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_addr_entry { 26 struct list_head list; 27 struct mptcp_addr_info addr; 28 struct rcu_head rcu; 29 }; 30 31 struct mptcp_pm_add_entry { 32 struct list_head list; 33 struct mptcp_addr_info addr; 34 struct timer_list add_timer; 35 struct mptcp_sock *sock; 36 u8 retrans_times; 37 }; 38 39 struct pm_nl_pernet { 40 /* protects pernet updates */ 41 spinlock_t lock; 42 struct list_head local_addr_list; 43 unsigned int addrs; 44 unsigned int add_addr_signal_max; 45 unsigned int add_addr_accept_max; 46 unsigned int local_addr_max; 47 unsigned int subflows_max; 48 unsigned int next_id; 49 }; 50 51 #define MPTCP_PM_ADDR_MAX 8 52 #define ADD_ADDR_RETRANS_MAX 3 53 54 static bool addresses_equal(const struct mptcp_addr_info *a, 55 struct mptcp_addr_info *b, bool use_port) 56 { 57 bool addr_equals = false; 58 59 if (a->family != b->family) 60 return false; 61 62 if (a->family == AF_INET) 63 addr_equals = a->addr.s_addr == b->addr.s_addr; 64 #if IS_ENABLED(CONFIG_MPTCP_IPV6) 65 else 66 addr_equals = !ipv6_addr_cmp(&a->addr6, &b->addr6); 67 #endif 68 69 if (!addr_equals) 70 return false; 71 if (!use_port) 72 return true; 73 74 return a->port == b->port; 75 } 76 77 static bool address_zero(const struct mptcp_addr_info *addr) 78 { 79 struct mptcp_addr_info zero; 80 81 memset(&zero, 0, sizeof(zero)); 82 zero.family = addr->family; 83 84 return addresses_equal(addr, &zero, false); 85 } 86 87 static void local_address(const struct sock_common *skc, 88 struct mptcp_addr_info *addr) 89 { 90 addr->port = 0; 91 addr->family = skc->skc_family; 92 if (addr->family == AF_INET) 93 addr->addr.s_addr = skc->skc_rcv_saddr; 94 #if IS_ENABLED(CONFIG_MPTCP_IPV6) 95 else if (addr->family == AF_INET6) 96 addr->addr6 = skc->skc_v6_rcv_saddr; 97 #endif 98 } 99 100 static void remote_address(const struct sock_common *skc, 101 struct mptcp_addr_info *addr) 102 { 103 addr->family = skc->skc_family; 104 addr->port = skc->skc_dport; 105 if (addr->family == AF_INET) 106 addr->addr.s_addr = skc->skc_daddr; 107 #if IS_ENABLED(CONFIG_MPTCP_IPV6) 108 else if (addr->family == AF_INET6) 109 addr->addr6 = skc->skc_v6_daddr; 110 #endif 111 } 112 113 static bool lookup_subflow_by_saddr(const struct list_head *list, 114 struct mptcp_addr_info *saddr) 115 { 116 struct mptcp_subflow_context *subflow; 117 struct mptcp_addr_info cur; 118 struct sock_common *skc; 119 120 list_for_each_entry(subflow, list, node) { 121 skc = (struct sock_common *)mptcp_subflow_tcp_sock(subflow); 122 123 local_address(skc, &cur); 124 if (addresses_equal(&cur, saddr, false)) 125 return true; 126 } 127 128 return false; 129 } 130 131 static struct mptcp_pm_addr_entry * 132 select_local_address(const struct pm_nl_pernet *pernet, 133 struct mptcp_sock *msk) 134 { 135 struct mptcp_pm_addr_entry *entry, *ret = NULL; 136 137 rcu_read_lock(); 138 spin_lock_bh(&msk->join_list_lock); 139 list_for_each_entry_rcu(entry, &pernet->local_addr_list, list) { 140 if (!(entry->addr.flags & MPTCP_PM_ADDR_FLAG_SUBFLOW)) 141 continue; 142 143 /* avoid any address already in use by subflows and 144 * pending join 145 */ 146 if (entry->addr.family == ((struct sock *)msk)->sk_family && 147 !lookup_subflow_by_saddr(&msk->conn_list, &entry->addr) && 148 !lookup_subflow_by_saddr(&msk->join_list, &entry->addr)) { 149 ret = entry; 150 break; 151 } 152 } 153 spin_unlock_bh(&msk->join_list_lock); 154 rcu_read_unlock(); 155 return ret; 156 } 157 158 static struct mptcp_pm_addr_entry * 159 select_signal_address(struct pm_nl_pernet *pernet, unsigned int pos) 160 { 161 struct mptcp_pm_addr_entry *entry, *ret = NULL; 162 int i = 0; 163 164 rcu_read_lock(); 165 /* do not keep any additional per socket state, just signal 166 * the address list in order. 167 * Note: removal from the local address list during the msk life-cycle 168 * can lead to additional addresses not being announced. 169 */ 170 list_for_each_entry_rcu(entry, &pernet->local_addr_list, list) { 171 if (!