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; 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 mptcp_pm_send_ack(msk, subflow, false, false); 796 break; 797 } 798 } 799 } 800 801 int mptcp_pm_nl_mp_prio_send_ack(struct mptcp_sock *msk, 802 struct mptcp_addr_info *addr, 803 struct mptcp_addr_info *rem, 804 u8 bkup) 805 { 806 struct mptcp_subflow_context *subflow; 807 808 pr_debug("bkup=%d\n", bkup); 809 810 mptcp_for_each_subflow(msk, subflow) { 811 struct sock *ssk = mptcp_subflow_tcp_sock(subflow); 812 struct mptcp_addr_info local, remote; 813 814 mptcp_local_address((struct sock_common *)ssk, &local); 815 if (!mptcp_addresses_equal(&local, addr, addr->port)) 816 continue; 817 818 if (rem && rem->family != AF_UNSPEC) { 819 remote_address((struct sock_common *)ssk, &remote); 820 if (!mptcp_addresses_equal(&remote, rem, rem->port)) 821 continue; 822 } 823 824 __mptcp_pm_send_ack(msk, subflow, true, bkup); 825 return 0; 826 } 827 828 return -EINVAL; 829 } 830 831 static void mptcp_pm_nl_rm_addr_or_subflow(struct mptcp_sock *msk, 832 const struct mptcp_rm_list *rm_list, 833 enum linux_mptcp_mib_field rm_type) 834 { 835 struct mptcp_subflow_context *subflow, *tmp; 836 struct sock *sk = (struct sock *)msk; 837 u8 i; 838 839 pr_debug("%s rm_list_nr %d\n", 840 rm_type == MPTCP_MIB_RMADDR ? "address" : "subflow", rm_list->nr); 841 842 msk_owned_by_me(msk); 843 844 if (sk->sk_state == TCP_LISTEN) 845 return; 846 847 if (!rm_list->nr) 848 return; 849 850 if (list_empty(&msk->conn_list)) 851 return; 852 853 for (i = 0; i < rm_list->nr; i++) { 854 u8 rm_id = rm_list->ids[i]; 855 bool removed = false; 856 857 mptcp_for_each_subflow_safe(msk, subflow, tmp) { 858 struct sock *ssk = mptcp_subflow_tcp_sock(subflow); 859 u8 remote_id = READ_ONCE(subflow->remote_id); 860 int how = RCV_SHUTDOWN | SEND_SHUTDOWN; 861 u8 id = subflow_get_local_id(subflow); 862 863 if (inet_sk_state_load(ssk) == TCP_CLOSE) 864 continue; 865 if (rm_type == MPTCP_MIB_RMADDR && remote_id != rm_id) 866 continue; 867 if (rm_type == MPTCP_MIB_RMSUBFLOW && id != rm_id) 868 continue; 869 870 pr_debug(" -> %s rm_list_ids[%d]=%u local_id=%u remote_id=%u mpc_id=%u\n", 871 rm_type == MPTCP_MIB_RMADDR ? "address" : "subflow", 872 i, rm_id, id, remote_id, msk->mpc_endpoint_id); 873 spin_unlock_bh(&msk->pm.lock); 874 mptcp_subflow_shutdown(sk, ssk, how); 875 876 /* the following takes care of updating the subflows counter */ 877 mptcp_close_ssk(sk, ssk, subflow); 878 spin_lock_bh(&msk->pm.lock); 879 880 removed |= subflow->request_join; 881 if (rm_type == MPTCP_MIB_RMSUBFLOW) 882 __MPTCP_INC_STATS(sock_net(sk), rm_type); 883 } 884 885 if (rm_type == MPTCP_MIB_RMADDR) 886 __MPTCP_INC_STATS(sock_net(sk), rm_type); 887 888 if (!removed) 889 continue; 890 891 if (!mptcp_pm_is_kernel(msk)) 892 continue; 893 894 if (rm_type == MPTCP_MIB_RMADDR && rm_id && 895 !WARN_ON_ONCE(msk->pm.add_addr_accepted == 0)) { 896 /* Note: if the subflow has been closed before, this 897 * add_addr_accepted counter will not be decremented. 898 */ 899 if (--msk->pm.add_addr_accepted < mptcp_pm_get_add_addr_accept_max(msk)) 900 WRITE_ONCE(msk->pm.accept_addr, true); 901 } 902 } 903 } 904 905 static void mptcp_pm_nl_rm_addr_received(struct mptcp_sock *msk) 906 { 907 mptcp_pm_nl_rm_addr_or_subflow(msk, &msk->pm.rm_list_rx, MPTCP_MIB_RMADDR); 908 } 909 910 static void mptcp_pm_nl_rm_subflow_received(struct mptcp_sock *msk, 911 const struct mptcp_rm_list *rm_list) 912 { 913 mptcp_pm_nl_rm_addr_or_subflow(msk, rm_list, MPTCP_MIB_RMSUBFLOW); 914 } 915 916 void mptcp_pm_nl_work(struct mptcp_sock *msk) 917 { 918 struct mptcp_pm_data *pm = &msk->pm; 919 920 msk_owned_by_me(msk); 921 922 if (!(pm->status & MPTCP_PM_WORK_MASK)) 923 return; 924 925 spin_lock_bh(&msk->pm.lock); 926 927 pr_debug("msk=%p status=%x\n", msk, pm->status); 928 if (pm->status & BIT(MPTCP_PM_ADD_ADDR_RECEIVED)) { 929 pm->status &= ~BIT(MPTCP_PM_ADD_ADDR_RECEIVED); 930 mptcp_pm_nl_add_addr_received(msk); 931 } 932 if (pm->status & BIT(MPTCP_PM_ADD_ADDR_SEND_ACK)) { 933 pm->status &= ~BIT(MPTCP_PM_ADD_ADDR_SEND_ACK); 934 mptcp_pm_nl_addr_send_ack(msk); 935 } 936 if (pm->status & BIT(MPTCP_PM_RM_ADDR_RECEIVED)) { 937 pm->status &= ~BIT(MPTCP_PM_RM_ADDR_RECEIVED); 938 mptcp_pm_nl_rm_addr_received(msk); 939 } 940 if (pm->status & BIT(MPTCP_PM_ESTABLISHED)) { 941 pm->status &= ~BIT(MPTCP_PM_ESTABLISHED); 942 mptcp_pm_nl_fully_established(msk); 943 } 944 if (pm->status & BIT(MPTCP_PM_SUBFLOW_ESTABLISHED)) { 945 pm->status &= ~BIT(MPTCP_PM_SUBFLOW_ESTABLISHED); 946 mptcp_pm_nl_subflow_established(msk); 947 } 948 949 spin_unlock_bh(&msk->pm.lock); 950 } 951 952 static bool address_use_port(struct mptcp_pm_addr_entry *entry) 953 { 954 return (entry->flags & 955 (MPTCP_PM_ADDR_FLAG_SIGNAL | MPTCP_PM_ADDR_FLAG_SUBFLOW)) == 956 MPTCP_PM_ADDR_FLAG_SIGNAL; 957 } 958 959 /* caller must ensure the RCU grace period is already elapsed */ 960 static void __mptcp_pm_release_addr_entry(struct mptcp_pm_addr_entry *entry) 961 { 962 if (entry->lsk) 963 sock_release(entry->lsk); 964 kfree(entry); 965 } 966 967 static int mptcp_pm_nl_append_new_local_addr(struct pm_nl_pernet *pernet, 968 struct mptcp_pm_addr_entry *entry, 969 bool needs_id) 970 { 971 struct mptcp_pm_addr_entry *cur, *del_entry = NULL; 972 unsigned int addr_max; 973 int ret = -EINVAL; 974 975 spin_lock_bh(&pernet->lock); 976 /* to keep the code simple, don't do IDR-like allocation for address ID, 977 * just bail when we exceed limits 978 */ 979 if (pernet->next_id == MPTCP_PM_MAX_ADDR_ID) 980 pernet->next_id = 1; 981 if (pernet->addrs >= MPTCP_PM_ADDR_MAX) { 982 ret = -ERANGE; 983 goto out; 984 } 985 if (test_bit(entry->addr.id, pernet->id_bitmap)) { 986 ret = -EBUSY; 987 goto out; 988 } 989 990 /* do not insert duplicate address, differentiate on port only 991 * singled addresses 992 */ 993 if (!address_use_port(entry)) 994 entry->addr.port = 0; 995 list_for_each_entry(cur, &pernet->local_addr_list, list) { 996 if (mptcp_addresses_equal(&cur->addr, &entry->addr, 997 cur->addr.port || entry->addr.port)) { 998 /* allow replacing the exiting endpoint only if such 999 * endpoint is an implicit one and the user-space 1000 * did not provide an endpoint id 1001 */ 1002 if (!