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/inet_common.h> 12 #include <net/netns/generic.h> 13 #include <net/mptcp.h> 14 15 #include "protocol.h" 16 #include "mib.h" 17 #include "mptcp_pm_gen.h" 18 19 static int pm_nl_pernet_id; 20 21 struct mptcp_pm_add_entry { 22 struct list_head list; 23 struct mptcp_addr_info addr; 24 u8 retrans_times; 25 struct timer_list add_timer; 26 struct mptcp_sock *sock; 27 }; 28 29 struct pm_nl_pernet { 30 /* protects pernet updates */ 31 spinlock_t lock; 32 struct list_head local_addr_list; 33 unsigned int addrs; 34 unsigned int stale_loss_cnt; 35 unsigned int add_addr_signal_max; 36 unsigned int add_addr_accept_max; 37 unsigned int local_addr_max; 38 unsigned int subflows_max; 39 unsigned int next_id; 40 DECLARE_BITMAP(id_bitmap, MPTCP_PM_MAX_ADDR_ID + 1); 41 }; 42 43 #define MPTCP_PM_ADDR_MAX 8 44 #define ADD_ADDR_RETRANS_MAX 3 45 46 static struct pm_nl_pernet *pm_nl_get_pernet(const struct net *net) 47 { 48 return net_generic(net, pm_nl_pernet_id); 49 } 50 51 static struct pm_nl_pernet * 52 pm_nl_get_pernet_from_msk(const struct mptcp_sock *msk) 53 { 54 return pm_nl_get_pernet(sock_net((struct sock *)msk)); 55 } 56 57 bool mptcp_addresses_equal(const struct mptcp_addr_info *a, 58 const struct mptcp_addr_info *b, bool use_port) 59 { 60 bool addr_equals = false; 61 62 if (a->family == b->family) { 63 if (a->family == AF_INET) 64 addr_equals = a->addr.s_addr == b->addr.s_addr; 65 #if IS_ENABLED(CONFIG_MPTCP_IPV6) 66 else 67 addr_equals = !ipv6_addr_cmp(&a->addr6, &b->addr6); 68 } else if (a->family == AF_INET) { 69 if (ipv6_addr_v4mapped(&b->addr6)) 70 addr_equals = a->addr.s_addr == b->addr6.s6_addr32[3]; 71 } else if (b->family == AF_INET) { 72 if (ipv6_addr_v4mapped(&a->addr6)) 73 addr_equals = a->addr6.s6_addr32[3] == b->addr.s_addr; 74 #endif 75 } 76 77 if (!addr_equals) 78 return false; 79 if (!use_port) 80 return true; 81 82 return a->port == b->port; 83 } 84 85 void mptcp_local_address(const struct sock_common *skc, struct mptcp_addr_info *addr) 86 { 87 addr->family = skc->skc_family; 88 addr->port = htons(skc->skc_num); 89 if (addr->family == AF_INET) 90 addr->addr.s_addr = skc->skc_rcv_saddr; 91 #if IS_ENABLED(CONFIG_MPTCP_IPV6) 92 else if (addr->family == AF_INET6) 93 addr->addr6 = skc->skc_v6_rcv_saddr; 94 #endif 95 } 96 97 static void remote_address(const struct sock_common *skc, 98 struct mptcp_addr_info *addr) 99 { 100 addr->family = skc->skc_family; 101 addr->port = skc->skc_dport; 102 if (addr->family == AF_INET) 103 addr->addr.s_addr = skc->skc_daddr; 104 #if IS_ENABLED(CONFIG_MPTCP_IPV6) 105 else if (addr->family == AF_INET6) 106 addr->addr6 = skc->skc_v6_daddr; 107 #endif 108 } 109 110 static bool lookup_subflow_by_saddr(const struct list_head *list, 111 const struct mptcp_addr_info *saddr) 112 { 113 struct mptcp_subflow_context *subflow; 114 struct mptcp_addr_info cur; 115 struct sock_common *skc; 116 117 list_for_each_entry(subflow, list, node) { 118 skc = (struct sock_common *)mptcp_subflow_tcp_sock(subflow); 119 120 mptcp_local_address(skc, &cur); 121 if (mptcp_addresses_equal(&cur, saddr, saddr->port)) 122 return true; 123 } 124 125 return false; 126 } 127 128 static bool lookup_subflow_by_daddr(const struct list_head *list, 129 const struct mptcp_addr_info *daddr) 130 { 131 struct mptcp_subflow_context *subflow; 132 struct mptcp_addr_info cur; 133 struct sock_common *skc; 134 135 list_for_each_entry(subflow, list, node) { 136 skc = (struct sock_common *)mptcp_subflow_tcp_sock(subflow); 137 138 remote_address(skc, &cur); 139 if (mptcp_addresses_equal(&cur, daddr, daddr->port)) 140 return true; 141 } 142 143 return false; 144 } 145 146 static struct mptcp_pm_addr_entry * 147 select_local_address(const struct pm_nl_pernet *pernet, 148 const struct mptcp_sock *msk) 149 { 150 struct mptcp_pm_addr_entry *entry, *ret = NULL; 151 152 msk_owned_by_me(msk); 153 154 rcu_read_lock(); 155 list_for_each_entry_rcu(entry, &pernet->local_addr_list, list) { 156 if (!(entry->flags & MPTCP_PM_ADDR_FLAG_SUBFLOW)) 157 continue; 158 159 if (!test_bit(entry->addr.id, msk->pm.id_avail_bitmap)) 160 continue; 161 162 ret = entry; 163 break; 164 } 165 rcu_read_unlock(); 166 return ret; 167 } 168 169 static struct mptcp_pm_addr_entry * 170 select_signal_address(struct pm_nl_pernet *pernet, const struct mptcp_sock *msk) 171 { 172 struct mptcp_pm_addr_entry *entry, *ret = NULL; 173 174 rcu_read_lock(); 175 /* do not keep any additional per socket state, just signal 176 * the address list in order. 177 * Note: removal from the local address list during the msk life-cycle 178 * can lead to additional addresses not being announced. 179 */ 180 list_for_each_entry_rcu(entry, &pernet->local_addr_list, list) { 181 if (!test_bit(entry->addr.id, msk->pm.id_avail_bitmap)) 182 continue; 183 184 if (!(entry->flags & MPTCP_PM_ADDR_FLAG_SIGNAL)) 185 continue; 186 187 ret = entry; 188 break; 189 } 190 rcu_read_unlock(); 191 return ret; 192 } 193 194 unsigned int mptcp_pm_get_add_addr_signal_max(const struct mptcp_sock *msk) 195 { 196 const struct pm_nl_pernet *pernet = pm_nl_get_pernet_from_msk(msk); 197 198 return READ_ONCE(pernet->add_addr_signal_max); 199 } 200 EXPORT_SYMBOL_GPL(mptcp_pm_get_add_addr_signal_max); 201 202 unsigned int mptcp_pm_get_add_addr_accept_max(const struct mptcp_sock *msk) 203 { 204 struct pm_nl_pernet *pernet = pm_nl_get_pernet_from_msk(msk); 205 206 return READ_ONCE(pernet->add_addr_accept_max); 207 } 208 EXPORT_SYMBOL_GPL(mptcp_pm_get_add_addr_accept_max); 209 210 unsigned int mptcp_pm_get_subflows_max(const struct mptcp_sock *msk) 211 { 212 struct pm_nl_pernet *pernet = pm_nl_get_pernet_from_msk(msk); 213 214 return READ_ONCE(pernet->subflows_max); 215 } 216 EXPORT_SYMBOL_GPL(mptcp_pm_get_subflows_max); 217 218 unsigned int mptcp_pm_get_local_addr_max(const struct mptcp_sock *msk) 219 { 220 struct pm_nl_pernet *pernet = pm_nl_get_pernet_from_msk(msk); 221 222 return READ_ONCE(pernet->local_addr_max); 223 } 224 EXPORT_SYMBOL_GPL(mptcp_pm_get_local_addr_max); 225 226 bool mptcp_pm_nl_check_work_pending(struct mptcp_sock *msk) 227 { 228 struct pm_nl_pernet *pernet = pm_nl_get_pernet_from_msk(msk); 229 230 if (msk->pm.subflows == mptcp_pm_get_subflows_max(msk) || 231 (find_next_and_bit(pernet->id_bitmap, msk->pm.id_avail_bitmap, 232 MPTCP_PM_MAX_ADDR_ID + 1, 0) == MPTCP_PM_MAX_ADDR_ID + 1)) { 233 WRITE_ONCE(msk->pm.work_pending, false); 234 return false; 235 } 236 return true; 237 } 238 239 struct mptcp_pm_add_entry * 240 mptcp_lookup_anno_list_by_saddr(const struct mptcp_sock *msk, 241 const struct mptcp_addr_info *addr) 242 { 243 struct mptcp_pm_add_entry *entry; 244 245 lockdep_assert_held(&msk->pm.lock); 246 247 list_for_each_entry(entry, &msk->pm.anno_list, list) { 248 if (mptcp_addresses_equal(&entry->addr, addr, true)) 249 return entry; 250 } 251 252 return NULL; 253 } 254 255 bool mptcp_pm_sport_in_anno_list(struct mptcp_sock *msk, const struct sock *sk) 256 { 257 struct mptcp_pm_add_entry *entry; 258 struct mptcp_addr_info saddr; 259 bool ret = false; 260 261 mptcp_local_address((struct sock_common *)sk, &saddr); 262 263 spin_lock_bh(&msk->pm.lock); 264 list_for_each_entry(entry, &msk->pm.anno_list, list) { 265 if (mptcp_addresses_equal(&entry->addr, &saddr, true)) { 266 ret = true; 267 goto out; 268 } 269 } 270 271 out: 272 spin_unlock_bh(&msk->pm.lock); 273 return ret; 274 } 275 276 static void mptcp_pm_add_timer(struct timer_list *timer) 277 { 278 struct mptcp_pm_add_entry *entry = from_timer(entry, timer, add_timer); 279 struct mptcp_sock *msk = entry->sock; 280 struct sock *sk = (struct sock *)msk; 281 282 pr_debug("msk=%p", msk); 283 284 if (!msk) 285 return; 286 287 if (inet_sk_state_load(sk) == TCP_CLOSE) 288 return; 289 290 if (!entry->addr.id) 291 return; 292 293 if (mptcp_pm_should_add_signal_addr(msk)) { 294 sk_reset_timer(sk, timer, jiffies + TCP_RTO_MAX / 8); 295 goto out; 296 } 297 298 spin_lock_bh(&msk->pm.lock); 299 300 if (!mptcp_pm_should_add_signal_addr(msk)) { 301 pr_debug("retransmit ADD_ADDR id=%d", entry->addr.id); 302 mptcp_pm_announce_addr(msk, &entry->addr, false); 303 mptcp_pm_add_addr_send_ack(msk); 304 entry->retrans_times++; 305 } 306 307 if (entry->retrans_times < ADD_ADDR_RETRANS_MAX) 308 sk_reset_timer(sk, timer, 309 jiffies + mptcp_get_add_addr_timeout(sock_net(sk))); 310 311 spin_unlock_bh(&msk->pm.lock); 312 313 if (entry->retrans_times == ADD_ADDR_RETRANS_MAX) 314 mptcp_pm_subflow_established(msk); 315 316 out: 317 __sock_put(sk); 318 } 319 320 struct mptcp_pm_add_entry * 321 mptcp_pm_del_add_timer(struct mptcp_sock *msk, 322 const struct mptcp_addr_info *addr, bool check_id) 323 { 324 struct mptcp_pm_add_entry *entry; 325 struct sock *sk = (struct sock *)msk; 326 327 spin_lock_bh(&msk->pm.lock); 328 entry = mptcp_lookup_anno_list_by_saddr(msk, addr); 329 if (entry && (!check_id || entry->addr.id == addr->id)) 330 entry->retrans_times = ADD_ADDR_RETRANS_MAX; 331 spin_unlock_bh(&msk->pm.lock); 332 333 if (entry && (!check_id || entry->addr.id == addr->id)) 334 sk_stop_timer_sync(sk, &entry->add_timer); 335 336 return entry; 337 } 338 339 bool mptcp_pm_alloc_anno_list(struct mptcp_sock *msk, 340 const struct mptcp_addr_info *addr) 341 { 342 struct mptcp_pm_add_entry *add_entry = NULL; 343 struct sock *sk = (struct sock *)msk; 344 struct net *net = sock_net(sk); 345 346 lockdep_assert_held(&msk->pm.lock); 347 348 add_entry = mptcp_lookup_anno_list_by_saddr(msk, addr); 349 350 if (add_entry) { 351 if (mptcp_pm_is_kernel(msk)) 352 return false; 353 354 sk_reset_timer(sk, &add_entry->add_timer, 355 jiffies + mptcp_get_add_addr_timeout(net)); 356 return true; 357 } 358 359 add_entry = kmalloc(sizeof(*add_entry), GFP_ATOMIC); 360 if (!add_entry) 361 return false; 362 363 list_add(&add_entry->list, &msk->pm.anno_list); 364 365 add_entry->addr = *addr; 366 add_entry->sock = msk; 367 add_entry->retrans_times = 0; 368 369 timer_setup(&add_entry->add_timer, mptcp_pm_add_timer, 0); 370 sk_reset_timer(sk, &add_entry->add_timer, 371 jiffies + mptcp_get_add_addr_timeout(net)); 372 373 return true; 374 } 375 376 void mptcp_pm_free_anno_list(struct mptcp_sock *msk) 377 { 378 struct mptcp_pm_add_entry *entry, *tmp; 379 struct sock *sk = (struct sock *)msk; 380 LIST_HEAD(free_list); 381 382 pr_debug("msk=%p", msk); 383 384 spin_lock_bh(&msk->pm.lock); 385 list_splice_init(&msk->pm.anno_list, &free_list); 386 spin_unlock_bh(&msk->pm.lock); 387 388 list_for_each_entry_safe(entry, tmp, &free_list, list) { 389 sk_stop_timer_sync(sk, &entry->add_timer); 390 kfree(entry); 391 } 392 } 393 394 /* Fill all the remote addresses into the array addrs[], 395 * and return the array size. 396 */ 397 static unsigned int fill_remote_addresses_vec(struct mptcp_sock *msk, 398 struct mptcp_addr_info *local, 399 bool fullmesh, 400 struct mptcp_addr_info *addrs) 401 { 402 bool deny_id0 = READ_ONCE(msk->pm.remote_deny_join_id0); 403 struct sock *sk = (struct sock *)msk, *ssk; 404 struct mptcp_subflow_context *subflow; 405 struct mptcp_addr_info remote = { 0 }; 406 unsigned int subflows_max; 407 int i = 0; 408 409 subflows_max = mptcp_pm_get_subflows_max(msk); 410 remote_address((struct sock_common *)sk, &remote); 411 412 /* Non-fullmesh endpoint, fill in the single entry 413 * corresponding to the primary MPC subflow remote address 414 */ 415 if (!fullmesh) { 416 if (deny_id0) 417 return 0; 418 419 if (!mptcp_pm_addr_families_match(sk, local, &remote)) 420 return 0; 421 422 msk->pm.subflows++; 423 addrs[i++] = remote; 424 } else { 425 DECLARE_BITMAP(unavail_id, MPTCP_PM_MAX_ADDR_ID + 1); 426 427 /* Forbid creation of new subflows matching existing 428 * ones, possibly already created by incoming ADD_ADDR 429 */ 430 bitmap_zero(unavail_id, MPTCP_PM_MAX_ADDR_ID + 1); 431 mptcp_for_each_subflow(msk, subflow) 432 if (READ_ONCE(subflow->local_id) == local->id) 433 __set_bit(subflow->remote_id, unavail_id); 434 435 mptcp_for_each_subflow(msk, subflow) { 436 ssk = mptcp_subflow_tcp_sock(subflow); 437 remote_address((struct sock_common *)ssk, &addrs[i]); 438 addrs[i].id = READ_ONCE(subflow->remote_id); 439 if (deny_id0 && !addrs[i].id) 440 continue; 441 442 if (test_bit(addrs[i].id, unavail_id)) 443 continue; 444 445 if (!mptcp_pm_addr_families_match(sk, local, &addrs[i])) 446 continue; 447 448 if (msk->pm.subflows < subflows_max) { 449 /* forbid creating multiple address towards 450 * this id 451 */ 452 __set_bit(addrs[i].id, unavail_id); 453 msk->pm.subflows++; 454 i++; 455 } 456 } 457 } 458 459 return i; 460 } 461 462 static void __mptcp_pm_send_ack(struct mptcp_sock *msk, struct mptcp_subflow_context *subflow, 463 bool prio, bool backup) 464 { 465 struct sock *ssk = mptcp_subflow_tcp_sock(subflow); 466 bool slow; 467 468 pr_debug("send ack for %s", 469 prio ? "mp_prio" : (mptcp_pm_should_add_signal(msk) ? "add_addr" : "rm_addr")); 470 471 slow = lock_sock_fast(ssk); 472 if (prio) { 473 subflow->send_mp_prio = 1; 474 subflow->backup = backup; 475 subflow->request_bkup = backup; 476 } 477 478 __mptcp_subflow_send_ack(ssk); 479 unlock_sock_fast(ssk, slow); 480 } 481 482 static void mptcp_pm_send_ack(struct mptcp_sock *msk, struct mptcp_subflow_context *subflow, 483 bool prio, bool backup) 484 { 485 spin_unlock_bh(&msk->pm.lock); 486 __mptcp_pm_send_ack(msk, subflow, prio, backup); 487 spin_lock_bh(&msk->pm.lock); 488 } 489 490 static struct mptcp_pm_addr_entry * 491 __lookup_addr_by_id(struct pm_nl_pernet *pernet, unsigned int id) 492 { 493 struct mptcp_pm_addr_entry *entry; 494 495 list_for_each_entry(entry, &pernet->local_addr_list, list) { 496 if (entry->addr.id == id) 497 return entry; 498 } 499 return NULL; 500 } 501 502 static struct mptcp_pm_addr_entry * 503 __lookup_addr(struct pm_nl_pernet *pernet, const struct mptcp_addr_info *info) 504 { 505 struct mptcp_pm_addr_entry *entry; 506 507 list_for_each_entry(entry, &pernet->local_addr_list, list) { 508 if (mptcp_addresses_equal(&entry->addr, info, entry->addr.port)) 509 return entry; 510 } 511 return NULL; 512 } 513 514 static void mptcp_pm_create_subflow_or_signal_addr(struct mptcp_sock *msk) 515 { 516 struct sock *sk = (struct sock *)msk; 517 struct mptcp_pm_addr_entry *local; 518 unsigned int add_addr_signal_max; 519 unsigned int local_addr_max; 520 struct pm_nl_pernet *pernet; 521 unsigned int subflows_max; 522 523 pernet = pm_nl_get_pernet(sock_net(sk)); 524 525 add_addr_signal_max = mptcp_pm_get_add_addr_signal_max(msk); 526 local_addr_max = mptcp_pm_get_local_addr_max(msk); 527 subflows_max = mptcp_pm_get_subflows_max(msk); 528 529 /* do lazy endpoint usage accounting for the MPC subflows */ 530 if (unlikely(!(msk->pm.status & BIT(MPTCP_PM_MPC_ENDPOINT_ACCOUNTED))) && msk->first) { 531 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(msk->first); 532 struct mptcp_pm_addr_entry *entry; 533 struct mptcp_addr_info mpc_addr; 534 bool backup = false; 535 536 mptcp_local_address((struct sock_common *)msk->first, &mpc_addr); 537 rcu_read_lock(); 538 entry = __lookup_addr(pernet, &mpc_addr); 539 if (entry) { 540 __clear_bit(entry->addr.id, msk->pm.id_avail_bitmap); 541 msk->mpc_endpoint_id = entry->addr.id; 542 backup = !!(entry->flags & MPTCP_PM_ADDR_FLAG_BACKUP); 543 } 544 rcu_read_unlock(); 545 546 if (backup) 547 mptcp_pm_send_ack(msk, subflow, true, backup); 548 549 msk->pm.status |= BIT(MPTCP_PM_MPC_ENDPOINT_ACCOUNTED); 550 } 551 552 pr_debug("local %d:%d signal %d:%d subflows %d:%d\n", 553 msk->pm.local_addr_used, local_addr_max, 554 msk->pm.add_addr_signaled, add_addr_signal_max, 555 msk->pm.subflows, subflows_max); 556 557 /* check first for announce */ 558 if (msk->pm.add_addr_signaled < add_addr_signal_max) { 559 local = select_signal_address(pernet, msk); 560 561 /* due to racing events on both ends we can reach here while 562 * previous add address is still running: if we invoke now 563 * mptcp_pm_announce_addr(), that will fail and the 564 * corresponding id will be marked as used. 565 * Instead let the PM machinery reschedule us when the 566 * current address announce will be completed. 