(entry->addr.flags & MPTCP_PM_ADDR_FLAG_SIGNAL)) 172 continue; 173 if (i++ == pos) { 174 ret = entry; 175 break; 176 } 177 } 178 rcu_read_unlock(); 179 return ret; 180 } 181 182 static void check_work_pending(struct mptcp_sock *msk) 183 { 184 if (msk->pm.add_addr_signaled == msk->pm.add_addr_signal_max && 185 (msk->pm.local_addr_used == msk->pm.local_addr_max || 186 msk->pm.subflows == msk->pm.subflows_max)) 187 WRITE_ONCE(msk->pm.work_pending, false); 188 } 189 190 static struct mptcp_pm_add_entry * 191 lookup_anno_list_by_saddr(struct mptcp_sock *msk, 192 struct mptcp_addr_info *addr) 193 { 194 struct mptcp_pm_add_entry *entry; 195 196 list_for_each_entry(entry, &msk->pm.anno_list, list) { 197 if (addresses_equal(&entry->addr, addr, false)) 198 return entry; 199 } 200 201 return NULL; 202 } 203 204 static void mptcp_pm_add_timer(struct timer_list *timer) 205 { 206 struct mptcp_pm_add_entry *entry = from_timer(entry, timer, add_timer); 207 struct mptcp_sock *msk = entry->sock; 208 struct sock *sk = (struct sock *)msk; 209 210 pr_debug("msk=%p", msk); 211 212 if (!msk) 213 return; 214 215 if (inet_sk_state_load(sk) == TCP_CLOSE) 216 return; 217 218 if (!entry->addr.id) 219 return; 220 221 if (mptcp_pm_should_add_signal(msk)) { 222 sk_reset_timer(sk, timer, jiffies + TCP_RTO_MAX / 8); 223 goto out; 224 } 225 226 spin_lock_bh(&msk->pm.lock); 227 228 if (!mptcp_pm_should_add_signal(msk)) { 229 pr_debug("retransmit ADD_ADDR id=%d", entry->addr.id); 230 mptcp_pm_announce_addr(msk, &entry->addr, false); 231 mptcp_pm_add_addr_send_ack(msk); 232 entry->retrans_times++; 233 } 234 235 if (entry->retrans_times < ADD_ADDR_RETRANS_MAX) 236 sk_reset_timer(sk, timer, 237 jiffies + mptcp_get_add_addr_timeout(sock_net(sk))); 238 239 spin_unlock_bh(&msk->pm.lock); 240 241 out: 242 __sock_put(sk); 243 } 244 245 struct mptcp_pm_add_entry * 246 mptcp_pm_del_add_timer(struct mptcp_sock *msk, 247 struct mptcp_addr_info *addr) 248 { 249 struct mptcp_pm_add_entry *entry; 250 struct sock *sk = (struct sock *)msk; 251 252 spin_lock_bh(&msk->pm.lock); 253 entry = lookup_anno_list_by_saddr(msk, addr); 254 if (entry) 255 entry->retrans_times = ADD_ADDR_RETRANS_MAX; 256 spin_unlock_bh(&msk->pm.lock); 257 258 if (entry) 259 sk_stop_timer_sync(sk, &entry->add_timer); 260 261 return entry; 262 } 263 264 static bool mptcp_pm_alloc_anno_list(struct mptcp_sock *msk, 265 struct mptcp_pm_addr_entry *entry) 266 { 267 struct mptcp_pm_add_entry *add_entry = NULL; 268 struct sock *sk = (struct sock *)msk; 269 struct net *net = sock_net(sk); 270 271 if (lookup_anno_list_by_saddr(msk, &entry->addr)) 272 return false; 273 274 add_entry = kmalloc(sizeof(*add_entry), GFP_ATOMIC); 275 if (!add_entry) 276 return false; 277 278 list_add(&add_entry->list, &msk->pm.anno_list); 279 280 add_entry->addr = entry->addr; 281 add_entry->sock = msk; 282 add_entry->retrans_times = 0; 283 284 timer_setup(&add_entry->add_timer, mptcp_pm_add_timer, 0); 285 sk_reset_timer(sk, &add_entry->add_timer, 286 jiffies + mptcp_get_add_addr_timeout(net)); 287 288 return true; 289 } 290 291 void mptcp_pm_free_anno_list(struct mptcp_sock *msk) 292 { 293 struct mptcp_pm_add_entry *entry, *tmp; 294 struct sock *sk = (struct sock *)msk; 295 LIST_HEAD(free_list); 296 297 pr_debug("msk=%p", msk); 298 299 spin_lock_bh(&msk->pm.lock); 300 list_splice_init(&msk->pm.anno_list, &free_list); 301 spin_unlock_bh(&msk->pm.lock); 302 303 list_for_each_entry_safe(entry, tmp, &free_list, list) { 304 sk_stop_timer_sync(sk, &entry->add_timer); 305 kfree(entry); 306 } 307 } 308 309 static void mptcp_pm_create_subflow_or_signal_addr(struct mptcp_sock *msk) 310 { 311 struct mptcp_addr_info remote = { 0 }; 312 struct sock *sk = (struct sock *)msk; 313 struct mptcp_pm_addr_entry *local; 314 struct pm_nl_pernet *pernet; 315 316 pernet = net_generic(sock_net((struct sock *)msk), pm_nl_pernet_id); 317 318 pr_debug("local %d:%d signal %d:%d subflows %d:%d\n", 