(cur->flags & MPTCP_PM_ADDR_FLAG_IMPLICIT)) { 1003 ret = -EEXIST; 1004 goto out; 1005 } 1006 if (entry->addr.id) 1007 goto out; 1008 1009 pernet->addrs--; 1010 entry->addr.id = cur->addr.id; 1011 list_del_rcu(&cur->list); 1012 del_entry = cur; 1013 break; 1014 } 1015 } 1016 1017 if (!entry->addr.id && needs_id) { 1018 find_next: 1019 entry->addr.id = find_next_zero_bit(pernet->id_bitmap, 1020 MPTCP_PM_MAX_ADDR_ID + 1, 1021 pernet->next_id); 1022 if (!entry->addr.id && pernet->next_id != 1) { 1023 pernet->next_id = 1; 1024 goto find_next; 1025 } 1026 } 1027 1028 if (!entry->addr.id && needs_id) 1029 goto out; 1030 1031 __set_bit(entry->addr.id, pernet->id_bitmap); 1032 if (entry->addr.id > pernet->next_id) 1033 pernet->next_id = entry->addr.id; 1034 1035 if (entry->flags & MPTCP_PM_ADDR_FLAG_SIGNAL) { 1036 addr_max = pernet->add_addr_signal_max; 1037 WRITE_ONCE(pernet->add_addr_signal_max, addr_max + 1); 1038 } 1039 if (entry->flags & MPTCP_PM_ADDR_FLAG_SUBFLOW) { 1040 addr_max = pernet->local_addr_max; 1041 WRITE_ONCE(pernet->local_addr_max, addr_max + 1); 1042 } 1043 1044 pernet->addrs++; 1045 if (!entry->addr.port) 1046 list_add_tail_rcu(&entry->list, &pernet->local_addr_list); 1047 else 1048 list_add_rcu(&entry->list, &pernet->local_addr_list); 1049 ret = entry->addr.id; 1050 1051 out: 1052 spin_unlock_bh(&pernet->lock); 1053 1054 /* just replaced an existing entry, free it */ 1055 if (del_entry) { 1056 synchronize_rcu(); 1057 __mptcp_pm_release_addr_entry(del_entry); 1058 } 1059 return ret; 1060 } 1061 1062 static struct lock_class_key mptcp_slock_keys[2]; 1063 static struct lock_class_key mptcp_keys[2]; 1064 1065 static int mptcp_pm_nl_create_listen_socket(struct sock *sk, 1066 struct mptcp_pm_addr_entry *entry) 1067 { 1068 bool is_ipv6 = sk->sk_family == AF_INET6; 1069 int addrlen = sizeof(struct sockaddr_in); 1070 struct sockaddr_storage addr; 1071 struct sock *newsk, *ssk; 1072 int backlog = 1024; 1073 int err; 1074 1075 err = sock_create_kern(sock_net(sk), entry->addr.family, 1076 SOCK_STREAM, IPPROTO_MPTCP, &entry->lsk); 1077 if (err) 1078 return err; 1079 1080 newsk = entry->lsk->sk; 1081 if (!newsk) 1082 return -EINVAL; 1083 1084 /* The subflow socket lock is acquired in a nested to the msk one 1085 * in several places, even by the TCP stack, and this msk is a kernel 1086 * socket: lockdep complains. Instead of propagating the _nested 1087 * modifiers in several places, re-init the lock class for the msk 1088 * socket to an mptcp specific one. 1089 */ 1090 sock_lock_init_class_and_name(newsk, 1091 is_ipv6 ? "mlock-AF_INET6" : "mlock-AF_INET", 1092 &mptcp_slock_keys[is_ipv6], 1093 is_ipv6 ? "msk_lock-AF_INET6" : "msk_lock-AF_INET", 1094 &mptcp_keys[is_ipv6]); 1095 1096 lock_sock(newsk); 1097 ssk = __mptcp_nmpc_sk(mptcp_sk(newsk)); 1098 release_sock(newsk); 1099 if (IS_ERR(ssk)) 1100 return PTR_ERR(ssk); 1101 1102 mptcp_info2sockaddr(&entry->addr, &addr, entry->addr.family); 1103 #if IS_ENABLED(CONFIG_MPTCP_IPV6) 1104 if (entry->addr.family == AF_INET6) 1105 addrlen = sizeof(struct sockaddr_in6); 1106 #endif 1107 if (ssk->sk_family == AF_INET) 1108 err = inet_bind_sk(ssk, (struct sockaddr *)&addr, addrlen); 1109 #if IS_ENABLED(CONFIG_MPTCP_IPV6) 1110 else if (ssk->sk_family == AF_INET6) 1111 err = inet6_bind_sk(ssk, (struct sockaddr *)&addr, addrlen); 1112 #endif 1113 if (err) 1114 return err; 1115 1116 /* We don't use mptcp_set_state() here because it needs to be called 1117 * under the msk socket lock. For the moment, that will not bring 1118 * anything more than only calling inet_sk_state_store(), because the 1119 * old status is known (TCP_CLOSE). 1120 */ 1121 inet_sk_state_store(newsk, TCP_LISTEN); 1122 lock_sock(ssk); 1123 err = __inet_listen_sk(ssk, backlog); 1124 if (!err) 1125 mptcp_event_pm_listener(ssk, MPTCP_EVENT_LISTENER_CREATED); 1126 release_sock(ssk); 1127 return err; 1128 } 1129 1130 int mptcp_pm_nl_get_local_id(struct mptcp_sock *msk, struct mptcp_addr_info *skc) 1131 { 1132 struct mptcp_pm_addr_entry *entry; 1133 struct pm_nl_pernet *pernet; 1134 int ret = -1; 1135 1136 pernet = pm_nl_get_pernet_from_msk(msk); 1137 1138 rcu_read_lock(); 1139 list_for_each_entry_rcu(entry, &pernet->local_addr_list, list) { 1140 if (mptcp_addresses_equal(&entry->addr, skc, entry->addr.port)) { 1141 ret = entry->addr.id; 1142 break; 1143 } 1144 } 1145 rcu_read_unlock(); 1146 if (ret >= 0) 1147 return ret; 1148 1149 /* address not found, add to local list */ 1150 entry = kmalloc(sizeof(*entry), GFP_ATOMIC); 1151 if (!entry) 1152 return -ENOMEM; 1153 1154 entry->addr = *skc; 1155 entry->addr.id = 0; 1156 entry->addr.port = 0; 1157 entry->ifindex = 0; 1158 entry->flags = MPTCP_PM_ADDR_FLAG_IMPLICIT; 1159 entry->lsk = NULL; 1160 ret = mptcp_pm_nl_append_new_local_addr(pernet, entry, true); 1161 if (ret < 0) 1162 kfree(entry); 1163 1164 return ret; 1165 } 1166 1167 bool mptcp_pm_nl_is_backup(struct mptcp_sock *msk, struct mptcp_addr_info *skc) 1168 { 1169 struct pm_nl_pernet *pernet = pm_nl_get_pernet_from_msk(msk); 1170 struct mptcp_pm_addr_entry *entry; 1171 bool backup = false; 1172 1173 rcu_read_lock(); 1174 list_for_each_entry_rcu(entry, &pernet->local_addr_list, list) { 1175 if (mptcp_addresses_equal(&entry->addr, skc, entry->addr.port)) { 1176 backup = !!(entry->flags & MPTCP_PM_ADDR_FLAG_BACKUP); 1177 break; 1178 } 1179 } 1180 rcu_read_unlock(); 1181 1182 return backup; 1183 } 1184 1185 #define MPTCP_PM_CMD_GRP_OFFSET 0 1186 #define MPTCP_PM_EV_GRP_OFFSET 1 1187 1188 static const struct genl_multicast_group mptcp_pm_mcgrps[] = { 1189 [MPTCP_PM_CMD_GRP_OFFSET] = { .name = MPTCP_PM_CMD_GRP_NAME, }, 1190 [MPTCP_PM_EV_GRP_OFFSET] = { .name = MPTCP_PM_EV_GRP_NAME, 1191 .flags = GENL_MCAST_CAP_NET_ADMIN, 1192 }, 1193 }; 1194 1195 void mptcp_pm_nl_subflow_chk_stale(const struct mptcp_sock *msk, struct sock *ssk) 1196 { 1197 struct mptcp_subflow_context *iter, *subflow = mptcp_subflow_ctx(ssk); 1198 struct sock *sk = (struct sock *)msk; 1199 unsigned int active_max_loss_cnt; 1200 struct net *net = sock_net(sk); 1201 unsigned int stale_loss_cnt; 1202 bool slow; 1203 1204 stale_loss_cnt = mptcp_stale_loss_cnt(net); 1205 if (subflow->stale || !stale_loss_cnt || subflow->stale_count <= stale_loss_cnt) 1206 return; 1207 1208 /* look for another available subflow not in loss state */ 1209 active_max_loss_cnt = max_t(int, stale_loss_cnt - 1, 1); 1210 mptcp_for_each_subflow(msk, iter) { 1211 if (iter != subflow && mptcp_subflow_active(iter) && 1212 iter->stale_count < active_max_loss_cnt) { 1213 /* we have some alternatives, try to mark this subflow as idle ...