *msk; 1466 1467 list.ids[list.nr++] = 0; 1468 1469 while ((msk = mptcp_token_iter_next(net, &s_slot, &s_num)) != NULL) { 1470 struct sock *sk = (struct sock *)msk; 1471 struct mptcp_addr_info msk_local; 1472 1473 if (list_empty(&msk->conn_list) || mptcp_pm_is_userspace(msk)) 1474 goto next; 1475 1476 local_address((struct sock_common *)msk, &msk_local); 1477 if (!mptcp_addresses_equal(&msk_local, addr, addr->port)) 1478 goto next; 1479 1480 lock_sock(sk); 1481 spin_lock_bh(&msk->pm.lock); 1482 mptcp_pm_remove_addr(msk, &list); 1483 mptcp_pm_nl_rm_subflow_received(msk, &list); 1484 spin_unlock_bh(&msk->pm.lock); 1485 release_sock(sk); 1486 1487 next: 1488 sock_put(sk); 1489 cond_resched(); 1490 } 1491 1492 return 0; 1493 } 1494 1495 static int mptcp_nl_cmd_del_addr(struct sk_buff *skb, struct genl_info *info) 1496 { 1497 struct nlattr *attr = info->attrs[MPTCP_PM_ATTR_ADDR]; 1498 struct pm_nl_pernet *pernet = genl_info_pm_nl(info); 1499 struct mptcp_pm_addr_entry addr, *entry; 1500 unsigned int addr_max; 1501 int ret; 1502 1503 ret = mptcp_pm_parse_entry(attr, info, false, &addr); 1504 if (ret < 0) 1505 return ret; 1506 1507 /* the zero id address is special: the first address used by the msk 1508 * always gets such an id, so different subflows can have different zero 1509 * id addresses. Additionally zero id is not accounted for in id_bitmap. 1510 * Let's use an 'mptcp_rm_list' instead of the common remove code. 1511 */ 1512 if (addr.addr.id == 0) 1513 return mptcp_nl_remove_id_zero_address(sock_net(skb->sk), &addr.addr); 1514 1515 spin_lock_bh(&pernet->lock); 1516 entry = __lookup_addr_by_id(pernet, addr.addr.id); 1517 if (!entry) { 1518 GENL_SET_ERR_MSG(info, "address not found"); 1519 spin_unlock_bh(&pernet->lock); 1520 return -EINVAL; 1521 } 1522 if (entry->flags & MPTCP_PM_ADDR_FLAG_SIGNAL) { 1523 addr_max = pernet->add_addr_signal_max; 1524 WRITE_ONCE(pernet->add_addr_signal_max, addr_max - 1); 1525 } 1526 if (entry->flags & MPTCP_PM_ADDR_FLAG_SUBFLOW) { 1527 addr_max = pernet->local_addr_max; 1528 WRITE_ONCE(pernet->local_addr_max, addr_max - 1); 1529 } 1530 1531 pernet->addrs--; 1532 list_del_rcu(&entry->list); 1533 __clear_bit(entry->addr.id, pernet->id_bitmap); 1534 spin_unlock_bh(&pernet->lock); 1535 1536 mptcp_nl_remove_subflow_and_signal_addr(sock_net(skb->sk), entry); 1537 synchronize_rcu(); 1538 __mptcp_pm_release_addr_entry(entry); 1539 1540 return ret; 1541 } 1542 1543 void mptcp_pm_remove_addrs_and_subflows(struct mptcp_sock *msk, 1544 struct list_head *rm_list) 1545 { 1546 struct mptcp_rm_list alist = { .nr = 0 }, slist = { .nr = 0 }; 1547 struct mptcp_pm_addr_entry *entry; 1548 1549 list_for_each_entry(entry, rm_list, list) { 1550 if (lookup_subflow_by_saddr(&msk->conn_list, &entry->addr) && 1551 slist.nr < MPTCP_RM_IDS_MAX) 1552 slist.ids[slist.nr++] = entry->addr.id; 1553 1554 if (remove_anno_list_by_saddr(msk, &entry->addr) && 1555 alist.nr < MPTCP_RM_IDS_MAX) 1556 alist.ids[alist.nr++] = entry->addr.id; 1557 } 1558 1559 if (alist.nr) { 1560 spin_lock_bh(&msk->pm.lock); 1561 mptcp_pm_remove_addr(msk, &alist); 1562 spin_unlock_bh(&msk->pm.lock); 1563 } 1564 if (slist.nr) 1565 mptcp_pm_remove_subflow(msk, &slist); 1566 } 1567 1568 static void mptcp_nl_remove_addrs_list(struct