319 msk->pm.local_addr_used, msk->pm.local_addr_max, 320 msk->pm.add_addr_signaled, msk->pm.add_addr_signal_max, 321 msk->pm.subflows, msk->pm.subflows_max); 322 323 /* check first for announce */ 324 if (msk->pm.add_addr_signaled < msk->pm.add_addr_signal_max) { 325 local = select_signal_address(pernet, 326 msk->pm.add_addr_signaled); 327 328 if (local) { 329 if (mptcp_pm_alloc_anno_list(msk, local)) { 330 msk->pm.add_addr_signaled++; 331 mptcp_pm_announce_addr(msk, &local->addr, false); 332 mptcp_pm_nl_add_addr_send_ack(msk); 333 } 334 } else { 335 /* pick failed, avoid fourther attempts later */ 336 msk->pm.local_addr_used = msk->pm.add_addr_signal_max; 337 } 338 339 check_work_pending(msk); 340 } 341 342 /* check if should create a new subflow */ 343 if (msk->pm.local_addr_used < msk->pm.local_addr_max && 344 msk->pm.subflows < msk->pm.subflows_max) { 345 remote_address((struct sock_common *)sk, &remote); 346 347 local = select_local_address(pernet, msk); 348 if (local) { 349 msk->pm.local_addr_used++; 350 msk->pm.subflows++; 351 check_work_pending(msk); 352 spin_unlock_bh(&msk->pm.lock); 353 __mptcp_subflow_connect(sk, &local->addr, &remote); 354 spin_lock_bh(&msk->pm.lock); 355 return; 356 } 357 358 /* lookup failed, avoid fourther attempts later */ 359 msk->pm.local_addr_used = msk->pm.local_addr_max; 360 check_work_pending(msk); 361 } 362 } 363 364 void mptcp_pm_nl_fully_established(struct mptcp_sock *msk) 365 { 366 mptcp_pm_create_subflow_or_signal_addr(msk); 367 } 368 369 void mptcp_pm_nl_subflow_established(struct mptcp_sock *msk) 370 { 371 mptcp_pm_create_subflow_or_signal_addr(msk); 372 } 373 374 void mptcp_pm_nl_add_addr_received(struct mptcp_sock *msk) 375 { 376 struct sock *sk = (struct sock *)msk; 377 struct mptcp_addr_info remote; 378 struct mptcp_addr_info local; 379 380 pr_debug("accepted %d:%d remote family %d", 381 msk->pm.add_addr_accepted, msk->pm.add_addr_accept_max, 382 msk->pm.remote.family); 383 msk->pm.add_addr_accepted++; 384 msk->pm.subflows++; 385 if (msk->pm.add_addr_accepted >= msk->pm.add_addr_accept_max || 386 msk->pm.subflows >= msk->pm.subflows_max) 387 WRITE_ONCE(msk->pm.accept_addr, false); 388 389 /* connect to the specified remote address, using whatever 390 * local address the routing configuration will pick. 391 */ 392 remote = msk->pm.remote; 393 if (!remote.port) 394 remote.port = sk->sk_dport; 395 memset(&local, 0, sizeof(local)); 396 local.family = remote.family; 397 398 spin_unlock_bh(&msk->pm.lock); 399 __mptcp_subflow_connect((struct sock *)msk, &local, &remote); 400 spin_lock_bh(&msk->pm.lock); 401 402 mptcp_pm_announce_addr(msk, &remote, true); 403 mptcp_pm_nl_add_addr_send_ack(msk); 404 } 405 406 void mptcp_pm_nl_add_addr_send_ack(struct mptcp_sock *msk) 407 { 408 struct mptcp_subflow_context *subflow; 409 410 if (!mptcp_pm_should_add_signal_ipv6(msk)) 411 return; 412 413 __mptcp_flush_join_list(msk); 414 subflow = list_first_entry_or_null(&msk->conn_list, typeof(*subflow), node); 415 if (subflow) { 416 struct sock *ssk = mptcp_subflow_tcp_sock(subflow); 417 u8 add_addr; 418 419 spin_unlock_bh(&msk->pm.lock); 420 pr_debug("send ack for add_addr6"); 421 lock_sock(ssk); 422 tcp_send_ack(ssk); 423 release_sock(ssk); 424 spin_lock_bh(&msk->pm.lock); 425 426 add_addr = READ_ONCE(msk->pm.add_addr_signal); 427 add_addr &= ~BIT(MPTCP_ADD_ADDR_IPV6); 428 WRITE_ONCE(msk->pm.add_addr_signal, add_addr); 429 } 430 } 431 432 void mptcp_pm_nl_rm_addr_received(struct mptcp_sock *msk) 433 { 434 struct mptcp_subflow_context *subflow, *tmp; 435 struct sock *sk = (struct sock *)msk; 436 437 pr_debug("address rm_id %d", msk->pm.rm_id); 438 439 if (!msk->pm.rm_id) 440 return; 441 442 if (list_empty(&msk->conn_list)) 443 return; 444 445 list_for_each_entry_safe(subflow, tmp, &msk->conn_list, node) { 446 struct sock *ssk = mptcp_subflow_tcp_sock(subflow); 447 int how = RCV_SHUTDOWN | SEND_SHUTDOWN; 448 449 if (msk->pm.rm_id != subflow->remote_id) 