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_equal(&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 bool mptcp_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_rcu(entry, &pernet->local_addr_list, list, 516 lockdep_is_held(&pernet->lock)) { 517 if (entry->addr.id == id) 518 return entry; 519 } 520 return NULL; 521 } 522 523 static struct mptcp_pm_addr_entry * 524 __lookup_addr(struct pm_nl_pernet *pernet, const struct mptcp_addr_info *info) 525 { 526 struct mptcp_pm_addr_entry *entry; 527 528 list_for_each_entry_rcu(entry, &pernet->local_addr_list, list, 529 lockdep_is_held(&pernet->lock)) { 530 if (mptcp_addresses_equal(&entry->addr, info, entry->addr.port)) 531 return entry; 532 } 533 return NULL; 534 } 535 536 static void mptcp_pm_create_subflow_or_signal_addr(struct mptcp_sock *msk) 537 { 538 struct sock *sk = (struct sock *)msk; 539 unsigned int add_addr_signal_max; 540 bool signal_and_subflow = false; 541 unsigned int local_addr_max; 542 struct pm_nl_pernet *pernet; 543 struct mptcp_pm_local local; 544 unsigned int subflows_max; 545 546 pernet = pm_nl_get_pernet(sock_net(sk)); 547 548 add_addr_signal_max = mptcp_pm_get_add_addr_signal_max(msk); 549 local_addr_max = mptcp_pm_get_local_addr_max(msk); 550 subflows_max = mptcp_pm_get_subflows_max(msk); 551 552 /* do lazy endpoint usage accounting for the MPC subflows */ 553 if (unlikely(!(msk->pm.status & BIT(MPTCP_PM_MPC_ENDPOINT_ACCOUNTED))) && msk->first) { 554 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(msk->first); 555 struct mptcp_pm_addr_entry *entry; 556 struct mptcp_addr_info mpc_addr; 557 bool backup = false; 558 559 mptcp_local_address((struct sock_common *)msk->first, &mpc_addr); 560 rcu_read_lock(); 561 entry = __lookup_addr(pernet, &mpc_addr); 562 if (entry) { 563 __clear_bit(entry->addr.id, msk->pm.id_avail_bitmap); 564 msk->mpc_endpoint_id = entry->addr.id; 565 backup = !!(entry->flags & MPTCP_PM_ADDR_FLAG_BACKUP); 566 } 567 rcu_read_unlock(); 568 569 if (backup) 570 mptcp_pm_send_ack(msk, subflow, true, backup); 571 572 msk->pm.status |= BIT(MPTCP_PM_MPC_ENDPOINT_ACCOUNTED); 573 } 574 575 pr_debug("local %d:%d signal %d:%d subflows %d:%d\n", 576 msk->pm.local_addr_used, local_addr_max, 577 msk->pm.add_addr_signaled, add_addr_signal_max, 578 msk->pm.subflows, subflows_max); 579 580 /* check first for announce */ 581 if (msk->pm.add_addr_signaled < add_addr_signal_max) { 582 /* due to racing events on both ends we can reach here while 583 * previous add address is still running: if we invoke now 584 * mptcp_pm_announce_addr(), that will fail and the 585 * corresponding id will be marked as used. 586 * Instead let the PM machinery reschedule us when the 587 * current address announce will be completed. 588 */ 589 if (msk->pm.addr_signal & BIT(MPTCP_ADD_ADDR_SIGNAL)) 590 return; 591 592 if (!select_signal_address(pernet, msk, &local)) 593 goto subflow; 594 595 /* If the alloc fails, we are on memory pressure, not worth 596 * continuing, and trying to create subflows. 597 */ 598 if (!mptcp_pm_alloc_anno_list(msk, &local.addr)) 599 return; 600 601 __clear_bit(local.addr.id, msk->pm.id_avail_bitmap); 602 msk->pm.add_addr_signaled++; 603 604 /* Special case for ID0: set the correct ID */ 605 if (local.addr.id == msk->mpc_endpoint_id) 606 local.addr.id = 0; 607 608 mptcp_pm_announce_addr(msk, &local.addr, false); 609 mptcp_pm_nl_addr_send_ack(msk); 610 611 if (local.flags & MPTCP_PM_ADDR_FLAG_SUBFLOW) 612 signal_and_subflow = true; 613 } 614 615 subflow: 616 /* check if should create a new subflow */ 617 while (msk->pm.local_addr_used < local_addr_max && 618 msk->pm.subflows < subflows_max) { 619 struct mptcp_addr_info addrs[MPTCP_PM_ADDR_MAX]; 620 bool fullmesh; 621 int i, nr; 622 623 if (signal_and_subflow) 624 signal_and_subflow = false; 625 else if (!select_local_address(pernet, msk, &local)) 626 break; 627 628 fullmesh = !!(local.flags & MPTCP_PM_ADDR_FLAG_FULLMESH); 629 630 __clear_bit(local.addr.id, msk->pm.id_avail_bitmap); 631 632 /* Special case for ID0: set the correct ID */ 633 if (local.addr.id == msk->mpc_endpoint_id) 634 local.addr.id = 0; 635 else /* local_addr_used is not decr for ID 0 */ 636 msk->pm.local_addr_used++; 637 638 nr = fill_remote_addresses_vec(msk, &local.addr, fullmesh, addrs); 639 if (nr == 0) 640 continue; 641 642 spin_unlock_bh(&msk->pm.lock); 643 for (i = 0; i < nr; i++) 644 __mptcp_subflow_connect(sk, &local, &addrs[i]); 645 spin_lock_bh(&msk->pm.lock); 646 } 647 mptcp_pm_nl_check_work_pending(msk); 648 } 649 650 static void mptcp_pm_nl_fully_established(struct mptcp_sock *msk) 651 { 652 mptcp_pm_create_subflow_or_signal_addr(msk); 653 } 654 655 static void mptcp_pm_nl_subflow_established(struct mptcp_sock *msk) 656 { 657 mptcp_pm_create_subflow_or_signal_addr(msk); 658 } 659 660 /* Fill all the local addresses into the array addrs[], 661 * and return the array size. 662 */ 663 static unsigned int fill_local_addresses_vec(struct mptcp_sock *msk, 664 struct mptcp_addr_info *remote, 665 struct mptcp_pm_local *locals) 666 { 667 struct sock *sk = (struct sock *)msk; 668 struct mptcp_pm_addr_entry *entry; 669 struct mptcp_addr_info mpc_addr; 670 struct pm_nl_pernet *pernet; 671 unsigned int subflows_max; 672 int i = 0; 673 674 pernet = pm_nl_get_pernet_from_msk(msk); 675 subflows_max = mptcp_pm_get_subflows_max(msk); 676 677 mptcp_local_address((struct sock_common *)msk, &mpc_addr); 678 679 rcu_read_lock(); 680 list_for_each_entry_rcu(entry, &pernet->local_addr_list, list) { 681 if (!(entry->flags & MPTCP_PM_ADDR_FLAG_FULLMESH)) 682 continue; 683 684 if (!mptcp_pm_addr_families_match(sk, &entry->addr, remote)) 685 continue; 686 687 if (msk->pm.subflows < subflows_max) { 688 locals[i].addr = entry->addr; 689 locals[i].flags = entry->flags; 690 locals[i].ifindex = entry->ifindex; 691 692 /* Special case for ID0: set the correct ID */ 693 if (mptcp_addresses_equal(&locals[i].addr, &mpc_addr, locals[i].addr.port)) 694 locals[i].addr.id = 0; 695 696 msk->pm.subflows++; 697 i++; 698 } 699 } 700 rcu_read_unlock(); 701 702 /* If the array is empty, fill in the single 703 * 'IPADDRANY' local address 704 */ 705 if (!i) { 706 memset(&locals[i], 0, sizeof(locals[i])); 707 locals[i].addr.family = 708 #if IS_ENABLED(CONFIG_MPTCP_IPV6) 709 remote->family == AF_INET6 && 710 ipv6_addr_v4mapped(&remote->addr6) ? AF_INET : 711 #endif 712 remote->family; 713 714 if (!mptcp_pm_addr_families_match(sk, &locals[i].addr, remote)) 715 return 0; 716 717 msk->pm.subflows++; 718 i++; 719 } 720 721 return i; 722 } 723 724 static void mptcp_pm_nl_add_addr_received(struct mptcp_sock *msk) 725 { 726 struct mptcp_pm_local locals[MPTCP_PM_ADDR_MAX]; 727 struct sock *sk = (struct sock *)msk; 728 unsigned int add_addr_accept_max; 729 struct mptcp_addr_info remote; 730 unsigned int subflows_max; 731 bool sf_created = false; 732 int i, nr; 733 734 add_addr_accept_max = mptcp_pm_get_add_addr_accept_max(msk); 735 subflows_max = mptcp_pm_get_subflows_max(msk); 736 737 pr_debug("accepted %d:%d remote family %d\n", 738 msk->pm.add_addr_accepted, add_addr_accept_max, 739 msk->pm.remote.family); 740 741 remote = msk->pm.remote; 742 mptcp_pm_announce_addr(msk, &remote, true); 743 mptcp_pm_nl_addr_send_ack(msk); 744 745 if (lookup_subflow_by_daddr(&msk->conn_list, &remote)) 746 return; 747 748 /* pick id 0 port, if none is provided the remote address */ 749 if (!remote.port) 750 remote.port = sk->sk_dport; 751 752 /* connect to the specified remote address, using whatever 753 * local address the routing configuration will pick. 754 */ 755 nr = fill_local_addresses_vec(msk, &remote, locals); 756 if (nr == 0) 757 return; 758 759 spin_unlock_bh(&msk->pm.lock); 760 for (i = 0; i < nr; i++) 761 if (__mptcp_subflow_connect(sk, &locals[i], &remote) == 0) 762 sf_created = true; 763 spin_lock_bh(&msk->pm.lock); 764 765 if (sf_created) { 766 /* add_addr_accepted is not decr for ID 0 */ 767 if (remote.id) 768 msk->pm.add_addr_accepted++; 769 if (msk->pm.add_addr_accepted >= add_addr_accept_max || 770 msk->pm.subflows >= subflows_max) 771 WRITE_ONCE(msk->pm.accept_addr, false); 772 } 773 } 774 775 bool mptcp_pm_nl_is_init_remote_addr(struct mptcp_sock *msk, 776 const struct mptcp_addr_info *remote) 777 { 778 struct mptcp_addr_info mpc_remote; 779 780 remote_address((struct sock_common *)msk, &mpc_remote); 781 return mptcp_addresses_equal(&mpc_remote, remote, remote->port); 782 } 783 784 void mptcp_pm_nl_addr_send_ack(struct mptcp_sock *msk) 785 { 786 struct mptcp_subflow_context *subflow, *alt = NULL; 787 788 msk_owned_by_me(msk); 789 lockdep_assert_held(&msk->pm.lock); 790 791 if (!mptcp_pm_should_add_signal(msk) && 792 !mptcp_pm_should_rm_signal(msk)) 793 return; 794 795 mptcp_for_each_subflow(msk, subflow) { 796 if (__mptcp_subflow_active(subflow)) { 797 if (!subflow->stale) { 798 mptcp_pm_send_ack(msk, subflow, false, false); 799 return; 800 } 801 802 if (!alt) 803 alt = subflow; 804 } 805 } 806 807 if (alt) 808 mptcp_pm_send_ack(msk, alt, false, false); 809 } 810 811 int mptcp_pm_nl_mp_prio_send_ack(struct mptcp_sock *msk, 812 struct mptcp_addr_info *addr, 813 struct mptcp_addr_info *rem, 814 u8 bkup) 815 { 816 struct mptcp_subflow_context *subflow; 817 818 pr_debug("bkup=%d\n", bkup); 819 820 mptcp_for_each_subflow(msk, subflow) { 821 struct sock *ssk = mptcp_subflow_tcp_sock(subflow); 822 struct mptcp_addr_info local, remote; 823 824 mptcp_local_address((struct sock_common *)ssk, &local); 825 if (!mptcp_addresses_equal(&local, addr, addr->port)) 826 continue; 827 828 if (rem && rem->family != AF_UNSPEC) { 829 remote_address((struct sock_common *)ssk, &remote); 830 if (!mptcp_addresses_equal(&remote, rem, rem->port)) 831 continue; 832 } 833 834 __mptcp_pm_send_ack(msk, subflow, true, bkup); 835 return 0; 836 } 837 838 return -EINVAL; 839 } 840 841 static void mptcp_pm_nl_rm_addr_or_subflow(struct mptcp_sock *msk, 842 const struct mptcp_rm_list *rm_list, 843 enum linux_mptcp_mib_field rm_type) 844 { 845 struct mptcp_subflow_context *subflow, *tmp; 846 struct sock *sk = (struct sock *)msk; 847 u8 i; 848 849 pr_debug("%s rm_list_nr %d\n", 850 rm_type == MPTCP_MIB_RMADDR ? "address" : "subflow", rm_list->nr); 851 852 msk_owned_by_me(msk); 853 854 if (sk->sk_state == TCP_LISTEN) 855 return; 856 857 if (!rm_list->nr) 858 return; 859 860 if (list_empty(&msk->conn_list)) 861 return; 862 863 for (i = 0; i < rm_list->nr; i++) { 864 u8 rm_id = rm_list->ids[i]; 865 bool removed = false; 866 867 mptcp_for_each_subflow_safe(msk, subflow, tmp) { 868 struct sock *ssk = mptcp_subflow_tcp_sock(subflow); 869 u8 remote_id = READ_ONCE(subflow->remote_id); 870 int how = RCV_SHUTDOWN | SEND_SHUTDOWN; 871 u8 id = subflow_get_local_id(subflow); 872 873 if ((1 << inet_sk_state_load(ssk)) & 874 (TCPF_FIN_WAIT1 | TCPF_FIN_WAIT2 | TCPF_CLOSING | TCPF_CLOSE)) 875 continue; 876 if (rm_type == MPTCP_MIB_RMADDR && remote_id != rm_id) 877 continue; 878 if (rm_type == MPTCP_MIB_RMSUBFLOW && id != rm_id) 879 continue; 880 881 pr_debug(" -> %s rm_list_ids[%d]=%u local_id=%u remote_id=%u mpc_id=%u\n", 882 rm_type == MPTCP_MIB_RMADDR ? "address" : "subflow", 883 i, rm_id, id, remote_id, msk->mpc_endpoint_id); 884 spin_unlock_bh(&msk->pm.lock); 885 mptcp_subflow_shutdown(sk, ssk, how); 886 removed |= subflow->request_join; 887 888 /* the following takes care of updating the subflows counter */ 889 mptcp_close_ssk(sk, ssk, subflow); 890 spin_lock_bh(&msk->pm.lock); 891 892 if (rm_type == MPTCP_MIB_RMSUBFLOW) 893 __MPTCP_INC_STATS(sock_net(sk), rm_type); 894 } 895 896 if (rm_type == MPTCP_MIB_RMADDR) 897 __MPTCP_INC_STATS(sock_net(sk), rm_type); 898 899 if (!removed) 900 continue; 901 902 if (!mptcp_pm_is_kernel(msk)) 903 continue; 904 905 if (rm_type == MPTCP_MIB_RMADDR && rm_id && 906 !WARN_ON_ONCE(msk->pm.add_addr_accepted == 0)) { 907 /* Note: if the subflow has been closed before, this 908 * add_addr_accepted counter will not be decremented. 909 */ 910 if (--msk->pm.add_addr_accepted < mptcp_pm_get_add_addr_accept_max(msk)) 911 WRITE_ONCE(msk->pm.accept_addr, true); 912 } 913 } 914 } 915 916 static void mptcp_pm_nl_rm_addr_received(struct mptcp_sock *msk) 917 { 918 mptcp_pm_nl_rm_addr_or_subflow(msk, &msk->pm.rm_list_rx, MPTCP_MIB_RMADDR); 919 } 920 921 static void mptcp_pm_nl_rm_subflow_received(struct mptcp_sock *msk, 922 const struct mptcp_rm_list *rm_list) 923 { 924 mptcp_pm_nl_rm_addr_or_subflow(msk, rm_list, MPTCP_MIB_RMSUBFLOW); 925 } 926 927 void mptcp_pm_nl_work(struct mptcp_sock *msk) 928 { 929 struct mptcp_pm_data *pm = &msk->pm; 930 931 msk_owned_by_me(msk); 932 933 if (!(pm->status & MPTCP_PM_WORK_MASK)) 934 return; 935 936 spin_lock_bh(&msk->pm.lock); 937 938 pr_debug("msk=%p status=%x\n", msk, pm->status); 939 if (pm->status & BIT(MPTCP_PM_ADD_ADDR_RECEIVED)) { 940 pm->status &= ~BIT(MPTCP_PM_ADD_ADDR_RECEIVED); 941 mptcp_pm_nl_add_addr_received(msk); 942 } 943 if (pm->status & BIT(MPTCP_PM_ADD_ADDR_SEND_ACK)) { 944 pm->status &= ~BIT(MPTCP_PM_ADD_ADDR_SEND_ACK); 945 mptcp_pm_nl_addr_send_ack(msk); 946 } 947 if (pm->status & BIT(MPTCP_PM_RM_ADDR_RECEIVED)) { 948 pm->status &= ~BIT(MPTCP_PM_RM_ADDR_RECEIVED); 949 mptcp_pm_nl_rm_addr_received(msk); 950 } 951 if (pm->status & BIT(MPTCP_PM_ESTABLISHED)) { 952 pm->status &= ~BIT(MPTCP_PM_ESTABLISHED); 953 mptcp_pm_nl_fully_established(msk); 954 } 955 if (pm->status & BIT(MPTCP_PM_SUBFLOW_ESTABLISHED)) { 956 pm->status &= ~BIT(MPTCP_PM_SUBFLOW_ESTABLISHED); 957 mptcp_pm_nl_subflow_established(msk); 958 } 959 960 spin_unlock_bh(&msk->pm.lock); 961 } 962 963 static bool address_use_port(struct mptcp_pm_addr_entry *entry) 964 { 965 return (entry->flags & 966 (MPTCP_PM_ADDR_FLAG_SIGNAL | MPTCP_PM_ADDR_FLAG_SUBFLOW)) == 967 MPTCP_PM_ADDR_FLAG_SIGNAL; 968 } 969 970 /* caller must ensure the RCU grace period is already elapsed */ 971 static void __mptcp_pm_release_addr_entry(struct mptcp_pm_addr_entry *entry) 972 { 973 if (entry->lsk) 