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 (WARN_ON_ONCE(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->request_bkup = backup; 475 } 476 477 __mptcp_subflow_send_ack(ssk); 478 unlock_sock_fast(ssk, slow); 479 } 480 481 static void mptcp_pm_send_ack(struct mptcp_sock *msk, struct mptcp_subflow_context *subflow, 482 bool prio, bool backup) 483 { 484 spin_unlock_bh(&msk->pm.lock); 485 __mptcp_pm_send_ack(msk, subflow, prio, backup); 486 spin_lock_bh(&msk->pm.lock); 487 } 488 489 static struct mptcp_pm_addr_entry * 490 __lookup_addr_by_id(struct pm_nl_pernet *pernet, unsigned int id) 491 { 492 struct mptcp_pm_addr_entry *entry; 493 494 list_for_each_entry(entry, &pernet->local_addr_list, list) { 495 if (entry->addr.id == id) 496 return entry; 497 } 498 return NULL; 499 } 500 501 static struct mptcp_pm_addr_entry * 502 __lookup_addr(struct pm_nl_pernet *pernet, const struct mptcp_addr_info *info) 503 { 504 struct mptcp_pm_addr_entry *entry; 505 506 list_for_each_entry(entry, &pernet->local_addr_list, list) { 507 if (mptcp_addresses_equal(&entry->addr, info, entry->addr.port)) 508 return entry; 509 } 510 return NULL; 511 } 512 513 static void mptcp_pm_create_subflow_or_signal_addr(struct mptcp_sock *msk) 514 { 515 struct mptcp_pm_addr_entry *local, *signal_and_subflow = NULL; 516 struct sock *sk = (struct sock *)msk; 517 unsigned int add_addr_signal_max; 518 unsigned int local_addr_max; 519 struct pm_nl_pernet *pernet; 520 unsigned int subflows_max; 521 522 pernet = pm_nl_get_pernet(sock_net(sk)); 523 524 add_addr_signal_max = mptcp_pm_get_add_addr_signal_max(msk); 525 local_addr_max = mptcp_pm_get_local_addr_max(msk); 526 subflows_max = mptcp_pm_get_subflows_max(msk); 527 528 /* do lazy endpoint usage accounting for the MPC subflows */ 529 if (unlikely(!(msk->pm.status & BIT(MPTCP_PM_MPC_ENDPOINT_ACCOUNTED))) && msk->first) { 530 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(msk->first); 531 struct mptcp_pm_addr_entry *entry; 532 struct mptcp_addr_info mpc_addr; 533 bool backup = false; 534 535 mptcp_local_address((struct sock_common *)msk->first, &mpc_addr); 536 rcu_read_lock(); 537 entry = __lookup_addr(pernet, &mpc_addr); 538 if (entry) { 539 __clear_bit(entry->addr.id, msk->pm.id_avail_bitmap); 540 msk->mpc_endpoint_id = entry->addr.id; 541 backup = !!(entry->flags & MPTCP_PM_ADDR_FLAG_BACKUP); 542 } 543 rcu_read_unlock(); 544 545 if (backup) 546 mptcp_pm_send_ack(msk, subflow, true, backup); 547 548 msk->pm.status |= BIT(MPTCP_PM_MPC_ENDPOINT_ACCOUNTED); 549 } 550 551 pr_debug("local %d:%d signal %d:%d subflows %d:%d\n", 552 msk->pm.local_addr_used, local_addr_max, 553 msk->pm.add_addr_signaled, add_addr_signal_max, 554 msk->pm.subflows, subflows_max); 555 556 /* check first for announce */ 557 if (msk->pm.add_addr_signaled < add_addr_signal_max) { 558 /* due to racing events on both ends we can reach here while 559 * previous add address is still running: if we invoke now 560 * mptcp_pm_announce_addr(), that will fail and the 561 * corresponding id will be marked as used. 562 * Instead let the PM machinery reschedule us when the 563 * current address announce will be completed. 564 */ 565 if (msk->pm.addr_signal & BIT(MPTCP_ADD_ADDR_SIGNAL)) 566 return; 567 568 local = select_signal_address(pernet, msk); 569 if (!local) 570 goto subflow; 571 572 /* If the alloc fails, we are on memory pressure, not worth 573 * continuing, and trying to create subflows. 574 */ 575 if (!mptcp_pm_alloc_anno_list(msk, &local->addr)) 576 return; 577 578 __clear_bit(local->addr.id, msk->pm.id_avail_bitmap); 579 msk->pm.add_addr_signaled++; 580 mptcp_pm_announce_addr(msk, &local->addr, false); 581 mptcp_pm_nl_addr_send_ack(msk); 582 583 if (local->flags & MPTCP_PM_ADDR_FLAG_SUBFLOW) 584 signal_and_subflow = local; 585 } 586 587 subflow: 588 /* check if should create a new subflow */ 589 while (msk->pm.local_addr_used < local_addr_max && 590 msk->pm.subflows < subflows_max) { 591 struct mptcp_addr_info addrs[MPTCP_PM_ADDR_MAX]; 592 bool fullmesh; 593 int i, nr; 594 595 if (signal_and_subflow) { 596 local = signal_and_subflow; 597 signal_and_subflow = NULL; 598 } else { 599 local = select_local_address(pernet, msk); 600 if (!local) 601 break; 602 } 603 604 fullmesh = !!(local->flags & MPTCP_PM_ADDR_FLAG_FULLMESH); 605 606 msk->pm.local_addr_used++; 607 __clear_bit(local->addr.id, msk->pm.id_avail_bitmap); 608 nr = fill_remote_addresses_vec(msk, &local->addr, fullmesh, addrs); 609 if (nr == 0) 610 continue; 611 612 spin_unlock_bh(&msk->pm.lock); 613 for (i = 0; i < nr; i++) 614 __mptcp_subflow_connect(sk, &local->addr, &addrs[i]); 615 spin_lock_bh(&msk->pm.lock); 616 } 617 mptcp_pm_nl_check_work_pending(msk); 618 } 619 620 static void mptcp_pm_nl_fully_established(struct mptcp_sock *msk) 621 { 622 mptcp_pm_create_subflow_or_signal_addr(msk); 623 } 624 625 static void mptcp_pm_nl_subflow_established(struct mptcp_sock *msk) 626 { 627 mptcp_pm_create_subflow_or_signal_addr(msk); 628 } 629 630 /* Fill all the local addresses into the array addrs[], 631 * and return the array size. 632 */ 633 static unsigned int fill_local_addresses_vec(struct mptcp_sock *msk, 634 struct mptcp_addr_info *remote, 635 struct mptcp_addr_info *addrs) 636 { 637 struct sock *sk = (struct sock *)msk; 638 struct mptcp_pm_addr_entry *entry; 639 struct pm_nl_pernet *pernet; 640 unsigned int subflows_max; 641 int i = 0; 642 643 pernet = pm_nl_get_pernet_from_msk(msk); 644 subflows_max = mptcp_pm_get_subflows_max(msk); 645 646 rcu_read_lock(); 647 list_for_each_entry_rcu(entry, &pernet->local_addr_list, list) { 648 if (!(entry->flags & MPTCP_PM_ADDR_FLAG_FULLMESH)) 649 continue; 650 651 if (!mptcp_pm_addr_families_match(sk, &entry->addr, remote)) 652 continue; 653 654 if (msk->pm.subflows < subflows_max) { 655 msk->pm.subflows++; 656 addrs[i++] = entry->addr; 657 } 658 } 659 rcu_read_unlock(); 660 661 /* If the array is empty, fill in the single 662 * 'IPADDRANY' local address 663 */ 664 if (!i) { 665 struct mptcp_addr_info local; 666 667 memset(&local, 0, sizeof(local)); 668 local.family = 669 #if IS_ENABLED(CONFIG_MPTCP_IPV6) 670 remote->family == AF_INET6 && 671 ipv6_addr_v4mapped(&remote->addr6) ? AF_INET : 672 #endif 673 remote->family; 674 675 if (!mptcp_pm_addr_families_match(sk, &local, remote)) 676 return 0; 677 678 msk->pm.subflows++; 679 addrs[i++] = local; 680 } 681 682 return i; 683 } 684 685 static void mptcp_pm_nl_add_addr_received(struct mptcp_sock *msk) 686 { 687 struct mptcp_addr_info addrs[MPTCP_PM_ADDR_MAX]; 688 struct sock *sk = (struct sock *)msk; 689 unsigned int add_addr_accept_max; 690 struct mptcp_addr_info remote; 691 unsigned int subflows_max; 692 bool sf_created = false; 693 int i, nr; 694 695 add_addr_accept_max = mptcp_pm_get_add_addr_accept_max(msk); 696 subflows_max = mptcp_pm_get_subflows_max(msk); 697 698 pr_debug("accepted %d:%d remote family %d", 699 msk->pm.add_addr_accepted, add_addr_accept_max, 700 msk->pm.remote.family); 