net *net, 1569 struct list_head *rm_list) 1570 { 1571 long s_slot = 0, s_num = 0; 1572 struct mptcp_sock *msk; 1573 1574 if (list_empty(rm_list)) 1575 return; 1576 1577 while ((msk = mptcp_token_iter_next(net, &s_slot, &s_num)) != NULL) { 1578 struct sock *sk = (struct sock *)msk; 1579 1580 if (!mptcp_pm_is_userspace(msk)) { 1581 lock_sock(sk); 1582 mptcp_pm_remove_addrs_and_subflows(msk, rm_list); 1583 release_sock(sk); 1584 } 1585 1586 sock_put(sk); 1587 cond_resched(); 1588 } 1589 } 1590 1591 /* caller must ensure the RCU grace period is already elapsed */ 1592 static void __flush_addrs(struct list_head *list) 1593 { 1594 while (!list_empty(list)) { 1595 struct mptcp_pm_addr_entry *cur; 1596 1597 cur = list_entry(list->next, 1598 struct mptcp_pm_addr_entry, list); 1599 list_del_rcu(&cur->list); 1600 __mptcp_pm_release_addr_entry(cur); 1601 } 1602 } 1603 1604 static void __reset_counters(struct pm_nl_pernet *pernet) 1605 { 1606 WRITE_ONCE(pernet->add_addr_signal_max, 0); 1607 WRITE_ONCE(pernet->add_addr_accept_max, 0); 1608 WRITE_ONCE(pernet->local_addr_max, 0); 1609 pernet->addrs = 0; 1610 } 1611 1612 static int mptcp_nl_cmd_flush_addrs(struct sk_buff *skb, struct genl_info *info) 1613 { 1614 struct pm_nl_pernet *pernet = genl_info_pm_nl(info); 1615 LIST_HEAD(free_list); 1616 1617 spin_lock_bh(&pernet->lock); 1618 list_splice_init(&pernet->local_addr_list, &free_list); 1619 __reset_counters(pernet); 1620 pernet->next_id = 1; 1621 bitmap_zero(pernet->id_bitmap, MPTCP_PM_MAX_ADDR_ID + 1); 1622 spin_unlock_bh(&pernet->lock); 1623 mptcp_nl_remove_addrs_list(sock_net(skb->sk), &free_list); 1624 synchronize_rcu(); 1625 __flush_addrs(&free_list); 1626 return 0; 1627 } 1628 1629 static int mptcp_nl_fill_addr(struct sk_buff *skb, 1630 struct mptcp_pm_addr_entry *entry) 1631 { 1632 struct mptcp_addr_info *addr = &entry->addr; 1633 struct nlattr *attr; 1634 1635 attr = nla_nest_start(skb, MPTCP_PM_ATTR_ADDR); 1636 if (!attr) 1637 return -EMSGSIZE; 1638 1639 if (nla_put_u16(skb, MPTCP_PM_ADDR_ATTR_FAMILY, addr->family)) 1640 goto nla_put_failure; 1641 if (nla_put_u16(skb, MPTCP_PM_ADDR_ATTR_PORT, ntohs(addr->port))) 1642 goto nla_put_failure; 1643 if (nla_put_u8(skb, MPTCP_PM_ADDR_ATTR_ID, addr->id)) 1644 goto nla_put_failure; 1645 if (nla_put_u32(skb, MPTCP_PM_ADDR_ATTR_FLAGS, entry->flags)) 1646 goto nla_put_failure; 1647 if (entry->ifindex && 1648 nla_put_s32(skb, MPTCP_PM_ADDR_ATTR_IF_IDX, entry->ifindex)) 1649 goto nla_put_failure; 1650 1651 if (addr->family == AF_INET && 1652 nla_put_in_addr(skb, MPTCP_PM_ADDR_ATTR_ADDR4, 1653 addr->addr.s_addr)) 1654 goto nla_put_failure; 1655 #if IS_ENABLED(CONFIG_MPTCP_IPV6) 1656 else if (addr->family == AF_INET6 && 1657 nla_put_in6_addr(skb, MPTCP_PM_ADDR_ATTR_ADDR6, &addr->addr6)) 1658 goto nla_put_failure; 1659 #endif 1660 nla_nest_end(skb, attr); 1661 return 0; 1662 1663 nla_put_failure: 1664 nla_nest_cancel(skb, attr); 1665 return -EMSGSIZE; 1666 } 1667 1668 static int mptcp_nl_cmd_get_addr(struct sk_buff *skb, struct genl_info *info) 1669 { 1670 struct nlattr *attr = info->attrs[MPTCP_PM_ATTR_ADDR]; 1671 struct pm_nl_pernet *pernet = genl_info_pm_nl(info); 1672 struct mptcp_pm_addr_entry addr, *entry; 1673 struct sk_buff *msg; 1674 void *reply; 1675 