974 sock_release(entry->lsk); 975 kfree(entry); 976 } 977 978 static int mptcp_pm_nl_append_new_local_addr(struct pm_nl_pernet *pernet, 979 struct mptcp_pm_addr_entry *entry, 980 bool needs_id, bool replace) 981 { 982 struct mptcp_pm_addr_entry *cur, *del_entry = NULL; 983 unsigned int addr_max; 984 int ret = -EINVAL; 985 986 spin_lock_bh(&pernet->lock); 987 /* to keep the code simple, don't do IDR-like allocation for address ID, 988 * just bail when we exceed limits 989 */ 990 if (pernet->next_id == MPTCP_PM_MAX_ADDR_ID) 991 pernet->next_id = 1; 992 if (pernet->addrs >= MPTCP_PM_ADDR_MAX) { 993 ret = -ERANGE; 994 goto out; 995 } 996 if (test_bit(entry->addr.id, pernet->id_bitmap)) { 997 ret = -EBUSY; 998 goto out; 999 } 1000 1001 /* do not insert duplicate address, differentiate on port only 1002 * singled addresses 1003 */ 1004 if (!address_use_port(entry)) 1005 entry->addr.port = 0; 1006 list_for_each_entry(cur, &pernet->local_addr_list, list) { 1007 if (mptcp_addresses_equal(&cur->addr, &entry->addr, 1008 cur->addr.port || entry->addr.port)) { 1009 /* allow replacing the exiting endpoint only if such 1010 * endpoint is an implicit one and the user-space 1011 * did not provide an endpoint id 1012 */ 1013 if (!(cur->flags & MPTCP_PM_ADDR_FLAG_IMPLICIT)) { 1014 ret = -EEXIST; 1015 goto out; 1016 } 1017 if (entry->addr.id) 1018 goto out; 1019 1020 /* allow callers that only need to look up the local 1021 * addr's id to skip replacement. This allows them to 1022 * avoid calling synchronize_rcu in the packet recv 1023 * path. 1024 */ 1025 if (!replace) { 1026 kfree(entry); 1027 ret = cur->addr.id; 1028 goto out; 1029 } 1030 1031 pernet->addrs--; 1032 entry->addr.id = cur->addr.id; 1033 list_del_rcu(&cur->list); 1034 del_entry = cur; 1035 break; 1036 } 1037 } 1038 1039 if (!entry->addr.id && needs_id) { 1040 find_next: 1041 entry->addr.id = find_next_zero_bit(pernet->id_bitmap, 1042 MPTCP_PM_MAX_ADDR_ID + 1, 1043 pernet->next_id); 1044 if (!entry->addr.id && pernet->next_id != 1) { 1045 pernet->next_id = 1; 1046 goto find_next; 1047 } 1048 } 1049 1050 if (!entry->addr.id && needs_id) 1051 goto out; 1052 1053 __set_bit(entry->addr.id, pernet->id_bitmap); 1054 if (entry->addr.id > pernet->next_id) 1055 pernet->next_id = entry->addr.id; 1056 1057 if (entry->flags & MPTCP_PM_ADDR_FLAG_SIGNAL) { 1058 addr_max = pernet->add_addr_signal_max; 1059 WRITE_ONCE(pernet->add_addr_signal_max, addr_max + 1); 1060 } 1061 if (entry->flags & MPTCP_PM_ADDR_FLAG_SUBFLOW) { 1062 addr_max = pernet->local_addr_max; 1063 WRITE_ONCE(pernet->local_addr_max, addr_max + 1); 1064 } 1065 1066 pernet->addrs++; 1067 if (!entry->addr.port) 1068 list_add_tail_rcu(&entry->list, &pernet->local_addr_list); 1069 else 1070 list_add_rcu(&entry->list, &pernet->local_addr_list); 1071 ret = entry->addr.id; 1072 1073 out: 1074 spin_unlock_bh(&pernet->lock); 1075 1076 /* just replaced an existing entry, free it */ 1077 if (del_entry) { 1078 synchronize_rcu(); 1079 __mptcp_pm_release_addr_entry(del_entry); 1080 } 1081 return ret; 1082 } 1083 1084 static struct lock_class_key mptcp_slock_keys[2]; 1085 static struct lock_class_key mptcp_keys[2]; 1086 1087 static int mptcp_pm_nl_create_listen_socket(struct sock *sk, 1088 struct mptcp_pm_addr_entry *entry) 1089 { 1090 bool is_ipv6 = sk->sk_family == AF_INET6; 1091 int addrlen = sizeof(struct sockaddr_in); 1092 struct sockaddr_storage addr; 1093 struct sock *newsk, *ssk; 1094 int backlog = 1024; 1095 int err; 1096 1097 err = sock_create_kern(sock_net(sk), entry->addr.family, 1098 SOCK_STREAM, IPPROTO_MPTCP, &entry->lsk); 1099 if (err) 1100 return err; 1101 1102 newsk = entry->lsk->sk; 1103 if (!newsk) 1104 return -EINVAL; 1105 1106 /* The subflow socket lock is acquired in a nested to the msk one 1107 * in several places, even by the TCP stack, and this msk is a kernel 1108 * socket: lockdep complains. Instead of propagating the _nested 1109 * modifiers in several places, re-init the lock class for the msk 1110 * socket to an mptcp specific one. 1111 */ 1112 sock_lock_init_class_and_name(newsk, 1113 is_ipv6 ? "mlock-AF_INET6" : "mlock-AF_INET", 1114 &mptcp_slock_keys[is_ipv6], 1115 is_ipv6 ? "msk_lock-AF_INET6" : "msk_lock-AF_INET", 1116 &mptcp_keys[is_ipv6]); 1117 1118 lock_sock(newsk); 1119 ssk = __mptcp_nmpc_sk(mptcp_sk(newsk)); 1120 release_sock(newsk); 1121 if (IS_ERR(ssk)) 1122 return PTR_ERR(ssk); 1123 1124 mptcp_info2sockaddr(&entry->addr, &addr, entry->addr.family); 1125 #if IS_ENABLED(CONFIG_MPTCP_IPV6) 1126 if (entry->addr.family == AF_INET6) 1127 addrlen = sizeof(struct sockaddr_in6); 1128 #endif 1129 if (ssk->sk_family == AF_INET) 1130 err = inet_bind_sk(ssk, (struct sockaddr *)&addr, addrlen); 1131 #if IS_ENABLED(CONFIG_MPTCP_IPV6) 1132 else if (ssk->sk_family == AF_INET6) 1133 err = inet6_bind_sk(ssk, (struct sockaddr *)&addr, addrlen); 1134 #endif 1135 if (err) 1136 return err; 1137 1138 /* We don't use mptcp_set_state() here because it needs to be called 1139 * under the msk socket lock. For the moment, that will not bring 1140 * anything more than only calling inet_sk_state_store(), because the 1141 * old status is known (TCP_CLOSE). 1142 */ 1143 inet_sk_state_store(newsk, TCP_LISTEN); 1144 lock_sock(ssk); 1145 WRITE_ONCE(mptcp_subflow_ctx(ssk)->pm_listener, true); 1146 err = __inet_listen_sk(ssk, backlog); 1147 if (!err) 1148 mptcp_event_pm_listener(ssk, MPTCP_EVENT_LISTENER_CREATED); 1149 release_sock(ssk); 1150 return err; 1151 } 1152 1153 int mptcp_pm_nl_get_local_id(struct mptcp_sock *msk, struct mptcp_addr_info *skc) 1154 { 1155 struct mptcp_pm_addr_entry *entry; 1156 struct pm_nl_pernet *pernet; 1157 int ret; 1158 1159 pernet = pm_nl_get_pernet_from_msk(msk); 1160 1161 rcu_read_lock(); 1162 entry = __lookup_addr(pernet, skc); 1163 ret = entry ? entry->addr.id : -1; 1164 rcu_read_unlock(); 1165 if (ret >= 0) 1166 return ret; 1167 1168 /* address not found, add to local list */ 1169 entry = kmalloc(sizeof(*entry), GFP_ATOMIC); 1170 if (!entry) 1171 return -ENOMEM; 1172 1173 entry->addr = *skc; 1174 entry->addr.id = 0; 1175 entry->addr.port = 0; 1176 entry->ifindex = 0; 1177 entry->flags = MPTCP_PM_ADDR_FLAG_IMPLICIT; 1178 entry->lsk = NULL; 1179 ret = mptcp_pm_nl_append_new_local_addr(pernet, entry, true, false); 1180 if (ret < 0) 1181 kfree(entry); 1182 1183 return ret; 1184 } 1185 1186 bool mptcp_pm_nl_is_backup(struct mptcp_sock *msk, struct mptcp_addr_info *skc) 1187 { 1188 struct pm_nl_pernet *pernet = pm_nl_get_pernet_from_msk(msk); 1189 struct mptcp_pm_addr_entry *entry; 1190 bool backup; 1191 1192 rcu_read_lock(); 1193 entry = __lookup_addr(pernet, skc); 1194 backup = entry && !!