701 702 remote = msk->pm.remote; 703 mptcp_pm_announce_addr(msk, &remote, true); 704 mptcp_pm_nl_addr_send_ack(msk); 705 706 if (lookup_subflow_by_daddr(&msk->conn_list, &remote)) 707 return; 708 709 /* pick id 0 port, if none is provided the remote address */ 710 if (!remote.port) 711 remote.port = sk->sk_dport; 712 713 /* connect to the specified remote address, using whatever 714 * local address the routing configuration will pick. 715 */ 716 nr = fill_local_addresses_vec(msk, &remote, addrs); 717 if (nr == 0) 718 return; 719 720 spin_unlock_bh(&msk->pm.lock); 721 for (i = 0; i < nr; i++) 722 if (__mptcp_subflow_connect(sk, &addrs[i], &remote) == 0) 723 sf_created = true; 724 spin_lock_bh(&msk->pm.lock); 725 726 if (sf_created) { 727 msk->pm.add_addr_accepted++; 728 if (msk->pm.add_addr_accepted >= add_addr_accept_max || 729 msk->pm.subflows >= subflows_max) 730 WRITE_ONCE(msk->pm.accept_addr, false); 731 } 732 } 733 734 void mptcp_pm_nl_addr_send_ack(struct mptcp_sock *msk) 735 { 736 struct mptcp_subflow_context *subflow; 737 738 msk_owned_by_me(msk); 739 lockdep_assert_held(&msk->pm.lock); 740 741 if (!mptcp_pm_should_add_signal(msk) && 742 !mptcp_pm_should_rm_signal(msk)) 743 return; 744 745 subflow = list_first_entry_or_null(&msk->conn_list, typeof(*subflow), node); 746 if (subflow) 747 mptcp_pm_send_ack(msk, subflow, false, false); 748 } 749 750 int mptcp_pm_nl_mp_prio_send_ack(struct mptcp_sock *msk, 751 struct mptcp_addr_info *addr, 752 struct mptcp_addr_info *rem, 753 u8 bkup) 754 { 755 struct mptcp_subflow_context *subflow; 756 757 pr_debug("bkup=%d", bkup); 758 759 mptcp_for_each_subflow(msk, subflow) { 760 struct sock *ssk = mptcp_subflow_tcp_sock(subflow); 761 struct mptcp_addr_info local, remote; 762 763 mptcp_local_address((struct sock_common *)ssk, &local); 764 if (!mptcp_addresses_equal(&local, addr, addr->port)) 765 continue; 766 767 if (rem && rem->family != AF_UNSPEC) { 768 remote_address((struct sock_common *)ssk, &remote); 769 if (!mptcp_addresses_equal(&remote, rem, rem->port)) 770 continue; 771 } 772 773 __mptcp_pm_send_ack(msk, subflow, true, bkup); 774 return 0; 775 } 776 777 return -EINVAL; 778 } 779 780 static bool mptcp_local_id_match(const struct mptcp_sock *msk, u8 local_id, u8 id) 781 { 782 return local_id == id || (!local_id && msk->mpc_endpoint_id == id); 783 } 784 785 static void mptcp_pm_nl_rm_addr_or_subflow(struct mptcp_sock *msk, 786 const struct mptcp_rm_list *rm_list, 787 enum linux_mptcp_mib_field rm_type) 788 { 789 struct mptcp_subflow_context *subflow, *tmp; 790 struct sock *sk = (struct sock *)msk; 791 u8 i; 792 793 pr_debug("%s rm_list_nr %d", 794 rm_type == MPTCP_MIB_RMADDR ? "address" : "subflow", rm_list->nr); 795 796 msk_owned_by_me(msk); 797 798 if (sk->sk_state == TCP_LISTEN) 799 return; 800 801 if (!rm_list->nr) 802 return; 803 804 if (list_empty(&msk->conn_list)) 805 return; 806 807 for (i = 0; i < rm_list->nr; i++) { 808 u8 rm_id = rm_list->ids[i]; 809 bool removed = false; 810 811 mptcp_for_each_subflow_safe(msk, subflow, tmp) { 812 struct sock *ssk = mptcp_subflow_tcp_sock(subflow); 813 u8 remote_id = READ_ONCE(subflow->remote_id); 814 int how = RCV_SHUTDOWN | SEND_SHUTDOWN; 815 u8 id = subflow_get_local_id(subflow); 816 817 if (rm_type == MPTCP_MIB_RMADDR && remote_id != rm_id) 818 continue; 819 if (rm_type == MPTCP_MIB_RMSUBFLOW && !mptcp_local_id_match(msk, id, rm_id)) 820 continue; 821 822 pr_debug(" -> %s rm_list_ids[%d]=%u local_id=%u remote_id=%u mpc_id=%u", 823 rm_type == MPTCP_MIB_RMADDR ? "address" : "subflow", 824 i, rm_id, id, remote_id, msk->mpc_endpoint_id); 825 spin_unlock_bh(&msk->pm.lock); 826 mptcp_subflow_shutdown(sk, ssk, how); 827 828 /* the following takes care of updating the subflows counter */ 829 mptcp_close_ssk(sk, ssk, subflow); 830 spin_lock_bh(&msk->pm.lock); 831 832 removed = true; 833 if (rm_type == MPTCP_MIB_RMSUBFLOW) 834 __MPTCP_INC_STATS(sock_net(sk), rm_type); 835 } 836 if (rm_type == MPTCP_MIB_RMSUBFLOW) 837 __set_bit(rm_id ? rm_id : msk->mpc_endpoint_id, msk->pm.id_avail_bitmap); 838 else if (rm_type == MPTCP_MIB_RMADDR) 839 __MPTCP_INC_STATS(sock_net(sk), rm_type); 840 if (!removed) 841 continue; 842 843 if (!mptcp_pm_is_kernel(msk)) 844 continue; 845 846 if (rm_type == MPTCP_MIB_RMADDR) { 847 msk->pm.add_addr_accepted--; 848 WRITE_ONCE(msk->pm.accept_addr, true); 849 } else if (rm_type == MPTCP_MIB_RMSUBFLOW) { 850 msk->pm.local_addr_used--; 851 } 852 } 853 } 854 855 static void mptcp_pm_nl_rm_addr_received(struct mptcp_sock *msk) 856 { 857 mptcp_pm_nl_rm_addr_or_subflow(msk, &msk->pm.rm_list_rx, MPTCP_MIB_RMADDR); 858 } 859 860 void mptcp_pm_nl_rm_subflow_received(struct mptcp_sock *msk, 861 const struct mptcp_rm_list *rm_list) 862 { 863 mptcp_pm_nl_rm_addr_or_subflow(msk, rm_list, MPTCP_MIB_RMSUBFLOW); 864 } 865 866 void mptcp_pm_nl_work(struct mptcp_sock *msk) 867 { 868 struct mptcp_pm_data *pm = &msk->pm; 869 870 msk_owned_by_me(msk); 871 872 if (!(pm->status & MPTCP_PM_WORK_MASK)) 873 return; 874 875 spin_lock_bh(&msk->pm.lock); 876 877 pr_debug("msk=%p status=%x", msk, pm->status); 878 if (pm->status & BIT(MPTCP_PM_ADD_ADDR_RECEIVED)) { 879 pm->status &= ~BIT(MPTCP_PM_ADD_ADDR_RECEIVED); 880 mptcp_pm_nl_add_addr_received(msk); 881 } 882 if (pm->status & BIT(MPTCP_PM_ADD_ADDR_SEND_ACK)) { 883 pm->status &= ~BIT(MPTCP_PM_ADD_ADDR_SEND_ACK); 884 mptcp_pm_nl_addr_send_ack(msk); 885 } 886 if (pm->status & BIT(MPTCP_PM_RM_ADDR_RECEIVED)) { 887 pm->status &= ~BIT(MPTCP_PM_RM_ADDR_RECEIVED); 888 mptcp_pm_nl_rm_addr_received(msk); 889 } 890 if (pm->status & BIT(MPTCP_PM_ESTABLISHED)) { 891 pm->status &= ~BIT(MPTCP_PM_ESTABLISHED); 892 mptcp_pm_nl_fully_established(msk); 893 } 894 if (pm->status & BIT(MPTCP_PM_SUBFLOW_ESTABLISHED)) { 895 pm->status &= ~BIT(MPTCP_PM_SUBFLOW_ESTABLISHED); 896 mptcp_pm_nl_subflow_established(msk); 897 } 898 899 spin_unlock_bh(&msk->pm.lock); 900 } 901 902 static bool address_use_port(struct mptcp_pm_addr_entry *entry) 903 { 904 return (entry->flags & 905 (MPTCP_PM_ADDR_FLAG_SIGNAL | MPTCP_PM_ADDR_FLAG_SUBFLOW)) == 906 MPTCP_PM_ADDR_FLAG_SIGNAL; 907 } 908 909 /* caller must ensure the RCU grace period is already elapsed */ 910 static void __mptcp_pm_release_addr_entry(struct mptcp_pm_addr_entry *entry) 911 { 912 if (entry->lsk) 913 sock_release(entry->lsk); 914 kfree(entry); 915 } 916 917 static int mptcp_pm_nl_append_new_local_addr(struct pm_nl_pernet *pernet, 918 struct mptcp_pm_addr_entry *entry, 919 bool needs_id) 920 { 921 struct mptcp_pm_addr_entry *cur, *del_entry = NULL; 922 unsigned int addr_max; 923 int ret = -EINVAL; 924 925 spin_lock_bh(&pernet->lock); 926 /* to keep the code simple, don't do IDR-like allocation for address ID, 927 * just bail when we exceed limits 928 */ 929 if (pernet->next_id == MPTCP_PM_MAX_ADDR_ID) 930 pernet->next_id = 1; 931 if (pernet->addrs >= MPTCP_PM_ADDR_MAX) { 932 ret = -ERANGE; 933 goto out; 934 } 935 if (test_bit(entry->addr.id, pernet->id_bitmap)) { 936 ret = -EBUSY; 937 goto out; 938 } 939 940 /* do not insert duplicate address, differentiate on port only 941 * singled addresses 942 */ 943 if (!address_use_port(entry)) 944 entry->addr.port = 0; 945 list_for_each_entry(cur, &pernet->local_addr_list, list) { 946 if (mptcp_addresses_equal(&cur->addr, &entry->addr, 947 cur->addr.port || entry->addr.port)) { 948 /* allow replacing the exiting endpoint only if such 949 * endpoint is an implicit one and the user-space 950 * did not provide an endpoint id 951 */ 952 if (!