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; 133 134 list_for_each_entry(subflow, list, node) { 135 struct sock *ssk = mptcp_subflow_tcp_sock(subflow); 136 137 if (!((1 << inet_sk_state_load(ssk)) & 138 (TCPF_ESTABLISHED | TCPF_SYN_SENT | TCPF_SYN_RECV))) 139 continue; 140 141 remote_address((struct sock_common *)ssk, &cur); 142 if (mptcp_addresses_equal(&cur, daddr, daddr->port)) 143 return true; 144 } 145 146 return false; 147 } 148 149 static bool 150 select_local_address(const struct pm_nl_pernet *pernet, 151 const struct mptcp_sock *msk, 152 struct mptcp_pm_local *new_local) 153 { 154 struct mptcp_pm_addr_entry *entry; 155 bool found = false; 156 157 msk_owned_by_me(msk); 158 159 rcu_read_lock(); 160 list_for_each_entry_rcu(entry, &pernet->local_addr_list, list) { 161 if (!(entry->flags & MPTCP_PM_ADDR_FLAG_SUBFLOW)) 162 continue; 163 164 if (!test_bit(entry->addr.id, msk->pm.id_avail_bitmap)) 165 continue; 166 167 new_local->addr = entry->addr; 168 new_local->flags = entry->flags; 169 new_local->ifindex = entry->ifindex; 170 found = true; 171 break; 172 } 173 rcu_read_unlock(); 174 175 return found; 176 } 177 178 static bool 179 select_signal_address(struct pm_nl_pernet *pernet, const struct mptcp_sock *msk, 180 struct mptcp_pm_local *new_local) 181 { 182 struct mptcp_pm_addr_entry *entry; 183 bool found = false; 184 185 rcu_read_lock(); 186 /* do not keep any additional per socket state, just signal 187 * the address list in order. 188 * Note: removal from the local address list during the msk life-cycle 189 * can lead to additional addresses not being announced. 190 */ 191 list_for_each_entry_rcu(entry, &pernet->local_addr_list, list) { 192 if (!test_bit(entry->addr.id, msk->pm.id_avail_bitmap)) 193 continue; 194 195 if (!(entry->flags & MPTCP_PM_ADDR_FLAG_SIGNAL)) 196 continue; 197 198 new_local->addr = entry->addr; 199 new_local->flags = entry->flags; 200 new_local->ifindex = entry->ifindex; 201 found = true; 202 break; 203 } 204 rcu_read_unlock(); 205 206 return found; 207 } 208 209 unsigned int mptcp_pm_get_add_addr_signal_max(const struct mptcp_sock *msk) 210 { 211 const struct pm_nl_pernet *pernet = pm_nl_get_pernet_from_msk(msk); 212 213 return READ_ONCE(pernet->add_addr_signal_max); 214 } 215 EXPORT_SYMBOL_GPL(mptcp_pm_get_add_addr_signal_max); 216 217 unsigned int mptcp_pm_get_add_addr_accept_max(const struct mptcp_sock *msk) 218 { 219 struct pm_nl_pernet *pernet = pm_nl_get_pernet_from_msk(msk); 220 221 return READ_ONCE(pernet->add_addr_accept_max); 222 } 223 EXPORT_SYMBOL_GPL(mptcp_pm_get_add_addr_accept_max); 224 225 unsigned int mptcp_pm_get_subflows_max(const struct mptcp_sock *msk) 226 { 227 struct pm_nl_pernet *pernet = pm_nl_get_pernet_from_msk(msk); 228 229 return READ_ONCE(pernet->subflows_max); 230 } 231 EXPORT_SYMBOL_GPL(mptcp_pm_get_subflows_max); 232 233 unsigned int mptcp_pm_get_local_addr_max(const struct mptcp_sock *msk) 234 { 235 struct pm_nl_pernet *pernet = pm_nl_get_pernet_from_msk(msk); 236 237 return READ_ONCE(pernet->local_addr_max); 238 } 239 EXPORT_SYMBOL_GPL(mptcp_pm_get_local_addr_max); 240 241 bool mptcp_pm_nl_check_work_pending(struct mptcp_sock *msk) 242 { 243 struct pm_nl_pernet *pernet = pm_nl_get_pernet_from_msk(msk); 244 245 if (msk->pm.subflows == mptcp_pm_get_subflows_max(msk) || 246 (find_next_and_bit(pernet->id_bitmap, msk->pm.id_avail_bitmap, 247 MPTCP_PM_MAX_ADDR_ID + 1, 0) == MPTCP_PM_MAX_ADDR_ID + 1)) { 248 WRITE_ONCE(msk->pm.work_pending, false); 249 return false; 250 } 251 return true; 252 } 253 254 struct mptcp_pm_add_entry * 255 mptcp_lookup_anno_list_by_saddr(const struct mptcp_sock *msk, 256 const struct mptcp_addr_info *addr) 257 { 258 struct mptcp_pm_add_entry *entry; 259 260 lockdep_assert_held(&msk->pm.lock); 261 262 list_for_each_entry(entry, &msk->pm.anno_list, list) { 263 if (mptcp_addresses_equal(&entry->addr, addr, true)) 264 return entry; 265 } 266 267 return NULL; 268 } 269 270 bool mptcp_pm_sport_in_anno_list(struct mptcp_sock *msk, const struct sock *sk) 271 { 272 struct mptcp_pm_add_entry *entry; 273 struct mptcp_addr_info saddr; 274 bool ret = false; 275 276 mptcp_local_address((struct sock_common *)sk, &saddr); 277 278 spin_lock_bh(&msk->pm.lock); 279 list_for_each_entry(entry, &msk->pm.anno_list, list) { 280 if (mptcp_addresses_equal(&entry->addr, &saddr, true)) { 281 ret = true; 282 goto out; 283 } 284 } 285 286 out: 287 spin_unlock_bh(&msk->pm.lock); 288 return ret; 289 } 290 291 static void mptcp_pm_add_timer(struct timer_list *timer) 292 { 293 struct mptcp_pm_add_entry *entry = from_timer(entry, timer, add_timer); 294 struct mptcp_sock *msk = entry->sock; 295 struct sock *sk = (struct sock *)msk; 296 297 pr_debug("msk=%p\n", msk); 298 299 if (!msk) 300 return; 301 302 if (inet_sk_state_load(sk) == TCP_CLOSE) 303 return; 304 305 if (!entry->addr.id) 306 return; 307 308 if (mptcp_pm_should_add_signal_addr(msk)) { 309 sk_reset_timer(sk, timer, jiffies + TCP_RTO_MAX / 8); 310 goto out; 311 } 312 313 spin_lock_bh(&msk->pm.lock); 314 315 if (!mptcp_pm_should_add_signal_addr(msk)) { 316 pr_debug("retransmit ADD_ADDR id=%d\n", entry->addr.id); 317 mptcp_pm_announce_addr(msk, &entry->addr, false); 318 mptcp_pm_add_addr_send_ack(msk); 319 entry->retrans_times++; 320 } 321 322 if (entry->retrans_times < ADD_ADDR_RETRANS_MAX) 323 sk_reset_timer(sk, timer, 324 jiffies + mptcp_get_add_addr_timeout(sock_net(sk))); 325 326 spin_unlock_bh(&msk->pm.lock); 327 328 if (entry->retrans_times == ADD_ADDR_RETRANS_MAX) 329 mptcp_pm_subflow_established(msk); 330 331 out: 332 __sock_put(sk); 333 } 334 335 struct mptcp_pm_add_entry * 336 mptcp_pm_del_add_timer(struct mptcp_sock *msk, 337 const struct mptcp_addr_info *addr, bool check_id) 338 { 339 struct mptcp_pm_add_entry *entry; 340 struct sock *sk = (struct sock *)msk; 341 struct timer_list *add_timer = NULL; 342 343 spin_lock_bh(&msk->pm.lock); 344 entry = mptcp_lookup_anno_list_by_saddr(msk, addr); 345 if (entry && (!check_id || entry->addr.id == addr->id)) { 346 entry->retrans_times = ADD_ADDR_RETRANS_MAX; 347 add_timer = &entry->add_timer; 348 } 349 if (!check_id && entry) 350 list_del(&entry->list); 351 spin_unlock_bh(&msk->pm.lock); 352 353 /* no lock, because sk_stop_timer_sync() is calling del_timer_sync() */ 354 if (add_timer) 355 sk_stop_timer_sync(sk, add_timer); 356 357 return entry; 358 } 359 360 bool mptcp_pm_alloc_anno_list(struct mptcp_sock *msk, 361 const struct mptcp_addr_info *addr) 362 { 363 struct mptcp_pm_add_entry *add_entry = NULL; 364 struct sock *sk = (struct sock *)msk; 365 struct net *net = sock_net(sk); 366 367 lockdep_assert_held(&msk->pm.lock); 368 369 add_entry = mptcp_lookup_anno_list_by_saddr(msk, addr); 370 371 if (add_entry) { 372 if (WARN_ON_ONCE(mptcp_pm_is_kernel(msk))) 373 return false; 374 375 sk_reset_timer(sk, &add_entry->add_timer, 376 jiffies + mptcp_get_add_addr_timeout(net)); 377 return true; 378 } 379 380 add_entry = kmalloc(sizeof(*add_entry), GFP_ATOMIC); 381 if (!add_entry) 382 return false; 383 384 list_add(&add_entry->list, &msk->pm.anno_list); 385 386 add_entry->addr = *addr; 387 add_entry->sock = msk; 388 add_entry->retrans_times = 0; 389 390 timer_setup(&add_entry->add_timer, mptcp_pm_add_timer, 0); 391 sk_reset_timer(sk, &add_entry->add_timer, 392 jiffies + mptcp_get_add_addr_timeout(net)); 393 394 return true; 395 } 396 397 void mptcp_pm_free_anno_list(struct mptcp_sock *msk) 398 { 399 struct mptcp_pm_add_entry *entry, *tmp; 400 struct sock *sk = (struct sock *)msk; 401 LIST_HEAD(free_list); 402 403 pr_debug("msk=%p\n", msk); 404 405 spin_lock_bh(&msk->pm.lock); 406 list_splice_init(&msk->pm.anno_list, &free_list); 407 spin_unlock_bh(&msk->pm.lock); 408 409 list_for_each_entry_safe(entry, tmp, &free_list, list) { 410 sk_stop_timer_sync(sk, &entry->add_timer); 411 kfree(entry); 412 } 413 } 414 415 /* Fill all the remote addresses into the array addrs[], 416 * and return the array size. 417 */ 418 static unsigned int fill_remote_addresses_vec(struct mptcp_sock *msk, 419 struct mptcp_addr_info *local, 420 bool fullmesh, 421 struct mptcp_addr_info *addrs) 422 { 423 bool deny_id0 = READ_ONCE(msk->pm.remote_deny_join_id0); 424 struct sock *sk = (struct sock *)msk, *ssk; 425 struct mptcp_subflow_context *subflow; 426 struct mptcp_addr_info remote = { 0 }; 427 unsigned int subflows_max; 428 int i = 0; 429 430 subflows_max = mptcp_pm_get_subflows_max(msk); 431 remote_address((struct sock_common *)sk, &remote); 432 433 /* Non-fullmesh endpoint, fill in the single entry 434 * corresponding