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/tcp.h> 12 #include <net/inet_common.h> 13 #include <net/netns/generic.h> 14 #include <net/mptcp.h> 15 #include <net/genetlink.h> 16 #include <uapi/linux/mptcp.h> 17 18 #include "protocol.h" 19 #include "mib.h" 20 21 /* forward declaration */ 22 static struct genl_family mptcp_genl_family; 23 24 static int pm_nl_pernet_id; 25 26 struct mptcp_pm_add_entry { 27 struct list_head list; 28 struct mptcp_addr_info addr; 29 u8 retrans_times; 30 struct timer_list add_timer; 31 struct mptcp_sock *sock; 32 }; 33 34 struct pm_nl_pernet { 35 /* protects pernet updates */ 36 spinlock_t lock; 37 struct list_head local_addr_list; 38 unsigned int addrs; 39 unsigned int stale_loss_cnt; 40 unsigned int add_addr_signal_max; 41 unsigned int add_addr_accept_max; 42 unsigned int local_addr_max; 43 unsigned int subflows_max; 44 unsigned int next_id; 45 DECLARE_BITMAP(id_bitmap, MPTCP_PM_MAX_ADDR_ID + 1); 46 }; 47 48 #define MPTCP_PM_ADDR_MAX 8 49 #define ADD_ADDR_RETRANS_MAX 3 50 51 static struct pm_nl_pernet *pm_nl_get_pernet(const struct net *net) 52 { 53 return net_generic(net, pm_nl_pernet_id); 54 } 55 56 static struct pm_nl_pernet * 57 pm_nl_get_pernet_from_msk(const struct mptcp_sock *msk) 58 { 59 return pm_nl_get_pernet(sock_net((struct sock *)msk)); 60 } 61 62 bool mptcp_addresses_equal(const struct mptcp_addr_info *a, 63 const struct mptcp_addr_info *b, bool use_port) 64 { 65 bool addr_equals = false; 66 67 if (a->family == b->family) { 68 if (a->family == AF_INET) 69 addr_equals = a->addr.s_addr == b->addr.s_addr; 70 #if IS_ENABLED(CONFIG_MPTCP_IPV6) 71 else 72 addr_equals = !ipv6_addr_cmp(&a->addr6, &b->addr6); 73 } else if (a->family == AF_INET) { 74 if (ipv6_addr_v4mapped(&b->addr6)) 75 addr_equals = a->addr.s_addr == b->addr6.s6_addr32[3]; 76 } else if (b->family == AF_INET) { 77 if (ipv6_addr_v4mapped(&a->addr6)) 78 addr_equals = a->addr6.s6_addr32[3] == b->addr.s_addr; 79 #endif 80 } 81 82 if (!addr_equals) 83 return false; 84 if (!use_port) 85 return true; 86 87 return a->port == b->port; 88 } 89 90 void mptcp_local_address(const struct sock_common *skc, struct mptcp_addr_info *addr) 91 { 92 addr->family = skc->skc_family; 93 addr->port = htons(skc->skc_num); 94 if (addr->family == AF_INET) 95 addr->addr.s_addr = skc->skc_rcv_saddr; 96 #if IS_ENABLED(CONFIG_MPTCP_IPV6) 97 else if (addr->family == AF_INET6) 98 addr->addr6 = skc->skc_v6_rcv_saddr; 99 #endif 100 } 101 102 static void remote_address(const struct sock_common *skc, 103 struct mptcp_addr_info *addr) 104 { 105 addr->family = skc->skc_family; 106 addr->port = skc->skc_dport; 107 if (addr->family == AF_INET) 108 addr->addr.s_addr = skc->skc_daddr; 109 #if IS_ENABLED(CONFIG_MPTCP_IPV6) 110 else if (addr->family == AF_INET6) 111 addr->addr6 = skc->skc_v6_daddr; 112 #endif 113 } 114 115 static bool lookup_subflow_by_saddr(const struct list_head *list, 116 const struct mptcp_addr_info *saddr) 117 { 118 struct mptcp_subflow_context *subflow; 119 struct mptcp_addr_info cur; 120 struct sock_common *skc; 121 122 list_for_each_entry(subflow, list, node) { 123 skc = (struct sock_common *)mptcp_subflow_tcp_sock(subflow); 124 125 mptcp_local_address(skc, &cur); 126 if (mptcp_addresses_equal(&cur, saddr, saddr->port)) 127 return true; 128 } 129 130 return false; 131 } 132 133 static bool lookup_subflow_by_daddr(const struct list_head *list, 134 const struct mptcp_addr_info *daddr) 135 { 136 struct mptcp_subflow_context *subflow; 137 struct mptcp_addr_info cur; 138 139 list_for_each_entry(subflow, list, node) { 140 struct sock *ssk = mptcp_subflow_tcp_sock(subflow); 141 142 if (!((1 << inet_sk_state_load(ssk)) & 143 (TCPF_ESTABLISHED | TCPF_SYN_SENT | TCPF_SYN_RECV))) 144 continue; 145 146 remote_address((struct sock_common *)ssk, &cur); 147 if (mptcp_addresses_equal(&cur, daddr, daddr->port)) 148 return true; 149 } 150 151 return false; 152 } 153 154 static bool 155 select_local_address(const struct pm_nl_pernet *pernet, 156 const struct mptcp_sock *msk, 157 struct mptcp_pm_addr_entry *new_entry) 158 { 159 struct mptcp_pm_addr_entry *entry; 160 bool found = false; 161 162 msk_owned_by_me(msk); 163 164 rcu_read_lock(); 165 list_for_each_entry_rcu(entry, &pernet->local_addr_list, list) { 166 if (!(entry->flags & MPTCP_PM_ADDR_FLAG_SUBFLOW)) 167 continue; 168 169 if (!test_bit(entry->addr.id, msk->pm.id_avail_bitmap)) 170 continue; 171 172 *new_entry = *entry; 173 found = true; 174 break; 175 } 176 rcu_read_unlock(); 177 178 return found; 179 } 180 181 static bool 182 select_signal_address(struct pm_nl_pernet *pernet, const struct mptcp_sock *msk, 183 struct mptcp_pm_addr_entry *new_entry) 184 { 185 struct mptcp_pm_addr_entry *entry; 186 bool found = false; 187 188 rcu_read_lock(); 189 /* do not keep any additional per socket state, just signal 190 * the address list in order. 191 * Note: removal from the local address list during the msk life-cycle 192 * can lead to additional addresses not being announced. 193 */ 194 list_for_each_entry_rcu(entry, &pernet->local_addr_list, list) { 195 if (!test_bit(entry->addr.id, msk->pm.id_avail_bitmap)) 196 continue; 197 198 if (!(entry->flags & MPTCP_PM_ADDR_FLAG_SIGNAL)) 199 continue; 200 201 *new_entry = *entry; 202 found = true; 203 break; 204 } 205 rcu_read_unlock(); 206 207 return found; 208 } 209 210 unsigned int mptcp_pm_get_add_addr_signal_max(const struct mptcp_sock *msk) 211 { 212 const struct pm_nl_pernet *pernet = pm_nl_get_pernet_from_msk(msk); 213 214 return READ_ONCE(pernet->add_addr_signal_max); 215 } 216 EXPORT_SYMBOL_GPL(mptcp_pm_get_add_addr_signal_max); 217 218 unsigned int mptcp_pm_get_add_addr_accept_max(const struct mptcp_sock *msk) 219 { 220 struct pm_nl_pernet *pernet = pm_nl_get_pernet_from_msk(msk); 221 222 return READ_ONCE(pernet->add_addr_accept_max); 223 } 224 EXPORT_SYMBOL_GPL(mptcp_pm_get_add_addr_accept_max); 225 226 unsigned int mptcp_pm_get_subflows_max(const struct mptcp_sock *msk) 227 { 228 struct pm_nl_pernet *pernet = pm_nl_get_pernet_from_msk(msk); 229 230 return READ_ONCE(pernet->subflows_max); 231 } 232 EXPORT_SYMBOL_GPL(mptcp_pm_get_subflows_max); 233 234 unsigned int mptcp_pm_get_local_addr_max(const struct mptcp_sock *msk) 235 { 236 struct pm_nl_pernet *pernet = pm_nl_get_pernet_from_msk(msk); 237 238 return READ_ONCE(pernet->local_addr_max); 239 } 240 EXPORT_SYMBOL_GPL(mptcp_pm_get_local_addr_max); 241 242 bool mptcp_pm_nl_check_work_pending(struct mptcp_sock *msk) 243 { 244 struct pm_nl_pernet *pernet = pm_nl_get_pernet_from_msk(msk); 245 246 if (msk->pm.subflows == mptcp_pm_get_subflows_max(msk) || 247 (find_next_and_bit(pernet->id_bitmap, msk->pm.id_avail_bitmap, 248 MPTCP_PM_MAX_ADDR_ID + 1, 0) == MPTCP_PM_MAX_ADDR_ID + 1)) { 249 WRITE_ONCE(msk->pm.work_pending, false); 250 return false; 251 } 252 return true; 253 } 254 255 struct mptcp_pm_add_entry * 256 mptcp_lookup_anno_list_by_saddr(const struct mptcp_sock *msk, 257 const struct mptcp_addr_info *addr) 258 { 259 struct mptcp_pm_add_entry *entry; 260 261 lockdep_assert_held(&msk->pm.lock); 262 263 list_for_each_entry(entry, &msk->pm.anno_list, list) { 264 if (mptcp_addresses_equal(&entry->addr, addr, true)) 265 return entry; 266 } 267 268 return NULL; 269 } 270 271 bool mptcp_pm_sport_in_anno_list(struct mptcp_sock *msk, const struct sock *sk) 272 { 273 struct mptcp_pm_add_entry *entry; 274 struct mptcp_addr_info saddr; 275 bool ret = false; 276 277 mptcp_local_address((struct sock_common *)sk, &saddr); 278 279 spin_lock_bh(&msk->pm.lock); 280 list_for_each_entry(entry, &msk->pm.anno_list, list) { 281 if (mptcp_addresses_equal(&entry->addr, &saddr, true)) { 282 ret = true; 283 goto out; 284 } 285 } 286 287 out: 288 spin_unlock_bh(&msk->pm.lock); 289 return ret; 290 } 291 292 static void mptcp_pm_add_timer(struct timer_list *timer) 293 { 294 struct mptcp_pm_add_entry *entry = from_timer(entry, timer, add_timer); 295 struct mptcp_sock *msk = entry->sock; 296 struct sock *sk = (struct sock *)msk; 297 298 pr_debug("msk=%p", msk); 299 300 if (!msk) 301 return; 302 303 if (inet_sk_state_load(sk) == TCP_CLOSE) 304 return; 305 306 if (!entry->addr.id) 307 return; 308 309 if (mptcp_pm_should_add_signal_addr(msk)) { 310 sk_reset_timer(sk, timer, jiffies + TCP_RTO_MAX / 8); 311 goto out; 312 } 313 314 spin_lock_bh(&msk->pm.lock); 315 316 if (!mptcp_pm_should_add_signal_addr(msk)) { 317 pr_debug("retransmit ADD_ADDR id=%d", entry->addr.id); 318 mptcp_pm_announce_addr(msk, &entry->addr, false); 319 mptcp_pm_add_addr_send_ack(msk); 320 entry->retrans_times++; 321 } 322 323 if (entry->retrans_times < ADD_ADDR_RETRANS_MAX) 324 sk_reset_timer(sk, timer, 325 jiffies + mptcp_get_add_addr_timeout(sock_net(sk))); 326 327 spin_unlock_bh(&msk->pm.lock); 328 329 if (entry->retrans_times == ADD_ADDR_RETRANS_MAX) 330 mptcp_pm_subflow_established(msk); 331 332 out: 333 __sock_put(sk); 334 } 335 336 struct mptcp_pm_add_entry * 337 mptcp_pm_del_add_timer(struct mptcp_sock *msk, 338 const struct mptcp_addr_info *addr, bool check_id) 339 { 340 struct mptcp_pm_add_entry *entry; 341 struct sock *sk = (struct sock *)msk; 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 spin_unlock_bh(&msk->pm.lock); 348 349 if (entry && (!check_id || entry->addr.id == addr->id)) 350 sk_stop_timer_sync(sk, &entry->add_timer); 351 352 return entry; 353 } 354 355 bool mptcp_pm_alloc_anno_list(struct mptcp_sock *msk, 356 const struct mptcp_addr_info *addr) 357 { 358 struct mptcp_pm_add_entry *add_entry = NULL; 359 struct sock *sk = (struct sock *)msk; 360 struct net *net = sock_net(sk); 361 362 lockdep_assert_held(&msk->pm.lock); 363 364 add_entry = mptcp_lookup_anno_list_by_saddr(msk, addr); 365 366 if (add_entry) { 367 if (WARN_ON_ONCE(mptcp_pm_is_kernel(msk))) 368 return false; 369 370 sk_reset_timer(sk, &add_entry->add_timer, 371 jiffies + mptcp_get_add_addr_timeout(net)); 372 return true; 373 } 374 375 add_entry = kmalloc(sizeof(*add_entry), GFP_ATOMIC); 376 if (!add_entry) 377 return false; 378 379 list_add(&add_entry->list, &msk->pm.anno_list); 380 381 add_entry->addr = *addr; 382 add_entry->sock = msk; 383 add_entry->retrans_times = 0; 384 385 timer_setup(&add_entry->add_timer, mptcp_pm_add_timer, 0); 386 sk_reset_timer(sk, &add_entry->add_timer, 387 jiffies + mptcp_get_add_addr_timeout(net)); 388 389 return true; 390 } 391 392 void mptcp_pm_free_anno_list(struct mptcp_sock *msk) 393 { 394 struct mptcp_pm_add_entry *entry, *tmp; 395 struct sock *sk = (struct sock *)msk; 396 LIST_HEAD(free_list); 397 398 pr_debug("msk=%p", msk); 399 400 spin_lock_bh(&msk->pm.lock); 401 list_splice_init(&msk->pm.anno_list, &free_list); 402 spin_unlock_bh(&msk->pm.lock); 403 404 list_for_each_entry_safe(entry, tmp, &free_list, list) { 405 sk_stop_timer_sync(sk, &entry->add_timer); 406 kfree(entry); 407 } 408 } 409 410 /* Fill all the remote addresses into the array addrs[], 411 * and return the array size. 412 */ 413 static unsigned int fill_remote_addresses_vec(struct mptcp_sock *msk, 414 struct mptcp_addr_info *local, 415 bool fullmesh, 416 struct mptcp_addr_info *addrs) 417 { 418 bool deny_id0 = READ_ONCE(msk->pm.remote_deny_join_id0); 419 struct sock *sk = (struct sock *)msk, *ssk; 420 struct mptcp_subflow_context *subflow; 421 struct mptcp_addr_info remote = { 0 }; 422 unsigned int subflows_max; 423 int i = 0; 424 425 subflows_max = mptcp_pm_get_subflows_max(msk); 426 remote_address((struct sock_common *)sk, &remote); 427 428 /* Non-fullmesh endpoint, fill in the single entry 429 * corresponding to the primary MPC subflow remote address 430 */ 431 if (!fullmesh) { 432 if (deny_id0) 433 return 0; 434 435 if (!mptcp_pm_addr_families_match(sk, local, &remote)) 436 return 0; 437 438 msk->pm.subflows++; 439 addrs[i++] = remote; 440 } else { 441 DECLARE_BITMAP(unavail_id, MPTCP_PM_MAX_ADDR_ID + 1); 442 443 /* Forbid creation of new subflows matching existing 444 * ones, possibly already created by incoming ADD_ADDR 445 */ 446 bitmap_zero(unavail_id, MPTCP_PM_MAX_ADDR_ID + 1); 447 mptcp_for_each_subflow(msk, subflow) 448 if (READ_ONCE(subflow->local_id) == local->id) 449 __set_bit(subflow->remote_id, unavail_id); 450 451 mptcp_for_each_subflow(msk, subflow) { 452 ssk = mptcp_subflow_tcp_sock(subflow); 453 remote_address((struct sock_common *)ssk, &addrs[i]); 454 addrs[i].id = READ_ONCE(subflow->remote_id); 455 if (deny_id0 && !addrs[i].id) 456 continue; 457 458 if (test_bit(addrs[i].id, unavail_id)) 459 continue; 460 461 if (!mptcp_pm_addr_families_match(sk, local, &addrs[i])) 462 continue; 463 464 if (msk->pm.subflows < subflows_max) { 465 /* forbid creating multiple address towards 466 * this id 467 */ 468 __set_bit(addrs[i].id, unavail_id); 469 msk->pm.subflows++; 470 i++; 471 } 472 } 473 } 474 475 return i; 476 } 477 478 static void __mptcp_pm_send_ack(struct mptcp_sock *msk, struct mptcp_subflow_context *subflow, 479 bool prio, bool backup) 480 { 481 struct sock *ssk = mptcp_subflow_tcp_sock(subflow); 482 bool slow; 483 484 pr_debug("send ack for %s", 485 prio ? "mp_prio" : (mptcp_pm_should_add_signal(msk) ? "add_addr" : "rm_addr")); 486 487 slow = lock_sock_fast(ssk); 488 if (prio) { 489 subflow->send_mp_prio = 1; 490 subflow->request_bkup = backup; 491 } 492 493 __mptcp_subflow_send_ack(ssk); 494 unlock_sock_fast(ssk, slow); 495 } 496 497 static void mptcp_pm_send_ack(struct mptcp_sock *msk, struct mptcp_subflow_context *subflow, 498 bool prio, bool backup) 499 { 500 spin_unlock_bh(&msk->pm.lock); 501 __mptcp_pm_send_ack(msk, subflow, prio, backup); 502 spin_lock_bh(&msk->pm.lock); 503 } 504 505 static struct mptcp_pm_addr_entry * 506 __lookup_addr_by_id(struct pm_nl_pernet *pernet, unsigned int id) 507 { 508 struct mptcp_pm_addr_entry *entry; 509 510 list_for_each_entry(entry, &pernet->local_addr_list, list) { 511 if (entry->addr.id == id) 512 return entry; 513 } 514 return NULL; 515 } 516 517 static struct mptcp_pm_addr_entry * 518 __lookup_addr(struct pm_nl_pernet *pernet, const struct mptcp_addr_info *info, 519 bool lookup_by_id) 520 { 521 struct mptcp_pm_addr_entry *entry; 522 523 list_for_each_entry(entry, &pernet->local_addr_list, list) { 524 if ((!lookup_by_id && 525 mptcp_addresses_equal(&entry->addr, info, entry->addr.port)) || 526 (lookup_by_id && entry->addr.id == info->id)) 527 return entry; 528 } 529 return NULL; 530 } 531 532 static void mptcp_pm_create_subflow_or_signal_addr(struct mptcp_sock *msk) 533 { 534 struct sock *sk = (struct sock *)msk; 535 struct mptcp_pm_addr_entry local; 536 unsigned int add_addr_signal_max; 537 bool signal_and_subflow = false; 538 unsigned int local_addr_max; 539 struct pm_nl_pernet *pernet; 540 unsigned int subflows_max; 541 542 pernet = pm_nl_get_pernet(sock_net(sk)); 543 544 add_addr_signal_max = mptcp_pm_get_add_addr_signal_max(msk); 545 local_addr_max = mptcp_pm_get_local_addr_max(msk); 546 subflows_max = mptcp_pm_get_subflows_max(msk); 547 548 /* do lazy endpoint usage accounting for the MPC subflows */ 549 if (unlikely(!(msk->pm.status & BIT(MPTCP_PM_MPC_ENDPOINT_ACCOUNTED))) && msk->first) { 550 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(msk->first); 551 struct mptcp_pm_addr_entry *entry; 552 struct mptcp_addr_info mpc_addr; 553 bool backup = false; 554 555 mptcp_local_address((struct sock_common *)msk->first, &mpc_addr); 556 rcu_read_lock(); 557 entry = __lookup_addr(pernet, &mpc_addr, false); 558 if (entry) { 559 __clear_bit(entry->addr.id, msk->pm.id_avail_bitmap); 560 msk->mpc_endpoint_id = entry->addr.id; 561 backup = !!(entry->flags & MPTCP_PM_ADDR_FLAG_BACKUP); 562 } 563 rcu_read_unlock(); 564 565 if (backup) 566 mptcp_pm_send_ack(msk, subflow, true, backup); 567 568 msk->pm.status |= BIT(MPTCP_PM_MPC_ENDPOINT_ACCOUNTED); 569 } 570 571 pr_debug("local %d:%d signal %d:%d subflows %d:%d\n", 572 msk->pm.local_addr_used, local_addr_max, 573 msk->pm.add_addr_signaled, add_addr_signal_max, 574 msk->pm.subflows, subflows_max); 575 576 /* check first for announce */ 577 if (msk->pm.add_addr_signaled < add_addr_signal_max) { 578 /* due to racing events on both ends we can reach here while 579 * previous add address is still running: if we invoke now 580 * mptcp_pm_announce_addr(), that will fail and the 581 * corresponding id will be marked as used. 582 * Instead let the PM machinery reschedule us when the 583 * current address announce will be completed. 584 */ 585 if (msk->pm.addr_signal & BIT(MPTCP_ADD_ADDR_SIGNAL)) 586 return; 587 588 if (!select_signal_address(pernet, msk, &local)) 589 goto subflow; 590 591 /* If the alloc fails, we are on memory pressure, not worth 592 * continuing, and trying to create subflows. 593 */ 594 if (!mptcp_pm_alloc_anno_list(msk, &local.addr)) 595 return; 596 597 __clear_bit(local.addr.id, msk->pm.id_avail_bitmap); 598 msk->pm.add_addr_signaled++; 599 600 /* Special case for ID0: set the correct ID */ 601 if (local.addr.id == msk->mpc_endpoint_id) 602 local.addr.id = 0; 603 604 mptcp_pm_announce_addr(msk, &local.addr, false); 605 mptcp_pm_nl_addr_send_ack(msk); 606 607 if (local.flags & MPTCP_PM_ADDR_FLAG_SUBFLOW) 608 signal_and_subflow = true; 609 } 610 611 subflow: 612 /* check if should create a new subflow */ 613 while (msk->pm.local_addr_used < local_addr_max && 614 msk->pm.subflows < subflows_max) { 615 struct mptcp_addr_info addrs[MPTCP_PM_ADDR_MAX]; 616 bool fullmesh; 617 int i, nr; 618 619 if (signal_and_subflow) 620 signal_and_subflow = false; 621 else if (!select_local_address(pernet, msk, &local)) 622 break; 623 624 fullmesh = !!