(cur->flags & MPTCP_PM_ADDR_FLAG_IMPLICIT)) { 953 ret = -EEXIST; 954 goto out; 955 } 956 if (entry->addr.id) 957 goto out; 958 959 pernet->addrs--; 960 entry->addr.id = cur->addr.id; 961 list_del_rcu(&cur->list); 962 del_entry = cur; 963 break; 964 } 965 } 966 967 if (!entry->addr.id && needs_id) { 968 find_next: 969 entry->addr.id = find_next_zero_bit(pernet->id_bitmap, 970 MPTCP_PM_MAX_ADDR_ID + 1, 971 pernet->next_id); 972 if (!entry->addr.id && pernet->next_id != 1) { 973 pernet->next_id = 1; 974 goto find_next; 975 } 976 } 977 978 if (!entry->addr.id && needs_id) 979 goto out; 980 981 __set_bit(entry->addr.id, pernet->id_bitmap); 982 if (entry->addr.id > pernet->next_id) 983 pernet->next_id = entry->addr.id; 984 985 if (entry->flags & MPTCP_PM_ADDR_FLAG_SIGNAL) { 986 addr_max = pernet->add_addr_signal_max; 987 WRITE_ONCE(pernet->add_addr_signal_max, addr_max + 1); 988 } 989 if (entry->flags & MPTCP_PM_ADDR_FLAG_SUBFLOW) { 990 addr_max = pernet->local_addr_max; 991 WRITE_ONCE(pernet->local_addr_max, addr_max + 1); 992 } 993 994 pernet->addrs++; 995 if (!entry->addr.port) 996 list_add_tail_rcu(&entry->list, &pernet->local_addr_list); 997 else 998 list_add_rcu(&entry->list, &pernet->local_addr_list); 999 ret = entry->addr.id; 1000 1001 out: 1002 spin_unlock_bh(&pernet->lock); 1003 1004 /* just replaced an existing entry, free it */ 1005 if (del_entry) { 1006 synchronize_rcu(); 1007 __mptcp_pm_release_addr_entry(del_entry); 1008 } 1009 return ret; 1010 } 1011 1012 static struct lock_class_key mptcp_slock_keys[2]; 1013 static struct lock_class_key mptcp_keys[2]; 1014 1015 static int mptcp_pm_nl_create_listen_socket(struct sock *sk, 1016 struct mptcp_pm_addr_entry *entry) 1017 { 1018 bool is_ipv6 = sk->sk_family == AF_INET6; 1019 int addrlen = sizeof(struct sockaddr_in); 1020 struct sockaddr_storage addr; 1021 struct sock *newsk, *ssk; 1022 int backlog = 1024; 1023 int err; 1024 1025 err = sock_create_kern(sock_net(sk), entry->addr.family, 1026 SOCK_STREAM, IPPROTO_MPTCP, &entry->lsk); 1027 if (err) 1028 return err; 1029 1030 newsk = entry->lsk->sk; 1031 if (!newsk) 1032 return -EINVAL; 1033 1034 /* The subflow socket lock is acquired in a nested to the msk one 1035 * in several places, even by the TCP stack, and this msk is a kernel 1036 * socket: lockdep complains. Instead of propagating the _nested 1037 * modifiers in several places, re-init the lock class for the msk 1038 * socket to an mptcp specific one. 1039 */ 1040 sock_lock_init_class_and_name(newsk, 1041 is_ipv6 ? "mlock-AF_INET6" : "mlock-AF_INET", 1042 &mptcp_slock_keys[is_ipv6], 1043 is_ipv6 ? "msk_lock-AF_INET6" : "msk_lock-AF_INET", 1044 &mptcp_keys[is_ipv6]); 1045 1046 lock_sock(newsk); 1047 ssk = __mptcp_nmpc_sk(mptcp_sk(newsk)); 1048 release_sock(newsk); 1049 if (IS_ERR(ssk)) 1050 return PTR_ERR(ssk); 1051 1052 mptcp_info2sockaddr(&entry->addr, &addr, entry->addr.family); 1053 #if IS_ENABLED(CONFIG_MPTCP_IPV6) 1054 if (entry->addr.family == AF_INET6) 1055 addrlen = sizeof(struct sockaddr_in6); 1056 #endif 1057 if (ssk->sk_family == AF_INET) 1058 err = inet_bind_sk(ssk, (struct sockaddr *)&addr, addrlen); 1059 #if IS_ENABLED(CONFIG_MPTCP_IPV6) 1060 else if (ssk->sk_family == AF_INET6) 1061 err = inet6_bind_sk(ssk, (struct sockaddr *)&addr, addrlen); 1062 #endif 1063 if (err) 1064 return err; 1065 1066 /* We don't use mptcp_set_state() here because it needs to be called 1067 * under the msk socket lock. For the moment, that will not bring 1068 * anything more than only calling inet_sk_state_store(), because the 1069 * old status is known (TCP_CLOSE). 1070 */ 1071 inet_sk_state_store(newsk, TCP_LISTEN); 1072 lock_sock(ssk); 1073 err = __inet_listen_sk(ssk, backlog); 1074 if (!err) 1075 mptcp_event_pm_listener(ssk, MPTCP_EVENT_LISTENER_CREATED); 1076 release_sock(ssk); 1077 return err; 1078 } 1079 1080 int mptcp_pm_nl_get_local_id(struct mptcp_sock *msk, struct mptcp_addr_info *skc) 1081 { 1082 struct mptcp_pm_addr_entry *entry; 1083 struct pm_nl_pernet *pernet; 1084 int ret = -1; 1085 1086 pernet = pm_nl_get_pernet_from_msk(msk); 1087 1088 rcu_read_lock(); 1089 list_for_each_entry_rcu(entry, &pernet->local_addr_list, list) { 1090 if (mptcp_addresses_equal(&entry->addr, skc, entry->addr.port)) { 1091 ret = entry->addr.id; 1092 break; 1093 } 1094 } 1095 rcu_read_unlock(); 1096 if (ret >= 0) 1097 return ret; 1098 1099 /* address not found, add to local list */ 1100 entry = kmalloc(sizeof(*entry), GFP_ATOMIC); 1101 if (!entry) 1102 return -ENOMEM; 1103 1104 entry->addr = *skc; 1105 entry->addr.id = 0; 1106 entry->addr.port = 0; 1107 entry->ifindex = 0; 1108 entry->flags = MPTCP_PM_ADDR_FLAG_IMPLICIT; 1109 entry->lsk = NULL; 1110 ret = mptcp_pm_nl_append_new_local_addr(pernet, entry, true); 1111 if (ret < 0) 1112 kfree(entry); 1113 1114 return ret; 1115 } 1116 1117 bool mptcp_pm_nl_is_backup(struct mptcp_sock *msk, struct mptcp_addr_info *skc) 1118 { 1119 struct pm_nl_pernet *pernet = pm_nl_get_pernet_from_msk(msk); 1120 struct mptcp_pm_addr_entry *entry; 1121 bool backup = false; 1122 1123 rcu_read_lock(); 1124 list_for_each_entry_rcu(entry, &pernet->local_addr_list, list) { 1125 if (mptcp_addresses_equal(&entry->addr, skc, entry->addr.port)) { 1126 backup = !!(entry->flags & MPTCP_PM_ADDR_FLAG_BACKUP); 1127 break; 1128 } 1129 } 1130 rcu_read_unlock(); 1131 1132 return backup; 1133 } 1134 1135 #define MPTCP_PM_CMD_GRP_OFFSET 0 1136 #define MPTCP_PM_EV_GRP_OFFSET 1 1137 1138 static const struct genl_multicast_group mptcp_pm_mcgrps[] = { 1139 [MPTCP_PM_CMD_GRP_OFFSET] = { .name = MPTCP_PM_CMD_GRP_NAME, }, 1140 [MPTCP_PM_EV_GRP_OFFSET] = { .name = MPTCP_PM_EV_GRP_NAME, 1141 .flags = GENL_MCAST_CAP_NET_ADMIN, 1142 }, 1143 }; 1144 1145 void mptcp_pm_nl_subflow_chk_stale(const struct mptcp_sock *msk, struct sock *ssk) 1146 { 1147 struct mptcp_subflow_context *iter, *subflow = mptcp_subflow_ctx(ssk); 1148 struct sock *sk = (struct sock *)msk; 1149 unsigned int active_max_loss_cnt; 1150 struct net *net = sock_net(sk); 1151 unsigned int stale_loss_cnt; 1152 bool slow; 1153 1154 stale_loss_cnt = mptcp_stale_loss_cnt(net); 1155 if (subflow->stale || !stale_loss_cnt || subflow->stale_count <= stale_loss_cnt) 1156 return; 1157 1158 /* look for another available subflow not in loss state */ 1159 active_max_loss_cnt = max_t(int, stale_loss_cnt - 1, 1); 1160 mptcp_for_each_subflow(msk, iter) { 1161 if (iter != subflow && mptcp_subflow_active(iter) && 1162 iter->stale_count < active_max_loss_cnt) { 1163 /* we have some alternatives, try to mark this subflow as idle ...