567 */ 568 if (msk->pm.addr_signal & BIT(MPTCP_ADD_ADDR_SIGNAL)) 569 return; 570 571 if (local) { 572 if (mptcp_pm_alloc_anno_list(msk, &local->addr)) { 573 __clear_bit(local->addr.id, msk->pm.id_avail_bitmap); 574 msk->pm.add_addr_signaled++; 575 mptcp_pm_announce_addr(msk, &local->addr, false); 576 mptcp_pm_nl_addr_send_ack(msk); 577 } 578 } 579 } 580 581 /* check if should create a new subflow */ 582 while (msk->pm.local_addr_used < local_addr_max && 583 msk->pm.subflows < subflows_max) { 584 struct mptcp_addr_info addrs[MPTCP_PM_ADDR_MAX]; 585 bool fullmesh; 586 int i, nr; 587 588 local = select_local_address(pernet, msk); 589 if (!local) 590 break; 591 592 fullmesh = !!(local->flags & MPTCP_PM_ADDR_FLAG_FULLMESH); 593 594 msk->pm.local_addr_used++; 595 __clear_bit(local->addr.id, msk->pm.id_avail_bitmap); 596 nr = fill_remote_addresses_vec(msk, &local->addr, fullmesh, addrs); 597 if (nr == 0) 598 continue; 599 600 spin_unlock_bh(&msk->pm.lock); 601 for (i = 0; i < nr; i++) 602 __mptcp_subflow_connect(sk, &local->addr, &addrs[i]); 603 spin_lock_bh(&msk->pm.lock); 604 } 605 mptcp_pm_nl_check_work_pending(msk); 606 } 607 608 static void mptcp_pm_nl_fully_established(struct mptcp_sock *msk) 609 { 610 mptcp_pm_create_subflow_or_signal_addr(msk); 611 } 612 613 static void mptcp_pm_nl_subflow_established(struct mptcp_sock *msk) 614 { 615 mptcp_pm_create_subflow_or_signal_addr(msk); 616 } 617 618 /* Fill all the local addresses into the array addrs[], 619 * and return the array size. 620 */ 621 static unsigned int fill_local_addresses_vec(struct mptcp_sock *msk, 622 struct mptcp_addr_info *remote, 623 struct mptcp_addr_info *addrs) 624 { 625 struct sock *sk = (struct sock *)msk; 626 struct mptcp_pm_addr_entry *entry; 627 struct pm_nl_pernet *pernet; 628 unsigned int subflows_max; 629 int i = 0; 630 631 pernet = pm_nl_get_pernet_from_msk(msk); 632 subflows_max = mptcp_pm_get_subflows_max(msk); 633 634 rcu_read_lock(); 635 list_for_each_entry_rcu(entry, &pernet->local_addr_list, list) { 636 if (!(entry->flags & MPTCP_PM_ADDR_FLAG_FULLMESH)) 637 continue; 638 639 if (!mptcp_pm_addr_families_match(sk, &entry->addr, remote)) 640 continue; 641 642 if (msk->pm.subflows < subflows_max) { 643 msk->pm.subflows++; 644 addrs[i++] = entry->addr; 645 } 646 } 647 rcu_read_unlock(); 648 649 /* If the array is empty, fill in the single 650 * 'IPADDRANY' local address 651 */ 652 if (!i) { 653 struct mptcp_addr_info local; 654 655 memset(&local, 0, sizeof(local)); 656 local.family = 657 #if IS_ENABLED(CONFIG_MPTCP_IPV6) 658 remote->family == AF_INET6 && 659 ipv6_addr_v4mapped(&remote->addr6) ? AF_INET : 660 #endif 661 remote->family; 662 663 if (!mptcp_pm_addr_families_match(sk, &local, remote)) 664 return 0; 665 666 msk->pm.subflows++; 667 addrs[i++] = local; 668 } 669 670 return i; 671 } 672 673 static void mptcp_pm_nl_add_addr_received(struct mptcp_sock *msk) 674 { 675 struct mptcp_addr_info addrs[MPTCP_PM_ADDR_MAX]; 676 struct sock *sk = (struct sock *)msk; 677 unsigned int add_addr_accept_max; 678 struct mptcp_addr_info remote; 679 unsigned int subflows_max; 680 int i, nr; 681 682 add_addr_accept_max = mptcp_pm_get_add_addr_accept_max(msk); 683 subflows_max = mptcp_pm_get_subflows_max(msk); 684 685 pr_debug("accepted %d:%d remote family %d", 686 msk->pm.add_addr_accepted, add_addr_accept_max, 687 msk->pm.remote.family); 688 689 remote = msk->pm.remote; 690 mptcp_pm_announce_addr(msk, &remote, true); 691 mptcp_pm_nl_addr_send_ack(msk); 692 693 if (lookup_subflow_by_daddr(&msk->conn_list, &remote)) 694 return; 695 696 /* pick id 0 port, if none is provided the remote address */ 697 if (!remote.port) 698 remote.port = sk->sk_dport; 699 700 /* connect to the specified remote address, using whatever 701 * local address the routing configuration will pick. 702 */ 703 nr = fill_local_addresses_vec(msk, &remote, addrs); 704 if (nr == 0) 705 return; 706 707 msk->pm.add_addr_accepted++; 708 if (msk->pm.add_addr_accepted >= add_addr_accept_max || 709 msk->pm.subflows >= subflows_max) 710 WRITE_ONCE(msk->pm.accept_addr, false); 711 712 spin_unlock_bh(&msk->pm.lock); 713 for (i = 0; i < nr; i++) 714 __mptcp_subflow_connect(sk, &addrs[i], &remote); 715 spin_lock_bh(&msk->pm.lock); 716 } 717 718 void mptcp_pm_nl_addr_send_ack(struct mptcp_sock *msk) 719 { 720 struct mptcp_subflow_context *subflow; 721 722 msk_owned_by_me(msk); 723 lockdep_assert_held(&msk->pm.lock); 724 725 if (!mptcp_pm_should_add_signal(msk) && 726 !mptcp_pm_should_rm_signal(msk)) 727 return; 728 729 subflow = list_first_entry_or_null(&msk->conn_list, typeof(*subflow), node); 730 if (subflow) 731 mptcp_pm_send_ack(msk, subflow, false, false); 732 } 733 734 int mptcp_pm_nl_mp_prio_send_ack(struct mptcp_sock *msk, 735 struct mptcp_addr_info *addr, 736 struct mptcp_addr_info *rem, 737 u8 bkup) 738 { 739 struct mptcp_subflow_context *subflow; 740 741 pr_debug("bkup=%d", bkup); 742 743 mptcp_for_each_subflow(msk, subflow) { 744 struct sock *ssk = mptcp_subflow_tcp_sock(subflow); 745 struct mptcp_addr_info local, remote; 746 747 mptcp_local_address((struct sock_common *)ssk, &local); 748 if (!mptcp_addresses_equal(&local, addr, addr->port)) 749 continue; 750 751 if (rem && rem->family != AF_UNSPEC) { 752 remote_address((struct sock_common *)ssk, &remote); 753 if (!mptcp_addresses_equal(&remote, rem, rem->port)) 754 continue; 755 } 756 757 __mptcp_pm_send_ack(msk, subflow, true, bkup); 758 return 0; 759 } 760 761 return -EINVAL; 762 } 763 764 static bool mptcp_local_id_match(const struct mptcp_sock *msk, u8 local_id, u8 id) 765 { 766 return local_id == id || (!local_id && msk->mpc_endpoint_id == id); 767 } 768 769 static void mptcp_pm_nl_rm_addr_or_subflow(struct mptcp_sock *msk, 770 const struct mptcp_rm_list *rm_list, 771 enum linux_mptcp_mib_field rm_type) 772 { 773 struct mptcp_subflow_context *subflow, *tmp; 774 struct sock *sk = (struct sock *)msk; 775 u8 i; 776 777 pr_debug("%s rm_list_nr %d", 778 rm_type == MPTCP_MIB_RMADDR ? "address" : "subflow", rm_list->nr); 779 780 msk_owned_by_me(msk); 781 782 if (sk->sk_state == TCP_LISTEN) 783 return; 784 785 if (!rm_list->nr) 786 return; 787 788 if (list_empty(&msk->conn_list)) 789 return; 790 791 for (i = 0; i < rm_list->nr; i++) { 792 u8 rm_id = rm_list->ids[i]; 793 bool removed = false; 794 795 mptcp_for_each_subflow_safe(msk, subflow, tmp) { 796 struct sock *ssk = mptcp_subflow_tcp_sock(subflow); 797 u8 remote_id = READ_ONCE(subflow->remote_id); 798 int how = RCV_SHUTDOWN | SEND_SHUTDOWN; 799 u8 id = subflow_get_local_id(subflow); 800 801 if (rm_type == MPTCP_MIB_RMADDR && remote_id != rm_id) 802 continue; 803 if (rm_type == MPTCP_MIB_RMSUBFLOW && !mptcp_local_id_match(msk, id, rm_id)) 804 continue; 805 806 pr_debug(" -> %s rm_list_ids[%d]=%u local_id=%u remote_id=%u mpc_id=%u", 807 rm_type == MPTCP_MIB_RMADDR ? "address" : "subflow", 808 i, rm_id, id, remote_id, msk->mpc_endpoint_id); 809 spin_unlock_bh(&msk->pm.lock); 810 mptcp_subflow_shutdown(sk, ssk, how); 811 812 /* the following takes care of updating the subflows counter */ 813 mptcp_close_ssk(sk, ssk, subflow); 814 spin_lock_bh(&msk->pm.lock); 815 816 removed = true; 817 __MPTCP_INC_STATS(sock_net(sk), rm_type); 818 } 819 if (rm_type == MPTCP_MIB_RMSUBFLOW) 820 __set_bit(rm_id ? rm_id : msk->mpc_endpoint_id, msk->pm.id_avail_bitmap); 821 if (!removed) 822 continue; 823 824 if (!mptcp_pm_is_kernel(msk)) 825 continue; 826 827 if (rm_type == MPTCP_MIB_RMADDR) { 828 msk->pm.add_addr_accepted--; 829 WRITE_ONCE(msk->pm.accept_addr, true); 830 } else if (rm_type == MPTCP_MIB_RMSUBFLOW) { 831 msk->pm.local_addr_used--; 832 } 833 } 834 } 835 836 static void mptcp_pm_nl_rm_addr_received(struct mptcp_sock *msk) 837 { 838 mptcp_pm_nl_rm_addr_or_subflow(msk, &msk->pm.rm_list_rx, MPTCP_MIB_RMADDR); 839 } 840 841 void mptcp_pm_nl_rm_subflow_received(struct mptcp_sock *msk, 842 const struct mptcp_rm_list *rm_list) 843 { 844 mptcp_pm_nl_rm_addr_or_subflow(msk, rm_list, MPTCP_MIB_RMSUBFLOW); 845 } 846 847 void mptcp_pm_nl_work(struct mptcp_sock *msk) 848 { 849 struct mptcp_pm_data *pm = &msk->pm; 850 851 msk_owned_by_me(msk); 852 853 if (!