(local.flags & MPTCP_PM_ADDR_FLAG_FULLMESH); 625 626 __clear_bit(local.addr.id, msk->pm.id_avail_bitmap); 627 628 /* Special case for ID0: set the correct ID */ 629 if (local.addr.id == msk->mpc_endpoint_id) 630 local.addr.id = 0; 631 else /* local_addr_used is not decr for ID 0 */ 632 msk->pm.local_addr_used++; 633 634 nr = fill_remote_addresses_vec(msk, &local.addr, fullmesh, addrs); 635 if (nr == 0) 636 continue; 637 638 spin_unlock_bh(&msk->pm.lock); 639 for (i = 0; i < nr; i++) 640 __mptcp_subflow_connect(sk, &local.addr, &addrs[i]); 641 spin_lock_bh(&msk->pm.lock); 642 } 643 mptcp_pm_nl_check_work_pending(msk); 644 } 645 646 static void mptcp_pm_nl_fully_established(struct mptcp_sock *msk) 647 { 648 mptcp_pm_create_subflow_or_signal_addr(msk); 649 } 650 651 static void mptcp_pm_nl_subflow_established(struct mptcp_sock *msk) 652 { 653 mptcp_pm_create_subflow_or_signal_addr(msk); 654 } 655 656 /* Fill all the local addresses into the array addrs[], 657 * and return the array size. 658 */ 659 static unsigned int fill_local_addresses_vec(struct mptcp_sock *msk, 660 struct mptcp_addr_info *remote, 661 struct mptcp_addr_info *addrs) 662 { 663 struct sock *sk = (struct sock *)msk; 664 struct mptcp_pm_addr_entry *entry; 665 struct mptcp_addr_info mpc_addr; 666 struct pm_nl_pernet *pernet; 667 unsigned int subflows_max; 668 int i = 0; 669 670 pernet = pm_nl_get_pernet_from_msk(msk); 671 subflows_max = mptcp_pm_get_subflows_max(msk); 672 673 mptcp_local_address((struct sock_common *)msk, &mpc_addr); 674 675 rcu_read_lock(); 676 list_for_each_entry_rcu(entry, &pernet->local_addr_list, list) { 677 if (!(entry->flags & MPTCP_PM_ADDR_FLAG_FULLMESH)) 678 continue; 679 680 if (!mptcp_pm_addr_families_match(sk, &entry->addr, remote)) 681 continue; 682 683 if (msk->pm.subflows < subflows_max) { 684 msk->pm.subflows++; 685 addrs[i] = entry->addr; 686 687 /* Special case for ID0: set the correct ID */ 688 if (mptcp_addresses_equal(&entry->addr, &mpc_addr, entry->addr.port)) 689 addrs[i].id = 0; 690 691 i++; 692 } 693 } 694 rcu_read_unlock(); 695 696 /* If the array is empty, fill in the single 697 * 'IPADDRANY' local address 698 */ 699 if (!i) { 700 struct mptcp_addr_info local; 701 702 memset(&local, 0, sizeof(local)); 703 local.family = 704 #if IS_ENABLED(CONFIG_MPTCP_IPV6) 705 remote->family == AF_INET6 && 706 ipv6_addr_v4mapped(&remote->addr6) ? AF_INET : 707 #endif 708 remote->family; 709 710 if (!mptcp_pm_addr_families_match(sk, &local, remote)) 711 return 0; 712 713 msk->pm.subflows++; 714 addrs[i++] = local; 715 } 716 717 return i; 718 } 719 720 static void mptcp_pm_nl_add_addr_received(struct mptcp_sock *msk) 721 { 722 struct mptcp_addr_info addrs[MPTCP_PM_ADDR_MAX]; 723 struct sock *sk = (struct sock *)msk; 724 unsigned int add_addr_accept_max; 725 struct mptcp_addr_info remote; 726 unsigned int subflows_max; 727 bool sf_created = false; 728 int i, nr; 729 730 add_addr_accept_max = mptcp_pm_get_add_addr_accept_max(msk); 731 subflows_max = mptcp_pm_get_subflows_max(msk); 732 733 pr_debug("accepted %d:%d remote family %d", 734 msk->pm.add_addr_accepted, add_addr_accept_max, 735 msk->pm.remote.family); 736 737 remote = msk->pm.remote; 738 mptcp_pm_announce_addr(msk, &remote, true); 739 mptcp_pm_nl_addr_send_ack(msk); 740 741 if (lookup_subflow_by_daddr(&msk->conn_list, &remote)) 742 return; 743 744 /* pick id 0 port, if none is provided the remote address */ 745 if (!remote.port) 746 remote.port = sk->sk_dport; 747 748 /* connect to the specified remote address, using whatever 749 * local address the routing configuration will pick. 750 */ 751 nr = fill_local_addresses_vec(msk, &remote, addrs); 752 if (nr == 0) 753 return; 754 755 spin_unlock_bh(&msk->pm.lock); 756 for (i = 0; i < nr; i++) 757 if (__mptcp_subflow_connect(sk, &addrs[i], &remote) == 0) 758 sf_created = true; 759 spin_lock_bh(&msk->pm.lock); 760 761 if (sf_created) { 762 /* add_addr_accepted is not decr for ID 0 */ 763 if (remote.id) 764 msk->pm.add_addr_accepted++; 765 if (msk->pm.add_addr_accepted >= add_addr_accept_max || 766 msk->pm.subflows >= subflows_max) 767 WRITE_ONCE(msk->pm.accept_addr, false); 768 } 769 } 770 771 bool mptcp_pm_nl_is_init_remote_addr(struct mptcp_sock *msk, 772 const struct mptcp_addr_info *remote) 773 { 774 struct mptcp_addr_info mpc_remote; 775 776 remote_address((struct sock_common *)msk, &mpc_remote); 777 return mptcp_addresses_equal(&mpc_remote, remote, remote->port); 778 } 779 780 void mptcp_pm_nl_addr_send_ack(struct mptcp_sock *msk) 781 { 782 struct mptcp_subflow_context *subflow; 783 784 msk_owned_by_me(msk); 785 lockdep_assert_held(&msk->pm.lock); 786 787 if (!mptcp_pm_should_add_signal(msk) && 788 !mptcp_pm_should_rm_signal(msk)) 789 return; 790 791 mptcp_for_each_subflow(msk, subflow) { 792 if (__mptcp_subflow_active(subflow)) { 793 mptcp_pm_send_ack(msk, subflow, false, false); 794 break; 795 } 796 } 797 } 798 799 int mptcp_pm_nl_mp_prio_send_ack(struct mptcp_sock *msk, 800 struct mptcp_addr_info *addr, 801 struct mptcp_addr_info *rem, 802 u8 bkup) 803 { 804 struct mptcp_subflow_context *subflow; 805 806 pr_debug("bkup=%d", bkup); 807 808 mptcp_for_each_subflow(msk, subflow) { 809 struct sock *ssk = mptcp_subflow_tcp_sock(subflow); 810 struct mptcp_addr_info local, remote; 811 812 mptcp_local_address((struct sock_common *)ssk, &local); 813 if (!mptcp_addresses_equal(&local, addr, addr->port)) 814 continue; 815 816 if (rem && rem->family != AF_UNSPEC) { 817 remote_address((struct sock_common *)ssk, &remote); 818 if (!mptcp_addresses_equal(&remote, rem, rem->port)) 819 continue; 820 } 821 822 __mptcp_pm_send_ack(msk, subflow, true, bkup); 823 return 0; 824 } 825 826 return -EINVAL; 827 } 828 829 static bool mptcp_local_id_match(const struct mptcp_sock *msk, u8 local_id, u8 id) 830 { 831 return local_id == id || (!local_id && msk->mpc_endpoint_id == id); 832 } 833 834 static void mptcp_pm_nl_rm_addr_or_subflow(struct mptcp_sock *msk, 835 const struct mptcp_rm_list *rm_list, 836 enum linux_mptcp_mib_field rm_type) 837 { 838 struct mptcp_subflow_context *subflow, *tmp; 839 struct sock *sk = (struct sock *)msk; 840 u8 i; 841 842 pr_debug("%s rm_list_nr %d", 843 rm_type == MPTCP_MIB_RMADDR ? "address" : "subflow", rm_list->nr); 844 845 msk_owned_by_me(msk); 846 847 if (sk->sk_state == TCP_LISTEN) 848 return; 849 850 if (!rm_list->nr) 851 return; 852 853 if (list_empty(&msk->conn_list)) 854 return; 855 856 for (i = 0; i < rm_list->nr; i++) { 857 u8 rm_id = rm_list->ids[i]; 858 bool removed = false; 859 860 mptcp_for_each_subflow_safe(msk, subflow, tmp) { 861 struct sock *ssk = mptcp_subflow_tcp_sock(subflow); 862 u8 remote_id = READ_ONCE(subflow->remote_id); 863 int how = RCV_SHUTDOWN | SEND_SHUTDOWN; 864 u8 id = subflow_get_local_id(subflow); 865 866 if (inet_sk_state_load(ssk) == TCP_CLOSE) 867 continue; 868 if (rm_type == MPTCP_MIB_RMADDR && remote_id != rm_id) 869 continue; 870 if (rm_type == MPTCP_MIB_RMSUBFLOW && !mptcp_local_id_match(msk, id, rm_id)) 871 continue; 872 873 pr_debug(" -> %s rm_list_ids[%d]=%u local_id=%u remote_id=%u mpc_id=%u", 874 rm_type == MPTCP_MIB_RMADDR ? "address" : "subflow", 875 i, rm_id, id, remote_id, msk->mpc_endpoint_id); 876 spin_unlock_bh(&msk->pm.lock); 877 mptcp_subflow_shutdown(sk, ssk, how); 878 879 /* the following takes care of updating the subflows counter */ 880 mptcp_close_ssk(sk, ssk, subflow); 881 spin_lock_bh(&msk->pm.lock); 882 883 removed |= subflow->request_join; 884 if (rm_type == MPTCP_MIB_RMSUBFLOW) 885 __MPTCP_INC_STATS(sock_net(sk), rm_type); 886 } 887 888 if (rm_type == MPTCP_MIB_RMADDR) 889 __MPTCP_INC_STATS(sock_net(sk), rm_type); 890 891 if (!removed) 892 continue; 893 894 if (!mptcp_pm_is_kernel(msk)) 895 continue; 896 897 if (rm_type == MPTCP_MIB_RMADDR && rm_id && 898 !WARN_ON_ONCE(msk->pm.add_addr_accepted == 0)) { 899 /* Note: if the subflow has been closed before, this 900 * add_addr_accepted counter will not be decremented. 901 */ 902 if (--msk->pm.add_addr_accepted < mptcp_pm_get_add_addr_accept_max(msk)) 903 WRITE_ONCE(msk->pm.accept_addr, true); 904 } 905 } 906 } 907 908 static void mptcp_pm_nl_rm_addr_received(struct mptcp_sock *msk) 909 { 910 mptcp_pm_nl_rm_addr_or_subflow(msk, &msk->pm.rm_list_rx, MPTCP_MIB_RMADDR); 911 } 912 913 static void mptcp_pm_nl_rm_subflow_received(struct mptcp_sock *msk, 914 const struct mptcp_rm_list *rm_list) 915 { 916 mptcp_pm_nl_rm_addr_or_subflow(msk, rm_list, MPTCP_MIB_RMSUBFLOW); 917 } 918 919 void mptcp_pm_nl_work(struct mptcp_sock *msk) 920 { 921 struct mptcp_pm_data *pm = &msk->pm; 922 923 msk_owned_by_me(msk); 924 925 if (!