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 u8 retrans_times; 29 struct timer_list add_timer; 30 struct mptcp_sock *sock; 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 void mptcp_local_address(const struct sock_common *skc, struct mptcp_addr_info *addr) 90 { 91 addr->family = skc->skc_family; 92 addr->port = htons(skc->skc_num); 93 if (addr->family == AF_INET) 94 addr->addr.s_addr = skc->skc_rcv_saddr; 95 #if IS_ENABLED(CONFIG_MPTCP_IPV6) 96 else if (addr->family == AF_INET6) 97 addr->addr6 = skc->skc_v6_rcv_saddr; 98 #endif 99 } 100 101 static void remote_address(const struct sock_common *skc, 102 struct mptcp_addr_info *addr) 103 { 104 addr->family = skc->skc_family; 105 addr->port = skc->skc_dport; 106 if (addr->family == AF_INET) 107 addr->addr.s_addr = skc->skc_daddr; 108 #if IS_ENABLED(CONFIG_MPTCP_IPV6) 109 else if (addr->family == AF_INET6) 110 addr->addr6 = skc->skc_v6_daddr; 111 #endif 112 } 113 114 static bool lookup_subflow_by_saddr(const struct list_head *list, 115 const struct mptcp_addr_info *saddr) 116 { 117 struct mptcp_subflow_context *subflow; 118 struct mptcp_addr_info cur; 119 struct sock_common *skc; 120 121 list_for_each_entry(subflow, list, node) { 122 skc = (struct sock_common *)mptcp_subflow_tcp_sock(subflow); 123 124 mptcp_local_address(skc, &cur); 125 if (mptcp_addresses_equal(&cur, saddr, saddr->port)) 126 return true; 127 } 128 129 return false; 130 } 131 132 static bool lookup_subflow_by_daddr(const struct list_head *list, 133 const struct mptcp_addr_info *daddr) 134 { 135 struct mptcp_subflow_context *subflow; 136 struct mptcp_addr_info cur; 137 struct sock_common *skc; 138 139 list_for_each_entry(subflow, list, node) { 140 skc = (struct sock_common *)mptcp_subflow_tcp_sock(subflow); 141 142 remote_address(skc, &cur); 143 if (mptcp_addresses_equal(&cur, daddr, daddr->port)) 144 return true; 145 } 146 147 return false; 148 } 149 150 static struct mptcp_pm_addr_entry * 151 select_local_address(const struct pm_nl_pernet *pernet, 152 const struct mptcp_sock *msk) 153 { 154 struct mptcp_pm_addr_entry *entry, *ret = NULL; 155 156 msk_owned_by_me(msk); 157 158 rcu_read_lock(); 159 list_for_each_entry_rcu(entry, &pernet->local_addr_list, list) { 160 if (!(entry->flags & MPTCP_PM_ADDR_FLAG_SUBFLOW)) 161 continue; 162 163 if (!test_bit(entry->addr.id, msk->pm.id_avail_bitmap)) 164 continue; 165 166 ret = entry; 167 break; 168 } 169 rcu_read_unlock(); 170 return ret; 171 } 172 173 static struct mptcp_pm_addr_entry * 174 select_signal_address(struct pm_nl_pernet *pernet, const struct mptcp_sock *msk) 175 { 176 struct mptcp_pm_addr_entry *entry, *ret = NULL; 177 178 rcu_read_lock(); 179 /* do not keep any additional per socket state, just signal 180 * the address list in order. 181 * Note: removal from the local address list during the msk life-cycle 182 * can lead to additional addresses not being announced. 183 */ 184 list_for_each_entry_rcu(entry, &pernet->local_addr_list, list) { 185 if (!test_bit(entry->addr.id, msk->pm.id_avail_bitmap)) 186 continue; 187 188 if (!(entry->flags & MPTCP_PM_ADDR_FLAG_SIGNAL)) 189 continue; 190 191 ret = entry; 192 break; 193 } 194 rcu_read_unlock(); 195 return ret; 196 } 197 198 unsigned int mptcp_pm_get_add_addr_signal_max(const struct mptcp_sock *msk) 199 { 200 const struct pm_nl_pernet *pernet = pm_nl_get_pernet_from_msk(msk); 201 202 return READ_ONCE(pernet->add_addr_signal_max); 203 } 204 EXPORT_SYMBOL_GPL(mptcp_pm_get_add_addr_signal_max); 205 206 unsigned int mptcp_pm_get_add_addr_accept_max(const struct mptcp_sock *msk) 207 { 208 struct pm_nl_pernet *pernet = pm_nl_get_pernet_from_msk(msk); 209 210 return READ_ONCE(pernet->add_addr_accept_max); 211 } 212 EXPORT_SYMBOL_GPL(mptcp_pm_get_add_addr_accept_max); 213 214 unsigned int mptcp_pm_get_subflows_max(const struct mptcp_sock *msk) 215 { 216 struct pm_nl_pernet *pernet = pm_nl_get_pernet_from_msk(msk); 217 218 return READ_ONCE(pernet->subflows_max); 219 } 220 EXPORT_SYMBOL_GPL(mptcp_pm_get_subflows_max); 221 222 unsigned int mptcp_pm_get_local_addr_max(const struct mptcp_sock *msk) 223 { 224 struct pm_nl_pernet *pernet = pm_nl_get_pernet_from_msk(msk); 225 226 return READ_ONCE(pernet->local_addr_max); 227 } 228 EXPORT_SYMBOL_GPL(mptcp_pm_get_local_addr_max); 229 230 bool mptcp_pm_nl_check_work_pending(struct mptcp_sock *msk) 231 { 232 struct pm_nl_pernet *pernet = pm_nl_get_pernet_from_msk(msk); 233 234 if (msk->pm.subflows == mptcp_pm_get_subflows_max(msk) || 235 (find_next_and_bit(pernet->id_bitmap, msk->pm.id_avail_bitmap, 236 MPTCP_PM_MAX_ADDR_ID + 1, 0) == MPTCP_PM_MAX_ADDR_ID + 1)) { 237 WRITE_ONCE(msk->pm.work_pending, false); 238 return false; 239 } 240 return true; 241 } 242 243 struct mptcp_pm_add_entry * 244 mptcp_lookup_anno_list_by_saddr(const struct mptcp_sock *msk, 245 const struct mptcp_addr_info *addr) 246 { 247 struct mptcp_pm_add_entry *entry; 248 249 lockdep_assert_held(&msk->pm.lock); 250 251 list_for_each_entry(entry, &msk->pm.anno_list, list) { 252 if (mptcp_addresses_equal(&entry->addr, addr, true)) 253 return entry; 254 } 255 256 return NULL; 257 } 258 259 bool mptcp_pm_sport_in_anno_list(struct mptcp_sock *msk, const struct sock *sk) 260 { 261 struct mptcp_pm_add_entry *entry; 262 struct mptcp_addr_info saddr; 263 bool ret = false; 264 265 mptcp_local_address((struct sock_common *)sk, &saddr); 266 267 spin_lock_bh(&msk->pm.lock); 268 list_for_each_entry(entry, &msk->pm.anno_list, list) { 269 if (mptcp_addresses_equal(&entry->addr, &saddr, true)) { 270 ret = true; 271 goto out; 272 } 273 } 274 275 out: 276 spin_unlock_bh(&msk->pm.lock); 277 return ret; 278 } 279 280 static void mptcp_pm_add_timer(struct timer_list *timer) 281 { 282 struct mptcp_pm_add_entry *entry = from_timer(entry, timer, add_timer); 283 struct mptcp_sock *msk = entry->sock; 284 struct sock *sk = (struct sock *)msk; 285 286 pr_debug("msk=%p", msk); 287 288 if (!msk) 289 return; 290 291 if (inet_sk_state_load(sk) == TCP_CLOSE) 292 return; 293 294 if (!entry->addr.id) 295 return; 296 297 if (mptcp_pm_should_add_signal_addr(msk)) { 298 sk_reset_timer(sk, timer, jiffies + TCP_RTO_MAX / 8); 299 goto out; 300 } 301 302 spin_lock_bh(&msk->pm.lock); 303 304 if (!mptcp_pm_should_add_signal_addr(msk)) { 305 pr_debug("retransmit ADD_ADDR id=%d", entry->addr.id); 306 mptcp_pm_announce_addr(msk, &entry->addr, false); 307 mptcp_pm_add_addr_send_ack(msk); 308 entry->retrans_times++; 309 } 310 311 if (entry->retrans_times < ADD_ADDR_RETRANS_MAX) 312 sk_reset_timer(sk, timer, 313 jiffies + mptcp_get_add_addr_timeout(sock_net(sk))); 314 315 spin_unlock_bh(&msk->pm.lock); 316 317 if (entry->retrans_times == ADD_ADDR_RETRANS_MAX) 318 mptcp_pm_subflow_established(msk); 319 320 out: 321 __sock_put(sk); 322 } 323 324 struct mptcp_pm_add_entry * 325 mptcp_pm_del_add_timer(struct mptcp_sock *msk, 326 const struct mptcp_addr_info *addr, bool check_id) 327 { 328 struct mptcp_pm_add_entry *entry; 329 struct sock *sk = (struct sock *)msk; 330 331 spin_lock_bh(&msk->pm.lock); 332 entry = mptcp_lookup_anno_list_by_saddr(msk, addr); 333 if (entry && (!check_id || entry->addr.id == addr->id)) 334 entry->retrans_times = ADD_ADDR_RETRANS_MAX; 335 spin_unlock_bh(&msk->pm.lock); 336 337 if (entry && (!check_id || entry->addr.id == addr->id)) 338 sk_stop_timer_sync(sk, &entry->add_timer); 339 340 return entry; 341 } 342 343 bool mptcp_pm_alloc_anno_list(struct mptcp_sock *msk, 344 const struct mptcp_addr_info *addr) 345 { 346 struct mptcp_pm_add_entry *add_entry = NULL; 347 struct sock *sk = (struct sock *)msk; 348 struct net *net = sock_net(sk); 349 350 lockdep_assert_held(&msk->pm.lock); 351 352 add_entry = mptcp_lookup_anno_list_by_saddr(msk, addr); 353 354 if (add_entry) { 355 if (mptcp_pm_is_kernel(msk)) 356 return false; 357 358 sk_reset_timer(sk, &add_entry->add_timer, 359 jiffies + mptcp_get_add_addr_timeout(net)); 360 return true; 361 } 362 363 add_entry = kmalloc(sizeof(*add_entry), GFP_ATOMIC); 364 if (!add_entry) 365 return false; 366 367 list_add(&add_entry->list, &msk->pm.anno_list); 368 369 add_entry->addr = *addr; 370 add_entry->sock = msk; 371 add_entry->retrans_times = 0; 372 373 timer_setup(&add_entry->add_timer, mptcp_pm_add_timer, 0); 374 sk_reset_timer(sk, &add_entry->add_timer, 375 jiffies + mptcp_get_add_addr_timeout(net)); 376 377 return