*/ 1164 slow = lock_sock_fast(ssk); 1165 if (!tcp_rtx_and_write_queues_empty(ssk)) { 1166 subflow->stale = 1; 1167 __mptcp_retransmit_pending_data(sk); 1168 MPTCP_INC_STATS(net, MPTCP_MIB_SUBFLOWSTALE); 1169 } 1170 unlock_sock_fast(ssk, slow); 1171 1172 /* always try to push the pending data regardless of re-injections: 1173 * we can possibly use backup subflows now, and subflow selection 1174 * is cheap under the msk socket lock 1175 */ 1176 __mptcp_push_pending(sk, 0); 1177 return; 1178 } 1179 } 1180 } 1181 1182 static int mptcp_pm_family_to_addr(int family) 1183 { 1184 #if IS_ENABLED(CONFIG_MPTCP_IPV6) 1185 if (family == AF_INET6) 1186 return MPTCP_PM_ADDR_ATTR_ADDR6; 1187 #endif 1188 return MPTCP_PM_ADDR_ATTR_ADDR4; 1189 } 1190 1191 static int mptcp_pm_parse_pm_addr_attr(struct nlattr *tb[], 1192 const struct nlattr *attr, 1193 struct genl_info *info, 1194 struct mptcp_addr_info *addr, 1195 bool require_family) 1196 { 1197 int err, addr_addr; 1198 1199 if (!attr) { 1200 GENL_SET_ERR_MSG(info, "missing address info"); 1201 return -EINVAL; 1202 } 1203 1204 /* no validation needed - was already done via nested policy */ 1205 err = nla_parse_nested_deprecated(tb, MPTCP_PM_ADDR_ATTR_MAX, attr, 1206 mptcp_pm_address_nl_policy, info->extack); 1207 if (err) 1208 return err; 1209 1210 if (tb[MPTCP_PM_ADDR_ATTR_ID]) 1211 addr->id = nla_get_u8(tb[MPTCP_PM_ADDR_ATTR_ID]); 1212 1213 if (!tb[MPTCP_PM_ADDR_ATTR_FAMILY]) { 1214 if (!require_family) 1215 return 0; 1216 1217 NL_SET_ERR_MSG_ATTR(info->extack, attr, 1218 "missing family"); 1219 return -EINVAL; 1220 } 1221 1222 addr->family = nla_get_u16(tb[MPTCP_PM_ADDR_ATTR_FAMILY]); 1223 if (addr->family != AF_INET 1224 #if IS_ENABLED(CONFIG_MPTCP_IPV6) 1225 && addr->family != AF_INET6 1226 #endif 1227 ) { 1228 NL_SET_ERR_MSG_ATTR(info->extack, attr, 1229 "unknown address family"); 1230 return -EINVAL; 1231 } 1232 addr_addr = mptcp_pm_family_to_addr(addr->family); 1233 if (!tb[addr_addr]) { 1234 NL_SET_ERR_MSG_ATTR(info->extack, attr, 1235 "missing address data"); 1236 return -EINVAL; 1237 } 1238 1239 #if IS_ENABLED(CONFIG_MPTCP_IPV6) 1240 if (addr->family == AF_INET6) 1241 addr->addr6 = nla_get_in6_addr(tb[addr_addr]); 1242 else 1243 #endif 1244 addr->addr.s_addr = nla_get_in_addr(tb[addr_addr]); 1245 1246 if (tb[MPTCP_PM_ADDR_ATTR_PORT]) 1247 addr->port = htons(nla_get_u16(tb[MPTCP_PM_ADDR_ATTR_PORT])); 1248 1249 return 0; 1250 } 1251 1252 int mptcp_pm_parse_addr(struct nlattr *attr, struct genl_info *info, 1253 struct mptcp_addr_info *addr) 1254 { 1255 struct nlattr *tb[MPTCP_PM_ADDR_ATTR_MAX + 1]; 1256 1257 memset(addr, 0, sizeof(*addr)); 1258 1259 return mptcp_pm_parse_pm_addr_attr(tb, attr, info, addr, true); 1260 } 1261 1262 int mptcp_pm_parse_entry(struct nlattr *attr, struct genl_info *info, 1263 bool require_family, 1264 struct mptcp_pm_addr_entry *entry) 1265 { 1266 struct nlattr *tb[MPTCP_PM_ADDR_ATTR_MAX + 1]; 1267 int err; 1268 1269 memset(entry, 0, sizeof(*entry)); 1270 1271 err = mptcp_pm_parse_pm_addr_attr(tb, attr, info, &entry->addr, require_family); 1272 if (err) 1273 return err; 1274 1275 if (tb[MPTCP_PM_ADDR_ATTR_IF_IDX]) { 1276 u32 val = nla_get_s32(tb[MPTCP_PM_ADDR_ATTR_IF_IDX]); 1277 1278 entry->ifindex = val; 1279 } 1280 1281 if (tb[MPTCP_PM_ADDR_ATTR_FLAGS]) 1282 entry->flags = nla_get_u32(tb[MPTCP_PM_ADDR_ATTR_FLAGS]); 1283 1284 if (tb[MPTCP_PM_ADDR_ATTR_PORT]) 1285 entry->addr.port = htons(nla_get_u16(tb[MPTCP_PM_ADDR_ATTR_PORT])); 1286 1287 return 0; 1288 } 1289 1290 static struct pm_nl_pernet *genl_info_pm_nl(struct genl_info *info) 1291 { 1292 return pm_nl_get_pernet(genl_info_net(info)); 1293 } 1294 1295 static int mptcp_nl_add_subflow_or_signal_addr(struct net *net) 1296 { 1297 struct mptcp_sock *msk; 1298 long s_slot = 0, s_num = 0; 1299 1300 while ((msk = mptcp_token_iter_next(net, &s_slot, &s_num)) != NULL) { 1301 struct sock *sk = (struct sock *)msk; 1302 1303 if (!READ_ONCE(msk->fully_established) || 1304 mptcp_pm_is_userspace(msk)) 1305 goto next; 1306 1307 lock_sock(sk); 1308 spin_lock_bh(&msk->pm.lock); 1309 mptcp_pm_create_subflow_or_signal_addr(msk); 1310 spin_unlock_bh(&msk->pm.lock); 1311 release_sock(sk); 1312 1313 next: 1314 sock_put(sk); 1315 cond_resched(); 1316 } 1317 1318 return 0; 1319 } 1320 1321 static bool mptcp_pm_has_addr_attr_id(const struct nlattr *attr, 1322 struct genl_info *info) 1323 { 1324 struct nlattr *tb[MPTCP_PM_ADDR_ATTR_MAX + 1]; 1325 1326 if (!nla_parse_nested_deprecated(tb, MPTCP_PM_ADDR_ATTR_MAX, attr, 1327 mptcp_pm_address_nl_policy, info->extack) && 1328 tb[MPTCP_PM_ADDR_ATTR_ID]) 1329 return true; 1330 return false; 1331 } 1332 1333 int mptcp_pm_nl_add_addr_doit(struct sk_buff *skb, struct genl_info *info) 1334 { 1335 struct nlattr *attr = info->attrs[MPTCP_PM_ENDPOINT_ADDR]; 1336 struct pm_nl_pernet *pernet = genl_info_pm_nl(info); 1337 struct mptcp_pm_addr_entry addr, *entry; 1338 int ret; 1339 1340 ret = mptcp_pm_parse_entry(attr, info, true, &addr); 1341 if (ret < 0) 1342 return ret; 1343 1344 if (addr.addr.port && !address_use_port(&addr)) { 1345 GENL_SET_ERR_MSG(info, "flags must have signal and not subflow when using port"); 1346 return -EINVAL; 1347 } 1348 1349 if (addr.flags & MPTCP_PM_ADDR_FLAG_SIGNAL && 1350 addr.flags & MPTCP_PM_ADDR_FLAG_FULLMESH) { 1351 GENL_SET_ERR_MSG(info, "flags mustn't have both signal and fullmesh"); 1352 return -EINVAL; 1353 } 1354 1355 if (addr.flags & MPTCP_PM_ADDR_FLAG_IMPLICIT) { 1356 GENL_SET_ERR_MSG(info, "can't create IMPLICIT endpoint"); 1357 return -EINVAL; 1358 } 1359 1360 entry = kzalloc(sizeof(*entry), GFP_KERNEL_ACCOUNT); 1361 if (!entry) { 1362 GENL_SET_ERR_MSG(info, "can't allocate addr"); 1363 return -ENOMEM; 1364 } 1365 1366 *entry = addr; 1367 if (entry->addr.port) { 1368 ret = mptcp_pm_nl_create_listen_socket(skb->sk, entry); 1369 if (ret) { 1370 GENL_SET_ERR_MSG_FMT(info, "create listen socket error: %d", ret); 1371 goto out_free; 1372 } 1373 } 1374 ret = mptcp_pm_nl_append_new_local_addr(pernet, entry, 1375 !mptcp_pm_has_addr_attr_id(attr, info)); 1376 if (ret < 0) { 1377 GENL_SET_ERR_MSG_FMT(info, "too many addresses or duplicate one: %d", ret); 1378 goto out_free; 1379 } 1380 1381 mptcp_nl_add_subflow_or_signal_addr(sock_net(skb->sk)); 1382 return 0; 1383 1384 out_free: 1385 __mptcp_pm_release_addr_entry(entry); 1386 return ret; 1387 } 1388 1389 int mptcp_pm_nl_get_flags_and_ifindex_by_id(struct mptcp_sock *msk, unsigned int id, 1390 u8 *flags, int *ifindex) 1391 { 1392 struct mptcp_pm_addr_entry *entry; 1393 struct sock *sk = (struct sock *)msk; 1394 struct net *net = sock_net(sk); 1395 1396 rcu_read_lock(); 1397 entry = __lookup_addr_by_id(pm_nl_get_pernet(net), id); 1398 if (entry) { 1399 *flags = entry->flags; 1400 *ifindex = entry->ifindex; 1401 } 1402 rcu_read_unlock(); 1403 1404 return 0; 1405 } 1406 1407 static bool