(pm->status & MPTCP_PM_WORK_MASK)) 926 return; 927 928 spin_lock_bh(&msk->pm.lock); 929 930 pr_debug("msk=%p status=%x", msk, pm->status); 931 if (pm->status & BIT(MPTCP_PM_ADD_ADDR_RECEIVED)) { 932 pm->status &= ~BIT(MPTCP_PM_ADD_ADDR_RECEIVED); 933 mptcp_pm_nl_add_addr_received(msk); 934 } 935 if (pm->status & BIT(MPTCP_PM_ADD_ADDR_SEND_ACK)) { 936 pm->status &= ~BIT(MPTCP_PM_ADD_ADDR_SEND_ACK); 937 mptcp_pm_nl_addr_send_ack(msk); 938 } 939 if (pm->status & BIT(MPTCP_PM_RM_ADDR_RECEIVED)) { 940 pm->status &= ~BIT(MPTCP_PM_RM_ADDR_RECEIVED); 941 mptcp_pm_nl_rm_addr_received(msk); 942 } 943 if (pm->status & BIT(MPTCP_PM_ESTABLISHED)) { 944 pm->status &= ~BIT(MPTCP_PM_ESTABLISHED); 945 mptcp_pm_nl_fully_established(msk); 946 } 947 if (pm->status & BIT(MPTCP_PM_SUBFLOW_ESTABLISHED)) { 948 pm->status &= ~BIT(MPTCP_PM_SUBFLOW_ESTABLISHED); 949 mptcp_pm_nl_subflow_established(msk); 950 } 951 952 spin_unlock_bh(&msk->pm.lock); 953 } 954 955 static bool address_use_port(struct mptcp_pm_addr_entry *entry) 956 { 957 return (entry->flags & 958 (MPTCP_PM_ADDR_FLAG_SIGNAL | MPTCP_PM_ADDR_FLAG_SUBFLOW)) == 959 MPTCP_PM_ADDR_FLAG_SIGNAL; 960 } 961 962 /* caller must ensure the RCU grace period is already elapsed */ 963 static void __mptcp_pm_release_addr_entry(struct mptcp_pm_addr_entry *entry) 964 { 965 if (entry->lsk) 966 sock_release(entry->lsk); 967 kfree(entry); 968 } 969 970 static int mptcp_pm_nl_append_new_local_addr(struct pm_nl_pernet *pernet, 971 struct mptcp_pm_addr_entry *entry, 972 bool needs_id) 973 { 974 struct mptcp_pm_addr_entry *cur, *del_entry = NULL; 975 unsigned int addr_max; 976 int ret = -EINVAL; 977 978 spin_lock_bh(&pernet->lock); 979 /* to keep the code simple, don't do IDR-like allocation for address ID, 980 * just bail when we exceed limits 981 */ 982 if (pernet->next_id == MPTCP_PM_MAX_ADDR_ID) 983 pernet->next_id = 1; 984 if (pernet->addrs >= MPTCP_PM_ADDR_MAX) { 985 ret = -ERANGE; 986 goto out; 987 } 988 if (test_bit(entry->addr.id, pernet->id_bitmap)) { 989 ret = -EBUSY; 990 goto out; 991 } 992 993 /* do not insert duplicate address, differentiate on port only 994 * singled addresses 995 */ 996 if (!address_use_port(entry)) 997 entry->addr.port = 0; 998 list_for_each_entry(cur, &pernet->local_addr_list, list) { 999 if (mptcp_addresses_equal(&cur->addr, &entry->addr, 1000 cur->addr.port || entry->addr.port)) { 1001 /* allow replacing the exiting endpoint only if such 1002 * endpoint is an implicit one and the user-space 1003 * did not provide an endpoint id 1004 */ 1005 if (!(cur->flags & MPTCP_PM_ADDR_FLAG_IMPLICIT)) { 1006 ret = -EEXIST; 1007 goto out; 1008 } 1009 if (entry->addr.id) 1010 goto out; 1011 1012 pernet->addrs--; 1013 entry->addr.id = cur->addr.id; 1014 list_del_rcu(&cur->list); 1015 del_entry = cur; 1016 break; 1017 } 1018 } 1019 1020 if (!entry->addr.id && needs_id) { 1021 find_next: 1022 entry->addr.id = find_next_zero_bit(pernet->id_bitmap, 1023 MPTCP_PM_MAX_ADDR_ID + 1, 1024 pernet->next_id); 1025 if (!entry->addr.id && pernet->next_id != 1) { 1026 pernet->next_id = 1; 1027 goto find_next; 1028 } 1029 } 1030 1031 if (!entry->addr.id && needs_id) 1032 goto out; 1033 1034 __set_bit(entry->addr.id, pernet->id_bitmap); 1035 if (entry->addr.id > pernet->next_id) 1036 pernet->next_id = entry->addr.id; 1037 1038 if (entry->flags & MPTCP_PM_ADDR_FLAG_SIGNAL) { 1039 addr_max = pernet->add_addr_signal_max; 1040 WRITE_ONCE(pernet->add_addr_signal_max, addr_max + 1); 1041 } 1042 if (entry->flags & MPTCP_PM_ADDR_FLAG_SUBFLOW) { 1043 addr_max = pernet->local_addr_max; 1044 WRITE_ONCE(pernet->local_addr_max, addr_max + 1); 1045 } 1046 1047 pernet->addrs++; 1048 if (!entry->addr.port) 1049 list_add_tail_rcu(&entry->list, &pernet->local_addr_list); 1050 else 1051 list_add_rcu(&entry->list, &pernet->local_addr_list); 1052 ret = entry->addr.id; 1053 1054 out: 1055 spin_unlock_bh(&pernet->lock); 1056 1057 /* just replaced an existing entry, free it */ 1058 if (del_entry) { 1059 synchronize_rcu(); 1060 __mptcp_pm_release_addr_entry(del_entry); 1061 } 1062 return ret; 1063 } 1064 1065 static struct lock_class_key mptcp_slock_keys[2]; 1066 static struct lock_class_key mptcp_keys[2]; 1067 1068 static int mptcp_pm_nl_create_listen_socket(struct sock *sk, 1069 struct mptcp_pm_addr_entry *entry) 1070 { 1071 bool is_ipv6 = sk->sk_family == AF_INET6; 1072 int addrlen = sizeof(struct sockaddr_in); 1073 struct sockaddr_storage addr; 1074 struct sock *newsk, *ssk; 1075 int backlog = 1024; 1076 int err; 1077 1078 err = sock_create_kern(sock_net(sk), entry->addr.family, 1079 SOCK_STREAM, IPPROTO_MPTCP, &entry->lsk); 1080 if (err) 1081 return err; 1082 1083 newsk = entry->lsk->sk; 1084 if (!newsk) 1085 return -EINVAL; 1086 1087 /* The subflow socket lock is acquired in a nested to the msk one 1088 * in several places, even by the TCP stack, and this msk is a kernel 1089 * socket: lockdep complains. Instead of propagating the _nested 1090 * modifiers in several places, re-init the lock class for the msk 1091 * socket to an mptcp specific one. 1092 */ 1093 sock_lock_init_class_and_name(newsk, 1094 is_ipv6 ? "mlock-AF_INET6" : "mlock-AF_INET", 1095 &mptcp_slock_keys[is_ipv6], 1096 is_ipv6 ? "msk_lock-AF_INET6" : "msk_lock-AF_INET", 1097 &mptcp_keys[is_ipv6]); 1098 1099 lock_sock(newsk); 1100 ssk = __mptcp_nmpc_sk(mptcp_sk(newsk)); 1101 release_sock(newsk); 1102 if (IS_ERR(ssk)) 1103 return PTR_ERR(ssk); 1104 1105 mptcp_info2sockaddr(&entry->addr, &addr, entry->addr.family); 1106 #if IS_ENABLED(CONFIG_MPTCP_IPV6) 1107 if (entry->addr.family == AF_INET6) 1108 addrlen = sizeof(struct sockaddr_in6); 1109 #endif 1110 if (ssk->sk_family == AF_INET) 1111 err = inet_bind_sk(ssk, (struct sockaddr *)&addr, addrlen); 1112 #if IS_ENABLED(CONFIG_MPTCP_IPV6) 1113 else if (ssk->sk_family == AF_INET6) 1114 err = inet6_bind_sk(ssk, (struct sockaddr *)&addr, addrlen); 1115 #endif 1116 if (err) 1117 return err; 1118 1119 /* We don't use mptcp_set_state() here because it needs to be called 1120 * under the msk socket lock. For the moment, that will not bring 1121 * anything more than only calling inet_sk_state_store(), because the 1122 * old status is known (TCP_CLOSE). 1123 */ 1124 inet_sk_state_store(newsk, TCP_LISTEN); 1125 lock_sock(ssk); 1126 err = __inet_listen_sk(ssk, backlog); 1127 if (!err) 1128 mptcp_event_pm_listener(ssk, MPTCP_EVENT_LISTENER_CREATED); 1129 release_sock(ssk); 1130 return err; 1131 } 1132 1133 int mptcp_pm_nl_get_local_id(struct mptcp_sock *msk, struct mptcp_addr_info *skc) 1134 { 1135 struct mptcp_pm_addr_entry *entry; 1136 struct pm_nl_pernet *pernet; 1137 int ret = -1; 1138 1139 pernet = pm_nl_get_pernet_from_msk(msk); 1140 1141 rcu_read_lock(); 1142 list_for_each_entry_rcu(entry, &pernet->local_addr_list, list) { 1143 if (mptcp_addresses_equal(&entry->addr, skc, entry->addr.port)) { 1144 ret = entry->addr.id; 1145 break; 1146 } 1147 } 1148 rcu_read_unlock(); 1149 if (ret >= 0) 1150 return ret; 1151 1152 /* address not found, add to local list */ 1153 entry = kmalloc(sizeof(*entry), GFP_ATOMIC); 1154 if (!entry) 1155 return -ENOMEM; 1156 1157 entry->addr = *skc; 1158 entry->addr.id = 0; 1159 entry->addr.port = 0; 1160 entry->ifindex = 0; 1161 entry->flags = MPTCP_PM_ADDR_FLAG_IMPLICIT; 1162 entry->lsk = NULL; 1163 ret = mptcp_pm_nl_append_new_local_addr(pernet, entry, true); 1164 if (ret < 0) 1165 kfree(entry); 1166 1167 return ret; 1168 } 1169 1170 bool mptcp_pm_nl_is_backup(struct mptcp_sock *msk, struct mptcp_addr_info *skc) 1171 { 1172 struct pm_nl_pernet *pernet = pm_nl_get_pernet_from_msk(msk); 1173 struct mptcp_pm_addr_entry *entry; 1174 bool backup = false; 1175 1176 rcu_read_lock(); 1177 list_for_each_entry_rcu(entry, &pernet->local_addr_list, list) { 1178 if (mptcp_addresses_equal(&entry->addr, skc, entry->addr.port)) { 1179 backup = !!