true; 378 } 379 380 void mptcp_pm_free_anno_list(struct mptcp_sock *msk) 381 { 382 struct mptcp_pm_add_entry *entry, *tmp; 383 struct sock *sk = (struct sock *)msk; 384 LIST_HEAD(free_list); 385 386 pr_debug("msk=%p", msk); 387 388 spin_lock_bh(&msk->pm.lock); 389 list_splice_init(&msk->pm.anno_list, &free_list); 390 spin_unlock_bh(&msk->pm.lock); 391 392 list_for_each_entry_safe(entry, tmp, &free_list, list) { 393 sk_stop_timer_sync(sk, &entry->add_timer); 394 kfree(entry); 395 } 396 } 397 398 static bool lookup_address_in_vec(const struct mptcp_addr_info *addrs, unsigned int nr, 399 const struct mptcp_addr_info *addr) 400 { 401 int i; 402 403 for (i = 0; i < nr; i++) { 404 if (addrs[i].id == addr->id) 405 return true; 406 } 407 408 return false; 409 } 410 411 /* Fill all the remote addresses into the array addrs[], 412 * and return the array size. 413 */ 414 static unsigned int fill_remote_addresses_vec(struct mptcp_sock *msk, 415 struct mptcp_addr_info *local, 416 bool fullmesh, 417 struct mptcp_addr_info *addrs) 418 { 419 bool deny_id0 = READ_ONCE(msk->pm.remote_deny_join_id0); 420 struct sock *sk = (struct sock *)msk, *ssk; 421 struct mptcp_subflow_context *subflow; 422 struct mptcp_addr_info remote = { 0 }; 423 unsigned int subflows_max; 424 int i = 0; 425 426 subflows_max = mptcp_pm_get_subflows_max(msk); 427 remote_address((struct sock_common *)sk, &remote); 428 429 /* Non-fullmesh endpoint, fill in the single entry 430 * corresponding to the primary MPC subflow remote address 431 */ 432 if (!fullmesh) { 433 if (deny_id0) 434 return 0; 435 436 if (!mptcp_pm_addr_families_match(sk, local, &remote)) 437 return 0; 438 439 msk->pm.subflows++; 440 addrs[i++] = remote; 441 } else { 442 mptcp_for_each_subflow(msk, subflow) { 443 ssk = mptcp_subflow_tcp_sock(subflow); 444 remote_address((struct sock_common *)ssk, &addrs[i]); 445 addrs[i].id = subflow->remote_id; 446 if (deny_id0 && !addrs[i].id) 447 continue; 448 449 if (!mptcp_pm_addr_families_match(sk, local, &addrs[i])) 450 continue; 451 452 if (!lookup_address_in_vec(addrs, i, &addrs[i]) && 453 msk->pm.subflows < subflows_max) { 454 msk->pm.subflows++; 455 i++; 456 } 457 } 458 } 459 460 return i; 461 } 462 463 static void __mptcp_pm_send_ack(struct mptcp_sock *msk, struct mptcp_subflow_context *subflow, 464 bool prio, bool backup) 465 { 466 struct sock *ssk = mptcp_subflow_tcp_sock(subflow); 467 bool slow; 468 469 pr_debug("send ack for %s", 470 prio ? "mp_prio" : (mptcp_pm_should_add_signal(msk) ? "add_addr" : "rm_addr")); 471 472 slow = lock_sock_fast(ssk); 473 if (prio) { 474 if (subflow->backup != backup) 475 msk->last_snd = NULL; 476 477 subflow->send_mp_prio = 1; 478 subflow->backup = backup; 479 subflow->request_bkup = backup; 480 } 481 482 __mptcp_subflow_send_ack(ssk); 483 unlock_sock_fast(ssk, slow); 484 } 485 486 static void mptcp_pm_send_ack(struct mptcp_sock *msk, struct mptcp_subflow_context *subflow, 487 bool prio, bool backup) 488 { 489 spin_unlock_bh(&msk->pm.lock); 490 __mptcp_pm_send_ack(msk, subflow, prio, backup); 491 spin_lock_bh(&msk->pm.lock); 492 } 493 494 static struct mptcp_pm_addr_entry * 495 __lookup_addr_by_id(struct pm_nl_pernet *pernet, unsigned int id) 496 { 497 struct mptcp_pm_addr_entry *entry; 498 499 list_for_each_entry(entry, &pernet->local_addr_list, list) { 500 if (entry->addr.id == id) 501 return entry; 502 } 503 return NULL; 504 } 505 506 static struct mptcp_pm_addr_entry * 507 __lookup_addr(struct pm_nl_pernet *pernet, const struct mptcp_addr_info *info, 508 bool lookup_by_id) 509 { 510 struct mptcp_pm_addr_entry *entry; 511 512 list_for_each_entry(entry, &pernet->local_addr_list, list) { 513 if ((!lookup_by_id && 514 mptcp_addresses_equal(&entry->addr, info, entry->addr.port)) || 515 (lookup_by_id && entry->addr.id == info->id)) 516 return entry; 517 } 518 return NULL; 519 } 520 521 static void mptcp_pm_create_subflow_or_signal_addr(struct mptcp_sock *msk) 522 { 523 struct sock *sk = (struct sock *)msk; 524 struct mptcp_pm_addr_entry *local; 525 unsigned int add_addr_signal_max; 526 unsigned int local_addr_max; 527 struct pm_nl_pernet *pernet; 528 unsigned int subflows_max; 529 530 pernet = pm_nl_get_pernet(sock_net(sk)); 531 532 add_addr_signal_max = mptcp_pm_get_add_addr_signal_max(msk); 533 local_addr_max = mptcp_pm_get_local_addr_max(msk); 534 subflows_max = mptcp_pm_get_subflows_max(msk); 535 536 /* do lazy endpoint usage accounting for the MPC subflows */ 537 if (unlikely(!(msk->pm.status & BIT(MPTCP_PM_MPC_ENDPOINT_ACCOUNTED))) && msk->first) { 538 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(msk->first); 539 struct mptcp_pm_addr_entry *entry; 540 struct mptcp_addr_info mpc_addr; 541 bool backup = false; 542 543 mptcp_local_address((struct sock_common *)msk->first, &mpc_addr); 544 rcu_read_lock(); 545 entry = __lookup_addr(pernet, &mpc_addr, false); 546 if (entry) { 547 __clear_bit(entry->addr.id, msk->pm.id_avail_bitmap); 548 msk->mpc_endpoint_id = entry->addr.id; 549 backup = !!(entry->flags & MPTCP_PM_ADDR_FLAG_BACKUP); 550 } 551 rcu_read_unlock(); 552 553 if (backup) 554 mptcp_pm_send_ack(msk, subflow, true, backup); 555 556 msk->pm.status |= BIT(MPTCP_PM_MPC_ENDPOINT_ACCOUNTED); 557 } 558 559 pr_debug("local %d:%d signal %d:%d subflows %d:%d\n", 560 msk->pm.local_addr_used, local_addr_max, 561 msk->pm.add_addr_signaled, add_addr_signal_max, 562 msk->pm.subflows, subflows_max); 563 564 /* check first for announce */ 565 if (msk->pm.add_addr_signaled < add_addr_signal_max) { 566 local = select_signal_address(pernet, msk); 567 568 /* due to racing events on both ends we can reach here while 569 * previous add address is still running: if we invoke now 570 * mptcp_pm_announce_addr(), that will fail and the 571 * corresponding id will be marked as used. 572 * Instead let the PM machinery reschedule us when the 573 * current address announce will be completed. 574 */ 575 if (msk->pm.addr_signal & BIT(MPTCP_ADD_ADDR_SIGNAL)) 576 return; 577 578 if (local) { 579 if (mptcp_pm_alloc_anno_list(msk, &local->addr)) { 580 __clear_bit(local->addr.id, msk->pm.id_avail_bitmap); 581 msk->pm.add_addr_signaled++; 582 mptcp_pm_announce_addr(msk, &local->addr, false); 583 mptcp_pm_nl_addr_send_ack(msk); 584 } 585 } 586 } 587 588 /* check if should create a new subflow */ 589 while (msk->pm.local_addr_used < local_addr_max && 590 msk->pm.subflows < subflows_max) { 591 struct mptcp_addr_info addrs[MPTCP_PM_ADDR_MAX]; 592 bool fullmesh; 593 int i, nr; 594 595 local = select_local_address(pernet, msk); 596 if (!local) 597 break; 598 599 fullmesh = !!(local->flags & MPTCP_PM_ADDR_FLAG_FULLMESH); 600 601 msk->pm.local_addr_used++; 602 __clear_bit(local->addr.id, msk->pm.id_avail_bitmap); 603 nr = fill_remote_addresses_vec(msk, &local->addr, fullmesh, addrs); 604 if (nr == 0) 605 continue; 606 607 spin_unlock_bh(&msk->pm.lock); 608 for (i = 0; i < nr; i++) 609 __mptcp_subflow_connect(sk, &local->addr, &addrs[i]); 610 spin_lock_bh(&msk->pm.lock); 611 } 612 mptcp_pm_nl_check_work_pending(msk); 613 } 614 615 static void mptcp_pm_nl_fully_established(struct mptcp_sock *msk) 616 { 617 mptcp_pm_create_subflow_or_signal_addr(msk); 618 } 619 620 static void mptcp_pm_nl_subflow_established(struct mptcp_sock *msk) 621 { 622 mptcp_pm_create_subflow_or_signal_addr(msk); 623 } 624 625 /* Fill all the local addresses into the array addrs[], 626 * and return the array size. 627 */ 628 static unsigned int fill_local_addresses_vec(struct mptcp_sock *msk, 629 struct mptcp_addr_info *remote, 630 struct mptcp_addr_info *addrs) 631 { 632 struct sock *sk = (struct sock *)msk; 633 struct mptcp_pm_addr_entry *entry; 634 struct pm_nl_pernet *pernet; 635 unsigned int subflows_max; 636 int i = 0; 637 638 pernet = pm_nl_get_pernet_from_msk(msk); 639 subflows_max = mptcp_pm_get_subflows_max(msk); 640 641 rcu_read_lock(); 642 list_for_each_entry_rcu(entry, &pernet->local_addr_list, list) { 643 if (!