(entry->flags & MPTCP_PM_ADDR_FLAG_BACKUP); 1195 rcu_read_unlock(); 1196 1197 return backup; 1198 } 1199 1200 #define MPTCP_PM_CMD_GRP_OFFSET 0 1201 #define MPTCP_PM_EV_GRP_OFFSET 1 1202 1203 static const struct genl_multicast_group mptcp_pm_mcgrps[] = { 1204 [MPTCP_PM_CMD_GRP_OFFSET] = { .name = MPTCP_PM_CMD_GRP_NAME, }, 1205 [MPTCP_PM_EV_GRP_OFFSET] = { .name = MPTCP_PM_EV_GRP_NAME, 1206 .flags = GENL_MCAST_CAP_NET_ADMIN, 1207 }, 1208 }; 1209 1210 void mptcp_pm_nl_subflow_chk_stale(const struct mptcp_sock *msk, struct sock *ssk) 1211 { 1212 struct mptcp_subflow_context *iter, *subflow = mptcp_subflow_ctx(ssk); 1213 struct sock *sk = (struct sock *)msk; 1214 unsigned int active_max_loss_cnt; 1215 struct net *net = sock_net(sk); 1216 unsigned int stale_loss_cnt; 1217 bool slow; 1218 1219 stale_loss_cnt = mptcp_stale_loss_cnt(net); 1220 if (subflow->stale || !stale_loss_cnt || subflow->stale_count <= stale_loss_cnt) 1221 return; 1222 1223 /* look for another available subflow not in loss state */ 1224 active_max_loss_cnt = max_t(int, stale_loss_cnt - 1, 1); 1225 mptcp_for_each_subflow(msk, iter) { 1226 if (iter != subflow && mptcp_subflow_active(iter) && 1227 iter->stale_count < active_max_loss_cnt) { 1228 /* we have some alternatives, try to mark this subflow as idle ...*/ 1229 slow = lock_sock_fast(ssk); 1230 if (!tcp_rtx_and_write_queues_empty(ssk)) { 1231 subflow->stale = 1; 1232 __mptcp_retransmit_pending_data(sk); 1233 MPTCP_INC_STATS(net, MPTCP_MIB_SUBFLOWSTALE); 1234 } 1235 unlock_sock_fast(ssk, slow); 1236 1237 /* always try to push the pending data regardless of re-injections: 1238 * we can possibly use backup subflows now, and subflow selection 1239 * is cheap under the msk socket lock 1240 */ 1241 __mptcp_push_pending(sk, 0); 1242 return; 1243 } 1244 } 1245 } 1246 1247 static int mptcp_pm_family_to_addr(int family) 1248 { 1249 #if IS_ENABLED(CONFIG_MPTCP_IPV6) 1250 if (family == AF_INET6) 1251 return MPTCP_PM_ADDR_ATTR_ADDR6; 1252 #endif 1253 return MPTCP_PM_ADDR_ATTR_ADDR4; 1254 } 1255 1256 static int mptcp_pm_parse_pm_addr_attr(struct nlattr *tb[], 1257 const struct nlattr *attr, 1258 struct genl_info *info, 1259 struct mptcp_addr_info *addr, 1260 bool require_family) 1261 { 1262 int err, addr_addr; 1263 1264 if (!attr) { 1265 GENL_SET_ERR_MSG(info, "missing address info"); 1266 return -EINVAL; 1267 } 1268 1269 /* no validation needed - was already done via nested policy */ 1270 err = nla_parse_nested_deprecated(tb, MPTCP_PM_ADDR_ATTR_MAX, attr, 1271 mptcp_pm_address_nl_policy, info->extack); 1272 if (err) 1273 return err; 1274 1275 if (tb[MPTCP_PM_ADDR_ATTR_ID]) 1276 addr->id = nla_get_u8(tb[MPTCP_PM_ADDR_ATTR_ID]); 1277 1278 if (!tb[MPTCP_PM_ADDR_ATTR_FAMILY]) { 1279 if (!require_family) 1280 return 0; 1281 1282 NL_SET_ERR_MSG_ATTR(info->extack, attr, 1283 "missing family"); 1284 return -EINVAL; 1285 } 1286 1287 addr->family = nla_get_u16(tb[MPTCP_PM_ADDR_ATTR_FAMILY]); 1288 if (addr->family != AF_INET 1289 #if IS_ENABLED(CONFIG_MPTCP_IPV6) 1290 && addr->family != AF_INET6 1291 #endif 1292 ) { 1293 NL_SET_ERR_MSG_ATTR(info->extack, attr, 1294 "unknown address family"); 1295 return -EINVAL; 1296 } 1297 addr_addr = mptcp_pm_family_to_addr(addr->family); 1298 if (!tb[addr_addr]) { 1299 NL_SET_ERR_MSG_ATTR(info->extack, attr, 1300 "missing address data"); 1301 return -EINVAL; 1302 } 1303 1304 #if IS_ENABLED(CONFIG_MPTCP_IPV6) 1305 if (addr->family == AF_INET6) 1306 addr->addr6 = nla_get_in6_addr(tb[addr_addr]); 1307 else 1308 #endif 1309 addr->addr.s_addr = nla_get_in_addr(tb[addr_addr]); 1310 1311 if (tb[MPTCP_PM_ADDR_ATTR_PORT]) 1312 addr->port = htons(nla_get_u16(tb[MPTCP_PM_ADDR_ATTR_PORT])); 1313 1314 return 0; 1315 } 1316 1317 int mptcp_pm_parse_addr(struct nlattr *attr, struct genl_info *info, 1318 struct mptcp_addr_info *addr) 1319 { 1320 struct nlattr *tb[MPTCP_PM_ADDR_ATTR_MAX + 1]; 1321 1322 memset(addr, 0, sizeof(*addr)); 1323 1324 return mptcp_pm_parse_pm_addr_attr(tb, attr, info, addr, true); 1325 } 1326 1327 int mptcp_pm_parse_entry(struct nlattr *attr, struct genl_info *info, 1328 bool require_family, 1329 struct mptcp_pm_addr_entry *entry) 1330 { 1331 struct nlattr *tb[MPTCP_PM_ADDR_ATTR_MAX + 1]; 1332 int err; 1333 1334 memset(entry, 0, sizeof(*entry)); 1335 1336 err = mptcp_pm_parse_pm_addr_attr(tb, attr, info, &entry->addr, require_family); 1337 if (err) 1338 return err; 1339 1340 if (tb[MPTCP_PM_ADDR_ATTR_IF_IDX]) { 1341 u32 val = nla_get_s32(tb[MPTCP_PM_ADDR_ATTR_IF_IDX]); 1342 1343 entry->ifindex = val; 1344 } 1345 1346 if (tb[MPTCP_PM_ADDR_ATTR_FLAGS]) 1347 entry->flags = nla_get_u32(tb[MPTCP_PM_ADDR_ATTR_FLAGS]); 1348 1349 if (tb[MPTCP_PM_ADDR_ATTR_PORT]) 1350 entry->addr.port = htons(nla_get_u16(tb[MPTCP_PM_ADDR_ATTR_PORT])); 1351 1352 return 0; 1353 } 1354 1355 static struct pm_nl_pernet *genl_info_pm_nl(struct genl_info *info) 1356 { 1357 return pm_nl_get_pernet(genl_info_net(info)); 1358 } 1359 1360 static int mptcp_nl_add_subflow_or_signal_addr(struct net *net, 1361 struct mptcp_addr_info *addr) 1362 { 1363 struct mptcp_sock *msk; 1364 long s_slot = 0, s_num = 0; 1365 1366 while ((msk = mptcp_token_iter_next(net, &s_slot, &s_num)) != NULL) { 1367 struct sock *sk = (struct sock *)msk; 1368 struct mptcp_addr_info mpc_addr; 1369 1370 if (!READ_ONCE(msk->fully_established) || 1371 mptcp_pm_is_userspace(msk)) 1372 goto next; 1373 1374 /* if the endp linked to the init sf is re-added with a != ID */ 1375 mptcp_local_address((struct sock_common *)msk, &mpc_addr); 1376 1377 lock_sock(sk); 1378 spin_lock_bh(&msk->pm.lock); 1379 if (mptcp_addresses_equal(addr, &mpc_addr, addr->port)) 1380 msk->mpc_endpoint_id = addr->id; 1381 mptcp_pm_create_subflow_or_signal_addr(msk); 1382 spin_unlock_bh(&msk->pm.lock); 1383 release_sock(sk); 1384 1385 next: 1386 sock_put(sk); 1387 cond_resched(); 1388 } 1389 1390 return 0; 1391 } 1392 1393 static bool mptcp_pm_has_addr_attr_id(const struct nlattr *attr, 1394 struct genl_info *info) 1395 { 1396 struct nlattr *tb[MPTCP_PM_ADDR_ATTR_MAX + 1]; 1397 1398 if (!nla_parse_nested_deprecated(tb, MPTCP_PM_ADDR_ATTR_MAX, attr, 1399 mptcp_pm_address_nl_policy, info->extack) && 1400 tb[MPTCP_PM_ADDR_ATTR_ID]) 1401 return true; 1402 return false; 1403 } 1404 1405 int mptcp_pm_nl_add_addr_doit(struct sk_buff *skb, struct genl_info *info) 1406 { 1407 struct pm_nl_pernet *pernet = genl_info_pm_nl(info); 1408 struct mptcp_pm_addr_entry addr, *entry; 1409 struct nlattr *attr; 1410 int ret; 1411 1412 if (GENL_REQ_ATTR_CHECK(info, MPTCP_PM_ENDPOINT_ADDR)) 1413 return -EINVAL; 1414 1415 attr = info->attrs[MPTCP_PM_ENDPOINT_ADDR]; 1416 ret = mptcp_pm_parse_entry(attr, info, true, &addr); 1417 if (ret < 0) 1418 return ret; 1419 1420 if (addr.addr.port && !address_use_port(&addr)) { 1421 NL_SET_ERR_MSG_ATTR(info->extack, attr, 1422 "flags must have signal and not subflow when using port"); 1423 return -EINVAL; 1424 } 1425 1426 if (addr.flags & MPTCP_PM_ADDR_FLAG_SIGNAL && 1427 addr.flags & MPTCP_PM_ADDR_FLAG_FULLMESH) { 1428 NL_SET_ERR_MSG_ATTR(info->extack, attr, 1429 "flags mustn't have both signal and fullmesh"); 1430 return -EINVAL; 1431 } 1432 1433 if (addr.flags & MPTCP_PM_ADDR_FLAG_IMPLICIT) { 1434 NL_SET_ERR_MSG_ATTR(info->extack, attr, 1435 "can't create IMPLICIT