int ret; 1676 1677 ret = mptcp_pm_parse_entry(attr, info, false, &addr); 1678 if (ret < 0) 1679 return ret; 1680 1681 msg = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL); 1682 if (!msg) 1683 return -ENOMEM; 1684 1685 reply = genlmsg_put_reply(msg, info, &mptcp_genl_family, 0, 1686 info->genlhdr->cmd); 1687 if (!reply) { 1688 GENL_SET_ERR_MSG(info, "not enough space in Netlink message"); 1689 ret = -EMSGSIZE; 1690 goto fail; 1691 } 1692 1693 spin_lock_bh(&pernet->lock); 1694 entry = __lookup_addr_by_id(pernet, addr.addr.id); 1695 if (!entry) { 1696 GENL_SET_ERR_MSG(info, "address not found"); 1697 ret = -EINVAL; 1698 goto unlock_fail; 1699 } 1700 1701 ret = mptcp_nl_fill_addr(msg, entry); 1702 if (ret) 1703 goto unlock_fail; 1704 1705 genlmsg_end(msg, reply); 1706 ret = genlmsg_reply(msg, info); 1707 spin_unlock_bh(&pernet->lock); 1708 return ret; 1709 1710 unlock_fail: 1711 spin_unlock_bh(&pernet->lock); 1712 1713 fail: 1714 nlmsg_free(msg); 1715 return ret; 1716 } 1717 1718 static int mptcp_nl_cmd_dump_addrs(struct sk_buff *msg, 1719 struct netlink_callback *cb) 1720 { 1721 struct net *net = sock_net(msg->sk); 1722 struct mptcp_pm_addr_entry *entry; 1723 struct pm_nl_pernet *pernet; 1724 int id = cb->args[0]; 1725 void *hdr; 1726 int i; 1727 1728 pernet = pm_nl_get_pernet(net); 1729 1730 spin_lock_bh(&pernet->lock); 1731 for (i = id; i < MPTCP_PM_MAX_ADDR_ID + 1; i++) { 1732 if (test_bit(i, pernet->id_bitmap)) { 1733 entry = __lookup_addr_by_id(pernet, i); 1734 if (!entry) 1735 break; 1736 1737 if (entry->addr.id <= id) 1738 continue; 1739 1740 hdr = genlmsg_put(msg, NETLINK_CB(cb->skb).portid, 1741 cb->nlh->nlmsg_seq, &mptcp_genl_family, 1742 NLM_F_MULTI, MPTCP_PM_CMD_GET_ADDR); 1743 if (!hdr) 1744 break; 1745 1746 if (mptcp_nl_fill_addr(msg, entry) < 0) { 1747 genlmsg_cancel(msg, hdr); 1748 break; 1749 } 1750 1751 id = entry->addr.id; 1752 genlmsg_end(msg, hdr); 1753 } 1754 } 1755 spin_unlock_bh(&pernet->lock); 1756 1757 cb->args[0] = id; 1758 return msg->len; 1759 } 1760 1761 static int parse_limit(struct genl_info *info, int id, unsigned int *limit) 1762 { 1763 struct nlattr *attr = info->attrs[id]; 1764 1765 if (!attr) 1766 return 0; 1767 1768 *limit = nla_get_u32(attr); 1769 if (*limit > MPTCP_PM_ADDR_MAX) { 1770 GENL_SET_ERR_MSG(info, "limit greater than maximum"); 1771 return -EINVAL; 1772 } 1773 return 0; 1774 } 1775 1776 static int 1777 mptcp_nl_cmd_set_limits(struct sk_buff *skb, struct genl_info *info) 1778 { 1779 struct pm_nl_pernet *pernet = genl_info_pm_nl(info); 1780 unsigned int rcv_addrs, subflows; 1781 int ret; 1782 1783 spin_lock_bh(&pernet->lock); 1784 rcv_addrs = pernet->add_addr_accept_max; 1785 ret = parse_limit(info, MPTCP_PM_ATTR_RCV_ADD_ADDRS, &rcv_addrs); 1786 if (ret) 1787 goto unlock; 1788 1789 subflows = pernet->subflows_max; 1790 ret = parse_limit(info, MPTCP_PM_ATTR_SUBFLOWS, &subflows); 1791 if (ret) 1792 goto unlock; 1793 1794 WRITE_ONCE(pernet->add_addr_accept_max, rcv_addrs); 1795 WRITE_ONCE(pernet->subflows_max, subflows); 1796 1797 unlock: 1798 spin_unlock_bh(&pernet->lock); 1799 return ret; 1800 } 1801 1802 static int 1803 mptcp_nl_cmd_get_limits(struct sk_buff *skb, struct genl_info *info) 1804 { 1805 struct pm_nl_pernet *pernet = genl_info_pm_nl(info); 