(entry->flags & MPTCP_PM_ADDR_FLAG_FULLMESH)) 644 continue; 645 646 if (!mptcp_pm_addr_families_match(sk, &entry->addr, remote)) 647 continue; 648 649 if (msk->pm.subflows < subflows_max) { 650 msk->pm.subflows++; 651 addrs[i++] = entry->addr; 652 } 653 } 654 rcu_read_unlock(); 655 656 /* If the array is empty, fill in the single 657 * 'IPADDRANY' local address 658 */ 659 if (!i) { 660 struct mptcp_addr_info local; 661 662 memset(&local, 0, sizeof(local)); 663 local.family = 664 #if IS_ENABLED(CONFIG_MPTCP_IPV6) 665 remote->family == AF_INET6 && 666 ipv6_addr_v4mapped(&remote->addr6) ? AF_INET : 667 #endif 668 remote->family; 669 670 if (!mptcp_pm_addr_families_match(sk, &local, remote)) 671 return 0; 672 673 msk->pm.subflows++; 674 addrs[i++] = local; 675 } 676 677 return i; 678 } 679 680 static void mptcp_pm_nl_add_addr_received(struct mptcp_sock *msk) 681 { 682 struct mptcp_addr_info addrs[MPTCP_PM_ADDR_MAX]; 683 struct sock *sk = (struct sock *)msk; 684 unsigned int add_addr_accept_max; 685 struct mptcp_addr_info remote; 686 unsigned int subflows_max; 687 int i, nr; 688 689 add_addr_accept_max = mptcp_pm_get_add_addr_accept_max(msk); 690 subflows_max = mptcp_pm_get_subflows_max(msk); 691 692 pr_debug("accepted %d:%d remote family %d", 693 msk->pm.add_addr_accepted, add_addr_accept_max, 694 msk->pm.remote.family); 695 696 remote = msk->pm.remote; 697 mptcp_pm_announce_addr(msk, &remote, true); 698 mptcp_pm_nl_addr_send_ack(msk); 699 700 if (lookup_subflow_by_daddr(&msk->conn_list, &remote)) 701 return; 702 703 /* pick id 0 port, if none is provided the remote address */ 704 if (!remote.port) 705 remote.port = sk->sk_dport; 706 707 /* connect to the specified remote address, using whatever 708 * local address the routing configuration will pick. 709 */ 710 nr = fill_local_addresses_vec(msk, &remote, addrs); 711 if (nr == 0) 712 return; 713 714 msk->pm.add_addr_accepted++; 715 if (msk->pm.add_addr_accepted >= add_addr_accept_max || 716 msk->pm.subflows >= subflows_max) 717 WRITE_ONCE(msk->pm.accept_addr, false); 718 719 spin_unlock_bh(&msk->pm.lock); 720 for (i = 0; i < nr; i++) 721 __mptcp_subflow_connect(sk, &addrs[i], &remote); 722 spin_lock_bh(&msk->pm.lock); 723 } 724 725 void mptcp_pm_nl_addr_send_ack(struct mptcp_sock *msk) 726 { 727 struct mptcp_subflow_context *subflow; 728 729 msk_owned_by_me(msk); 730 lockdep_assert_held(&msk->pm.lock); 731 732 if (!mptcp_pm_should_add_signal(msk) && 733 !mptcp_pm_should_rm_signal(msk)) 734 return; 735 736 subflow = list_first_entry_or_null(&msk->conn_list, typeof(*subflow), node); 737 if (subflow) 738 mptcp_pm_send_ack(msk, subflow, false, false); 739 } 740 741 int mptcp_pm_nl_mp_prio_send_ack(struct mptcp_sock *msk, 742 struct mptcp_addr_info *addr, 743 struct mptcp_addr_info *rem, 744 u8 bkup) 745 { 746 struct mptcp_subflow_context *subflow; 747 748 pr_debug("bkup=%d", bkup); 749 750 mptcp_for_each_subflow(msk, subflow) { 751 struct sock *ssk = mptcp_subflow_tcp_sock(subflow); 752 struct mptcp_addr_info local, remote; 753 754 mptcp_local_address((struct sock_common *)ssk, &local); 755 if (!mptcp_addresses_equal(&local, addr, addr->port)) 756 continue; 757 758 if (rem && rem->family != AF_UNSPEC) { 759 remote_address((struct sock_common *)ssk, &remote); 760 if (!mptcp_addresses_equal(&remote, rem, rem->port)) 761 continue; 762 } 763 764 __mptcp_pm_send_ack(msk, subflow, true, bkup); 765 return 0; 766 } 767 768 return -EINVAL; 769 } 770 771 static bool mptcp_local_id_match(const struct mptcp_sock *msk, u8 local_id, u8 id) 772 { 773 return local_id == id || (!local_id && msk->mpc_endpoint_id == id); 774 } 775 776 static void mptcp_pm_nl_rm_addr_or_subflow(struct mptcp_sock *msk, 777 const struct mptcp_rm_list *rm_list, 778 enum linux_mptcp_mib_field rm_type) 779 { 780 struct mptcp_subflow_context *subflow, *tmp; 781 struct sock *sk = (struct sock *)msk; 782 u8 i; 783 784 pr_debug("%s rm_list_nr %d", 785 rm_type == MPTCP_MIB_RMADDR ? "address" : "subflow", rm_list->nr); 786 787 msk_owned_by_me(msk); 788 789 if (sk->sk_state == TCP_LISTEN) 790 return; 791 792 if (!rm_list->nr) 793 return; 794 795 if (list_empty(&msk->conn_list)) 796 return; 797 798 for (i = 0; i < rm_list->nr; i++) { 799 u8 rm_id = rm_list->ids[i]; 800 bool removed = false; 801 802 mptcp_for_each_subflow_safe(msk, subflow, tmp) { 803 struct sock *ssk = mptcp_subflow_tcp_sock(subflow); 804 int how = RCV_SHUTDOWN | SEND_SHUTDOWN; 805 u8 id = subflow->local_id; 806 807 if (rm_type == MPTCP_MIB_RMADDR && subflow->remote_id != rm_id) 808 continue; 809 if (rm_type == MPTCP_MIB_RMSUBFLOW && !mptcp_local_id_match(msk, id, rm_id)) 810 continue; 811 812 pr_debug(" -> %s rm_list_ids[%d]=%u local_id=%u remote_id=%u mpc_id=%u", 813 rm_type == MPTCP_MIB_RMADDR ? "address" : "subflow", 814 i, rm_id, subflow->local_id, subflow->remote_id, 815 msk->mpc_endpoint_id); 816 spin_unlock_bh(&msk->pm.lock); 817 mptcp_subflow_shutdown(sk, ssk, how); 818 819 /* the following takes care of updating the subflows counter */ 820 mptcp_close_ssk(sk, ssk, subflow); 821 spin_lock_bh(&msk->pm.lock); 822 823 removed = true; 824 __MPTCP_INC_STATS(sock_net(sk), rm_type); 825 } 826 if (rm_type == MPTCP_MIB_RMSUBFLOW) 827 __set_bit(rm_id ? rm_id : msk->mpc_endpoint_id, msk->pm.id_avail_bitmap); 828 if (!removed) 829 continue; 830 831 if (!mptcp_pm_is_kernel(msk)) 832 continue; 833 834 if (rm_type == MPTCP_MIB_RMADDR) { 835 msk->pm.add_addr_accepted--; 836 WRITE_ONCE(msk->pm.accept_addr, true); 837 } else if (rm_type == MPTCP_MIB_RMSUBFLOW) { 838 msk->pm.local_addr_used--; 839 } 840 } 841 } 842 843 static void mptcp_pm_nl_rm_addr_received(struct mptcp_sock *msk) 844 { 845 mptcp_pm_nl_rm_addr_or_subflow(msk, &msk->pm.rm_list_rx, MPTCP_MIB_RMADDR); 846 } 847 848 void mptcp_pm_nl_rm_subflow_received(struct mptcp_sock *msk, 849 const struct mptcp_rm_list *rm_list) 850 { 851 mptcp_pm_nl_rm_addr_or_subflow(msk, rm_list, MPTCP_MIB_RMSUBFLOW); 852 } 853 854 void mptcp_pm_nl_work(struct mptcp_sock *msk) 855 { 856 struct mptcp_pm_data *pm = &msk->pm; 857 858 msk_owned_by_me(msk); 859 860 if (!(pm->status & MPTCP_PM_WORK_MASK)) 861 return; 862 863 spin_lock_bh(&msk->pm.lock); 864 865 pr_debug("msk=%p status=%x", msk, pm->status); 866 if (pm->status & BIT(MPTCP_PM_ADD_ADDR_RECEIVED)) { 867 pm->status &= ~BIT(MPTCP_PM_ADD_ADDR_RECEIVED); 868 mptcp_pm_nl_add_addr_received(msk); 869 } 870 if (pm->status & BIT(MPTCP_PM_ADD_ADDR_SEND_ACK)) { 871 pm->status &= ~BIT(MPTCP_PM_ADD_ADDR_SEND_ACK); 872 mptcp_pm_nl_addr_send_ack(msk); 873 } 874 if (pm->status & BIT(MPTCP_PM_RM_ADDR_RECEIVED)) { 875 pm->status &= ~BIT(MPTCP_PM_RM_ADDR_RECEIVED); 876 mptcp_pm_nl_rm_addr_received(msk); 877 } 878 if (pm->status & BIT(MPTCP_PM_ESTABLISHED)) { 879 pm->status &= ~BIT(MPTCP_PM_ESTABLISHED); 880 mptcp_pm_nl_fully_established(msk); 881 } 882 if (pm->status & BIT(MPTCP_PM_SUBFLOW_ESTABLISHED)) { 883 pm->status &= ~BIT(MPTCP_PM_SUBFLOW_ESTABLISHED); 884 mptcp_pm_nl_subflow_established(msk); 885 } 886 887 spin_unlock_bh(&msk->pm.lock); 888 } 889 890 static bool address_use_port(struct mptcp_pm_addr_entry *entry) 891 { 892 return (entry->flags & 893 (MPTCP_PM_ADDR_FLAG_SIGNAL | MPTCP_PM_ADDR_FLAG_SUBFLOW)) == 894 MPTCP_PM_ADDR_FLAG_SIGNAL; 895 } 896 897 /* caller must ensure the RCU grace period is already elapsed */ 898 static void __mptcp_pm_release_addr_entry(struct mptcp_pm_addr_entry *entry) 899 { 900 if (entry->lsk) 901 sock_release(entry->lsk); 902 kfree(entry); 903 } 904 905 static int mptcp_pm_nl_append_new_local_addr(struct pm_nl_pernet *pernet, 906 struct mptcp_pm_addr_entry *entry) 907 { 908 struct mptcp_pm_addr_entry *cur, *del_entry = NULL; 909 unsigned int addr_max; 910 int ret = -EINVAL; 911 912 spin_lock_bh(&pernet->lock); 913 /* to keep the code simple, don't do IDR-like allocation for address ID, 914 * just bail when we exceed limits 915 */ 916 if (pernet->next_id == MPTCP_PM_MAX_ADDR_ID) 917 pernet->next_id = 1; 918 if (pernet->addrs >= MPTCP_PM_ADDR_MAX) { 919 ret = -ERANGE; 920 goto out; 921 } 922 if (test_bit(entry->addr.id, pernet->id_bitmap)) { 923 ret = -EBUSY; 924 goto out; 925 } 926 927 /* do not insert duplicate address, differentiate on port only 928 * singled addresses 929 */ 930 if (!address_use_port(entry)) 931 entry->addr.port = 0; 932 list_for_each_entry(cur, &pernet->local_addr_list, list) { 933 if (mptcp_addresses_equal(&cur->addr, &entry->addr, 934 cur->addr.port || entry->addr.port)) { 935 /* allow replacing the exiting endpoint only if such 936 * endpoint is an implicit one and the user-space 937 * did not provide an endpoint id 938 */ 939 if (!