remove_anno_list_by_saddr(struct mptcp_sock *msk, 1408 const struct mptcp_addr_info *addr) 1409 { 1410 struct mptcp_pm_add_entry *entry; 1411 1412 entry = mptcp_pm_del_add_timer(msk, addr, false); 1413 if (entry) { 1414 list_del(&entry->list); 1415 kfree(entry); 1416 return true; 1417 } 1418 1419 return false; 1420 } 1421 1422 static bool mptcp_pm_remove_anno_addr(struct mptcp_sock *msk, 1423 const struct mptcp_addr_info *addr, 1424 bool force) 1425 { 1426 struct mptcp_rm_list list = { .nr = 0 }; 1427 bool ret; 1428 1429 list.ids[list.nr++] = addr->id; 1430 1431 ret = remove_anno_list_by_saddr(msk, addr); 1432 if (ret || force) { 1433 spin_lock_bh(&msk->pm.lock); 1434 msk->pm.add_addr_signaled -= ret; 1435 mptcp_pm_remove_addr(msk, &list); 1436 spin_unlock_bh(&msk->pm.lock); 1437 } 1438 return ret; 1439 } 1440 1441 static int mptcp_nl_remove_subflow_and_signal_addr(struct net *net, 1442 const struct mptcp_pm_addr_entry *entry) 1443 { 1444 const struct mptcp_addr_info *addr = &entry->addr; 1445 struct mptcp_rm_list list = { .nr = 0 }; 1446 long s_slot = 0, s_num = 0; 1447 struct mptcp_sock *msk; 1448 1449 pr_debug("remove_id=%d", addr->id); 1450 1451 list.ids[list.nr++] = addr->id; 1452 1453 while ((msk = mptcp_token_iter_next(net, &s_slot, &s_num)) != NULL) { 1454 struct sock *sk = (struct sock *)msk; 1455 bool remove_subflow; 1456 1457 if (mptcp_pm_is_userspace(msk)) 1458 goto next; 1459 1460 if (list_empty(&msk->conn_list)) { 1461 mptcp_pm_remove_anno_addr(msk, addr, false); 1462 goto next; 1463 } 1464 1465 lock_sock(sk); 1466 remove_subflow = lookup_subflow_by_saddr(&msk->conn_list, addr); 1467 mptcp_pm_remove_anno_addr(msk, addr, remove_subflow && 1468 !(entry->flags & MPTCP_PM_ADDR_FLAG_IMPLICIT)); 1469 if (remove_subflow) 1470 mptcp_pm_remove_subflow(msk, &list); 1471 release_sock(sk); 1472 1473 next: 1474 sock_put(sk); 1475 cond_resched(); 1476 } 1477 1478 return 0; 1479 } 1480 1481 static int mptcp_nl_remove_id_zero_address(struct net *net, 1482 struct mptcp_addr_info *addr) 1483 { 1484 struct mptcp_rm_list list = { .nr = 0 }; 1485 long s_slot = 0, s_num = 0; 1486 struct mptcp_sock *msk; 1487 1488 list.ids[list.nr++] = 0; 1489 1490 while ((msk = mptcp_token_iter_next(net, &s_slot, &s_num)) != NULL) { 1491 struct sock *sk = (struct sock *)msk; 1492 struct mptcp_addr_info msk_local; 1493 1494 if (list_empty(&msk->conn_list) || mptcp_pm_is_userspace(msk)) 1495 goto next; 1496 1497 mptcp_local_address((struct sock_common *)msk, &msk_local); 1498 if (!mptcp_addresses_equal(&msk_local, addr, addr->port)) 1499 goto next; 1500 1501 lock_sock(sk); 1502 spin_lock_bh(&msk->pm.lock); 1503 mptcp_pm_remove_addr(msk, &list); 1504 mptcp_pm_nl_rm_subflow_received(msk, &list); 1505 spin_unlock_bh(&msk->pm.lock); 1506 release_sock(sk); 1507 1508 next: 1509 sock_put(sk); 1510 cond_resched(); 1511 } 1512 1513 return 0; 1514 } 1515 1516 int mptcp_pm_nl_del_addr_doit(struct sk_buff *skb, struct genl_info *info) 1517 { 1518 struct nlattr *attr = info->attrs[MPTCP_PM_ENDPOINT_ADDR]; 1519 struct pm_nl_pernet *pernet = genl_info_pm_nl(info); 1520 struct mptcp_pm_addr_entry addr, *entry; 1521 unsigned int addr_max; 1522 int ret; 1523 1524 ret = mptcp_pm_parse_entry(attr, info, false, &addr); 1525 if (ret < 0) 1526 return ret; 1527 1528 /* the zero id address is special: the first address used by the msk 1529 * always gets such an id, so different subflows can have different zero 1530 * id addresses. Additionally zero id is not accounted for in id_bitmap. 1531 * Let's use an 'mptcp_rm_list' instead of the common remove code. 1532 */ 1533 if (addr.addr.id == 0) 1534 return mptcp_nl_remove_id_zero_address(sock_net(skb->sk), &addr.addr); 1535 1536 spin_lock_bh(&pernet->lock); 1537 entry = __lookup_addr_by_id(pernet, addr.addr.id); 1538 if (!entry) { 1539 GENL_SET_ERR_MSG(info, "address not found"); 1540 spin_unlock_bh(&pernet->lock); 1541 return -EINVAL; 1542 } 1543 if (entry->flags & MPTCP_PM_ADDR_FLAG_SIGNAL) { 1544 addr_max = pernet->add_addr_signal_max; 1545 WRITE_ONCE(pernet->add_addr_signal_max, addr_max - 1); 1546 } 1547 if (entry->flags & MPTCP_PM_ADDR_FLAG_SUBFLOW) { 1548 addr_max = pernet->local_addr_max; 1549 WRITE_ONCE(pernet->local_addr_max, addr_max - 1); 1550 } 1551 1552 pernet->addrs--; 1553 list_del_rcu(&entry->list); 1554 __clear_bit(entry->addr.id, pernet->id_bitmap); 1555 spin_unlock_bh(&pernet->lock); 1556 1557 mptcp_nl_remove_subflow_and_signal_addr(sock_net(skb->sk), entry); 1558 synchronize_rcu(); 1559 __mptcp_pm_release_addr_entry(entry); 1560 1561 return ret; 1562 } 1563 1564 void mptcp_pm_remove_addrs(struct mptcp_sock *msk, struct list_head *rm_list) 1565 { 1566 struct mptcp_rm_list alist = { .nr = 0 }; 1567 struct mptcp_pm_addr_entry *entry; 1568 int anno_nr = 0; 1569 1570 list_for_each_entry(entry, rm_list, list) { 1571 if (alist.nr >= MPTCP_RM_IDS_MAX) 1572 break; 1573 1574 /* only delete if either announced or matching a subflow */ 1575 if (remove_anno_list_by_saddr(msk, &entry->addr)) 1576 anno_nr++; 1577 else if (!lookup_subflow_by_saddr(&msk->conn_list, 1578 &entry->addr)) 1579 continue; 1580 1581 alist.ids[alist.nr++] = entry->addr.id; 1582 } 1583 1584 if (alist.nr) { 1585 spin_lock_bh(&msk->pm.lock); 1586 msk->pm.add_addr_signaled -= anno_nr; 1587 mptcp_pm_remove_addr(msk, &alist); 1588 spin_unlock_bh(&msk->pm.lock); 1589 } 1590 } 1591 1592 static void mptcp_pm_remove_addrs_and_subflows(struct mptcp_sock *msk, 1593 struct list_head *rm_list) 1594 { 1595 struct mptcp_rm_list alist = { .nr = 0 }, slist = { .nr = 0 }; 1596 struct mptcp_pm_addr_entry *entry; 1597 1598 list_for_each_entry(entry, rm_list, list) { 1599 if (slist.nr < MPTCP_RM_IDS_MAX && 1600 lookup_subflow_by_saddr(&msk->conn_list, &entry->addr)) 1601 slist.ids[slist.nr++] = entry->addr.id; 1602 1603 if (alist.nr < MPTCP_RM_IDS_MAX && 1604 remove_anno_list_by_saddr(msk, &entry->addr)) 1605 alist.ids[alist.nr++] = entry->addr.id; 1606 } 1607 1608 if (alist.nr) { 1609 spin_lock_bh(&msk->pm.lock); 1610 msk->pm.add_addr_signaled -= alist.nr; 1611 mptcp_pm_remove_addr(msk, &alist); 1612 spin_unlock_bh(&msk->pm.lock); 1613 } 1614 if (slist.nr) 1615 mptcp_pm_remove_subflow(msk, &slist); 1616 } 1617 1618 static void mptcp_nl_remove_addrs_list(struct net *net, 1619 struct list_head *rm_list) 1620 { 1621 long s_slot = 0, s_num = 0; 1622 struct mptcp_sock *msk; 1623 1624 if (list_empty(rm_list)) 1625 return; 1626 1627 while ((msk = mptcp_token_iter_next(net, &s_slot, &s_num)) != NULL) { 1628 struct sock *sk = (struct sock *)msk; 1629 1630 if (!mptcp_pm_is_userspace(msk)) { 1631 lock_sock(sk); 1632 mptcp_pm_remove_addrs_and_subflows(msk, rm_list); 1633 release_sock(sk); 1634 } 1635 1636 sock_put(sk); 1637 cond_resched(); 1638 } 1639 } 1640 1641 /* caller must ensure the RCU grace period is already elapsed */ 1642 static void __flush_addrs(struct list_head *list) 1643 { 1644 while (!list_empty(list)) { 1645 struct mptcp_pm_addr_entry *cur; 1646 1647 cur = list_entry(list->next, 