to the primary MPC subflow remote address 435 */ 436 if (!fullmesh) { 437 if (deny_id0) 438 return 0; 439 440 if (!mptcp_pm_addr_families_match(sk, local, &remote)) 441 return 0; 442 443 msk->pm.subflows++; 444 addrs[i++] = remote; 445 } else { 446 DECLARE_BITMAP(unavail_id, MPTCP_PM_MAX_ADDR_ID + 1); 447 448 /* Forbid creation of new subflows matching existing 449 * ones, possibly already created by incoming ADD_ADDR 450 */ 451 bitmap_zero(unavail_id, MPTCP_PM_MAX_ADDR_ID + 1); 452 mptcp_for_each_subflow(msk, subflow) 453 if (READ_ONCE(subflow->local_id) == local->id) 454 __set_bit(subflow->remote_id, unavail_id); 455 456 mptcp_for_each_subflow(msk, subflow) { 457 ssk = mptcp_subflow_tcp_sock(subflow); 458 remote_address((struct sock_common *)ssk, &addrs[i]); 459 addrs[i].id = READ_ONCE(subflow->remote_id); 460 if (deny_id0 && !addrs[i].id) 461 continue; 462 463 if (test_bit(addrs[i].id, unavail_id)) 464 continue; 465 466 if (!mptcp_pm_addr_families_match(sk, local, &addrs[i])) 467 continue; 468 469 if (msk->pm.subflows < subflows_max) { 470 /* forbid creating multiple address towards 471 * this id 472 */ 473 __set_bit(addrs[i].id, unavail_id); 474 msk->pm.subflows++; 475 i++; 476 } 477 } 478 } 479 480 return i; 481 } 482 483 static void __mptcp_pm_send_ack(struct mptcp_sock *msk, struct mptcp_subflow_context *subflow, 484 bool prio, bool backup) 485 { 486 struct sock *ssk = mptcp_subflow_tcp_sock(subflow); 487 bool slow; 488 489 pr_debug("send ack for %s\n", 490 prio ? "mp_prio" : (mptcp_pm_should_add_signal(msk) ? "add_addr" : "rm_addr")); 491 492 slow = lock_sock_fast(ssk); 493 if (prio) { 494 subflow->send_mp_prio = 1; 495 subflow->request_bkup = backup; 496 } 497 498 __mptcp_subflow_send_ack(ssk); 499 unlock_sock_fast(ssk, slow); 500 } 501 502 static void mptcp_pm_send_ack(struct mptcp_sock *msk, struct mptcp_subflow_context *subflow, 503 bool prio, bool backup) 504 { 505 spin_unlock_bh(&msk->pm.lock); 506 __mptcp_pm_send_ack(msk, subflow, prio, backup); 507 spin_lock_bh(&msk->pm.lock); 508 } 509 510 static struct mptcp_pm_addr_entry * 511 __lookup_addr_by_id(struct pm_nl_pernet *pernet, unsigned int id) 512 { 513 struct mptcp_pm_addr_entry *entry; 514 515 list_for_each_entry(entry, &pernet->local_addr_list, list) { 516 if (entry->addr.id == id) 517 return entry; 518 } 519 return NULL; 520 } 521 522 static struct mptcp_pm_addr_entry * 523 __lookup_addr(struct pm_nl_pernet *pernet, const struct mptcp_addr_info *info) 524 { 525 struct mptcp_pm_addr_entry *entry; 526 527 list_for_each_entry(entry, &pernet->local_addr_list, list) { 528 if (mptcp_addresses_equal(&entry->addr, info, entry->addr.port)) 529 return entry; 530 } 531 return NULL; 532 } 533 534 static void mptcp_pm_create_subflow_or_signal_addr(struct mptcp_sock *msk) 535 { 536 struct sock *sk = (struct sock *)msk; 537 unsigned int add_addr_signal_max; 538 bool signal_and_subflow = false; 539 unsigned int local_addr_max; 540 struct pm_nl_pernet *pernet; 541 struct mptcp_pm_local local; 542 unsigned int subflows_max; 543 544 pernet = pm_nl_get_pernet(sock_net(sk)); 545 546 add_addr_signal_max = mptcp_pm_get_add_addr_signal_max(msk); 547 local_addr_max = mptcp_pm_get_local_addr_max(msk); 548 subflows_max = mptcp_pm_get_subflows_max(msk); 549 550 /* do lazy endpoint usage accounting for the MPC subflows */ 551 if (unlikely(!(msk->pm.status & BIT(MPTCP_PM_MPC_ENDPOINT_ACCOUNTED))) && msk->first) { 552 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(msk->first); 553 struct mptcp_pm_addr_entry *entry; 554 struct mptcp_addr_info mpc_addr; 555 bool backup = false; 556 557 mptcp_local_address((struct sock_common *)msk->first, &mpc_addr); 558 rcu_read_lock(); 559 entry = __lookup_addr(pernet, &mpc_addr); 560 if (entry) { 561 __clear_bit(entry->addr.id, msk->pm.id_avail_bitmap); 562 msk->mpc_endpoint_id = entry->addr.id; 563 backup = !!(entry->flags & MPTCP_PM_ADDR_FLAG_BACKUP); 564 } 565 rcu_read_unlock(); 566 567 if (backup) 568 mptcp_pm_send_ack(msk, subflow, true, backup); 569 570 msk->pm.status |= BIT(MPTCP_PM_MPC_ENDPOINT_ACCOUNTED); 571 } 572 573 pr_debug("local %d:%d signal %d:%d subflows %d:%d\n", 574 msk->pm.local_addr_used, local_addr_max, 575 msk->pm.add_addr_signaled, add_addr_signal_max, 576 msk->pm.subflows, subflows_max); 577 578 /* check first for announce */ 579 if (msk->pm.add_addr_signaled < add_addr_signal_max) { 580 /* due to racing events on both ends we can reach here while 581 * previous add address is still running: if we invoke now 582 * mptcp_pm_announce_addr(), that will fail and the 583 * corresponding id will be marked as used. 584 * Instead let the PM machinery reschedule us when the 585 * current address announce will be completed. 586 */ 587 if (msk->pm.addr_signal & BIT(MPTCP_ADD_ADDR_SIGNAL)) 588 return; 589 590 if (!select_signal_address(pernet, msk, &local)) 591 goto subflow; 592 593 /* If the alloc fails, we are on memory pressure, not worth 594 * continuing, and trying to create subflows. 595 */ 596 if (!mptcp_pm_alloc_anno_list(msk, &local.addr)) 597 return; 598 599 __clear_bit(local.addr.id, msk->pm.id_avail_bitmap); 600 msk->pm.add_addr_signaled++; 601 602 /* Special case for ID0: set the correct ID */ 603 if (local.addr.id == msk->mpc_endpoint_id) 604 local.addr.id = 0; 605 606 mptcp_pm_announce_addr(msk, &local.addr, false); 607 mptcp_pm_nl_addr_send_ack(msk); 608 609 if (local.flags & MPTCP_PM_ADDR_FLAG_SUBFLOW) 610 signal_and_subflow = true; 611 } 612 613 subflow: 614 /* check if should create a new subflow */ 615 while (msk->pm.local_addr_used < local_addr_max && 616 msk->pm.subflows < subflows_max) { 617 struct mptcp_addr_info addrs[MPTCP_PM_ADDR_MAX]; 618 bool fullmesh; 619 int i, nr; 620 621 if (signal_and_subflow) 622 signal_and_subflow = false; 623 else if (!select_local_address(pernet, msk, &local)) 624 break; 625 626 fullmesh = !!(local.flags & MPTCP_PM_ADDR_FLAG_FULLMESH); 627 628 __clear_bit(local.addr.id, msk->pm.id_avail_bitmap); 629 630 /* Special case for ID0: set the correct ID */ 631 if (local.addr.id == msk->mpc_endpoint_id) 632 local.addr.id = 0; 633 else /* local_addr_used is not decr for ID 0 */ 634 msk->pm.local_addr_used++; 635 636 nr = fill_remote_addresses_vec(msk, &local.addr, fullmesh, addrs); 637 if (nr == 0) 638 continue; 639 640 spin_unlock_bh(&msk->pm.lock); 641 for (i = 0; i < nr; i++) 642 __mptcp_subflow_connect(sk, &local, &addrs[i]); 643 spin_lock_bh(&msk->pm.lock); 644 } 645 mptcp_pm_nl_check_work_pending(msk); 646 } 647 648 static void mptcp_pm_nl_fully_established(struct mptcp_sock *msk) 649 { 650 mptcp_pm_create_subflow_or_signal_addr(msk); 651 } 652 653 static void mptcp_pm_nl_subflow_established(struct mptcp_sock *msk) 654 { 655 mptcp_pm_create_subflow_or_signal_addr(msk); 656 } 657 658 /* Fill all the local addresses into the array addrs[], 659 * and return the array size. 660 */ 661 static unsigned int fill_local_addresses_vec(struct mptcp_sock *msk, 662 struct mptcp_addr_info *remote, 663 struct mptcp_pm_local *locals) 664 { 665 struct sock *sk = (struct sock *)msk; 666 struct mptcp_pm_addr_entry *entry; 667 struct mptcp_addr_info mpc_addr; 668 struct pm_nl_pernet *pernet; 669 unsigned int subflows_max; 670 int i = 0; 671 672 pernet = pm_nl_get_pernet_from_msk(msk); 673 subflows_max = mptcp_pm_get_subflows_max(msk); 674 675 mptcp_local_address((struct sock_common *)msk, &mpc_addr); 676 677 rcu_read_lock(); 678 list_for_each_entry_rcu(entry, &pernet->local_addr_list, list) { 679 if (!