450 continue; 451 452 spin_unlock_bh(&msk->pm.lock); 453 mptcp_subflow_shutdown(sk, ssk, how); 454 __mptcp_close_ssk(sk, ssk, subflow); 455 spin_lock_bh(&msk->pm.lock); 456 457 msk->pm.add_addr_accepted--; 458 msk->pm.subflows--; 459 WRITE_ONCE(msk->pm.accept_addr, true); 460 461 __MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_RMADDR); 462 463 break; 464 } 465 } 466 467 void mptcp_pm_nl_rm_subflow_received(struct mptcp_sock *msk, u8 rm_id) 468 { 469 struct mptcp_subflow_context *subflow, *tmp; 470 struct sock *sk = (struct sock *)msk; 471 472 pr_debug("subflow rm_id %d", rm_id); 473 474 if (!rm_id) 475 return; 476 477 if (list_empty(&msk->conn_list)) 478 return; 479 480 list_for_each_entry_safe(subflow, tmp, &msk->conn_list, node) { 481 struct sock *ssk = mptcp_subflow_tcp_sock(subflow); 482 int how = RCV_SHUTDOWN | SEND_SHUTDOWN; 483 484 if (rm_id != subflow->local_id) 485 continue; 486 487 spin_unlock_bh(&msk->pm.lock); 488 mptcp_subflow_shutdown(sk, ssk, how); 489 __mptcp_close_ssk(sk, ssk, subflow); 490 spin_lock_bh(&msk->pm.lock); 491 492 msk->pm.local_addr_used--; 493 msk->pm.subflows--; 494 495 __MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_RMSUBFLOW); 496 497 break; 498 } 499 } 500 501 static bool address_use_port(struct mptcp_pm_addr_entry *entry) 502 { 503 return (entry->addr.flags & 504 (MPTCP_PM_ADDR_FLAG_SIGNAL | MPTCP_PM_ADDR_FLAG_SUBFLOW)) == 505 MPTCP_PM_ADDR_FLAG_SIGNAL; 506 } 507 508 static int mptcp_pm_nl_append_new_local_addr(struct pm_nl_pernet *pernet, 509 struct mptcp_pm_addr_entry *entry) 510 { 511 struct mptcp_pm_addr_entry *cur; 512 int ret = -EINVAL; 513 514 spin_lock_bh(&pernet->lock); 515 /* to keep the code simple, don't do IDR-like allocation for address ID, 516 * just bail when we exceed limits 517 */ 518 if (pernet->next_id > 255) 519 goto out; 520 if (pernet->addrs >= MPTCP_PM_ADDR_MAX) 521 goto out; 522 523 /* do not insert duplicate address, differentiate on port only 524 * singled addresses 525 */ 526 list_for_each_entry(cur, &pernet->local_addr_list, list) { 527 if (addresses_equal(&cur->addr, &entry->addr, 528 address_use_port(entry) && 529 address_use_port(cur))) 530 goto out; 531 } 532 533 if (entry->addr.flags & MPTCP_PM_ADDR_FLAG_SIGNAL) 534 pernet->add_addr_signal_max++; 535 if (entry->addr.flags & MPTCP_PM_ADDR_FLAG_SUBFLOW) 536 pernet->local_addr_max++; 537 538 entry->addr.id = pernet->next_id++; 539 pernet->addrs++; 540 list_add_tail_rcu(&entry->list, &pernet->local_addr_list); 541 ret = entry->addr.id; 542 543 out: 544 spin_unlock_bh(&pernet->lock); 545 return ret; 546 } 547 548 int mptcp_pm_nl_get_local_id(struct mptcp_sock *msk, struct sock_common *skc) 549 { 550 struct mptcp_pm_addr_entry *entry; 551 struct mptcp_addr_info skc_local; 552 struct mptcp_addr_info msk_local; 553 struct pm_nl_pernet *pernet; 554 int ret = -1; 555 556 if (WARN_ON_ONCE(!msk)) 557 return -1; 558 559 /* The 0 ID mapping is defined by the first subflow, copied into the msk 560 * addr 561 */ 562 local_address((struct sock_common *)msk, &msk_local); 563 local_address((struct sock_common *)skc, &skc_local); 564 if (addresses_equal(&msk_local, &skc_local, false)) 565 return 0; 566 567 if (address_zero(&skc_local)) 568 return 0; 569 570 pernet = net_generic(sock_net((struct sock *)msk), pm_nl_pernet_id); 571 572 rcu_read_lock(); 573 list_for_each_entry_rcu(entry, &pernet->local_addr_list, list) { 574 if (addresses_equal(&entry->addr, &skc_local, false)) { 575 ret = entry->addr.id; 576 break; 577 } 578 } 579 rcu_read_unlock(); 580 if (ret >= 0) 581 return ret; 582 583 /* address not found, add to local list */ 584 entry = kmalloc(sizeof(*entry), GFP_ATOMIC); 585 if (!entry) 586 return -ENOMEM; 587 588 entry->addr = skc_local; 589 entry->addr.ifindex = 0; 590 entry->addr.flags = 0; 591 ret = mptcp_pm_nl_append_new_local_addr(pernet, entry); 592 if (ret < 0) 593 kfree(entry); 594 595 return ret; 596 } 597 598 void mptcp_pm_nl_data_init(struct