1648 struct mptcp_pm_addr_entry, list); 1649 list_del_rcu(&cur->list); 1650 __mptcp_pm_release_addr_entry(cur); 1651 } 1652 } 1653 1654 static void __reset_counters(struct pm_nl_pernet *pernet) 1655 { 1656 WRITE_ONCE(pernet->add_addr_signal_max, 0); 1657 WRITE_ONCE(pernet->add_addr_accept_max, 0); 1658 WRITE_ONCE(pernet->local_addr_max, 0); 1659 pernet->addrs = 0; 1660 } 1661 1662 int mptcp_pm_nl_flush_addrs_doit(struct sk_buff *skb, struct genl_info *info) 1663 { 1664 struct pm_nl_pernet *pernet = genl_info_pm_nl(info); 1665 LIST_HEAD(free_list); 1666 1667 spin_lock_bh(&pernet->lock); 1668 list_splice_init(&pernet->local_addr_list, &free_list); 1669 __reset_counters(pernet); 1670 pernet->next_id = 1; 1671 bitmap_zero(pernet->id_bitmap, MPTCP_PM_MAX_ADDR_ID + 1); 1672 spin_unlock_bh(&pernet->lock); 1673 mptcp_nl_remove_addrs_list(sock_net(skb->sk), &free_list); 1674 synchronize_rcu(); 1675 __flush_addrs(&free_list); 1676 return 0; 1677 } 1678 1679 int mptcp_nl_fill_addr(struct sk_buff *skb, 1680 struct mptcp_pm_addr_entry *entry) 1681 { 1682 struct mptcp_addr_info *addr = &entry->addr; 1683 struct nlattr *attr; 1684 1685 attr = nla_nest_start(skb, MPTCP_PM_ATTR_ADDR); 1686 if (!attr) 1687 return -EMSGSIZE; 1688 1689 if (nla_put_u16(skb, MPTCP_PM_ADDR_ATTR_FAMILY, addr->family)) 1690 goto nla_put_failure; 1691 if (nla_put_u16(skb, MPTCP_PM_ADDR_ATTR_PORT, ntohs(addr->port))) 1692 goto nla_put_failure; 1693 if (nla_put_u8(skb, MPTCP_PM_ADDR_ATTR_ID, addr->id)) 1694 goto nla_put_failure; 1695 if (nla_put_u32(skb, MPTCP_PM_ADDR_ATTR_FLAGS, entry->flags)) 1696 goto nla_put_failure; 1697 if (entry->ifindex && 1698 nla_put_s32(skb, MPTCP_PM_ADDR_ATTR_IF_IDX, entry->ifindex)) 1699 goto nla_put_failure; 1700 1701 if (addr->family == AF_INET && 1702 nla_put_in_addr(skb, MPTCP_PM_ADDR_ATTR_ADDR4, 1703 addr->addr.s_addr)) 1704 goto nla_put_failure; 1705 #if IS_ENABLED(CONFIG_MPTCP_IPV6) 1706 else if (addr->family == AF_INET6 && 1707 nla_put_in6_addr(skb, MPTCP_PM_ADDR_ATTR_ADDR6, &addr->addr6)) 1708 goto nla_put_failure; 1709 #endif 1710 nla_nest_end(skb, attr); 1711 return 0; 1712 1713 nla_put_failure: 1714 nla_nest_cancel(skb, attr); 1715 return -EMSGSIZE; 1716 } 1717 1718 int mptcp_pm_nl_get_addr(struct sk_buff *skb, struct genl_info *info) 1719 { 1720 struct nlattr *attr = info->attrs[MPTCP_PM_ENDPOINT_ADDR]; 1721 struct pm_nl_pernet *pernet = genl_info_pm_nl(info); 1722 struct mptcp_pm_addr_entry addr, *entry; 1723 struct sk_buff *msg; 1724 void *reply; 1725 int ret; 1726 1727 ret = mptcp_pm_parse_entry(attr, info, false, &addr); 1728 if (ret < 0) 1729 return ret; 1730 1731 msg = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL); 1732 if (!msg) 1733 return -ENOMEM; 1734 1735 reply = genlmsg_put_reply(msg, info, &mptcp_genl_family, 0, 1736 info->genlhdr->cmd); 1737 if (!reply) { 1738 GENL_SET_ERR_MSG(info, "not enough space in Netlink message"); 1739 ret = -EMSGSIZE; 1740 goto fail; 1741 } 1742 1743 spin_lock_bh(&pernet->lock); 1744 entry = __lookup_addr_by_id(pernet, addr.addr.id); 1745 if (!entry) { 1746 GENL_SET_ERR_MSG(info, "address not found"); 1747 ret = -EINVAL; 1748 goto unlock_fail; 1749 } 1750 1751 ret = mptcp_nl_fill_addr(msg, entry); 1752 if (ret) 1753 goto unlock_fail; 1754 1755 genlmsg_end(msg, reply); 1756 ret = genlmsg_reply(msg, info); 1757 spin_unlock_bh(&pernet->lock); 1758 return ret; 1759 1760 unlock_fail: 1761 spin_unlock_bh(&pernet->lock); 1762 1763 fail: 1764 nlmsg_free(msg); 1765 return ret; 1766 } 1767 1768 int mptcp_pm_nl_get_addr_doit(struct sk_buff *skb, struct genl_info *info) 1769 { 1770 return mptcp_pm_get_addr(skb, info); 1771 } 1772 1773 int mptcp_pm_nl_dump_addr(struct sk_buff *msg, 1774 struct netlink_callback *cb) 1775 { 1776 struct net *net = sock_net(msg->sk); 1777 struct mptcp_pm_addr_entry *entry; 1778 struct pm_nl_pernet *pernet; 1779 int id = cb->args[0]; 1780 void *hdr; 1781 int i; 1782 1783 pernet = pm_nl_get_pernet(net); 1784 1785 spin_lock_bh(&pernet->lock); 1786 for (i = id; i < MPTCP_PM_MAX_ADDR_ID + 1; i++) { 1787 if (test_bit(i, pernet->id_bitmap)) { 1788 entry = __lookup_addr_by_id(pernet, i); 1789 if (!entry) 1790 break; 1791 1792 if (entry->addr.id <= id) 1793 continue; 1794 1795 hdr = genlmsg_put(msg, NETLINK_CB(cb->skb).portid, 1796 cb->nlh->nlmsg_seq, &mptcp_genl_family, 1797 NLM_F_MULTI, MPTCP_PM_CMD_GET_ADDR); 1798 if (!hdr) 1799 break; 1800 1801 if (mptcp_nl_fill_addr(msg, entry) < 0) { 1802 genlmsg_cancel(msg, hdr); 1803 break; 1804 } 1805 1806 id = entry->addr.id; 1807 genlmsg_end(msg, hdr); 1808 } 1809 } 1810 spin_unlock_bh(&pernet->lock); 1811 1812 cb->args[0] = id; 1813 return msg->len; 1814 } 1815 1816 int mptcp_pm_nl_get_addr_dumpit(struct sk_buff *msg, 1817 struct netlink_callback *cb) 1818 { 1819 return mptcp_pm_dump_addr(msg, cb); 1820 } 1821 1822 static int parse_limit(struct genl_info *info, int id, unsigned int *limit) 1823 { 1824 struct nlattr *attr = info->attrs[id]; 1825 1826 if (!attr) 1827 return 0; 1828 1829 *limit = nla_get_u32(attr); 1830 if (*limit > MPTCP_PM_ADDR_MAX) { 1831 GENL_SET_ERR_MSG(info, "limit greater than maximum"); 1832 return -EINVAL; 1833 } 1834 return 0; 1835 } 1836 1837 int mptcp_pm_nl_set_limits_doit(struct sk_buff *skb, struct genl_info *info) 1838 { 1839 struct pm_nl_pernet *pernet = genl_info_pm_nl(info); 1840 unsigned int rcv_addrs, subflows; 1841 int ret; 1842 1843 spin_lock_bh(&pernet->lock); 1844 rcv_addrs = pernet->add_addr_accept_max; 1845 ret = parse_limit(info, MPTCP_PM_ATTR_RCV_ADD_ADDRS, &rcv_addrs); 1846 if (ret) 1847 goto unlock; 1848 1849 subflows = pernet->subflows_max; 1850 ret = parse_limit(info, MPTCP_PM_ATTR_SUBFLOWS, &subflows); 1851 if (ret) 1852 goto unlock; 1853 1854 WRITE_ONCE(pernet->add_addr_accept_max, rcv_addrs); 1855 WRITE_ONCE(pernet->subflows_max, subflows); 1856 1857 unlock: 1858 spin_unlock_bh(&pernet->lock); 1859 return ret; 1860 } 1861 1862 int mptcp_pm_nl_get_limits_doit(struct sk_buff *skb, struct genl_info *info) 1863 { 1864 struct pm_nl_pernet *pernet = genl_info_pm_nl(info); 1865 struct sk_buff *msg; 1866 void *reply; 1867 1868 msg = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL); 1869 if (!msg) 1870 return -ENOMEM; 1871 1872 reply = genlmsg_put_reply(msg, info, &mptcp_genl_family, 0, 1873 MPTCP_PM_CMD_GET_LIMITS); 1874 if (!reply) 1875 goto fail; 1876 1877 if (nla_put_u32(msg, MPTCP_PM_ATTR_RCV_ADD_ADDRS, 1878 READ_ONCE(pernet->add_addr_accept_max))) 1879 goto fail; 1880 1881 if (nla_put_u32(msg, MPTCP_PM_ATTR_SUBFLOWS, 1882 READ_ONCE(pernet->subflows_max))) 1883 goto fail; 1884 1885 genlmsg_end(msg, reply); 1886 return genlmsg_reply(msg, info); 1887 1888 fail: 1889 GENL_SET_ERR_MSG(info, "not enough space in Netlink message"); 1890 nlmsg_free(msg); 1891 return -EMSGSIZE; 1892 } 1893 1894 static void mptcp_pm_nl_fullmesh(struct mptcp_sock *msk, 1895 struct mptcp_addr_info *addr) 1896 { 1897 struct mptcp_rm_list list = { .nr = 0 }; 1898 1899 list.ids[list.nr++] = addr->id; 