1 // SPDX-License-Identifier: GPL-2.0 2 /* Multipath TCP 3 * 4 * Copyright (c) 2020, Red Hat, Inc. 5 */ 6 7 #define pr_fmt(fmt) "MPTCP: " fmt 8 9 #include <linux/inet.h> 10 #include <linux/kernel.h> 11 #include <net/tcp.h> 12 #include <net/netns/generic.h> 13 #include <net/mptcp.h> 14 #include <net/genetlink.h> 15 #include <uapi/linux/mptcp.h> 16 17 #include "protocol.h" 18 #include "mib.h" 19 20 /* forward declaration */ 21 static struct genl_family mptcp_genl_family; 22 23 static int pm_nl_pernet_id; 24 25 struct mptcp_pm_add_entry { 26 struct list_head list; 27 struct mptcp_addr_info addr; 28 struct timer_list add_timer; 29 struct mptcp_sock *sock; 30 u8 retrans_times; 31 }; 32 33 struct pm_nl_pernet { 34 /* protects pernet updates */ 35 spinlock_t lock; 36 struct list_head local_addr_list; 37 unsigned int addrs; 38 unsigned int stale_loss_cnt; 39 unsigned int add_addr_signal_max; 40 unsigned int add_addr_accept_max; 41 unsigned int local_addr_max; 42 unsigned int subflows_max; 43 unsigned int next_id; 44 DECLARE_BITMAP(id_bitmap, MPTCP_PM_MAX_ADDR_ID + 1); 45 }; 46 47 #define MPTCP_PM_ADDR_MAX 8 48 #define ADD_ADDR_RETRANS_MAX 3 49 50 static struct pm_nl_pernet *pm_nl_get_pernet(const struct net *net) 51 { 52 return net_generic(net, pm_nl_pernet_id); 53 } 54 55 static struct pm_nl_pernet * 56 pm_nl_get_pernet_from_msk(const struct mptcp_sock *msk) 57 { 58 return pm_nl_get_pernet(sock_net((struct sock *)msk)); 59 } 60 61 bool mptcp_addresses_equal(const struct mptcp_addr_info *a, 62 const struct mptcp_addr_info *b, bool use_port) 63 { 64 bool addr_equals = false; 65 66 if (a->family == b->family) { 67 if (a->family == AF_INET) 68 addr_equals = a->addr.s_addr == b->addr.s_addr; 69 #if IS_ENABLED(CONFIG_MPTCP_IPV6) 70 else 71 addr_equals = !ipv6_addr_cmp(&a->addr6, &b->addr6); 72 } else if (a->family == AF_INET) { 73 if (ipv6_addr_v4mapped(&b->addr6)) 74 addr_equals = a->addr.s_addr == b->addr6.s6_addr32[3]; 75 } else if (b->family == AF_INET) { 76 if (ipv6_addr_v4mapped(&a->addr6)) 77 addr_equals = a->addr6.s6_addr32[3] == b->addr.s_addr; 78 #endif 79 } 80 81 if (!addr_equals) 82 return false; 83 if (!use_port) 84 return true; 85 86 return a->port == b->port; 87 } 88 89 static void local_address(const struct sock_common *skc, 90 struct mptcp_addr_info *addr) 91 { 92 addr->family = skc->skc_family; 93 addr->port = htons(skc->skc_num); 94 if (addr->family == AF_INET) 95 addr->addr.s_addr = skc->skc_rcv_saddr; 96 #if IS_ENABLED(CONFIG_MPTCP_IPV6) 97 else if (addr->family == AF_INET6) 98 addr->addr6 = skc->skc_v6_rcv_saddr; 99 #endif 100 } 101 102 static void remote_address(const struct sock_common *skc, 103 struct mptcp_addr_info *addr) 104 { 105 addr->family = skc->skc_family; 106 addr->port = skc->skc_dport; 107 if (addr->family == AF_INET) 108 addr->addr.s_addr = skc->skc_daddr; 109 #if IS_ENABLED(CONFIG_MPTCP_IPV6) 110 else if (addr->family == AF_INET6) 111 addr->addr6 = skc->skc_v6_daddr; 112 #endif 113 } 114 115 static bool lookup_subflow_by_saddr(const struct list_head *list, 116 const struct mptcp_addr_info *saddr) 117 { 118 struct mptcp_subflow_context *subflow; 119 struct mptcp_addr_info cur; 120 struct sock_common *skc; 121 122 list_for_each_entry(subflow, list, node) { 123 skc = (struct sock_common *)mptcp_subflow_tcp_sock(subflow); 124 125 local_address(skc, &cur); 126 if (mptcp_addresses_equal(&cur, saddr, saddr->port)) 127 return true; 128 } 129 130 return false; 131 } 132 133 static bool lookup_subflow_by_daddr(const struct list_head *list, 134 const struct mptcp_addr_info *daddr) 135 { 136 struct mptcp_subflow_context *subflow; 137 struct mptcp_addr_info cur; 138 struct sock_common *skc; 139 140 list_for_each_entry(subflow, list, node) { 141 skc = (struct sock_common *)mptcp_subflow_tcp_sock(subflow); 142 143 remote_address(skc, &cur); 144 if (mptcp_addresses_equal(&cur, daddr, daddr->port)) 145 return true; 146 } 147 148 return false; 149 } 150 151 static struct mptcp_pm_addr_entry * 152 select_local_address(const struct pm_nl_pernet *pernet, 153 const struct mptcp_sock *msk) 154 { 155 struct mptcp_pm_addr_entry *entry, *ret = NULL; 156 157 msk_owned_by_me(msk); 158 159 rcu_read_lock(); 160 list_for_each_entry_rcu(entry, &pernet->local_addr_list, list) { 161 if (!(entry->flags & MPTCP_PM_ADDR_FLAG_SUBFLOW)) 162 continue; 163 164 if (!test_bit(entry->addr.id, msk->pm.id_avail_bitmap)) 165 continue; 166 167 ret = entry; 168 break; 169 } 170 rcu_read_unlock(); 171 return ret; 172 } 173 174 static struct mptcp_pm_addr_entry * 175 select_signal_address(struct pm_nl_pernet *pernet, const struct mptcp_sock *msk) 176 { 177 struct mptcp_pm_addr_entry *entry, *ret = NULL; 178 179 rcu_read_lock(); 180 /* do not keep any additional per socket state, just signal 181 * the address list in order. 182 * Note: removal from the local address list during the msk life-cycle 183 * can lead to additional addresses not being announced. 184 */ 185 list_for_each_entry_rcu(entry, &pernet->local_addr_list, list) { 186 if (!test_bit(entry->addr.id, msk->pm.id_avail_bitmap)) 187 continue; 188 189 if (!(entry->flags & MPTCP_PM_ADDR_FLAG_SIGNAL)) 190 continue; 191 192 ret = entry; 193 break; 194 } 195 rcu_read_unlock(); 196 return ret; 197 } 198 199 unsigned int mptcp_pm_get_add_addr_signal_max(const struct mptcp_sock *msk) 200 { 201 const struct pm_nl_pernet *pernet = pm_nl_get_pernet_from_msk(msk); 202 203 return READ_ONCE(pernet->add_addr_signal_max); 204 } 205 EXPORT_SYMBOL_GPL(mptcp_pm_get_add_addr_signal_max); 206 207 unsigned int mptcp_pm_get_add_addr_accept_max(const struct mptcp_sock *msk) 208 { 209 struct pm_nl_pernet *pernet = pm_nl_get_pernet_from_msk(msk); 210 211 return READ_ONCE(pernet->add_addr_accept_max); 212 } 213 EXPORT_SYMBOL_GPL(mptcp_pm_get_add_addr_accept_max); 214 215 unsigned int mptcp_pm_get_subflows_max(const struct mptcp_sock *msk) 216 { 217 struct pm_nl_pernet *pernet = pm_nl_get_pernet_from_msk(msk); 218 219 return READ_ONCE(pernet->subflows_max); 220 } 221 EXPORT_SYMBOL_GPL(mptcp_pm_get_subflows_max); 222 223 unsigned int mptcp_pm_get_local_addr_max(const struct mptcp_sock *msk) 224 { 225 struct pm_nl_pernet *pernet = pm_nl_get_pernet_from_msk(msk); 226 227 return READ_ONCE(pernet->local_addr_max); 228 } 229 EXPORT_SYMBOL_GPL(mptcp_pm_get_local_addr_max); 230 231 bool mptcp_pm_nl_check_work_pending(struct mptcp_sock *msk) 232 { 233 struct pm_nl_pernet *pernet = pm_nl_get_pernet_from_msk(msk); 234 235 if (msk->pm.subflows == mptcp_pm_get_subflows_max(msk) || 236 (find_next_and_bit(pernet->id_bitmap, msk->pm.id_avail_bitmap, 237 MPTCP_PM_MAX_ADDR_ID + 1, 0) == MPTCP_PM_MAX_ADDR_ID + 1)) { 238 WRITE_ONCE(msk->pm.work_pending, false); 239 return false; 240 } 241 return true; 242 } 243 244 struct mptcp_pm_add_entry * 245 mptcp_lookup_anno_list_by_saddr(const struct mptcp_sock *msk, 246 const struct mptcp_addr_info *addr) 247 { 248 struct mptcp_pm_add_entry *entry; 249 250 lockdep_assert_held(&msk->pm.lock); 251 252 list_for_each_entry(entry, &msk->pm.anno_list, list) { 253 if (mptcp_addresses_equal(&entry->addr, addr, true)) 254 return entry; 255 } 256 257 return NULL; 258 } 259 260 bool mptcp_pm_sport_in_anno_list(struct mptcp_sock *msk, const struct sock *sk) 261 { 262 struct mptcp_pm_add_entry *entry; 263 struct mptcp_addr_info saddr; 264 bool ret = false; 265 266 local_address((struct sock_common *)sk, &saddr); 