(pm->status & MPTCP_PM_WORK_MASK)) 854 return; 855 856 spin_lock_bh(&msk->pm.lock); 857 858 pr_debug("msk=%p status=%x", msk, pm->status); 859 if (pm->status & BIT(MPTCP_PM_ADD_ADDR_RECEIVED)) { 860 pm->status &= ~BIT(MPTCP_PM_ADD_ADDR_RECEIVED); 861 mptcp_pm_nl_add_addr_received(msk); 862 } 863 if (pm->status & BIT(MPTCP_PM_ADD_ADDR_SEND_ACK)) { 864 pm->status &= ~BIT(MPTCP_PM_ADD_ADDR_SEND_ACK); 865 mptcp_pm_nl_addr_send_ack(msk); 866 } 867 if (pm->status & BIT(MPTCP_PM_RM_ADDR_RECEIVED)) { 868 pm->status &= ~BIT(MPTCP_PM_RM_ADDR_RECEIVED); 869 mptcp_pm_nl_rm_addr_received(msk); 870 } 871 if (pm->status & BIT(MPTCP_PM_ESTABLISHED)) { 872 pm->status &= ~BIT(MPTCP_PM_ESTABLISHED); 873 mptcp_pm_nl_fully_established(msk); 874 } 875 if (pm->status & BIT(MPTCP_PM_SUBFLOW_ESTABLISHED)) { 876 pm->status &= ~BIT(MPTCP_PM_SUBFLOW_ESTABLISHED); 877 mptcp_pm_nl_subflow_established(msk); 878 } 879 880 spin_unlock_bh(&msk->pm.lock); 881 } 882 883 static bool address_use_port(struct mptcp_pm_addr_entry *entry) 884 { 885 return (entry->flags & 886 (MPTCP_PM_ADDR_FLAG_SIGNAL | MPTCP_PM_ADDR_FLAG_SUBFLOW)) == 887 MPTCP_PM_ADDR_FLAG_SIGNAL; 888 } 889 890 /* caller must ensure the RCU grace period is already elapsed */ 891 static void __mptcp_pm_release_addr_entry(struct mptcp_pm_addr_entry *entry) 892 { 893 if (entry->lsk) 894 sock_release(entry->lsk); 895 kfree(entry); 896 } 897 898 static int mptcp_pm_nl_append_new_local_addr(struct pm_nl_pernet *pernet, 899 struct mptcp_pm_addr_entry *entry, 900 bool needs_id) 901 { 902 struct mptcp_pm_addr_entry *cur, *del_entry = NULL; 903 unsigned int addr_max; 904 int ret = -EINVAL; 905 906 spin_lock_bh(&pernet->lock); 907 /* to keep the code simple, don't do IDR-like allocation for address ID, 908 * just bail when we exceed limits 909 */ 910 if (pernet->next_id == MPTCP_PM_MAX_ADDR_ID) 911 pernet->next_id = 1; 912 if (pernet->addrs >= MPTCP_PM_ADDR_MAX) { 913 ret = -ERANGE; 914 goto out; 915 } 916 if (test_bit(entry->addr.id, pernet->id_bitmap)) { 917 ret = -EBUSY; 918 goto out; 919 } 920 921 /* do not insert duplicate address, differentiate on port only 922 * singled addresses 923 */ 924 if (!address_use_port(entry)) 925 entry->addr.port = 0; 926 list_for_each_entry(cur, &pernet->local_addr_list, list) { 927 if (mptcp_addresses_equal(&cur->addr, &entry->addr, 928 cur->addr.port || entry->addr.port)) { 929 /* allow replacing the exiting endpoint only if such 930 * endpoint is an implicit one and the user-space 931 * did not provide an endpoint id 932 */ 933 if (!(cur->flags & MPTCP_PM_ADDR_FLAG_IMPLICIT)) { 934 ret = -EEXIST; 935 goto out; 936 } 937 if (entry->addr.id) 938 goto out; 939 940 pernet->addrs--; 941 entry->addr.id = cur->addr.id; 942 list_del_rcu(&cur->list); 943 del_entry = cur; 944 break; 945 } 946 } 947 948 if (!entry->addr.id && needs_id) { 949 find_next: 950 entry->addr.id = find_next_zero_bit(pernet->id_bitmap, 951 MPTCP_PM_MAX_ADDR_ID + 1, 952 pernet->next_id); 953 if (!entry->addr.id && pernet->next_id != 1) { 954 pernet->next_id = 1; 955 goto find_next; 956 } 957 } 958 959 if (!entry->addr.id && needs_id) 960 goto out; 961 962 __set_bit(entry->addr.id, pernet->id_bitmap); 963 if (entry->addr.id > pernet->next_id) 964 pernet->next_id = entry->addr.id; 965 966 if (entry->flags & MPTCP_PM_ADDR_FLAG_SIGNAL) { 967 addr_max = pernet->add_addr_signal_max; 968 WRITE_ONCE(pernet->add_addr_signal_max, addr_max + 1); 969 } 970 if (entry->flags & MPTCP_PM_ADDR_FLAG_SUBFLOW) { 971 addr_max = pernet->local_addr_max; 972 WRITE_ONCE(pernet->local_addr_max, addr_max + 1); 973 } 974 975 pernet->addrs++; 976 if (!entry->addr.port) 977 list_add_tail_rcu(&entry->list, &pernet->local_addr_list); 978 else 979 list_add_rcu(&entry->list, &pernet->local_addr_list); 980 ret = entry->addr.id; 981 982 out: 983 spin_unlock_bh(&pernet->lock); 984 985 /* just replaced an existing entry, free it */ 986 if (del_entry) { 987 synchronize_rcu(); 988 __mptcp_pm_release_addr_entry(del_entry); 989 } 990 return ret; 991 } 992 993 static struct lock_class_key mptcp_slock_keys[2]; 994 static struct lock_class_key mptcp_keys[2]; 995 996 static int mptcp_pm_nl_create_listen_socket(struct sock *sk, 997 struct mptcp_pm_addr_entry *entry) 998 { 999 bool is_ipv6 = sk->sk_family == AF_INET6; 1000 int addrlen = sizeof(struct sockaddr_in); 1001 struct sockaddr_storage addr; 1002 struct sock *newsk, *ssk; 1003 int backlog = 1024; 1004 int err; 1005 1006 err = sock_create_kern(sock_net(sk), entry->addr.family, 1007 SOCK_STREAM, IPPROTO_MPTCP, &entry->lsk); 1008 if (err) 1009 return err; 1010 1011 newsk = entry->lsk->sk; 1012 if (!newsk) 1013 return -EINVAL; 1014 1015 /* The subflow socket lock is acquired in a nested to the msk one 1016 * in several places, even by the TCP stack, and this msk is a kernel 1017 * socket: lockdep complains. Instead of propagating the _nested 1018 * modifiers in several places, re-init the lock class for the msk 1019 * socket to an mptcp specific one. 1020 */ 1021 sock_lock_init_class_and_name(newsk, 1022 is_ipv6 ? "mlock-AF_INET6" : "mlock-AF_INET", 1023 &mptcp_slock_keys[is_ipv6], 1024 is_ipv6 ? "msk_lock-AF_INET6" : "msk_lock-AF_INET", 1025 &mptcp_keys[is_ipv6]); 1026 1027 lock_sock(newsk); 1028 ssk = __mptcp_nmpc_sk(mptcp_sk(newsk)); 1029 release_sock(newsk); 1030 if (IS_ERR(ssk)) 1031 return PTR_ERR(ssk); 1032 1033 mptcp_info2sockaddr(&entry->addr, &addr, entry->addr.family); 1034 #if IS_ENABLED(CONFIG_MPTCP_IPV6) 1035 if (entry->addr.family == AF_INET6) 1036 addrlen = sizeof(struct sockaddr_in6); 1037 #endif 1038 if (ssk->sk_family == AF_INET) 1039 err = inet_bind_sk(ssk, (struct sockaddr *)&addr, addrlen); 1040 #if IS_ENABLED(CONFIG_MPTCP_IPV6) 1041 else if (ssk->sk_family == AF_INET6) 1042 err = inet6_bind_sk(ssk, (struct sockaddr *)&addr, addrlen); 1043 #endif 1044 if (err) 1045 return err; 1046 1047 /* We don't use mptcp_set_state() here because it needs to be called 1048 * under the msk socket lock. For the moment, that will not bring 1049 * anything more than only calling inet_sk_state_store(), because the 1050 * old status is known (TCP_CLOSE). 