1900 1901 spin_lock_bh(&msk->pm.lock); 1902 mptcp_pm_nl_rm_subflow_received(msk, &list); 1903 mptcp_pm_create_subflow_or_signal_addr(msk); 1904 spin_unlock_bh(&msk->pm.lock); 1905 } 1906 1907 static int mptcp_nl_set_flags(struct net *net, 1908 struct mptcp_addr_info *addr, 1909 u8 bkup, u8 changed) 1910 { 1911 long s_slot = 0, s_num = 0; 1912 struct mptcp_sock *msk; 1913 int ret = -EINVAL; 1914 1915 while ((msk = mptcp_token_iter_next(net, &s_slot, &s_num)) != NULL) { 1916 struct sock *sk = (struct sock *)msk; 1917 1918 if (list_empty(&msk->conn_list) || mptcp_pm_is_userspace(msk)) 1919 goto next; 1920 1921 lock_sock(sk); 1922 if (changed & MPTCP_PM_ADDR_FLAG_BACKUP) 1923 ret = mptcp_pm_nl_mp_prio_send_ack(msk, addr, NULL, bkup); 1924 if (changed & MPTCP_PM_ADDR_FLAG_FULLMESH) 1925 mptcp_pm_nl_fullmesh(msk, addr); 1926 release_sock(sk); 1927 1928 next: 1929 sock_put(sk); 1930 cond_resched(); 1931 } 1932 1933 return ret; 1934 } 1935 1936 int mptcp_pm_nl_set_flags(struct sk_buff *skb, struct genl_info *info) 1937 { 1938 struct mptcp_pm_addr_entry addr = { .addr = { .family = AF_UNSPEC }, }; 1939 struct nlattr *attr = info->attrs[MPTCP_PM_ATTR_ADDR]; 1940 u8 changed, mask = MPTCP_PM_ADDR_FLAG_BACKUP | 1941 MPTCP_PM_ADDR_FLAG_FULLMESH; 1942 struct net *net = sock_net(skb->sk); 1943 struct mptcp_pm_addr_entry *entry; 1944 struct pm_nl_pernet *pernet; 1945 u8 lookup_by_id = 0; 1946 u8 bkup = 0; 1947 int ret; 1948 1949 pernet = pm_nl_get_pernet(net); 1950 1951 ret = mptcp_pm_parse_entry(attr, info, false, &addr); 1952 if (ret < 0) 1953 return ret; 1954 1955 if (addr.addr.family == AF_UNSPEC) { 1956 lookup_by_id = 1; 1957 if (!addr.addr.id) { 1958 GENL_SET_ERR_MSG(info, "missing required inputs"); 1959 return -EOPNOTSUPP; 1960 } 1961 } 1962 1963 if (addr.flags & MPTCP_PM_ADDR_FLAG_BACKUP) 1964 bkup = 1; 1965 1966 spin_lock_bh(&pernet->lock); 1967 entry = lookup_by_id ? __lookup_addr_by_id(pernet, addr.addr.id) : 1968 __lookup_addr(pernet, &addr.addr); 1969 if (!entry) { 1970 spin_unlock_bh(&pernet->lock); 1971 GENL_SET_ERR_MSG(info, "address not found"); 1972 return -EINVAL; 1973 } 1974 if ((addr.flags & MPTCP_PM_ADDR_FLAG_FULLMESH) && 1975 (entry->flags & MPTCP_PM_ADDR_FLAG_SIGNAL)) { 1976 spin_unlock_bh(&pernet->lock); 1977 GENL_SET_ERR_MSG(info, "invalid addr flags"); 1978 return -EINVAL; 1979 } 1980 1981 changed = (addr.flags ^ entry->flags) & mask; 1982 entry->flags = (entry->flags & ~mask) | (addr.flags & mask); 1983 addr = *entry; 1984 spin_unlock_bh(&pernet->lock); 1985 1986 mptcp_nl_set_flags(net, &addr.addr, bkup, changed); 1987 return 0; 1988 } 1989 1990 int mptcp_pm_nl_set_flags_doit(struct sk_buff *skb, struct genl_info *info) 1991 { 1992 return mptcp_pm_set_flags(skb, info); 1993 } 1994 1995 static void mptcp_nl_mcast_send(struct net *net, struct sk_buff *nlskb, gfp_t gfp) 1996 { 1997 genlmsg_multicast_netns(&mptcp_genl_family, net, 1998 nlskb, 0, MPTCP_PM_EV_GRP_OFFSET, gfp); 1999 } 2000 2001 bool mptcp_userspace_pm_active(const struct mptcp_sock *msk) 2002 { 2003 return genl_has_listeners(&mptcp_genl_family, 2004 sock_net((const struct sock *)msk), 2005 MPTCP_PM_EV_GRP_OFFSET); 2006 } 2007 2008 static int mptcp_event_add_subflow(struct sk_buff *skb, const struct sock *ssk) 2009 { 2010 const struct inet_sock *issk = inet_sk(ssk); 2011 const struct mptcp_subflow_context *sf; 2012 2013 if (nla_put_u16(skb, MPTCP_ATTR_FAMILY, ssk->sk_family)) 2014 return -EMSGSIZE; 2015 2016 switch (ssk->sk_family) { 2017 case AF_INET: 2018 if (nla_put_in_addr(skb, MPTCP_ATTR_SADDR4, issk->inet_saddr)) 2019 return -EMSGSIZE; 2020 if (nla_put_in_addr(skb, MPTCP_ATTR_DADDR4, issk->inet_daddr)) 2021 return -EMSGSIZE; 2022 break; 2023 #if IS_ENABLED(CONFIG_MPTCP_IPV6) 2024 case AF_INET6: { 2025 const struct ipv6_pinfo *np = inet6_sk(ssk); 2026 2027 if (nla_put_in6_addr(skb, MPTCP_ATTR_SADDR6, &np->saddr)) 2028 return -EMSGSIZE; 2029 if (nla_put_in6_addr(skb, MPTCP_ATTR_DADDR6, &ssk->sk_v6_daddr)) 2030 return -EMSGSIZE; 2031 break; 2032 } 2033 #endif 2034 default: 2035 WARN_ON_ONCE(1); 2036 return -EMSGSIZE; 2037 } 2038 2039 if (nla_put_be16(skb, MPTCP_ATTR_SPORT, issk->inet_sport)) 2040 return -EMSGSIZE; 2041 if (nla_put_be16(skb, MPTCP_ATTR_DPORT, issk->inet_dport)) 2042 return -EMSGSIZE; 2043 2044 sf = mptcp_subflow_ctx(ssk); 2045 if (WARN_ON_ONCE(!sf)) 2046 return -EINVAL; 2047 2048 if (nla_put_u8(skb, MPTCP_ATTR_LOC_ID, subflow_get_local_id(sf))) 2049 return -EMSGSIZE; 2050 2051 if (nla_put_u8(skb, MPTCP_ATTR_REM_ID, sf->remote_id)) 2052 return -EMSGSIZE; 2053 2054 return 0; 2055 } 2056 2057 static int mptcp_event_put_token_and_ssk(struct sk_buff *skb, 2058 const struct mptcp_sock *msk, 2059 const struct sock *ssk) 2060 { 2061 const struct sock *sk = (const struct sock *)msk; 2062 const struct mptcp_subflow_context *sf; 2063 u8 sk_err; 2064 2065 if (nla_put_u32(skb, MPTCP_ATTR_TOKEN, READ_ONCE(msk->token))) 2066 return -EMSGSIZE; 2067 2068 if (mptcp_event_add_subflow(skb, ssk)) 2069 return -EMSGSIZE; 2070 2071 sf = mptcp_subflow_ctx(ssk); 2072 if (WARN_ON_ONCE(!sf)) 2073 return -EINVAL; 2074 2075 if (nla_put_u8(skb, MPTCP_ATTR_BACKUP, sf->backup)) 2076 return -EMSGSIZE; 2077 2078 if (ssk->sk_bound_dev_if && 2079 nla_put_s32(skb, MPTCP_ATTR_IF_IDX, ssk->sk_bound_dev_if)) 2080 return -EMSGSIZE; 2081 2082 sk_err = READ_ONCE(ssk->sk_err); 2083 if (sk_err && sk->sk_state == TCP_ESTABLISHED && 2084 nla_put_u8(skb, MPTCP_ATTR_ERROR, sk_err)) 2085 return -EMSGSIZE; 2086 2087 return 0; 2088 } 2089 2090 static int mptcp_event_sub_established(struct sk_buff *skb, 2091 const struct mptcp_sock *msk, 2092 const struct sock *ssk) 2093 { 2094 return mptcp_event_put_token_and_ssk(skb, msk, ssk); 2095 } 2096 2097 static int mptcp_event_sub_closed(struct sk_buff *skb, 2098 const struct mptcp_sock *msk, 2099 const struct sock *ssk) 2100 { 2101 const struct mptcp_subflow_context *sf; 2102 2103 if (mptcp_event_put_token_and_ssk(skb, msk, ssk)) 2104 return -EMSGSIZE; 2105 2106 sf = mptcp_subflow_ctx(ssk); 2107 if (!sf->reset_seen) 2108 return 0; 2109 2110 if (nla_put_u32(skb, MPTCP_ATTR_RESET_REASON, sf->reset_reason)) 2111 return -EMSGSIZE; 2112 2113 if (nla_put_u32(skb, MPTCP_ATTR_RESET_FLAGS, sf->reset_transient)) 2114 return -EMSGSIZE; 2115 2116 return 0; 2117 } 2118 2119 static int mptcp_event_created(struct sk_buff *skb, 2120 const struct mptcp_sock *msk, 2121 const struct sock *ssk) 2122 { 2123 int err = nla_put_u32(skb, MPTCP_ATTR_TOKEN, READ_ONCE(msk->token)); 2124 2125 if (err) 2126 return err; 2127 2128 if (nla_put_u8(skb, MPTCP_ATTR_SERVER_SIDE, READ_ONCE(msk->pm.server_side))) 2129 return -EMSGSIZE; 2130 2131 return mptcp_event_add_subflow(skb, ssk); 2132 } 2133 2134 void mptcp_event_addr_removed(const struct mptcp_sock *msk, uint8_t id) 2135 { 2136 struct net *net = sock_net((const struct sock *)msk); 2137 struct nlmsghdr *nlh; 2138 struct sk_buff *skb; 2139 2140 if (!genl_has_listeners(&mptcp_genl_family, net, MPTCP_PM_EV_GRP_OFFSET)) 