(entry->flags & MPTCP_PM_ADDR_FLAG_BACKUP); 1180 break; 1181 } 1182 } 1183 rcu_read_unlock(); 1184 1185 return backup; 1186 } 1187 1188 #define MPTCP_PM_CMD_GRP_OFFSET 0 1189 #define MPTCP_PM_EV_GRP_OFFSET 1 1190 1191 static const struct genl_multicast_group mptcp_pm_mcgrps[] = { 1192 [MPTCP_PM_CMD_GRP_OFFSET] = { .name = MPTCP_PM_CMD_GRP_NAME, }, 1193 [MPTCP_PM_EV_GRP_OFFSET] = { .name = MPTCP_PM_EV_GRP_NAME, 1194 .flags = GENL_UNS_ADMIN_PERM, 1195 }, 1196 }; 1197 1198 static const struct nla_policy 1199 mptcp_pm_addr_policy[MPTCP_PM_ADDR_ATTR_MAX + 1] = { 1200 [MPTCP_PM_ADDR_ATTR_FAMILY] = { .type = NLA_U16, }, 1201 [MPTCP_PM_ADDR_ATTR_ID] = { .type = NLA_U8, }, 1202 [MPTCP_PM_ADDR_ATTR_ADDR4] = { .type = NLA_U32, }, 1203 [MPTCP_PM_ADDR_ATTR_ADDR6] = 1204 NLA_POLICY_EXACT_LEN(sizeof(struct in6_addr)), 1205 [MPTCP_PM_ADDR_ATTR_PORT] = { .type = NLA_U16 }, 1206 [MPTCP_PM_ADDR_ATTR_FLAGS] = { .type = NLA_U32 }, 1207 [MPTCP_PM_ADDR_ATTR_IF_IDX] = { .type = NLA_S32 }, 1208 }; 1209 1210 static const struct nla_policy mptcp_pm_policy[MPTCP_PM_ATTR_MAX + 1] = { 1211 [MPTCP_PM_ATTR_ADDR] = 1212 NLA_POLICY_NESTED(mptcp_pm_addr_policy), 1213 [MPTCP_PM_ATTR_RCV_ADD_ADDRS] = { .type = NLA_U32, }, 1214 [MPTCP_PM_ATTR_SUBFLOWS] = { .type = NLA_U32, }, 1215 [MPTCP_PM_ATTR_TOKEN] = { .type = NLA_U32, }, 1216 [MPTCP_PM_ATTR_LOC_ID] = { .type = NLA_U8, }, 1217 [MPTCP_PM_ATTR_ADDR_REMOTE] = 1218 NLA_POLICY_NESTED(mptcp_pm_addr_policy), 1219 }; 1220 1221 void mptcp_pm_nl_subflow_chk_stale(const struct mptcp_sock *msk, struct sock *ssk) 1222 { 1223 struct mptcp_subflow_context *iter, *subflow = mptcp_subflow_ctx(ssk); 1224 struct sock *sk = (struct sock *)msk; 1225 unsigned int active_max_loss_cnt; 1226 struct net *net = sock_net(sk); 1227 unsigned int stale_loss_cnt; 1228 bool slow; 1229 1230 stale_loss_cnt = mptcp_stale_loss_cnt(net); 1231 if (subflow->stale || !stale_loss_cnt || subflow->stale_count <= stale_loss_cnt) 1232 return; 1233 1234 /* look for another available subflow not in loss state */ 1235 active_max_loss_cnt = max_t(int, stale_loss_cnt - 1, 1); 1236 mptcp_for_each_subflow(msk, iter) { 1237 if (iter != subflow && mptcp_subflow_active(iter) && 1238 iter->stale_count < active_max_loss_cnt) { 1239 /* we have some alternatives, try to mark this subflow as idle ...*/ 1240 slow = lock_sock_fast(ssk); 1241 if (!tcp_rtx_and_write_queues_empty(ssk)) { 1242 subflow->stale = 1; 1243 __mptcp_retransmit_pending_data(sk); 1244 MPTCP_INC_STATS(net, MPTCP_MIB_SUBFLOWSTALE); 1245 } 1246 unlock_sock_fast(ssk, slow); 1247 1248 /* always try to push the pending data regardless of re-injections: 1249 * we can possibly use backup subflows now, and subflow selection 1250 * is cheap under the msk socket lock 1251 */ 1252 __mptcp_push_pending(sk, 0); 1253 return; 1254 } 1255 } 1256 } 1257 1258 static int mptcp_pm_family_to_addr(int family) 1259 { 1260 #if IS_ENABLED(CONFIG_MPTCP_IPV6) 1261 if (family == AF_INET6) 1262 return MPTCP_PM_ADDR_ATTR_ADDR6; 1263 #endif 1264 return MPTCP_PM_ADDR_ATTR_ADDR4; 1265 } 1266 1267 static int mptcp_pm_parse_pm_addr_attr(struct nlattr *tb[], 1268 const struct nlattr *attr, 1269 struct genl_info *info, 1270 struct mptcp_addr_info *addr, 1271 bool require_family) 1272 { 1273 int err, addr_addr; 1274 1275 if (!attr) { 1276 GENL_SET_ERR_MSG(info, "missing address info"); 1277 return -EINVAL; 1278 } 1279 1280 /* no validation needed - was already done via nested policy */ 1281 err = nla_parse_nested_deprecated(tb, MPTCP_PM_ADDR_ATTR_MAX, attr, 1282 mptcp_pm_addr_policy, info->extack); 1283 if (err) 1284 return err; 1285 1286 if (tb[MPTCP_PM_ADDR_ATTR_ID]) 1287 addr->id = nla_get_u8(tb[MPTCP_PM_ADDR_ATTR_ID]); 1288 1289 if (!tb[MPTCP_PM_ADDR_ATTR_FAMILY]) { 1290 if (!require_family) 1291 return 0; 1292 1293 NL_SET_ERR_MSG_ATTR(info->extack, attr, 1294 "missing family"); 1295 return -EINVAL; 1296 } 1297 1298 addr->family = nla_get_u16(tb[MPTCP_PM_ADDR_ATTR_FAMILY]); 1299 if (addr->family != AF_INET 1300 #if IS_ENABLED(CONFIG_MPTCP_IPV6) 1301 && addr->family != AF_INET6 1302 #endif 1303 ) { 1304 NL_SET_ERR_MSG_ATTR(info->extack, attr, 1305 "unknown address family"); 1306 return -EINVAL; 1307 } 1308 addr_addr = mptcp_pm_family_to_addr(addr->family); 1309 if (!tb[addr_addr]) { 1310 NL_SET_ERR_MSG_ATTR(info->extack, attr, 1311 "missing address data"); 1312 return -EINVAL; 1313 } 1314 1315 #if IS_ENABLED(CONFIG_MPTCP_IPV6) 1316 if (addr->family == AF_INET6) 1317 addr->addr6 = nla_get_in6_addr(tb[addr_addr]); 1318 else 1319 #endif 1320 addr->addr.s_addr = nla_get_in_addr(tb[addr_addr]); 1321 1322 if (tb[MPTCP_PM_ADDR_ATTR_PORT]) 1323 addr->port = htons(nla_get_u16(tb[MPTCP_PM_ADDR_ATTR_PORT])); 1324 1325 return 0; 1326 } 1327 1328 int mptcp_pm_parse_addr(struct nlattr *attr, struct genl_info *info, 1329 struct mptcp_addr_info *addr) 1330 { 1331 struct nlattr *tb[MPTCP_PM_ADDR_ATTR_MAX + 1]; 1332 1333 memset(addr, 0, sizeof(*addr)); 1334 1335 return mptcp_pm_parse_pm_addr_attr(tb, attr, info, addr, true); 1336 } 1337 1338 int mptcp_pm_parse_entry(struct nlattr *attr, struct genl_info *info, 1339 bool require_family, 1340 struct mptcp_pm_addr_entry *entry) 1341 { 1342 struct nlattr *tb[MPTCP_PM_ADDR_ATTR_MAX + 1]; 1343 int err; 1344 1345 memset(entry, 0, sizeof(*entry)); 1346 1347 err = mptcp_pm_parse_pm_addr_attr(tb, attr, info, &entry->addr, require_family); 1348 if (err) 1349 return err; 1350 1351 if (tb[MPTCP_PM_ADDR_ATTR_IF_IDX]) { 1352 u32 val = nla_get_s32(tb[MPTCP_PM_ADDR_ATTR_IF_IDX]); 1353 1354 entry->ifindex = val; 1355 } 1356 1357 if (tb[MPTCP_PM_ADDR_ATTR_FLAGS]) 1358 entry->flags = nla_get_u32(tb[MPTCP_PM_ADDR_ATTR_FLAGS]); 1359 1360 if (tb[MPTCP_PM_ADDR_ATTR_PORT]) 1361 entry->addr.port = htons(nla_get_u16(tb[MPTCP_PM_ADDR_ATTR_PORT])); 1362 1363 return 0; 1364 } 1365 1366 static struct pm_nl_pernet *genl_info_pm_nl(struct genl_info *info) 1367 { 1368 return pm_nl_get_pernet(genl_info_net(info)); 1369 } 1370 1371 static int mptcp_nl_add_subflow_or_signal_addr(struct net *net, 1372 struct mptcp_addr_info *addr) 1373 { 1374 struct mptcp_sock *msk; 1375 long s_slot = 0, s_num = 0; 1376 1377 while ((msk = mptcp_token_iter_next(net, &s_slot, &s_num)) != NULL) { 1378 struct sock *sk = (struct sock *)msk; 1379 struct mptcp_addr_info mpc_addr; 1380 1381 if (!READ_ONCE(msk->fully_established) || 1382 mptcp_pm_is_userspace(msk)) 1383 goto next; 1384 1385 /* if the endp linked to the init sf is re-added with a != ID */ 1386 mptcp_local_address((struct sock_common *)msk, &mpc_addr); 1387 1388 lock_sock(sk); 1389 spin_lock_bh(&msk->pm.lock); 1390 if (mptcp_addresses_equal(addr, &mpc_addr, addr->port)) 1391 msk->mpc_endpoint_id = addr->id; 1392 mptcp_pm_create_subflow_or_signal_addr(msk); 1393 spin_unlock_bh(&msk->pm.lock); 1394 release_sock(sk); 1395 1396 next: 1397 sock_put(sk); 1398 cond_resched(); 1399 } 1400 1401 return 0; 1402 } 1403 1404 static bool mptcp_pm_has_addr_attr_id(const struct nlattr *attr, 1405 struct genl_info *info) 1406 { 1407 struct nlattr *tb[MPTCP_PM_ADDR_ATTR_MAX + 1]; 1408 1409 if (!nla_parse_nested_deprecated(tb, MPTCP_PM_ADDR_ATTR_MAX, attr, 1410 mptcp_pm_addr_policy, info->extack) && 1411 tb[MPTCP_PM_ADDR_ATTR_ID]) 1412 return true; 1413 return false; 1414 } 1415 1416 static int mptcp_nl_cmd_add_addr(struct sk_buff *skb, struct genl_info *info) 1417 { 1418 struct nlattr *attr = info->attrs[MPTCP_PM_ATTR_ADDR]; 1419 struct pm_nl_pernet *pernet = genl_info_pm_nl(info); 1420 struct mptcp_pm_addr_entry addr, *entry; 1421 int ret; 1422 1423 ret = mptcp_pm_parse_entry(attr, info, true, &addr); 1424 if (ret < 0) 1425 return ret; 1426 1427 if (addr.addr.port && !address_use_port(&addr)) { 1428 GENL_SET_ERR_MSG(info, "flags must have signal and not subflow when using port"); 1429 return -EINVAL; 1430 } 1431 1432 if (addr.flags & MPTCP_PM_ADDR_FLAG_SIGNAL && 1433 addr.flags & MPTCP_PM_ADDR_FLAG_FULLMESH) { 1434 GENL_SET_ERR_MSG(info, "flags mustn't have both signal and fullmesh"); 1435 return -EINVAL; 1436 } 1437 1438 if (addr.flags & MPTCP_PM_ADDR_FLAG_IMPLICIT) { 1439 GENL_SET_ERR_MSG(info, "can't create IMPLICIT endpoint"); 1440 return -EINVAL; 1441 } 1442 1443 entry = kzalloc(sizeof(*entry), GFP_KERNEL_ACCOUNT); 1444 if (!entry) { 1445 GENL_SET_ERR_MSG(info, "can't allocate addr"); 1446 return -ENOMEM; 1447 } 1448 1449 *entry = addr; 1450 if (entry->addr.port) { 1451 ret = mptcp_pm_nl_create_listen_socket(skb->sk, entry); 1452 if (ret) { 1453 GENL_SET_ERR_MSG_FMT(info, "create listen socket error: %d", ret); 1454 goto out_free; 1455 } 1456 } 1457 ret = mptcp_pm_nl_append_new_local_addr(pernet, entry, 1458 !mptcp_pm_has_addr_attr_id(attr, info)); 1459 if (ret < 0) { 1460 GENL_SET_ERR_MSG_FMT(info, "too many addresses or duplicate one: %d", ret); 1461 goto out_free; 1462 } 1463 1464 mptcp_nl_add_subflow_or_signal_addr(sock_net(skb->sk), &entry->addr); 1465 return 0; 1466 1467 out_free: 1468 __mptcp_pm_release_addr_entry(entry); 1469 return ret; 1470 } 1471 1472 int mptcp_pm_nl_get_flags_and_ifindex_by_id(struct mptcp_sock *msk, unsigned int id, 1473 u8 *flags, int *ifindex) 1474 { 1475 struct mptcp_pm_addr_entry *entry; 1476 struct sock *sk = (struct sock *)msk; 1477 struct net *net = sock_net(sk); 1478 1479 /* No entries with ID 0 */ 1480 if (id == 0) 1481 return 0; 1482 1483 rcu_read_lock(); 1484 entry = __lookup_addr_by_id(pm_nl_get_pernet(net), id); 1485 if (entry) { 1486 *flags = entry->flags; 1487 *ifindex = entry->ifindex; 1488 } 1489 rcu_read_unlock(); 1490 1491 return 0; 1492 } 1493 1494 static bool remove_anno_list_by_saddr(struct mptcp_sock *msk, 1495 const struct mptcp_addr_info *addr) 1496 { 1497 struct mptcp_pm_add_entry *entry; 1498 1499 entry = mptcp_pm_del_add_timer(msk, addr, false); 1500 if (entry) { 1501 list_del(&entry->list); 1502 kfree(entry); 1503 return true; 1504 } 1505 1506 return false; 1507 } 1508 1509 static bool mptcp_pm_remove_anno_addr(struct mptcp_sock *msk, 1510 const struct mptcp_addr_info *addr, 1511 bool force) 1512 { 1513 struct mptcp_rm_list list = { .nr = 0 }; 1514 bool ret; 1515 1516 list.ids[list.nr++] = addr->id; 1517 1518 ret = remove_anno_list_by_saddr(msk, addr); 1519 if (ret || force) { 1520 spin_lock_bh(&msk->pm.lock); 1521 if (ret) { 1522 __set_bit(addr->id, msk->pm.id_avail_bitmap); 1523 msk->pm.add_addr_signaled--; 1524 } 1525 mptcp_pm_remove_addr(msk, &list); 1526 spin_unlock_bh(&msk->pm.lock); 1527 } 1528 return ret; 1529 } 1530 1531 static void __mark_subflow_endp_available(struct mptcp_sock *msk, u8 id) 1532 { 1533 /* If it was marked as used, and not ID 0, decrement local_addr_used */ 1534 if (!__test_and_set_bit(id ? : msk->mpc_endpoint_id, msk->pm.id_avail_bitmap) && 1535 id && !WARN_ON_ONCE(msk->pm.local_addr_used == 0)) 1536 msk->pm.local_addr_used--; 1537 } 1538 1539 static int mptcp_nl_remove_subflow_and_signal_addr(struct net *net, 1540 const struct mptcp_pm_addr_entry *entry) 1541 { 1542 const struct mptcp_addr_info *addr = &entry->addr; 1543 struct mptcp_rm_list list = { .nr = 0 }; 1544 long s_slot = 0, s_num = 0; 1545 struct mptcp_sock *msk; 1546 1547 pr_debug("remove_id=%d", addr->id); 1548 1549 list.ids[list.nr++] = addr->id; 1550 1551 while ((msk = mptcp_token_iter_next(net, &s_slot, &s_num)) != NULL) { 1552 struct sock *sk = (struct sock *)msk; 1553 bool remove_subflow; 1554 1555 if (mptcp_pm_is_userspace(msk)) 1556 goto next; 1557 1558 if (list_empty(&msk->conn_list)) { 1559 mptcp_pm_remove_anno_addr(msk, addr, false); 1560 goto next; 1561 } 1562 1563 lock_sock(sk); 1564 remove_subflow = lookup_subflow_by_saddr(&msk->conn_list, addr); 1565 mptcp_pm_remove_anno_addr(msk, addr, remove_subflow && 1566 !(entry->flags & MPTCP_PM_ADDR_FLAG_IMPLICIT)); 1567 1568 if (remove_subflow) { 1569 spin_lock_bh(&msk->pm.lock); 1570 mptcp_pm_nl_rm_subflow_received(msk, &list); 1571 spin_unlock_bh(&msk->pm.lock); 1572 } 1573 1574 if (entry->flags & MPTCP_PM_ADDR_FLAG_SUBFLOW) { 1575 spin_lock_bh(&msk->pm.lock); 1576 __mark_subflow_endp_available(msk, list.ids[0]); 1577 spin_unlock_bh(&msk->pm.lock); 1578 } 1579 1580 if (msk->mpc_endpoint_id == entry->addr.id) 1581 msk->mpc_endpoint_id = 0; 1582 release_sock(sk); 1583 1584 next: 1585 sock_put(sk); 1586 cond_resched(); 1587 } 1588 1589 return 0; 1590 } 1591 1592 static int mptcp_nl_remove_id_zero_address(struct net *net, 1593 struct mptcp_addr_info *addr) 1594 { 1595 struct mptcp_rm_list list = { .nr = 0 }; 1596 long s_slot = 0, s_num = 0; 1597 struct mptcp_sock *msk; 1598 1599 list.ids[list.nr++] = 0; 1600 1601 while ((msk = mptcp_token_iter_next(net, &s_slot, &s_num)) != NULL) { 1602 struct sock *sk = (struct sock *)msk; 1603 struct mptcp_addr_info msk_local; 1604 1605 if (list_empty(&msk->conn_list) || mptcp_pm_is_userspace(msk)) 1606 goto next; 1607 1608 mptcp_local_address((struct sock_common *)msk, &msk_local); 1609 if (!mptcp_addresses_equal(&msk_local, addr, addr->port)) 1610 goto next; 1611 1612 lock_sock(sk); 1613 spin_lock_bh(&msk->pm.lock); 1614 mptcp_pm_remove_addr(msk, &list); 1615 mptcp_pm_nl_rm_subflow_received(msk, &list); 1616 __mark_subflow_endp_available(msk, 0); 1617 spin_unlock_bh(&msk->pm.lock); 1618 release_sock(sk); 1619 1620 next: 1621 sock_put(sk); 1622 cond_resched(); 1623 } 1624 1625 return 0; 1626 } 1627 1628 static int mptcp_nl_cmd_del_addr(struct sk_buff *skb, struct genl_info *info) 1629 { 1630 struct nlattr *attr = info->attrs[MPTCP_PM_ATTR_ADDR]; 1631 struct pm_nl_pernet *pernet = genl_info_pm_nl(info); 1632 struct mptcp_pm_addr_entry addr, *entry; 1633 unsigned int addr_max; 1634 int ret; 1635 1636 ret = mptcp_pm_parse_entry(attr, info, false, &addr); 1637 if (ret < 0) 1638 return ret; 1639 1640 /* the zero id address is special: the first address used by the msk 1641 * always gets such an id, so different subflows can have different zero 1642 * id addresses. Additionally zero id is not accounted for in id_bitmap. 1643 * Let's use an 'mptcp_rm_list' instead of the common remove code. 1644 */ 1645 if (addr.addr.id == 0) 1646 return mptcp_nl_remove_id_zero_address(sock_net(skb->sk), &addr.addr); 1647 1648 spin_lock_bh(&pernet->lock); 1649 entry = __lookup_addr_by_id(pernet, addr.addr.id); 1650 if (!entry) { 1651 GENL_SET_ERR_MSG(info, "address not found"); 1652 spin_unlock_bh(&pernet->lock); 1653 return -EINVAL; 1654 } 1655 if (entry->flags & MPTCP_PM_ADDR_FLAG_SIGNAL) { 1656 addr_max = pernet->add_addr_signal_max; 1657 WRITE_ONCE(pernet->add_addr_signal_max, addr_max - 1); 1658 } 1659 if (entry->flags & MPTCP_PM_ADDR_FLAG_SUBFLOW) { 1660 addr_max = pernet->local_addr_max; 1661 WRITE_ONCE(pernet->local_addr_max, addr_max - 1); 1662 } 1663 1664 pernet->addrs--; 1665 list_del_rcu(&entry->list); 1666 __clear_bit(entry->addr.id, pernet->id_bitmap); 1667 spin_unlock_bh(&pernet->lock); 1668 1669 mptcp_nl_remove_subflow_and_signal_addr(sock_net(skb->sk), entry); 1670 synchronize_rcu(); 1671 __mptcp_pm_release_addr_entry(entry); 1672 1673 return ret; 1674 } 1675 1676 void mptcp_pm_remove_addrs(struct mptcp_sock *msk, struct list_head *rm_list) 1677 { 1678 struct mptcp_rm_list alist = { .nr = 0 }; 1679 struct mptcp_pm_addr_entry *entry; 1680 int anno_nr = 0; 1681 1682 list_for_each_entry(entry, rm_list, list) { 1683 if (alist.nr >= MPTCP_RM_IDS_MAX) 1684 break; 1685 1686 /* only delete if either announced or matching a subflow */ 1687 if (remove_anno_list_by_saddr(msk, &entry->addr)) 1688 anno_nr++; 1689 else if (!lookup_subflow_by_saddr(&msk->conn_list, 1690 &entry->addr)) 1691 continue; 1692 1693 alist.ids[alist.nr++] = entry->addr.id; 1694 } 1695 1696 if (alist.nr) { 1697 spin_lock_bh(&msk->pm.lock); 1698 msk->pm.add_addr_signaled -= anno_nr; 1699 mptcp_pm_remove_addr(msk, &alist); 1700 spin_unlock_bh(&msk->pm.lock); 1701 } 1702 } 1703 1704 void mptcp_pm_remove_addrs_and_subflows(struct mptcp_sock *msk, 1705 struct list_head *rm_list) 1706 { 1707 struct mptcp_rm_list alist = { .nr = 0 }, slist = { .nr = 0 }; 1708 struct mptcp_pm_addr_entry *entry; 1709 1710 list_for_each_entry(entry, rm_list, list) { 1711 if (slist.nr < MPTCP_RM_IDS_MAX && 1712 lookup_subflow_by_saddr(&msk->conn_list, &entry->addr)) 1713 slist.ids[slist.nr++] = entry->addr.id; 1714 1715 if (alist.nr < MPTCP_RM_IDS_MAX && 1716 remove_anno_list_by_saddr(msk, &entry->addr)) 1717 alist.ids[alist.nr++] = entry->addr.id; 1718 } 1719 1720 spin_lock_bh(&msk->pm.lock); 1721 if (alist.nr) { 1722 msk->pm.add_addr_signaled -= alist.nr; 1723 mptcp_pm_remove_addr(msk, &alist); 1724 } 1725 if (slist.nr) 1726 mptcp_pm_nl_rm_subflow_received(msk, &slist); 1727 /* Reset counters: maybe some subflows have been removed before */ 1728 bitmap_fill(msk->pm.id_avail_bitmap, MPTCP_PM_MAX_ADDR_ID + 1); 1729 msk->pm.local_addr_used = 0; 1730 spin_unlock_bh(&msk->pm.lock); 1731 } 1732 1733 static void mptcp_nl_remove_addrs_list(struct net *net, 1734 struct list_head *rm_list) 1735 { 1736 long s_slot = 0, s_num = 0; 1737 struct mptcp_sock *msk; 1738 1739 if (list_empty(rm_list)) 1740 return; 1741 1742 while ((msk = mptcp_token_iter_next(net, &s_slot, &s_num)) != NULL) { 1743 struct sock *sk = (struct sock *)msk; 1744 1745 if (!mptcp_pm_is_userspace(msk)) { 1746 lock_sock(sk); 1747 mptcp_pm_remove_addrs_and_subflows(msk, rm_list); 1748 release_sock(sk); 1749 } 1750 1751 sock_put(sk); 1752 cond_resched(); 1753 } 1754 } 1755 1756 /* caller must ensure the RCU grace period is already elapsed */ 1757 static void __flush_addrs(struct list_head *list) 1758 { 1759 while (!list_empty(list)) { 1760 struct mptcp_pm_addr_entry *cur; 1761 1762 cur = list_entry(list->next, 1763 struct mptcp_pm_addr_entry, list); 1764 list_del_rcu(&cur->list); 1765 __mptcp_pm_release_addr_entry(cur); 1766 } 1767 } 1768 1769 static