endpoint"); 1436 return -EINVAL; 1437 } 1438 1439 entry = kzalloc(sizeof(*entry), GFP_KERNEL_ACCOUNT); 1440 if (!entry) { 1441 GENL_SET_ERR_MSG(info, "can't allocate addr"); 1442 return -ENOMEM; 1443 } 1444 1445 *entry = addr; 1446 if (entry->addr.port) { 1447 ret = mptcp_pm_nl_create_listen_socket(skb->sk, entry); 1448 if (ret) { 1449 GENL_SET_ERR_MSG_FMT(info, "create listen socket error: %d", ret); 1450 goto out_free; 1451 } 1452 } 1453 ret = mptcp_pm_nl_append_new_local_addr(pernet, entry, 1454 !mptcp_pm_has_addr_attr_id(attr, info), 1455 true); 1456 if (ret < 0) { 1457 GENL_SET_ERR_MSG_FMT(info, "too many addresses or duplicate one: %d", ret); 1458 goto out_free; 1459 } 1460 1461 mptcp_nl_add_subflow_or_signal_addr(sock_net(skb->sk), &entry->addr); 1462 return 0; 1463 1464 out_free: 1465 __mptcp_pm_release_addr_entry(entry); 1466 return ret; 1467 } 1468 1469 bool mptcp_remove_anno_list_by_saddr(struct mptcp_sock *msk, 1470 const struct mptcp_addr_info *addr) 1471 { 1472 struct mptcp_pm_add_entry *entry; 1473 1474 entry = mptcp_pm_del_add_timer(msk, addr, false); 1475 if (entry) { 1476 kfree(entry); 1477 return true; 1478 } 1479 1480 return false; 1481 } 1482 1483 static u8 mptcp_endp_get_local_id(struct mptcp_sock *msk, 1484 const struct mptcp_addr_info *addr) 1485 { 1486 return msk->mpc_endpoint_id == addr->id ? 0 : addr->id; 1487 } 1488 1489 static bool mptcp_pm_remove_anno_addr(struct mptcp_sock *msk, 1490 const struct mptcp_addr_info *addr, 1491 bool force) 1492 { 1493 struct mptcp_rm_list list = { .nr = 0 }; 1494 bool ret; 1495 1496 list.ids[list.nr++] = mptcp_endp_get_local_id(msk, addr); 1497 1498 ret = mptcp_remove_anno_list_by_saddr(msk, addr); 1499 if (ret || force) { 1500 spin_lock_bh(&msk->pm.lock); 1501 if (ret) { 1502 __set_bit(addr->id, msk->pm.id_avail_bitmap); 1503 msk->pm.add_addr_signaled--; 1504 } 1505 mptcp_pm_remove_addr(msk, &list); 1506 spin_unlock_bh(&msk->pm.lock); 1507 } 1508 return ret; 1509 } 1510 1511 static void __mark_subflow_endp_available(struct mptcp_sock *msk, u8 id) 1512 { 1513 /* If it was marked as used, and not ID 0, decrement local_addr_used */ 1514 if (!__test_and_set_bit(id ? : msk->mpc_endpoint_id, msk->pm.id_avail_bitmap) && 1515 id && !WARN_ON_ONCE(msk->pm.local_addr_used == 0)) 1516 msk->pm.local_addr_used--; 1517 } 1518 1519 static int mptcp_nl_remove_subflow_and_signal_addr(struct net *net, 1520 const struct mptcp_pm_addr_entry *entry) 1521 { 1522 const struct mptcp_addr_info *addr = &entry->addr; 1523 struct mptcp_rm_list list = { .nr = 1 }; 1524 long s_slot = 0, s_num = 0; 1525 struct mptcp_sock *msk; 1526 1527 pr_debug("remove_id=%d\n", addr->id); 1528 1529 while ((msk = mptcp_token_iter_next(net, &s_slot, &s_num)) != NULL) { 1530 struct sock *sk = (struct sock *)msk; 1531 bool remove_subflow; 1532 1533 if (mptcp_pm_is_userspace(msk)) 1534 goto next; 1535 1536 lock_sock(sk); 1537 remove_subflow = mptcp_lookup_subflow_by_saddr(&msk->conn_list, addr); 1538 mptcp_pm_remove_anno_addr(msk, addr, remove_subflow && 1539 !(entry->flags & MPTCP_PM_ADDR_FLAG_IMPLICIT)); 1540 1541 list.ids[0] = mptcp_endp_get_local_id(msk, addr); 1542 if (remove_subflow) { 1543 spin_lock_bh(&msk->pm.lock); 1544 mptcp_pm_nl_rm_subflow_received(msk, &list); 1545 spin_unlock_bh(&msk->pm.lock); 1546 } 1547 1548 if (entry->flags & MPTCP_PM_ADDR_FLAG_SUBFLOW) { 1549 spin_lock_bh(&msk->pm.lock); 1550 __mark_subflow_endp_available(msk, list.ids[0]); 1551 spin_unlock_bh(&msk->pm.lock); 1552 } 1553 1554 if (msk->mpc_endpoint_id == entry->addr.id) 1555 msk->mpc_endpoint_id = 0; 1556 release_sock(sk); 1557 1558 next: 1559 sock_put(sk); 1560 cond_resched(); 1561 } 1562 1563 return 0; 1564 } 1565 1566 static int mptcp_nl_remove_id_zero_address(struct net *net, 1567 struct mptcp_addr_info *addr) 1568 { 1569 struct mptcp_rm_list list = { .nr = 0 }; 1570 long s_slot = 0, s_num = 0; 1571 struct mptcp_sock *msk; 1572 1573 list.ids[list.nr++] = 0; 1574 1575 while ((msk = mptcp_token_iter_next(net, &s_slot, &s_num)) != NULL) { 1576 struct sock *sk = (struct sock *)msk; 1577 struct mptcp_addr_info msk_local; 1578 1579 if (list_empty(&msk->conn_list) || mptcp_pm_is_userspace(msk)) 1580 goto next; 1581 1582 mptcp_local_address((struct sock_common *)msk, &msk_local); 1583 if (!mptcp_addresses_equal(&msk_local, addr, addr->port)) 1584 goto next; 1585 1586 lock_sock(sk); 1587 spin_lock_bh(&msk->pm.lock); 1588 mptcp_pm_remove_addr(msk, &list); 1589 mptcp_pm_nl_rm_subflow_received(msk, &list); 1590 __mark_subflow_endp_available(msk, 0); 1591 spin_unlock_bh(&msk->pm.lock); 1592 release_sock(sk); 1593 1594 next: 1595 sock_put(sk); 1596 cond_resched(); 1597 } 1598 1599 return 0; 1600 } 1601 1602 int mptcp_pm_nl_del_addr_doit(struct sk_buff *skb, struct genl_info *info) 1603 { 1604 struct pm_nl_pernet *pernet = genl_info_pm_nl(info); 1605 struct mptcp_pm_addr_entry addr, *entry; 1606 unsigned int addr_max; 1607 struct nlattr *attr; 1608 int ret; 1609 1610 if (GENL_REQ_ATTR_CHECK(info, MPTCP_PM_ENDPOINT_ADDR)) 1611 return -EINVAL; 1612 1613 attr = info->attrs[MPTCP_PM_ENDPOINT_ADDR]; 1614 ret = mptcp_pm_parse_entry(attr, info, false, &addr); 1615 if (ret < 0) 1616 return ret; 1617 1618 /* the zero id address is special: the first address used by the msk 1619 * always gets such an id, so different subflows can have different zero 1620 * id addresses. Additionally zero id is not accounted for in id_bitmap. 1621 * Let's use an 'mptcp_rm_list' instead of the common remove code. 1622 */ 1623 if (addr.addr.id == 0) 1624 return mptcp_nl_remove_id_zero_address(sock_net(skb->sk), &addr.addr); 1625 1626 spin_lock_bh(&pernet->lock); 1627 entry = __lookup_addr_by_id(pernet, addr.addr.id); 1628 if (!entry) { 1629 NL_SET_ERR_MSG_ATTR(info->extack, attr, "address not found"); 1630 spin_unlock_bh(&pernet->lock); 1631 return -EINVAL; 1632 } 1633 if (entry->flags & MPTCP_PM_ADDR_FLAG_SIGNAL) { 1634 addr_max = pernet->add_addr_signal_max; 1635 WRITE_ONCE(pernet->add_addr_signal_max, addr_max - 1); 1636 } 1637 if (entry->flags & MPTCP_PM_ADDR_FLAG_SUBFLOW) { 1638 addr_max = pernet->local_addr_max; 1639 WRITE_ONCE(pernet->local_addr_max, addr_max - 1); 1640 } 1641 1642 pernet->addrs--; 1643 list_del_rcu(&entry->list); 1644 __clear_bit(entry->addr.id, pernet->id_bitmap); 1645 spin_unlock_bh(&pernet->lock); 1646 1647 mptcp_nl_remove_subflow_and_signal_addr(sock_net(skb->sk), entry); 1648 synchronize_rcu(); 1649 __mptcp_pm_release_addr_entry(entry); 1650 1651 return ret; 1652 } 1653 1654 static void mptcp_pm_flush_addrs_and_subflows(struct mptcp_sock *msk, 1655 struct list_head *rm_list) 1656 { 1657 struct mptcp_rm_list alist = { .nr = 0 }, slist = { .nr = 0 }; 1658 struct mptcp_pm_addr_entry *entry; 1659 1660 list_for_each_entry(entry, rm_list, list) { 1661 if (slist.nr < MPTCP_RM_IDS_MAX && 1662 mptcp_lookup_subflow_by_saddr(&msk->conn_list, &entry->addr)) 1663 slist.ids[slist.nr++] = mptcp_endp_get_local_id(msk, &entry->addr); 1664 