(entry->flags & MPTCP_PM_ADDR_FLAG_FULLMESH)) 680 continue; 681 682 if (!mptcp_pm_addr_families_match(sk, &entry->addr, remote)) 683 continue; 684 685 if (msk->pm.subflows < subflows_max) { 686 locals[i].addr = entry->addr; 687 locals[i].flags = entry->flags; 688 locals[i].ifindex = entry->ifindex; 689 690 /* Special case for ID0: set the correct ID */ 691 if (mptcp_addresses_equal(&locals[i].addr, &mpc_addr, locals[i].addr.port)) 692 locals[i].addr.id = 0; 693 694 msk->pm.subflows++; 695 i++; 696 } 697 } 698 rcu_read_unlock(); 699 700 /* If the array is empty, fill in the single 701 * 'IPADDRANY' local address 702 */ 703 if (!i) { 704 memset(&locals[i], 0, sizeof(locals[i])); 705 locals[i].addr.family = 706 #if IS_ENABLED(CONFIG_MPTCP_IPV6) 707 remote->family == AF_INET6 && 708 ipv6_addr_v4mapped(&remote->addr6) ? AF_INET : 709 #endif 710 remote->family; 711 712 if (!mptcp_pm_addr_families_match(sk, &locals[i].addr, remote)) 713 return 0; 714 715 msk->pm.subflows++; 716 i++; 717 } 718 719 return i; 720 } 721 722 static void mptcp_pm_nl_add_addr_received(struct mptcp_sock *msk) 723 { 724 struct mptcp_pm_local locals[MPTCP_PM_ADDR_MAX]; 725 struct sock *sk = (struct sock *)msk; 726 unsigned int add_addr_accept_max; 727 struct mptcp_addr_info remote; 728 unsigned int subflows_max; 729 bool sf_created = false; 730 int i, nr; 731 732 add_addr_accept_max = mptcp_pm_get_add_addr_accept_max(msk); 733 subflows_max = mptcp_pm_get_subflows_max(msk); 734 735 pr_debug("accepted %d:%d remote family %d\n", 736 msk->pm.add_addr_accepted, add_addr_accept_max, 737 msk->pm.remote.family); 738 739 remote = msk->pm.remote; 740 mptcp_pm_announce_addr(msk, &remote, true); 741 mptcp_pm_nl_addr_send_ack(msk); 742 743 if (lookup_subflow_by_daddr(&msk->conn_list, &remote)) 744 return; 745 746 /* pick id 0 port, if none is provided the remote address */ 747 if (!remote.port) 748 remote.port = sk->sk_dport; 749 750 /* connect to the specified remote address, using whatever 751 * local address the routing configuration will pick. 752 */ 753 nr = fill_local_addresses_vec(msk, &remote, locals); 754 if (nr == 0) 755 return; 756 757 spin_unlock_bh(&msk->pm.lock); 758 for (i = 0; i < nr; i++) 759 if (__mptcp_subflow_connect(sk, &locals[i], &remote) == 0) 760 sf_created = true; 761 spin_lock_bh(&msk->pm.lock); 762 763 if (sf_created) { 764 /* add_addr_accepted is not decr for ID 0 */ 765 if (remote.id) 766 msk->pm.add_addr_accepted++; 767 if (msk->pm.add_addr_accepted >= add_addr_accept_max || 768 msk->pm.subflows >= subflows_max) 769 WRITE_ONCE(msk->pm.accept_addr, false); 770 } 771 } 772 773 bool mptcp_pm_nl_is_init_remote_addr(struct mptcp_sock *msk, 774 const struct mptcp_addr_info *remote) 775 { 776 struct mptcp_addr_info mpc_remote; 777 778 remote_address((struct sock_common *)msk, &mpc_remote); 779 return mptcp_addresses_equal(&mpc_remote, remote, remote->port); 780 } 781 782 void mptcp_pm_nl_addr_send_ack(struct mptcp_sock *msk) 783 { 784 struct mptcp_subflow_context *subflow, *alt = NULL; 785 786 msk_owned_by_me(msk); 787 lockdep_assert_held(&msk->pm.lock); 788 789 if (!mptcp_pm_should_add_signal(msk) && 790 !mptcp_pm_should_rm_signal(msk)) 791 return; 792 793 mptcp_for_each_subflow(msk, subflow) { 794 if (__mptcp_subflow_active(subflow)) { 795 if (!subflow->stale) { 796 mptcp_pm_send_ack(msk, subflow, false, false); 797 return; 798 } 799 800 if (!alt) 801 alt = subflow; 802 } 803 } 804 805 if (alt) 806 mptcp_pm_send_ack(msk, alt, false, false); 807 } 808 809 int mptcp_pm_nl_mp_prio_send_ack(struct mptcp_sock *msk, 810 struct mptcp_addr_info *addr, 811 struct mptcp_addr_info *rem, 812 u8 bkup) 813 { 814 struct mptcp_subflow_context *subflow; 815 816 pr_debug("bkup=%d\n", bkup); 817 818 mptcp_for_each_subflow(msk, subflow) { 819 struct sock *ssk = mptcp_subflow_tcp_sock(subflow); 820 struct mptcp_addr_info local, remote; 821 822 mptcp_local_address((struct sock_common *)ssk, &local); 823 if (!mptcp_addresses_equal(&local, addr, addr->port)) 824 continue; 825 826 if (rem && rem->family != AF_UNSPEC) { 827 remote_address((struct sock_common *)ssk, &remote); 828 if (!mptcp_addresses_equal(&remote, rem, rem->port)) 829 continue; 830 } 831 832 __mptcp_pm_send_ack(msk, subflow, true, bkup); 833 return 0; 834 } 835 836 return -EINVAL; 837 } 838 839 static void mptcp_pm_nl_rm_addr_or_subflow(struct mptcp_sock *msk, 840 const struct mptcp_rm_list *rm_list, 841 enum linux_mptcp_mib_field rm_type) 842 { 843 struct mptcp_subflow_context *subflow, *tmp; 844 struct sock *sk = (struct sock *)msk; 845 u8 i; 846 847 pr_debug("%s rm_list_nr %d\n", 848 rm_type == MPTCP_MIB_RMADDR ? "address" : "subflow", rm_list->nr); 849 850 msk_owned_by_me(msk); 851 852 if (sk->sk_state == TCP_LISTEN) 853 return; 854 855 if (!rm_list->nr) 856 return; 857 858 if (list_empty(&msk->conn_list)) 859 return; 860 861 for (i = 0; i < rm_list->nr; i++) { 862 u8 rm_id = rm_list->ids[i]; 863 bool removed = false; 864 865 mptcp_for_each_subflow_safe(msk, subflow, tmp) { 866 struct sock *ssk = mptcp_subflow_tcp_sock(subflow); 867 u8 remote_id = READ_ONCE(subflow->remote_id); 868 int how = RCV_SHUTDOWN | SEND_SHUTDOWN; 869 u8 id = subflow_get_local_id(subflow); 870 871 if ((1 << inet_sk_state_load(ssk)) & 872 (TCPF_FIN_WAIT1 | TCPF_FIN_WAIT2 | TCPF_CLOSING | TCPF_CLOSE)) 873 continue; 874 if (rm_type == MPTCP_MIB_RMADDR && remote_id != rm_id) 875 continue; 876 if (rm_type == MPTCP_MIB_RMSUBFLOW && id != rm_id) 877 continue; 878 879 pr_debug(" -> %s rm_list_ids[%d]=%u local_id=%u remote_id=%u mpc_id=%u\n", 880 rm_type == MPTCP_MIB_RMADDR ? "address" : "subflow", 881 i, rm_id, id, remote_id, msk->mpc_endpoint_id); 882 spin_unlock_bh(&msk->pm.lock); 883 mptcp_subflow_shutdown(sk, ssk, how); 884 removed |= subflow->request_join; 885 886 /* the following takes care of updating the subflows counter */ 887 mptcp_close_ssk(sk, ssk, subflow); 888 spin_lock_bh(&msk->pm.lock); 889 890 if (rm_type == MPTCP_MIB_RMSUBFLOW) 891 __MPTCP_INC_STATS(sock_net(sk), rm_type); 892 } 893 894 if (rm_type == MPTCP_MIB_RMADDR) 895 __MPTCP_INC_STATS(sock_net(sk), rm_type); 896 897 if (!removed) 898 continue; 899 900 if (!mptcp_pm_is_kernel(msk)) 901 continue; 902 903 if (rm_type == MPTCP_MIB_RMADDR && rm_id && 904 !WARN_ON_ONCE(msk->pm.add_addr_accepted == 0)) { 905 /* Note: if the subflow has been closed before, this 906 * add_addr_accepted counter will not be decremented. 907 */ 908 if (--msk->pm.add_addr_accepted < mptcp_pm_get_add_addr_accept_max(msk)) 909 WRITE_ONCE(msk->pm.accept_addr, true); 910 } 911 } 912 } 913 914 static void mptcp_pm_nl_rm_addr_received(struct mptcp_sock *msk) 915 { 916 mptcp_pm_nl_rm_addr_or_subflow(msk, &msk->pm.rm_list_rx, MPTCP_MIB_RMADDR); 917 } 918 919 static void mptcp_pm_nl_rm_subflow_received(struct mptcp_sock *msk, 920 const struct mptcp_rm_list *rm_list) 921 { 922 mptcp_pm_nl_rm_addr_or_subflow(msk, rm_list, MPTCP_MIB_RMSUBFLOW); 923 } 924 925 void mptcp_pm_nl_work(struct mptcp_sock *msk) 926 { 927 struct mptcp_pm_data *pm = &msk->pm; 928 929 msk_owned_by_me(msk); 930 931 if (!(pm->status & MPTCP_PM_WORK_MASK)) 932 return; 933 934 spin_lock_bh(&msk->pm.lock); 935 936 pr_debug("msk=%p status=%x\n", msk, pm->status); 937 if (pm->status & BIT(MPTCP_PM_ADD_ADDR_RECEIVED)) { 938 pm->status &= ~BIT(MPTCP_PM_ADD_ADDR_RECEIVED); 939 mptcp_pm_nl_add_addr_received(msk); 940 } 941 if (pm->status & BIT(MPTCP_PM_ADD_ADDR_SEND_ACK)) { 942 pm->status &= ~BIT(MPTCP_PM_ADD_ADDR_SEND_ACK); 943 mptcp_pm_nl_addr_send_ack(msk); 944 } 945 if (pm->status & BIT(MPTCP_PM_RM_ADDR_RECEIVED)) { 946 pm->status &= ~BIT(MPTCP_PM_RM_ADDR_RECEIVED); 947 mptcp_pm_nl_rm_addr_received(msk); 948 } 949 if (pm->status & BIT(MPTCP_PM_ESTABLISHED)) { 950 pm->status &= ~BIT(MPTCP_PM_ESTABLISHED); 951 mptcp_pm_nl_fully_established(msk); 952 } 953 if (pm->status & BIT(MPTCP_PM_SUBFLOW_ESTABLISHED)) { 954 pm->status &= ~BIT(MPTCP_PM_SUBFLOW_ESTABLISHED); 955 mptcp_pm_nl_subflow_established(msk); 956 } 957 958 spin_unlock_bh(&msk->pm.lock); 959 } 960 961 static bool address_use_port(struct mptcp_pm_addr_entry *entry) 962 { 963 return (entry->flags & 964 (MPTCP_PM_ADDR_FLAG_SIGNAL | MPTCP_PM_ADDR_FLAG_SUBFLOW)) == 965 MPTCP_PM_ADDR_FLAG_SIGNAL; 966 } 967 968 /* caller must ensure the RCU grace period is already elapsed */ 969 static void __mptcp_pm_release_addr_entry(struct mptcp_pm_addr_entry *entry) 970 { 971 if (entry->lsk) 972 sock_release(entry->lsk); 973 kfree(entry); 974 } 975 976 static int mptcp_pm_nl_append_new_local_addr(struct pm_nl_pernet *pernet, 977 struct mptcp_pm_addr_entry *entry, 978 bool needs_id) 979 { 980 struct mptcp_pm_addr_entry *cur, *del_entry = NULL; 981 unsigned int addr_max; 982 int ret = -EINVAL; 983 984 spin_lock_bh(&pernet->lock); 985 /* to keep the code simple, don't do IDR-like allocation for address ID, 986 * just bail when we exceed limits 987 */ 988 if (pernet->next_id == MPTCP_PM_MAX_ADDR_ID) 989 pernet->next_id = 1; 990 if (pernet->addrs >= MPTCP_PM_ADDR_MAX) { 991 ret = -ERANGE; 992 goto out; 993 } 994 if (test_bit(entry->addr.id, pernet->id_bitmap)) { 995 ret = -EBUSY; 996 goto out; 997 } 998 999 /* do not insert duplicate address, differentiate on port only 1000 * singled addresses 1001 */ 1002 if (!address_use_port(entry)) 1003 entry->addr.port = 0; 1004 list_for_each_entry(cur, &pernet->local_addr_list, list) { 1005 if (mptcp_addresses_equal(&cur->addr, &entry->addr, 1006 cur->addr.port || entry->addr.port)) { 1007 /* allow replacing the exiting endpoint only if such 1008 * endpoint is an implicit one and the user-space 1009 * did not provide an endpoint id 1010 */ 1011 if (!