267 268 spin_lock_bh(&msk->pm.lock); 269 list_for_each_entry(entry, &msk->pm.anno_list, list) { 270 if (mptcp_addresses_equal(&entry->addr, &saddr, true)) { 271 ret = true; 272 goto out; 273 } 274 } 275 276 out: 277 spin_unlock_bh(&msk->pm.lock); 278 return ret; 279 } 280 281 static void mptcp_pm_add_timer(struct timer_list *timer) 282 { 283 struct mptcp_pm_add_entry *entry = from_timer(entry, timer, add_timer); 284 struct mptcp_sock *msk = entry->sock; 285 struct sock *sk = (struct sock *)msk; 286 287 pr_debug("msk=%p", msk); 288 289 if (!msk) 290 return; 291 292 if (inet_sk_state_load(sk) == TCP_CLOSE) 293 return; 294 295 if (!entry->addr.id) 296 return; 297 298 if (mptcp_pm_should_add_signal_addr(msk)) { 299 sk_reset_timer(sk, timer, jiffies + TCP_RTO_MAX / 8); 300 goto out; 301 } 302 303 spin_lock_bh(&msk->pm.lock); 304 305 if (!mptcp_pm_should_add_signal_addr(msk)) { 306 pr_debug("retransmit ADD_ADDR id=%d", entry->addr.id); 307 mptcp_pm_announce_addr(msk, &entry->addr, false); 308 mptcp_pm_add_addr_send_ack(msk); 309 entry->retrans_times++; 310 } 311 312 if (entry->retrans_times < ADD_ADDR_RETRANS_MAX) 313 sk_reset_timer(sk, timer, 314 jiffies + mptcp_get_add_addr_timeout(sock_net(sk))); 315 316 spin_unlock_bh(&msk->pm.lock); 317 318 if (entry->retrans_times == ADD_ADDR_RETRANS_MAX) 319 mptcp_pm_subflow_established(msk); 320 321 out: 322 __sock_put(sk); 323 } 324 325 struct mptcp_pm_add_entry * 326 mptcp_pm_del_add_timer(struct mptcp_sock *msk, 327 const struct mptcp_addr_info *addr, bool check_id) 328 { 329 struct mptcp_pm_add_entry *entry; 330 struct sock *sk = (struct sock *)msk; 331 332 spin_lock_bh(&msk->pm.lock); 333 entry = mptcp_lookup_anno_list_by_saddr(msk, addr); 334 if (entry && (!check_id || entry->addr.id == addr->id)) 335 entry->retrans_times = ADD_ADDR_RETRANS_MAX; 336 spin_unlock_bh(&msk->pm.lock); 337 338 if (entry && (!check_id || entry->addr.id == addr->id)) 339 sk_stop_timer_sync(sk, &entry->add_timer); 340 341 return entry; 342 } 343 344 bool mptcp_pm_alloc_anno_list(struct mptcp_sock *msk, 345 const struct mptcp_pm_addr_entry *entry) 346 { 347 struct mptcp_pm_add_entry *add_entry = NULL; 348 struct sock *sk = (struct sock *)msk; 349 struct net *net = sock_net(sk); 350 351 lockdep_assert_held(&msk->pm.lock); 352 353 add_entry = mptcp_lookup_anno_list_by_saddr(msk, &entry->addr); 354 355 if (add_entry) { 356 if (mptcp_pm_is_kernel(msk)) 357 return false; 358 359 sk_reset_timer(sk, &add_entry->add_timer, 360 jiffies + mptcp_get_add_addr_timeout(net)); 361 return true; 362 } 363 364 add_entry = kmalloc(sizeof(*add_entry), GFP_ATOMIC); 365 if (!add_entry) 366 return false; 367 368 list_add(&add_entry->list, &msk->pm.anno_list); 369 370 add_entry->addr = entry->addr; 371 add_entry->sock = msk; 372 add_entry->retrans_times = 0; 373 374 timer_setup(&add_entry->add_timer, mptcp_pm_add_timer, 0); 375 sk_reset_timer(sk, &add_entry->add_timer, 376 jiffies + mptcp_get_add_addr_timeout(net)); 377 378 return true; 379 } 380 381 void mptcp_pm_free_anno_list(struct mptcp_sock *msk) 382 { 383 struct mptcp_pm_add_entry *entry, *tmp; 384 struct sock *sk = (struct sock *)msk; 385 LIST_HEAD(free_list); 386 387 pr_debug("msk=%p", msk); 388 389 spin_lock_bh(&msk->pm.lock); 390 list_splice_init(&msk->pm.anno_list, &free_list); 391 spin_unlock_bh(&msk->pm.lock); 392 393 list_for_each_entry_safe(entry, tmp, &free_list, list) { 394 sk_stop_timer_sync(sk, &entry->add_timer); 395 kfree(entry); 396 } 397 } 398 399 static bool lookup_address_in_vec(const struct mptcp_addr_info *addrs, unsigned int nr, 400 const struct mptcp_addr_info *addr) 401 { 402 int i; 403 404 for (i = 0; i < nr; i++) { 405 if (addrs[i].id == addr->id) 406 return true; 407 } 408 409 return false; 410 } 411 412 /* Fill all the remote addresses into the array addrs[], 413 * and return the array size. 414 */ 415 static unsigned int fill_remote_addresses_vec(struct mptcp_sock *msk, 416 struct mptcp_addr_info *local, 417 bool fullmesh, 418 struct mptcp_addr_info *addrs) 419 { 420 bool deny_id0 = READ_ONCE(msk->pm.remote_deny_join_id0); 421 struct sock *sk = (struct sock *)msk, *ssk; 422 struct mptcp_subflow_context *subflow; 423 struct mptcp_addr_info remote = { 0 }; 424 unsigned int subflows_max; 425 int i = 0; 426 427 subflows_max = mptcp_pm_get_subflows_max(msk); 428 remote_address((struct sock_common *)sk, &remote); 429 430 /* Non-fullmesh endpoint, fill in the single entry 431 * corresponding to the primary MPC subflow remote address 432 */ 433 if (!fullmesh) { 434 if (deny_id0) 435 return 0; 436 437 if (!mptcp_pm_addr_families_match(sk, local, &remote)) 438 return 0; 439 440 msk->pm.subflows++; 441 addrs[i++] = remote; 442 } else { 443 mptcp_for_each_subflow(msk, subflow) { 444 ssk = mptcp_subflow_tcp_sock(subflow); 445 remote_address((struct sock_common *)ssk, &addrs[i]); 446 addrs[i].id = subflow->remote_id; 447 if (deny_id0 && !addrs[i].id) 448 continue; 449 450 if (!mptcp_pm_addr_families_match(sk, local, &addrs[i])) 451 continue; 452 453 if (!lookup_address_in_vec(addrs, i, &addrs[i]) && 454 msk->pm.subflows < subflows_max) { 455 msk->pm.subflows++; 456 i++; 457 } 458 } 459 } 460 461 return i; 462 } 463 464 static void __mptcp_pm_send_ack(struct mptcp_sock *msk, struct mptcp_subflow_context *subflow, 465 bool prio, bool backup) 466 { 467 struct sock *ssk = mptcp_subflow_tcp_sock(subflow); 468 bool slow; 469 470 pr_debug("send ack for %s", 471 prio ? "mp_prio" : (mptcp_pm_should_add_signal(msk) ? "add_addr" : "rm_addr")); 472 473 slow = lock_sock_fast(ssk); 474 if (prio) { 475 if (subflow->backup != backup) 476 msk->last_snd = NULL; 477 478 subflow->send_mp_prio = 1; 479 subflow->backup = backup; 480 subflow->request_bkup = backup; 481 } 482 483 __mptcp_subflow_send_ack(ssk); 484 unlock_sock_fast(ssk, slow); 485 } 486 487 static void mptcp_pm_send_ack(struct mptcp_sock *msk, struct mptcp_subflow_context *subflow, 488 bool prio, bool backup) 489 { 490 spin_unlock_bh(&msk->pm.lock); 491 __mptcp_pm_send_ack(msk, subflow, prio, backup); 492 spin_lock_bh(&msk->pm.lock); 493 } 494 495 static struct mptcp_pm_addr_entry * 496 __lookup_addr_by_id(struct pm_nl_pernet *pernet, unsigned int id) 497 { 498 struct mptcp_pm_addr_entry *entry; 499 500 list_for_each_entry(entry, &pernet->local_addr_list, list) { 501 if (entry->addr.id == id) 502 return entry; 503 } 504 return NULL; 505 } 506 507 static struct mptcp_pm_addr_entry * 508 __lookup_addr(struct pm_nl_pernet *pernet, const struct mptcp_addr_info *info, 509 bool lookup_by_id) 510 { 511 struct mptcp_pm_addr_entry *entry; 512 513 list_for_each_entry(entry, &pernet->local_addr_list, list) { 514 if ((!lookup_by_id && 515 mptcp_addresses_equal(&entry->addr, info, entry->addr.port)) || 516 (lookup_by_id && entry->addr.id == info->id)) 517 return entry; 518 } 519 return NULL; 520 } 521 522 static void mptcp_pm_create_subflow_or_signal_addr(struct mptcp_sock *msk) 523 { 524 struct sock *sk = (struct sock *)msk; 525 struct mptcp_pm_addr_entry *local; 526 unsigned int add_addr_signal_max; 527 unsigned int local_addr_max; 528 struct pm_nl_pernet *pernet; 529 unsigned int subflows_max; 530 531 pernet = pm_nl_get_pernet(sock_net(sk)); 532 533 add_addr_signal_max = mptcp_pm_get_add_addr_signal_max(msk); 534 local_addr_max = mptcp_pm_get_local_addr_max(msk); 535 subflows_max = mptcp_pm_get_subflows_max(msk); 536 537 /* do lazy endpoint usage accounting for the MPC subflows */ 538 if (unlikely(!