(cur->flags & MPTCP_PM_ADDR_FLAG_IMPLICIT)) { 940 ret = -EEXIST; 941 goto out; 942 } 943 if (entry->addr.id) 944 goto out; 945 946 pernet->addrs--; 947 entry->addr.id = cur->addr.id; 948 list_del_rcu(&cur->list); 949 del_entry = cur; 950 break; 951 } 952 } 953 954 if (!entry->addr.id) { 955 find_next: 956 entry->addr.id = find_next_zero_bit(pernet->id_bitmap, 957 MPTCP_PM_MAX_ADDR_ID + 1, 958 pernet->next_id); 959 if (!entry->addr.id && pernet->next_id != 1) { 960 pernet->next_id = 1; 961 goto find_next; 962 } 963 } 964 965 if (!entry->addr.id) 966 goto out; 967 968 __set_bit(entry->addr.id, pernet->id_bitmap); 969 if (entry->addr.id > pernet->next_id) 970 pernet->next_id = entry->addr.id; 971 972 if (entry->flags & MPTCP_PM_ADDR_FLAG_SIGNAL) { 973 addr_max = pernet->add_addr_signal_max; 974 WRITE_ONCE(pernet->add_addr_signal_max, addr_max + 1); 975 } 976 if (entry->flags & MPTCP_PM_ADDR_FLAG_SUBFLOW) { 977 addr_max = pernet->local_addr_max; 978 WRITE_ONCE(pernet->local_addr_max, addr_max + 1); 979 } 980 981 pernet->addrs++; 982 if (!entry->addr.port) 983 list_add_tail_rcu(&entry->list, &pernet->local_addr_list); 984 else 985 list_add_rcu(&entry->list, &pernet->local_addr_list); 986 ret = entry->addr.id; 987 988 out: 989 spin_unlock_bh(&pernet->lock); 990 991 /* just replaced an existing entry, free it */ 992 if (del_entry) { 993 synchronize_rcu(); 994 __mptcp_pm_release_addr_entry(del_entry); 995 } 996 return ret; 997 } 998 999 static struct lock_class_key mptcp_slock_keys[2]; 1000 static struct lock_class_key mptcp_keys[2]; 1001 1002 static int mptcp_pm_nl_create_listen_socket(struct sock *sk, 1003 struct mptcp_pm_addr_entry *entry) 1004 { 1005 bool is_ipv6 = sk->sk_family == AF_INET6; 1006 int addrlen = sizeof(struct sockaddr_in); 1007 struct sockaddr_storage addr; 1008 struct socket *ssock; 1009 struct sock *newsk; 1010 int backlog = 1024; 1011 int err; 1012 1013 err = sock_create_kern(sock_net(sk), entry->addr.family, 1014 SOCK_STREAM, IPPROTO_MPTCP, &entry->lsk); 1015 if (err) 1016 return err; 1017 1018 newsk = entry->lsk->sk; 1019 if (!newsk) 1020 return -EINVAL; 1021 1022 /* The subflow socket lock is acquired in a nested to the msk one 1023 * in several places, even by the TCP stack, and this msk is a kernel 1024 * socket: lockdep complains. Instead of propagating the _nested 1025 * modifiers in several places, re-init the lock class for the msk 1026 * socket to an mptcp specific one. 1027 */ 1028 sock_lock_init_class_and_name(newsk, 1029 is_ipv6 ? "mlock-AF_INET6" : "mlock-AF_INET", 1030 &mptcp_slock_keys[is_ipv6], 1031 is_ipv6 ? "msk_lock-AF_INET6" : "msk_lock-AF_INET", 1032 &mptcp_keys[is_ipv6]); 1033 1034 lock_sock(newsk); 1035 ssock = __mptcp_nmpc_socket(mptcp_sk(newsk)); 1036 release_sock(newsk); 1037 if (IS_ERR(ssock)) 1038 return PTR_ERR(ssock); 1039 1040 mptcp_info2sockaddr(&entry->addr, &addr, entry->addr.family); 1041 #if IS_ENABLED(CONFIG_MPTCP_IPV6) 1042 if (entry->addr.family == AF_INET6) 1043 addrlen = sizeof(struct sockaddr_in6); 1044 #endif 1045 err = kernel_bind(ssock, (struct sockaddr *)&addr, addrlen); 1046 if (err) 1047 return err; 1048 1049 inet_sk_state_store(newsk, TCP_LISTEN); 1050 err = kernel_listen(ssock, backlog); 1051 if (err) 1052 return err; 1053 1054 mptcp_event_pm_listener(ssock->sk, MPTCP_EVENT_LISTENER_CREATED); 1055 1056 return 0; 1057 } 1058 1059 int mptcp_pm_nl_get_local_id(struct mptcp_sock *msk, struct mptcp_addr_info *skc) 1060 { 1061 struct mptcp_pm_addr_entry *entry; 1062 struct pm_nl_pernet *pernet; 1063 int ret = -1; 1064 1065 pernet = pm_nl_get_pernet_from_msk(msk); 1066 1067 rcu_read_lock(); 1068 list_for_each_entry_rcu(entry, &pernet->local_addr_list, list) { 1069 if (mptcp_addresses_equal(&entry->addr, skc, entry->addr.port)) { 1070 ret = entry->addr.id; 1071 break; 1072 } 1073 } 1074 rcu_read_unlock(); 1075 if (ret >= 0) 1076 return ret; 1077 1078 /* address not found, add to local list */ 1079 entry = kmalloc(sizeof(*entry), GFP_ATOMIC); 1080 if (!entry) 1081 return -ENOMEM; 1082 1083 entry->addr = *skc; 1084 entry->addr.id = 0; 1085 entry->addr.port = 0; 1086 entry->ifindex = 0; 1087 entry->flags = MPTCP_PM_ADDR_FLAG_IMPLICIT; 1088 entry->lsk = NULL; 1089 ret = mptcp_pm_nl_append_new_local_addr(pernet, entry); 1090 if (ret < 0) 1091 kfree(entry); 1092 1093 return ret; 1094 } 1095 1096 #define MPTCP_PM_CMD_GRP_OFFSET 0 1097 #define MPTCP_PM_EV_GRP_OFFSET 1 1098 1099 static const struct genl_multicast_group mptcp_pm_mcgrps[] = { 1100 [MPTCP_PM_CMD_GRP_OFFSET] = { .name = MPTCP_PM_CMD_GRP_NAME, }, 1101 [MPTCP_PM_EV_GRP_OFFSET] = { .name = MPTCP_PM_EV_GRP_NAME, 1102 .flags = GENL_UNS_ADMIN_PERM, 1103 }, 1104 }; 1105 1106 static const struct nla_policy 1107 mptcp_pm_addr_policy[MPTCP_PM_ADDR_ATTR_MAX + 1] = { 1108 [MPTCP_PM_ADDR_ATTR_FAMILY] = { .type = NLA_U16, }, 1109 [MPTCP_PM_ADDR_ATTR_ID] = { .type = NLA_U8, }, 1110 [MPTCP_PM_ADDR_ATTR_ADDR4] = { .type = NLA_U32, }, 1111 [MPTCP_PM_ADDR_ATTR_ADDR6] = 1112 NLA_POLICY_EXACT_LEN(sizeof(struct in6_addr)), 1113 [MPTCP_PM_ADDR_ATTR_PORT] = { .type = NLA_U16 }, 1114 [MPTCP_PM_ADDR_ATTR_FLAGS] = { .type = NLA_U32 }, 1115 [MPTCP_PM_ADDR_ATTR_IF_IDX] = { .type = NLA_S32 }, 1116 }; 1117 1118 static const struct nla_policy mptcp_pm_policy[MPTCP_PM_ATTR_MAX + 1] = { 1119 [MPTCP_PM_ATTR_ADDR] = 1120 NLA_POLICY_NESTED(mptcp_pm_addr_policy), 1121 [MPTCP_PM_ATTR_RCV_ADD_ADDRS] = { .type = NLA_U32, }, 1122 [MPTCP_PM_ATTR_SUBFLOWS] = { .type = NLA_U32, }, 1123 [MPTCP_PM_ATTR_TOKEN] = { .type = NLA_U32, }, 1124 [MPTCP_PM_ATTR_LOC_ID] = { .type = NLA_U8, }, 1125 [MPTCP_PM_ATTR_ADDR_REMOTE] = 1126 NLA_POLICY_NESTED(mptcp_pm_addr_policy), 1127 }; 1128 1129 void mptcp_pm_nl_subflow_chk_stale(const struct mptcp_sock *msk, struct sock *ssk) 1130 { 1131 struct mptcp_subflow_context *iter, *subflow = mptcp_subflow_ctx(ssk); 1132 struct sock *sk = (struct sock *)msk; 1133 unsigned int active_max_loss_cnt; 1134 struct net *net = sock_net(sk); 1135 unsigned int stale_loss_cnt; 1136 bool slow; 1137 1138 stale_loss_cnt = mptcp_stale_loss_cnt(net); 1139 if (subflow->stale || !stale_loss_cnt || subflow->stale_count <= stale_loss_cnt) 1140 return; 1141 1142 /* look for another available subflow not in loss state */ 1143 active_max_loss_cnt = max_t(int, stale_loss_cnt - 1, 1); 1144 mptcp_for_each_subflow(msk, iter) { 1145 if (iter != subflow && mptcp_subflow_active(iter) && 1146 iter->stale_count < active_max_loss_cnt) { 1147 /* we have some alternatives, try to mark this subflow as idle ...*/ 1148 slow = lock_sock_fast(ssk); 1149 if (!tcp_rtx_and_write_queues_empty(ssk)) { 1150 subflow->stale = 1; 1151 __mptcp_retransmit_pending_data(sk); 1152 MPTCP_INC_STATS(net, MPTCP_MIB_SUBFLOWSTALE); 1153 } 1154 unlock_sock_fast(ssk, slow); 1155 1156 /* always try to push the pending data regardless of re-injections: 1157 * we can possibly use backup subflows now, and subflow selection 1158 * is cheap under the msk socket lock 1159 */ 1160 __mptcp_push_pending(sk, 0); 1161 return; 1162 } 1163 } 1164 } 1165 1166 static int mptcp_pm_family_to_addr(int family) 1167 { 1168 #if IS_ENABLED(CONFIG_MPTCP_IPV6) 1169 if (family == AF_INET6) 1170 return MPTCP_PM_ADDR_ATTR_ADDR6; 1171 #endif 1172 return MPTCP_PM_ADDR_ATTR_ADDR4; 1173 } 1174 1175 static int mptcp_pm_parse_pm_addr_attr(struct nlattr *tb[], 1176 const struct nlattr *attr, 1177 struct genl_info *info, 1178 struct mptcp_addr_info *addr, 1179 bool require_family) 1180 { 1181 int err, addr_addr; 1182 1183 if (!attr) { 1184 GENL_SET_ERR_MSG(info, "missing address info"); 1185 return -EINVAL; 1186 } 1187 1188 /* no validation needed - was already done via nested policy */ 1189 err = nla_parse_nested_deprecated(tb, MPTCP_PM_ADDR_ATTR_MAX, attr, 1190 mptcp_pm_addr_policy, info->extack); 1191 if (err) 1192 return err; 1193 1194 if (tb[MPTCP_PM_ADDR_ATTR_ID]) 1195 addr->id = nla_get_u8(tb[MPTCP_PM_ADDR_ATTR_ID]); 1196 1197 if (!tb[MPTCP_PM_ADDR_ATTR_FAMILY]) { 1198 if (!require_family) 1199 return 0; 1200 1201 NL_SET_ERR_MSG_ATTR(info->extack, attr, 1202 "missing family"); 1203 return -EINVAL; 1204 } 1205 1206 addr->family = nla_get_u16(tb[MPTCP_PM_ADDR_ATTR_FAMILY]); 1207 if (addr->family != AF_INET 1208 #if IS_ENABLED(CONFIG_MPTCP_IPV6) 1209 && addr->family != AF_INET6 1210 #endif 1211 ) { 1212 NL_SET_ERR_MSG_ATTR(info->extack, attr, 1213 "unknown address family"); 1214 return -EINVAL; 1215 } 1216 addr_addr = mptcp_pm_family_to_addr(addr->family); 1217 if (!tb[addr_addr]) { 