(cur->flags & MPTCP_PM_ADDR_FLAG_IMPLICIT)) { 1012 ret = -EEXIST; 1013 goto out; 1014 } 1015 if (entry->addr.id) 1016 goto out; 1017 1018 pernet->addrs--; 1019 entry->addr.id = cur->addr.id; 1020 list_del_rcu(&cur->list); 1021 del_entry = cur; 1022 break; 1023 } 1024 } 1025 1026 if (!entry->addr.id && needs_id) { 1027 find_next: 1028 entry->addr.id = find_next_zero_bit(pernet->id_bitmap, 1029 MPTCP_PM_MAX_ADDR_ID + 1, 1030 pernet->next_id); 1031 if (!entry->addr.id && pernet->next_id != 1) { 1032 pernet->next_id = 1; 1033 goto find_next; 1034 } 1035 } 1036 1037 if (!entry->addr.id && needs_id) 1038 goto out; 1039 1040 __set_bit(entry->addr.id, pernet->id_bitmap); 1041 if (entry->addr.id > pernet->next_id) 1042 pernet->next_id = entry->addr.id; 1043 1044 if (entry->flags & MPTCP_PM_ADDR_FLAG_SIGNAL) { 1045 addr_max = pernet->add_addr_signal_max; 1046 WRITE_ONCE(pernet->add_addr_signal_max, addr_max + 1); 1047 } 1048 if (entry->flags & MPTCP_PM_ADDR_FLAG_SUBFLOW) { 1049 addr_max = pernet->local_addr_max; 1050 WRITE_ONCE(pernet->local_addr_max, addr_max + 1); 1051 } 1052 1053 pernet->addrs++; 1054 if (!entry->addr.port) 1055 list_add_tail_rcu(&entry->list, &pernet->local_addr_list); 1056 else 1057 list_add_rcu(&entry->list, &pernet->local_addr_list); 1058 ret = entry->addr.id; 1059 1060 out: 1061 spin_unlock_bh(&pernet->lock); 1062 1063 /* just replaced an existing entry, free it */ 1064 if (del_entry) { 1065 synchronize_rcu(); 1066 __mptcp_pm_release_addr_entry(del_entry); 1067 } 1068 return ret; 1069 } 1070 1071 static struct lock_class_key mptcp_slock_keys[2]; 1072 static struct lock_class_key mptcp_keys[2]; 1073 1074 static int mptcp_pm_nl_create_listen_socket(struct sock *sk, 1075 struct mptcp_pm_addr_entry *entry) 1076 { 1077 bool is_ipv6 = sk->sk_family == AF_INET6; 1078 int addrlen = sizeof(struct sockaddr_in); 1079 struct sockaddr_storage addr; 1080 struct sock *newsk, *ssk; 1081 int backlog = 1024; 1082 int err; 1083 1084 err = sock_create_kern(sock_net(sk), entry->addr.family, 1085 SOCK_STREAM, IPPROTO_MPTCP, &entry->lsk); 1086 if (err) 1087 return err; 1088 1089 newsk = entry->lsk->sk; 1090 if (!newsk) 1091 return -EINVAL; 1092 1093 /* The subflow socket lock is acquired in a nested to the msk one 1094 * in several places, even by the TCP stack, and this msk is a kernel 1095 * socket: lockdep complains. Instead of propagating the _nested 1096 * modifiers in several places, re-init the lock class for the msk 1097 * socket to an mptcp specific one. 1098 */ 1099 sock_lock_init_class_and_name(newsk, 1100 is_ipv6 ? "mlock-AF_INET6" : "mlock-AF_INET", 1101 &mptcp_slock_keys[is_ipv6], 1102 is_ipv6 ? "msk_lock-AF_INET6" : "msk_lock-AF_INET", 1103 &mptcp_keys[is_ipv6]); 1104 1105 lock_sock(newsk); 1106 ssk = __mptcp_nmpc_sk(mptcp_sk(newsk)); 1107 release_sock(newsk); 1108 if (IS_ERR(ssk)) 1109 return PTR_ERR(ssk); 1110 1111 mptcp_info2sockaddr(&entry->addr, &addr, entry->addr.family); 1112 #if IS_ENABLED(CONFIG_MPTCP_IPV6) 1113 if (entry->addr.family == AF_INET6) 1114 addrlen = sizeof(struct sockaddr_in6); 1115 #endif 1116 if (ssk->sk_family == AF_INET) 1117 err = inet_bind_sk(ssk, (struct sockaddr *)&addr, addrlen); 1118 #if IS_ENABLED(CONFIG_MPTCP_IPV6) 1119 else if (ssk->sk_family == AF_INET6) 1120 err = inet6_bind_sk(ssk, (struct sockaddr *)&addr, addrlen); 1121 #endif 1122 if (err) 1123 return err; 1124 1125 /* We don't use mptcp_set_state() here because it needs to be called 1126 * under the msk socket lock. For the moment, that will not bring 1127 * anything more than only calling inet_sk_state_store(), because the 1128 * old status is known (TCP_CLOSE). 1129 */ 1130 inet_sk_state_store(newsk, TCP_LISTEN); 1131 lock_sock(ssk); 1132 WRITE_ONCE(mptcp_subflow_ctx(ssk)->pm_listener, true); 1133 err = __inet_listen_sk(ssk, backlog); 1134 if (!err) 1135 mptcp_event_pm_listener(ssk, MPTCP_EVENT_LISTENER_CREATED); 1136 release_sock(ssk); 1137 return err; 1138 } 1139 1140 int mptcp_pm_nl_get_local_id(struct mptcp_sock *msk, struct mptcp_addr_info *skc) 1141 { 1142 struct mptcp_pm_addr_entry *entry; 1143 struct pm_nl_pernet *pernet; 1144 int ret = -1; 1145 1146 pernet = pm_nl_get_pernet_from_msk(msk); 1147 1148 rcu_read_lock(); 1149 list_for_each_entry_rcu(entry, &pernet->local_addr_list, list) { 1150 if (mptcp_addresses_equal(&entry->addr, skc, entry->addr.port)) { 1151 ret = entry->addr.id; 1152 break; 1153 } 1154 } 1155 rcu_read_unlock(); 1156 if (ret >= 0) 1157 return ret; 1158 1159 /* address not found, add to local list */ 1160 entry = kmalloc(sizeof(*entry), GFP_ATOMIC); 1161 if (!entry) 1162 return -ENOMEM; 1163 1164 entry->addr = *skc; 1165 entry->addr.id = 0; 1166 entry->addr.port = 0; 1167 entry->ifindex = 0; 1168 entry->flags = MPTCP_PM_ADDR_FLAG_IMPLICIT; 1169 entry->lsk = NULL; 1170 ret = mptcp_pm_nl_append_new_local_addr(pernet, entry, true); 1171 if (ret < 0) 1172 kfree(entry); 1173 1174 return ret; 1175 } 1176 1177 bool mptcp_pm_nl_is_backup(struct mptcp_sock *msk, struct mptcp_addr_info *skc) 1178 { 1179 struct pm_nl_pernet *pernet = pm_nl_get_pernet_from_msk(msk); 1180 struct mptcp_pm_addr_entry *entry; 1181 bool backup = false; 1182 1183 rcu_read_lock(); 1184 list_for_each_entry_rcu(entry, &pernet->local_addr_list, list) { 1185 if (mptcp_addresses_equal(&entry->addr, skc, entry->addr.port)) { 1186 backup = !!(entry->flags & MPTCP_PM_ADDR_FLAG_BACKUP); 1187 break; 1188 } 1189 } 1190 rcu_read_unlock(); 1191 1192 return backup; 1193 } 1194 1195 #define MPTCP_PM_CMD_GRP_OFFSET 0 1196 #define MPTCP_PM_EV_GRP_OFFSET 1 1197 1198 static const struct genl_multicast_group mptcp_pm_mcgrps[] = { 1199 [MPTCP_PM_CMD_GRP_OFFSET] = { .name = MPTCP_PM_CMD_GRP_NAME, }, 1200 [MPTCP_PM_EV_GRP_OFFSET] = { .name = MPTCP_PM_EV_GRP_NAME, 1201 .flags = GENL_MCAST_CAP_NET_ADMIN, 1202 }, 1203 }; 1204 1205 void mptcp_pm_nl_subflow_chk_stale(const struct mptcp_sock *msk, struct sock *ssk) 1206 { 1207 struct mptcp_subflow_context *iter, *subflow = mptcp_subflow_ctx(ssk); 1208 struct sock *sk = (struct sock *)msk; 1209 unsigned int active_max_loss_cnt; 1210 struct net *net = sock_net(sk); 1211 unsigned int stale_loss_cnt; 1212 bool slow; 1213 1214 stale_loss_cnt = mptcp_stale_loss_cnt(net); 1215 if (subflow->stale || !stale_loss_cnt || subflow->stale_count <= stale_loss_cnt) 1216 return; 1217 1218 /* look for another available subflow not in loss state */ 1219 active_max_loss_cnt = max_t(int, stale_loss_cnt - 1, 1); 1220 mptcp_for_each_subflow(msk, iter) { 1221 if (iter != subflow && mptcp_subflow_active(iter) && 1222 iter->stale_count < active_max_loss_cnt) { 1223 /* we have some alternatives, try to mark this subflow as idle ...*/ 1224 slow = lock_sock_fast(ssk); 1225 if (!tcp_rtx_and_write_queues_empty(ssk)) { 1226 subflow->stale = 1; 1227 __mptcp_retransmit_pending_data(sk); 1228 MPTCP_INC_STATS(net, MPTCP_MIB_SUBFLOWSTALE); 1229 } 1230 unlock_sock_fast(ssk, slow); 1231 1232 /* always try to push the pending data regardless of re-injections: 1233 * we can possibly use backup subflows now, and subflow selection 1234 * is cheap under the msk socket lock 1235 */ 1236 __mptcp_push_pending(sk, 0); 1237 return; 1238 } 1239 } 1240 } 1241 1242 static int mptcp_pm_family_to_addr(int family) 1243 { 1244 #if IS_ENABLED(CONFIG_MPTCP_IPV6) 1245 if (family == AF_INET6) 1246 return MPTCP_PM_ADDR_ATTR_ADDR6; 1247 #endif 1248 return MPTCP_PM_ADDR_ATTR_ADDR4; 1249 } 1250 1251 static int mptcp_pm_parse_pm_addr_attr(struct nlattr *tb[], 1252 const struct nlattr *attr, 1253 struct genl_info *info, 1254 struct mptcp_addr_info *addr, 1255 bool require_family) 1256 { 1257 int err, addr_addr; 1258 1259 if (!attr) { 1260 GENL_SET_ERR_MSG(info, "missing address info"); 1261 return -EINVAL; 1262 } 1263 1264 /* no validation needed - was already done via nested policy */ 1265 err = nla_parse_nested_deprecated(tb, MPTCP_PM_ADDR_ATTR_MAX, attr, 1266 mptcp_pm_address_nl_policy, info->extack); 1267 if (err) 1268 return err; 1269 1270 if (tb[MPTCP_PM_ADDR_ATTR_ID]) 1271 addr->id = nla_get_u8(tb[MPTCP_PM_ADDR_ATTR_ID]); 1272 1273 if (!tb[MPTCP_PM_ADDR_ATTR_FAMILY]) { 1274 if (!require_family) 1275 return 0; 1276 1277 NL_SET_ERR_MSG_ATTR(info->extack, attr, 1278 "missing family"); 1279 return -EINVAL; 1280 } 1281 1282 addr->family = nla_get_u16(tb[MPTCP_PM_ADDR_ATTR_FAMILY]); 1283 if (addr->family != AF_INET 1284 #if IS_ENABLED(CONFIG_MPTCP_IPV6) 1285 && addr->family != AF_INET6 1286 #endif 1287 ) { 1288 NL_SET_ERR_MSG_ATTR(info->extack, attr, 1289 "unknown address family"); 