(msk->pm.status & BIT(MPTCP_PM_MPC_ENDPOINT_ACCOUNTED))) && msk->first) { 539 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(msk->first); 540 struct mptcp_pm_addr_entry *entry; 541 struct mptcp_addr_info mpc_addr; 542 bool backup = false; 543 544 local_address((struct sock_common *)msk->first, &mpc_addr); 545 rcu_read_lock(); 546 entry = __lookup_addr(pernet, &mpc_addr, false); 547 if (entry) { 548 __clear_bit(entry->addr.id, msk->pm.id_avail_bitmap); 549 msk->mpc_endpoint_id = entry->addr.id; 550 backup = !!(entry->flags & MPTCP_PM_ADDR_FLAG_BACKUP); 551 } 552 rcu_read_unlock(); 553 554 if (backup) 555 mptcp_pm_send_ack(msk, subflow, true, backup); 556 557 msk->pm.status |= BIT(MPTCP_PM_MPC_ENDPOINT_ACCOUNTED); 558 } 559 560 pr_debug("local %d:%d signal %d:%d subflows %d:%d\n", 561 msk->pm.local_addr_used, local_addr_max, 562 msk->pm.add_addr_signaled, add_addr_signal_max, 563 msk->pm.subflows, subflows_max); 564 565 /* check first for announce */ 566 if (msk->pm.add_addr_signaled < add_addr_signal_max) { 567 local = select_signal_address(pernet, msk); 568 569 /* due to racing events on both ends we can reach here while 570 * previous add address is still running: if we invoke now 571 * mptcp_pm_announce_addr(), that will fail and the 572 * corresponding id will be marked as used. 573 * Instead let the PM machinery reschedule us when the 574 * current address announce will be completed. 575 */ 576 if (msk->pm.addr_signal & BIT(MPTCP_ADD_ADDR_SIGNAL)) 577 return; 578 579 if (local) { 580 if (mptcp_pm_alloc_anno_list(msk, local)) { 581 __clear_bit(local->addr.id, msk->pm.id_avail_bitmap); 582 msk->pm.add_addr_signaled++; 583 mptcp_pm_announce_addr(msk, &local->addr, false); 584 mptcp_pm_nl_addr_send_ack(msk); 585 } 586 } 587 } 588 589 /* check if should create a new subflow */ 590 while (msk->pm.local_addr_used < local_addr_max && 591 msk->pm.subflows < subflows_max) { 592 struct mptcp_addr_info addrs[MPTCP_PM_ADDR_MAX]; 593 bool fullmesh; 594 int i, nr; 595 596 local = select_local_address(pernet, msk); 597 if (!local) 598 break; 599 600 fullmesh = !!(local->flags & MPTCP_PM_ADDR_FLAG_FULLMESH); 601 602 msk->pm.local_addr_used++; 603 __clear_bit(local->addr.id, msk->pm.id_avail_bitmap); 604 nr = fill_remote_addresses_vec(msk, &local->addr, fullmesh, addrs); 605 if (nr == 0) 606 continue; 607 608 spin_unlock_bh(&msk->pm.lock); 609 for (i = 0; i < nr; i++) 610 __mptcp_subflow_connect(sk, &local->addr, &addrs[i]); 611 spin_lock_bh(&msk->pm.lock); 612 } 613 mptcp_pm_nl_check_work_pending(msk); 614 } 615 616 static void mptcp_pm_nl_fully_established(struct mptcp_sock *msk) 617 { 618 mptcp_pm_create_subflow_or_signal_addr(msk); 619 } 620 621 static void mptcp_pm_nl_subflow_established(struct mptcp_sock *msk) 622 { 623 mptcp_pm_create_subflow_or_signal_addr(msk); 624 } 625 626 /* Fill all the local addresses into the array addrs[], 627 * and return the array size. 628 */ 629 static unsigned int fill_local_addresses_vec(struct mptcp_sock *msk, 630 struct mptcp_addr_info *remote, 631 struct mptcp_addr_info *addrs) 632 { 633 struct sock *sk = (struct sock *)msk; 634 struct mptcp_pm_addr_entry *entry; 635 struct pm_nl_pernet *pernet; 636 unsigned int subflows_max; 637 int i = 0; 638 639 pernet = pm_nl_get_pernet_from_msk(msk); 640 subflows_max = mptcp_pm_get_subflows_max(msk); 641 642 rcu_read_lock(); 643 list_for_each_entry_rcu(entry, &pernet->local_addr_list, list) { 644 if (!(entry->flags & MPTCP_PM_ADDR_FLAG_FULLMESH)) 645 continue; 646 647 if (!mptcp_pm_addr_families_match(sk, &entry->addr, remote)) 648 continue; 649 650 if (msk->pm.subflows < subflows_max) { 651 msk->pm.subflows++; 652 addrs[i++] = entry->addr; 653 } 654 } 655 rcu_read_unlock(); 656 657 /* If the array is empty, fill in the single 658 * 'IPADDRANY' local address 659 */ 660 if (!i) { 661 struct mptcp_addr_info local; 662 663 memset(&local, 0, sizeof(local)); 664 local.family = 665 #if IS_ENABLED(CONFIG_MPTCP_IPV6) 666 remote->family == AF_INET6 && 667 ipv6_addr_v4mapped(&remote->addr6) ? AF_INET : 668 #endif 669 remote->family; 670 671 if (!mptcp_pm_addr_families_match(sk, &local, remote)) 672 return 0; 673 674 msk->pm.subflows++; 675 addrs[i++] = local; 676 } 677 678 return i; 679 } 680 681 static void mptcp_pm_nl_add_addr_received(struct mptcp_sock *msk) 682 { 683 struct mptcp_addr_info addrs[MPTCP_PM_ADDR_MAX]; 684 struct sock *sk = (struct sock *)msk; 685 unsigned int add_addr_accept_max; 686 struct mptcp_addr_info remote; 687 unsigned int subflows_max; 688 int i, nr; 689 690 add_addr_accept_max = mptcp_pm_get_add_addr_accept_max(msk); 691 subflows_max = mptcp_pm_get_subflows_max(msk); 692 693 pr_debug("accepted %d:%d remote family %d", 694 msk->pm.add_addr_accepted, add_addr_accept_max, 695 msk->pm.remote.family); 696 697 remote = msk->pm.remote; 698 mptcp_pm_announce_addr(msk, &remote, true); 699 mptcp_pm_nl_addr_send_ack(msk); 700 701 if (lookup_subflow_by_daddr(&msk->conn_list, &remote)) 702 return; 703 704 /* pick id 0 port, if none is provided the remote address */ 705 if (!remote.port) 706 remote.port = sk->sk_dport; 707 708 /* connect to the specified remote address, using whatever 709 * local address the routing configuration will pick. 710 */ 711 nr = fill_local_addresses_vec(msk, &remote, addrs); 712 if (nr == 0) 713 return; 714 715 msk->pm.add_addr_accepted++; 716 if (msk->pm.add_addr_accepted >= add_addr_accept_max || 717 msk->pm.subflows >= subflows_max) 718 WRITE_ONCE(msk->pm.accept_addr, false); 719 720 spin_unlock_bh(&msk->pm.lock); 721 for (i = 0; i < nr; i++) 722 __mptcp_subflow_connect(sk, &addrs[i], &remote); 723 spin_lock_bh(&msk->pm.lock); 724 } 725 726 void mptcp_pm_nl_addr_send_ack(struct mptcp_sock *msk) 727 { 728 struct mptcp_subflow_context *subflow; 729 730 msk_owned_by_me(msk); 731 lockdep_assert_held(&msk->pm.lock); 732 733 if (!mptcp_pm_should_add_signal(msk) && 734 !mptcp_pm_should_rm_signal(msk)) 735 return; 736 737 subflow = list_first_entry_or_null(&msk->conn_list, typeof(*subflow), node); 738 if (subflow) 739 mptcp_pm_send_ack(msk, subflow, false, false); 740 } 741 742 int mptcp_pm_nl_mp_prio_send_ack(struct mptcp_sock *msk, 743 struct mptcp_addr_info *addr, 744 struct mptcp_addr_info *rem, 745 u8 bkup) 746 { 747 struct mptcp_subflow_context *subflow; 748 749 pr_debug("bkup=%d", bkup); 750 751 mptcp_for_each_subflow(msk, subflow) { 752 struct sock *ssk = mptcp_subflow_tcp_sock(subflow); 753 struct mptcp_addr_info local, remote; 754 755 local_address((struct sock_common *)ssk, &local); 756 if (!mptcp_addresses_equal(&local, addr, addr->port)) 757 continue; 758 759 if (rem && rem->family != AF_UNSPEC) { 760 remote_address((struct sock_common *)ssk, &remote); 761 if (!mptcp_addresses_equal(&remote, rem, rem->port)) 762 continue; 763 } 764 765 __mptcp_pm_send_ack(msk, subflow, true, bkup); 766 return 0; 767 } 768 769 return -EINVAL; 770 } 771 772 static bool mptcp_local_id_match(const struct mptcp_sock *msk, u8 local_id, u8 id) 773 { 774 return local_id == id || (!local_id && msk->mpc_endpoint_id == id); 775 } 776 777 static void mptcp_pm_nl_rm_addr_or_subflow(struct mptcp_sock *msk, 778 const struct mptcp_rm_list *rm_list, 779 enum linux_mptcp_mib_field rm_type) 780 { 781 struct mptcp_subflow_context *subflow, *tmp; 782 struct sock *sk = (struct sock *)msk; 783 u8 i; 784 785 pr_debug("%s rm_list_nr %d", 786 rm_type == MPTCP_MIB_RMADDR ? "address" : "subflow", rm_list->nr); 787 788 msk_owned_by_me(msk); 789 790 if (sk->sk_state == TCP_LISTEN) 791 return; 792 793 if (!rm_list->nr) 794 return; 795 796 if (list_empty(&msk->conn_list)) 797 return; 798 799 for (i = 0; i < rm_list->nr; i++) { 800 u8 rm_id = rm_list->ids[i]; 801 bool removed = false; 802 803 mptcp_for_each_subflow_safe(msk, subflow, tmp) { 804 struct sock *ssk = mptcp_subflow_tcp_sock(subflow); 805 int how = RCV_SHUTDOWN | SEND_SHUTDOWN; 806 u8 id = subflow->local_id; 807 808 if (rm_type == MPTCP_MIB_RMADDR && subflow->remote_id != rm_id) 809 continue; 810 if (rm_type == MPTCP_MIB_RMSUBFLOW && !mptcp_local_id_match(msk, id, rm_id)) 811 continue; 812 813 pr_debug(" -> %s rm_list_ids[%d]=%u local_id=%u remote_id=%u mpc_id=%u", 814 rm_type == MPTCP_MIB_RMADDR ? "address" : "subflow", 815 i, rm_id, subflow->local_id, subflow->remote_id, 816 msk->mpc_endpoint_id); 817 spin_unlock_bh(&msk->pm.lock); 818 mptcp_subflow_shutdown(sk, ssk, how); 819 820 /* the following takes care of updating the subflows counter */ 821 mptcp_close_ssk(sk, ssk, subflow); 822 spin_lock_bh(&msk->pm.lock); 823 824 removed = true; 825 __MPTCP_INC_STATS(sock_net(sk), rm_type); 826 } 827 if (rm_type == MPTCP_MIB_RMSUBFLOW) 828 __set_bit(rm_id ? rm_id : msk->mpc_endpoint_id, msk->pm.id_avail_bitmap); 829 if (!removed) 830 continue; 831 832 if (!mptcp_pm_is_kernel(msk)) 833 continue; 834 835 if (rm_type == MPTCP_MIB_RMADDR) { 836 msk->pm.add_addr_accepted--; 837 WRITE_ONCE(msk->pm.accept_addr, true); 838 } else if (rm_type == MPTCP_MIB_RMSUBFLOW) { 839 msk->pm.local_addr_used--; 840 } 841 } 842 } 843 844 static void mptcp_pm_nl_rm_addr_received(struct mptcp_sock *msk) 845 { 846 mptcp_pm_nl_rm_addr_or_subflow(msk, &msk->pm.rm_list_rx, MPTCP_MIB_RMADDR); 847 } 848 849 void mptcp_pm_nl_rm_subflow_received(struct mptcp_sock *msk, 850 const struct mptcp_rm_list *rm_list) 851 { 852 mptcp_pm_nl_rm_addr_or_subflow(msk, rm_list, MPTCP_MIB_RMSUBFLOW); 853 } 854 855 void mptcp_pm_nl_work(struct mptcp_sock *msk) 856 { 857 struct mptcp_pm_data *pm = &msk->pm; 858 859 msk_owned_by_me(msk); 860 861 if (!(pm->status & MPTCP_PM_WORK_MASK)) 862 return; 863 864 spin_lock_bh(&msk->pm.lock); 865 866 pr_debug("msk=%p status=%x", msk, pm->status); 867 if (pm->status & BIT(MPTCP_PM_ADD_ADDR_RECEIVED)) { 868 pm->status &= ~BIT(MPTCP_PM_ADD_ADDR_RECEIVED); 869 mptcp_pm_nl_add_addr_received(msk); 870 } 871 if (pm->status & BIT(MPTCP_PM_ADD_ADDR_SEND_ACK)) { 872 pm->status &= ~BIT(MPTCP_PM_ADD_ADDR_SEND_ACK); 873 mptcp_pm_nl_addr_send_ack(msk); 874 } 875 if (pm->status & BIT(MPTCP_PM_RM_ADDR_RECEIVED)) { 876 pm->status &= ~BIT(MPTCP_PM_RM_ADDR_RECEIVED); 877 mptcp_pm_nl_rm_addr_received(msk); 878 } 879 if (pm->status & BIT(MPTCP_PM_ESTABLISHED)) { 880 pm->status &= ~BIT(MPTCP_PM_ESTABLISHED); 881 mptcp_pm_nl_fully_established(msk); 882 } 883 if (pm->status & BIT(MPTCP_PM_SUBFLOW_ESTABLISHED)) { 884 pm->status &= ~BIT(MPTCP_PM_SUBFLOW_ESTABLISHED); 885 mptcp_pm_nl_subflow_established(msk); 886 } 887 888 spin_unlock_bh(&msk->pm.lock); 889 } 890 891 static bool address_use_port(struct mptcp_pm_addr_entry *entry) 892 { 893 return (entry->flags & 894 (MPTCP_PM_ADDR_FLAG_SIGNAL | MPTCP_PM_ADDR_FLAG_SUBFLOW)) == 895 MPTCP_PM_ADDR_FLAG_SIGNAL; 896 } 897 898 /* caller must ensure the RCU grace period is already elapsed */ 899 static void __mptcp_pm_release_addr_entry(struct mptcp_pm_addr_entry *entry) 900 { 901 if (entry->lsk) 902 sock_release(entry->lsk); 903 kfree(entry); 904 } 905 906 static int mptcp_pm_nl_append_new_local_addr(struct pm_nl_pernet *pernet, 907 struct mptcp_pm_addr_entry *entry) 908 { 909 struct mptcp_pm_addr_entry *cur, *del_entry = NULL; 910 unsigned int addr_max; 911 int ret = -EINVAL; 912 913 spin_lock_bh(&pernet->lock); 914 /* to keep the code simple, don't do IDR-like allocation for address ID, 915 * just bail when we exceed limits 916 */ 917 if (pernet->next_id == MPTCP_PM_MAX_ADDR_ID) 918 pernet->next_id = 1; 919 if (pernet->addrs >= MPTCP_PM_ADDR_MAX) { 920 ret = -ERANGE; 921 goto out; 922 } 923 if (test_bit(entry->addr.id, pernet->id_bitmap)) { 924 ret = -EBUSY; 925 goto out; 926 } 927 928 /* do not insert duplicate address, differentiate on port only 929 * singled addresses 930 */ 931 if (!address_use_port(entry)) 932 entry->addr.port = 0; 933 list_for_each_entry(cur, &pernet->local_addr_list, list) { 934 if (mptcp_addresses_equal(&cur->addr, &entry->addr, 935 cur->addr.port || entry->addr.port)) { 936 /* allow replacing the exiting endpoint only if such 937 * endpoint is an implicit one and the user-space 938 * did not provide an endpoint id 939 */ 940 if (!(cur->flags & MPTCP_PM_ADDR_FLAG_IMPLICIT)) { 941 ret = -EEXIST; 942 goto out; 943 } 944 if (entry->addr.id) 945 goto out; 946 947 pernet->addrs--; 948 entry->addr.id = cur->addr.id; 949 list_del_rcu(&cur->list); 950 del_entry = cur; 951 break; 952 } 953 } 954 955 if (!entry->addr.id) { 956 find_next: 957 entry->addr.id = find_next_zero_bit(pernet->id_bitmap, 958 MPTCP_PM_MAX_ADDR_ID + 1, 959 pernet->next_id); 960 if (!entry->addr.id && pernet->next_id != 1) { 961 pernet->next_id = 1; 962 goto find_next; 963 } 964 } 965 966 if (!entry->addr.id) 967 goto out; 968 969 __set_bit(entry->addr.id, pernet->id_bitmap); 970 if (entry->addr.id > pernet->next_id) 971 pernet->next_id = entry->addr.id; 972 973 if (entry->flags & MPTCP_PM_ADDR_FLAG_SIGNAL) { 974 addr_max = pernet->add_addr_signal_max; 975 WRITE_ONCE(pernet->add_addr_signal_max, addr_max + 1); 976 } 977 if (entry->flags & MPTCP_PM_ADDR_FLAG_SUBFLOW) { 978 addr_max = pernet->local_addr_max; 979 WRITE_ONCE(pernet->local_addr_max, addr_max + 1); 980 } 981 982 pernet->addrs++; 983 if (!entry->addr.port) 984 list_add_tail_rcu(&entry->list, &pernet->local_addr_list); 985 else 986 list_add_rcu(&entry->list, &pernet->local_addr_list); 987 ret = entry->addr.id; 988 989 out: 990 spin_unlock_bh(&pernet->lock); 991 992 /* just replaced an existing entry, free it */ 993 if (del_entry) { 994 synchronize_rcu(); 995 __mptcp_pm_release_addr_entry(del_entry); 996 } 997 return ret; 998 } 999 1000 static int mptcp_pm_nl_create_listen_socket(struct sock *sk, 1001 struct mptcp_pm_addr_entry *entry) 1002 { 1003 int addrlen = sizeof(struct sockaddr_in); 1004 struct sockaddr_storage addr; 1005 struct socket *ssock; 1006 struct sock *newsk; 1007 int backlog = 1024; 1008 int err; 1009 1010 err = sock_create_kern(sock_net(sk), entry->addr.family, 1011 SOCK_STREAM, IPPROTO_MPTCP, &entry->lsk); 1012 if (err) 1013 return err; 1014 1015 newsk = entry->lsk->sk; 1016 if (!newsk) 1017 return -EINVAL; 1018 1019 lock_sock(newsk); 1020 ssock = __mptcp_nmpc_socket(mptcp_sk(newsk)); 1021 release_sock(newsk); 1022 if (!ssock) 1023 return -EINVAL; 1024 1025 mptcp_info2sockaddr(&entry->addr, &addr, entry->addr.family); 1026 #if IS_ENABLED(CONFIG_MPTCP_IPV6) 1027 if (entry->addr.family == AF_INET6) 1028 addrlen = sizeof(struct sockaddr_in6); 1029 #endif 1030 err = kernel_bind(ssock, (struct sockaddr *)&addr, addrlen); 1031 if (err) 1032 return err; 1033 1034 err = kernel_listen(ssock, backlog); 1035 if (err) 1036 return err; 1037 1038 mptcp_event_pm_listener(ssock->sk, MPTCP_EVENT_LISTENER_CREATED); 1039 1040 return 0; 1041 } 1042 1043 int mptcp_pm_nl_get_local_id(struct