mptcp_sock *msk) 599 { 600 struct mptcp_pm_data *pm = &msk->pm; 601 struct pm_nl_pernet *pernet; 602 bool subflows; 603 604 pernet = net_generic(sock_net((struct sock *)msk), pm_nl_pernet_id); 605 606 pm->add_addr_signal_max = READ_ONCE(pernet->add_addr_signal_max); 607 pm->add_addr_accept_max = READ_ONCE(pernet->add_addr_accept_max); 608 pm->local_addr_max = READ_ONCE(pernet->local_addr_max); 609 pm->subflows_max = READ_ONCE(pernet->subflows_max); 610 subflows = !!pm->subflows_max; 611 WRITE_ONCE(pm->work_pending, (!!pm->local_addr_max && subflows) || 612 !!pm->add_addr_signal_max); 613 WRITE_ONCE(pm->accept_addr, !!pm->add_addr_accept_max && subflows); 614 WRITE_ONCE(pm->accept_subflow, subflows); 615 } 616 617 #define MPTCP_PM_CMD_GRP_OFFSET 0 618 619 static const struct genl_multicast_group mptcp_pm_mcgrps[] = { 620 [MPTCP_PM_CMD_GRP_OFFSET] = { .name = MPTCP_PM_CMD_GRP_NAME, }, 621 }; 622 623 static const struct nla_policy 624 mptcp_pm_addr_policy[MPTCP_PM_ADDR_ATTR_MAX + 1] = { 625 [MPTCP_PM_ADDR_ATTR_FAMILY] = { .type = NLA_U16, }, 626 [MPTCP_PM_ADDR_ATTR_ID] = { .type = NLA_U8, }, 627 [MPTCP_PM_ADDR_ATTR_ADDR4] = { .type = NLA_U32, }, 628 [MPTCP_PM_ADDR_ATTR_ADDR6] = 629 NLA_POLICY_EXACT_LEN(sizeof(struct in6_addr)), 630 [MPTCP_PM_ADDR_ATTR_PORT] = { .type = NLA_U16 }, 631 [MPTCP_PM_ADDR_ATTR_FLAGS] = { .type = NLA_U32 }, 632 [MPTCP_PM_ADDR_ATTR_IF_IDX] = { .type = NLA_S32 }, 633 }; 634 635 static const struct nla_policy mptcp_pm_policy[MPTCP_PM_ATTR_MAX + 1] = { 636 [MPTCP_PM_ATTR_ADDR] = 637 NLA_POLICY_NESTED(mptcp_pm_addr_policy), 638 [MPTCP_PM_ATTR_RCV_ADD_ADDRS] = { .type = NLA_U32, }, 639 [MPTCP_PM_ATTR_SUBFLOWS] = { .type = NLA_U32, }, 640 }; 641 642 static int mptcp_pm_family_to_addr(int family) 643 { 644 #if IS_ENABLED(CONFIG_MPTCP_IPV6) 645 if (family == AF_INET6) 646 return MPTCP_PM_ADDR_ATTR_ADDR6; 647 #endif 648 return MPTCP_PM_ADDR_ATTR_ADDR4; 649 } 650 651 static int mptcp_pm_parse_addr(struct nlattr *attr, struct genl_info *info, 652 bool require_family, 653 struct mptcp_pm_addr_entry *entry) 654 { 655 struct nlattr *tb[MPTCP_PM_ADDR_ATTR_MAX + 1]; 656 int err, addr_addr; 657 658 if (!attr) { 659 GENL_SET_ERR_MSG(info, "missing address info"); 660 return -EINVAL; 661 } 662 663 /* no validation needed - was already done via nested policy */ 664 err = nla_parse_nested_deprecated(tb, MPTCP_PM_ADDR_ATTR_MAX, attr, 665 mptcp_pm_addr_policy, info->extack); 666 if (err) 667 return err; 668 669 memset(entry, 0, sizeof(*entry)); 670 if (!tb[MPTCP_PM_ADDR_ATTR_FAMILY]) { 671 if (!require_family) 672 goto skip_family; 673 674 NL_SET_ERR_MSG_ATTR(info->extack, attr, 675 "missing family"); 676 return -EINVAL; 677 } 678 679 entry->addr.family = nla_get_u16(tb[MPTCP_PM_ADDR_ATTR_FAMILY]); 680 if (entry->addr.family != AF_INET 681 #if IS_ENABLED(CONFIG_MPTCP_IPV6) 682 && entry->addr.family != AF_INET6 683 #endif 684 ) { 685 NL_SET_ERR_MSG_ATTR(info->extack, attr, 686 "unknown address family"); 687 return -EINVAL; 688 } 689 addr_addr = mptcp_pm_family_to_addr(entry->addr.family); 690 if (!tb[addr_addr]) { 691 NL_SET_ERR_MSG_ATTR(info->extack, attr, 692 "missing address data"); 693 return -EINVAL; 694 } 695 696 #if IS_ENABLED(CONFIG_MPTCP_IPV6) 697 if (entry->addr.family == AF_INET6) 698 entry->addr.addr6 = nla_get_in6_addr(tb[addr_addr]); 699 else 700 #endif 701 entry->addr.addr.s_addr = nla_get_in_addr(tb[addr_addr]); 702 703 skip_family: 704 if (tb[MPTCP_PM_ADDR_ATTR_IF_IDX]) { 705 u32 val = nla_get_s32(tb[MPTCP_PM_ADDR_ATTR_IF_IDX]); 706 707 entry->addr.ifindex = val; 708 } 709 710 if (tb[MPTCP_PM_ADDR_ATTR_ID]) 711 entry->addr.id = nla_get_u8(tb[MPTCP_PM_ADDR_ATTR_ID]); 712 713 if (tb[MPTCP_PM_ADDR_ATTR_FLAGS]) 714 entry->addr.flags = nla_get_u32(tb[MPTCP_PM_ADDR_ATTR_FLAGS]); 715 716 return 0; 717 } 718 719 static struct pm_nl_pernet *genl_info_pm_nl(struct genl_info *info) 720 { 721 return net_generic(genl_info_net(info), pm_nl_pernet_id); 722 } 723 724 static int