*/ 1214 slow = lock_sock_fast(ssk); 1215 if (!tcp_rtx_and_write_queues_empty(ssk)) { 1216 subflow->stale = 1; 1217 __mptcp_retransmit_pending_data(sk); 1218 MPTCP_INC_STATS(net, MPTCP_MIB_SUBFLOWSTALE); 1219 } 1220 unlock_sock_fast(ssk, slow); 1221 1222 /* always try to push the pending data regardless of re-injections: 1223 * we can possibly use backup subflows now, and subflow selection 1224 * is cheap under the msk socket lock 1225 */ 1226 __mptcp_push_pending(sk, 0); 1227 return; 1228 } 1229 } 1230 } 1231 1232 static int mptcp_pm_family_to_addr(int family) 1233 { 1234 #if IS_ENABLED(CONFIG_MPTCP_IPV6) 1235 if (family == AF_INET6) 1236 return MPTCP_PM_ADDR_ATTR_ADDR6; 1237 #endif 1238 return MPTCP_PM_ADDR_ATTR_ADDR4; 1239 } 1240 1241 static int mptcp_pm_parse_pm_addr_attr(struct nlattr *tb[], 1242 const struct nlattr *attr, 1243 struct genl_info *info, 1244 struct mptcp_addr_info *addr, 1245 bool require_family) 1246 { 1247 int err, addr_addr; 1248 1249 if (!attr) { 1250 GENL_SET_ERR_MSG(info, "missing address info"); 1251 return -EINVAL; 1252 } 1253 1254 /* no validation needed - was already done via nested policy */ 1255 err = nla_parse_nested_deprecated(tb, MPTCP_PM_ADDR_ATTR_MAX, attr, 1256 mptcp_pm_address_nl_policy, info->extack); 1257 if (err) 1258 return err; 1259 1260 if (tb[MPTCP_PM_ADDR_ATTR_ID]) 1261 addr->id = nla_get_u8(tb[MPTCP_PM_ADDR_ATTR_ID]); 1262 1263 if (!tb[MPTCP_PM_ADDR_ATTR_FAMILY]) { 1264 if (!require_family) 1265 return 0; 1266 1267 NL_SET_ERR_MSG_ATTR(info->extack, attr, 1268 "missing family"); 1269 return -EINVAL; 1270 } 1271 1272 addr->family = nla_get_u16(tb[MPTCP_PM_ADDR_ATTR_FAMILY]); 1273 if (addr->family != AF_INET 1274 #if IS_ENABLED(CONFIG_MPTCP_IPV6) 1275 && addr->family != AF_INET6 1276 #endif 1277 ) { 1278 NL_SET_ERR_MSG_ATTR(info->extack, attr, 1279 "unknown address family"); 1280 return -EINVAL; 1281 } 1282 addr_addr = mptcp_pm_family_to_addr(addr->family); 1283 if (!tb[addr_addr]) { 1284 NL_SET_ERR_MSG_ATTR(info->extack, attr, 1285 "missing address data"); 1286 return -EINVAL; 1287 } 1288 1289 #if IS_ENABLED(CONFIG_MPTCP_IPV6) 1290 if (addr->family == AF_INET6) 1291 addr->addr6 = nla_get_in6_addr(tb[addr_addr]); 1292 else 1293 #endif 1294 addr->addr.s_addr = nla_get_in_addr(tb[addr_addr]); 1295 1296 if (tb[MPTCP_PM_ADDR_ATTR_PORT]) 1297 addr->port = htons(nla_get_u16(tb[MPTCP_PM_ADDR_ATTR_PORT])); 1298 1299 return 0; 1300 } 1301 1302 int mptcp_pm_parse_addr(struct nlattr *attr, struct genl_info *info, 1303 struct mptcp_addr_info *addr) 1304 { 1305 struct nlattr *tb[MPTCP_PM_ADDR_ATTR_MAX + 1]; 1306 1307 memset(addr, 0, sizeof(*addr)); 1308 1309 return mptcp_pm_parse_pm_addr_attr(tb, attr, info, addr, true); 1310 } 1311 1312 int mptcp_pm_parse_entry(struct nlattr *attr, struct genl_info *info, 1313 bool require_family, 1314 struct mptcp_pm_addr_entry *entry) 1315 { 1316 struct nlattr *tb[MPTCP_PM_ADDR_ATTR_MAX + 1]; 1317 int err; 1318 1319 memset(entry, 0, sizeof(*entry)); 1320 1321 err = mptcp_pm_parse_pm_addr_attr(tb, attr, info, &entry->addr, require_family); 1322 if (err) 1323 return err; 1324 1325 if (tb[MPTCP_PM_ADDR_ATTR_IF_IDX]) { 1326 u32 val = nla_get_s32(tb[MPTCP_PM_ADDR_ATTR_IF_IDX]); 1327 1328 entry->ifindex = val; 1329 } 1330 1331 if (tb[MPTCP_PM_ADDR_ATTR_FLAGS]) 1332 entry->flags = nla_get_u32(tb[MPTCP_PM_ADDR_ATTR_FLAGS]); 1333 1334 if (tb[MPTCP_PM_ADDR_ATTR_PORT]) 1335 entry->addr.port = htons(nla_get_u16(tb[MPTCP_PM_ADDR_ATTR_PORT])); 1336 1337 return 0; 1338 } 1339 1340 static struct pm_nl_pernet *genl_info_pm_nl(struct genl_info *info) 1341 { 1342 return pm_nl_get_pernet(genl_info_net(info)); 1343 } 1344 1345 static int mptcp_nl_add_subflow_or_signal_addr(struct net *net, 1346 struct mptcp_addr_info *addr) 1347 { 1348 struct mptcp_sock *msk; 1349 long s_slot = 0, s_num = 0; 1350 1351 while ((msk = mptcp_token_iter_next(net, &s_slot, &s_num)) != NULL) { 1352 struct sock *sk = (struct sock *)msk; 1353 struct mptcp_addr_info mpc_addr; 1354 1355 if (!READ_ONCE(msk->fully_established) || 1356 mptcp_pm_is_userspace(msk)) 1357 goto next; 1358 1359 /* if the endp linked to the init sf is re-added with a != ID */ 1360 mptcp_local_address((struct sock_common *)msk, &mpc_addr); 1361 1362 lock_sock(sk); 1363 spin_lock_bh(&msk->pm.lock); 1364 if (mptcp_addresses_equal(addr, &mpc_addr, addr->port)) 1365 msk->mpc_endpoint_id = addr->id; 1366 mptcp_pm_create_subflow_or_signal_addr(msk); 1367 spin_unlock_bh(&msk->pm.lock); 1368 release_sock(sk); 1369 1370 next: 1371 sock_put(sk); 1372 cond_resched(); 1373 } 1374 1375 return 0; 1376 } 1377 1378 static bool mptcp_pm_has_addr_attr_id(const struct nlattr *attr, 1379 struct genl_info *info) 1380 { 1381 struct nlattr *tb[MPTCP_PM_ADDR_ATTR_MAX + 1]; 1382 1383 if (!nla_parse_nested_deprecated(tb, MPTCP_PM_ADDR_ATTR_MAX, attr, 1384 mptcp_pm_address_nl_policy, info->extack) && 1385 tb[MPTCP_PM_ADDR_ATTR_ID]) 1386 return true; 1387 return false; 1388 } 1389 1390 int mptcp_pm_nl_add_addr_doit(struct sk_buff *skb, struct genl_info *info) 1391 { 1392 struct nlattr *attr = info->attrs[MPTCP_PM_ENDPOINT_ADDR]; 1393 struct pm_nl_pernet *pernet = genl_info_pm_nl(info); 1394 struct mptcp_pm_addr_entry addr, *entry; 1395 int ret; 1396 1397 ret = mptcp_pm_parse_entry(attr, info, true, &addr); 1398 if (ret < 0) 1399 return ret; 1400 1401 if (addr.addr.port && !address_use_port(&addr)) { 1402 GENL_SET_ERR_MSG(info, "flags must have signal and not subflow when using port"); 1403 return -EINVAL; 