1665 if (alist.nr < MPTCP_RM_IDS_MAX && 1666 mptcp_remove_anno_list_by_saddr(msk, &entry->addr)) 1667 alist.ids[alist.nr++] = mptcp_endp_get_local_id(msk, &entry->addr); 1668 } 1669 1670 spin_lock_bh(&msk->pm.lock); 1671 if (alist.nr) { 1672 msk->pm.add_addr_signaled -= alist.nr; 1673 mptcp_pm_remove_addr(msk, &alist); 1674 } 1675 if (slist.nr) 1676 mptcp_pm_nl_rm_subflow_received(msk, &slist); 1677 /* Reset counters: maybe some subflows have been removed before */ 1678 bitmap_fill(msk->pm.id_avail_bitmap, MPTCP_PM_MAX_ADDR_ID + 1); 1679 msk->pm.local_addr_used = 0; 1680 spin_unlock_bh(&msk->pm.lock); 1681 } 1682 1683 static void mptcp_nl_flush_addrs_list(struct net *net, 1684 struct list_head *rm_list) 1685 { 1686 long s_slot = 0, s_num = 0; 1687 struct mptcp_sock *msk; 1688 1689 if (list_empty(rm_list)) 1690 return; 1691 1692 while ((msk = mptcp_token_iter_next(net, &s_slot, &s_num)) != NULL) { 1693 struct sock *sk = (struct sock *)msk; 1694 1695 if (!mptcp_pm_is_userspace(msk)) { 1696 lock_sock(sk); 1697 mptcp_pm_flush_addrs_and_subflows(msk, rm_list); 1698 release_sock(sk); 1699 } 1700 1701 sock_put(sk); 1702 cond_resched(); 1703 } 1704 } 1705 1706 /* caller must ensure the RCU grace period is already elapsed */ 1707 static void __flush_addrs(struct list_head *list) 1708 { 1709 while (!list_empty(list)) { 1710 struct mptcp_pm_addr_entry *cur; 1711 1712 cur = list_entry(list->next, 1713 struct mptcp_pm_addr_entry, list); 1714 list_del_rcu(&cur->list); 1715 __mptcp_pm_release_addr_entry(cur); 1716 } 1717 } 1718 1719 static void __reset_counters(struct pm_nl_pernet *pernet) 1720 { 1721 WRITE_ONCE(pernet->add_addr_signal_max, 0); 1722 WRITE_ONCE(pernet->add_addr_accept_max, 0); 1723 WRITE_ONCE(pernet->local_addr_max, 0); 1724 pernet->addrs = 0; 1725 } 1726 1727 int mptcp_pm_nl_flush_addrs_doit(struct sk_buff *skb, struct genl_info *info) 1728 { 1729 struct pm_nl_pernet *pernet = genl_info_pm_nl(info); 1730 LIST_HEAD(free_list); 1731 1732 spin_lock_bh(&pernet->lock); 1733 list_splice_init(&pernet->local_addr_list, &free_list); 1734 __reset_counters(pernet); 1735 pernet->next_id = 1; 1736 bitmap_zero(pernet->id_bitmap, MPTCP_PM_MAX_ADDR_ID + 1); 1737 spin_unlock_bh(&pernet->lock); 1738 mptcp_nl_flush_addrs_list(sock_net(skb->sk), &free_list); 1739 synchronize_rcu(); 1740 __flush_addrs(&free_list); 1741 return 0; 1742 } 1743 1744 int mptcp_nl_fill_addr(struct sk_buff *skb, 1745 struct mptcp_pm_addr_entry *entry) 1746 { 1747 struct mptcp_addr_info *addr = &entry->addr; 1748 struct nlattr *attr; 1749 1750 attr = nla_nest_start(skb, MPTCP_PM_ATTR_ADDR); 1751 if (!attr) 1752 return -EMSGSIZE; 1753 1754 if (nla_put_u16(skb, MPTCP_PM_ADDR_ATTR_FAMILY, addr->family)) 1755 goto nla_put_failure; 1756 if (nla_put_u16(skb, MPTCP_PM_ADDR_ATTR_PORT, ntohs(addr->port))) 1757 goto nla_put_failure; 1758 if (nla_put_u8(skb, MPTCP_PM_ADDR_ATTR_ID, addr->id)) 1759 goto nla_put_failure; 1760 if (nla_put_u32(skb, MPTCP_PM_ADDR_ATTR_FLAGS, entry->flags)) 1761 goto nla_put_failure; 1762 if (entry->ifindex && 1763 nla_put_s32(skb, MPTCP_PM_ADDR_ATTR_IF_IDX, entry->ifindex)) 1764 goto nla_put_failure; 1765 1766 if (addr->family == AF_INET && 1767 nla_put_in_addr(skb, MPTCP_PM_ADDR_ATTR_ADDR4, 1768 addr->addr.s_addr)) 1769 goto nla_put_failure; 1770 #if IS_ENABLED(CONFIG_MPTCP_IPV6) 1771 else if (addr->family == AF_INET6 && 1772 nla_put_in6_addr(skb, MPTCP_PM_ADDR_ATTR_ADDR6, &addr->addr6)) 1773 goto nla_put_failure; 1774 #endif 1775 nla_nest_end(skb, attr); 1776 return 0; 1777 1778 nla_put_failure: 1779 nla_nest_cancel(skb, attr); 1780 return -EMSGSIZE; 1781 } 1782 1783 int mptcp_pm_nl_get_addr(u8 id, struct mptcp_pm_addr_entry *addr, 1784 struct genl_info *info) 1785 { 1786 struct pm_nl_pernet *pernet = genl_info_pm_nl(info); 1787 struct mptcp_pm_addr_entry *entry; 1788 int ret = -EINVAL; 1789 1790 rcu_read_lock(); 1791 entry = __lookup_addr_by_id(pernet, id); 1792 if (entry) { 1793 *addr = *entry; 1794 ret = 0; 1795 } 1796 rcu_read_unlock(); 1797 1798 return ret; 1799 } 1800 1801 int mptcp_pm_nl_dump_addr(struct sk_buff *msg, 1802 struct netlink_callback *cb) 1803 { 1804 struct net *net = sock_net(msg->sk); 1805 struct mptcp_pm_addr_entry *entry; 1806 struct pm_nl_pernet *pernet; 1807 int id = cb->args[0]; 1808 int i; 1809 1810 pernet = pm_nl_get_pernet(net); 1811 1812 rcu_read_lock(); 1813 for (i = id; i < MPTCP_PM_MAX_ADDR_ID + 1; i++) { 1814 if (test_bit(i, pernet->id_bitmap)) { 1815 entry = __lookup_addr_by_id(pernet, i); 1816 if (!entry) 1817 break; 1818 1819 if (entry->addr.id <= id) 1820 continue; 1821 1822 if (mptcp_pm_genl_fill_addr(msg, cb, entry) < 0) 1823 break; 1824 1825 id = entry->addr.id; 1826 } 1827 } 1828 rcu_read_unlock(); 1829 1830 cb->args[0] = id; 1831 return msg->len; 1832 } 1833 1834 static int parse_limit(struct genl_info *info, int id, unsigned int *limit) 1835 { 1836 struct nlattr *attr = info->attrs[id]; 1837 1838 if (!attr) 1839 return 0; 1840 1841 *limit = nla_get_u32(attr); 1842 if (*limit > MPTCP_PM_ADDR_MAX) { 1843 NL_SET_ERR_MSG_ATTR_FMT(info->extack, attr, 1844 "limit greater than maximum (%u)", 1845 MPTCP_PM_ADDR_MAX); 1846 return -EINVAL; 1847 } 1848 return 0; 1849 } 1850 1851 int mptcp_pm_nl_set_limits_doit(struct sk_buff *skb, struct genl_info *info) 1852 { 1853 struct pm_nl_pernet *pernet = genl_info_pm_nl(info); 1854 unsigned int rcv_addrs, subflows; 1855 int ret; 1856 1857 spin_lock_bh(&pernet->lock); 1858 rcv_addrs = pernet->add_addr_accept_max; 1859 ret = parse_limit(info, MPTCP_PM_ATTR_RCV_ADD_ADDRS, &rcv_addrs); 1860 if (ret) 1861 goto unlock; 1862 1863 subflows = pernet->subflows_max; 1864 ret = parse_limit(info, MPTCP_PM_ATTR_SUBFLOWS, &subflows); 1865 if (ret) 1866 goto unlock; 1867 1868 WRITE_ONCE(pernet->add_addr_accept_max, rcv_addrs); 1869 WRITE_ONCE(pernet->subflows_max, subflows); 1870 1871 unlock: 1872 spin_unlock_bh(&pernet->lock); 1873 return ret; 1874 } 1875 1876 int mptcp_pm_nl_get_limits_doit(struct sk_buff *skb, struct genl_info *info) 1877 { 1878 struct pm_nl_pernet *pernet = genl_info_pm_nl(info); 1879 struct sk_buff *msg; 1880 void *reply; 1881 1882 msg = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL); 1883 if (!msg) 1884 return -ENOMEM; 1885 1886 reply = genlmsg_put_reply(msg, info, &mptcp_genl_family, 0, 1887 MPTCP_PM_CMD_GET_LIMITS); 1888 if (!reply) 1889 goto fail; 1890 1891 if (nla_put_u32(msg, MPTCP_PM_ATTR_RCV_ADD_ADDRS, 1892 READ_ONCE(pernet->add_addr_accept_max))) 1893 goto fail; 1894 1895 if (nla_put_u32(msg, MPTCP_PM_ATTR_SUBFLOWS, 1896 READ_ONCE(pernet->subflows_max))) 1897 goto fail; 1898 1899 genlmsg_end(msg, reply); 1900 return genlmsg_reply(msg, info); 1901 1902 fail: 1903 GENL_SET_ERR_MSG(info, "not enough space in Netlink message"); 1904 nlmsg_free(msg); 1905 return -EMSGSIZE; 1906 } 1907 1908 static void mptcp_pm_nl_fullmesh(struct mptcp_sock *msk, 1909 struct mptcp_addr_info *addr) 1910 { 1911 struct mptcp_rm_list list = { .nr = 0 }; 1912 1913 list.ids[list.nr++] = mptcp_endp_get_local_id(msk, addr); 1914 1915 spin_lock_bh(&msk->pm.lock); 1916 mptcp_pm_nl_rm_subflow_received(msk, &list); 1917 __mark_subflow_endp_available(msk, list.ids[0]); 1918 mptcp_pm_create_subflow_or_signal_addr(msk); 1919 spin_unlock_bh(&msk->pm.lock); 1920 } 1921 1922 static void mptcp_nl_set_flags(struct net *net, 1923 struct mptcp_pm_addr_entry *local, 1924 u8 changed) 1925 { 1926 u8 is_subflow = !!