mptcp_nl_cmd_add_addr(struct sk_buff *skb, struct genl_info *info) 725 { 726 struct nlattr *attr = info->attrs[MPTCP_PM_ATTR_ADDR]; 727 struct pm_nl_pernet *pernet = genl_info_pm_nl(info); 728 struct mptcp_pm_addr_entry addr, *entry; 729 int ret; 730 731 ret = mptcp_pm_parse_addr(attr, info, true, &addr); 732 if (ret < 0) 733 return ret; 734 735 entry = kmalloc(sizeof(*entry), GFP_KERNEL); 736 if (!entry) { 737 GENL_SET_ERR_MSG(info, "can't allocate addr"); 738 return -ENOMEM; 739 } 740 741 *entry = addr; 742 ret = mptcp_pm_nl_append_new_local_addr(pernet, entry); 743 if (ret < 0) { 744 GENL_SET_ERR_MSG(info, "too many addresses or duplicate one"); 745 kfree(entry); 746 return ret; 747 } 748 749 return 0; 750 } 751 752 static struct mptcp_pm_addr_entry * 753 __lookup_addr_by_id(struct pm_nl_pernet *pernet, unsigned int id) 754 { 755 struct mptcp_pm_addr_entry *entry; 756 757 list_for_each_entry(entry, &pernet->local_addr_list, list) { 758 if (entry->addr.id == id) 759 return entry; 760 } 761 return NULL; 762 } 763 764 static bool remove_anno_list_by_saddr(struct mptcp_sock *msk, 765 struct mptcp_addr_info *addr) 766 { 767 struct mptcp_pm_add_entry *entry; 768 769 entry = mptcp_pm_del_add_timer(msk, addr); 770 if (entry) { 771 list_del(&entry->list); 772 kfree(entry); 773 return true; 774 } 775 776 return false; 777 } 778 779 static bool mptcp_pm_remove_anno_addr(struct mptcp_sock *msk, 780 struct mptcp_addr_info *addr, 781 bool force) 782 { 783 bool ret; 784 785 ret = remove_anno_list_by_saddr(msk, addr); 786 if (ret || force) { 787 spin_lock_bh(&msk->pm.lock); 788 mptcp_pm_remove_addr(msk, addr->id); 789 spin_unlock_bh(&msk->pm.lock); 790 } 791 return ret; 792 } 793 794 static int mptcp_nl_remove_subflow_and_signal_addr(struct net *net, 795 struct mptcp_addr_info *addr) 796 { 797 struct mptcp_sock *msk; 798 long s_slot = 0, s_num = 0; 799 800 pr_debug("remove_id=%d", addr->id); 801 802 while ((msk = mptcp_token_iter_next(net, &s_slot, &s_num)) != NULL) { 803 struct sock *sk = (struct sock *)msk; 804 bool remove_subflow; 805 806 if (list_empty(&msk->conn_list)) { 807 mptcp_pm_remove_anno_addr(msk, addr, false); 808 goto next; 809 } 810 811 lock_sock(sk); 812 remove_subflow = lookup_subflow_by_saddr(&msk->conn_list, addr); 813 mptcp_pm_remove_anno_addr(msk, addr, remove_subflow); 814 if (remove_subflow) 815 mptcp_pm_remove_subflow(msk, addr->id); 816 release_sock(sk); 817 818 next: 819 sock_put(sk); 820 cond_resched(); 821 } 822 823 return 0; 824 } 825 826 static int mptcp_nl_cmd_del_addr(struct sk_buff *skb, struct genl_info *info) 827 { 828 struct nlattr *attr = info->attrs[MPTCP_PM_ATTR_ADDR]; 829 struct pm_nl_pernet *pernet = genl_info_pm_nl(info); 830 struct mptcp_pm_addr_entry addr, *entry; 831 int ret; 832 833 ret = mptcp_pm_parse_addr(attr, info, false, &addr); 834 if (ret < 0) 835 return ret; 836 837 spin_lock_bh(&pernet->lock); 838 entry = __lookup_addr_by_id(pernet, addr.addr.id); 839 if (!entry) { 840 GENL_SET_ERR_MSG(info, "address not found"); 841 spin_unlock_bh(&pernet->lock); 842 return -EINVAL; 843 } 844 if (entry->addr.flags & MPTCP_PM_ADDR_FLAG_SIGNAL) 845 pernet->add_addr_signal_max--; 846 if (entry->addr.flags & MPTCP_PM_ADDR_FLAG_SUBFLOW) 847 pernet->local_addr_max--; 848 849 pernet->addrs--; 850 list_del_rcu(&entry->list); 851 spin_unlock_bh(&pernet->lock); 852 853 mptcp_nl_remove_subflow_and_signal_addr(sock_net(skb->sk), &entry->addr); 854 kfree_rcu(entry, rcu); 855 856 return ret; 857 } 858 859 static void __flush_addrs(struct pm_nl_pernet *pernet) 860 { 861 while (!list_empty(&pernet->local_addr_list)) { 862 struct mptcp_pm_addr_entry *cur; 863 864 cur = list_entry(pernet->local_addr_list.next, 865 struct mptcp_pm_addr_entry, list); 866 list_del_rcu(&cur->list); 867 kfree_rcu(cur, rcu); 868 } 869 } 870 871 static void __reset_counters(struct pm_nl_pernet *pernet) 872 { 873 pernet->add_addr_signal_max = 0; 874 pernet->add_addr_accept_max = 0; 875 