1404 } 1405 1406 if (addr.flags & MPTCP_PM_ADDR_FLAG_SIGNAL && 1407 addr.flags & MPTCP_PM_ADDR_FLAG_FULLMESH) { 1408 GENL_SET_ERR_MSG(info, "flags mustn't have both signal and fullmesh"); 1409 return -EINVAL; 1410 } 1411 1412 if (addr.flags & MPTCP_PM_ADDR_FLAG_IMPLICIT) { 1413 GENL_SET_ERR_MSG(info, "can't create IMPLICIT endpoint"); 1414 return -EINVAL; 1415 } 1416 1417 entry = kzalloc(sizeof(*entry), GFP_KERNEL_ACCOUNT); 1418 if (!entry) { 1419 GENL_SET_ERR_MSG(info, "can't allocate addr"); 1420 return -ENOMEM; 1421 } 1422 1423 *entry = addr; 1424 if (entry->addr.port) { 1425 ret = mptcp_pm_nl_create_listen_socket(skb->sk, entry); 1426 if (ret) { 1427 GENL_SET_ERR_MSG_FMT(info, "create listen socket error: %d", ret); 1428 goto out_free; 1429 } 1430 } 1431 ret = mptcp_pm_nl_append_new_local_addr(pernet, entry, 1432 !mptcp_pm_has_addr_attr_id(attr, info)); 1433 if (ret < 0) { 1434 GENL_SET_ERR_MSG_FMT(info, "too many addresses or duplicate one: %d", ret); 1435 goto out_free; 1436 } 1437 1438 mptcp_nl_add_subflow_or_signal_addr(sock_net(skb->sk), &entry->addr); 1439 return 0; 1440 1441 out_free: 1442 __mptcp_pm_release_addr_entry(entry); 1443 return ret; 1444 } 1445 1446 static bool remove_anno_list_by_saddr(struct mptcp_sock *msk, 1447 const struct mptcp_addr_info *addr) 1448 { 1449 struct mptcp_pm_add_entry *entry; 1450 1451 entry = mptcp_pm_del_add_timer(msk, addr, false); 1452 if (entry) { 1453 kfree(entry); 1454 return true; 1455 } 1456 1457 return false; 1458 } 1459 1460 static u8 mptcp_endp_get_local_id(struct mptcp_sock *msk, 1461 const struct mptcp_addr_info *addr) 1462 { 1463 return msk->mpc_endpoint_id == addr->id ? 0 : addr->id; 1464 } 1465 1466 static bool mptcp_pm_remove_anno_addr(struct mptcp_sock *msk, 1467 const struct mptcp_addr_info *addr, 1468 bool force) 1469 { 1470 struct mptcp_rm_list list = { .nr = 0 }; 1471 bool ret; 1472 1473 list.ids[list.nr++] = mptcp_endp_get_local_id(msk, addr); 1474 1475 ret = remove_anno_list_by_saddr(msk, addr); 1476 if (ret || force) { 1477 spin_lock_bh(&msk->pm.lock); 1478 if (ret) { 1479 __set_bit(addr->id, msk->pm.id_avail_bitmap); 1480 msk->pm.add_addr_signaled--; 1481 } 1482 mptcp_pm_remove_addr(msk, &list); 1483 spin_unlock_bh(&msk->pm.lock); 1484 } 1485 return ret; 1486 } 1487 1488 static void __mark_subflow_endp_available(struct mptcp_sock *msk, u8 id) 1489 { 1490 /* If it was marked as used, and not ID 0, decrement local_addr_used */ 1491 if (!__test_and_set_bit(id ? : msk->mpc_endpoint_id, msk->pm.id_avail_bitmap) && 1492 id && !WARN_ON_ONCE(msk->pm.local_addr_used == 0)) 1493 msk->pm.local_addr_used--; 1494 } 1495 1496 static int mptcp_nl_remove_subflow_and_signal_addr(struct net *net, 1497 const struct mptcp_pm_addr_entry *entry) 1498 { 1499 const struct mptcp_addr_info *addr = &entry->addr; 1500 struct mptcp_rm_list list = { .nr = 1 }; 1501 long s_slot = 0, s_num = 0; 1502 struct mptcp_sock *msk; 1503 1504 pr_debug("remove_id=%d\n", addr->id); 1505 1506 while ((msk = mptcp_token_iter_next(net, &s_slot, &s_num)) != NULL) { 1507 struct sock *sk = (struct sock *)msk; 1508 bool remove_subflow; 1509 1510 if (mptcp_pm_is_userspace(msk)) 1511 goto next; 1512 1513 if (list_empty(&msk->conn_list)) { 1514 mptcp_pm_remove_anno_addr(msk, addr, false); 1515 goto next; 1516 } 1517 1518 lock_sock(sk); 1519 remove_subflow = lookup_subflow_by_saddr(&msk->conn_list, addr); 1520 mptcp_pm_remove_anno_addr(msk, addr, remove_subflow && 1521 !(entry->flags & MPTCP_PM_ADDR_FLAG_IMPLICIT)); 1522 1523 list.ids[0] = mptcp_endp_get_local_id(msk, addr); 1524 if (remove_subflow) { 1525 spin_lock_bh(&msk->pm.lock); 1526 mptcp_pm_nl_rm_subflow_received(msk, &list); 1527 spin_unlock_bh(&msk->pm.lock); 1528 } 1529 1530 if (entry->flags & MPTCP_PM_ADDR_FLAG_SUBFLOW) { 1531 spin_lock_bh(&msk->pm.lock); 1532 __mark_subflow_endp_available(msk, list.ids[0]); 1533 spin_unlock_bh(&msk->pm.lock); 1534 } 1535 1536 if (msk->mpc_endpoint_id == entry->addr.id) 1537 msk->mpc_endpoint_id = 0; 1538 release_sock(sk); 1539 1540 next: 1541 sock_put(sk); 1542 cond_resched(); 1543 } 1544 1545 return 0; 1546 } 1547 1548 static int mptcp_nl_remove_id_zero_address(struct net *net, 1549 struct mptcp_addr_info *addr) 1550 { 1551 struct mptcp_rm_list list = { .nr = 0 }; 1552 long s_slot = 0, s_num = 0; 1553 struct mptcp_sock *msk; 1554 1555 list.ids[list.nr++] = 0; 1556 1557 while ((msk = mptcp_token_iter_next(net, &s_slot, &s_num)) != NULL) { 1558 struct sock *sk = (struct sock *)msk; 1559 struct mptcp_addr_info msk_local; 1560 1561 if (list_empty(&msk->conn_list) || mptcp_pm_is_userspace(msk)) 1562 goto next; 1563 1564 mptcp_local_address((struct sock_common *)msk, &msk_local); 1565 if (!mptcp_addresses_equal(&msk_local, addr, addr->port)) 1566 goto next; 1567 1568 lock_sock(sk); 1569 spin_lock_bh(&msk->pm.lock); 1570 mptcp_pm_remove_addr(msk, &list); 1571 mptcp_pm_nl_rm_subflow_received(msk, &list); 1572 __mark_subflow_endp_available(msk, 0); 1573 spin_unlock_bh(&msk->pm.lock); 1574 release_sock(sk); 1575 1576 next: 1577 sock_put(sk); 1578 cond_resched(); 1579 } 1580 1581 return 0; 1582 } 1583 1584 int mptcp_pm_nl_del_addr_doit(struct sk_buff *skb, struct genl_info *info) 1585 { 1586 struct nlattr *attr = info->attrs[MPTCP_PM_ENDPOINT_ADDR]; 1587 struct pm_nl_pernet *pernet = genl_info_pm_nl(info); 1588 struct mptcp_pm_addr_entry addr, *entry; 1589 unsigned int addr_max; 1590 int ret; 1591 1592 ret = mptcp_pm_parse_entry(attr, info, false, &addr); 1593 if (ret < 0) 1594 return ret; 1595 1596 /* the zero id address is special: the first address used by the msk 1597 * always gets such an id, so different subflows can have different zero 1598 * id addresses. Additionally zero id is not accounted for in id_bitmap. 