1218 NL_SET_ERR_MSG_ATTR(info->extack, attr, 1219 "missing address data"); 1220 return -EINVAL; 1221 } 1222 1223 #if IS_ENABLED(CONFIG_MPTCP_IPV6) 1224 if (addr->family == AF_INET6) 1225 addr->addr6 = nla_get_in6_addr(tb[addr_addr]); 1226 else 1227 #endif 1228 addr->addr.s_addr = nla_get_in_addr(tb[addr_addr]); 1229 1230 if (tb[MPTCP_PM_ADDR_ATTR_PORT]) 1231 addr->port = htons(nla_get_u16(tb[MPTCP_PM_ADDR_ATTR_PORT])); 1232 1233 return 0; 1234 } 1235 1236 int mptcp_pm_parse_addr(struct nlattr *attr, struct genl_info *info, 1237 struct mptcp_addr_info *addr) 1238 { 1239 struct nlattr *tb[MPTCP_PM_ADDR_ATTR_MAX + 1]; 1240 1241 memset(addr, 0, sizeof(*addr)); 1242 1243 return mptcp_pm_parse_pm_addr_attr(tb, attr, info, addr, true); 1244 } 1245 1246 int mptcp_pm_parse_entry(struct nlattr *attr, struct genl_info *info, 1247 bool require_family, 1248 struct mptcp_pm_addr_entry *entry) 1249 { 1250 struct nlattr *tb[MPTCP_PM_ADDR_ATTR_MAX + 1]; 1251 int err; 1252 1253 memset(entry, 0, sizeof(*entry)); 1254 1255 err = mptcp_pm_parse_pm_addr_attr(tb, attr, info, &entry->addr, require_family); 1256 if (err) 1257 return err; 1258 1259 if (tb[MPTCP_PM_ADDR_ATTR_IF_IDX]) { 1260 u32 val = nla_get_s32(tb[MPTCP_PM_ADDR_ATTR_IF_IDX]); 1261 1262 entry->ifindex = val; 1263 } 1264 1265 if (tb[MPTCP_PM_ADDR_ATTR_FLAGS]) 1266 entry->flags = nla_get_u32(tb[MPTCP_PM_ADDR_ATTR_FLAGS]); 1267 1268 if (tb[MPTCP_PM_ADDR_ATTR_PORT]) 1269 entry->addr.port = htons(nla_get_u16(tb[MPTCP_PM_ADDR_ATTR_PORT])); 1270 1271 return 0; 1272 } 1273 1274 static struct pm_nl_pernet *genl_info_pm_nl(struct genl_info *info) 1275 { 1276 return pm_nl_get_pernet(genl_info_net(info)); 1277 } 1278 1279 static int mptcp_nl_add_subflow_or_signal_addr(struct net *net) 1280 { 1281 struct mptcp_sock *msk; 1282 long s_slot = 0, s_num = 0; 1283 1284 while ((msk = mptcp_token_iter_next(net, &s_slot, &s_num)) != NULL) { 1285 struct sock *sk = (struct sock *)msk; 1286 1287 if (!READ_ONCE(msk->fully_established) || 1288 mptcp_pm_is_userspace(msk)) 1289 goto next; 1290 1291 lock_sock(sk); 1292 spin_lock_bh(&msk->pm.lock); 1293 mptcp_pm_create_subflow_or_signal_addr(msk); 1294 spin_unlock_bh(&msk->pm.lock); 1295 release_sock(sk); 1296 1297 next: 1298 sock_put(sk); 1299 cond_resched(); 1300 } 1301 1302 return 0; 1303 } 1304 1305 static int mptcp_nl_cmd_add_addr(struct sk_buff *skb, struct genl_info *info) 1306 { 1307 struct nlattr *attr = info->attrs[MPTCP_PM_ATTR_ADDR]; 1308 struct pm_nl_pernet *pernet = genl_info_pm_nl(info); 1309 struct mptcp_pm_addr_entry addr, *entry; 1310 int ret; 1311 1312 ret = mptcp_pm_parse_entry(attr, info, true, &addr); 1313 if (ret < 0) 1314 return ret; 1315 1316 if (addr.addr.port && !(addr.flags & MPTCP_PM_ADDR_FLAG_SIGNAL)) { 1317 GENL_SET_ERR_MSG(info, "flags must have signal when using port"); 1318 return -EINVAL; 1319 } 1320 1321 if (addr.flags & MPTCP_PM_ADDR_FLAG_SIGNAL && 1322 addr.flags & MPTCP_PM_ADDR_FLAG_FULLMESH) { 1323 GENL_SET_ERR_MSG(info, "flags mustn't have both signal and fullmesh"); 1324 return -EINVAL; 1325 } 1326 1327 if (addr.flags & MPTCP_PM_ADDR_FLAG_IMPLICIT) { 1328 GENL_SET_ERR_MSG(info, "can't create IMPLICIT endpoint"); 1329 return -EINVAL; 1330 } 1331 1332 entry = kzalloc(sizeof(*entry), GFP_KERNEL_ACCOUNT); 1333 if (!entry) { 1334 GENL_SET_ERR_MSG(info, "can't allocate addr"); 1335 return -ENOMEM; 1336 } 1337 1338 *entry = addr; 1339 if (entry->addr.port) { 1340 ret = mptcp_pm_nl_create_listen_socket(skb->sk, entry); 1341 if (ret) { 1342 GENL_SET_ERR_MSG_FMT(info, "create listen socket error: %d", ret); 1343 goto out_free; 1344 } 1345 } 1346 ret = mptcp_pm_nl_append_new_local_addr(pernet, entry); 1347 if (ret < 0) { 1348 GENL_SET_ERR_MSG_FMT(info, "too many addresses or duplicate one: %d", ret); 1349 goto out_free; 1350 } 1351 1352 mptcp_nl_add_subflow_or_signal_addr(sock_net(skb->sk)); 1353 return 0; 1354 1355 out_free: 1356 __mptcp_pm_release_addr_entry(entry); 1357 return ret; 1358 } 1359 1360 int mptcp_pm_nl_get_flags_and_ifindex_by_id(struct mptcp_sock *msk, unsigned int id, 1361 u8 *flags, int *ifindex) 1362 { 1363 struct mptcp_pm_addr_entry *entry; 1364 struct sock *sk = (struct sock *)msk; 1365 struct net *net = sock_net(sk); 1366 1367 rcu_read_lock(); 1368 entry = __lookup_addr_by_id(pm_nl_get_pernet(net), id); 1369 if (entry) { 1370 *flags = entry->flags; 1371 *ifindex = entry->ifindex; 1372 } 1373 rcu_read_unlock(); 1374 1375 return 0; 1376 } 1377 1378 static bool remove_anno_list_by_saddr(struct mptcp_sock *msk, 1379 const struct mptcp_addr_info *addr) 1380 { 1381 struct mptcp_pm_add_entry *entry; 1382 1383 entry = mptcp_pm_del_add_timer(msk, addr, false); 1384 if (entry) { 1385 list_del(&entry->list); 1386 kfree(entry); 1387 return true; 1388 } 1389 1390 return false; 1391 } 1392 1393 static bool mptcp_pm_remove_anno_addr(struct mptcp_sock *msk, 1394 const struct mptcp_addr_info *addr, 1395 bool force) 1396 { 1397 struct mptcp_rm_list list = { .nr = 0 }; 1398 bool ret; 1399 1400 list.ids[list.nr++] = addr->id; 1401 1402 ret = remove_anno_list_by_saddr(msk, addr); 1403 if (ret || force) { 1404 spin_lock_bh(&msk->pm.lock); 1405 mptcp_pm_remove_addr(msk, &list); 1406 spin_unlock_bh(&msk->pm.lock); 1407 } 1408 return ret; 1409 } 1410 1411 static int mptcp_nl_remove_subflow_and_signal_addr(struct net *net, 1412 const struct mptcp_pm_addr_entry *entry) 1413 { 1414 const struct mptcp_addr_info *addr = &entry->addr; 1415 struct mptcp_rm_list list = { .nr = 0 }; 1416 long s_slot = 0, s_num = 0; 1417 struct mptcp_sock *msk; 1418 1419 pr_debug("remove_id=%d", addr->id); 1420 1421 list.ids[list.nr++] = addr->id; 1422 1423 while ((msk = mptcp_token_iter_next(net, &s_slot, &s_num)) != NULL) { 1424 struct sock *sk = (struct sock *)msk; 1425 bool remove_subflow; 1426 1427 if (mptcp_pm_is_userspace(msk)) 1428 goto next; 1429 1430 if (list_empty(&msk->conn_list)) { 1431 mptcp_pm_remove_anno_addr(msk, addr, false); 1432 goto next; 1433 } 1434 1435 lock_sock(sk); 1436 remove_subflow = lookup_subflow_by_saddr(&msk->conn_list, addr); 1437 mptcp_pm_remove_anno_addr(msk, addr, remove_subflow && 1438 !(entry->flags & MPTCP_PM_ADDR_FLAG_IMPLICIT)); 1439 if (remove_subflow) 1440 mptcp_pm_remove_subflow(msk, &list); 1441 release_sock(sk); 1442 1443 next: 1444 sock_put(sk); 1445 cond_resched(); 1446 } 1447 1448 return 0; 1449 } 1450 1451 static int mptcp_nl_remove_id_zero_address(struct net *net, 1452 struct mptcp_addr_info *addr) 1453 { 1454 struct mptcp_rm_list list = { .nr = 0 }; 1455 long s_slot = 0, s_num = 0; 1456 struct mptcp_sock *msk; 1457 1458 list.ids[list.nr++] = 0; 1459 1460 while ((msk = mptcp_token_iter_next(net, &s_slot, &s_num)) != NULL) { 1461 struct sock *sk = (struct sock *)msk; 1462 struct mptcp_addr_info msk_local; 1463 1464 if (list_empty(&msk->conn_list) || mptcp_pm_is_userspace(msk)) 1465 goto next; 1466 1467 mptcp_local_address((struct sock_common *)msk, &msk_local); 1468 if (!mptcp_addresses_equal(&msk_local, addr, addr->port)) 1469 goto next; 1470 1471 lock_sock(sk); 1472 spin_lock_bh(&msk->pm.lock); 1473 mptcp_pm_remove_addr(msk, &list); 1474 mptcp_pm_nl_rm_subflow_received(msk, &list); 1475 spin_unlock_bh(&msk->pm.lock); 1476 release_sock(sk); 1477 1478 next: 1479 sock_put(sk); 1480 cond_resched(); 1481 } 1482 1483 return 0; 1484 } 1485 1486 static int mptcp_nl_cmd_del_addr(struct sk_buff *skb, struct genl_info *info) 1487 { 1488 struct nlattr *attr = info->attrs[MPTCP_PM_ATTR_ADDR]; 1489 struct pm_nl_pernet *pernet = genl_info_pm_nl(info); 1490 struct mptcp_pm_addr_entry addr, *entry; 1491 unsigned int addr_max; 1492 int ret; 1493 1494 ret = mptcp_pm_parse_entry(attr, info, false, &addr); 1495 if (ret < 0) 1496 return ret; 1497 1498 /* the zero id address is special: the first address used by the msk 1499 * always gets such an id, so different subflows can have different zero 1500 * id addresses. Additionally zero id is not accounted for in id_bitmap. 