1051 */ 1052 inet_sk_state_store(newsk, TCP_LISTEN); 1053 lock_sock(ssk); 1054 err = __inet_listen_sk(ssk, backlog); 1055 if (!err) 1056 mptcp_event_pm_listener(ssk, MPTCP_EVENT_LISTENER_CREATED); 1057 release_sock(ssk); 1058 return err; 1059 } 1060 1061 int mptcp_pm_nl_get_local_id(struct mptcp_sock *msk, struct mptcp_addr_info *skc) 1062 { 1063 struct mptcp_pm_addr_entry *entry; 1064 struct pm_nl_pernet *pernet; 1065 int ret = -1; 1066 1067 pernet = pm_nl_get_pernet_from_msk(msk); 1068 1069 rcu_read_lock(); 1070 list_for_each_entry_rcu(entry, &pernet->local_addr_list, list) { 1071 if (mptcp_addresses_equal(&entry->addr, skc, entry->addr.port)) { 1072 ret = entry->addr.id; 1073 break; 1074 } 1075 } 1076 rcu_read_unlock(); 1077 if (ret >= 0) 1078 return ret; 1079 1080 /* address not found, add to local list */ 1081 entry = kmalloc(sizeof(*entry), GFP_ATOMIC); 1082 if (!entry) 1083 return -ENOMEM; 1084 1085 entry->addr = *skc; 1086 entry->addr.id = 0; 1087 entry->addr.port = 0; 1088 entry->ifindex = 0; 1089 entry->flags = MPTCP_PM_ADDR_FLAG_IMPLICIT; 1090 entry->lsk = NULL; 1091 ret = mptcp_pm_nl_append_new_local_addr(pernet, entry, true); 1092 if (ret < 0) 1093 kfree(entry); 1094 1095 return ret; 1096 } 1097 1098 #define MPTCP_PM_CMD_GRP_OFFSET 0 1099 #define MPTCP_PM_EV_GRP_OFFSET 1 1100 1101 static const struct genl_multicast_group mptcp_pm_mcgrps[] = { 1102 [MPTCP_PM_CMD_GRP_OFFSET] = { .name = MPTCP_PM_CMD_GRP_NAME, }, 1103 [MPTCP_PM_EV_GRP_OFFSET] = { .name = MPTCP_PM_EV_GRP_NAME, 1104 .flags = GENL_MCAST_CAP_NET_ADMIN, 1105 }, 1106 }; 1107 1108 void mptcp_pm_nl_subflow_chk_stale(const struct mptcp_sock *msk, struct sock *ssk) 1109 { 1110 struct mptcp_subflow_context *iter, *subflow = mptcp_subflow_ctx(ssk); 1111 struct sock *sk = (struct sock *)msk; 1112 unsigned int active_max_loss_cnt; 1113 struct net *net = sock_net(sk); 1114 unsigned int stale_loss_cnt; 1115 bool slow; 1116 1117 stale_loss_cnt = mptcp_stale_loss_cnt(net); 1118 if (subflow->stale || !stale_loss_cnt || subflow->stale_count <= stale_loss_cnt) 1119 return; 1120 1121 /* look for another available subflow not in loss state */ 1122 active_max_loss_cnt = max_t(int, stale_loss_cnt - 1, 1); 1123 mptcp_for_each_subflow(msk, iter) { 1124 if (iter != subflow && mptcp_subflow_active(iter) && 1125 iter->stale_count < active_max_loss_cnt) { 1126 /* we have some alternatives, try to mark this subflow as idle ...*/ 1127 slow = lock_sock_fast(ssk); 1128 if (!tcp_rtx_and_write_queues_empty(ssk)) { 1129 subflow->stale = 1; 1130 __mptcp_retransmit_pending_data(sk); 1131 MPTCP_INC_STATS(net, MPTCP_MIB_SUBFLOWSTALE); 1132 } 1133 unlock_sock_fast(ssk, slow); 1134 1135 /* always try to push the pending data regardless of re-injections: 1136 * we can possibly use backup subflows now, and subflow selection 1137 * is cheap under the msk socket lock 1138 */ 1139 __mptcp_push_pending(sk, 0); 1140 return; 1141 } 1142 } 1143 } 1144 1145 static int mptcp_pm_family_to_addr(int family) 1146 { 1147 #if IS_ENABLED(CONFIG_MPTCP_IPV6) 1148 if (family == AF_INET6) 1149 return MPTCP_PM_ADDR_ATTR_ADDR6; 1150 #endif 1151 return MPTCP_PM_ADDR_ATTR_ADDR4; 1152 } 1153 1154 static int mptcp_pm_parse_pm_addr_attr(struct nlattr *tb[], 1155 const struct nlattr *attr, 1156 struct genl_info *info, 1157 struct mptcp_addr_info *addr, 1158 bool require_family) 1159 { 1160 int err, addr_addr; 1161 1162 if (!attr) { 1163 GENL_SET_ERR_MSG(info, "missing address info"); 1164 return -EINVAL; 1165 } 1166 1167 /* no validation needed - was already done via nested policy */ 1168 err = nla_parse_nested_deprecated(tb, MPTCP_PM_ADDR_ATTR_MAX, attr, 1169 mptcp_pm_address_nl_policy, info->extack); 1170 if (err) 1171 return err; 1172 1173 if (tb[MPTCP_PM_ADDR_ATTR_ID]) 1174 addr->id = nla_get_u8(tb[MPTCP_PM_ADDR_ATTR_ID]); 1175 1176 if (!tb[MPTCP_PM_ADDR_ATTR_FAMILY]) { 1177 if (!require_family) 1178 return 0; 1179 1180 NL_SET_ERR_MSG_ATTR(info->extack, attr, 1181 "missing family"); 1182 return -EINVAL; 1183 } 1184 1185 addr->family = nla_get_u16(tb[MPTCP_PM_ADDR_ATTR_FAMILY]); 1186 if (addr->family != AF_INET 1187 #if IS_ENABLED(CONFIG_MPTCP_IPV6) 1188 && addr->family != AF_INET6 1189 #endif 1190 ) { 1191 NL_SET_ERR_MSG_ATTR(info->extack, attr, 1192 "unknown address family"); 1193 return -EINVAL; 1194 } 1195 addr_addr = mptcp_pm_family_to_addr(addr->family); 1196 if (!tb[addr_addr]) { 1197 NL_SET_ERR_MSG_ATTR(info->extack, attr, 1198 "missing address data"); 1199 return -EINVAL; 1200 } 1201 1202 #if IS_ENABLED(CONFIG_MPTCP_IPV6) 1203 if (addr->family == AF_INET6) 1204 addr->addr6 = nla_get_in6_addr(tb[addr_addr]); 1205 else 1206 #endif 1207 addr->addr.s_addr = nla_get_in_addr(tb[addr_addr]); 1208 1209 if (tb[MPTCP_PM_ADDR_ATTR_PORT]) 1210 addr->port = htons(nla_get_u16(tb[MPTCP_PM_ADDR_ATTR_PORT])); 1211 1212 return 0; 1213 } 1214 1215 int mptcp_pm_parse_addr(struct nlattr *attr, struct genl_info *info, 1216 struct mptcp_addr_info *addr) 1217 { 1218 struct nlattr *tb[MPTCP_PM_ADDR_ATTR_MAX + 1]; 1219 1220 memset(addr, 0, sizeof(*addr)); 1221 1222 return mptcp_pm_parse_pm_addr_attr(tb, attr, info, addr, true); 1223 } 1224 1225 int mptcp_pm_parse_entry(struct nlattr *attr, struct genl_info *info, 1226 bool require_family, 1227 struct mptcp_pm_addr_entry *entry) 1228 { 1229 struct nlattr *tb[MPTCP_PM_ADDR_ATTR_MAX + 1]; 1230 int err; 1231 1232 memset(entry, 0, sizeof(*entry)); 1233 1234 err = mptcp_pm_parse_pm_addr_attr(tb, attr, info, &entry->addr, require_family); 1235 if (err) 1236 return err; 1237 1238 if (tb[MPTCP_PM_ADDR_ATTR_IF_IDX]) { 1239 u32 val = nla_get_s32(tb[MPTCP_PM_ADDR_ATTR_IF_IDX]); 1240 1241 entry->ifindex = val; 1242 } 1243 1244 if (tb[MPTCP_PM_ADDR_ATTR_FLAGS]) 1245 entry->flags = nla_get_u32(tb[MPTCP_PM_ADDR_ATTR_FLAGS]); 1246 1247 if (tb[MPTCP_PM_ADDR_ATTR_PORT]) 1248 entry->addr.port = htons(nla_get_u16(tb[MPTCP_PM_ADDR_ATTR_PORT])); 1249 1250 return 0; 1251 } 1252 1253 static struct pm_nl_pernet *genl_info_pm_nl(struct genl_info *info) 1254 { 1255 return pm_nl_get_pernet(genl_info_net(info)); 1256 } 1257 1258 static int mptcp_nl_add_subflow_or_signal_addr(struct net *net) 1259 { 1260 struct mptcp_sock *msk; 1261 long s_slot = 0, s_num = 0; 1262 1263 while ((msk = mptcp_token_iter_next(net, &s_slot, &s_num)) != NULL) { 1264 struct sock *sk = (struct sock *)msk; 1265 1266 if (!READ_ONCE(msk->fully_established) || 1267 mptcp_pm_is_userspace(msk)) 1268 goto next; 1269 1270 lock_sock(sk); 1271 spin_lock_bh(&msk->pm.lock); 1272 mptcp_pm_create_subflow_or_signal_addr(msk); 1273 spin_unlock_bh(&msk->pm.lock); 1274 release_sock(sk); 1275 1276 next: 1277 sock_put(sk); 1278 cond_resched(); 1279 } 1280 1281 return 0; 1282 } 1283 1284 static bool mptcp_pm_has_addr_attr_id(const struct nlattr *attr, 1285 struct genl_info *info) 1286 { 1287 struct nlattr *tb[MPTCP_PM_ADDR_ATTR_MAX + 1]; 1288 1289 if (!nla_parse_nested_deprecated(tb, MPTCP_PM_ADDR_ATTR_MAX, attr, 1290 mptcp_pm_address_nl_policy, info->extack) && 1291 tb[MPTCP_PM_ADDR_ATTR_ID]) 1292 return true; 1293 return false; 1294 } 1295 1296 int mptcp_pm_nl_add_addr_doit(struct sk_buff *skb, struct genl_info *info) 1297 { 1298 struct nlattr *attr = info->attrs[MPTCP_PM_ENDPOINT_ADDR]; 1299 struct pm_nl_pernet *pernet = genl_info_pm_nl(info); 1300 struct mptcp_pm_addr_entry addr, *entry; 1301 int ret; 1302 1303 ret = mptcp_pm_parse_entry(attr, info, true, &addr); 1304 if (ret < 0) 1305 return ret; 1306 1307 if (addr.addr.port && !(addr.flags & MPTCP_PM_ADDR_FLAG_SIGNAL)) { 1308 GENL_SET_ERR_MSG(info, "flags must have signal when using port"); 1309 return -EINVAL; 1310 } 1311 1312 if (addr.flags & MPTCP_PM_ADDR_FLAG_SIGNAL && 1313 addr.flags & MPTCP_PM_ADDR_FLAG_FULLMESH) { 1314 GENL_SET_ERR_MSG(info, "flags mustn't have both signal and fullmesh"); 1315 return -EINVAL; 1316 } 1317 1318 if (addr.flags & MPTCP_PM_ADDR_FLAG_IMPLICIT) { 1319 GENL_SET_ERR_MSG(info, "can't create IMPLICIT endpoint"); 1320 return -EINVAL; 1321 } 1322 1323 entry = kzalloc(sizeof(*entry), GFP_KERNEL_ACCOUNT); 1324 if (!entry) { 1325 GENL_SET_ERR_MSG(info, "can't allocate addr"); 1326 return -ENOMEM; 1327 } 1328 1329 *entry = addr; 1330 if (entry->addr.port) { 1331 ret = mptcp_pm_nl_create_listen_socket(skb->sk, entry); 1332 if (ret) { 1333 GENL_SET_ERR_MSG_FMT(info, "create listen socket error: %d", ret); 1334 goto out_free; 1335 } 1336 } 1337 ret = mptcp_pm_nl_append_new_local_addr(pernet, entry, 1338 !mptcp_pm_has_addr_attr_id(attr, info)); 1339 if (ret < 0) { 1340 GENL_SET_ERR_MSG_FMT(info, "too many addresses or duplicate one: %d", ret); 1341 goto out_free; 1342 } 1343 1344 mptcp_nl_add_subflow_or_signal_addr(sock_net(skb->sk)); 1345 return 0; 1346 1347 out_free: 1348 __mptcp_pm_release_addr_entry(entry); 1349 