mptcp_sock *msk, struct sock_common *skc) 1044 { 1045 struct mptcp_pm_addr_entry *entry; 1046 struct mptcp_addr_info skc_local; 1047 struct mptcp_addr_info msk_local; 1048 struct pm_nl_pernet *pernet; 1049 int ret = -1; 1050 1051 if (WARN_ON_ONCE(!msk)) 1052 return -1; 1053 1054 /* The 0 ID mapping is defined by the first subflow, copied into the msk 1055 * addr 1056 */ 1057 local_address((struct sock_common *)msk, &msk_local); 1058 local_address((struct sock_common *)skc, &skc_local); 1059 if (mptcp_addresses_equal(&msk_local, &skc_local, false)) 1060 return 0; 1061 1062 if (mptcp_pm_is_userspace(msk)) 1063 return mptcp_userspace_pm_get_local_id(msk, &skc_local); 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_local, 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_local; 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_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 *flags = 0; 1368 *ifindex = 0; 1369 1370 if (id) { 1371 if (mptcp_pm_is_userspace(msk)) 1372 return mptcp_userspace_pm_get_flags_and_ifindex_by_id(msk, 1373 id, 1374 flags, 1375 ifindex); 1376 1377 rcu_read_lock(); 1378 entry = __lookup_addr_by_id(pm_nl_get_pernet(net), id); 1379 if (entry) { 1380 *flags = entry->flags; 1381 *ifindex = entry->ifindex; 1382 } 1383 rcu_read_unlock(); 1384 } 1385 1386 return 0; 1387 } 1388 1389 static bool remove_anno_list_by_saddr(struct mptcp_sock *msk, 1390 const struct mptcp_addr_info *addr) 1391 { 1392 struct mptcp_pm_add_entry *entry; 1393 1394 entry = mptcp_pm_del_add_timer(msk, addr, false); 1395 if (entry) { 1396 list_del(&entry->list); 1397 kfree(entry); 1398 return true; 1399 } 1400 1401 return false; 1402 } 1403 1404 static bool mptcp_pm_remove_anno_addr(struct mptcp_sock *msk, 1405 const struct mptcp_addr_info *addr, 1406 bool force) 1407 { 1408 struct mptcp_rm_list list = { .nr = 0 }; 1409 bool ret; 1410 1411 list.ids[list.nr++] = addr->id; 1412 1413 ret = remove_anno_list_by_saddr(msk, addr); 1414 if (ret || force) { 1415 spin_lock_bh(&msk->pm.lock); 1416 mptcp_pm_remove_addr(msk, &list); 1417 spin_unlock_bh(&msk->pm.lock); 1418 } 1419 return ret; 1420 } 1421 1422 static int mptcp_nl_remove_subflow_and_signal_addr(struct net *net, 1423 const struct mptcp_pm_addr_entry *entry) 1424 { 1425 const struct mptcp_addr_info *addr = &entry->addr; 1426 struct mptcp_rm_list list = { .nr = 0 }; 1427 long s_slot = 0, s_num = 0; 1428 struct mptcp_sock *msk; 1429 1430 pr_debug("remove_id=%d", addr->id); 1431 1432 list.ids[list.nr++] = addr->id; 1433 1434 while ((msk = mptcp_token_iter_next(net, &s_slot, &s_num)) != NULL) { 1435 struct sock *sk = (struct sock *)msk; 1436 bool remove_subflow; 1437 1438 if (mptcp_pm_is_userspace(msk)) 1439 goto next; 1440 1441 if (list_empty(&msk->conn_list)) { 1442 mptcp_pm_remove_anno_addr(msk, addr, false); 1443 goto next; 1444 } 1445 1446 lock_sock(sk); 1447 remove_subflow = lookup_subflow_by_saddr(&msk->conn_list, addr); 1448 mptcp_pm_remove_anno_addr(msk, addr, remove_subflow && 1449 !(entry->flags & MPTCP_PM_ADDR_FLAG_IMPLICIT)); 1450 if (remove_subflow) 1451 mptcp_pm_remove_subflow(msk, &list); 1452 release_sock(sk); 1453 1454 next: 1455 sock_put(sk); 1456 cond_resched(); 1457 } 1458 1459 return 0; 1460 } 1461 1462 static int mptcp_nl_remove_id_zero_address(struct net *net, 1463 struct mptcp_addr_info *addr) 1464 { 1465 struct mptcp_rm_list list = { .nr = 0 }; 1466 long s_slot = 0, s_num = 0; 1467 struct mptcp_sock *msk; 1468 1469 list.ids[list.nr++] = 0; 1470 1471 while ((msk = mptcp_token_iter_next(net, &s_slot, &s_num)) != NULL) { 1472 struct sock *sk = (struct sock *)msk; 1473 struct mptcp_addr_info msk_local; 1474 1475 if (list_empty(&msk->conn_list) || mptcp_pm_is_userspace(msk)) 1476 goto next; 1477 1478 local_address((struct sock_common *)msk, &msk_local); 1479 if (!mptcp_addresses_equal(&msk_local, addr, addr->port)) 1480 goto next; 1481 1482 lock_sock(sk); 1483 spin_lock_bh(&msk->pm.lock); 1484 mptcp_pm_remove_addr(msk, &list); 1485 mptcp_pm_nl_rm_subflow_received(msk, &list); 1486 spin_unlock_bh(&msk->pm.lock); 1487 release_sock(sk); 1488 1489 next: 1490 sock_put(sk); 1491 cond_resched(); 1492 } 1493 1494 return 0; 1495 } 1496 1497 static int mptcp_nl_cmd_del_addr(struct sk_buff *skb, struct genl_info *info) 1498 { 1499 struct nlattr *attr = info->attrs[MPTCP_PM_ATTR_ADDR]; 1500 struct pm_nl_pernet *pernet = genl_info_pm_nl(info); 1501 struct mptcp_pm_addr_entry addr, *entry; 1502 unsigned int addr_max; 1503 int ret; 1504 1505 ret = mptcp_pm_parse_entry(attr, info, false, &addr); 1506 if (ret < 0) 1507 return ret; 1508 1509 /* the zero id address is special: the first address used by the msk 1510 * always gets such an id, so different subflows can have different zero 1511 * id addresses. Additionally zero id is not accounted for in id_bitmap. 1512 * Let's use an 'mptcp_rm_list' instead of the common remove code. 1513 */ 1514 if (addr.addr.id == 0) 1515 return mptcp_nl_remove_id_zero_address(sock_net(skb->sk), &addr.addr); 1516 1517 spin_lock_bh(&pernet->lock); 1518 entry = __lookup_addr_by_id(pernet, addr.addr.id); 1519 if (!entry) { 1520 GENL_SET_ERR_MSG(info, "address not found"); 1521 spin_unlock_bh(&pernet->lock); 1522 return -EINVAL; 1523 } 1524 if (entry->flags & MPTCP_PM_ADDR_FLAG_SIGNAL) { 1525 addr_max = pernet->add_addr_signal_max; 1526 WRITE_ONCE(pernet->add_addr_signal_max, addr_max - 1); 1527 } 1528 if (entry->flags & MPTCP_PM_ADDR_FLAG_SUBFLOW) { 1529 addr_max = pernet->local_addr_max; 1530 WRITE_ONCE(pernet->local_addr_max, addr_max - 1); 1531 } 1532 1533 pernet->addrs--; 1534 list_del_rcu(&entry->list); 1535 __clear_bit(entry->addr.id, pernet->id_bitmap); 1536 spin_unlock_bh(&pernet->lock); 1537 1538 mptcp_nl_remove_subflow_and_signal_addr(sock_net(skb->sk), entry); 1539 synchronize_rcu(); 1540 __mptcp_pm_release_addr_entry(entry); 1541 1542 return ret; 1543 } 1544 1545 void mptcp_pm_remove_addrs_and_subflows(struct mptcp_sock *msk, 1546 struct list_head *rm_list) 1547 { 1548 struct mptcp_rm_list alist = { .nr = 0 }, slist = { .nr = 0 }; 1549 struct mptcp_pm_addr_entry *entry; 1550 1551 list_for_each_entry(entry, rm_list, list) { 1552 if (lookup_subflow_by_saddr(&msk->conn_list, &entry->addr) && 1553 slist.nr < MPTCP_RM_IDS_MAX) 1554 slist.ids[slist.nr++] = entry->addr.id; 1555 1556 if (remove_anno_list_by_saddr(msk, &entry->addr) && 1557 alist.nr < MPTCP_RM_IDS_MAX) 1558 alist.ids[alist.nr++] = entry->addr.id; 1559 } 1560 1561 if (alist.nr) { 1562 spin_lock_bh(&msk->pm.lock); 1563 mptcp_pm_remove_addr(msk, &alist); 1564 spin_unlock_bh(&msk->pm.lock); 1565 } 1566 if (slist.nr) 1567 mptcp_pm_remove_subflow(msk, &slist); 1568 } 1569 1570 static void mptcp_nl_remove_addrs_list(struct net *net, 1571 struct list_head *rm_list) 1572 { 1573 long s_slot = 0, s_num = 0; 1574 struct mptcp_sock *msk; 1575 1576 if (list_empty(rm_list)) 1577 return; 1578 1579 while ((msk = mptcp_token_iter_next(net, &s_slot, &s_num)) != NULL) { 1580 struct sock *sk = (struct sock *)msk; 1581 1582 if (!mptcp_pm_is_userspace(msk)) { 1583 lock_sock(sk); 1584 mptcp_pm_remove_addrs_and_subflows(msk, rm_list); 1585 release_sock(sk); 1586 } 1587 1588 sock_put(sk); 1589 cond_resched(); 1590 } 1591 } 1592 1593 /* caller must ensure the RCU grace period is already elapsed */ 1594 static void __flush_addrs(struct list_head *list) 1595 { 1596 while (!list_empty(list)) { 1597 struct