1599 * Let's use an 'mptcp_rm_list' instead of the common remove code. 1600 */ 1601 if (addr.addr.id == 0) 1602 return mptcp_nl_remove_id_zero_address(sock_net(skb->sk), &addr.addr); 1603 1604 spin_lock_bh(&pernet->lock); 1605 entry = __lookup_addr_by_id(pernet, addr.addr.id); 1606 if (!entry) { 1607 GENL_SET_ERR_MSG(info, "address not found"); 1608 spin_unlock_bh(&pernet->lock); 1609 return -EINVAL; 1610 } 1611 if (entry->flags & MPTCP_PM_ADDR_FLAG_SIGNAL) { 1612 addr_max = pernet->add_addr_signal_max; 1613 WRITE_ONCE(pernet->add_addr_signal_max, addr_max - 1); 1614 } 1615 if (entry->flags & MPTCP_PM_ADDR_FLAG_SUBFLOW) { 1616 addr_max = pernet->local_addr_max; 1617 WRITE_ONCE(pernet->local_addr_max, addr_max - 1); 1618 } 1619 1620 pernet->addrs--; 1621 list_del_rcu(&entry->list); 1622 __clear_bit(entry->addr.id, pernet->id_bitmap); 1623 spin_unlock_bh(&pernet->lock); 1624 1625 mptcp_nl_remove_subflow_and_signal_addr(sock_net(skb->sk), entry); 1626 synchronize_rcu(); 1627 __mptcp_pm_release_addr_entry(entry); 1628 1629 return ret; 1630 } 1631 1632 /* Called from the userspace PM only */ 1633 void mptcp_pm_remove_addrs(struct mptcp_sock *msk, struct list_head *rm_list) 1634 { 1635 struct mptcp_rm_list alist = { .nr = 0 }; 1636 struct mptcp_pm_addr_entry *entry; 1637 int anno_nr = 0; 1638 1639 list_for_each_entry(entry, rm_list, list) { 1640 if (alist.nr >= MPTCP_RM_IDS_MAX) 1641 break; 1642 1643 /* only delete if either announced or matching a subflow */ 1644 if (remove_anno_list_by_saddr(msk, &entry->addr)) 1645 anno_nr++; 1646 else if (!lookup_subflow_by_saddr(&msk->conn_list, 1647 &entry->addr)) 1648 continue; 1649 1650 alist.ids[alist.nr++] = entry->addr.id; 1651 } 1652 1653 if (alist.nr) { 1654 spin_lock_bh(&msk->pm.lock); 1655 msk->pm.add_addr_signaled -= anno_nr; 1656 mptcp_pm_remove_addr(msk, &alist); 1657 spin_unlock_bh(&msk->pm.lock); 1658 } 1659 } 1660 1661 /* Called from the in-kernel PM only */ 1662 static void mptcp_pm_flush_addrs_and_subflows(struct mptcp_sock *msk, 1663 struct list_head *rm_list) 1664 { 1665 struct mptcp_rm_list alist = { .nr = 0 }, slist = { .nr = 0 }; 1666 struct mptcp_pm_addr_entry *entry; 1667 1668 list_for_each_entry(entry, rm_list, list) { 1669 if (slist.nr < MPTCP_RM_IDS_MAX && 1670 lookup_subflow_by_saddr(&msk->conn_list, &entry->addr)) 1671 slist.ids[slist.nr++] = mptcp_endp_get_local_id(msk, &entry->addr); 1672 1673 if (alist.nr < MPTCP_RM_IDS_MAX && 1674 remove_anno_list_by_saddr(msk, &entry->addr)) 1675 alist.ids[alist.nr++] = mptcp_endp_get_local_id(msk, &entry->addr); 1676 } 1677 1678 spin_lock_bh(&msk->pm.lock); 1679 if (alist.nr) { 1680 msk->pm.add_addr_signaled -= alist.nr; 1681 mptcp_pm_remove_addr(msk, &alist); 1682 } 1683 if (slist.nr) 1684 mptcp_pm_nl_rm_subflow_received(msk, &slist); 1685 /* Reset counters: maybe some subflows have been removed before */ 1686 bitmap_fill(msk->pm.id_avail_bitmap, MPTCP_PM_MAX_ADDR_ID + 1); 1687 msk->pm.local_addr_used = 0; 1688 spin_unlock_bh(&msk->pm.lock); 1689 } 1690 1691 static void mptcp_nl_flush_addrs_list(struct net *net, 1692 struct list_head *rm_list) 1693 { 1694 long s_slot = 0, s_num = 0; 1695 struct mptcp_sock *msk; 1696 1697 if (list_empty(rm_list)) 1698 return; 1699 1700 while ((msk = mptcp_token_iter_next(net, &s_slot, &s_num)) != NULL) { 1701 struct sock *sk = (struct sock *)msk; 1702 1703 if (!mptcp_pm_is_userspace(msk)) { 1704 lock_sock(sk); 1705 mptcp_pm_flush_addrs_and_subflows(msk, rm_list); 1706 release_sock(sk); 1707 } 1708 1709 sock_put(sk); 1710 cond_resched(); 1711 } 1712 } 1713 1714 /* caller must ensure the RCU grace period is already elapsed */ 1715 static void __flush_addrs(struct list_head *list) 1716 { 1717 while (!list_empty(list)) { 1718 struct mptcp_pm_addr_entry *cur; 1719 1720 cur = list_entry(list->next, 1721 struct mptcp_pm_addr_entry, list); 1722 list_del_rcu(&cur->list); 1723 __mptcp_pm_release_addr_entry(cur); 1724 } 1725 } 1726 1727 static void __reset_counters(struct pm_nl_pernet *pernet) 1728 { 1729 WRITE_ONCE(pernet->add_addr_signal_max, 0); 1730 WRITE_ONCE(pernet->add_addr_accept_max, 0); 1731 WRITE_ONCE(pernet->local_addr_max, 0); 1732 pernet->addrs = 0; 1733 } 1734 1735 int mptcp_pm_nl_flush_addrs_doit(struct sk_buff *skb, struct genl_info *info) 1736 { 1737 struct pm_nl_pernet *pernet = genl_info_pm_nl(info); 1738 LIST_HEAD(free_list); 1739 1740 spin_lock_bh(&pernet->lock); 1741 list_splice_init(&pernet->local_addr_list, &free_list); 1742 __reset_counters(pernet); 1743 pernet->next_id = 1; 1744 bitmap_zero(pernet->id_bitmap, MPTCP_PM_MAX_ADDR_ID + 1); 1745 spin_unlock_bh(&pernet->lock); 1746 mptcp_nl_flush_addrs_list(sock_net(skb->sk), &free_list); 1747 synchronize_rcu(); 1748 __flush_addrs(&free_list); 1749 return 0; 1750 } 1751 1752 int mptcp_nl_fill_addr(struct sk_buff *skb, 1753 struct mptcp_pm_addr_entry *entry) 1754 { 1755 struct mptcp_addr_info *addr = &entry->addr; 1756 struct nlattr *attr; 1757 1758 attr = nla_nest_start(skb, MPTCP_PM_ATTR_ADDR); 1759 if (!attr) 1760 return -EMSGSIZE; 1761 1762 if (nla_put_u16(skb, MPTCP_PM_ADDR_ATTR_FAMILY, addr->family)) 1763 goto nla_put_failure; 1764 if (nla_put_u16(skb, MPTCP_PM_ADDR_ATTR_PORT, ntohs(addr->port))) 1765 goto nla_put_failure; 1766 if (nla_put_u8(skb, MPTCP_PM_ADDR_ATTR_ID, addr->id)) 1767 goto nla_put_failure; 1768 if (nla_put_u32(skb, MPTCP_PM_ADDR_ATTR_FLAGS, entry->flags)) 1769 goto nla_put_failure; 1770 if (entry->ifindex && 1771 nla_put_s32(skb, MPTCP_PM_ADDR_ATTR_IF_IDX, entry->ifindex)) 1772 goto nla_put_failure; 1773 1774 if (addr->family == AF_INET && 1775 nla_put_in_addr(skb, MPTCP_PM_ADDR_ATTR_ADDR4, 1776 addr->addr.s_addr)) 1777 goto nla_put_failure; 1778 #if IS_ENABLED(CONFIG_MPTCP_IPV6) 1779 else if (addr->family == AF_INET6 && 1780 nla_put_in6_addr(skb, MPTCP_PM_ADDR_ATTR_ADDR6, &addr->addr6)) 1781 goto nla_put_failure; 1782 #endif 1783 nla_nest_end(skb, attr); 1784 return 0; 1785 1786 nla_put_failure: 1787 nla_nest_cancel(skb, attr); 1788 return -EMSGSIZE; 1789 } 1790 1791 int mptcp_pm_nl_get_addr(struct sk_buff *skb, struct genl_info *info) 1792 { 1793 struct nlattr *attr = info->attrs[MPTCP_PM_ENDPOINT_ADDR]; 1794 struct pm_nl_pernet *pernet = genl_info_pm_nl(info); 1795 struct mptcp_pm_addr_entry addr, *entry; 1796 struct sk_buff *msg; 1797 void *reply; 1798 int ret; 1799 1800 ret = mptcp_pm_parse_entry(attr, info, false, &addr); 1801 if (ret < 0) 1802 return ret; 1803 1804 msg = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL); 