1806 struct sk_buff *msg; 1807 void *reply; 1808 1809 msg = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL); 1810 if (!msg) 1811 return -ENOMEM; 1812 1813 reply = genlmsg_put_reply(msg, info, &mptcp_genl_family, 0, 1814 MPTCP_PM_CMD_GET_LIMITS); 1815 if (!reply) 1816 goto fail; 1817 1818 if (nla_put_u32(msg, MPTCP_PM_ATTR_RCV_ADD_ADDRS, 1819 READ_ONCE(pernet->add_addr_accept_max))) 1820 goto fail; 1821 1822 if (nla_put_u32(msg, MPTCP_PM_ATTR_SUBFLOWS, 1823 READ_ONCE(pernet->subflows_max))) 1824 goto fail; 1825 1826 genlmsg_end(msg, reply); 1827 return genlmsg_reply(msg, info); 1828 1829 fail: 1830 GENL_SET_ERR_MSG(info, "not enough space in Netlink message"); 1831 nlmsg_free(msg); 1832 return -EMSGSIZE; 1833 } 1834 1835 static void mptcp_pm_nl_fullmesh(struct mptcp_sock *msk, 1836 struct mptcp_addr_info *addr) 1837 { 1838 struct mptcp_rm_list list = { .nr = 0 }; 1839 1840 list.ids[list.nr++] = addr->id; 1841 1842 spin_lock_bh(&msk->pm.lock); 1843 mptcp_pm_nl_rm_subflow_received(msk, &list); 1844 mptcp_pm_create_subflow_or_signal_addr(msk); 1845 spin_unlock_bh(&msk->pm.lock); 1846 } 1847 1848 static int mptcp_nl_set_flags(struct net *net, 1849 struct mptcp_addr_info *addr, 1850 u8 bkup, u8 changed) 1851 { 1852 long s_slot = 0, s_num = 0; 1853 struct mptcp_sock *msk; 1854 int ret = -EINVAL; 1855 1856 while ((msk = mptcp_token_iter_next(net, &s_slot, &s_num)) != NULL) { 1857 struct sock *sk = (struct sock *)msk; 1858 1859 if (list_empty(&msk->conn_list) || mptcp_pm_is_userspace(msk)) 1860 goto next; 1861 1862 lock_sock(sk); 1863 if (changed & MPTCP_PM_ADDR_FLAG_BACKUP) 1864 ret = mptcp_pm_nl_mp_prio_send_ack(msk, addr, NULL, bkup); 1865 if (changed & MPTCP_PM_ADDR_FLAG_FULLMESH) 1866 mptcp_pm_nl_fullmesh(msk, addr); 1867 release_sock(sk); 1868 1869 next: 1870 sock_put(sk); 1871 cond_resched(); 1872 } 1873 1874 return ret; 1875 } 1876 1877 static int mptcp_nl_cmd_set_flags(struct sk_buff *skb, struct genl_info *info) 1878 { 1879 struct mptcp_pm_addr_entry addr = { .addr = { .family = AF_UNSPEC }, }, *entry; 1880 struct mptcp_pm_addr_entry remote = { .addr = { .family = AF_UNSPEC }, }; 1881 struct nlattr *attr_rem = info->attrs[MPTCP_PM_ATTR_ADDR_REMOTE]; 1882 struct nlattr *token = info->attrs[MPTCP_PM_ATTR_TOKEN]; 1883 struct nlattr *attr = info->attrs[MPTCP_PM_ATTR_ADDR]; 1884 struct pm_nl_pernet *pernet = genl_info_pm_nl(info); 1885 u8 changed, mask = MPTCP_PM_ADDR_FLAG_BACKUP | 1886 MPTCP_PM_ADDR_FLAG_FULLMESH; 1887 struct net *net = sock_net(skb->sk); 1888 u8 bkup = 0, lookup_by_id = 0; 1889 int ret; 1890 1891 ret = mptcp_pm_parse_entry(attr, info, false, &addr); 1892 if (ret < 0) 1893 return ret; 1894 1895 if (attr_rem) { 1896 ret = mptcp_pm_parse_entry(attr_rem, info, false, &remote); 1897 if (ret < 0) 1898 return ret; 1899 } 1900 1901 if (addr.flags & MPTCP_PM_ADDR_FLAG_BACKUP) 1902 bkup = 1; 1903 if (addr.addr.family == AF_UNSPEC) { 1904 lookup_by_id = 1; 1905 if (!addr.addr.id) 1906 return -EOPNOTSUPP; 1907 } 1908 1909 if (token) 1910 return mptcp_userspace_pm_set_flags(net, token, &addr, &remote, bkup); 1911 1912 spin_lock_bh(&pernet->lock); 1913 entry = __lookup_addr(pernet, &addr.addr, lookup_by_id); 1914 if (!entry) { 1915 spin_unlock_bh(&pernet->lock); 1916 return -EINVAL; 1917 } 1918 if ((addr.flags & MPTCP_PM_ADDR_FLAG_FULLMESH) && 1919 (entry->flags & MPTCP_PM_ADDR_FLAG_SIGNAL)) { 1920 spin_unlock_bh(&pernet->lock); 1921 return -EINVAL; 1922 } 1923 1924 changed = (addr.flags ^ entry->flags) & mask; 1925 entry->flags = (entry->flags & ~mask) | (addr.flags & mask); 1926 addr = *entry; 1927 spin_unlock_bh(&pernet->lock); 1928 1929 mptcp_nl_set_flags(net, &addr.addr, bkup, changed); 1930 return 0; 1931 } 1932 1933 static void mptcp_nl_mcast_send(struct net *net, struct sk_buff *nlskb, gfp_t gfp) 1934 { 1935 genlmsg_multicast_netns(&mptcp_genl_family, net, 1936 nlskb, 0, MPTCP_PM_EV_GRP_OFFSET, gfp); 1937 } 1938 1939 bool mptcp_userspace_pm_active(const struct mptcp_sock *msk) 1940 { 1941 return genl_has_listeners(&mptcp_genl_family, 1942 sock_net((const struct sock *)msk), 1943 MPTCP_PM_EV_GRP_OFFSET); 1944 } 1945 1946 static int mptcp_event_add_subflow(struct sk_buff *skb, const struct sock *ssk) 1947 { 1948 const struct inet_sock *issk = inet_sk(ssk); 1949 const struct mptcp_subflow_context *sf; 1950 1951 if (nla_put_u16(skb, MPTCP_ATTR_FAMILY, ssk->sk_family)) 1952 return -EMSGSIZE; 1953 1954 switch (ssk->sk_family) { 1955 case AF_INET: 1956 if (nla_put_in_addr(skb, MPTCP_ATTR_SADDR4, issk->inet_saddr)) 1957 return -EMSGSIZE; 1958 if (nla_put_in_addr(skb, MPTCP_ATTR_DADDR4, issk->inet_daddr)) 1959 return -EMSGSIZE; 1960 break; 1961 #if IS_ENABLED(CONFIG_MPTCP_IPV6) 1962 case AF_INET6: { 1963 const struct ipv6_pinfo *np = inet6_sk(ssk); 1964 1965 if (nla_put_in6_addr(skb, MPTCP_ATTR_SADDR6, &np->saddr)) 1966 return -EMSGSIZE; 1967 if (nla_put_in6_addr(skb, MPTCP_ATTR_DADDR6, &ssk->sk_v6_daddr)) 1968 return -EMSGSIZE; 1969 break; 1970 } 1971 #endif 1972 default: 1973 WARN_ON_ONCE(1); 1974 return -EMSGSIZE; 1975 } 1976 1977 if (nla_put_be16(skb, MPTCP_ATTR_SPORT, issk->inet_sport)) 1978 return -EMSGSIZE; 1979 if (nla_put_be16(skb, MPTCP_ATTR_DPORT, issk->inet_dport)) 1980 return -EMSGSIZE; 1981 1982 sf = mptcp_subflow_ctx(ssk); 1983 if (WARN_ON_ONCE(!sf)) 1984 return -EINVAL; 1985 1986 if (nla_put_u8(skb, MPTCP_ATTR_LOC_ID, sf->local_id)) 1987 return -EMSGSIZE; 1988 1989 if (nla_put_u8(skb, MPTCP_ATTR_REM_ID, sf->remote_id)) 1990 return -EMSGSIZE; 1991 1992 return 0; 1993 } 1994 1995 static int mptcp_event_put_token_and_ssk(struct sk_buff *skb, 1996 const struct mptcp_sock *msk, 1997 const struct sock *ssk) 1998 { 1999 const struct sock *sk = (const struct sock *)msk; 2000 const struct mptcp_subflow_context *sf; 2001 u8 sk_err; 2002 2003 if (nla_put_u32(skb, MPTCP_ATTR_TOKEN, msk->token)) 2004 return -EMSGSIZE; 2005 2006 if (mptcp_event_add_subflow(skb, ssk)) 2007 return -EMSGSIZE; 2008 2009 sf = mptcp_subflow_ctx(ssk); 2010 if (WARN_ON_ONCE(!sf)) 2011 return -EINVAL; 2012 2013 if (nla_put_u8(skb, MPTCP_ATTR_BACKUP, sf->backup)) 2014 return -EMSGSIZE; 2015 2016 if (ssk->sk_bound_dev_if && 2017 nla_put_s32(skb, MPTCP_ATTR_IF_IDX, ssk->sk_bound_dev_if)) 2018 return -EMSGSIZE; 2019 2020 sk_err = ssk->sk_err; 2021 if (sk_err && sk->sk_state == TCP_ESTABLISHED && 2022 nla_put_u8(skb, MPTCP_ATTR_ERROR, sk_err)) 2023 return -EMSGSIZE; 2024 2025 return 0; 2026 } 2027 2028 static int mptcp_event_sub_established(struct sk_buff *skb, 2029 const struct mptcp_sock *msk, 2030 const struct sock *ssk) 2031 { 2032 return mptcp_event_put_token_and_ssk(skb, msk, ssk); 2033 } 2034 2035 static int mptcp_event_sub_closed(struct sk_buff *skb, 2036 const struct mptcp_sock *msk, 2037 const struct sock *ssk) 2038 { 2039 const struct mptcp_subflow_context *sf; 2040 2041 if (mptcp_event_put_token_and_ssk(skb, msk, ssk)) 2042 return -EMSGSIZE; 2043 2044 sf = mptcp_subflow_ctx(ssk); 2045 if (!sf->reset_seen) 2046 return 0; 2047 2048 if (nla_put_u32(skb, MPTCP_ATTR_RESET_REASON, sf->reset_reason)) 2049 return -EMSGSIZE; 2050 2051 if (nla_put_u32(skb, MPTCP_ATTR_RESET_FLAGS, sf->reset_transient)) 2052 return -EMSGSIZE; 2053 2054 return 0; 2055 } 2056 2057 static int mptcp_event_created(struct sk_buff *skb, 2058 const struct mptcp_sock *msk, 2059 const struct sock *ssk) 2060 { 2061 int err = nla_put_u32(skb, MPTCP_ATTR_TOKEN, msk->token); 2062 2063 if (err) 2064 return err; 2065 2066 if (nla_put_u8(skb, MPTCP_ATTR_SERVER_SIDE, READ_ONCE(msk->pm.server_side))) 2067 return -EMSGSIZE; 2068 2069 return mptcp_event_add_subflow(skb, ssk); 2070 } 2071 2072 void mptcp_event_addr_removed(const struct mptcp_sock *msk, uint8_t id) 2073 { 2074 struct net *net = sock_net((const struct sock *)msk); 2075 struct nlmsghdr *nlh; 2076 struct sk_buff *skb; 2077 2078 if (!genl_has_listeners(&mptcp_genl_family, net, MPTCP_PM_EV_GRP_OFFSET)) 2079 return; 2080 2081 skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_ATOMIC); 2082 if (!skb) 2083 return; 2084 2085 nlh = genlmsg_put(skb, 0, 0, &mptcp_genl_family, 0, MPTCP_EVENT_REMOVED); 2086 if (!nlh) 2087 goto nla_put_failure; 2088 2089 if (nla_put_u32(skb, MPTCP_ATTR_TOKEN, msk->token)) 2090 goto nla_put_failure; 2091 2092 if (nla_put_u8(skb, MPTCP_ATTR_REM_ID, id)) 2093 goto nla_put_failure; 2094 2095 genlmsg_end(skb, nlh); 2096 mptcp_nl_mcast_send(net, skb, GFP_ATOMIC); 2097 return; 2098 2099 nla_put_failure: 2100 nlmsg_free(skb); 2101 } 2102 2103 void mptcp_event_addr_announced(const struct sock *ssk, 2104 const struct mptcp_addr_info *info) 2105 { 2106 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk); 2107 struct mptcp_sock *msk = mptcp_sk(subflow->conn); 2108 struct net *net = sock_net(ssk); 2109 struct nlmsghdr *nlh; 2110 struct sk_buff *skb; 2111 2112 if (!genl_has_listeners(&mptcp_genl_family, net, MPTCP_PM_EV_GRP_OFFSET)) 2113 return; 2114 2115 skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_ATOMIC); 2116 if (!skb) 2117 return; 2118 2119 nlh = genlmsg_put(skb, 0, 0, &mptcp_genl_family, 0, 2120 MPTCP_EVENT_ANNOUNCED); 2121 if (!nlh) 2122 goto nla_put_failure; 2123 2124 if (nla_put_u32(skb, MPTCP_ATTR_TOKEN, msk->token)) 2125 goto nla_put_failure; 2126 2127 if (nla_put_u8(skb, MPTCP_ATTR_REM_ID, info->id)) 2128 goto nla_put_failure; 2129 2130 if (nla_put_be16(skb, MPTCP_ATTR_DPORT, 2131 info->port == 0 ? 