return ret; 1350 } 1351 1352 int mptcp_pm_nl_get_flags_and_ifindex_by_id(struct mptcp_sock *msk, unsigned int id, 1353 u8 *flags, int *ifindex) 1354 { 1355 struct mptcp_pm_addr_entry *entry; 1356 struct sock *sk = (struct sock *)msk; 1357 struct net *net = sock_net(sk); 1358 1359 rcu_read_lock(); 1360 entry = __lookup_addr_by_id(pm_nl_get_pernet(net), id); 1361 if (entry) { 1362 *flags = entry->flags; 1363 *ifindex = entry->ifindex; 1364 } 1365 rcu_read_unlock(); 1366 1367 return 0; 1368 } 1369 1370 static bool remove_anno_list_by_saddr(struct mptcp_sock *msk, 1371 const struct mptcp_addr_info *addr) 1372 { 1373 struct mptcp_pm_add_entry *entry; 1374 1375 entry = mptcp_pm_del_add_timer(msk, addr, false); 1376 if (entry) { 1377 list_del(&entry->list); 1378 kfree(entry); 1379 return true; 1380 } 1381 1382 return false; 1383 } 1384 1385 static bool mptcp_pm_remove_anno_addr(struct mptcp_sock *msk, 1386 const struct mptcp_addr_info *addr, 1387 bool force) 1388 { 1389 struct mptcp_rm_list list = { .nr = 0 }; 1390 bool ret; 1391 1392 list.ids[list.nr++] = addr->id; 1393 1394 ret = remove_anno_list_by_saddr(msk, addr); 1395 if (ret || force) { 1396 spin_lock_bh(&msk->pm.lock); 1397 mptcp_pm_remove_addr(msk, &list); 1398 spin_unlock_bh(&msk->pm.lock); 1399 } 1400 return ret; 1401 } 1402 1403 static int mptcp_nl_remove_subflow_and_signal_addr(struct net *net, 1404 const struct mptcp_pm_addr_entry *entry) 1405 { 1406 const struct mptcp_addr_info *addr = &entry->addr; 1407 struct mptcp_rm_list list = { .nr = 0 }; 1408 long s_slot = 0, s_num = 0; 1409 struct mptcp_sock *msk; 1410 1411 pr_debug("remove_id=%d", addr->id); 1412 1413 list.ids[list.nr++] = addr->id; 1414 1415 while ((msk = mptcp_token_iter_next(net, &s_slot, &s_num)) != NULL) { 1416 struct sock *sk = (struct sock *)msk; 1417 bool remove_subflow; 1418 1419 if (mptcp_pm_is_userspace(msk)) 1420 goto next; 1421 1422 if (list_empty(&msk->conn_list)) { 1423 mptcp_pm_remove_anno_addr(msk, addr, false); 1424 goto next; 1425 } 1426 1427 lock_sock(sk); 1428 remove_subflow = lookup_subflow_by_saddr(&msk->conn_list, addr); 1429 mptcp_pm_remove_anno_addr(msk, addr, remove_subflow && 1430 !(entry->flags & MPTCP_PM_ADDR_FLAG_IMPLICIT)); 1431 if (remove_subflow) 1432 mptcp_pm_remove_subflow(msk, &list); 1433 release_sock(sk); 1434 1435 next: 1436 sock_put(sk); 1437 cond_resched(); 1438 } 1439 1440 return 0; 1441 } 1442 1443 static int mptcp_nl_remove_id_zero_address(struct net *net, 1444 struct mptcp_addr_info *addr) 1445 { 1446 struct mptcp_rm_list list = { .nr = 0 }; 1447 long s_slot = 0, s_num = 0; 1448 struct mptcp_sock *msk; 1449 1450 list.ids[list.nr++] = 0; 1451 1452 while ((msk = mptcp_token_iter_next(net, &s_slot, &s_num)) != NULL) { 1453 struct sock *sk = (struct sock *)msk; 1454 struct mptcp_addr_info msk_local; 1455 1456 if (list_empty(&msk->conn_list) || mptcp_pm_is_userspace(msk)) 1457 goto next; 1458 1459 mptcp_local_address((struct sock_common *)msk, &msk_local); 1460 if (!mptcp_addresses_equal(&msk_local, addr, addr->port)) 1461 goto next; 1462 1463 lock_sock(sk); 1464 spin_lock_bh(&msk->pm.lock); 1465 mptcp_pm_remove_addr(msk, &list); 1466 mptcp_pm_nl_rm_subflow_received(msk, &list); 1467 spin_unlock_bh(&msk->pm.lock); 1468 release_sock(sk); 1469 1470 next: 1471 sock_put(sk); 1472 cond_resched(); 1473 } 1474 1475 return 0; 1476 } 1477 1478 int mptcp_pm_nl_del_addr_doit(struct sk_buff *skb, struct genl_info *info) 1479 { 1480 struct nlattr *attr = info->attrs[MPTCP_PM_ENDPOINT_ADDR]; 1481 struct pm_nl_pernet *pernet = genl_info_pm_nl(info); 1482 struct mptcp_pm_addr_entry addr, *entry; 1483 unsigned int addr_max; 1484 int ret; 1485 1486 ret = mptcp_pm_parse_entry(attr, info, false, &addr); 1487 if (ret < 0) 1488 return ret; 1489 1490 /* the zero id address is special: the first address used by the msk 1491 * always gets such an id, so different subflows can have different zero 1492 * id addresses. Additionally zero id is not accounted for in id_bitmap. 1493 * Let's use an 'mptcp_rm_list' instead of the common remove code. 1494 */ 1495 if (addr.addr.id == 0) 1496 return mptcp_nl_remove_id_zero_address(sock_net(skb->sk), &addr.addr); 1497 1498 spin_lock_bh(&pernet->lock); 1499 entry = __lookup_addr_by_id(pernet, addr.addr.id); 1500 if (!entry) { 1501 GENL_SET_ERR_MSG(info, "address not found"); 1502 spin_unlock_bh(&pernet->lock); 1503 return -EINVAL; 1504 } 1505 if (entry->flags & MPTCP_PM_ADDR_FLAG_SIGNAL) { 1506 addr_max = pernet->add_addr_signal_max; 1507 WRITE_ONCE(pernet->add_addr_signal_max, addr_max - 1); 1508 } 1509 if (entry->flags & MPTCP_PM_ADDR_FLAG_SUBFLOW) { 1510 addr_max = pernet->local_addr_max; 1511 WRITE_ONCE(pernet->local_addr_max, addr_max - 1); 1512 } 1513 1514 pernet->addrs--; 1515 list_del_rcu(&entry->list); 1516 __clear_bit(entry->addr.id, pernet->id_bitmap); 1517 spin_unlock_bh(&pernet->lock); 1518 1519 mptcp_nl_remove_subflow_and_signal_addr(sock_net(skb->sk), entry); 1520 synchronize_rcu(); 1521 __mptcp_pm_release_addr_entry(entry); 1522 1523 return ret; 1524 } 1525 1526 void mptcp_pm_remove_addrs(struct mptcp_sock *msk, struct list_head *rm_list) 1527 { 1528 struct mptcp_rm_list alist = { .nr = 0 }; 1529 struct mptcp_pm_addr_entry *entry; 1530 1531 list_for_each_entry(entry, rm_list, list) { 1532 if ((remove_anno_list_by_saddr(msk, &entry->addr) || 1533 lookup_subflow_by_saddr(&msk->conn_list, &entry->addr)) && 1534 alist.nr < MPTCP_RM_IDS_MAX) 1535 alist.ids[alist.nr++] = entry->addr.id; 1536 } 1537 1538 if (alist.nr) { 1539 spin_lock_bh(&msk->pm.lock); 1540 mptcp_pm_remove_addr(msk, &alist); 1541 spin_unlock_bh(&msk->pm.lock); 1542 } 1543 } 1544 1545 static 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 int mptcp_pm_nl_flush_addrs_doit(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 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 int mptcp_pm_nl_get_addr(struct sk_buff *skb, struct genl_info *info) 1671 { 1672 struct nlattr *attr = info->attrs[MPTCP_PM_ENDPOINT_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 int mptcp_pm_nl_get_addr_doit(struct sk_buff *skb, struct genl_info *info) 1721 { 1722 return mptcp_pm_get_addr(skb, info); 1723 } 1724 1725 int mptcp_pm_nl_dump_addr(struct sk_buff *msg, 1726 struct netlink_callback *cb) 1727 { 1728 struct net *net = sock_net(msg->sk); 1729 struct mptcp_pm_addr_entry *entry; 1730 struct pm_nl_pernet *pernet; 1731 int id = cb->args[0]; 1732 void *hdr; 1733 int i; 1734 1735 pernet = pm_nl_get_pernet(net); 1736 1737 spin_lock_bh(&pernet->lock); 1738 for (i = id; i < MPTCP_PM_MAX_ADDR_ID + 1; i++) { 1739 if (test_bit(i, pernet->id_bitmap)) { 1740 entry = __lookup_addr_by_id(pernet, i); 1741 if (!entry) 1742 break; 1743 1744 if (entry->addr.id <= id) 1745 continue; 1746 1747 hdr = genlmsg_put(msg, NETLINK_CB(cb->skb).portid, 1748 cb->nlh->nlmsg_seq, &mptcp_genl_family, 1749 NLM_F_MULTI, MPTCP_PM_CMD_GET_ADDR); 1750 if (!hdr) 1751 break; 1752 1753 if (mptcp_nl_fill_addr(msg, entry) < 0) { 1754 genlmsg_cancel(msg, hdr); 1755 break; 1756 } 1757 1758 id = entry->addr.id; 1759 genlmsg_end(msg, hdr); 1760 } 1761 } 1762 spin_unlock_bh(&pernet->lock); 1763 1764 cb->args[0] = id; 1765 return msg->len; 1766 } 1767 1768 int mptcp_pm_nl_get_addr_dumpit(struct sk_buff *msg, 1769 struct netlink_callback *cb) 1770 { 1771 return mptcp_pm_dump_addr(msg, cb); 1772 } 1773 1774 static int parse_limit(struct genl_info *info, int id, unsigned int *limit) 1775 { 1776 struct nlattr *attr = info->attrs[id]; 1777 1778 if (!attr) 1779 return 0; 1780 1781 *limit = nla_get_u32(attr); 1782 if (*limit > MPTCP_PM_ADDR_MAX) { 1783 GENL_SET_ERR_MSG(info, "limit greater than maximum"); 1784 return -EINVAL; 1785 } 1786 return 0; 1787 } 1788 1789 int mptcp_pm_nl_set_limits_doit(struct sk_buff *skb, struct genl_info *info) 1790 { 1791 struct