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