2132 inet_sk(ssk)->inet_dport : 2133 info->port)) 2134 goto nla_put_failure; 2135 2136 switch (info->family) { 2137 case AF_INET: 2138 if (nla_put_in_addr(skb, MPTCP_ATTR_DADDR4, info->addr.s_addr)) 2139 goto nla_put_failure; 2140 break; 2141 #if IS_ENABLED(CONFIG_MPTCP_IPV6) 2142 case AF_INET6: 2143 if (nla_put_in6_addr(skb, MPTCP_ATTR_DADDR6, &info->addr6)) 2144 goto nla_put_failure; 2145 break; 2146 #endif 2147 default: 2148 WARN_ON_ONCE(1); 2149 goto nla_put_failure; 2150 } 2151 2152 genlmsg_end(skb, nlh); 2153 mptcp_nl_mcast_send(net, skb, GFP_ATOMIC); 2154 return; 2155 2156 nla_put_failure: 2157 nlmsg_free(skb); 2158 } 2159 2160 void mptcp_event_pm_listener(const struct sock *ssk, 2161 enum mptcp_event_type event) 2162 { 2163 const struct inet_sock *issk = inet_sk(ssk); 2164 struct net *net = sock_net(ssk); 2165 struct nlmsghdr *nlh; 2166 struct sk_buff *skb; 2167 2168 if (!genl_has_listeners(&mptcp_genl_family, net, MPTCP_PM_EV_GRP_OFFSET)) 2169 return; 2170 2171 skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL); 2172 if (!skb) 2173 return; 2174 2175 nlh = genlmsg_put(skb, 0, 0, &mptcp_genl_family, 0, event); 2176 if (!nlh) 2177 goto nla_put_failure; 2178 2179 if (nla_put_u16(skb, MPTCP_ATTR_FAMILY, ssk->sk_family)) 2180 goto nla_put_failure; 2181 2182 if (nla_put_be16(skb, MPTCP_ATTR_SPORT, issk->inet_sport)) 2183 goto nla_put_failure; 2184 2185 switch (ssk->sk_family) { 2186 case AF_INET: 2187 if (nla_put_in_addr(skb, MPTCP_ATTR_SADDR4, issk->inet_saddr)) 2188 goto nla_put_failure; 2189 break; 2190 #if IS_ENABLED(CONFIG_MPTCP_IPV6) 2191 case AF_INET6: { 2192 const struct ipv6_pinfo *np = inet6_sk(ssk); 2193 2194 if (nla_put_in6_addr(skb, MPTCP_ATTR_SADDR6, &np->saddr)) 2195 goto nla_put_failure; 2196 break; 2197 } 2198 #endif 2199 default: 2200 WARN_ON_ONCE(1); 2201 goto nla_put_failure; 2202 } 2203 2204 genlmsg_end(skb, nlh); 2205 mptcp_nl_mcast_send(net, skb, GFP_KERNEL); 2206 return; 2207 2208 nla_put_failure: 2209 nlmsg_free(skb); 2210 } 2211 2212 void mptcp_event(enum mptcp_event_type type, const struct mptcp_sock *msk, 2213 const struct sock *ssk, gfp_t gfp) 2214 { 2215 struct net *net = sock_net((const struct sock *)msk); 2216 struct nlmsghdr *nlh; 2217 struct sk_buff *skb; 2218 2219 if (!genl_has_listeners(&mptcp_genl_family, net, MPTCP_PM_EV_GRP_OFFSET)) 2220 return; 2221 2222 skb = nlmsg_new(NLMSG_DEFAULT_SIZE, gfp); 2223 if (!skb) 2224 return; 2225 2226 nlh = genlmsg_put(skb, 0, 0, &mptcp_genl_family, 0, type); 2227 if (!nlh) 2228 goto nla_put_failure; 2229 2230 switch (type) { 2231 case MPTCP_EVENT_UNSPEC: 2232 WARN_ON_ONCE(1); 2233 break; 2234 case MPTCP_EVENT_CREATED: 2235 case MPTCP_EVENT_ESTABLISHED: 2236 if (mptcp_event_created(skb, msk, ssk) < 0) 2237 goto nla_put_failure; 2238 break; 2239 case MPTCP_EVENT_CLOSED: 2240 if (nla_put_u32(skb, MPTCP_ATTR_TOKEN, msk->token) < 0) 2241 goto nla_put_failure; 2242 break; 2243 case MPTCP_EVENT_ANNOUNCED: 2244 case MPTCP_EVENT_REMOVED: 2245 /* call mptcp_event_addr_announced()/removed instead */ 2246 WARN_ON_ONCE(1); 2247 break; 2248 case MPTCP_EVENT_SUB_ESTABLISHED: 2249 case MPTCP_EVENT_SUB_PRIORITY: 2250 if (mptcp_event_sub_established(skb, msk, ssk) < 0) 2251 goto nla_put_failure; 2252 break; 2253 case MPTCP_EVENT_SUB_CLOSED: 2254 if (mptcp_event_sub_closed(skb, msk, ssk) < 0) 2255 goto nla_put_failure; 2256 break; 2257 case MPTCP_EVENT_LISTENER_CREATED: 2258 case MPTCP_EVENT_LISTENER_CLOSED: 2259 break; 2260 } 2261 2262 genlmsg_end(skb, nlh); 2263 mptcp_nl_mcast_send(net, skb, gfp); 2264 return; 2265 2266 nla_put_failure: 2267 nlmsg_free(skb); 2268 } 2269 2270 static const struct genl_small_ops mptcp_pm_ops[] = { 2271 { 2272 .cmd = MPTCP_PM_CMD_ADD_ADDR, 2273 .doit = mptcp_nl_cmd_add_addr, 2274 .flags = GENL_UNS_ADMIN_PERM, 2275 }, 2276 { 2277 .cmd = MPTCP_PM_CMD_DEL_ADDR, 2278 .doit = mptcp_nl_cmd_del_addr, 2279 .flags = GENL_UNS_ADMIN_PERM, 2280 }, 2281 { 2282 .cmd = MPTCP_PM_CMD_FLUSH_ADDRS, 2283 .doit = mptcp_nl_cmd_flush_addrs, 2284 .flags = GENL_UNS_ADMIN_PERM, 2285 }, 2286 { 2287 .cmd = MPTCP_PM_CMD_GET_ADDR, 2288 .doit = mptcp_nl_cmd_get_addr, 2289 .dumpit = mptcp_nl_cmd_dump_addrs, 2290 }, 2291 { 2292 .cmd = MPTCP_PM_CMD_SET_LIMITS, 2293 .doit = mptcp_nl_cmd_set_limits, 2294 .flags = GENL_UNS_ADMIN_PERM, 2295 }, 2296 { 2297 .cmd = MPTCP_PM_CMD_GET_LIMITS, 2298 .doit = mptcp_nl_cmd_get_limits, 2299 }, 2300 { 2301 .cmd = MPTCP_PM_CMD_SET_FLAGS, 2302 .doit = mptcp_nl_cmd_set_flags, 2303 .flags = GENL_UNS_ADMIN_PERM, 2304 }, 2305 { 2306 .cmd = MPTCP_PM_CMD_ANNOUNCE, 2307 .doit = mptcp_nl_cmd_announce, 2308 .flags = GENL_UNS_ADMIN_PERM, 2309 }, 2310 { 2311 .cmd = MPTCP_PM_CMD_REMOVE, 2312 .doit = mptcp_nl_cmd_remove, 2313 .flags = GENL_UNS_ADMIN_PERM, 2314 }, 2315 { 2316 .cmd = MPTCP_PM_CMD_SUBFLOW_CREATE, 2317 .doit = mptcp_nl_cmd_sf_create, 2318 .flags = GENL_UNS_ADMIN_PERM, 2319 }, 2320 { 2321 .cmd = MPTCP_PM_CMD_SUBFLOW_DESTROY, 2322 .doit = mptcp_nl_cmd_sf_destroy, 2323 .flags = GENL_UNS_ADMIN_PERM, 2324 }, 2325 }; 2326 2327 static struct genl_family mptcp_genl_family __ro_after_init = { 2328 .name = MPTCP_PM_NAME, 2329 .version = MPTCP_PM_VER, 2330 .maxattr = MPTCP_PM_ATTR_MAX, 2331 .policy = mptcp_pm_policy, 2332 .netnsok = true, 2333 .module = THIS_MODULE, 2334 .small_ops = mptcp_pm_ops, 2335 .n_small_ops = ARRAY_SIZE(mptcp_pm_ops), 2336 .resv_start_op = MPTCP_PM_CMD_SUBFLOW_DESTROY + 1, 2337 .mcgrps = mptcp_pm_mcgrps, 2338 .n_mcgrps = ARRAY_SIZE(mptcp_pm_mcgrps), 2339 }; 2340 2341 static int __net_init pm_nl_init_net(struct net *net) 2342 { 2343 struct pm_nl_pernet *pernet = pm_nl_get_pernet(net); 2344 2345 INIT_LIST_HEAD_RCU(&pernet->local_addr_list); 2346 2347 /* Cit. 2 subflows ought to be enough for anybody. */ 2348 pernet->subflows_max = 2; 2349 pernet->next_id = 1; 2350 pernet->stale_loss_cnt = 4; 2351 spin_lock_init(&pernet->lock); 2352 2353 /* No need to initialize other pernet fields, the struct is zeroed at 2354 * allocation time. 2355 */ 2356 2357 return 0; 2358 } 2359 2360 static void __net_exit pm_nl_exit_net(struct list_head *net_list) 2361 { 2362 struct net *net; 2363 2364 list_for_each_entry(net, net_list, exit_list) { 2365 struct pm_nl_pernet *pernet = pm_nl_get_pernet(net); 2366 2367 /* net is removed from namespace list, can't race with 2368 * other modifiers, also netns core already waited for a 2369 * RCU grace period. 2370 */ 2371 __flush_addrs(&pernet->local_addr_list); 2372 } 2373 } 2374 2375 static struct pernet_operations mptcp_pm_pernet_ops = { 2376 .init = pm_nl_init_net, 2377 .exit_batch = pm_nl_exit_net, 2378 .id = &pm_nl_pernet_id, 2379 .size = sizeof(struct pm_nl_pernet), 2380 }; 2381 2382 void __init mptcp_pm_nl_init(void) 2383 { 2384 if (register_pernet_subsys(&mptcp_pm_pernet_ops) < 0) 2385 panic("Failed to register MPTCP PM pernet subsystem.\n"); 2386 2387 if (genl_register_family(&mptcp_genl_family)) 2388 panic("Failed to register MPTCP PM netlink family\n"); 2389 } 2390