1501 * Let's use an 'mptcp_rm_list' instead of the common remove code. 1502 */ 1503 if (addr.addr.id == 0) 1504 return mptcp_nl_remove_id_zero_address(sock_net(skb->sk), &addr.addr); 1505 1506 spin_lock_bh(&pernet->lock); 1507 entry = __lookup_addr_by_id(pernet, addr.addr.id); 1508 if (!entry) { 1509 GENL_SET_ERR_MSG(info, "address not found"); 1510 spin_unlock_bh(&pernet->lock); 1511 return -EINVAL; 1512 } 1513 if (entry->flags & MPTCP_PM_ADDR_FLAG_SIGNAL) { 1514 addr_max = pernet->add_addr_signal_max; 1515 WRITE_ONCE(pernet->add_addr_signal_max, addr_max - 1); 1516 } 1517 if (entry->flags & MPTCP_PM_ADDR_FLAG_SUBFLOW) { 1518 addr_max = pernet->local_addr_max; 1519 WRITE_ONCE(pernet->local_addr_max, addr_max - 1); 1520 } 1521 1522 pernet->addrs--; 1523 list_del_rcu(&entry->list); 1524 __clear_bit(entry->addr.id, pernet->id_bitmap); 1525 spin_unlock_bh(&pernet->lock); 1526 1527 mptcp_nl_remove_subflow_and_signal_addr(sock_net(skb->sk), entry); 1528 synchronize_rcu(); 1529 __mptcp_pm_release_addr_entry(entry); 1530 1531 return ret; 1532 } 1533 1534 void mptcp_pm_remove_addrs(struct mptcp_sock *msk, struct list_head *rm_list) 1535 { 1536 struct mptcp_rm_list alist = { .nr = 0 }; 1537 struct mptcp_pm_addr_entry *entry; 1538 1539 list_for_each_entry(entry, rm_list, list) { 1540 remove_anno_list_by_saddr(msk, &entry->addr); 1541 if (alist.nr < MPTCP_RM_IDS_MAX) 1542 alist.ids[alist.nr++] = entry->addr.id; 1543 } 1544 1545 if (alist.nr) { 1546 spin_lock_bh(&msk->pm.lock); 1547 mptcp_pm_remove_addr(msk, &alist); 1548 spin_unlock_bh(&msk->pm.lock); 1549 } 1550 } 1551 1552 void mptcp_pm_remove_addrs_and_subflows(struct mptcp_sock *msk, 1553 struct list_head *rm_list) 1554 { 1555 struct mptcp_rm_list alist = { .nr = 0 }, slist = { .nr = 0 }; 1556 struct mptcp_pm_addr_entry *entry; 1557 1558 list_for_each_entry(entry, rm_list, list) { 1559 if (lookup_subflow_by_saddr(&msk->conn_list, &entry->addr) && 1560 slist.nr < MPTCP_RM_IDS_MAX) 1561 slist.ids[slist.nr++] = entry->addr.id; 1562 1563 if (remove_anno_list_by_saddr(msk, &entry->addr) && 1564 alist.nr < MPTCP_RM_IDS_MAX) 1565 alist.ids[alist.nr++] = entry->addr.id; 1566 } 1567 1568 if (alist.nr) { 1569 spin_lock_bh(&msk->pm.lock); 1570 mptcp_pm_remove_addr(msk, &alist); 1571 spin_unlock_bh(&msk->pm.lock); 1572 } 1573 if (slist.nr) 1574 mptcp_pm_remove_subflow(msk, &slist); 1575 } 1576 1577 static void mptcp_nl_remove_addrs_list(struct net *net, 1578 struct list_head *rm_list) 1579 { 1580 long s_slot = 0, s_num = 0; 1581 struct mptcp_sock *msk; 1582 1583 if (list_empty(rm_list)) 1584 return; 1585 1586 while ((msk = mptcp_token_iter_next(net, &s_slot, &s_num)) != NULL) { 1587 struct sock *sk = (struct sock *)msk; 1588 1589 if (!mptcp_pm_is_userspace(msk)) { 1590 lock_sock(sk); 1591 mptcp_pm_remove_addrs_and_subflows(msk, rm_list); 1592 release_sock(sk); 1593 } 1594 1595 sock_put(sk); 1596 cond_resched(); 1597 } 1598 } 1599 1600 /* caller must ensure the RCU grace period is already elapsed */ 1601 static void __flush_addrs(struct list_head *list) 1602 { 1603 while (!list_empty(list)) { 1604 struct mptcp_pm_addr_entry *cur; 1605 1606 cur = list_entry(list->next, 1607 struct mptcp_pm_addr_entry, list); 1608 list_del_rcu(&cur->list); 1609 __mptcp_pm_release_addr_entry(cur); 1610 } 1611 } 1612 1613 static void __reset_counters(struct pm_nl_pernet *pernet) 1614 { 1615 WRITE_ONCE(pernet->add_addr_signal_max, 0); 1616 WRITE_ONCE(pernet->add_addr_accept_max, 0); 1617 WRITE_ONCE(pernet->local_addr_max, 0); 1618 pernet->addrs = 0; 1619 } 1620 1621 static int mptcp_nl_cmd_flush_addrs(struct sk_buff *skb, struct genl_info *info) 1622 { 1623 struct pm_nl_pernet *pernet = genl_info_pm_nl(info); 1624 LIST_HEAD(free_list); 1625 1626 spin_lock_bh(&pernet->lock); 1627 list_splice_init(&pernet->local_addr_list, &free_list); 1628 __reset_counters(pernet); 1629 pernet->next_id = 1; 1630 bitmap_zero(pernet->id_bitmap, MPTCP_PM_MAX_ADDR_ID + 1); 1631 spin_unlock_bh(&pernet->lock); 1632 mptcp_nl_remove_addrs_list(sock_net(skb->sk), &free_list); 1633 synchronize_rcu(); 1634 __flush_addrs(&free_list); 1635 return 0; 1636 } 1637 1638 static int mptcp_nl_fill_addr(struct sk_buff *skb, 1639 struct mptcp_pm_addr_entry *entry) 1640 { 1641 struct mptcp_addr_info *addr = &entry->addr; 1642 struct nlattr *attr; 1643 1644 attr = nla_nest_start(skb, MPTCP_PM_ATTR_ADDR); 1645 if (!attr) 1646 return -EMSGSIZE; 1647 1648 if (nla_put_u16(skb, MPTCP_PM_ADDR_ATTR_FAMILY, addr->family)) 1649 goto nla_put_failure; 1650 if (nla_put_u16(skb, MPTCP_PM_ADDR_ATTR_PORT, ntohs(addr->port))) 1651 goto nla_put_failure; 1652 if (nla_put_u8(skb, MPTCP_PM_ADDR_ATTR_ID, addr->id)) 1653 goto nla_put_failure; 1654 if (nla_put_u32(skb, MPTCP_PM_ADDR_ATTR_FLAGS, entry->flags)) 1655 goto nla_put_failure; 1656 if (entry->ifindex && 1657 nla_put_s32(skb, MPTCP_PM_ADDR_ATTR_IF_IDX, entry->ifindex)) 1658 goto nla_put_failure; 1659 1660 if (addr->family == AF_INET && 1661 nla_put_in_addr(skb, MPTCP_PM_ADDR_ATTR_ADDR4, 1662 addr->addr.s_addr)) 1663 goto nla_put_failure; 1664 #if IS_ENABLED(CONFIG_MPTCP_IPV6) 1665 else if (addr->family == AF_INET6 && 1666 nla_put_in6_addr(skb, MPTCP_PM_ADDR_ATTR_ADDR6, &addr->addr6)) 1667 goto nla_put_failure; 1668 #endif 1669 nla_nest_end(skb, attr); 1670 return 0; 1671 1672 nla_put_failure: 1673 nla_nest_cancel(skb, attr); 1674 return -EMSGSIZE; 1675 } 1676 1677 static int mptcp_nl_cmd_get_addr(struct sk_buff *skb, struct genl_info *info) 1678 { 1679 struct nlattr *attr = info->attrs[MPTCP_PM_ATTR_ADDR]; 1680 struct pm_nl_pernet *pernet = genl_info_pm_nl(info); 1681 struct mptcp_pm_addr_entry addr, *entry; 1682 struct sk_buff *msg; 1683 void *reply; 1684 int ret; 1685 1686 ret = mptcp_pm_parse_entry(attr, info, false, &addr); 1687 if (ret < 0) 1688 return ret; 1689 1690 msg = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL); 1691 if (!msg) 1692 return -ENOMEM; 1693 1694 reply = genlmsg_put_reply(msg, info, &mptcp_genl_family, 0, 1695 info->genlhdr->cmd); 1696 if (!reply) { 1697 GENL_SET_ERR_MSG(info, "not enough space in Netlink message"); 1698 ret = -EMSGSIZE; 1699 goto fail; 1700 } 1701 1702 spin_lock_bh(&pernet->lock); 1703 entry = __lookup_addr_by_id(pernet, addr.addr.id); 1704 if (!entry) { 1705 GENL_SET_ERR_MSG(info, "address not found"); 1706 ret = -EINVAL; 1707 goto unlock_fail; 1708 } 1709 1710 ret = mptcp_nl_fill_addr(msg, entry); 1711 if (ret) 1712 goto unlock_fail; 1713 1714 genlmsg_end(msg, reply); 1715 ret = genlmsg_reply(msg, info); 1716 spin_unlock_bh(&pernet->lock); 1717 return ret; 1718 1719 unlock_fail: 1720 spin_unlock_bh(&pernet->lock); 1721 1722 fail: 1723 nlmsg_free(msg); 1724 return ret; 1725 } 1726 1727 static int mptcp_nl_cmd_dump_addrs(struct sk_buff *msg, 1728 struct netlink_callback *cb) 1729 { 1730 struct net *net = sock_net(msg->sk); 1731 struct mptcp_pm_addr_entry *entry; 1732 struct pm_nl_pernet *pernet; 1733 int id = cb->args[0]; 1734 void *hdr; 1735 int i; 1736 1737 pernet = pm_nl_get_pernet(net); 1738 1739 spin_lock_bh(&pernet->lock); 1740 for (i = id; i < MPTCP_PM_MAX_ADDR_ID + 1; i++) { 1741 if (test_bit(i, pernet->id_bitmap)) { 1742 entry = __lookup_addr_by_id(pernet, i); 1743 if (!entry) 1744 break; 1745 1746 if (entry->addr.id <= id) 1747 continue; 1748 1749 hdr = genlmsg_put(msg, NETLINK_CB(cb->skb).portid, 1750 cb->nlh->nlmsg_seq, &mptcp_genl_family, 1751 NLM_F_MULTI, MPTCP_PM_CMD_GET_ADDR); 1752 if (!hdr) 1753 break; 1754 1755 if (mptcp_nl_fill_addr(msg, entry) < 0) { 1756 genlmsg_cancel(msg, hdr); 1757 break; 1758 } 1759 1760 id = entry->addr.id; 1761 genlmsg_end(msg, hdr); 1762 } 1763 } 1764 spin_unlock_bh(&pernet->lock); 1765 1766 cb->args[0] = id; 1767 return msg->len; 1768 } 1769 1770 static int parse_limit(struct genl_info *info, int id, unsigned int *limit) 1771 { 1772 struct nlattr *attr = info->attrs[id]; 1773 1774 if (!attr) 1775 return 0; 1776 1777 *limit = nla_get_u32(attr); 1778 if (*limit > MPTCP_PM_ADDR_MAX) { 1779 GENL_SET_ERR_MSG(info, "limit greater than maximum"); 1780 return -EINVAL; 1781 } 1782 return 0; 1783 } 1784 1785 static int 1786 mptcp_nl_cmd_set_limits(struct sk_buff *skb, struct genl_info *info) 1787 { 1788 struct pm_nl_pernet *pernet = genl_info_pm_nl(info); 1789 unsigned int rcv_addrs, subflows; 1790 int ret; 1791 1792 spin_lock_bh(&pernet->lock); 1793 rcv_addrs = pernet->add_addr_accept_max; 1794 ret = parse_limit(info, MPTCP_PM_ATTR_RCV_ADD_ADDRS, &rcv_addrs); 1795 if (ret) 1796 goto unlock; 1797 1798 subflows = pernet->subflows_max; 1799 ret = parse_limit(info, MPTCP_PM_ATTR_SUBFLOWS, &subflows); 1800 if (ret) 1801 goto unlock; 1802 1803 WRITE_ONCE(pernet->add_addr_accept_max, rcv_addrs); 1804 WRITE_ONCE(pernet->subflows_max, subflows); 1805 1806 unlock: 