2141 return; 2142 2143 skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_ATOMIC); 2144 if (!skb) 2145 return; 2146 2147 nlh = genlmsg_put(skb, 0, 0, &mptcp_genl_family, 0, MPTCP_EVENT_REMOVED); 2148 if (!nlh) 2149 goto nla_put_failure; 2150 2151 if (nla_put_u32(skb, MPTCP_ATTR_TOKEN, READ_ONCE(msk->token))) 2152 goto nla_put_failure; 2153 2154 if (nla_put_u8(skb, MPTCP_ATTR_REM_ID, id)) 2155 goto nla_put_failure; 2156 2157 genlmsg_end(skb, nlh); 2158 mptcp_nl_mcast_send(net, skb, GFP_ATOMIC); 2159 return; 2160 2161 nla_put_failure: 2162 nlmsg_free(skb); 2163 } 2164 2165 void mptcp_event_addr_announced(const struct sock *ssk, 2166 const struct mptcp_addr_info *info) 2167 { 2168 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk); 2169 struct mptcp_sock *msk = mptcp_sk(subflow->conn); 2170 struct net *net = sock_net(ssk); 2171 struct nlmsghdr *nlh; 2172 struct sk_buff *skb; 2173 2174 if (!genl_has_listeners(&mptcp_genl_family, net, MPTCP_PM_EV_GRP_OFFSET)) 2175 return; 2176 2177 skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_ATOMIC); 2178 if (!skb) 2179 return; 2180 2181 nlh = genlmsg_put(skb, 0, 0, &mptcp_genl_family, 0, 2182 MPTCP_EVENT_ANNOUNCED); 2183 if (!nlh) 2184 goto nla_put_failure; 2185 2186 if (nla_put_u32(skb, MPTCP_ATTR_TOKEN, READ_ONCE(msk->token))) 2187 goto nla_put_failure; 2188 2189 if (nla_put_u8(skb, MPTCP_ATTR_REM_ID, info->id)) 2190 goto nla_put_failure; 2191 2192 if (nla_put_be16(skb, MPTCP_ATTR_DPORT, 2193 info->port == 0 ? 2194 inet_sk(ssk)->inet_dport : 2195 info->port)) 2196 goto nla_put_failure; 2197 2198 switch (info->family) { 2199 case AF_INET: 2200 if (nla_put_in_addr(skb, MPTCP_ATTR_DADDR4, info->addr.s_addr)) 2201 goto nla_put_failure; 2202 break; 2203 #if IS_ENABLED(CONFIG_MPTCP_IPV6) 2204 case AF_INET6: 2205 if (nla_put_in6_addr(skb, MPTCP_ATTR_DADDR6, &info->addr6)) 2206 goto nla_put_failure; 2207 break; 2208 #endif 2209 default: 2210 WARN_ON_ONCE(1); 2211 goto nla_put_failure; 2212 } 2213 2214 genlmsg_end(skb, nlh); 2215 mptcp_nl_mcast_send(net, skb, GFP_ATOMIC); 2216 return; 2217 2218 nla_put_failure: 2219 nlmsg_free(skb); 2220 } 2221 2222 void mptcp_event_pm_listener(const struct sock *ssk, 2223 enum mptcp_event_type event) 2224 { 2225 const struct inet_sock *issk = inet_sk(ssk); 2226 struct net *net = sock_net(ssk); 2227 struct nlmsghdr *nlh; 2228 struct sk_buff *skb; 2229 2230 if (!genl_has_listeners(&mptcp_genl_family, net, MPTCP_PM_EV_GRP_OFFSET)) 2231 return; 2232 2233 skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL); 2234 if (!skb) 2235 return; 2236 2237 nlh = genlmsg_put(skb, 0, 0, &mptcp_genl_family, 0, event); 2238 if (!nlh) 2239 goto nla_put_failure; 2240 2241 if (nla_put_u16(skb, MPTCP_ATTR_FAMILY, ssk->sk_family)) 2242 goto nla_put_failure; 2243 2244 if (nla_put_be16(skb, MPTCP_ATTR_SPORT, issk->inet_sport)) 2245 goto nla_put_failure; 2246 2247 switch (ssk->sk_family) { 2248 case AF_INET: 2249 if (nla_put_in_addr(skb, MPTCP_ATTR_SADDR4, issk->inet_saddr)) 2250 goto nla_put_failure; 2251 break; 2252 #if IS_ENABLED(CONFIG_MPTCP_IPV6) 2253 case AF_INET6: { 2254 const struct ipv6_pinfo *np = inet6_sk(ssk); 2255 2256 if (nla_put_in6_addr(skb, MPTCP_ATTR_SADDR6, &np->saddr)) 2257 goto nla_put_failure; 2258 break; 2259 } 2260 #endif 2261 default: 2262 WARN_ON_ONCE(1); 2263 goto nla_put_failure; 2264 } 2265 2266 genlmsg_end(skb, nlh); 2267 mptcp_nl_mcast_send(net, skb, GFP_KERNEL); 2268 return; 2269 2270 nla_put_failure: 2271 nlmsg_free(skb); 2272 } 2273 2274 void mptcp_event(enum mptcp_event_type type, const struct mptcp_sock *msk, 2275 const struct sock *ssk, gfp_t gfp) 2276 { 2277 struct net *net = sock_net((const struct sock *)msk); 2278 struct nlmsghdr *nlh; 2279 struct sk_buff *skb; 2280 2281 if (!genl_has_listeners(&mptcp_genl_family, net, MPTCP_PM_EV_GRP_OFFSET)) 2282 return; 2283 2284 skb = nlmsg_new(NLMSG_DEFAULT_SIZE, gfp); 2285 if (!skb) 2286 return; 2287 2288 nlh = genlmsg_put(skb, 0, 0, &mptcp_genl_family, 0, type); 2289 if (!nlh) 2290 goto nla_put_failure; 2291 2292 switch (type) { 2293 case MPTCP_EVENT_UNSPEC: 2294 WARN_ON_ONCE(1); 2295 break; 2296 case MPTCP_EVENT_CREATED: 2297 case MPTCP_EVENT_ESTABLISHED: 2298 if (mptcp_event_created(skb, msk, ssk) < 0) 2299 goto nla_put_failure; 2300 break; 2301 case MPTCP_EVENT_CLOSED: 2302 if (nla_put_u32(skb, MPTCP_ATTR_TOKEN, READ_ONCE(msk->token)) < 0) 2303 goto nla_put_failure; 2304 break; 2305 case MPTCP_EVENT_ANNOUNCED: 2306 case MPTCP_EVENT_REMOVED: 2307 /* call mptcp_event_addr_announced()/removed instead */ 2308 WARN_ON_ONCE(1); 2309 break; 2310 case MPTCP_EVENT_SUB_ESTABLISHED: 2311 case MPTCP_EVENT_SUB_PRIORITY: 2312 if (mptcp_event_sub_established(skb, msk, ssk) < 0) 2313 goto nla_put_failure; 2314 break; 2315 case MPTCP_EVENT_SUB_CLOSED: 2316 if (mptcp_event_sub_closed(skb, msk, ssk) < 0) 2317 goto nla_put_failure; 2318 break; 2319 case MPTCP_EVENT_LISTENER_CREATED: 2320 case MPTCP_EVENT_LISTENER_CLOSED: 2321 break; 2322 } 2323 2324 genlmsg_end(skb, nlh); 2325 mptcp_nl_mcast_send(net, skb, gfp); 2326 return; 2327 2328 nla_put_failure: 2329 nlmsg_free(skb); 2330 } 2331 2332 struct genl_family mptcp_genl_family __ro_after_init = { 2333 .name = MPTCP_PM_NAME, 2334 .version = MPTCP_PM_VER, 2335 .netnsok = true, 2336 .module = THIS_MODULE, 2337 .ops = mptcp_pm_nl_ops, 2338 .n_ops = ARRAY_SIZE(mptcp_pm_nl_ops), 2339 .resv_start_op = MPTCP_PM_CMD_SUBFLOW_DESTROY + 1, 2340 .mcgrps = mptcp_pm_mcgrps, 2341 .n_mcgrps = ARRAY_SIZE(mptcp_pm_mcgrps), 2342 }; 2343 2344 static int __net_init pm_nl_init_net(struct net *net) 2345 { 2346 struct pm_nl_pernet *pernet = pm_nl_get_pernet(net); 2347 2348 INIT_LIST_HEAD_RCU(&pernet->local_addr_list); 2349 2350 /* Cit. 2 subflows ought to be enough for anybody. */ 2351 pernet->subflows_max = 2; 2352 pernet->next_id = 1; 2353 pernet->stale_loss_cnt = 4; 2354 spin_lock_init(&pernet->lock); 2355 2356 /* No need to initialize other pernet fields, the struct is zeroed at 2357 * allocation time. 2358 */ 2359 2360 return 0; 2361 } 2362 2363 static void __net_exit pm_nl_exit_net(struct list_head *net_list) 2364 { 2365 struct net *net; 2366 2367 list_for_each_entry(net, net_list, exit_list) { 2368 struct pm_nl_pernet *pernet = pm_nl_get_pernet(net); 2369 2370 /* net is removed from namespace list, can't race with 2371 * other modifiers, also netns core already waited for a 2372 * RCU grace period. 2373 */ 2374 __flush_addrs(&pernet->local_addr_list); 2375 } 2376 } 2377 2378 static struct pernet_operations mptcp_pm_pernet_ops = { 2379 .init = pm_nl_init_net, 2380 .exit_batch = pm_nl_exit_net, 2381 .id = &pm_nl_pernet_id, 2382 .size = sizeof(struct pm_nl_pernet), 2383 }; 2384 2385 void __init mptcp_pm_nl_init(void) 2386 { 2387 if (register_pernet_subsys(&mptcp_pm_pernet_ops) < 0) 2388 panic("Failed to register MPTCP PM pernet subsystem.\n"); 2389 2390 if (genl_register_family(&mptcp_genl_family)) 2391 panic("Failed to register MPTCP PM netlink family\n"); 2392 } 2393