pernet->local_addr_max = 0; 876 pernet->addrs = 0; 877 } 878 879 static int mptcp_nl_cmd_flush_addrs(struct sk_buff *skb, struct genl_info *info) 880 { 881 struct pm_nl_pernet *pernet = genl_info_pm_nl(info); 882 883 spin_lock_bh(&pernet->lock); 884 __flush_addrs(pernet); 885 __reset_counters(pernet); 886 spin_unlock_bh(&pernet->lock); 887 return 0; 888 } 889 890 static int mptcp_nl_fill_addr(struct sk_buff *skb, 891 struct mptcp_pm_addr_entry *entry) 892 { 893 struct mptcp_addr_info *addr = &entry->addr; 894 struct nlattr *attr; 895 896 attr = nla_nest_start(skb, MPTCP_PM_ATTR_ADDR); 897 if (!attr) 898 return -EMSGSIZE; 899 900 if (nla_put_u16(skb, MPTCP_PM_ADDR_ATTR_FAMILY, addr->family)) 901 goto nla_put_failure; 902 if (nla_put_u8(skb, MPTCP_PM_ADDR_ATTR_ID, addr->id)) 903 goto nla_put_failure; 904 if (nla_put_u32(skb, MPTCP_PM_ADDR_ATTR_FLAGS, entry->addr.flags)) 905 goto nla_put_failure; 906 if (entry->addr.ifindex && 907 nla_put_s32(skb, MPTCP_PM_ADDR_ATTR_IF_IDX, entry->addr.ifindex)) 908 goto nla_put_failure; 909 910 if (addr->family == AF_INET && 911 nla_put_in_addr(skb, MPTCP_PM_ADDR_ATTR_ADDR4, 912 addr->addr.s_addr)) 913 goto nla_put_failure; 914 #if IS_ENABLED(CONFIG_MPTCP_IPV6) 915 else if (addr->family == AF_INET6 && 916 nla_put_in6_addr(skb, MPTCP_PM_ADDR_ATTR_ADDR6, &addr->addr6)) 917 goto nla_put_failure; 918 #endif 919 nla_nest_end(skb, attr); 920 return 0; 921 922 nla_put_failure: 923 nla_nest_cancel(skb, attr); 924 return -EMSGSIZE; 925 } 926 927 static int mptcp_nl_cmd_get_addr(struct sk_buff *skb, struct genl_info *info) 928 { 929 struct nlattr *attr = info->attrs[MPTCP_PM_ATTR_ADDR]; 930 struct pm_nl_pernet *pernet = genl_info_pm_nl(info); 931 struct mptcp_pm_addr_entry addr, *entry; 932 struct sk_buff *msg; 933 void *reply; 934 int ret; 935 936 ret = mptcp_pm_parse_addr(attr, info, false, &addr); 937 if (ret < 0) 938 return ret; 939 940 msg = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL); 941 if (!msg) 942 return -ENOMEM; 943 944 reply = genlmsg_put_reply(msg, info, &mptcp_genl_family, 0, 945 info->genlhdr->cmd); 946 if (!reply) { 947 GENL_SET_ERR_MSG(info, "not enough space in Netlink message"); 948 ret = -EMSGSIZE; 949 goto fail; 950 } 951 952 spin_lock_bh(&pernet->lock); 953 entry = __lookup_addr_by_id(pernet, addr.addr.id); 954 if (!entry) { 955 GENL_SET_ERR_MSG(info, "address not found"); 956 ret = -EINVAL; 957 goto unlock_fail; 958 } 959 960 ret = mptcp_nl_fill_addr(msg, entry); 961 if (ret) 962 goto unlock_fail; 963 964 genlmsg_end(msg, reply); 965 ret = genlmsg_reply(msg, info); 966 spin_unlock_bh(&pernet->lock); 967 return ret; 968 969 unlock_fail: 970 spin_unlock_bh(&pernet->lock); 971 972 fail: 973 nlmsg_free(msg); 974 return ret; 975 } 976 977 static int mptcp_nl_cmd_dump_addrs(struct sk_buff *msg, 978 struct netlink_callback *cb) 979 { 980 struct net *net = sock_net(msg->sk); 981 struct mptcp_pm_addr_entry *entry; 982 struct pm_nl_pernet *pernet; 983 int id = cb->args[0]; 984 void *hdr; 985 986 pernet = net_generic(net, pm_nl_pernet_id); 987 988 spin_lock_bh(&pernet->lock); 989 list_for_each_entry(entry, &pernet->local_addr_list, list) { 990 if (entry->addr.id <= id) 991 continue; 992 993 hdr = genlmsg_put(msg, NETLINK_CB(cb->skb).portid, 994 cb->nlh->nlmsg_seq, &mptcp_genl_family, 995 NLM_F_MULTI, MPTCP_PM_CMD_GET_ADDR); 996 if (!hdr) 997 break; 998 999 if (mptcp_nl_fill_addr(msg, entry) < 0) { 1000 genlmsg_cancel(msg, hdr); 1001 break; 1002 } 1003 1004 id = entry->addr.id; 1005 genlmsg_end(msg, hdr); 1006 } 1007 spin_unlock_bh(&pernet->lock); 1008 1009 cb->args[0] = id; 1010 return msg->len; 1011 } 1012 1013 static int parse_limit(struct genl_info *info, int id, unsigned int *limit) 1014 { 1015 struct nlattr *attr = info->attrs[id]; 1016 1017 if (!attr) 1018 return 0; 1019 1020 *limit = nla_get_u32(attr); 1021 if (*limit > MPTCP_PM_ADDR_MAX) { 1022 GENL_SET_ERR_MSG(info, "limit greater than maximum"); 1023 return -EINVAL; 1024 } 1025 return 