1807 spin_unlock_bh(&pernet->lock); 1808 return ret; 1809 } 1810 1811 static int 1812 mptcp_nl_cmd_get_limits(struct sk_buff *skb, struct genl_info *info) 1813 { 1814 struct pm_nl_pernet *pernet = genl_info_pm_nl(info); 1815 struct sk_buff *msg; 1816 void *reply; 1817 1818 msg = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL); 1819 if (!msg) 1820 return -ENOMEM; 1821 1822 reply = genlmsg_put_reply(msg, info, &mptcp_genl_family, 0, 1823 MPTCP_PM_CMD_GET_LIMITS); 1824 if (!reply) 1825 goto fail; 1826 1827 if (nla_put_u32(msg, MPTCP_PM_ATTR_RCV_ADD_ADDRS, 1828 READ_ONCE(pernet->add_addr_accept_max))) 1829 goto fail; 1830 1831 if (nla_put_u32(msg, MPTCP_PM_ATTR_SUBFLOWS, 1832 READ_ONCE(pernet->subflows_max))) 1833 goto fail; 1834 1835 genlmsg_end(msg, reply); 1836 return genlmsg_reply(msg, info); 1837 1838 fail: 1839 GENL_SET_ERR_MSG(info, "not enough space in Netlink message"); 1840 nlmsg_free(msg); 1841 return -EMSGSIZE; 1842 } 1843 1844 static void mptcp_pm_nl_fullmesh(struct mptcp_sock *msk, 1845 struct mptcp_addr_info *addr) 1846 { 1847 struct mptcp_rm_list list = { .nr = 0 }; 1848 1849 list.ids[list.nr++] = addr->id; 1850 1851 spin_lock_bh(&msk->pm.lock); 1852 mptcp_pm_nl_rm_subflow_received(msk, &list); 1853 mptcp_pm_create_subflow_or_signal_addr(msk); 1854 spin_unlock_bh(&msk->pm.lock); 1855 } 1856 1857 static int mptcp_nl_set_flags(struct net *net, 1858 struct mptcp_addr_info *addr, 1859 u8 bkup, u8 changed) 1860 { 1861 long s_slot = 0, s_num = 0; 1862 struct mptcp_sock *msk; 1863 int ret = -EINVAL; 1864 1865 while ((msk = mptcp_token_iter_next(net, &s_slot, &s_num)) != NULL) { 1866 struct sock *sk = (struct sock *)msk; 1867 1868 if (list_empty(&msk->conn_list) || mptcp_pm_is_userspace(msk)) 1869 goto next; 1870 1871 lock_sock(sk); 1872 if (changed & MPTCP_PM_ADDR_FLAG_BACKUP) 1873 ret = mptcp_pm_nl_mp_prio_send_ack(msk, addr, NULL, bkup); 1874 if (changed & MPTCP_PM_ADDR_FLAG_FULLMESH) 1875 mptcp_pm_nl_fullmesh(msk, addr); 1876 release_sock(sk); 1877 1878 next: 1879 sock_put(sk); 1880 cond_resched(); 1881 } 1882 1883 return ret; 1884 } 1885 1886 int mptcp_pm_nl_set_flags(struct net *net, struct mptcp_pm_addr_entry *addr, u8 bkup) 1887 { 1888 struct pm_nl_pernet *pernet = pm_nl_get_pernet(net); 1889 u8 changed, mask = MPTCP_PM_ADDR_FLAG_BACKUP | 1890 MPTCP_PM_ADDR_FLAG_FULLMESH; 1891 struct mptcp_pm_addr_entry *entry; 1892 u8 lookup_by_id = 0; 1893 1894 if (addr->addr.family == AF_UNSPEC) { 1895 lookup_by_id = 1; 1896 if (!addr->addr.id) 1897 return -EOPNOTSUPP; 1898 } 1899 1900 spin_lock_bh(&pernet->lock); 1901 entry = __lookup_addr(pernet, &addr->addr, lookup_by_id); 1902 if (!entry) { 1903 spin_unlock_bh(&pernet->lock); 1904 return -EINVAL; 1905 } 1906 if ((addr->flags & MPTCP_PM_ADDR_FLAG_FULLMESH) && 1907 (entry->flags & MPTCP_PM_ADDR_FLAG_SIGNAL)) { 1908 spin_unlock_bh(&pernet->lock); 1909 return -EINVAL; 1910 } 1911 1912 changed = (addr->flags ^ entry->flags) & mask; 1913 entry->flags = (entry->flags & ~mask) | (addr->flags & mask); 1914 *addr = *entry; 1915 spin_unlock_bh(&pernet->lock); 1916 1917 mptcp_nl_set_flags(net, &addr->addr, bkup, changed); 1918 return 0; 1919 } 1920 1921 static int mptcp_nl_cmd_set_flags(struct sk_buff *skb, struct genl_info *info) 1922 { 1923 struct mptcp_pm_addr_entry remote = { .addr = { .family = AF_UNSPEC }, }; 1924 struct mptcp_pm_addr_entry addr = { .addr = { .family = AF_UNSPEC }, }; 1925 struct nlattr *attr_rem = info->attrs[MPTCP_PM_ATTR_ADDR_REMOTE]; 1926 struct nlattr *token = info->attrs[MPTCP_PM_ATTR_TOKEN]; 1927 struct nlattr *attr = info->attrs[MPTCP_PM_ATTR_ADDR]; 1928 struct net *net = sock_net(skb->sk); 1929 u8 bkup = 0; 1930 int ret; 1931 1932 ret = mptcp_pm_parse_entry(attr, info, false, &addr); 1933 if (ret < 0) 1934 return ret; 1935 1936 if (attr_rem) { 1937 ret = mptcp_pm_parse_entry(attr_rem, info, false, &remote); 1938 if (ret < 0) 1939 return ret; 1940 } 1941 1942 if (addr.flags & MPTCP_PM_ADDR_FLAG_BACKUP) 1943 bkup = 1; 1944 1945 return mptcp_pm_set_flags(net, token, &addr, &remote, bkup); 1946 } 1947 1948 static void mptcp_nl_mcast_send(struct net *net, struct sk_buff *nlskb, gfp_t gfp) 1949 { 1950 genlmsg_multicast_netns(&mptcp_genl_family, net, 1951 nlskb, 0, MPTCP_PM_EV_GRP_OFFSET, gfp); 1952 } 1953 1954 bool mptcp_userspace_pm_active(const struct mptcp_sock *msk) 1955 { 1956 return genl_has_listeners(&mptcp_genl_family, 1957 sock_net((const struct sock *)msk), 1958 MPTCP_PM_EV_GRP_OFFSET); 1959 } 1960 1961 static int mptcp_event_add_subflow(struct sk_buff *skb, const struct sock *ssk) 1962 { 1963 const struct inet_sock *issk = inet_sk(ssk); 1964 const struct mptcp_subflow_context *sf; 1965 1966 if (nla_put_u16(skb, MPTCP_ATTR_FAMILY, ssk->sk_family)) 1967 return -EMSGSIZE; 1968 1969 switch (ssk->sk_family) { 1970 case AF_INET: 1971 if (nla_put_in_addr(skb, MPTCP_ATTR_SADDR4, issk->inet_saddr)) 1972 return -EMSGSIZE; 1973 if (nla_put_in_addr(skb, MPTCP_ATTR_DADDR4, issk->inet_daddr)) 1974 return -EMSGSIZE; 1975 break; 1976 #if IS_ENABLED(CONFIG_MPTCP_IPV6) 1977 case AF_INET6: { 1978 const struct ipv6_pinfo *np = inet6_sk(ssk); 1979 1980 if (nla_put_in6_addr(skb, MPTCP_ATTR_SADDR6, &np->saddr)) 1981 return -EMSGSIZE; 1982 if (nla_put_in6_addr(skb, MPTCP_ATTR_DADDR6, &ssk->sk_v6_daddr)) 1983 return -EMSGSIZE; 1984 break; 1985 } 1986 #endif 1987 default: 1988 WARN_ON_ONCE(1); 1989 return -EMSGSIZE; 1990 } 1991 1992 if (nla_put_be16(skb, MPTCP_ATTR_SPORT, issk->inet_sport)) 1993 return -EMSGSIZE; 1994 if (nla_put_be16(skb, MPTCP_ATTR_DPORT, issk->inet_dport)) 1995 return -EMSGSIZE; 1996 1997 sf = mptcp_subflow_ctx(ssk); 1998 if (WARN_ON_ONCE(!sf)) 1999 return -EINVAL; 2000 2001 if (nla_put_u8(skb, MPTCP_ATTR_LOC_ID, sf->local_id)) 2002 return -EMSGSIZE; 2003 2004 if (nla_put_u8(skb, MPTCP_ATTR_REM_ID, sf->remote_id)) 2005 return -EMSGSIZE; 2006 2007 return 0; 2008 } 2009 2010 static int mptcp_event_put_token_and_ssk(struct sk_buff *skb, 2011 const struct mptcp_sock *msk, 2012 const struct sock *ssk) 2013 { 2014 const struct sock *sk = (const struct sock *)msk; 2015 const struct mptcp_subflow_context *sf; 2016 u8 sk_err; 2017 2018 if (nla_put_u32(skb, MPTCP_ATTR_TOKEN, msk->token)) 2019 return -EMSGSIZE; 2020 2021 if (mptcp_event_add_subflow(skb, ssk)) 2022 return -EMSGSIZE; 2023 2024 sf = mptcp_subflow_ctx(ssk); 2025 if (WARN_ON_ONCE(!sf)) 2026 return -EINVAL; 2027 2028 if (nla_put_u8(skb, MPTCP_ATTR_BACKUP, sf->backup)) 2029 return -EMSGSIZE; 2030 2031 if (ssk->sk_bound_dev_if && 2032 nla_put_s32(skb, MPTCP_ATTR_IF_IDX, ssk->sk_bound_dev_if)) 2033 return -EMSGSIZE; 2034 2035 sk_err = READ_ONCE(ssk->sk_err); 2036 if (sk_err && sk->sk_state == TCP_ESTABLISHED && 2037 nla_put_u8(skb, MPTCP_ATTR_ERROR, sk_err)) 2038 return -EMSGSIZE; 2039 2040 return 0; 2041 } 2042 2043 static int mptcp_event_sub_established(struct sk_buff *skb, 2044 const struct mptcp_sock *msk, 2045 const struct sock *ssk) 2046 { 2047 return mptcp_event_put_token_and_ssk(skb, msk, ssk); 2048 } 2049 2050 static int mptcp_event_sub_closed(struct sk_buff *skb, 2051 const struct mptcp_sock *msk, 2052 const struct sock *ssk) 2053 { 2054 const struct mptcp_subflow_context *sf; 2055 2056 if (mptcp_event_put_token_and_ssk(skb, msk, ssk)) 2057 return -EMSGSIZE; 2058 2059 sf = mptcp_subflow_ctx(ssk); 2060 if (!sf->reset_seen) 2061 return 0; 2062 2063 if (nla_put_u32(skb, MPTCP_ATTR_RESET_REASON, sf->reset_reason)) 2064 return -EMSGSIZE; 2065 2066 if (nla_put_u32(skb, MPTCP_ATTR_RESET_FLAGS, sf->reset_transient)) 2067 return -EMSGSIZE; 2068 2069 return 0; 2070 } 2071 2072 static int mptcp_event_created(struct sk_buff *skb, 2073 const struct mptcp_sock *msk, 2074 const struct sock *ssk) 2075 { 2076 int err = nla_put_u32(skb, MPTCP_ATTR_TOKEN, msk->token); 2077 2078 if (err) 2079 return err; 2080 2081 if (nla_put_u8(skb, MPTCP_ATTR_SERVER_SIDE, READ_ONCE(msk->pm.server_side))) 2082 return -EMSGSIZE; 2083 2084 return mptcp_event_add_subflow(skb, ssk); 2085 } 2086 2087 void mptcp_event_addr_removed(const struct mptcp_sock *msk, uint8_t id) 2088 { 2089 struct net *net = sock_net((const struct sock *)msk); 2090 struct nlmsghdr *nlh; 2091 struct sk_buff *skb; 2092 2093 if (!genl_has_listeners(&mptcp_genl_family, net, MPTCP_PM_EV_GRP_OFFSET)) 2094 return; 2095 2096 skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_ATOMIC); 