1805 if (!msg) 1806 return -ENOMEM; 1807 1808 reply = genlmsg_put_reply(msg, info, &mptcp_genl_family, 0, 1809 info->genlhdr->cmd); 1810 if (!reply) { 1811 GENL_SET_ERR_MSG(info, "not enough space in Netlink message"); 1812 ret = -EMSGSIZE; 1813 goto fail; 1814 } 1815 1816 spin_lock_bh(&pernet->lock); 1817 entry = __lookup_addr_by_id(pernet, addr.addr.id); 1818 if (!entry) { 1819 GENL_SET_ERR_MSG(info, "address not found"); 1820 ret = -EINVAL; 1821 goto unlock_fail; 1822 } 1823 1824 ret = mptcp_nl_fill_addr(msg, entry); 1825 if (ret) 1826 goto unlock_fail; 1827 1828 genlmsg_end(msg, reply); 1829 ret = genlmsg_reply(msg, info); 1830 spin_unlock_bh(&pernet->lock); 1831 return ret; 1832 1833 unlock_fail: 1834 spin_unlock_bh(&pernet->lock); 1835 1836 fail: 1837 nlmsg_free(msg); 1838 return ret; 1839 } 1840 1841 int mptcp_pm_nl_get_addr_doit(struct sk_buff *skb, struct genl_info *info) 1842 { 1843 return mptcp_pm_get_addr(skb, info); 1844 } 1845 1846 int mptcp_pm_nl_dump_addr(struct sk_buff *msg, 1847 struct netlink_callback *cb) 1848 { 1849 struct net *net = sock_net(msg->sk); 1850 struct mptcp_pm_addr_entry *entry; 1851 struct pm_nl_pernet *pernet; 1852 int id = cb->args[0]; 1853 void *hdr; 1854 int i; 1855 1856 pernet = pm_nl_get_pernet(net); 1857 1858 spin_lock_bh(&pernet->lock); 1859 for (i = id; i < MPTCP_PM_MAX_ADDR_ID + 1; i++) { 1860 if (test_bit(i, pernet->id_bitmap)) { 1861 entry = __lookup_addr_by_id(pernet, i); 1862 if (!entry) 1863 break; 1864 1865 if (entry->addr.id <= id) 1866 continue; 1867 1868 hdr = genlmsg_put(msg, NETLINK_CB(cb->skb).portid, 1869 cb->nlh->nlmsg_seq, &mptcp_genl_family, 1870 NLM_F_MULTI, MPTCP_PM_CMD_GET_ADDR); 1871 if (!hdr) 1872 break; 1873 1874 if (mptcp_nl_fill_addr(msg, entry) < 0) { 1875 genlmsg_cancel(msg, hdr); 1876 break; 1877 } 1878 1879 id = entry->addr.id; 1880 genlmsg_end(msg, hdr); 1881 } 1882 } 1883 spin_unlock_bh(&pernet->lock); 1884 1885 cb->args[0] = id; 1886 return msg->len; 1887 } 1888 1889 int mptcp_pm_nl_get_addr_dumpit(struct sk_buff *msg, 1890 struct netlink_callback *cb) 1891 { 1892 return mptcp_pm_dump_addr(msg, cb); 1893 } 1894 1895 static int parse_limit(struct genl_info *info, int id, unsigned int *limit) 1896 { 1897 struct nlattr *attr = info->attrs[id]; 1898 1899 if (!attr) 1900 return 0; 1901 1902 *limit = nla_get_u32(attr); 1903 if (*limit > MPTCP_PM_ADDR_MAX) { 1904 GENL_SET_ERR_MSG(info, "limit greater than maximum"); 1905 return -EINVAL; 1906 } 1907 return 0; 1908 } 1909 1910 int mptcp_pm_nl_set_limits_doit(struct sk_buff *skb, struct genl_info *info) 1911 { 1912 struct pm_nl_pernet *pernet = genl_info_pm_nl(info); 1913 unsigned int rcv_addrs, subflows; 1914 int ret; 1915 1916 spin_lock_bh(&pernet->lock); 1917 rcv_addrs = pernet->add_addr_accept_max; 1918 ret = parse_limit(info, MPTCP_PM_ATTR_RCV_ADD_ADDRS, &rcv_addrs); 1919 if (ret) 1920 goto unlock; 1921 1922 subflows = pernet->subflows_max; 1923 ret = parse_limit(info, MPTCP_PM_ATTR_SUBFLOWS, &subflows); 1924 if (ret) 1925 goto unlock; 1926 1927 WRITE_ONCE(pernet->add_addr_accept_max, rcv_addrs); 1928 WRITE_ONCE(pernet->subflows_max, subflows); 1929 1930 unlock: 1931 spin_unlock_bh(&pernet->lock); 1932 return ret; 1933 } 1934 1935 int mptcp_pm_nl_get_limits_doit(struct sk_buff *skb, struct genl_info *info) 1936 { 1937 struct pm_nl_pernet *pernet = genl_info_pm_nl(info); 1938 struct sk_buff *msg; 1939 void *reply; 1940 1941 msg = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL); 1942 if (!msg) 1943 return -ENOMEM; 1944 1945 reply = genlmsg_put_reply(msg, info, &mptcp_genl_family, 0, 1946 MPTCP_PM_CMD_GET_LIMITS); 1947 if (!reply) 1948 goto fail; 1949 1950 if (nla_put_u32(msg, MPTCP_PM_ATTR_RCV_ADD_ADDRS, 1951 READ_ONCE(pernet->add_addr_accept_max))) 1952 goto fail; 1953 1954 if (nla_put_u32(msg, MPTCP_PM_ATTR_SUBFLOWS, 1955 READ_ONCE(pernet->subflows_max))) 1956 goto fail; 1957 1958 genlmsg_end(msg, reply); 1959 return genlmsg_reply(msg, info); 1960 1961 fail: 1962 GENL_SET_ERR_MSG(info, "not enough space in Netlink message"); 1963 nlmsg_free(msg); 1964 return -EMSGSIZE; 1965 } 1966 1967 static void mptcp_pm_nl_fullmesh(struct mptcp_sock *msk, 1968 struct mptcp_addr_info *addr) 1969 { 1970 struct mptcp_rm_list list = { .nr = 0 }; 1971 1972 list.ids[list.nr++] = mptcp_endp_get_local_id(msk, addr); 1973 1974 spin_lock_bh(&msk->pm.lock); 1975 mptcp_pm_nl_rm_subflow_received(msk, &list); 1976 __mark_subflow_endp_available(msk, list.ids[0]); 1977 mptcp_pm_create_subflow_or_signal_addr(msk); 1978 spin_unlock_bh(&msk->pm.lock); 1979 } 1980 1981 static int mptcp_nl_set_flags(struct net *net, 1982 struct mptcp_addr_info *addr, 1983 u8 bkup, u8 changed) 1984 { 1985 long s_slot = 0, s_num = 0; 1986 struct mptcp_sock *msk; 1987 int ret = -EINVAL; 1988 1989 while ((msk = mptcp_token_iter_next(net, &s_slot, &s_num)) != NULL) { 1990 struct sock *sk = (struct sock *)msk; 1991 1992 if (list_empty(&msk->conn_list) || mptcp_pm_is_userspace(msk)) 1993 goto next; 1994 1995 lock_sock(sk); 1996 if (changed & MPTCP_PM_ADDR_FLAG_BACKUP) 1997 ret = mptcp_pm_nl_mp_prio_send_ack(msk, addr, NULL, bkup); 1998 if (changed & MPTCP_PM_ADDR_FLAG_FULLMESH) 1999 mptcp_pm_nl_fullmesh(msk, addr); 2000 release_sock(sk); 2001 2002 next: 2003 sock_put(sk); 2004 cond_resched(); 2005 } 2006 2007 return ret; 2008 } 2009 2010 int mptcp_pm_nl_set_flags(struct sk_buff *skb, struct genl_info *info) 2011 { 2012 struct mptcp_pm_addr_entry addr = { .addr = { .family = AF_UNSPEC }, }; 2013 struct nlattr *attr = info->attrs[MPTCP_PM_ATTR_ADDR]; 2014 u8 changed, mask = MPTCP_PM_ADDR_FLAG_BACKUP | 2015 MPTCP_PM_ADDR_FLAG_FULLMESH; 2016 struct net *net = sock_net(skb->sk); 2017 struct mptcp_pm_addr_entry *entry; 2018 struct pm_nl_pernet *pernet; 2019 u8 lookup_by_id = 0; 2020 u8 bkup = 0; 2021 int ret; 2022 2023 pernet = pm_nl_get_pernet(net); 2024 2025 ret = mptcp_pm_parse_entry(attr, info, false, &addr); 2026 if (ret < 0) 2027 return ret; 2028 2029 if (addr.addr.family == AF_UNSPEC) { 2030 lookup_by_id = 1; 2031 if (!addr.addr.id) { 2032 GENL_SET_ERR_MSG(info, "missing required