pm_nl_pernet *pernet = genl_info_pm_nl(info); 1792 unsigned int rcv_addrs, subflows; 1793 int ret; 1794 1795 spin_lock_bh(&pernet->lock); 1796 rcv_addrs = pernet->add_addr_accept_max; 1797 ret = parse_limit(info, MPTCP_PM_ATTR_RCV_ADD_ADDRS, &rcv_addrs); 1798 if (ret) 1799 goto unlock; 1800 1801 subflows = pernet->subflows_max; 1802 ret = parse_limit(info, MPTCP_PM_ATTR_SUBFLOWS, &subflows); 1803 if (ret) 1804 goto unlock; 1805 1806 WRITE_ONCE(pernet->add_addr_accept_max, rcv_addrs); 1807 WRITE_ONCE(pernet->subflows_max, subflows); 1808 1809 unlock: 1810 spin_unlock_bh(&pernet->lock); 1811 return ret; 1812 } 1813 1814 int mptcp_pm_nl_get_limits_doit(struct sk_buff *skb, struct genl_info *info) 1815 { 1816 struct pm_nl_pernet *pernet = genl_info_pm_nl(info); 1817 struct sk_buff *msg; 1818 void *reply; 1819 1820 msg = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL); 1821 if (!msg) 1822 return -ENOMEM; 1823 1824 reply = genlmsg_put_reply(msg, info, &mptcp_genl_family, 0, 1825 MPTCP_PM_CMD_GET_LIMITS); 1826 if (!reply) 1827 goto fail; 1828 1829 if (nla_put_u32(msg, MPTCP_PM_ATTR_RCV_ADD_ADDRS, 1830 READ_ONCE(pernet->add_addr_accept_max))) 1831 goto fail; 1832 1833 if (nla_put_u32(msg, MPTCP_PM_ATTR_SUBFLOWS, 1834 READ_ONCE(pernet->subflows_max))) 1835 goto fail; 1836 1837 genlmsg_end(msg, reply); 1838 return genlmsg_reply(msg, info); 1839 1840 fail: 1841 GENL_SET_ERR_MSG(info, "not enough space in Netlink message"); 1842 nlmsg_free(msg); 1843 return -EMSGSIZE; 1844 } 1845 1846 static void mptcp_pm_nl_fullmesh(struct mptcp_sock *msk, 1847 struct mptcp_addr_info *addr) 1848 { 1849 struct mptcp_rm_list list = { .nr = 0 }; 1850 1851 list.ids[list.nr++] = addr->id; 1852 1853 spin_lock_bh(&msk->pm.lock); 1854 mptcp_pm_nl_rm_subflow_received(msk, &list); 1855 mptcp_pm_create_subflow_or_signal_addr(msk); 1856 spin_unlock_bh(&msk->pm.lock); 1857 } 1858 1859 static int mptcp_nl_set_flags(struct net *net, 1860 struct mptcp_addr_info *addr, 1861 u8 bkup, u8 changed) 1862 { 1863 long s_slot = 0, s_num = 0; 1864 struct mptcp_sock *msk; 1865 int ret = -EINVAL; 1866 1867 while ((msk = mptcp_token_iter_next(net, &s_slot, &s_num)) != NULL) { 1868 struct sock *sk = (struct sock *)msk; 1869 1870 if (list_empty(&msk->conn_list) || mptcp_pm_is_userspace(msk)) 1871 goto next; 1872 1873 lock_sock(sk); 1874 if (changed & MPTCP_PM_ADDR_FLAG_BACKUP) 1875 ret = mptcp_pm_nl_mp_prio_send_ack(msk, addr, NULL, bkup); 1876 if (changed & MPTCP_PM_ADDR_FLAG_FULLMESH) 1877 mptcp_pm_nl_fullmesh(msk, addr); 1878 release_sock(sk); 1879 1880 next: 1881 sock_put(sk); 1882 cond_resched(); 1883 } 1884 1885 return ret; 1886 } 1887 1888 int mptcp_pm_nl_set_flags(struct sk_buff *skb, struct genl_info *info) 1889 { 1890 struct mptcp_pm_addr_entry addr = { .addr = { .family = AF_UNSPEC }, }; 1891 struct nlattr *attr = info->attrs[MPTCP_PM_ATTR_ADDR]; 1892 u8 changed, mask = MPTCP_PM_ADDR_FLAG_BACKUP | 1893 MPTCP_PM_ADDR_FLAG_FULLMESH; 1894 struct net *net = sock_net(skb->sk); 1895 struct mptcp_pm_addr_entry *entry; 1896 struct pm_nl_pernet *pernet; 1897 u8 lookup_by_id = 0; 1898 u8 bkup = 0; 1899 int ret; 1900 1901 pernet = pm_nl_get_pernet(net); 1902 1903 ret = mptcp_pm_parse_entry(attr, info, false, &addr); 1904 if (ret < 0) 1905 return ret; 1906 1907 if (addr.addr.family == AF_UNSPEC) { 1908 lookup_by_id = 1; 1909 if (!addr.addr.id) { 1910 GENL_SET_ERR_MSG(info, "missing required inputs"); 1911 return -EOPNOTSUPP; 1912 } 1913 } 1914 1915 if (addr.flags & MPTCP_PM_ADDR_FLAG_BACKUP) 1916 bkup = 1; 1917 1918 spin_lock_bh(&pernet->lock); 1919 entry = lookup_by_id ? __lookup_addr_by_id(pernet, addr.addr.id) : 1920 __lookup_addr(pernet, &addr.addr); 1921 if (!entry) { 1922 spin_unlock_bh(&pernet->lock); 1923 GENL_SET_ERR_MSG(info, "address not found"); 1924 return -EINVAL; 1925 } 1926 if ((addr.flags & MPTCP_PM_ADDR_FLAG_FULLMESH) && 1927 (entry->flags & MPTCP_PM_ADDR_FLAG_SIGNAL)) { 1928 spin_unlock_bh(&pernet->lock); 1929 GENL_SET_ERR_MSG(info, "invalid addr flags"); 1930 return -EINVAL; 1931 } 1932 1933 changed = (addr.flags ^ entry->flags) & mask; 1934 entry->flags = (entry->flags & ~mask) | (addr.flags & mask); 1935 addr = *entry; 1936 spin_unlock_bh(&pernet->lock); 1937 1938 mptcp_nl_set_flags(net, &addr.addr, bkup, changed); 1939 return 0; 1940 } 1941 1942 int mptcp_pm_nl_set_flags_doit(struct sk_buff *skb, struct genl_info *info) 1943 { 1944 return mptcp_pm_set_flags(skb, info); 1945 } 1946 1947 static void mptcp_nl_mcast_send(struct net *net, struct sk_buff *nlskb, gfp_t gfp) 1948 { 1949 genlmsg_multicast_netns(&mptcp_genl_family, net, 1950 nlskb, 0, MPTCP_PM_EV_GRP_OFFSET, gfp); 1951 } 1952 1953 bool mptcp_userspace_pm_active(const struct mptcp_sock *msk) 1954 { 1955 return genl_has_listeners(&mptcp_genl_family, 1956 sock_net((const struct sock *)msk), 1957 MPTCP_PM_EV_GRP_OFFSET); 1958 } 1959 1960 static int mptcp_event_add_subflow(struct sk_buff *skb, const struct sock *ssk) 1961 { 1962 const struct inet_sock *issk = inet_sk(ssk); 1963 const struct mptcp_subflow_context *sf; 1964 1965 if (nla_put_u16(skb, MPTCP_ATTR_FAMILY, ssk->sk_family)) 1966 return -EMSGSIZE; 1967 1968 switch (ssk->sk_family) { 1969 case AF_INET: 1970 if (nla_put_in_addr(skb, MPTCP_ATTR_SADDR4, issk->inet_saddr)) 1971 return -EMSGSIZE; 1972 if (nla_put_in_addr(skb, MPTCP_ATTR_DADDR4, issk->inet_daddr)) 1973 return -EMSGSIZE; 1974 break; 1975 #if IS_ENABLED(CONFIG_MPTCP_IPV6) 1976 case AF_INET6: { 1977 const struct ipv6_pinfo *np = inet6_sk(ssk); 1978 1979 if (nla_put_in6_addr(skb, MPTCP_ATTR_SADDR6, &np->saddr)) 1980 return -EMSGSIZE; 1981 if (nla_put_in6_addr(skb, MPTCP_ATTR_DADDR6, &ssk->sk_v6_daddr)) 1982 return -EMSGSIZE; 1983 break; 1984 } 1985 #endif 1986 default: 1987 WARN_ON_ONCE(1); 1988 return -EMSGSIZE; 1989 } 1990 1991 if (nla_put_be16(skb, MPTCP_ATTR_SPORT, issk->inet_sport)) 1992 return -EMSGSIZE; 1993 if (nla_put_be16(skb, MPTCP_ATTR_DPORT, issk->inet_dport)) 1994 return -EMSGSIZE; 1995 1996 sf = mptcp_subflow_ctx(ssk); 1997 if (WARN_ON_ONCE(!sf)) 1998 return -EINVAL; 1999 2000 if (nla_put_u8(skb, MPTCP_ATTR_LOC_ID, subflow_get_local_id(sf))) 2001 return -EMSGSIZE; 2002 2003 if (nla_put_u8(skb, MPTCP_ATTR_REM_ID, sf->remote_id)) 2004 return -EMSGSIZE; 2005 2006 return 0; 2007 } 2008 2009 static int mptcp_event_put_token_and_ssk(struct sk_buff *skb, 2010 const struct mptcp_sock *msk, 2011 const struct sock *ssk) 2012 { 2013 const struct sock *sk = (const struct sock *)msk; 2014 const struct mptcp_subflow_context *sf; 2015 u8 sk_err; 2016 2017 if (nla_put_u32(skb, MPTCP_ATTR_TOKEN, READ_ONCE(msk->token))) 2018 return -EMSGSIZE; 2019 2020 if (mptcp_event_add_subflow(skb, ssk)) 2021 return -EMSGSIZE; 2022 2023 sf = mptcp_subflow_ctx(ssk); 2024 if (WARN_ON_ONCE(!sf)) 2025 return -EINVAL; 2026 2027 if (nla_put_u8(skb, MPTCP_ATTR_BACKUP, sf->backup)) 2028 return -EMSGSIZE; 2029 2030 if (ssk->sk_bound_dev_if && 2031 nla_put_s32(skb, MPTCP_ATTR_IF_IDX, ssk->sk_bound_dev_if)) 2032 return -EMSGSIZE; 2033 2034 sk_err = READ_ONCE(ssk->sk_err); 2035 if (sk_err && sk->sk_state == TCP_ESTABLISHED && 2036 nla_put_u8(skb, MPTCP_ATTR_ERROR, sk_err)) 2037 return -EMSGSIZE; 2038 2039 return 0; 2040 } 2041 2042 static int mptcp_event_sub_established(struct sk_buff *skb, 2043 const struct mptcp_sock *msk, 2044 const struct sock *ssk) 2045 { 2046 return mptcp_event_put_token_and_ssk(skb, msk, ssk); 2047 } 2048 2049 static int mptcp_event_sub_closed(struct sk_buff *skb, 2050 const struct mptcp_sock *msk, 2051 const struct sock *ssk) 2052 { 2053 const struct mptcp_subflow_context *sf; 2054 2055 if (mptcp_event_put_token_and_ssk(skb, msk, ssk)) 2056 return -EMSGSIZE; 2057 2058 sf = mptcp_subflow_ctx(ssk); 2059 if (!sf->reset_seen) 2060 return 0; 2061 2062 if (nla_put_u32(skb, MPTCP_ATTR_RESET_REASON, sf->reset_reason)) 2063 