mptcp_pm_addr_entry *cur; 1598 1599 cur = list_entry(list->next, 1600 struct mptcp_pm_addr_entry, list); 1601 list_del_rcu(&cur->list); 1602 __mptcp_pm_release_addr_entry(cur); 1603 } 1604 } 1605 1606 static void __reset_counters(struct pm_nl_pernet *pernet) 1607 { 1608 WRITE_ONCE(pernet->add_addr_signal_max, 0); 1609 WRITE_ONCE(pernet->add_addr_accept_max, 0); 1610 WRITE_ONCE(pernet->local_addr_max, 0); 1611 pernet->addrs = 0; 1612 } 1613 1614 static int mptcp_nl_cmd_flush_addrs(struct sk_buff *skb, struct genl_info *info) 1615 { 1616 struct pm_nl_pernet *pernet = genl_info_pm_nl(info); 1617 LIST_HEAD(free_list); 1618 1619 spin_lock_bh(&pernet->lock); 1620 list_splice_init(&pernet->local_addr_list, &free_list); 1621 __reset_counters(pernet); 1622 pernet->next_id = 1; 1623 bitmap_zero(pernet->id_bitmap, MPTCP_PM_MAX_ADDR_ID + 1); 1624 spin_unlock_bh(&pernet->lock); 1625 mptcp_nl_remove_addrs_list(sock_net(skb->sk), &free_list); 1626 synchronize_rcu(); 1627 __flush_addrs(&free_list); 1628 return 0; 1629 } 1630 1631 static int mptcp_nl_fill_addr(struct sk_buff *skb, 1632 struct mptcp_pm_addr_entry *entry) 1633 { 1634 struct mptcp_addr_info *addr = &entry->addr; 1635 struct nlattr *attr; 1636 1637 attr = nla_nest_start(skb, MPTCP_PM_ATTR_ADDR); 1638 if (!attr) 1639 return -EMSGSIZE; 1640 1641 if (nla_put_u16(skb, MPTCP_PM_ADDR_ATTR_FAMILY, addr->family)) 1642 goto nla_put_failure; 1643 if (nla_put_u16(skb, MPTCP_PM_ADDR_ATTR_PORT, ntohs(addr->port))) 1644 goto nla_put_failure; 1645 if (nla_put_u8(skb, MPTCP_PM_ADDR_ATTR_ID, addr->id)) 1646 goto nla_put_failure; 1647 if (nla_put_u32(skb, MPTCP_PM_ADDR_ATTR_FLAGS, entry->flags)) 1648 goto nla_put_failure; 1649 if (entry->ifindex && 1650 nla_put_s32(skb, MPTCP_PM_ADDR_ATTR_IF_IDX, entry->ifindex)) 1651 goto nla_put_failure; 1652 1653 if (addr->family == AF_INET && 1654 nla_put_in_addr(skb, MPTCP_PM_ADDR_ATTR_ADDR4, 1655 addr->addr.s_addr)) 1656 goto nla_put_failure; 1657 #if IS_ENABLED(CONFIG_MPTCP_IPV6) 1658 else if (addr->family == AF_INET6 && 1659 nla_put_in6_addr(skb, MPTCP_PM_ADDR_ATTR_ADDR6, &addr->addr6)) 1660 goto nla_put_failure; 1661 #endif 1662 nla_nest_end(skb, attr); 1663 return 0; 1664 1665 nla_put_failure: 1666 nla_nest_cancel(skb, attr); 1667 return -EMSGSIZE; 1668 } 1669 1670 static int mptcp_nl_cmd_get_addr(struct sk_buff *skb, struct genl_info *info) 1671 { 1672 struct nlattr *attr = info->attrs[MPTCP_PM_ATTR_ADDR]; 1673 struct pm_nl_pernet *pernet = genl_info_pm_nl(info); 1674 struct mptcp_pm_addr_entry addr, *entry; 1675 struct sk_buff *msg; 1676 void *reply; 1677 int ret; 1678 1679 ret = mptcp_pm_parse_entry(attr, info, false, &addr); 1680 if (ret < 0) 1681 return ret; 1682 1683 msg = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL); 1684 if (!msg) 1685 return -ENOMEM; 1686 1687 reply = genlmsg_put_reply(msg, info, &mptcp_genl_family, 0, 1688 info->genlhdr->cmd); 1689 if (!reply) { 1690 GENL_SET_ERR_MSG(info, "not enough space in Netlink message"); 1691 ret = -EMSGSIZE; 1692 goto fail; 1693 } 1694 1695 spin_lock_bh(&pernet->lock); 1696 entry = __lookup_addr_by_id(pernet, addr.addr.id); 1697 if (!entry) { 1698 GENL_SET_ERR_MSG(info, "address not found"); 1699 ret = -EINVAL; 1700 goto unlock_fail; 1701 } 1702 1703 ret = mptcp_nl_fill_addr(msg, entry); 1704 if (ret) 1705 goto unlock_fail; 1706 1707 genlmsg_end(msg, reply); 1708 ret = genlmsg_reply(msg, info); 1709 spin_unlock_bh(&pernet->lock); 1710 return ret; 1711 1712 unlock_fail: 1713 spin_unlock_bh(&pernet->lock); 1714 1715 fail: 1716 nlmsg_free(msg); 1717 return ret; 1718 } 1719 1720 static int mptcp_nl_cmd_dump_addrs(struct sk_buff *msg, 1721 struct netlink_callback *cb) 1722 { 1723 struct net *net = sock_net(msg->sk); 1724 struct mptcp_pm_addr_entry *entry; 1725 struct pm_nl_pernet *pernet; 1726 int id = cb->args[0]; 1727 void *hdr; 1728 int i; 1729 1730 pernet = pm_nl_get_pernet(net); 1731 1732 spin_lock_bh(&pernet->lock); 1733 for (i = id; i < MPTCP_PM_MAX_ADDR_ID + 1; i++) { 1734 if (test_bit(i, pernet->id_bitmap)) { 1735 entry = __lookup_addr_by_id(pernet, i); 1736 if (!entry) 1737 break; 1738 1739 if (entry->addr.id <= id) 1740 continue; 1741 1742 hdr = genlmsg_put(msg, NETLINK_CB(cb->skb).portid, 1743 cb->nlh->nlmsg_seq, &mptcp_genl_family, 1744 NLM_F_MULTI, MPTCP_PM_CMD_GET_ADDR); 1745 if (!hdr) 1746 break; 1747 1748 if (mptcp_nl_fill_addr(msg, entry) < 0) { 1749 genlmsg_cancel(msg, hdr); 1750 break; 1751 } 1752 1753 id = entry->addr.id; 1754 genlmsg_end(msg, hdr); 1755 } 1756 } 1757 spin_unlock_bh(&pernet->lock); 1758 1759 cb->args[0] = id; 1760 return msg->len; 1761 } 1762 1763 static int parse_limit(struct genl_info *info, int id, unsigned int *limit) 1764 { 1765 struct nlattr *attr = info->attrs[id]; 1766 1767 if (!attr) 1768 return 0; 1769 1770 *limit = nla_get_u32(attr); 1771 if (*limit > MPTCP_PM_ADDR_MAX) { 1772 GENL_SET_ERR_MSG(info, "limit greater than maximum"); 1773 return -EINVAL; 1774 } 1775 return 0; 1776 } 1777 1778 static int 1779 mptcp_nl_cmd_set_limits(struct sk_buff *skb, struct genl_info *info) 1780 { 1781 struct pm_nl_pernet *pernet = genl_info_pm_nl(info); 1782 unsigned int rcv_addrs, subflows; 1783 int ret; 1784 1785 spin_lock_bh(&pernet->lock); 1786 rcv_addrs = pernet->add_addr_accept_max; 1787 ret = parse_limit(info, MPTCP_PM_ATTR_RCV_ADD_ADDRS, &rcv_addrs); 1788 if (ret) 1789 goto unlock; 1790 1791 subflows = pernet->subflows_max; 1792 ret = parse_limit(info, MPTCP_PM_ATTR_SUBFLOWS, &subflows); 1793 if (ret) 1794 goto unlock; 1795 1796 WRITE_ONCE(pernet->add_addr_accept_max, rcv_addrs); 1797 WRITE_ONCE(pernet->subflows_max, subflows); 1798 1799 unlock: 1800 spin_unlock_bh(&pernet->lock); 1801 return ret; 1802 } 1803 1804 static int 1805 mptcp_nl_cmd_get_limits(struct sk_buff *skb, struct genl_info *info) 1806 { 1807 struct pm_nl_pernet *pernet = genl_info_pm_nl(info); 1808 struct sk_buff *msg; 1809 void *reply; 1810 1811 msg = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL); 1812 if (!msg) 1813 return -ENOMEM; 1814 1815 reply = genlmsg_put_reply(msg, info, &mptcp_genl_family, 0, 1816 MPTCP_PM_CMD_GET_LIMITS); 1817 if (!reply) 1818 goto fail; 1819 1820 if (nla_put_u32(msg, MPTCP_PM_ATTR_RCV_ADD_ADDRS, 1821 READ_ONCE(pernet->add_addr_accept_max))) 1822 goto fail; 1823 1824 if (nla_put_u32(msg, MPTCP_PM_ATTR_SUBFLOWS, 1825 READ_ONCE(pernet->subflows_max))) 1826 goto fail; 1827 1828 genlmsg_end(msg, reply); 1829 return genlmsg_reply(msg, info); 1830 1831 fail: 1832 GENL_SET_ERR_MSG(info, "not enough space in Netlink message"); 1833 nlmsg_free(msg); 1834 return -EMSGSIZE; 1835 } 1836 1837 static void mptcp_pm_nl_fullmesh(struct mptcp_sock *msk, 1838 struct mptcp_addr_info *addr) 1839 { 1840 struct mptcp_rm_list list = { .nr = 0 }; 1841 1842 list.ids[list.nr++] = addr->id; 1843 1844 spin_lock_bh(&msk->pm.lock); 1845 mptcp_pm_nl_rm_subflow_received(msk, &list); 1846 mptcp_pm_create_subflow_or_signal_addr(msk); 1847 spin_unlock_bh(&msk->pm.lock); 1848 } 1849 1850 static int mptcp_nl_set_flags(struct net *net, 1851 struct mptcp_addr_info *addr, 1852 u8 bkup, u8 changed) 1853 { 1854 long s_slot = 0, s_num = 0; 1855 struct mptcp_sock *msk; 1856 int ret = -EINVAL; 1857 1858 while ((msk = mptcp_token_iter_next(net, &s_slot, &s_num)) != NULL) { 1859 struct sock *sk = (struct sock *)msk; 1860 1861 if (list_empty(&msk->conn_list) || mptcp_pm_is_userspace(msk)) 1862 goto