inputs"); 2033 return -EOPNOTSUPP; 2034 } 2035 } 2036 2037 if (addr.flags & MPTCP_PM_ADDR_FLAG_BACKUP) 2038 bkup = 1; 2039 2040 spin_lock_bh(&pernet->lock); 2041 entry = lookup_by_id ? __lookup_addr_by_id(pernet, addr.addr.id) : 2042 __lookup_addr(pernet, &addr.addr); 2043 if (!entry) { 2044 spin_unlock_bh(&pernet->lock); 2045 GENL_SET_ERR_MSG(info, "address not found"); 2046 return -EINVAL; 2047 } 2048 if ((addr.flags & MPTCP_PM_ADDR_FLAG_FULLMESH) && 2049 (entry->flags & MPTCP_PM_ADDR_FLAG_SIGNAL)) { 2050 spin_unlock_bh(&pernet->lock); 2051 GENL_SET_ERR_MSG(info, "invalid addr flags"); 2052 return -EINVAL; 2053 } 2054 2055 changed = (addr.flags ^ entry->flags) & mask; 2056 entry->flags = (entry->flags & ~mask) | (addr.flags & mask); 2057 addr = *entry; 2058 spin_unlock_bh(&pernet->lock); 2059 2060 mptcp_nl_set_flags(net, &addr.addr, bkup, changed); 2061 return 0; 2062 } 2063 2064 int mptcp_pm_nl_set_flags_doit(struct sk_buff *skb, struct genl_info *info) 2065 { 2066 return mptcp_pm_set_flags(skb, info); 2067 } 2068 2069 static void mptcp_nl_mcast_send(struct net *net, struct sk_buff *nlskb, gfp_t gfp) 2070 { 2071 genlmsg_multicast_netns(&mptcp_genl_family, net, 2072 nlskb, 0, MPTCP_PM_EV_GRP_OFFSET, gfp); 2073 } 2074 2075 bool mptcp_userspace_pm_active(const struct mptcp_sock *msk) 2076 { 2077 return genl_has_listeners(&mptcp_genl_family, 2078 sock_net((const struct sock *)msk), 2079 MPTCP_PM_EV_GRP_OFFSET); 2080 } 2081 2082 static int mptcp_event_add_subflow(struct sk_buff *skb, const struct sock *ssk) 2083 { 2084 const struct inet_sock *issk = inet_sk(ssk); 2085 const struct mptcp_subflow_context *sf; 2086 2087 if (nla_put_u16(skb, MPTCP_ATTR_FAMILY, ssk->sk_family)) 2088 return -EMSGSIZE; 2089 2090 switch (ssk->sk_family) { 2091 case AF_INET: 2092 if (nla_put_in_addr(skb, MPTCP_ATTR_SADDR4, issk->inet_saddr)) 2093 return -EMSGSIZE; 2094 if (nla_put_in_addr(skb, MPTCP_ATTR_DADDR4, issk->inet_daddr)) 2095 return -EMSGSIZE; 2096 break; 2097 #if IS_ENABLED(CONFIG_MPTCP_IPV6) 2098 case AF_INET6: { 2099 const struct ipv6_pinfo *np = inet6_sk(ssk); 2100 2101 if (nla_put_in6_addr(skb, MPTCP_ATTR_SADDR6, &np->saddr)) 2102 return -EMSGSIZE; 2103 if (nla_put_in6_addr(skb, MPTCP_ATTR_DADDR6, &ssk->sk_v6_daddr)) 2104 return -EMSGSIZE; 2105 break; 2106 } 2107 #endif 2108 default: 2109 WARN_ON_ONCE(1); 2110 return -EMSGSIZE; 2111 } 2112 2113 if (nla_put_be16(skb, MPTCP_ATTR_SPORT, issk->inet_sport)) 2114 return -EMSGSIZE; 2115 if (nla_put_be16(skb, MPTCP_ATTR_DPORT, issk->inet_dport)) 2116 return -EMSGSIZE; 2117 2118 sf = mptcp_subflow_ctx(ssk); 2119 if (WARN_ON_ONCE(!sf)) 2120 return -EINVAL; 2121 2122 if (nla_put_u8(skb, MPTCP_ATTR_LOC_ID, subflow_get_local_id(sf))) 2123 return -EMSGSIZE; 2124 2125 if (nla_put_u8(skb, MPTCP_ATTR_REM_ID, sf->remote_id)) 2126 return -EMSGSIZE; 2127 2128 return 0; 2129 } 2130 2131 static int mptcp_event_put_token_and_ssk(struct sk_buff *skb, 2132 const struct mptcp_sock *msk, 2133 const struct sock *ssk) 2134 { 2135 const struct sock *sk = (const struct sock *)msk; 2136 const struct mptcp_subflow_context *sf; 2137 u8 sk_err; 2138 2139 if (nla_put_u32(skb, MPTCP_ATTR_TOKEN, READ_ONCE(msk->token))) 2140 return -EMSGSIZE; 2141 2142 if (mptcp_event_add_subflow(skb, ssk)) 2143 return -EMSGSIZE; 2144 2145 sf = mptcp_subflow_ctx(ssk); 2146 if (WARN_ON_ONCE(!sf)) 2147 return -EINVAL; 2148 2149 if (nla_put_u8(skb, MPTCP_ATTR_BACKUP, sf->backup)) 2150 return -EMSGSIZE; 2151 2152 if (ssk->sk_bound_dev_if && 2153 nla_put_s32(skb, MPTCP_ATTR_IF_IDX, ssk->sk_bound_dev_if)) 2154 return -EMSGSIZE; 2155 2156 sk_err = READ_ONCE(ssk->sk_err); 2157 if (sk_err && sk->sk_state == TCP_ESTABLISHED && 2158 nla_put_u8(skb, MPTCP_ATTR_ERROR, sk_err)) 2159 return -EMSGSIZE; 2160 2161 return 0; 2162 } 2163 2164 static int mptcp_event_sub_established(struct sk_buff *skb, 2165 const struct mptcp_sock *msk, 2166 const struct sock *ssk) 2167 { 2168 return mptcp_event_put_token_and_ssk(skb, msk, ssk); 2169 } 2170 2171 static int mptcp_event_sub_closed(struct sk_buff *skb, 2172 const struct mptcp_sock *msk, 2173 const struct sock *ssk) 2174 { 2175 const struct mptcp_subflow_context *sf; 2176 2177 if (mptcp_event_put_token_and_ssk(skb, msk, ssk)) 2178 return -EMSGSIZE; 2179 2180 sf = mptcp_subflow_ctx(ssk); 2181 if (!sf->reset_seen) 2182 return 0; 2183 2184 if (nla_put_u32(skb, MPTCP_ATTR_RESET_REASON, sf->reset_reason)) 2185 return -EMSGSIZE; 2186 2187 if (nla_put_u32(skb, MPTCP_ATTR_RESET_FLAGS, sf->reset_transient)) 2188 return -EMSGSIZE; 2189 2190 return 0; 2191 } 2192 2193 static int mptcp_event_created(struct sk_buff *skb, 2194 const struct mptcp_sock *msk, 2195 const struct sock *ssk) 2196 { 2197 int err = nla_put_u32(skb, MPTCP_ATTR_TOKEN, READ_ONCE(msk->token)); 2198 2199 if (err) 2200 return err; 2201 2202 if (nla_put_u8(skb, MPTCP_ATTR_SERVER_SIDE, READ_ONCE(msk->pm.server_side))) 2203 return -EMSGSIZE; 2204 2205 return mptcp_event_add_subflow(skb, ssk); 2206 } 2207 2208 void mptcp_event_addr_removed(const struct mptcp_sock *msk, uint8_t id) 2209 { 2210 struct net *net = sock_net((const struct sock *)msk); 2211 struct nlmsghdr *nlh; 2212 struct sk_buff *skb; 2213 2214 if (!genl_has_listeners(&mptcp_genl_family, net, MPTCP_PM_EV_GRP_OFFSET)) 2215 return; 2216 2217 skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_ATOMIC); 2218 if (!skb) 2219 return; 2220 2221 nlh = genlmsg_put(skb, 0, 0, &mptcp_genl_family, 0, MPTCP_EVENT_REMOVED); 2222 if (!nlh) 2223 goto nla_put_failure; 2224 2225 if (nla_put_u32(skb, MPTCP_ATTR_TOKEN, READ_ONCE(msk->token))) 2226 goto nla_put_failure; 2227 2228 if (nla_put_u8(skb, MPTCP_ATTR_REM_ID, id)) 2229 goto nla_put_failure; 2230 2231 genlmsg_end(skb, nlh); 2232 mptcp_nl_mcast_send(net, skb, GFP_ATOMIC); 2233 return; 2234 2235 nla_put_failure: 2236 nlmsg_free(skb); 2237 } 2238 2239 void mptcp_event_addr_announced(const struct sock *ssk, 2240 const struct mptcp_addr_info *info) 2241 { 2242 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk); 2243 struct mptcp_sock *msk = mptcp_sk(subflow->conn); 2244 struct net *net = sock_net(ssk); 2245 struct nlmsghdr *nlh; 2246 struct sk_buff *skb; 2247 2248 if (!genl_has_listeners(&mptcp_genl_family, net, MPTCP_PM_EV_GRP_OFFSET)) 2249 return; 2250 2251 skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_ATOMIC); 2252 if (!skb) 2253 return; 2254 2255 nlh = genlmsg_put(skb, 0, 0, &mptcp_genl_family, 0, 2256 MPTCP_EVENT_ANNOUNCED); 2257 if (!nlh) 2258 goto nla_put_failure; 2259 2260 if (nla_put_u32(skb, MPTCP_ATTR_TOKEN, READ_ONCE(msk->token))) 2261 goto nla_put_failure; 2262 2263 if (nla_put_u8(skb, MPTCP_ATTR_REM_ID, info->id)) 2264 goto nla_put_failure; 2265 2266 if (nla_put_be16(skb, MPTCP_ATTR_DPORT, 2267 info->port == 0 ? 