1290 return -EINVAL; 1291 } 1292 addr_addr = mptcp_pm_family_to_addr(addr->family); 1293 if (!tb[addr_addr]) { 1294 NL_SET_ERR_MSG_ATTR(info->extack, attr, 1295 "missing address data"); 1296 return -EINVAL; 1297 } 1298 1299 #if IS_ENABLED(CONFIG_MPTCP_IPV6) 1300 if (addr->family == AF_INET6) 1301 addr->addr6 = nla_get_in6_addr(tb[addr_addr]); 1302 else 1303 #endif 1304 addr->addr.s_addr = nla_get_in_addr(tb[addr_addr]); 1305 1306 if (tb[MPTCP_PM_ADDR_ATTR_PORT]) 1307 addr->port = htons(nla_get_u16(tb[MPTCP_PM_ADDR_ATTR_PORT])); 1308 1309 return 0; 1310 } 1311 1312 int mptcp_pm_parse_addr(struct nlattr *attr, struct genl_info *info, 1313 struct mptcp_addr_info *addr) 1314 { 1315 struct nlattr *tb[MPTCP_PM_ADDR_ATTR_MAX + 1]; 1316 1317 memset(addr, 0, sizeof(*addr)); 1318 1319 return mptcp_pm_parse_pm_addr_attr(tb, attr, info, addr, true); 1320 } 1321 1322 int mptcp_pm_parse_entry(struct nlattr *attr, struct genl_info *info, 1323 bool require_family, 1324 struct mptcp_pm_addr_entry *entry) 1325 { 1326 struct nlattr *tb[MPTCP_PM_ADDR_ATTR_MAX + 1]; 1327 int err; 1328 1329 memset(entry, 0, sizeof(*entry)); 1330 1331 err = mptcp_pm_parse_pm_addr_attr(tb, attr, info, &entry->addr, require_family); 1332 if (err) 1333 return err; 1334 1335 if (tb[MPTCP_PM_ADDR_ATTR_IF_IDX]) { 1336 u32 val = nla_get_s32(tb[MPTCP_PM_ADDR_ATTR_IF_IDX]); 1337 1338 entry->ifindex = val; 1339 } 1340 1341 if (tb[MPTCP_PM_ADDR_ATTR_FLAGS]) 1342 entry->flags = nla_get_u32(tb[MPTCP_PM_ADDR_ATTR_FLAGS]); 1343 1344 if (tb[MPTCP_PM_ADDR_ATTR_PORT]) 1345 entry->addr.port = htons(nla_get_u16(tb[MPTCP_PM_ADDR_ATTR_PORT])); 1346 1347 return 0; 1348 } 1349 1350 static struct pm_nl_pernet *genl_info_pm_nl(struct genl_info *info) 1351 { 1352 return pm_nl_get_pernet(genl_info_net(info)); 1353 } 1354 1355 static int mptcp_nl_add_subflow_or_signal_addr(struct net *net, 1356 struct mptcp_addr_info *addr) 1357 { 1358 struct mptcp_sock *msk; 1359 long s_slot = 0, s_num = 0; 1360 1361 while ((msk = mptcp_token_iter_next(net, &s_slot, &s_num)) != NULL) { 1362 struct sock *sk = (struct sock *)msk; 1363 struct mptcp_addr_info mpc_addr; 1364 1365 if (!READ_ONCE(msk->fully_established) || 1366 mptcp_pm_is_userspace(msk)) 1367 goto next; 1368 1369 /* if the endp linked to the init sf is re-added with a != ID */ 1370 mptcp_local_address((struct sock_common *)msk, &mpc_addr); 1371 1372 lock_sock(sk); 1373 spin_lock_bh(&msk->pm.lock); 1374 if (mptcp_addresses_equal(addr, &mpc_addr, addr->port)) 1375 msk->mpc_endpoint_id = addr->id; 1376 mptcp_pm_create_subflow_or_signal_addr(msk); 1377 spin_unlock_bh(&msk->pm.lock); 1378 release_sock(sk); 1379 1380 next: 1381 sock_put(sk); 1382 cond_resched(); 1383 } 1384 1385 return 0; 1386 } 1387 1388 static bool mptcp_pm_has_addr_attr_id(const struct nlattr *attr, 1389 struct genl_info *info) 1390 { 1391 struct nlattr *tb[MPTCP_PM_ADDR_ATTR_MAX + 1]; 1392 1393 if (!nla_parse_nested_deprecated(tb, MPTCP_PM_ADDR_ATTR_MAX, attr, 1394 mptcp_pm_address_nl_policy, info->extack) && 1395 tb[MPTCP_PM_ADDR_ATTR_ID]) 1396 return true; 1397 return false; 1398 } 1399 1400 int mptcp_pm_nl_add_addr_doit(struct sk_buff *skb, struct genl_info *info) 1401 { 1402 struct nlattr *attr = info->attrs[MPTCP_PM_ENDPOINT_ADDR]; 1403 struct pm_nl_pernet *pernet = genl_info_pm_nl(info); 1404 struct mptcp_pm_addr_entry addr, *entry; 1405 int ret; 1406 1407 ret = mptcp_pm_parse_entry(attr, info, true, &addr); 1408 if (ret < 0) 1409 return ret; 1410 1411 if (addr.addr.port && !address_use_port(&addr)) { 1412 GENL_SET_ERR_MSG(info, "flags must have signal and not subflow when using port"); 1413 return -EINVAL; 1414 } 1415 1416 if (addr.flags & MPTCP_PM_ADDR_FLAG_SIGNAL && 1417 addr.flags & MPTCP_PM_ADDR_FLAG_FULLMESH) { 1418 GENL_SET_ERR_MSG(info, "flags mustn't have both signal and fullmesh"); 1419 return -EINVAL; 1420 } 1421 1422 if (addr.flags & MPTCP_PM_ADDR_FLAG_IMPLICIT) { 1423 GENL_SET_ERR_MSG(info, "can't create IMPLICIT endpoint"); 1424 return -EINVAL; 1425 } 1426 1427 entry = kzalloc(sizeof(*entry), GFP_KERNEL_ACCOUNT); 1428 if (!entry) { 1429 GENL_SET_ERR_MSG(info, "can't allocate addr"); 1430 return -ENOMEM; 1431 } 1432 1433 *entry = addr; 1434 if (entry->addr.port) { 1435 ret = mptcp_pm_nl_create_listen_socket(skb->sk, entry); 1436 if (ret) { 1437 GENL_SET_ERR_MSG_FMT(info, "create listen socket error: %d", ret); 1438 goto out_free; 1439 } 1440 } 1441 ret = mptcp_pm_nl_append_new_local_addr(pernet, entry, 1442 !mptcp_pm_has_addr_attr_id(attr, info)); 1443 if (ret < 0) { 1444 GENL_SET_ERR_MSG_FMT(info, "too many addresses or duplicate one: %d", ret); 1445 goto out_free; 1446 } 1447 1448 mptcp_nl_add_subflow_or_signal_addr(sock_net(skb->sk), &entry->addr); 1449 return 0; 1450 1451 out_free: 1452 __mptcp_pm_release_addr_entry(entry); 1453 return ret; 1454 } 1455 1456 static bool remove_anno_list_by_saddr(struct mptcp_sock *msk, 1457 const struct mptcp_addr_info *addr) 1458 { 1459 struct mptcp_pm_add_entry *entry; 1460 1461 entry = mptcp_pm_del_add_timer(msk, addr, false); 1462 if (entry) { 1463 kfree(entry); 1464 return true; 1465 } 1466 1467 return false; 1468 } 1469 1470 static u8 mptcp_endp_get_local_id(struct mptcp_sock *msk, 1471 const struct mptcp_addr_info *addr) 1472 { 1473 return msk->mpc_endpoint_id == addr->id ? 0 : addr->id; 1474 } 1475 1476 static bool mptcp_pm_remove_anno_addr(struct mptcp_sock *msk, 1477 const struct mptcp_addr_info *addr, 1478 bool force) 1479 { 1480 struct mptcp_rm_list list = { .nr = 0 }; 1481 bool ret; 1482 1483 list.ids[list.nr++] = mptcp_endp_get_local_id(msk, addr); 1484 1485 ret = remove_anno_list_by_saddr(msk, addr); 1486 if (ret || force) { 1487 spin_lock_bh(&msk->pm.lock); 1488 if (ret) { 1489 __set_bit(addr->id, msk->pm.id_avail_bitmap); 1490 msk->pm.add_addr_signaled--; 1491 } 1492 mptcp_pm_remove_addr(msk, &list); 1493 spin_unlock_bh(&msk->pm.lock); 1494 } 1495 return ret; 1496 } 1497 1498 static void __mark_subflow_endp_available(struct mptcp_sock *msk, u8 id) 1499 { 1500 /* If it was marked as used, and not ID 0, decrement local_addr_used */ 1501 if (!__test_and_set_bit(id ? : msk->mpc_endpoint_id, msk->pm.id_avail_bitmap) && 1502 id && !WARN_ON_ONCE(msk->pm.local_addr_used == 0)) 1503 msk->pm.local_addr_used--; 1504 } 1505 1506 static int mptcp_nl_remove_subflow_and_signal_addr(struct net *net, 1507 const struct mptcp_pm_addr_entry *entry) 1508 { 1509 const struct mptcp_addr_info *addr = &entry->addr; 1510 struct mptcp_rm_list list = { .nr = 1 }; 1511 long s_slot = 0, s_num = 0; 1512 struct mptcp_sock *msk; 1513 1514 pr_debug("remove_id=%d\n", addr->id); 1515 1516 while ((msk = mptcp_token_iter_next(net, &s_slot, &s_num)) != NULL) { 1517 struct sock *sk = (struct sock *)msk; 1518 bool remove_subflow; 1519 1520 if (mptcp_pm_is_userspace(msk)) 1521 goto next; 1522 1523 if (list_empty(&msk->conn_list)) { 1524 mptcp_pm_remove_anno_addr(msk, addr, false); 1525 goto next; 1526 } 1527 1528 lock_sock(sk); 1529 remove_subflow = lookup_subflow_by_saddr(&msk->conn_list, addr); 1530 mptcp_pm_remove_anno_addr(msk, addr, remove_subflow && 1531 !(entry->flags & MPTCP_PM_ADDR_FLAG_IMPLICIT)); 1532 1533 list.ids[0] = mptcp_endp_get_local_id(msk, addr); 1534 if (remove_subflow) { 1535 spin_lock_bh(&msk->pm.lock); 1536 mptcp_pm_nl_rm_subflow_received(msk, &list); 1537 spin_unlock_bh(&msk->pm.lock); 1538 } 1539 1540 if (entry->flags & MPTCP_PM_ADDR_FLAG_SUBFLOW) { 1541 spin_lock_bh(&msk->pm.lock); 1542 __mark_subflow_endp_available(msk, list.ids[0]); 1543 spin_unlock_bh(&msk->pm.lock); 1544 } 1545 1546 if (msk->mpc_endpoint_id == entry->addr.id) 1547 msk->mpc_endpoint_id = 0; 1548 release_sock(sk); 1549 1550 next: 1551 sock_put(sk); 1552 cond_resched(); 1553 } 1554 1555 return 0; 1556 } 1557 1558 static int mptcp_nl_remove_id_zero_address(struct net *net, 1559 struct mptcp_addr_info *addr) 1560 { 1561 struct mptcp_rm_list list = { .nr = 0 }; 1562 long s_slot = 0, s_num = 0; 1563 struct mptcp_sock *msk; 1564 1565 list.ids[list.nr++] = 0; 1566 1567 while ((msk = mptcp_token_iter_next(net, &s_slot, &s_num)) != NULL) { 1568 struct sock *sk = (struct sock *)msk; 1569 struct mptcp_addr_info msk_local; 1570 1571 if (list_empty(&msk->conn_list) || mptcp_pm_is_userspace(msk)) 1572 goto next; 1573 1574 mptcp_local_address((struct sock_common *)msk, &msk_local); 1575 if (!mptcp_addresses_equal(&msk_local, addr, addr->port)) 1576 goto next; 1577 1578 lock_sock(sk); 1579 spin_lock_bh(&msk->pm.lock); 1580 mptcp_pm_remove_addr(msk, &list); 1581 mptcp_pm_nl_rm_subflow_received(msk, &list); 1582 __mark_subflow_endp_available(msk, 0); 1583 spin_unlock_bh(&msk->pm.lock); 1584 release_sock(sk); 1585 1586 next: 1587 sock_put(sk); 1588 cond_resched(); 1589 } 1590 1591 return 0; 1592 } 1593 1594 int mptcp_pm_nl_del_addr_doit(struct sk_buff *skb, struct genl_info *info) 1595 { 1596 struct nlattr *attr = info->attrs[MPTCP_PM_ENDPOINT_ADDR]; 1597 struct pm_nl_pernet *pernet = genl_info_pm_nl(info); 1598 struct mptcp_pm_addr_entry addr, *entry; 1599 unsigned int addr_max; 1600 int ret; 1601 1602 ret = mptcp_pm_parse_entry(attr, info, false, &addr); 1603 if (ret < 0) 1604 return ret; 1605 1606 /* the zero id address is special: the first address used by the msk 1607 * always gets such an id, so different subflows can have different zero 1608 * id addresses. Additionally zero id is not accounted for in id_bitmap. 