next; 1863 1864 lock_sock(sk); 1865 if (changed & MPTCP_PM_ADDR_FLAG_BACKUP) 1866 ret = mptcp_pm_nl_mp_prio_send_ack(msk, addr, NULL, bkup); 1867 if (changed & MPTCP_PM_ADDR_FLAG_FULLMESH) 1868 mptcp_pm_nl_fullmesh(msk, addr); 1869 release_sock(sk); 1870 1871 next: 1872 sock_put(sk); 1873 cond_resched(); 1874 } 1875 1876 return ret; 1877 } 1878 1879 static int mptcp_nl_cmd_set_flags(struct sk_buff *skb, struct genl_info *info) 1880 { 1881 struct mptcp_pm_addr_entry addr = { .addr = { .family = AF_UNSPEC }, }, *entry; 1882 struct mptcp_pm_addr_entry remote = { .addr = { .family = AF_UNSPEC }, }; 1883 struct nlattr *attr_rem = info->attrs[MPTCP_PM_ATTR_ADDR_REMOTE]; 1884 struct nlattr *token = info->attrs[MPTCP_PM_ATTR_TOKEN]; 1885 struct nlattr *attr = info->attrs[MPTCP_PM_ATTR_ADDR]; 1886 struct pm_nl_pernet *pernet = genl_info_pm_nl(info); 1887 u8 changed, mask = MPTCP_PM_ADDR_FLAG_BACKUP | 1888 MPTCP_PM_ADDR_FLAG_FULLMESH; 1889 struct net *net = sock_net(skb->sk); 1890 u8 bkup = 0, lookup_by_id = 0; 1891 int ret; 1892 1893 ret = mptcp_pm_parse_entry(attr, info, false, &addr); 1894 if (ret < 0) 1895 return ret; 1896 1897 if (attr_rem) { 1898 ret = mptcp_pm_parse_entry(attr_rem, info, false, &remote); 1899 if (ret < 0) 1900 return ret; 1901 } 1902 1903 if (addr.flags & MPTCP_PM_ADDR_FLAG_BACKUP) 1904 bkup = 1; 1905 if (addr.addr.family == AF_UNSPEC) { 1906 lookup_by_id = 1; 1907 if (!addr.addr.id) 1908 return -EOPNOTSUPP; 1909 } 1910 1911 if (token) 1912 return mptcp_userspace_pm_set_flags(net, token, &addr, &remote, bkup); 1913 1914 spin_lock_bh(&pernet->lock); 1915 entry = __lookup_addr(pernet, &addr.addr, lookup_by_id); 1916 if (!entry) { 1917 spin_unlock_bh(&pernet->lock); 1918 return -EINVAL; 1919 } 1920 if ((addr.flags & MPTCP_PM_ADDR_FLAG_FULLMESH) && 1921 (entry->flags & MPTCP_PM_ADDR_FLAG_SIGNAL)) { 1922 spin_unlock_bh(&pernet->lock); 1923 return -EINVAL; 1924 } 1925 1926 changed = (addr.flags ^ entry->flags) & mask; 1927 entry->flags = (entry->flags & ~mask) | (addr.flags & mask); 1928 addr = *entry; 1929 spin_unlock_bh(&pernet->lock); 1930 1931 mptcp_nl_set_flags(net, &addr.addr, bkup, changed); 1932 return 0; 1933 } 1934 1935 static void mptcp_nl_mcast_send(struct net *net, struct sk_buff *nlskb, gfp_t gfp) 1936 { 1937 genlmsg_multicast_netns(&mptcp_genl_family, net, 1938 nlskb, 0, MPTCP_PM_EV_GRP_OFFSET, gfp); 1939 } 1940 1941 bool mptcp_userspace_pm_active(const struct mptcp_sock *msk) 1942 { 1943 return genl_has_listeners(&mptcp_genl_family, 1944 sock_net((const struct sock *)msk), 1945 MPTCP_PM_EV_GRP_OFFSET); 1946 } 1947 1948 static int mptcp_event_add_subflow(struct sk_buff *skb, const struct sock *ssk) 1949 { 1950 const struct inet_sock *issk = inet_sk(ssk); 1951 const struct mptcp_subflow_context *sf; 1952 1953 if (nla_put_u16(skb, MPTCP_ATTR_FAMILY, ssk->sk_family)) 1954 return -EMSGSIZE; 1955 1956 switch (ssk->sk_family) { 1957 case AF_INET: 1958 if (nla_put_in_addr(skb, MPTCP_ATTR_SADDR4, issk->inet_saddr)) 1959 return -EMSGSIZE; 1960 if (nla_put_in_addr(skb, MPTCP_ATTR_DADDR4, issk->inet_daddr)) 1961 return -EMSGSIZE; 1962 break; 1963 #if IS_ENABLED(CONFIG_MPTCP_IPV6) 1964 case AF_INET6: { 1965 const struct ipv6_pinfo *np = inet6_sk(ssk); 1966 1967 if (nla_put_in6_addr(skb, MPTCP_ATTR_SADDR6, &np->saddr)) 1968 return -EMSGSIZE; 1969 if (nla_put_in6_addr(skb, MPTCP_ATTR_DADDR6, &ssk->sk_v6_daddr)) 1970 return -EMSGSIZE; 1971 break; 1972 } 1973 #endif 1974 default: 1975 WARN_ON_ONCE(1); 1976 return -EMSGSIZE; 1977 } 1978 1979 if (nla_put_be16(skb, MPTCP_ATTR_SPORT, issk->inet_sport)) 1980 return -EMSGSIZE; 1981 if (nla_put_be16(skb, MPTCP_ATTR_DPORT, issk->inet_dport)) 1982 return -EMSGSIZE; 1983 1984 sf = mptcp_subflow_ctx(ssk); 1985 if (WARN_ON_ONCE(!sf)) 1986 return -EINVAL; 1987 1988 if (nla_put_u8(skb, MPTCP_ATTR_LOC_ID, sf->local_id)) 1989 return -EMSGSIZE; 1990 1991 if (nla_put_u8(skb, MPTCP_ATTR_REM_ID, sf->remote_id)) 1992 return -EMSGSIZE; 1993 1994 return 0; 1995 } 1996 1997 static int mptcp_event_put_token_and_ssk(struct sk_buff *skb, 1998 const struct mptcp_sock *msk, 1999 const struct sock *ssk) 2000 { 2001 const struct sock *sk = (const struct sock *)msk; 2002 const struct mptcp_subflow_context *sf; 2003 u8 sk_err; 2004 2005 if (nla_put_u32(skb, MPTCP_ATTR_TOKEN, msk->token)) 2006 return -EMSGSIZE; 2007 2008 if (mptcp_event_add_subflow(skb, ssk)) 2009 return -EMSGSIZE; 2010 2011 sf = mptcp_subflow_ctx(ssk); 2012 if (WARN_ON_ONCE(!sf)) 2013 return -EINVAL; 2014 2015 if (nla_put_u8(skb, MPTCP_ATTR_BACKUP, sf->backup)) 2016 return -EMSGSIZE; 2017 2018 if (ssk->sk_bound_dev_if && 2019 nla_put_s32(skb, MPTCP_ATTR_IF_IDX, ssk->sk_bound_dev_if)) 2020 return -EMSGSIZE; 2021 2022 sk_err = ssk->sk_err; 2023 if (sk_err && sk->sk_state == TCP_ESTABLISHED && 2024 nla_put_u8(skb, MPTCP_ATTR_ERROR, sk_err)) 2025 return -EMSGSIZE; 2026 2027 return 0; 2028 } 2029 2030 static int mptcp_event_sub_established(struct sk_buff *skb, 2031 const struct mptcp_sock *msk, 2032 const struct sock *ssk) 2033 { 2034 return mptcp_event_put_token_and_ssk(skb, msk, ssk); 2035 } 2036 2037 static int mptcp_event_sub_closed(struct sk_buff *skb, 2038 const struct mptcp_sock *msk, 2039 const struct sock *ssk) 2040 { 2041 const struct mptcp_subflow_context *sf; 2042 2043 if (mptcp_event_put_token_and_ssk(skb, msk, ssk)) 2044 return -EMSGSIZE; 2045 2046 sf = mptcp_subflow_ctx(ssk); 2047 if (!sf->reset_seen) 2048 return 0; 2049 2050 if (nla_put_u32(skb, MPTCP_ATTR_RESET_REASON, sf->reset_reason)) 2051 return -EMSGSIZE; 2052 2053 if (nla_put_u32(skb, MPTCP_ATTR_RESET_FLAGS, sf->reset_transient)) 2054 return -EMSGSIZE; 2055 2056 return 0; 2057 } 2058 2059 static int mptcp_event_created(struct sk_buff *skb, 2060 const struct mptcp_sock *msk, 2061 const struct sock *ssk) 2062 { 2063 int err = nla_put_u32(skb, MPTCP_ATTR_TOKEN, msk->token); 2064 2065 if (err) 2066 return err; 2067 2068 if (nla_put_u8(skb, MPTCP_ATTR_SERVER_SIDE, READ_ONCE(msk->pm.server_side))) 2069 return -EMSGSIZE; 2070 2071 return mptcp_event_add_subflow(skb, ssk); 2072 } 2073 2074 void mptcp_event_addr_removed(const struct mptcp_sock *msk, uint8_t id) 2075 { 2076 struct net *net = sock_net((const struct sock *)msk); 2077 struct nlmsghdr *nlh; 2078 struct sk_buff *skb; 2079 2080 if (!genl_has_listeners(&mptcp_genl_family, net, MPTCP_PM_EV_GRP_OFFSET)) 2081 return; 2082 2083 skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_ATOMIC); 2084 if (!skb) 2085 return; 2086 2087 nlh = genlmsg_put(skb, 0, 0, &mptcp_genl_family, 0, MPTCP_EVENT_REMOVED); 2088 if (!nlh) 2089 goto nla_put_failure; 2090 2091 if (nla_put_u32(skb, MPTCP_ATTR_TOKEN, msk->token)) 2092 goto nla_put_failure; 2093 2094 if (nla_put_u8(skb, MPTCP_ATTR_REM_ID, id)) 2095 goto nla_put_failure; 2096 2097 genlmsg_end(skb, nlh); 2098 mptcp_nl_mcast_send(net, skb, GFP_ATOMIC); 2099 return; 2100 2101 nla_put_failure: 2102 nlmsg_free(skb); 2103 } 2104 2105 void mptcp_event_addr_announced(const struct sock *ssk, 2106 const struct mptcp_addr_info *info) 2107 { 2108 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk); 2109 struct mptcp_sock *msk = mptcp_sk(subflow->conn); 2110 struct net *net = sock_net(ssk); 2111 struct nlmsghdr *nlh; 2112 struct sk_buff *skb; 2113 2114 if (!genl_has_listeners(&mptcp_genl_family, net, MPTCP_PM_EV_GRP_OFFSET)) 2115 return; 2116 2117 skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_ATOMIC); 2118 if (!skb) 2119 return; 2120 2121 nlh = genlmsg_put(skb, 0, 0, &mptcp_genl_family, 0, 2122 MPTCP_EVENT_ANNOUNCED); 2123 if (!nlh) 2124 goto nla_put_failure; 2125 2126 if (nla_put_u32(skb, MPTCP_ATTR_TOKEN, msk->token)) 2127 goto nla_put_failure; 2128 2129 if (nla_put_u8(skb, MPTCP_ATTR_REM_ID, info->id)) 2130 goto nla_put_failure; 2131 2132 if (nla_put_be16(skb, MPTCP_ATTR_DPORT, 2133 info->port == 0 ? 