2268 inet_sk(ssk)->inet_dport : 2269 info->port)) 2270 goto nla_put_failure; 2271 2272 switch (info->family) { 2273 case AF_INET: 2274 if (nla_put_in_addr(skb, MPTCP_ATTR_DADDR4, info->addr.s_addr)) 2275 goto nla_put_failure; 2276 break; 2277 #if IS_ENABLED(CONFIG_MPTCP_IPV6) 2278 case AF_INET6: 2279 if (nla_put_in6_addr(skb, MPTCP_ATTR_DADDR6, &info->addr6)) 2280 goto nla_put_failure; 2281 break; 2282 #endif 2283 default: 2284 WARN_ON_ONCE(1); 2285 goto nla_put_failure; 2286 } 2287 2288 genlmsg_end(skb, nlh); 2289 mptcp_nl_mcast_send(net, skb, GFP_ATOMIC); 2290 return; 2291 2292 nla_put_failure: 2293 nlmsg_free(skb); 2294 } 2295 2296 void mptcp_event_pm_listener(const struct sock *ssk, 2297 enum mptcp_event_type event) 2298 { 2299 const struct inet_sock *issk = inet_sk(ssk); 2300 struct net *net = sock_net(ssk); 2301 struct nlmsghdr *nlh; 2302 struct sk_buff *skb; 2303 2304 if (!genl_has_listeners(&mptcp_genl_family, net, MPTCP_PM_EV_GRP_OFFSET)) 2305 return; 2306 2307 skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL); 2308 if (!skb) 2309 return; 2310 2311 nlh = genlmsg_put(skb, 0, 0, &mptcp_genl_family, 0, event); 2312 if (!nlh) 2313 goto nla_put_failure; 2314 2315 if (nla_put_u16(skb, MPTCP_ATTR_FAMILY, ssk->sk_family)) 2316 goto nla_put_failure; 2317 2318 if (nla_put_be16(skb, MPTCP_ATTR_SPORT, issk->inet_sport)) 2319 goto nla_put_failure; 2320 2321 switch (ssk->sk_family) { 2322 case AF_INET: 2323 if (nla_put_in_addr(skb, MPTCP_ATTR_SADDR4, issk->inet_saddr)) 2324 goto nla_put_failure; 2325 break; 2326 #if IS_ENABLED(CONFIG_MPTCP_IPV6) 2327 case AF_INET6: { 2328 const struct ipv6_pinfo *np = inet6_sk(ssk); 2329 2330 if (nla_put_in6_addr(skb, MPTCP_ATTR_SADDR6, &np->saddr)) 2331 goto nla_put_failure; 2332 break; 2333 } 2334 #endif 2335 default: 2336 WARN_ON_ONCE(1); 2337 goto nla_put_failure; 2338 } 2339 2340 genlmsg_end(skb, nlh); 2341 mptcp_nl_mcast_send(net, skb, GFP_KERNEL); 2342 return; 2343 2344 nla_put_failure: 2345 nlmsg_free(skb); 2346 } 2347 2348 void mptcp_event(enum mptcp_event_type type, const struct mptcp_sock *msk, 2349 const struct sock *ssk, gfp_t gfp) 2350 { 2351 struct net *net = sock_net((const struct sock *)msk); 2352 struct nlmsghdr *nlh; 2353 struct sk_buff *skb; 2354 2355 if (!genl_has_listeners(&mptcp_genl_family, net, MPTCP_PM_EV_GRP_OFFSET)) 2356 return; 2357 2358 skb = nlmsg_new(NLMSG_DEFAULT_SIZE, gfp); 2359 if (!skb) 2360 return; 2361 2362 nlh = genlmsg_put(skb, 0, 0, &mptcp_genl_family, 0, type); 2363 if (!nlh) 2364 goto nla_put_failure; 2365 2366 switch (type) { 2367 case MPTCP_EVENT_UNSPEC: 2368 WARN_ON_ONCE(1); 2369 break; 2370 case MPTCP_EVENT_CREATED: 2371 case MPTCP_EVENT_ESTABLISHED: 2372 if (mptcp_event_created(skb, msk, ssk) < 0) 2373 goto nla_put_failure; 2374 break; 2375 case MPTCP_EVENT_CLOSED: 2376 if (nla_put_u32(skb, MPTCP_ATTR_TOKEN, READ_ONCE(msk->token)) < 0) 2377 goto nla_put_failure; 2378 break; 2379 case MPTCP_EVENT_ANNOUNCED: 2380 case MPTCP_EVENT_REMOVED: 2381 /* call mptcp_event_addr_announced()/removed instead */ 2382 WARN_ON_ONCE(1); 2383 break; 2384 case MPTCP_EVENT_SUB_ESTABLISHED: 2385 case MPTCP_EVENT_SUB_PRIORITY: 2386 if (mptcp_event_sub_established(skb, msk, ssk) < 0) 2387 goto nla_put_failure; 2388 break; 2389 case MPTCP_EVENT_SUB_CLOSED: 2390 if (mptcp_event_sub_closed(skb, msk, ssk) < 0) 2391 goto nla_put_failure; 2392 break; 2393 case MPTCP_EVENT_LISTENER_CREATED: 2394 case MPTCP_EVENT_LISTENER_CLOSED: 2395 break; 2396 } 2397 2398 genlmsg_end(skb, nlh); 2399 mptcp_nl_mcast_send(net, skb, gfp); 2400 return; 2401 2402 nla_put_failure: 2403 nlmsg_free(skb); 2404 } 2405 2406 struct genl_family mptcp_genl_family __ro_after_init = { 2407 .name = MPTCP_PM_NAME, 2408 .version = MPTCP_PM_VER, 2409 .netnsok = true, 2410 .module = THIS_MODULE, 2411 .ops = mptcp_pm_nl_ops, 2412 .n_ops = ARRAY_SIZE(mptcp_pm_nl_ops), 2413 .resv_start_op = MPTCP_PM_CMD_SUBFLOW_DESTROY + 1, 2414 .mcgrps = mptcp_pm_mcgrps, 2415 .n_mcgrps = ARRAY_SIZE(mptcp_pm_mcgrps), 2416 }; 2417 2418 static int __net_init pm_nl_init_net(struct net *net) 2419 { 2420 struct pm_nl_pernet *pernet = pm_nl_get_pernet(net); 2421 2422 INIT_LIST_HEAD_RCU(&pernet->local_addr_list); 2423 2424 /* Cit. 2 subflows ought to be enough for anybody. */ 2425 pernet->subflows_max = 2; 2426 pernet->next_id = 1; 2427 pernet->stale_loss_cnt = 4; 2428 spin_lock_init(&pernet->lock); 2429 2430 /* No need to initialize other pernet fields, the struct is zeroed at 2431 * allocation time. 2432 */ 2433 2434 return 0; 2435 } 2436 2437 static void __net_exit pm_nl_exit_net(struct list_head *net_list) 2438 { 2439 struct net *net; 2440 2441 list_for_each_entry(net, net_list, exit_list) { 2442 struct pm_nl_pernet *pernet = pm_nl_get_pernet(net); 2443 2444 /* net is removed from namespace list, can't race with 2445 * other modifiers, also netns core already waited for a 2446 * RCU grace period. 2447 */ 2448 __flush_addrs(&pernet->local_addr_list); 2449 } 2450 } 2451 2452 static struct pernet_operations mptcp_pm_pernet_ops = { 2453 .init = pm_nl_init_net, 2454 .exit_batch = pm_nl_exit_net, 2455 .id = &pm_nl_pernet_id, 2456 .size = sizeof(struct pm_nl_pernet), 2457 }; 2458 2459 void __init mptcp_pm_nl_init(void) 2460 { 2461 if (register_pernet_subsys(&mptcp_pm_pernet_ops) < 0) 2462 panic("Failed to register MPTCP PM pernet subsystem.\n"); 2463 2464 if (genl_register_family(&mptcp_genl_family)) 2465 panic("Failed to register MPTCP PM netlink family\n"); 2466 } 2467