1609 * Let's use an 'mptcp_rm_list' instead of the common remove code. 1610 */ 1611 if (addr.addr.id == 0) 1612 return mptcp_nl_remove_id_zero_address(sock_net(skb->sk), &addr.addr); 1613 1614 spin_lock_bh(&pernet->lock); 1615 entry = __lookup_addr_by_id(pernet, addr.addr.id); 1616 if (!entry) { 1617 GENL_SET_ERR_MSG(info, "address not found"); 1618 spin_unlock_bh(&pernet->lock); 1619 return -EINVAL; 1620 } 1621 if (entry->flags & MPTCP_PM_ADDR_FLAG_SIGNAL) { 1622 addr_max = pernet->add_addr_signal_max; 1623 WRITE_ONCE(pernet->add_addr_signal_max, addr_max - 1); 1624 } 1625 if (entry->flags & MPTCP_PM_ADDR_FLAG_SUBFLOW) { 1626 addr_max = pernet->local_addr_max; 1627 WRITE_ONCE(pernet->local_addr_max, addr_max - 1); 1628 } 1629 1630 pernet->addrs--; 1631 list_del_rcu(&entry->list); 1632 __clear_bit(entry->addr.id, pernet->id_bitmap); 1633 spin_unlock_bh(&pernet->lock); 1634 1635 mptcp_nl_remove_subflow_and_signal_addr(sock_net(skb->sk), entry); 1636 synchronize_rcu(); 1637 __mptcp_pm_release_addr_entry(entry); 1638 1639 return ret; 1640 } 1641 1642 /* Called from the userspace PM only */ 1643 void mptcp_pm_remove_addrs(struct mptcp_sock *msk, struct list_head *rm_list) 1644 { 1645 struct mptcp_rm_list alist = { .nr = 0 }; 1646 struct mptcp_pm_addr_entry *entry; 1647 int anno_nr = 0; 1648 1649 list_for_each_entry(entry, rm_list, list) { 1650 if (alist.nr >= MPTCP_RM_IDS_MAX) 1651 break; 1652 1653 /* only delete if either announced or matching a subflow */ 1654 if (remove_anno_list_by_saddr(msk, &entry->addr)) 1655 anno_nr++; 1656 else if (!lookup_subflow_by_saddr(&msk->conn_list, 1657 &entry->addr)) 1658 continue; 1659 1660 alist.ids[alist.nr++] = entry->addr.id; 1661 } 1662 1663 if (alist.nr) { 1664 spin_lock_bh(&msk->pm.lock); 1665 msk->pm.add_addr_signaled -= anno_nr; 1666 mptcp_pm_remove_addr(msk, &alist); 1667 spin_unlock_bh(&msk->pm.lock); 1668 } 1669 } 1670 1671 /* Called from the in-kernel PM only */ 1672 static void mptcp_pm_flush_addrs_and_subflows(struct mptcp_sock *msk, 1673 struct list_head *rm_list) 1674 { 1675 struct mptcp_rm_list alist = { .nr = 0 }, slist = { .nr = 0 }; 1676 struct mptcp_pm_addr_entry *entry; 1677 1678 list_for_each_entry(entry, rm_list, list) { 1679 if (slist.nr < MPTCP_RM_IDS_MAX && 1680 lookup_subflow_by_saddr(&msk->conn_list, &entry->addr)) 1681 slist.ids[slist.nr++] = mptcp_endp_get_local_id(msk, &entry->addr); 1682 1683 if (alist.nr < MPTCP_RM_IDS_MAX && 1684 remove_anno_list_by_saddr(msk, &entry->addr)) 1685 alist.ids[alist.nr++] = mptcp_endp_get_local_id(msk, &entry->addr); 1686 } 1687 1688 spin_lock_bh(&msk->pm.lock); 1689 if (alist.nr) { 1690 msk->pm.add_addr_signaled -= alist.nr; 1691 mptcp_pm_remove_addr(msk, &alist); 1692 } 1693 if (slist.nr) 1694 mptcp_pm_nl_rm_subflow_received(msk, &slist); 1695 /* Reset counters: maybe some subflows have been removed before */ 1696 bitmap_fill(msk->pm.id_avail_bitmap, MPTCP_PM_MAX_ADDR_ID + 1); 1697 msk->pm.local_addr_used = 0; 1698 spin_unlock_bh(&msk->pm.lock); 1699 } 1700 1701 static void mptcp_nl_flush_addrs_list(struct net *net, 1702 struct list_head *rm_list) 1703 { 1704 long s_slot = 0, s_num = 0; 1705 struct mptcp_sock *msk; 1706 1707 if (list_empty(rm_list)) 1708 return; 1709 1710 while ((msk = mptcp_token_iter_next(net, &s_slot, &s_num)) != NULL) { 1711 struct sock *sk = (struct sock *)msk; 1712 1713 if (!mptcp_pm_is_userspace(msk)) { 1714 lock_sock(sk); 1715 mptcp_pm_flush_addrs_and_subflows(msk, rm_list); 1716 release_sock(sk); 1717 } 1718 1719 sock_put(sk); 1720 cond_resched(); 1721 } 1722 } 1723 1724 /* caller must ensure the RCU grace period is already elapsed */ 1725 static void __flush_addrs(struct list_head *list) 1726 { 1727 while (!list_empty(list)) { 1728 struct mptcp_pm_addr_entry *cur; 1729 1730 cur = list_entry(list->next, 1731 struct mptcp_pm_addr_entry, list); 1732 list_del_rcu(&cur->list); 1733 __mptcp_pm_release_addr_entry(cur); 1734 } 1735 } 1736 1737 static void __reset_counters(struct pm_nl_pernet *pernet) 1738 { 1739 WRITE_ONCE(pernet->add_addr_signal_max, 0); 1740 WRITE_ONCE(pernet->add_addr_accept_max, 0); 1741 WRITE_ONCE(pernet->local_addr_max, 0); 1742 pernet->addrs = 0; 1743 } 1744 1745 int mptcp_pm_nl_flush_addrs_doit(struct sk_buff *skb, struct genl_info *info) 1746 { 1747 struct pm_nl_pernet *pernet = genl_info_pm_nl(info); 1748 LIST_HEAD(free_list); 1749 1750 spin_lock_bh(&pernet->lock); 1751 list_splice_init(&pernet->local_addr_list, &free_list); 1752 __reset_counters(pernet); 1753 pernet->next_id = 1; 1754 bitmap_zero(pernet->id_bitmap, MPTCP_PM_MAX_ADDR_ID + 1); 1755 spin_unlock_bh(&pernet->lock); 1756 mptcp_nl_flush_addrs_list(sock_net(skb->sk), &free_list); 1757 synchronize_rcu(); 1758 __flush_addrs(&free_list); 1759 return 0; 1760 } 1761 1762 int mptcp_nl_fill_addr(struct sk_buff *skb, 1763 struct mptcp_pm_addr_entry *entry) 1764 { 1765 struct mptcp_addr_info *addr = &entry->addr; 1766 struct nlattr *attr; 1767 1768 attr = nla_nest_start(skb, MPTCP_PM_ATTR_ADDR); 1769 if (!attr) 1770 return -EMSGSIZE; 1771 1772 if (nla_put_u16(skb, MPTCP_PM_ADDR_ATTR_FAMILY, addr->family)) 1773 goto nla_put_failure; 1774 if (nla_put_u16(skb, MPTCP_PM_ADDR_ATTR_PORT, ntohs(addr->port))) 1775 goto nla_put_failure; 1776 if (nla_put_u8(skb, MPTCP_PM_ADDR_ATTR_ID, addr->id)) 1777 goto nla_put_failure; 1778 if (nla_put_u32(skb, MPTCP_PM_ADDR_ATTR_FLAGS, entry->flags)) 1779 goto nla_put_failure; 1780 if (entry->ifindex && 1781 nla_put_s32(skb, MPTCP_PM_ADDR_ATTR_IF_IDX, entry->ifindex)) 1782 goto nla_put_failure; 1783 1784 if (addr->family == AF_INET && 1785 nla_put_in_addr(skb, MPTCP_PM_ADDR_ATTR_ADDR4, 1786 addr->addr.s_addr)) 1787 goto nla_put_failure; 1788 #if IS_ENABLED(CONFIG_MPTCP_IPV6) 1789 else if (addr->family == AF_INET6 && 1790 nla_put_in6_addr(skb, MPTCP_PM_ADDR_ATTR_ADDR6, &addr->addr6)) 1791 goto nla_put_failure; 1792 #endif 1793 nla_nest_end(skb, attr); 1794 return 0; 1795 1796 nla_put_failure: 1797 nla_nest_cancel(skb, attr); 1798 return -EMSGSIZE; 1799 } 1800 1801 int mptcp_pm_nl_get_addr(struct sk_buff *skb, struct genl_info *info) 1802 { 1803 struct nlattr *attr = info->attrs[MPTCP_PM_ENDPOINT_ADDR]; 1804 struct pm_nl_pernet *pernet = genl_info_pm_nl(info); 1805 struct mptcp_pm_addr_entry addr, *entry; 1806 struct sk_buff *msg; 1807 void *reply; 1808 int ret; 1809 1810 ret = mptcp_pm_parse_entry(attr, info, false, &addr); 1811 if (ret < 0) 1812 return ret; 1813 1814 msg = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL); 1815 if (!msg) 1816 return -ENOMEM; 1817 1818 reply = genlmsg_put_reply(msg, info, &mptcp_genl_family, 0, 1819 info->genlhdr->cmd); 1820 if (!reply) { 1821 GENL_SET_ERR_MSG(info, "not enough space in Netlink message"); 1822 ret = -EMSGSIZE; 1823 goto fail; 1824 } 1825 1826 spin_lock_bh(&pernet->lock); 1827 entry = __lookup_addr_by_id(pernet, addr.addr.id); 1828 if (!entry) { 1829 GENL_SET_ERR_MSG(info, "address not found"); 1830 ret = -EINVAL; 1831 goto unlock_fail; 1832 } 1833 1834 ret = mptcp_nl_fill_addr(msg, entry); 1835 if (ret) 1836 goto unlock_fail; 1837 1838 genlmsg_end(msg, reply); 1839 ret = genlmsg_reply(msg, info); 1840 spin_unlock_bh(&pernet->lock); 1841 return ret; 1842 1843 unlock_fail: 1844 spin_unlock_bh(&pernet->lock); 1845 1846 fail: 1847 nlmsg_free(msg); 1848 return ret; 1849 } 1850 1851 int mptcp_pm_nl_get_addr_doit(struct sk_buff *skb, struct genl_info *info) 1852 { 1853 return mptcp_pm_get_addr(skb, info); 1854 } 1855 1856 int mptcp_pm_nl_dump_addr(struct sk_buff *msg, 1857 struct netlink_callback *cb) 1858 { 1859 struct net *net = sock_net(msg->sk); 1860 struct mptcp_pm_addr_entry *entry; 1861 struct pm_nl_pernet *pernet; 1862 int id = cb->args[0]; 1863 void *hdr; 1864 