0; 1026 } 1027 1028 static int 1029 mptcp_nl_cmd_set_limits(struct sk_buff *skb, struct genl_info *info) 1030 { 1031 struct pm_nl_pernet *pernet = genl_info_pm_nl(info); 1032 unsigned int rcv_addrs, subflows; 1033 int ret; 1034 1035 spin_lock_bh(&pernet->lock); 1036 rcv_addrs = pernet->add_addr_accept_max; 1037 ret = parse_limit(info, MPTCP_PM_ATTR_RCV_ADD_ADDRS, &rcv_addrs); 1038 if (ret) 1039 goto unlock; 1040 1041 subflows = pernet->subflows_max; 1042 ret = parse_limit(info, MPTCP_PM_ATTR_SUBFLOWS, &subflows); 1043 if (ret) 1044 goto unlock; 1045 1046 WRITE_ONCE(pernet->add_addr_accept_max, rcv_addrs); 1047 WRITE_ONCE(pernet->subflows_max, subflows); 1048 1049 unlock: 1050 spin_unlock_bh(&pernet->lock); 1051 return ret; 1052 } 1053 1054 static int 1055 mptcp_nl_cmd_get_limits(struct sk_buff *skb, struct genl_info *info) 1056 { 1057 struct pm_nl_pernet *pernet = genl_info_pm_nl(info); 1058 struct sk_buff *msg; 1059 void *reply; 1060 1061 msg = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL); 1062 if (!msg) 1063 return -ENOMEM; 1064 1065 reply = genlmsg_put_reply(msg, info, &mptcp_genl_family, 0, 1066 MPTCP_PM_CMD_GET_LIMITS); 1067 if (!reply) 1068 goto fail; 1069 1070 if (nla_put_u32(msg, MPTCP_PM_ATTR_RCV_ADD_ADDRS, 1071 READ_ONCE(pernet->add_addr_accept_max))) 1072 goto fail; 1073 1074 if (nla_put_u32(msg, MPTCP_PM_ATTR_SUBFLOWS, 1075 READ_ONCE(pernet->subflows_max))) 1076 goto fail; 1077 1078 genlmsg_end(msg, reply); 1079 return genlmsg_reply(msg, info); 1080 1081 fail: 1082 GENL_SET_ERR_MSG(info, "not enough space in Netlink message"); 1083 nlmsg_free(msg); 1084 return -EMSGSIZE; 1085 } 1086 1087 static const struct genl_small_ops mptcp_pm_ops[] = { 1088 { 1089 .cmd = MPTCP_PM_CMD_ADD_ADDR, 1090 .doit = mptcp_nl_cmd_add_addr, 1091 .flags = GENL_ADMIN_PERM, 1092 }, 1093 { 1094 .cmd = MPTCP_PM_CMD_DEL_ADDR, 1095 .doit = mptcp_nl_cmd_del_addr, 1096 .flags = GENL_ADMIN_PERM, 1097 }, 1098 { 1099 .cmd = MPTCP_PM_CMD_FLUSH_ADDRS, 1100 .doit = mptcp_nl_cmd_flush_addrs, 1101 .flags = GENL_ADMIN_PERM, 1102 }, 1103 { 1104 .cmd = MPTCP_PM_CMD_GET_ADDR, 1105 .doit = mptcp_nl_cmd_get_addr, 1106 .dumpit = mptcp_nl_cmd_dump_addrs, 1107 }, 1108 { 1109 .cmd = MPTCP_PM_CMD_SET_LIMITS, 1110 .doit = mptcp_nl_cmd_set_limits, 1111 .flags = GENL_ADMIN_PERM, 1112 }, 1113 { 1114 .cmd = MPTCP_PM_CMD_GET_LIMITS, 1115 .doit = mptcp_nl_cmd_get_limits, 1116 }, 1117 }; 1118 1119 static struct genl_family mptcp_genl_family __ro_after_init = { 1120 .name = MPTCP_PM_NAME, 1121 .version = MPTCP_PM_VER, 1122 .maxattr = MPTCP_PM_ATTR_MAX, 1123 .policy = mptcp_pm_policy, 1124 .netnsok = true, 1125 .module = THIS_MODULE, 1126 .small_ops = mptcp_pm_ops, 1127 .n_small_ops = ARRAY_SIZE(mptcp_pm_ops), 1128 .mcgrps = mptcp_pm_mcgrps, 1129 .n_mcgrps = ARRAY_SIZE(mptcp_pm_mcgrps), 1130 }; 1131 1132 static int __net_init pm_nl_init_net(struct net *net) 1133 { 1134 struct pm_nl_pernet *pernet = net_generic(net, pm_nl_pernet_id); 1135 1136 INIT_LIST_HEAD_RCU(&pernet->local_addr_list); 1137 __reset_counters(pernet); 1138 pernet->next_id = 1; 1139 spin_lock_init(&pernet->lock); 1140 return 0; 1141 } 1142 1143 static void __net_exit pm_nl_exit_net(struct list_head *net_list) 1144 { 1145 struct net *net; 1146 1147 list_for_each_entry(net, net_list, exit_list) { 1148 /* net is removed from namespace list, can't race with 1149 * other modifiers 1150 */ 1151 __flush_addrs(net_generic(net, pm_nl_pernet_id)); 1152 } 1153 } 1154 1155 static struct pernet_operations mptcp_pm_pernet_ops = { 1156 .init = pm_nl_init_net, 1157 .exit_batch = pm_nl_exit_net, 1158 .id = &pm_nl_pernet_id, 1159 .size = sizeof(struct pm_nl_pernet), 1160 }; 1161 1162 void __init mptcp_pm_nl_init(void) 1163 { 1164 if (register_pernet_subsys(&mptcp_pm_pernet_ops) < 0) 1165 panic("Failed to register MPTCP PM pernet subsystem.\n"); 1166 1167 if (genl_register_family(&mptcp_genl_family)) 1168 panic("Failed to register MPTCP PM netlink family\n"); 1169 } 1170