void __reset_counters(struct pm_nl_pernet *pernet) 1770 { 1771 WRITE_ONCE(pernet->add_addr_signal_max, 0); 1772 WRITE_ONCE(pernet->add_addr_accept_max, 0); 1773 WRITE_ONCE(pernet->local_addr_max, 0); 1774 pernet->addrs = 0; 1775 } 1776 1777 static int mptcp_nl_cmd_flush_addrs(struct sk_buff *skb, struct genl_info *info) 1778 { 1779 struct pm_nl_pernet *pernet = genl_info_pm_nl(info); 1780 LIST_HEAD(free_list); 1781 1782 spin_lock_bh(&pernet->lock); 1783 list_splice_init(&pernet->local_addr_list, &free_list); 1784 __reset_counters(pernet); 1785 pernet->next_id = 1; 1786 bitmap_zero(pernet->id_bitmap, MPTCP_PM_MAX_ADDR_ID + 1); 1787 spin_unlock_bh(&pernet->lock); 1788 mptcp_nl_remove_addrs_list(sock_net(skb->sk), &free_list); 1789 synchronize_rcu(); 1790 __flush_addrs(&free_list); 1791 return 0; 1792 } 1793 1794 static int mptcp_nl_fill_addr(struct sk_buff *skb, 1795 struct mptcp_pm_addr_entry *entry) 1796 { 1797 struct mptcp_addr_info *addr = &entry->addr; 1798 struct nlattr *attr; 1799 1800 attr = nla_nest_start(skb, MPTCP_PM_ATTR_ADDR); 1801 if (!attr) 1802 return -EMSGSIZE; 1803 1804 if (nla_put_u16(skb, MPTCP_PM_ADDR_ATTR_FAMILY, addr->family)) 1805 goto nla_put_failure; 1806 if (nla_put_u16(skb, MPTCP_PM_ADDR_ATTR_PORT, ntohs(addr->port))) 1807 goto nla_put_failure; 1808 if (nla_put_u8(skb, MPTCP_PM_ADDR_ATTR_ID, addr->id)) 1809 goto nla_put_failure; 1810 if (nla_put_u32(skb, MPTCP_PM_ADDR_ATTR_FLAGS, entry->flags)) 1811 goto nla_put_failure; 1812 if (entry->ifindex && 1813 nla_put_s32(skb, MPTCP_PM_ADDR_ATTR_IF_IDX, entry->ifindex)) 1814 goto nla_put_failure; 1815 1816 if (addr->family == AF_INET && 1817 nla_put_in_addr(skb, MPTCP_PM_ADDR_ATTR_ADDR4, 1818 addr->addr.s_addr)) 1819 goto nla_put_failure; 1820 #if IS_ENABLED(CONFIG_MPTCP_IPV6) 1821 else if (addr->family == AF_INET6 && 1822 nla_put_in6_addr(skb, MPTCP_PM_ADDR_ATTR_ADDR6, &addr->addr6)) 1823 goto nla_put_failure; 1824 #endif 1825 nla_nest_end(skb, attr); 1826 return 0; 1827 1828 nla_put_failure: 1829 nla_nest_cancel(skb, attr); 1830 return -EMSGSIZE; 1831 } 1832 1833 static int mptcp_nl_cmd_get_addr(struct sk_buff *skb, struct genl_info *info) 1834 { 1835 struct nlattr *attr = info->attrs[MPTCP_PM_ATTR_ADDR]; 1836 struct pm_nl_pernet *pernet = genl_info_pm_nl(info); 1837 struct mptcp_pm_addr_entry addr, *entry; 1838 struct sk_buff *msg; 1839 void *reply; 1840 int ret; 1841 1842 ret = mptcp_pm_parse_entry(attr, info, false, &addr); 1843 if (ret < 0) 1844 return ret; 1845 1846 msg = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL); 1847 if (!msg) 1848 return -ENOMEM; 1849 1850 reply = genlmsg_put_reply(msg, info, &mptcp_genl_family, 0, 1851 info->genlhdr->cmd); 1852 if (!reply) { 1853 GENL_SET_ERR_MSG(info, "not enough space in Netlink message"); 1854 ret = -EMSGSIZE; 1855 goto fail; 1856 } 1857 1858 spin_lock_bh(&pernet->lock); 1859 entry = __lookup_addr_by_id(pernet, addr.addr.id); 1860 if (!entry) { 1861 GENL_SET_ERR_MSG(info, "address not found"); 1862 ret = -EINVAL; 1863 goto unlock_fail; 1864 } 1865 1866 ret = mptcp_nl_fill_addr(msg, entry); 1867 if (ret) 1868 goto unlock_fail; 1869 1870 genlmsg_end(msg, reply); 1871 ret = genlmsg_reply(msg, info); 1872 spin_unlock_bh(&pernet->lock); 1873 return ret; 1874 1875 unlock_fail: 1876 spin_unlock_bh(&pernet->lock); 1877 1878 fail: 1879 nlmsg_free(msg); 1880 return ret; 1881 } 1882 1883 static int mptcp_nl_cmd_dump_addrs(struct sk_buff *msg, 1884 struct netlink_callback *cb) 1885 { 1886 struct net *net = sock_net(msg->sk); 1887 struct mptcp_pm_addr_entry *entry; 1888 struct pm_nl_pernet *pernet; 1889 int id = cb->args[0]; 1890 void *hdr; 1891 int i; 1892 1893 pernet = pm_nl_get_pernet(net); 1894 1895 spin_lock_bh(&pernet->lock); 1896 for (i = id; i < MPTCP_PM_MAX_ADDR_ID + 1; i++) { 1897 if (test_bit(i, pernet->id_bitmap)) { 1898 entry = __lookup_addr_by_id(pernet, i); 1899 if (!entry) 1900 break; 1901 1902 if (entry->addr.id <= id) 1903 continue; 1904 1905 hdr = genlmsg_put(msg, NETLINK_CB(cb->skb).portid, 1906 cb->nlh->nlmsg_seq, &mptcp_genl_family, 1907 NLM_F_MULTI, MPTCP_PM_CMD_GET_ADDR); 1908 if (!hdr) 1909 break; 1910 1911 if (mptcp_nl_fill_addr(msg, entry) < 0) { 1912 genlmsg_cancel(msg, hdr); 1913 break; 1914 } 1915 1916 id = entry->addr.id; 1917 genlmsg_end(msg, hdr); 1918 } 1919 } 1920 spin_unlock_bh(&pernet->lock); 1921 1922 cb->args[0] = id; 1923 return msg->len; 1924 } 1925 1926 static int parse_limit(struct genl_info *info, int id, unsigned int *limit) 1927 { 1928 struct nlattr *attr = info->attrs[id]; 1929 1930 if (!attr) 1931 return 0; 1932 1933 *limit = nla_get_u32(attr); 1934 if (*limit > MPTCP_PM_ADDR_MAX) { 1935 GENL_SET_ERR_MSG(info, "limit greater than maximum"); 1936 return -EINVAL; 1937 } 1938 return 0; 1939 } 1940 1941 static int 1942 mptcp_nl_cmd_set_limits(struct sk_buff *skb, struct genl_info *info) 1943 { 1944 struct pm_nl_pernet *pernet = genl_info_pm_nl(info); 1945 unsigned int rcv_addrs, subflows; 1946 int ret; 1947 1948 spin_lock_bh(&pernet->lock); 1949 rcv_addrs = pernet->add_addr_accept_max; 1950 ret = parse_limit(info, MPTCP_PM_ATTR_RCV_ADD_ADDRS, &rcv_addrs); 1951 if (ret) 1952 goto unlock; 1953 1954 subflows = pernet->subflows_max; 1955 ret = parse_limit(info, MPTCP_PM_ATTR_SUBFLOWS, &subflows); 1956 if (ret) 1957 goto unlock; 1958 1959 WRITE_ONCE(pernet->add_addr_accept_max, rcv_addrs); 1960 WRITE_ONCE(pernet->subflows_max, subflows); 1961 1962 unlock: 1963 spin_unlock_bh(&pernet->lock); 1964 return ret; 1965 } 1966 1967 static int 1968 mptcp_nl_cmd_get_limits(struct sk_buff *skb, struct genl_info *info) 1969 { 1970 struct pm_nl_pernet *pernet = genl_info_pm_nl(info); 1971 struct sk_buff *msg; 1972 void *reply; 1973 1974 msg = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL); 1975 if (!msg) 1976 return -ENOMEM; 1977 1978 reply = genlmsg_put_reply(msg, info, &mptcp_genl_family, 0, 1979 MPTCP_PM_CMD_GET_LIMITS); 1980 if (!reply) 1981 goto fail; 1982 1983 if (nla_put_u32(msg, MPTCP_PM_ATTR_RCV_ADD_ADDRS, 1984 READ_ONCE(pernet->add_addr_accept_max))) 1985 goto fail; 1986 1987 if (nla_put_u32(msg, MPTCP_PM_ATTR_SUBFLOWS, 1988 READ_ONCE(pernet->subflows_max))) 1989 goto fail; 1990 1991 genlmsg_end(msg, reply); 1992 return genlmsg_reply(msg, info); 1993 1994 fail: 1995 GENL_SET_ERR_MSG(info, "not enough space in Netlink message"); 1996 nlmsg_free(msg); 1997 return -EMSGSIZE; 1998 } 1999 2000 static void mptcp_pm_nl_fullmesh(struct mptcp_sock *msk, 2001 struct mptcp_addr_info *addr) 2002 { 2003 struct mptcp_rm_list list = { .nr = 0 }; 2004 2005 list.ids[list.nr++] = addr->id; 2006 2007 spin_lock_bh(&msk->pm.lock); 2008 mptcp_pm_nl_rm_subflow_received(msk, &list); 2009 __mark_subflow_endp_available(msk, list.ids[0]); 2010 mptcp_pm_create_subflow_or_signal_addr(msk); 2011 spin_unlock_bh(&msk->pm.lock); 2012 } 2013 2014 static int mptcp_nl_set_flags(struct net *net, 2015 struct mptcp_addr_info *addr, 2016 u8 bkup, u8 changed) 2017 { 2018 long s_slot = 0, s_num = 0; 2019 struct mptcp_sock *msk; 2020 int ret = -EINVAL; 2021 2022 while ((msk = mptcp_token_iter_next(net, &s_slot, &s_num)) != NULL) { 2023 struct sock *sk = (struct sock *)msk; 2024 2025 if (list_empty(&msk->conn_list) || mptcp_pm_is_userspace(msk)) 2026 goto next; 2027 2028 lock_sock(sk); 2029 if (changed & MPTCP_PM_ADDR_FLAG_BACKUP) 2030 ret = mptcp_pm_nl_mp_prio_send_ack(msk, addr, NULL, bkup); 2031 if (changed & MPTCP_PM_ADDR_FLAG_FULLMESH) 2032 mptcp_pm_nl_fullmesh(msk, addr); 2033 release_sock(sk); 2034 2035 next: 2036 sock_put(sk); 2037 cond_resched(); 2038 } 2039 2040 return ret; 2041 } 2042 2043 int mptcp_pm_nl_set_flags(struct net *net, struct mptcp_pm_addr_entry *addr, u8 bkup) 2044 { 2045 struct pm_nl_pernet *pernet = pm_nl_get_pernet(net); 2046 u8 changed, mask = MPTCP_PM_ADDR_FLAG_BACKUP | 2047 MPTCP_PM_ADDR_FLAG_FULLMESH; 2048 struct mptcp_pm_addr_entry *entry; 2049 u8 lookup_by_id = 0; 2050 2051 if (addr->addr.family == AF_UNSPEC) { 2052 lookup_by_id = 1; 2053 if (!addr->addr.id) 2054 return -EOPNOTSUPP; 2055 } 2056 2057 spin_lock_bh(&pernet->lock); 2058 entry = __lookup_addr(pernet, &addr->addr, lookup_by_id); 2059 if (!entry) { 2060 spin_unlock_bh(&pernet->lock); 2061 return -EINVAL; 2062 } 2063 if ((addr->flags & MPTCP_PM_ADDR_FLAG_FULLMESH) && 2064 (entry->flags & MPTCP_PM_ADDR_FLAG_SIGNAL)) { 2065 spin_unlock_bh(&pernet->lock); 2066 return -EINVAL; 2067 } 2068 2069 changed = (addr->flags ^ entry->flags) & mask; 2070 entry->flags = (entry->flags & ~mask) | (addr->flags & mask); 2071 *addr = *entry; 2072 spin_unlock_bh(&pernet->lock); 2073 2074 mptcp_nl_set_flags(net, &addr->addr, bkup, changed); 