int i; 1865 1866 pernet = pm_nl_get_pernet(net); 1867 1868 spin_lock_bh(&pernet->lock); 1869 for (i = id; i < MPTCP_PM_MAX_ADDR_ID + 1; i++) { 1870 if (test_bit(i, pernet->id_bitmap)) { 1871 entry = __lookup_addr_by_id(pernet, i); 1872 if (!entry) 1873 break; 1874 1875 if (entry->addr.id <= id) 1876 continue; 1877 1878 hdr = genlmsg_put(msg, NETLINK_CB(cb->skb).portid, 1879 cb->nlh->nlmsg_seq, &mptcp_genl_family, 1880 NLM_F_MULTI, MPTCP_PM_CMD_GET_ADDR); 1881 if (!hdr) 1882 break; 1883 1884 if (mptcp_nl_fill_addr(msg, entry) < 0) { 1885 genlmsg_cancel(msg, hdr); 1886 break; 1887 } 1888 1889 id = entry->addr.id; 1890 genlmsg_end(msg, hdr); 1891 } 1892 } 1893 spin_unlock_bh(&pernet->lock); 1894 1895 cb->args[0] = id; 1896 return msg->len; 1897 } 1898 1899 int mptcp_pm_nl_get_addr_dumpit(struct sk_buff *msg, 1900 struct netlink_callback *cb) 1901 { 1902 return mptcp_pm_dump_addr(msg, cb); 1903 } 1904 1905 static int parse_limit(struct genl_info *info, int id, unsigned int *limit) 1906 { 1907 struct nlattr *attr = info->attrs[id]; 1908 1909 if (!attr) 1910 return 0; 1911 1912 *limit = nla_get_u32(attr); 1913 if (*limit > MPTCP_PM_ADDR_MAX) { 1914 GENL_SET_ERR_MSG(info, "limit greater than maximum"); 1915 return -EINVAL; 1916 } 1917 return 0; 1918 } 1919 1920 int mptcp_pm_nl_set_limits_doit(struct sk_buff *skb, struct genl_info *info) 1921 { 1922 struct pm_nl_pernet *pernet = genl_info_pm_nl(info); 1923 unsigned int rcv_addrs, subflows; 1924 int ret; 1925 1926 spin_lock_bh(&pernet->lock); 1927 rcv_addrs = pernet->add_addr_accept_max; 1928 ret = parse_limit(info, MPTCP_PM_ATTR_RCV_ADD_ADDRS, &rcv_addrs); 1929 if (ret) 1930 goto unlock; 1931 1932 subflows = pernet->subflows_max; 1933 ret = parse_limit(info, MPTCP_PM_ATTR_SUBFLOWS, &subflows); 1934 if (ret) 1935 goto unlock; 1936 1937 WRITE_ONCE(pernet->add_addr_accept_max, rcv_addrs); 1938 WRITE_ONCE(pernet->subflows_max, subflows); 1939 1940 unlock: 1941 spin_unlock_bh(&pernet->lock); 1942 return ret; 1943 } 1944 1945 int mptcp_pm_nl_get_limits_doit(struct sk_buff *skb, struct genl_info *info) 1946 { 1947 struct pm_nl_pernet *pernet = genl_info_pm_nl(info); 1948 struct sk_buff *msg; 1949 void *reply; 1950 1951 msg = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL); 1952 if (!msg) 1953 return -ENOMEM; 1954 1955 reply = genlmsg_put_reply(msg, info, &mptcp_genl_family, 0, 1956 MPTCP_PM_CMD_GET_LIMITS); 1957 if (!reply) 1958 goto fail; 1959 1960 if (nla_put_u32(msg, MPTCP_PM_ATTR_RCV_ADD_ADDRS, 1961 READ_ONCE(pernet->add_addr_accept_max))) 1962 goto fail; 1963 1964 if (nla_put_u32(msg, MPTCP_PM_ATTR_SUBFLOWS, 1965 READ_ONCE(pernet->subflows_max))) 1966 goto fail; 1967 1968 genlmsg_end(msg, reply); 1969 return genlmsg_reply(msg, info); 1970 1971 fail: 1972 GENL_SET_ERR_MSG(info, "not enough space in Netlink message"); 1973 nlmsg_free(msg); 1974 return -EMSGSIZE; 1975 } 1976 1977 static void mptcp_pm_nl_fullmesh(struct mptcp_sock *msk, 1978 struct mptcp_addr_info *addr) 1979 { 1980 struct mptcp_rm_list list = { .nr = 0 }; 1981 1982 list.ids[list.nr++] = mptcp_endp_get_local_id(msk, addr); 1983 1984 spin_lock_bh(&msk->pm.lock); 1985 mptcp_pm_nl_rm_subflow_received(msk, &list); 1986 __mark_subflow_endp_available(msk, list.ids[0]); 1987 mptcp_pm_create_subflow_or_signal_addr(msk); 1988 spin_unlock_bh(&msk->pm.lock); 1989 } 1990 1991 static int mptcp_nl_set_flags(struct net *net, 1992 struct mptcp_addr_info *addr, 1993 u8 bkup, u8 changed) 1994 { 1995 long s_slot = 0, s_num = 0; 1996 struct mptcp_sock *msk; 1997 int ret = -EINVAL; 1998 1999 while ((msk = mptcp_token_iter_next(net, &s_slot, &s_num)) != NULL) { 2000 struct sock *sk = (struct sock *)msk; 2001 2002 if (list_empty(&msk->conn_list) || mptcp_pm_is_userspace(msk)) 2003 goto next; 2004 2005 lock_sock(sk); 2006 if (changed & MPTCP_PM_ADDR_FLAG_BACKUP) 2007 ret = mptcp_pm_nl_mp_prio_send_ack(msk, addr, NULL, bkup); 2008 if (changed & MPTCP_PM_ADDR_FLAG_FULLMESH) 2009 mptcp_pm_nl_fullmesh(msk, addr); 2010 release_sock(sk); 2011 2012 next: 2013 sock_put(sk); 2014 cond_resched(); 2015 } 2016 2017 return ret; 2018 } 2019 2020 int mptcp_pm_nl_set_flags(struct sk_buff *skb, struct genl_info *info) 2021 { 2022 struct mptcp_pm_addr_entry addr = { .addr = { .family = AF_UNSPEC }, }; 2023 struct nlattr *attr = info->attrs[MPTCP_PM_ATTR_ADDR]; 2024 u8 changed, mask = MPTCP_PM_ADDR_FLAG_BACKUP | 2025 MPTCP_PM_ADDR_FLAG_FULLMESH; 2026 struct net *net = sock_net(skb->sk); 2027 struct mptcp_pm_addr_entry *entry; 2028 struct pm_nl_pernet *pernet; 2029 u8 lookup_by_id = 0; 2030 u8 bkup = 0; 2031 int ret; 2032 2033 pernet = pm_nl_get_pernet(net); 2034 2035 ret = mptcp_pm_parse_entry(attr, info, false, &addr); 2036 if (ret < 0) 2037 return ret; 2038 2039 if (addr.addr.family == AF_UNSPEC) { 2040 lookup_by_id = 1; 2041 if (!addr.addr.id) { 2042 GENL_SET_ERR_MSG(info, "missing required inputs"); 2043 return -EOPNOTSUPP; 2044 } 2045 } 2046 2047 if (addr.flags & MPTCP_PM_ADDR_FLAG_BACKUP) 2048 bkup = 1; 2049 2050 spin_lock_bh(&pernet->lock); 2051 entry = lookup_by_id ? __lookup_addr_by_id(pernet, addr.addr.id) : 2052 __lookup_addr(pernet, &addr.addr); 2053 if (!entry) { 2054 spin_unlock_bh(&pernet->lock); 2055 GENL_SET_ERR_MSG(info, "address not found"); 2056 return -EINVAL; 2057 } 2058 if ((addr.flags & MPTCP_PM_ADDR_FLAG_FULLMESH) && 2059 (entry->flags & MPTCP_PM_ADDR_FLAG_SIGNAL)) { 2060 spin_unlock_bh(&pernet->lock); 2061 GENL_SET_ERR_MSG(info, "invalid addr flags"); 2062 return -EINVAL; 2063 } 2064 2065 changed = (addr.flags ^ entry->flags) & mask; 2066 entry->flags = (entry->flags & ~mask) | (addr.flags & mask); 2067 addr = *entry; 2068 spin_unlock_bh(&pernet->lock); 2069 2070 mptcp_nl_set_flags(net, &addr.addr, bkup, changed); 2071 return 0; 2072 } 2073 2074 int mptcp_pm_nl_set_flags_doit(struct sk_buff *skb, struct genl_info *info) 2075 { 2076 return mptcp_pm_set_flags(skb, info); 2077 } 2078 2079 static void mptcp_nl_mcast_send(struct net *net, struct sk_buff *nlskb, gfp_t gfp) 2080 { 2081 genlmsg_multicast_netns(&mptcp_genl_family, net, 2082 nlskb, 0, MPTCP_PM_EV_GRP_OFFSET, gfp); 2083 } 2084 2085 bool mptcp_userspace_pm_active(const struct mptcp_sock *msk) 2086 { 2087 return genl_has_listeners(&mptcp_genl_family, 2088 sock_net((const struct sock *)msk), 2089 MPTCP_PM_EV_GRP_OFFSET); 2090 } 2091 2092 static int mptcp_event_add_subflow(struct sk_buff *skb, const struct sock *ssk) 2093 { 2094 const struct inet_sock *issk = inet_sk(ssk); 2095 const struct mptcp_subflow_context *sf; 2096 2097 if (nla_put_u16(skb, MPTCP_ATTR_FAMILY, ssk->sk_family)) 2098 return -EMSGSIZE; 2099 2100 switch (ssk->sk_family) { 2101 case AF_INET: 2102 if (nla_put_in_addr(skb, MPTCP_ATTR_SADDR4, issk->inet_saddr)) 2103 return -EMSGSIZE; 2104 if (nla_put_in_addr(skb, MPTCP_ATTR_DADDR4, issk->inet_daddr)) 2105 return -EMSGSIZE; 2106 break; 2107 #if IS_ENABLED(CONFIG_MPTCP_IPV6) 2108 case AF_INET6: { 2109 const struct ipv6_pinfo *np = inet6_sk(ssk); 2110 2111 if (nla_put_in6_addr(skb, MPTCP_ATTR_SADDR6, &np->saddr)) 2112 return -EMSGSIZE; 2113 if (nla_put_in6_addr(skb, MPTCP_ATTR_DADDR6, &ssk->sk_v6_daddr)) 2114 return -EMSGSIZE; 2115 break; 2116 } 2117 #endif 2118 default: 2119 WARN_ON_ONCE(1); 2120 return -EMSGSIZE; 2121 } 2122 2123 if (nla_put_be16(skb, MPTCP_ATTR_SPORT, issk->inet_sport)) 2124 return -EMSGSIZE; 2125 if (nla_put_be16(skb, MPTCP_ATTR_DPORT, issk->inet_dport)) 2126 return -EMSGSIZE; 2127 2128 sf = mptcp_subflow_ctx(ssk); 2129 if (WARN_ON_ONCE(!sf)) 2130 return -EINVAL; 2131 2132 if (nla_put_u8(skb, MPTCP_ATTR_LOC_ID, subflow_get_local_id(sf))) 2133 return -EMSGSIZE; 2134 2135 if (nla_put_u8(skb, MPTCP_ATTR_REM_ID, sf->remote_id)) 2136 return -EMSGSIZE; 2137 2138 return 0; 2139 } 2140 2141 static int mptcp_event_put_token_and_ssk(struct sk_buff *skb, 2142 const struct mptcp_sock *msk, 2143 const struct sock *ssk) 2144 { 2145 const struct sock *sk = (const struct sock *)msk; 2146 const struct mptcp_subflow_context *sf; 2147 u8 sk_err; 2148 2149 if (nla_put_u32(skb, MPTCP_ATTR_TOKEN, READ_ONCE(msk->token))) 2150 return -EMSGSIZE; 2151 2152 if (mptcp_event_add_subflow(skb, ssk)) 2153 return -EMSGSIZE; 2154 2155 sf = mptcp_subflow_ctx(ssk); 2156 if (WARN_ON_ONCE(!sf)) 2157 return -EINVAL; 2158 2159 if (nla_put_u8(skb, MPTCP_ATTR_BACKUP, sf->backup)) 2160 return -EMSGSIZE; 2161 2162 if (ssk->sk_bound_dev_if && 2163 nla_put_s32(skb, MPTCP_ATTR_IF_IDX, ssk->sk_bound_dev_if)) 2164 return -EMSGSIZE; 2165 2166 sk_err = READ_ONCE(ssk->sk_err); 2167 if (sk_err && sk->sk_state == TCP_ESTABLISHED && 