2134 inet_sk(ssk)->inet_dport : 2135 info->port)) 2136 goto nla_put_failure; 2137 2138 switch (info->family) { 2139 case AF_INET: 2140 if (nla_put_in_addr(skb, MPTCP_ATTR_DADDR4, info->addr.s_addr)) 2141 goto nla_put_failure; 2142 break; 2143 #if IS_ENABLED(CONFIG_MPTCP_IPV6) 2144 case AF_INET6: 2145 if (nla_put_in6_addr(skb, MPTCP_ATTR_DADDR6, &info->addr6)) 2146 goto nla_put_failure; 2147 break; 2148 #endif 2149 default: 2150 WARN_ON_ONCE(1); 2151 goto nla_put_failure; 2152 } 2153 2154 genlmsg_end(skb, nlh); 2155 mptcp_nl_mcast_send(net, skb, GFP_ATOMIC); 2156 return; 2157 2158 nla_put_failure: 2159 nlmsg_free(skb); 2160 } 2161 2162 void mptcp_event_pm_listener(const struct sock *ssk, 2163 enum mptcp_event_type event) 2164 { 2165 const struct inet_sock *issk = inet_sk(ssk); 2166 struct net *net = sock_net(ssk); 2167 struct nlmsghdr *nlh; 2168 struct sk_buff *skb; 2169 2170 if (!genl_has_listeners(&mptcp_genl_family, net, MPTCP_PM_EV_GRP_OFFSET)) 2171 return; 2172 2173 skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL); 2174 if (!skb) 2175 return; 2176 2177 nlh = genlmsg_put(skb, 0, 0, &mptcp_genl_family, 0, event); 2178 if (!nlh) 2179 goto nla_put_failure; 2180 2181 if (nla_put_u16(skb, MPTCP_ATTR_FAMILY, ssk->sk_family)) 2182 goto nla_put_failure; 2183 2184 if (nla_put_be16(skb, MPTCP_ATTR_SPORT, issk->inet_sport)) 2185 goto nla_put_failure; 2186 2187 switch (ssk->sk_family) { 2188 case AF_INET: 2189 if (nla_put_in_addr(skb, MPTCP_ATTR_SADDR4, issk->inet_saddr)) 2190 goto nla_put_failure; 2191 break; 2192 #if IS_ENABLED(CONFIG_MPTCP_IPV6) 2193 case AF_INET6: { 2194 const struct ipv6_pinfo *np = inet6_sk(ssk); 2195 2196 if (nla_put_in6_addr(skb, MPTCP_ATTR_SADDR6, &np->saddr)) 2197 goto nla_put_failure; 2198 break; 2199 } 2200 #endif 2201 default: 2202 WARN_ON_ONCE(1); 2203 goto nla_put_failure; 2204 } 2205 2206 genlmsg_end(skb, nlh); 2207 mptcp_nl_mcast_send(net, skb, GFP_KERNEL); 2208 return; 2209 2210 nla_put_failure: 2211 nlmsg_free(skb); 2212 } 2213 2214 void mptcp_event(enum mptcp_event_type type, const struct mptcp_sock *msk, 2215 const struct sock *ssk, gfp_t gfp) 2216 { 2217 struct net *net = sock_net((const struct sock *)msk); 2218 struct nlmsghdr *nlh; 2219 struct sk_buff *skb; 2220 2221 if (!genl_has_listeners(&mptcp_genl_family, net, MPTCP_PM_EV_GRP_OFFSET)) 2222 return; 2223 2224 skb = nlmsg_new(NLMSG_DEFAULT_SIZE, gfp); 2225 if (!skb) 2226 return; 2227 2228 nlh = genlmsg_put(skb, 0, 0, &mptcp_genl_family, 0, type); 2229 if (!nlh) 2230 goto nla_put_failure; 2231 2232 switch (type) { 2233 case MPTCP_EVENT_UNSPEC: 2234 WARN_ON_ONCE(1); 2235 break; 2236 case MPTCP_EVENT_CREATED: 2237 case MPTCP_EVENT_ESTABLISHED: 2238 if (mptcp_event_created(skb, msk, ssk) < 0) 2239 goto nla_put_failure; 2240 break; 2241 case MPTCP_EVENT_CLOSED: 2242 if (nla_put_u32(skb, MPTCP_ATTR_TOKEN, msk->token) < 0) 2243 goto nla_put_failure; 2244 break; 2245 case MPTCP_EVENT_ANNOUNCED: 2246 case MPTCP_EVENT_REMOVED: 2247 /* call mptcp_event_addr_announced()/removed instead */ 2248 WARN_ON_ONCE(1); 2249 break; 2250 case MPTCP_EVENT_SUB_ESTABLISHED: 2251 case MPTCP_EVENT_SUB_PRIORITY: 2252 if (mptcp_event_sub_established(skb, msk, ssk) < 0) 2253 goto nla_put_failure; 2254 break; 2255 case MPTCP_EVENT_SUB_CLOSED: 2256 if (mptcp_event_sub_closed(skb, msk, ssk) < 0) 2257 goto nla_put_failure; 2258 break; 2259 case MPTCP_EVENT_LISTENER_CREATED: 2260 case MPTCP_EVENT_LISTENER_CLOSED: 2261 break; 2262 } 2263 2264 genlmsg_end(skb, nlh); 2265 mptcp_nl_mcast_send(net, skb, gfp); 2266 return; 2267 2268 nla_put_failure: 2269 nlmsg_free(skb); 2270 } 2271 2272 static const struct genl_small_ops mptcp_pm_ops[] = { 2273 { 2274 .cmd = MPTCP_PM_CMD_ADD_ADDR, 2275 .doit = mptcp_nl_cmd_add_addr, 2276 .flags = GENL_UNS_ADMIN_PERM, 2277 }, 2278 { 2279 .cmd = MPTCP_PM_CMD_DEL_ADDR, 2280 .doit = mptcp_nl_cmd_del_addr, 2281 .flags = GENL_UNS_ADMIN_PERM, 2282 }, 2283 { 2284 .cmd = MPTCP_PM_CMD_FLUSH_ADDRS, 2285 .doit = mptcp_nl_cmd_flush_addrs, 2286 .flags = GENL_UNS_ADMIN_PERM, 2287 }, 2288 { 2289 .cmd = MPTCP_PM_CMD_GET_ADDR, 2290 .doit = mptcp_nl_cmd_get_addr, 2291 .dumpit = mptcp_nl_cmd_dump_addrs, 2292 }, 2293 { 2294 .cmd = MPTCP_PM_CMD_SET_LIMITS, 2295 .doit = mptcp_nl_cmd_set_limits, 2296 .flags = GENL_UNS_ADMIN_PERM, 2297 }, 2298 { 2299 .cmd = MPTCP_PM_CMD_GET_LIMITS, 2300 .doit = mptcp_nl_cmd_get_limits, 2301 }, 2302 { 2303 .cmd = MPTCP_PM_CMD_SET_FLAGS, 2304 .doit = mptcp_nl_cmd_set_flags, 2305 .flags = GENL_UNS_ADMIN_PERM, 2306 }, 2307 { 2308 .cmd = MPTCP_PM_CMD_ANNOUNCE, 2309 .doit = mptcp_nl_cmd_announce, 2310 .flags = GENL_UNS_ADMIN_PERM, 2311 }, 2312 { 2313 .cmd = MPTCP_PM_CMD_REMOVE, 2314 .doit = mptcp_nl_cmd_remove, 2315 .flags = GENL_UNS_ADMIN_PERM, 2316 }, 2317 { 2318 .cmd = MPTCP_PM_CMD_SUBFLOW_CREATE, 2319 .doit = mptcp_nl_cmd_sf_create, 2320 .flags = GENL_UNS_ADMIN_PERM, 2321 }, 2322 { 2323 .cmd = MPTCP_PM_CMD_SUBFLOW_DESTROY, 2324 .doit = mptcp_nl_cmd_sf_destroy, 2325 .flags = GENL_UNS_ADMIN_PERM, 2326 }, 2327 }; 2328 2329 static struct genl_family mptcp_genl_family __ro_after_init = { 2330 .name = MPTCP_PM_NAME, 2331 .version = MPTCP_PM_VER, 2332 .maxattr = MPTCP_PM_ATTR_MAX, 2333 .policy = mptcp_pm_policy, 2334 .netnsok = true, 2335 .module = THIS_MODULE, 2336 .small_ops = mptcp_pm_ops, 2337 .n_small_ops = ARRAY_SIZE(mptcp_pm_ops), 2338 .resv_start_op = MPTCP_PM_CMD_SUBFLOW_DESTROY + 1, 2339 .mcgrps = mptcp_pm_mcgrps, 2340 .n_mcgrps = ARRAY_SIZE(mptcp_pm_mcgrps), 2341 }; 2342 2343 static int __net_init pm_nl_init_net(struct net *net) 2344 { 2345 struct pm_nl_pernet *pernet = pm_nl_get_pernet(net); 2346 2347 INIT_LIST_HEAD_RCU(&pernet->local_addr_list); 2348 2349 /* Cit. 2 subflows ought to be enough for anybody. */ 2350 pernet->subflows_max = 2; 2351 pernet->next_id = 1; 2352 pernet->stale_loss_cnt = 4; 2353 spin_lock_init(&pernet->lock); 2354 2355 /* No need to initialize other pernet fields, the struct is zeroed at 2356 * allocation time. 2357 */ 2358 2359 return 0; 2360 } 2361 2362 static void __net_exit pm_nl_exit_net(struct list_head *net_list) 2363 { 2364 struct net *net; 2365 2366 list_for_each_entry(net, net_list, exit_list) { 2367 struct pm_nl_pernet *pernet = pm_nl_get_pernet(net); 2368 2369 /* net is removed from namespace list, can't race with 2370 * other modifiers, also netns core already waited for a 2371 * RCU grace period. 2372 */ 2373 __flush_addrs(&pernet->local_addr_list); 2374 } 2375 } 2376 2377 static struct pernet_operations mptcp_pm_pernet_ops = { 2378 .init = pm_nl_init_net, 2379 .exit_batch = pm_nl_exit_net, 2380 .id = &pm_nl_pernet_id, 2381 .size = sizeof(struct pm_nl_pernet), 2382 }; 2383 2384 void __init mptcp_pm_nl_init(void) 2385 { 2386 if (register_pernet_subsys(&mptcp_pm_pernet_ops) < 0) 2387 panic("Failed to register MPTCP PM pernet subsystem.\n"); 2388 2389 if (genl_register_family(&mptcp_genl_family)) 2390 panic("Failed to register MPTCP PM netlink family\n"); 2391 } 2392