return -EMSGSIZE; 2064 2065 if (nla_put_u32(skb, MPTCP_ATTR_RESET_FLAGS, sf->reset_transient)) 2066 return -EMSGSIZE; 2067 2068 return 0; 2069 } 2070 2071 static int mptcp_event_created(struct sk_buff *skb, 2072 const struct mptcp_sock *msk, 2073 const struct sock *ssk) 2074 { 2075 int err = nla_put_u32(skb, MPTCP_ATTR_TOKEN, READ_ONCE(msk->token)); 2076 2077 if (err) 2078 return err; 2079 2080 if (nla_put_u8(skb, MPTCP_ATTR_SERVER_SIDE, READ_ONCE(msk->pm.server_side))) 2081 return -EMSGSIZE; 2082 2083 return mptcp_event_add_subflow(skb, ssk); 2084 } 2085 2086 void mptcp_event_addr_removed(const struct mptcp_sock *msk, uint8_t id) 2087 { 2088 struct net *net = sock_net((const struct sock *)msk); 2089 struct nlmsghdr *nlh; 2090 struct sk_buff *skb; 2091 2092 if (!genl_has_listeners(&mptcp_genl_family, net, MPTCP_PM_EV_GRP_OFFSET)) 2093 return; 2094 2095 skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_ATOMIC); 2096 if (!skb) 2097 return; 2098 2099 nlh = genlmsg_put(skb, 0, 0, &mptcp_genl_family, 0, MPTCP_EVENT_REMOVED); 2100 if (!nlh) 2101 goto nla_put_failure; 2102 2103 if (nla_put_u32(skb, MPTCP_ATTR_TOKEN, READ_ONCE(msk->token))) 2104 goto nla_put_failure; 2105 2106 if (nla_put_u8(skb, MPTCP_ATTR_REM_ID, id)) 2107 goto nla_put_failure; 2108 2109 genlmsg_end(skb, nlh); 2110 mptcp_nl_mcast_send(net, skb, GFP_ATOMIC); 2111 return; 2112 2113 nla_put_failure: 2114 nlmsg_free(skb); 2115 } 2116 2117 void mptcp_event_addr_announced(const struct sock *ssk, 2118 const struct mptcp_addr_info *info) 2119 { 2120 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk); 2121 struct mptcp_sock *msk = mptcp_sk(subflow->conn); 2122 struct net *net = sock_net(ssk); 2123 struct nlmsghdr *nlh; 2124 struct sk_buff *skb; 2125 2126 if (!genl_has_listeners(&mptcp_genl_family, net, MPTCP_PM_EV_GRP_OFFSET)) 2127 return; 2128 2129 skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_ATOMIC); 2130 if (!skb) 2131 return; 2132 2133 nlh = genlmsg_put(skb, 0, 0, &mptcp_genl_family, 0, 2134 MPTCP_EVENT_ANNOUNCED); 2135 if (!nlh) 2136 goto nla_put_failure; 2137 2138 if (nla_put_u32(skb, MPTCP_ATTR_TOKEN, READ_ONCE(msk->token))) 2139 goto nla_put_failure; 2140 2141 if (nla_put_u8(skb, MPTCP_ATTR_REM_ID, info->id)) 2142 goto nla_put_failure; 2143 2144 if (nla_put_be16(skb, MPTCP_ATTR_DPORT, 2145 info->port == 0 ? 2146 inet_sk(ssk)->inet_dport : 2147 info->port)) 2148 goto nla_put_failure; 2149 2150 switch (info->family) { 2151 case AF_INET: 2152 if (nla_put_in_addr(skb, MPTCP_ATTR_DADDR4, info->addr.s_addr)) 2153 goto nla_put_failure; 2154 break; 2155 #if IS_ENABLED(CONFIG_MPTCP_IPV6) 2156 case AF_INET6: 2157 if (nla_put_in6_addr(skb, MPTCP_ATTR_DADDR6, &info->addr6)) 2158 goto nla_put_failure; 2159 break; 2160 #endif 2161 default: 2162 WARN_ON_ONCE(1); 2163 goto nla_put_failure; 2164 } 2165 2166 genlmsg_end(skb, nlh); 2167 mptcp_nl_mcast_send(net, skb, GFP_ATOMIC); 2168 return; 2169 2170 nla_put_failure: 2171 nlmsg_free(skb); 2172 } 2173 2174 void mptcp_event_pm_listener(const struct sock *ssk, 2175 enum mptcp_event_type event) 2176 { 2177 const struct inet_sock *issk = inet_sk(ssk); 2178 struct net *net = sock_net(ssk); 2179 struct nlmsghdr *nlh; 2180 struct sk_buff *skb; 2181 2182 if (!genl_has_listeners(&mptcp_genl_family, net, MPTCP_PM_EV_GRP_OFFSET)) 2183 return; 2184 2185 skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL); 2186 if (!skb) 2187 return; 2188 2189 nlh = genlmsg_put(skb, 0, 0, &mptcp_genl_family, 0, event); 2190 if (!nlh) 2191 goto nla_put_failure; 2192 2193 if (nla_put_u16(skb, MPTCP_ATTR_FAMILY, ssk->sk_family)) 2194 goto nla_put_failure; 2195 2196 if (nla_put_be16(skb, MPTCP_ATTR_SPORT, issk->inet_sport)) 2197 goto nla_put_failure; 2198 2199 switch (ssk->sk_family) { 2200 case AF_INET: 2201 if (nla_put_in_addr(skb, MPTCP_ATTR_SADDR4, issk->inet_saddr)) 2202 goto nla_put_failure; 2203 break; 2204 #if IS_ENABLED(CONFIG_MPTCP_IPV6) 2205 case AF_INET6: { 2206 const struct ipv6_pinfo *np = inet6_sk(ssk); 2207 2208 if (nla_put_in6_addr(skb, MPTCP_ATTR_SADDR6, &np->saddr)) 2209 goto nla_put_failure; 2210 break; 2211 } 2212 #endif 2213 default: 2214 WARN_ON_ONCE(1); 2215 goto nla_put_failure; 2216 } 2217 2218 genlmsg_end(skb, nlh); 2219 mptcp_nl_mcast_send(net, skb, GFP_KERNEL); 2220 return; 2221 2222 nla_put_failure: 2223 nlmsg_free(skb); 2224 } 2225 2226 void mptcp_event(enum mptcp_event_type type, const struct mptcp_sock *msk, 2227 const struct sock *ssk, gfp_t gfp) 2228 { 2229 struct net *net = sock_net((const struct sock *)msk); 2230 struct nlmsghdr *nlh; 2231 struct sk_buff *skb; 2232 2233 if (!genl_has_listeners(&mptcp_genl_family, net, MPTCP_PM_EV_GRP_OFFSET)) 2234 return; 2235 2236 skb = nlmsg_new(NLMSG_DEFAULT_SIZE, gfp); 2237 if (!skb) 2238 return; 2239 2240 nlh = genlmsg_put(skb, 0, 0, &mptcp_genl_family, 0, type); 2241 if (!nlh) 2242 goto nla_put_failure; 2243 2244 switch (type) { 2245 case MPTCP_EVENT_UNSPEC: 2246 WARN_ON_ONCE(1); 2247 break; 2248 case MPTCP_EVENT_CREATED: 2249 case MPTCP_EVENT_ESTABLISHED: 2250 if (mptcp_event_created(skb, msk, ssk) < 0) 2251 goto nla_put_failure; 2252 break; 2253 case MPTCP_EVENT_CLOSED: 2254 if (nla_put_u32(skb, MPTCP_ATTR_TOKEN, READ_ONCE(msk->token)) < 0) 2255 goto nla_put_failure; 2256 break; 2257 case MPTCP_EVENT_ANNOUNCED: 2258 case MPTCP_EVENT_REMOVED: 2259 /* call mptcp_event_addr_announced()/removed instead */ 2260 WARN_ON_ONCE(1); 2261 break; 2262 case MPTCP_EVENT_SUB_ESTABLISHED: 2263 case MPTCP_EVENT_SUB_PRIORITY: 2264 if (mptcp_event_sub_established(skb, msk, ssk) < 0) 2265 goto nla_put_failure; 2266 break; 2267 case MPTCP_EVENT_SUB_CLOSED: 2268 if (mptcp_event_sub_closed(skb, msk, ssk) < 0) 2269 goto nla_put_failure; 2270 break; 2271 case MPTCP_EVENT_LISTENER_CREATED: 2272 case MPTCP_EVENT_LISTENER_CLOSED: 2273 break; 2274 } 2275 2276 genlmsg_end(skb, nlh); 2277 mptcp_nl_mcast_send(net, skb, gfp); 2278 return; 2279 2280 nla_put_failure: 2281 nlmsg_free(skb); 2282 } 2283 2284 struct genl_family mptcp_genl_family __ro_after_init = { 2285 .name = MPTCP_PM_NAME, 2286 .version = MPTCP_PM_VER, 2287 .netnsok = true, 2288 .module = THIS_MODULE, 2289 .ops = mptcp_pm_nl_ops, 2290 .n_ops = ARRAY_SIZE(mptcp_pm_nl_ops), 2291 .resv_start_op = MPTCP_PM_CMD_SUBFLOW_DESTROY + 1, 2292 .mcgrps = mptcp_pm_mcgrps, 2293 .n_mcgrps = ARRAY_SIZE(mptcp_pm_mcgrps), 2294 }; 2295 2296 static int __net_init pm_nl_init_net(struct net *net) 2297 { 2298 struct pm_nl_pernet *pernet = pm_nl_get_pernet(net); 2299 2300 INIT_LIST_HEAD_RCU(&pernet->local_addr_list); 2301 2302 /* Cit. 2 subflows ought to be enough for anybody. */ 2303 pernet->subflows_max = 2; 2304 pernet->next_id = 1; 2305 pernet->stale_loss_cnt = 4; 2306 spin_lock_init(&pernet->lock); 2307 2308 /* No need to initialize other pernet fields, the struct is zeroed at 2309 * allocation time. 2310 */ 2311 2312 return 0; 2313 } 2314 2315 static void __net_exit pm_nl_exit_net(struct list_head *net_list) 2316 { 2317 struct net *net; 2318 2319 list_for_each_entry(net, net_list, exit_list) { 2320 struct pm_nl_pernet *pernet = pm_nl_get_pernet(net); 2321 2322 /* net is removed from namespace list, can't race with 2323 * other modifiers, also netns core already waited for a 2324 * RCU grace period. 2325 */ 2326 __flush_addrs(&pernet->local_addr_list); 2327 } 2328 } 2329 2330 static struct pernet_operations mptcp_pm_pernet_ops = { 2331 .init = pm_nl_init_net, 2332 .exit_batch = pm_nl_exit_net, 2333 .id = &pm_nl_pernet_id, 2334 .size = sizeof(struct pm_nl_pernet), 2335 }; 2336 2337 void __init mptcp_pm_nl_init(void) 2338 { 2339 if (register_pernet_subsys(&mptcp_pm_pernet_ops) < 0) 2340 panic("Failed to register MPTCP PM pernet subsystem.\n"); 2341 2342 if (genl_register_family(&mptcp_genl_family)) 2343 panic("Failed to register MPTCP PM netlink family\n"); 2344 } 2345