2168 nla_put_u8(skb, MPTCP_ATTR_ERROR, sk_err)) 2169 return -EMSGSIZE; 2170 2171 return 0; 2172 } 2173 2174 static int mptcp_event_sub_established(struct sk_buff *skb, 2175 const struct mptcp_sock *msk, 2176 const struct sock *ssk) 2177 { 2178 return mptcp_event_put_token_and_ssk(skb, msk, ssk); 2179 } 2180 2181 static int mptcp_event_sub_closed(struct sk_buff *skb, 2182 const struct mptcp_sock *msk, 2183 const struct sock *ssk) 2184 { 2185 const struct mptcp_subflow_context *sf; 2186 2187 if (mptcp_event_put_token_and_ssk(skb, msk, ssk)) 2188 return -EMSGSIZE; 2189 2190 sf = mptcp_subflow_ctx(ssk); 2191 if (!sf->reset_seen) 2192 return 0; 2193 2194 if (nla_put_u32(skb, MPTCP_ATTR_RESET_REASON, sf->reset_reason)) 2195 return -EMSGSIZE; 2196 2197 if (nla_put_u32(skb, MPTCP_ATTR_RESET_FLAGS, sf->reset_transient)) 2198 return -EMSGSIZE; 2199 2200 return 0; 2201 } 2202 2203 static int mptcp_event_created(struct sk_buff *skb, 2204 const struct mptcp_sock *msk, 2205 const struct sock *ssk) 2206 { 2207 int err = nla_put_u32(skb, MPTCP_ATTR_TOKEN, READ_ONCE(msk->token)); 2208 2209 if (err) 2210 return err; 2211 2212 if (nla_put_u8(skb, MPTCP_ATTR_SERVER_SIDE, READ_ONCE(msk->pm.server_side))) 2213 return -EMSGSIZE; 2214 2215 return mptcp_event_add_subflow(skb, ssk); 2216 } 2217 2218 void mptcp_event_addr_removed(const struct mptcp_sock *msk, uint8_t id) 2219 { 2220 struct net *net = sock_net((const struct sock *)msk); 2221 struct nlmsghdr *nlh; 2222 struct sk_buff *skb; 2223 2224 if (!genl_has_listeners(&mptcp_genl_family, net, MPTCP_PM_EV_GRP_OFFSET)) 2225 return; 2226 2227 skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_ATOMIC); 2228 if (!skb) 2229 return; 2230 2231 nlh = genlmsg_put(skb, 0, 0, &mptcp_genl_family, 0, MPTCP_EVENT_REMOVED); 2232 if (!nlh) 2233 goto nla_put_failure; 2234 2235 if (nla_put_u32(skb, MPTCP_ATTR_TOKEN, READ_ONCE(msk->token))) 2236 goto nla_put_failure; 2237 2238 if (nla_put_u8(skb, MPTCP_ATTR_REM_ID, id)) 2239 goto nla_put_failure; 2240 2241 genlmsg_end(skb, nlh); 2242 mptcp_nl_mcast_send(net, skb, GFP_ATOMIC); 2243 return; 2244 2245 nla_put_failure: 2246 nlmsg_free(skb); 2247 } 2248 2249 void mptcp_event_addr_announced(const struct sock *ssk, 2250 const struct mptcp_addr_info *info) 2251 { 2252 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk); 2253 struct mptcp_sock *msk = mptcp_sk(subflow->conn); 2254 struct net *net = sock_net(ssk); 2255 struct nlmsghdr *nlh; 2256 struct sk_buff *skb; 2257 2258 if (!genl_has_listeners(&mptcp_genl_family, net, MPTCP_PM_EV_GRP_OFFSET)) 2259 return; 2260 2261 skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_ATOMIC); 2262 if (!skb) 2263 return; 2264 2265 nlh = genlmsg_put(skb, 0, 0, &mptcp_genl_family, 0, 2266 MPTCP_EVENT_ANNOUNCED); 2267 if (!nlh) 2268 goto nla_put_failure; 2269 2270 if (nla_put_u32(skb, MPTCP_ATTR_TOKEN, READ_ONCE(msk->token))) 2271 goto nla_put_failure; 2272 2273 if (nla_put_u8(skb, MPTCP_ATTR_REM_ID, info->id)) 2274 goto nla_put_failure; 2275 2276 if (nla_put_be16(skb, MPTCP_ATTR_DPORT, 2277 info->port == 0 ? 2278 inet_sk(ssk)->inet_dport : 2279 info->port)) 2280 goto nla_put_failure; 2281 2282 switch (info->family) { 2283 case AF_INET: 2284 if (nla_put_in_addr(skb, MPTCP_ATTR_DADDR4, info->addr.s_addr)) 2285 goto nla_put_failure; 2286 break; 2287 #if IS_ENABLED(CONFIG_MPTCP_IPV6) 2288 case AF_INET6: 2289 if (nla_put_in6_addr(skb, MPTCP_ATTR_DADDR6, &info->addr6)) 2290 goto nla_put_failure; 2291 break; 2292 #endif 2293 default: 2294 WARN_ON_ONCE(1); 2295 goto nla_put_failure; 2296 } 2297 2298 genlmsg_end(skb, nlh); 2299 mptcp_nl_mcast_send(net, skb, GFP_ATOMIC); 2300 return; 2301 2302 nla_put_failure: 2303 nlmsg_free(skb); 2304 } 2305 2306 void mptcp_event_pm_listener(const struct sock *ssk, 2307 enum mptcp_event_type event) 2308 { 2309 const struct inet_sock *issk = inet_sk(ssk); 2310 struct net *net = sock_net(ssk); 2311 struct nlmsghdr *nlh; 2312 struct sk_buff *skb; 2313 2314 if (!genl_has_listeners(&mptcp_genl_family, net, MPTCP_PM_EV_GRP_OFFSET)) 2315 return; 2316 2317 skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL); 2318 if (!skb) 2319 return; 2320 2321 nlh = genlmsg_put(skb, 0, 0, &mptcp_genl_family, 0, event); 2322 if (!nlh) 2323 goto nla_put_failure; 2324 2325 if (nla_put_u16(skb, MPTCP_ATTR_FAMILY, ssk->sk_family)) 2326 goto nla_put_failure; 2327 2328 if (nla_put_be16(skb, MPTCP_ATTR_SPORT, issk->inet_sport)) 2329 goto nla_put_failure; 2330 2331 switch (ssk->sk_family) { 2332 case AF_INET: 2333 if (nla_put_in_addr(skb, MPTCP_ATTR_SADDR4, issk->inet_saddr)) 2334 goto nla_put_failure; 2335 break; 2336 #if IS_ENABLED(CONFIG_MPTCP_IPV6) 2337 case AF_INET6: { 2338 const struct ipv6_pinfo *np = inet6_sk(ssk); 2339 2340 if (nla_put_in6_addr(skb, MPTCP_ATTR_SADDR6, &np->saddr)) 2341 goto nla_put_failure; 2342 break; 2343 } 2344 #endif 2345 default: 2346 WARN_ON_ONCE(1); 2347 goto nla_put_failure; 2348 } 2349 2350 genlmsg_end(skb, nlh); 2351 mptcp_nl_mcast_send(net, skb, GFP_KERNEL); 2352 return; 2353 2354 nla_put_failure: 2355 nlmsg_free(skb); 2356 } 2357 2358 void mptcp_event(enum mptcp_event_type type, const struct mptcp_sock *msk, 2359 const struct sock *ssk, gfp_t gfp) 2360 { 2361 struct net *net = sock_net((const struct sock *)msk); 2362 struct nlmsghdr *nlh; 2363 struct sk_buff *skb; 2364 2365 if (!genl_has_listeners(&mptcp_genl_family, net, MPTCP_PM_EV_GRP_OFFSET)) 2366 return; 2367 2368 skb = nlmsg_new(NLMSG_DEFAULT_SIZE, gfp); 2369 if (!skb) 2370 return; 2371 2372 nlh = genlmsg_put(skb, 0, 0, &mptcp_genl_family, 0, type); 2373 if (!nlh) 2374 goto nla_put_failure; 2375 2376 switch (type) { 2377 case MPTCP_EVENT_UNSPEC: 2378 WARN_ON_ONCE(1); 2379 break; 2380 case MPTCP_EVENT_CREATED: 2381 case MPTCP_EVENT_ESTABLISHED: 2382 if (mptcp_event_created(skb, msk, ssk) < 0) 2383 goto nla_put_failure; 2384 break; 2385 case MPTCP_EVENT_CLOSED: 2386 if (nla_put_u32(skb, MPTCP_ATTR_TOKEN, READ_ONCE(msk->token)) < 0) 2387 goto nla_put_failure; 2388 break; 2389 case MPTCP_EVENT_ANNOUNCED: 2390 case MPTCP_EVENT_REMOVED: 2391 /* call mptcp_event_addr_announced()/removed instead */ 2392 WARN_ON_ONCE(1); 2393 break; 2394 case MPTCP_EVENT_SUB_ESTABLISHED: 2395 case MPTCP_EVENT_SUB_PRIORITY: 2396 if (mptcp_event_sub_established(skb, msk, ssk) < 0) 2397 goto nla_put_failure; 2398 break; 2399 case MPTCP_EVENT_SUB_CLOSED: 2400 if (mptcp_event_sub_closed(skb, msk, ssk) < 0) 2401 goto nla_put_failure; 2402 break; 2403 case MPTCP_EVENT_LISTENER_CREATED: 2404 case MPTCP_EVENT_LISTENER_CLOSED: 2405 break; 2406 } 2407 2408 genlmsg_end(skb, nlh); 2409 mptcp_nl_mcast_send(net, skb, gfp); 2410 return; 2411 2412 nla_put_failure: 2413 nlmsg_free(skb); 2414 } 2415 2416 struct genl_family mptcp_genl_family __ro_after_init = { 2417 .name = MPTCP_PM_NAME, 2418 .version = MPTCP_PM_VER, 2419 .netnsok = true, 2420 .module = THIS_MODULE, 2421 .ops = mptcp_pm_nl_ops, 2422 .n_ops = ARRAY_SIZE(mptcp_pm_nl_ops), 2423 .resv_start_op = MPTCP_PM_CMD_SUBFLOW_DESTROY + 1, 2424 .mcgrps = mptcp_pm_mcgrps, 2425 .n_mcgrps = ARRAY_SIZE(mptcp_pm_mcgrps), 2426 }; 2427 2428 static int __net_init pm_nl_init_net(struct net *net) 2429 { 2430 struct pm_nl_pernet *pernet = pm_nl_get_pernet(net); 2431 2432 INIT_LIST_HEAD_RCU(&pernet->local_addr_list); 2433 2434 /* Cit. 2 subflows ought to be enough for anybody. */ 2435 pernet->subflows_max = 2; 2436 pernet->next_id = 1; 2437 pernet->stale_loss_cnt = 4; 2438 spin_lock_init(&pernet->lock); 2439 2440 /* No need to initialize other pernet fields, the struct is zeroed at 2441 * allocation time. 2442 */ 2443 2444 return 0; 2445 } 2446 2447 static void __net_exit pm_nl_exit_net(struct list_head *net_list) 2448 { 2449 struct net *net; 2450 2451 list_for_each_entry(net, net_list, exit_list) { 2452 struct pm_nl_pernet *pernet = pm_nl_get_pernet(net); 2453 2454 /* net is removed from namespace list, can't race with 2455 * other modifiers, also netns core already waited for a 2456 * RCU grace period. 2457 */ 2458 __flush_addrs(&pernet->local_addr_list); 2459 } 2460 } 2461 2462 static struct pernet_operations mptcp_pm_pernet_ops = { 2463 .init = pm_nl_init_net, 2464 .exit_batch = pm_nl_exit_net, 2465 .id = &pm_nl_pernet_id, 2466 .size = sizeof(struct pm_nl_pernet), 2467 }; 2468 2469 void __init mptcp_pm_nl_init(void) 2470 { 2471 if (register_pernet_subsys(&mptcp_pm_pernet_ops) < 0) 2472 panic("Failed to register MPTCP PM pernet subsystem.\n"); 2473 2474 if (genl_register_family(&mptcp_genl_family)) 2475 panic("Failed to register MPTCP PM netlink family\n"); 2476 } 2477