(local->flags & MPTCP_PM_ADDR_FLAG_SUBFLOW); 1927 u8 bkup = !!(local->flags & MPTCP_PM_ADDR_FLAG_BACKUP); 1928 long s_slot = 0, s_num = 0; 1929 struct mptcp_sock *msk; 1930 1931 if (changed == MPTCP_PM_ADDR_FLAG_FULLMESH && !is_subflow) 1932 return; 1933 1934 while ((msk = mptcp_token_iter_next(net, &s_slot, &s_num)) != NULL) { 1935 struct sock *sk = (struct sock *)msk; 1936 1937 if (list_empty(&msk->conn_list) || mptcp_pm_is_userspace(msk)) 1938 goto next; 1939 1940 lock_sock(sk); 1941 if (changed & MPTCP_PM_ADDR_FLAG_BACKUP) 1942 mptcp_pm_nl_mp_prio_send_ack(msk, &local->addr, NULL, bkup); 1943 /* Subflows will only be recreated if the SUBFLOW flag is set */ 1944 if (is_subflow && (changed & MPTCP_PM_ADDR_FLAG_FULLMESH)) 1945 mptcp_pm_nl_fullmesh(msk, &local->addr); 1946 release_sock(sk); 1947 1948 next: 1949 sock_put(sk); 1950 cond_resched(); 1951 } 1952 1953 return; 1954 } 1955 1956 int mptcp_pm_nl_set_flags(struct mptcp_pm_addr_entry *local, 1957 struct genl_info *info) 1958 { 1959 struct nlattr *attr = info->attrs[MPTCP_PM_ATTR_ADDR]; 1960 u8 changed, mask = MPTCP_PM_ADDR_FLAG_BACKUP | 1961 MPTCP_PM_ADDR_FLAG_FULLMESH; 1962 struct net *net = genl_info_net(info); 1963 struct mptcp_pm_addr_entry *entry; 1964 struct pm_nl_pernet *pernet; 1965 u8 lookup_by_id = 0; 1966 1967 pernet = pm_nl_get_pernet(net); 1968 1969 if (local->addr.family == AF_UNSPEC) { 1970 lookup_by_id = 1; 1971 if (!local->addr.id) { 1972 NL_SET_ERR_MSG_ATTR(info->extack, attr, 1973 "missing address ID"); 1974 return -EOPNOTSUPP; 1975 } 1976 } 1977 1978 spin_lock_bh(&pernet->lock); 1979 entry = lookup_by_id ? __lookup_addr_by_id(pernet, local->addr.id) : 1980 __lookup_addr(pernet, &local->addr); 1981 if (!entry) { 1982 spin_unlock_bh(&pernet->lock); 1983 NL_SET_ERR_MSG_ATTR(info->extack, attr, "address not found"); 1984 return -EINVAL; 1985 } 1986 if ((local->flags & MPTCP_PM_ADDR_FLAG_FULLMESH) && 1987 (entry->flags & (MPTCP_PM_ADDR_FLAG_SIGNAL | 1988 MPTCP_PM_ADDR_FLAG_IMPLICIT))) { 1989 spin_unlock_bh(&pernet->lock); 1990 NL_SET_ERR_MSG_ATTR(info->extack, attr, "invalid addr flags"); 1991 return -EINVAL; 1992 } 1993 1994 changed = (local->flags ^ entry->flags) & mask; 1995 entry->flags = (entry->flags & ~mask) | (local->flags & mask); 1996 *local = *entry; 1997 spin_unlock_bh(&pernet->lock); 1998 1999 mptcp_nl_set_flags(net, local, changed); 2000 return 0; 2001 } 2002 2003 static void mptcp_nl_mcast_send(struct net *net, struct sk_buff *nlskb, gfp_t gfp) 2004 { 2005 genlmsg_multicast_netns(&mptcp_genl_family, net, 2006 nlskb, 0, MPTCP_PM_EV_GRP_OFFSET, gfp); 2007 } 2008 2009 bool mptcp_userspace_pm_active(const struct mptcp_sock *msk) 2010 { 2011 return genl_has_listeners(&mptcp_genl_family, 2012 sock_net((const struct sock *)msk), 2013 MPTCP_PM_EV_GRP_OFFSET); 2014 } 2015 2016 static int mptcp_event_add_subflow(struct sk_buff *skb, const struct sock *ssk) 2017 { 2018 const struct inet_sock *issk = inet_sk(ssk); 2019 const struct mptcp_subflow_context *sf; 2020 2021 if (nla_put_u16(skb, MPTCP_ATTR_FAMILY, ssk->sk_family)) 2022 return -EMSGSIZE; 2023 2024 switch (ssk->sk_family) { 2025 case AF_INET: 2026 if (nla_put_in_addr(skb, MPTCP_ATTR_SADDR4, issk->inet_saddr)) 2027 return -EMSGSIZE; 2028 if (nla_put_in_addr(skb, MPTCP_ATTR_DADDR4, issk->inet_daddr)) 2029 return -EMSGSIZE; 2030 break; 2031 #if IS_ENABLED(CONFIG_MPTCP_IPV6) 2032 case AF_INET6: { 2033 if (nla_put_in6_addr(skb, MPTCP_ATTR_SADDR6, &issk->pinet6->saddr)) 2034 return -EMSGSIZE; 2035 if (nla_put_in6_addr(skb, MPTCP_ATTR_DADDR6, &ssk->sk_v6_daddr)) 2036 return -EMSGSIZE; 2037 break; 2038 } 2039 #endif 2040 default: 2041 WARN_ON_ONCE(1); 2042 return -EMSGSIZE; 2043 } 2044 2045 if (nla_put_be16(skb, MPTCP_ATTR_SPORT, issk->inet_sport)) 2046 return -EMSGSIZE; 2047 if (nla_put_be16(skb, MPTCP_ATTR_DPORT, issk->inet_dport)) 2048 return -EMSGSIZE; 2049 2050 sf = mptcp_subflow_ctx(ssk); 2051 if (WARN_ON_ONCE(!sf)) 2052 return -EINVAL; 2053 2054 if (nla_put_u8(skb, MPTCP_ATTR_LOC_ID, subflow_get_local_id(sf))) 2055 return -EMSGSIZE; 2056 2057 if (nla_put_u8(skb, MPTCP_ATTR_REM_ID, sf->remote_id)) 2058 return -EMSGSIZE; 2059 2060 return 0; 2061 } 2062 2063 static int mptcp_event_put_token_and_ssk(struct sk_buff *skb, 2064 const struct mptcp_sock *msk, 2065 const struct sock *ssk) 2066 { 2067 const struct sock *sk = (const struct sock *)msk; 2068 const struct mptcp_subflow_context *sf; 2069 u8 sk_err; 2070 2071 if (nla_put_u32(skb, MPTCP_ATTR_TOKEN, READ_ONCE(msk->token))) 2072 return -EMSGSIZE; 2073 2074 if (mptcp_event_add_subflow(skb, ssk)) 2075 return -EMSGSIZE; 2076 2077 sf = mptcp_subflow_ctx(ssk); 2078 if (WARN_ON_ONCE(!sf)) 2079 return -EINVAL; 2080 2081 if (nla_put_u8(skb, MPTCP_ATTR_BACKUP, sf->backup)) 2082 return -EMSGSIZE; 2083 2084 if (ssk->sk_bound_dev_if && 2085 nla_put_s32(skb, MPTCP_ATTR_IF_IDX, ssk->sk_bound_dev_if)) 2086 return -EMSGSIZE; 2087 2088 sk_err = READ_ONCE(ssk->sk_err); 2089 if (sk_err && sk->sk_state == TCP_ESTABLISHED && 2090 nla_put_u8(skb, MPTCP_ATTR_ERROR, sk_err)) 2091 return -EMSGSIZE; 2092 2093 return 0; 2094 } 2095 2096 static int mptcp_event_sub_established(struct sk_buff *skb, 2097 const struct mptcp_sock *msk, 2098 const struct sock *ssk) 2099 { 2100 return mptcp_event_put_token_and_ssk(skb, msk, ssk); 2101 } 2102 2103 static int mptcp_event_sub_closed(struct sk_buff *skb, 2104 const struct mptcp_sock *msk, 2105 const struct sock *ssk) 2106 { 2107 const struct mptcp_subflow_context *sf; 2108 2109 if (mptcp_event_put_token_and_ssk(skb, msk, ssk)) 2110 return -EMSGSIZE; 2111 2112 sf = mptcp_subflow_ctx(ssk); 2113 if (!sf->reset_seen) 2114 return 0; 2115 2116 if (nla_put_u32(skb, MPTCP_ATTR_RESET_REASON, sf->reset_reason)) 2117 return -EMSGSIZE; 2118 2119 if (nla_put_u32(skb, MPTCP_ATTR_RESET_FLAGS, sf->reset_transient)) 2120 return -EMSGSIZE; 2121 2122 return 0; 2123 } 2124 2125 static int mptcp_event_created(struct sk_buff *skb, 2126 const struct mptcp_sock *msk, 2127 const struct sock *ssk) 2128 { 2129 int err = nla_put_u32(skb, MPTCP_ATTR_TOKEN, READ_ONCE(msk->token)); 2130 2131 if (err) 2132 return err; 2133 2134 if (nla_put_u8(skb, MPTCP_ATTR_SERVER_SIDE, READ_ONCE(msk->pm.server_side))) 2135 return -EMSGSIZE; 2136 2137 return mptcp_event_add_subflow(skb, ssk); 2138 } 2139 2140 void mptcp_event_addr_removed(const struct mptcp_sock *msk, uint8_t id) 2141 { 2142 struct net *net = sock_net((const struct sock *)msk); 2143 struct nlmsghdr *nlh; 2144 struct sk_buff *skb; 2145 2146 if (!genl_has_listeners(&mptcp_genl_family, net, MPTCP_PM_EV_GRP_OFFSET)) 2147 return; 2148 2149 skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_ATOMIC); 2150 if (!skb) 2151 return; 2152 2153 nlh = genlmsg_put(skb, 0, 0, &mptcp_genl_family, 0, MPTCP_EVENT_REMOVED); 2154 if (!nlh) 2155 goto nla_put_failure; 2156 2157 if (nla_put_u32(skb, MPTCP_ATTR_TOKEN, READ_ONCE(msk->token))) 2158 goto nla_put_failure; 2159 2160 if (nla_put_u8(skb, MPTCP_ATTR_REM_ID, id)) 2161 goto nla_put_failure; 2162 2163 genlmsg_end(skb, nlh); 2164 mptcp_nl_mcast_send(net, skb, GFP_ATOMIC); 2165 return; 2166 2167 nla_put_failure: 2168 nlmsg_free(skb); 2169 } 2170 2171 void mptcp_event_addr_announced(const struct sock *ssk, 2172 const struct mptcp_addr_info *info) 2173 { 2174 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk); 2175 struct mptcp_sock *msk = mptcp_sk(subflow->conn); 2176 struct net *net = sock_net(ssk); 2177 struct nlmsghdr *nlh; 2178 struct sk_buff *skb; 2179 2180 if (!genl_has_listeners(&mptcp_genl_family, net, MPTCP_PM_EV_GRP_OFFSET)) 2181 return; 2182 2183 skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_ATOMIC); 2184 if (!skb) 2185 return; 2186 2187 nlh = genlmsg_put(skb, 0, 0, &mptcp_genl_family, 0, 2188 MPTCP_EVENT_ANNOUNCED); 2189 if (!nlh) 2190 goto nla_put_failure; 2191 2192 if (nla_put_u32(skb, MPTCP_ATTR_TOKEN, READ_ONCE(msk->token))) 2193 goto nla_put_failure; 2194 2195 if (nla_put_u8(skb, MPTCP_ATTR_REM_ID, info->id)) 2196 goto nla_put_failure; 2197 2198 if (nla_put_be16(skb, MPTCP_ATTR_DPORT, 2199 info->port == 0 ? 2200 inet_sk(ssk)->inet_dport : 2201 info->port)) 2202 goto nla_put_failure; 2203 2204 switch (info->family) { 2205 case AF_INET: 2206 if (nla_put_in_addr(skb, MPTCP_ATTR_DADDR4, info->addr.s_addr)) 2207 goto nla_put_failure; 2208 break; 2209 #if IS_ENABLED(CONFIG_MPTCP_IPV6) 2210 case AF_INET6: 2211 if (nla_put_in6_addr(skb, MPTCP_ATTR_DADDR6, &info->addr6)) 2212 goto nla_put_failure; 2213 break; 2214 #endif 2215 default: 2216 WARN_ON_ONCE(1); 2217 goto nla_put_failure; 2218 } 2219 2220 genlmsg_end(skb, nlh); 2221 mptcp_nl_mcast_send(net, skb, GFP_ATOMIC); 2222 return; 2223 2224 nla_put_failure: 2225 nlmsg_free(skb); 2226 } 2227 2228 void mptcp_event_pm_listener(const struct sock *ssk, 2229 enum mptcp_event_type event) 2230 { 2231 const struct inet_sock *issk = inet_sk(ssk); 2232 struct net *net = sock_net(ssk); 2233 struct nlmsghdr *nlh; 2234 struct sk_buff *skb; 2235 2236 if (!genl_has_listeners(&mptcp_genl_family, net, MPTCP_PM_EV_GRP_OFFSET)) 2237 return; 2238 2239 skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL); 2240 if (!skb) 2241 return; 2242 2243 nlh = genlmsg_put(skb, 0, 0, &mptcp_genl_family, 0, event); 2244 if (!nlh) 2245 goto nla_put_failure; 2246 2247 if (nla_put_u16(skb, MPTCP_ATTR_FAMILY, ssk->sk_family)) 2248 goto nla_put_failure; 2249 2250 if (nla_put_be16(skb, MPTCP_ATTR_SPORT, issk->inet_sport)) 2251 goto nla_put_failure; 2252 2253 switch (ssk->sk_family) { 2254 case AF_INET: 2255 if (nla_put_in_addr(skb, MPTCP_ATTR_SADDR4, issk->inet_saddr)) 2256 goto nla_put_failure; 2257 break; 2258 #if IS_ENABLED(CONFIG_MPTCP_IPV6) 2259 case AF_INET6: { 2260 if (nla_put_in6_addr(skb, MPTCP_ATTR_SADDR6, &issk->pinet6->saddr)) 2261 goto nla_put_failure; 2262 break; 2263 } 2264 #endif 2265 default: 2266 WARN_ON_ONCE(1); 2267 goto nla_put_failure; 2268 } 2269 2270 genlmsg_end(skb, nlh); 2271 mptcp_nl_mcast_send(net, skb, GFP_KERNEL); 2272 return; 2273 2274 nla_put_failure: 2275 nlmsg_free(skb); 2276 } 2277 2278 void mptcp_event(enum mptcp_event_type type, const struct mptcp_sock *msk, 2279 const struct sock *ssk, gfp_t gfp) 2280 { 2281 struct net *net = sock_net((const struct sock *)msk); 2282 struct nlmsghdr *nlh; 2283 struct sk_buff *skb; 2284 2285 if (!genl_has_listeners(&mptcp_genl_family, net, MPTCP_PM_EV_GRP_OFFSET)) 2286 return; 2287 2288 skb = nlmsg_new(NLMSG_DEFAULT_SIZE, gfp); 2289 if (!skb) 2290 return; 2291 2292 nlh = genlmsg_put(skb, 0, 0, &mptcp_genl_family, 0, type); 2293 if (!nlh) 2294 goto nla_put_failure; 2295 2296 switch (type) { 2297 case MPTCP_EVENT_UNSPEC: 2298 WARN_ON_ONCE(1); 2299 break; 2300 case MPTCP_EVENT_CREATED: 2301 case MPTCP_EVENT_ESTABLISHED: 2302 if (mptcp_event_created(skb, msk, ssk) < 0) 2303 goto nla_put_failure; 2304 break; 2305 case MPTCP_EVENT_CLOSED: 2306 if (nla_put_u32(skb, MPTCP_ATTR_TOKEN, READ_ONCE(msk->token)) < 0) 2307 goto nla_put_failure; 2308 break; 2309 case MPTCP_EVENT_ANNOUNCED: 2310 case MPTCP_EVENT_REMOVED: 2311 /* call mptcp_event_addr_announced()/removed instead */ 2312 WARN_ON_ONCE(1); 2313 break; 2314 case MPTCP_EVENT_SUB_ESTABLISHED: 2315 case MPTCP_EVENT_SUB_PRIORITY: 2316 if (mptcp_event_sub_established(skb, msk, ssk) < 0) 2317 goto nla_put_failure; 2318 break; 2319 case MPTCP_EVENT_SUB_CLOSED: 2320 if (mptcp_event_sub_closed(skb, msk, ssk) < 0) 2321 goto nla_put_failure; 2322 break; 2323 case MPTCP_EVENT_LISTENER_CREATED: 2324 case MPTCP_EVENT_LISTENER_CLOSED: 2325 break; 2326 } 2327 2328 genlmsg_end(skb, nlh); 2329 mptcp_nl_mcast_send(net, skb, gfp); 2330 return; 2331 2332 nla_put_failure: 2333 nlmsg_free(skb); 2334 } 2335 2336 struct genl_family mptcp_genl_family __ro_after_init = { 2337 .name = MPTCP_PM_NAME, 2338 .version = MPTCP_PM_VER, 2339 .netnsok = true, 2340 .module = THIS_MODULE, 2341 .ops = mptcp_pm_nl_ops, 2342 .n_ops = ARRAY_SIZE(mptcp_pm_nl_ops), 2343 .resv_start_op = MPTCP_PM_CMD_SUBFLOW_DESTROY + 1, 2344 .mcgrps = mptcp_pm_mcgrps, 2345 .n_mcgrps = ARRAY_SIZE(mptcp_pm_mcgrps), 2346 }; 2347 2348 static int __net_init pm_nl_init_net(struct net *net) 2349 { 2350 struct pm_nl_pernet *pernet = pm_nl_get_pernet(net); 2351 2352 INIT_LIST_HEAD_RCU(&pernet->local_addr_list); 2353 2354 /* Cit. 2 subflows ought to be enough for anybody. */ 2355 pernet->subflows_max = 2; 2356 pernet->next_id = 1; 2357 pernet->stale_loss_cnt = 4; 2358 spin_lock_init(&pernet->lock); 2359 2360 /* No need to initialize other pernet fields, the struct is zeroed at 2361 * allocation time. 2362 */ 2363 2364 return 0; 2365 } 2366 2367 static void __net_exit pm_nl_exit_net(struct list_head *net_list) 2368 { 2369 struct net *net; 2370 2371 list_for_each_entry(net, net_list, exit_list) { 2372 struct pm_nl_pernet *pernet = pm_nl_get_pernet(net); 2373 2374 /* net is removed from namespace list, can't race with 2375 * other modifiers, also netns core already waited for a 2376 * RCU grace period. 2377 */ 2378 __flush_addrs(&pernet->local_addr_list); 2379 } 2380 } 2381 2382 static struct pernet_operations mptcp_pm_pernet_ops = { 2383 .init = pm_nl_init_net, 2384 .exit_batch = pm_nl_exit_net, 2385 .id = &pm_nl_pernet_id, 2386 .size = sizeof(struct pm_nl_pernet), 2387 }; 2388 2389 void __init mptcp_pm_nl_init(void) 2390 { 2391 if (register_pernet_subsys(&mptcp_pm_pernet_ops) < 0) 2392 panic("Failed to register MPTCP PM pernet subsystem.\n"); 2393 2394 if (genl_register_family(&mptcp_genl_family)) 2395 panic("Failed to register MPTCP PM netlink family\n"); 2396 } 2397