2097 if (!skb) 2098 return; 2099 2100 nlh = genlmsg_put(skb, 0, 0, &mptcp_genl_family, 0, MPTCP_EVENT_REMOVED); 2101 if (!nlh) 2102 goto nla_put_failure; 2103 2104 if (nla_put_u32(skb, MPTCP_ATTR_TOKEN, msk->token)) 2105 goto nla_put_failure; 2106 2107 if (nla_put_u8(skb, MPTCP_ATTR_REM_ID, id)) 2108 goto nla_put_failure; 2109 2110 genlmsg_end(skb, nlh); 2111 mptcp_nl_mcast_send(net, skb, GFP_ATOMIC); 2112 return; 2113 2114 nla_put_failure: 2115 nlmsg_free(skb); 2116 } 2117 2118 void mptcp_event_addr_announced(const struct sock *ssk, 2119 const struct mptcp_addr_info *info) 2120 { 2121 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk); 2122 struct mptcp_sock *msk = mptcp_sk(subflow->conn); 2123 struct net *net = sock_net(ssk); 2124 struct nlmsghdr *nlh; 2125 struct sk_buff *skb; 2126 2127 if (!genl_has_listeners(&mptcp_genl_family, net, MPTCP_PM_EV_GRP_OFFSET)) 2128 return; 2129 2130 skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_ATOMIC); 2131 if (!skb) 2132 return; 2133 2134 nlh = genlmsg_put(skb, 0, 0, &mptcp_genl_family, 0, 2135 MPTCP_EVENT_ANNOUNCED); 2136 if (!nlh) 2137 goto nla_put_failure; 2138 2139 if (nla_put_u32(skb, MPTCP_ATTR_TOKEN, msk->token)) 2140 goto nla_put_failure; 2141 2142 if (nla_put_u8(skb, MPTCP_ATTR_REM_ID, info->id)) 2143 goto nla_put_failure; 2144 2145 if (nla_put_be16(skb, MPTCP_ATTR_DPORT, 2146 info->port == 0 ? 2147 inet_sk(ssk)->inet_dport : 2148 info->port)) 2149 goto nla_put_failure; 2150 2151 switch (info->family) { 2152 case AF_INET: 2153 if (nla_put_in_addr(skb, MPTCP_ATTR_DADDR4, info->addr.s_addr)) 2154 goto nla_put_failure; 2155 break; 2156 #if IS_ENABLED(CONFIG_MPTCP_IPV6) 2157 case AF_INET6: 2158 if (nla_put_in6_addr(skb, MPTCP_ATTR_DADDR6, &info->addr6)) 2159 goto nla_put_failure; 2160 break; 2161 #endif 2162 default: 2163 WARN_ON_ONCE(1); 2164 goto nla_put_failure; 2165 } 2166 2167 genlmsg_end(skb, nlh); 2168 mptcp_nl_mcast_send(net, skb, GFP_ATOMIC); 2169 return; 2170 2171 nla_put_failure: 2172 nlmsg_free(skb); 2173 } 2174 2175 void mptcp_event_pm_listener(const struct sock *ssk, 2176 enum mptcp_event_type event) 2177 { 2178 const struct inet_sock *issk = inet_sk(ssk); 2179 struct net *net = sock_net(ssk); 2180 struct nlmsghdr *nlh; 2181 struct sk_buff *skb; 2182 2183 if (!genl_has_listeners(&mptcp_genl_family, net, MPTCP_PM_EV_GRP_OFFSET)) 2184 return; 2185 2186 skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL); 2187 if (!skb) 2188 return; 2189 2190 nlh = genlmsg_put(skb, 0, 0, &mptcp_genl_family, 0, event); 2191 if (!nlh) 2192 goto nla_put_failure; 2193 2194 if (nla_put_u16(skb, MPTCP_ATTR_FAMILY, ssk->sk_family)) 2195 goto nla_put_failure; 2196 2197 if (nla_put_be16(skb, MPTCP_ATTR_SPORT, issk->inet_sport)) 2198 goto nla_put_failure; 2199 2200 switch (ssk->sk_family) { 2201 case AF_INET: 2202 if (nla_put_in_addr(skb, MPTCP_ATTR_SADDR4, issk->inet_saddr)) 2203 goto nla_put_failure; 2204 break; 2205 #if IS_ENABLED(CONFIG_MPTCP_IPV6) 2206 case AF_INET6: { 2207 const struct ipv6_pinfo *np = inet6_sk(ssk); 2208 2209 if (nla_put_in6_addr(skb, MPTCP_ATTR_SADDR6, &np->saddr)) 2210 goto nla_put_failure; 2211 break; 2212 } 2213 #endif 2214 default: 2215 WARN_ON_ONCE(1); 2216 goto nla_put_failure; 2217 } 2218 2219 genlmsg_end(skb, nlh); 2220 mptcp_nl_mcast_send(net, skb, GFP_KERNEL); 2221 return; 2222 2223 nla_put_failure: 2224 nlmsg_free(skb); 2225 } 2226 2227 void mptcp_event(enum mptcp_event_type type, const struct mptcp_sock *msk, 2228 const struct sock *ssk, gfp_t gfp) 2229 { 2230 struct net *net = sock_net((const struct sock *)msk); 2231 struct nlmsghdr *nlh; 2232 struct sk_buff *skb; 2233 2234 if (!genl_has_listeners(&mptcp_genl_family, net, MPTCP_PM_EV_GRP_OFFSET)) 2235 return; 2236 2237 skb = nlmsg_new(NLMSG_DEFAULT_SIZE, gfp); 2238 if (!skb) 2239 return; 2240 2241 nlh = genlmsg_put(skb, 0, 0, &mptcp_genl_family, 0, type); 2242 if (!nlh) 2243 goto nla_put_failure; 2244 2245 switch (type) { 2246 case MPTCP_EVENT_UNSPEC: 2247 WARN_ON_ONCE(1); 2248 break; 2249 case MPTCP_EVENT_CREATED: 2250 case MPTCP_EVENT_ESTABLISHED: 2251 if (mptcp_event_created(skb, msk, ssk) < 0) 2252 goto nla_put_failure; 2253 break; 2254 case MPTCP_EVENT_CLOSED: 2255 if (nla_put_u32(skb, MPTCP_ATTR_TOKEN, msk->token) < 0) 2256 goto nla_put_failure; 2257 break; 2258 case MPTCP_EVENT_ANNOUNCED: 2259 case MPTCP_EVENT_REMOVED: 2260 /* call mptcp_event_addr_announced()/removed instead */ 2261 WARN_ON_ONCE(1); 2262 break; 2263 case MPTCP_EVENT_SUB_ESTABLISHED: 2264 case MPTCP_EVENT_SUB_PRIORITY: 2265 if (mptcp_event_sub_established(skb, msk, ssk) < 0) 2266 goto nla_put_failure; 2267 break; 2268 case MPTCP_EVENT_SUB_CLOSED: 2269 if (mptcp_event_sub_closed(skb, msk, ssk) < 0) 2270 goto nla_put_failure; 2271 break; 2272 case MPTCP_EVENT_LISTENER_CREATED: 2273 case MPTCP_EVENT_LISTENER_CLOSED: 2274 break; 2275 } 2276 2277 genlmsg_end(skb, nlh); 2278 mptcp_nl_mcast_send(net, skb, gfp); 2279 return; 2280 2281 nla_put_failure: 2282 nlmsg_free(skb); 2283 } 2284 2285 static const struct genl_small_ops mptcp_pm_ops[] = { 2286 { 2287 .cmd = MPTCP_PM_CMD_ADD_ADDR, 2288 .doit = mptcp_nl_cmd_add_addr, 2289 .flags = GENL_UNS_ADMIN_PERM, 2290 }, 2291 { 2292 .cmd = MPTCP_PM_CMD_DEL_ADDR, 2293 .doit = mptcp_nl_cmd_del_addr, 2294 .flags = GENL_UNS_ADMIN_PERM, 2295 }, 2296 { 2297 .cmd = MPTCP_PM_CMD_FLUSH_ADDRS, 2298 .doit = mptcp_nl_cmd_flush_addrs, 2299 .flags = GENL_UNS_ADMIN_PERM, 2300 }, 2301 { 2302 .cmd = MPTCP_PM_CMD_GET_ADDR, 2303 .doit = mptcp_nl_cmd_get_addr, 2304 .dumpit = mptcp_nl_cmd_dump_addrs, 2305 }, 2306 { 2307 .cmd = MPTCP_PM_CMD_SET_LIMITS, 2308 .doit = mptcp_nl_cmd_set_limits, 2309 .flags = GENL_UNS_ADMIN_PERM, 2310 }, 2311 { 2312 .cmd = MPTCP_PM_CMD_GET_LIMITS, 2313 .doit = mptcp_nl_cmd_get_limits, 2314 }, 2315 { 2316 .cmd = MPTCP_PM_CMD_SET_FLAGS, 2317 .doit = mptcp_nl_cmd_set_flags, 2318 .flags = GENL_UNS_ADMIN_PERM, 2319 }, 2320 { 2321 .cmd = MPTCP_PM_CMD_ANNOUNCE, 2322 .doit = mptcp_nl_cmd_announce, 2323 .flags = GENL_UNS_ADMIN_PERM, 2324 }, 2325 { 2326 .cmd = MPTCP_PM_CMD_REMOVE, 2327 .doit = mptcp_nl_cmd_remove, 2328 .flags = GENL_UNS_ADMIN_PERM, 2329 }, 2330 { 2331 .cmd = MPTCP_PM_CMD_SUBFLOW_CREATE, 2332 .doit = mptcp_nl_cmd_sf_create, 2333 .flags = GENL_UNS_ADMIN_PERM, 2334 }, 2335 { 2336 .cmd = MPTCP_PM_CMD_SUBFLOW_DESTROY, 2337 .doit = mptcp_nl_cmd_sf_destroy, 2338 .flags = GENL_UNS_ADMIN_PERM, 2339 }, 2340 }; 2341 2342 static struct genl_family mptcp_genl_family __ro_after_init = { 2343 .name = MPTCP_PM_NAME, 2344 .version = MPTCP_PM_VER, 2345 .maxattr = MPTCP_PM_ATTR_MAX, 2346 .policy = mptcp_pm_policy, 2347 .netnsok = true, 2348 .module = THIS_MODULE, 2349 .small_ops = mptcp_pm_ops, 2350 .n_small_ops = ARRAY_SIZE(mptcp_pm_ops), 2351 .resv_start_op = MPTCP_PM_CMD_SUBFLOW_DESTROY + 1, 2352 .mcgrps = mptcp_pm_mcgrps, 2353 .n_mcgrps = ARRAY_SIZE(mptcp_pm_mcgrps), 2354 }; 2355 2356 static int __net_init pm_nl_init_net(struct net *net) 2357 { 2358 struct pm_nl_pernet *pernet = pm_nl_get_pernet(net); 2359 2360 INIT_LIST_HEAD_RCU(&pernet->local_addr_list); 2361 2362 /* Cit. 2 subflows ought to be enough for anybody. */ 2363 pernet->subflows_max = 2; 2364 pernet->next_id = 1; 2365 pernet->stale_loss_cnt = 4; 2366 spin_lock_init(&pernet->lock); 2367 2368 /* No need to initialize other pernet fields, the struct is zeroed at 2369 * allocation time. 2370 */ 2371 2372 return 0; 2373 } 2374 2375 static void __net_exit pm_nl_exit_net(struct list_head *net_list) 2376 { 2377 struct net *net; 2378 2379 list_for_each_entry(net, net_list, exit_list) { 2380 struct pm_nl_pernet *pernet = pm_nl_get_pernet(net); 2381 2382 /* net is removed from namespace list, can't race with 2383 * other modifiers, also netns core already waited for a 2384 * RCU grace period. 2385 */ 2386 __flush_addrs(&pernet->local_addr_list); 2387 } 2388 } 2389 2390 static struct pernet_operations mptcp_pm_pernet_ops = { 2391 .init = pm_nl_init_net, 2392 .exit_batch = pm_nl_exit_net, 2393 .id = &pm_nl_pernet_id, 2394 .size = sizeof(struct pm_nl_pernet), 2395 }; 2396 2397 void __init mptcp_pm_nl_init(void) 2398 { 2399 if (register_pernet_subsys(&mptcp_pm_pernet_ops) < 0) 2400 panic("Failed to register MPTCP PM pernet subsystem.\n"); 2401 2402 if (genl_register_family(&mptcp_genl_family)) 2403 panic("Failed to register MPTCP PM netlink family\n"); 2404 } 2405