2075 return 0; 2076 } 2077 2078 static int mptcp_nl_cmd_set_flags(struct sk_buff *skb, struct genl_info *info) 2079 { 2080 struct mptcp_pm_addr_entry remote = { .addr = { .family = AF_UNSPEC }, }; 2081 struct mptcp_pm_addr_entry addr = { .addr = { .family = AF_UNSPEC }, }; 2082 struct nlattr *attr_rem = info->attrs[MPTCP_PM_ATTR_ADDR_REMOTE]; 2083 struct nlattr *token = info->attrs[MPTCP_PM_ATTR_TOKEN]; 2084 struct nlattr *attr = info->attrs[MPTCP_PM_ATTR_ADDR]; 2085 struct net *net = sock_net(skb->sk); 2086 u8 bkup = 0; 2087 int ret; 2088 2089 ret = mptcp_pm_parse_entry(attr, info, false, &addr); 2090 if (ret < 0) 2091 return ret; 2092 2093 if (attr_rem) { 2094 ret = mptcp_pm_parse_entry(attr_rem, info, false, &remote); 2095 if (ret < 0) 2096 return ret; 2097 } 2098 2099 if (addr.flags & MPTCP_PM_ADDR_FLAG_BACKUP) 2100 bkup = 1; 2101 2102 return mptcp_pm_set_flags(net, token, &addr, &remote, bkup); 2103 } 2104 2105 static void mptcp_nl_mcast_send(struct net *net, struct sk_buff *nlskb, gfp_t gfp) 2106 { 2107 genlmsg_multicast_netns(&mptcp_genl_family, net, 2108 nlskb, 0, MPTCP_PM_EV_GRP_OFFSET, gfp); 2109 } 2110 2111 bool mptcp_userspace_pm_active(const struct mptcp_sock *msk) 2112 { 2113 return genl_has_listeners(&mptcp_genl_family, 2114 sock_net((const struct sock *)msk), 2115 MPTCP_PM_EV_GRP_OFFSET); 2116 } 2117 2118 static int mptcp_event_add_subflow(struct sk_buff *skb, const struct sock *ssk) 2119 { 2120 const struct inet_sock *issk = inet_sk(ssk); 2121 const struct mptcp_subflow_context *sf; 2122 2123 if (nla_put_u16(skb, MPTCP_ATTR_FAMILY, ssk->sk_family)) 2124 return -EMSGSIZE; 2125 2126 switch (ssk->sk_family) { 2127 case AF_INET: 2128 if (nla_put_in_addr(skb, MPTCP_ATTR_SADDR4, issk->inet_saddr)) 2129 return -EMSGSIZE; 2130 if (nla_put_in_addr(skb, MPTCP_ATTR_DADDR4, issk->inet_daddr)) 2131 return -EMSGSIZE; 2132 break; 2133 #if IS_ENABLED(CONFIG_MPTCP_IPV6) 2134 case AF_INET6: { 2135 const struct ipv6_pinfo *np = inet6_sk(ssk); 2136 2137 if (nla_put_in6_addr(skb, MPTCP_ATTR_SADDR6, &np->saddr)) 2138 return -EMSGSIZE; 2139 if (nla_put_in6_addr(skb, MPTCP_ATTR_DADDR6, &ssk->sk_v6_daddr)) 2140 return -EMSGSIZE; 2141 break; 2142 } 2143 #endif 2144 default: 2145 WARN_ON_ONCE(1); 2146 return -EMSGSIZE; 2147 } 2148 2149 if (nla_put_be16(skb, MPTCP_ATTR_SPORT, issk->inet_sport)) 2150 return -EMSGSIZE; 2151 if (nla_put_be16(skb, MPTCP_ATTR_DPORT, issk->inet_dport)) 2152 return -EMSGSIZE; 2153 2154 sf = mptcp_subflow_ctx(ssk); 2155 if (WARN_ON_ONCE(!sf)) 2156 return -EINVAL; 2157 2158 if (nla_put_u8(skb, MPTCP_ATTR_LOC_ID, subflow_get_local_id(sf))) 2159 return -EMSGSIZE; 2160 2161 if (nla_put_u8(skb, MPTCP_ATTR_REM_ID, sf->remote_id)) 2162 return -EMSGSIZE; 2163 2164 return 0; 2165 } 2166 2167 static int mptcp_event_put_token_and_ssk(struct sk_buff *skb, 2168 const struct mptcp_sock *msk, 2169 const struct sock *ssk) 2170 { 2171 const struct sock *sk = (const struct sock *)msk; 2172 const struct mptcp_subflow_context *sf; 2173 u8 sk_err; 2174 2175 if (nla_put_u32(skb, MPTCP_ATTR_TOKEN, msk->token)) 2176 return -EMSGSIZE; 2177 2178 if (mptcp_event_add_subflow(skb, ssk)) 2179 return -EMSGSIZE; 2180 2181 sf = mptcp_subflow_ctx(ssk); 2182 if (WARN_ON_ONCE(!sf)) 2183 return -EINVAL; 2184 2185 if (nla_put_u8(skb, MPTCP_ATTR_BACKUP, sf->backup)) 2186 return -EMSGSIZE; 2187 2188 if (ssk->sk_bound_dev_if && 2189 nla_put_s32(skb, MPTCP_ATTR_IF_IDX, ssk->sk_bound_dev_if)) 2190 return -EMSGSIZE; 2191 2192 sk_err = READ_ONCE(ssk->sk_err); 2193 if (sk_err && sk->sk_state == TCP_ESTABLISHED && 2194 nla_put_u8(skb, MPTCP_ATTR_ERROR, sk_err)) 2195 return -EMSGSIZE; 2196 2197 return 0; 2198 } 2199 2200 static int mptcp_event_sub_established(struct sk_buff *skb, 2201 const struct mptcp_sock *msk, 2202 const struct sock *ssk) 2203 { 2204 return mptcp_event_put_token_and_ssk(skb, msk, ssk); 2205 } 2206 2207 static int mptcp_event_sub_closed(struct sk_buff *skb, 2208 const struct mptcp_sock *msk, 2209 const struct sock *ssk) 2210 { 2211 const struct mptcp_subflow_context *sf; 2212 2213 if (mptcp_event_put_token_and_ssk(skb, msk, ssk)) 2214 return -EMSGSIZE; 2215 2216 sf = mptcp_subflow_ctx(ssk); 2217 if (!sf->reset_seen) 2218 return 0; 2219 2220 if (nla_put_u32(skb, MPTCP_ATTR_RESET_REASON, sf->reset_reason)) 2221 return -EMSGSIZE; 2222 2223 if (nla_put_u32(skb, MPTCP_ATTR_RESET_FLAGS, sf->reset_transient)) 2224 return -EMSGSIZE; 2225 2226 return 0; 2227 } 2228 2229 static int mptcp_event_created(struct sk_buff *skb, 2230 const struct mptcp_sock *msk, 2231 const struct sock *ssk) 2232 { 2233 int err = nla_put_u32(skb, MPTCP_ATTR_TOKEN, msk->token); 2234 2235 if (err) 2236 return err; 2237 2238 if (nla_put_u8(skb, MPTCP_ATTR_SERVER_SIDE, READ_ONCE(msk->pm.server_side))) 2239 return -EMSGSIZE; 2240 2241 return mptcp_event_add_subflow(skb, ssk); 2242 } 2243 2244 void mptcp_event_addr_removed(const struct mptcp_sock *msk, uint8_t id) 2245 { 2246 struct net *net = sock_net((const struct sock *)msk); 2247 struct nlmsghdr *nlh; 2248 struct sk_buff *skb; 2249 2250 if (!genl_has_listeners(&mptcp_genl_family, net, MPTCP_PM_EV_GRP_OFFSET)) 2251 return; 2252 2253 skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_ATOMIC); 2254 if (!skb) 2255 return; 2256 2257 nlh = genlmsg_put(skb, 0, 0, &mptcp_genl_family, 0, MPTCP_EVENT_REMOVED); 2258 if (!nlh) 2259 goto nla_put_failure; 2260 2261 if (nla_put_u32(skb, MPTCP_ATTR_TOKEN, msk->token)) 2262 goto nla_put_failure; 2263 2264 if (nla_put_u8(skb, MPTCP_ATTR_REM_ID, id)) 2265 goto nla_put_failure; 2266 2267 genlmsg_end(skb, nlh); 2268 mptcp_nl_mcast_send(net, skb, GFP_ATOMIC); 2269 return; 2270 2271 nla_put_failure: 2272 nlmsg_free(skb); 2273 } 2274 2275 void mptcp_event_addr_announced(const struct sock *ssk, 2276 const struct mptcp_addr_info *info) 2277 { 2278 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk); 2279 struct mptcp_sock *msk = mptcp_sk(subflow->conn); 2280 struct net *net = sock_net(ssk); 2281 struct nlmsghdr *nlh; 2282 struct sk_buff *skb; 2283 2284 if (!genl_has_listeners(&mptcp_genl_family, net, MPTCP_PM_EV_GRP_OFFSET)) 2285 return; 2286 2287 skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_ATOMIC); 2288 if (!skb) 2289 return; 2290 2291 nlh = genlmsg_put(skb, 0, 0, &mptcp_genl_family, 0, 2292 MPTCP_EVENT_ANNOUNCED); 2293 if (!nlh) 2294 goto nla_put_failure; 2295 2296 if (nla_put_u32(skb, MPTCP_ATTR_TOKEN, msk->token)) 2297 goto nla_put_failure; 2298 2299 if (nla_put_u8(skb, MPTCP_ATTR_REM_ID, info->id)) 2300 goto nla_put_failure; 2301 2302 if (nla_put_be16(skb, MPTCP_ATTR_DPORT, 2303 info->port == 0 ? 2304 inet_sk(ssk)->inet_dport : 2305 info->port)) 2306 goto nla_put_failure; 2307 2308 switch (info->family) { 2309 case AF_INET: 2310 if (nla_put_in_addr(skb, MPTCP_ATTR_DADDR4, info->addr.s_addr)) 2311 goto nla_put_failure; 2312 break; 2313 #if IS_ENABLED(CONFIG_MPTCP_IPV6) 2314 case AF_INET6: 2315 if (nla_put_in6_addr(skb, MPTCP_ATTR_DADDR6, &info->addr6)) 2316 goto nla_put_failure; 2317 break; 2318 #endif 2319 default: 2320 WARN_ON_ONCE(1); 2321 goto nla_put_failure; 2322 } 2323 2324 genlmsg_end(skb, nlh); 2325 mptcp_nl_mcast_send(net, skb, GFP_ATOMIC); 2326 return; 2327 2328 nla_put_failure: 2329 nlmsg_free(skb); 2330 } 2331 2332 void mptcp_event_pm_listener(const struct sock *ssk, 2333 enum mptcp_event_type event) 2334 { 2335 const struct inet_sock *issk = inet_sk(ssk); 2336 struct net *net = sock_net(ssk); 2337 struct nlmsghdr *nlh; 2338 struct sk_buff *skb; 2339 2340 if (!genl_has_listeners(&mptcp_genl_family, net, MPTCP_PM_EV_GRP_OFFSET)) 2341 return; 2342 2343 skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL); 2344 if (!skb) 2345 return; 2346 2347 nlh = genlmsg_put(skb, 0, 0, &mptcp_genl_family, 0, event); 2348 if (!nlh) 2349 goto nla_put_failure; 2350 2351 if (nla_put_u16(skb, MPTCP_ATTR_FAMILY, ssk->sk_family)) 2352 goto nla_put_failure; 2353 2354 if (nla_put_be16(skb, MPTCP_ATTR_SPORT, issk->inet_sport)) 2355 goto nla_put_failure; 2356 2357 switch (ssk->sk_family) { 2358 case AF_INET: 2359 if (nla_put_in_addr(skb, MPTCP_ATTR_SADDR4, issk->inet_saddr)) 2360 goto nla_put_failure; 2361 break; 2362 #if IS_ENABLED(CONFIG_MPTCP_IPV6) 2363 case AF_INET6: { 2364 const struct ipv6_pinfo *np = inet6_sk(ssk); 2365 2366 if (nla_put_in6_addr(skb, MPTCP_ATTR_SADDR6, &np->saddr)) 2367 goto nla_put_failure; 2368 break; 2369 } 2370 #endif 2371 default: 2372 WARN_ON_ONCE(1); 2373 goto nla_put_failure; 2374 } 2375 2376 genlmsg_end(skb, nlh); 2377 mptcp_nl_mcast_send(net, skb, GFP_KERNEL); 2378 return; 2379 2380 nla_put_failure: 2381 nlmsg_free(skb); 2382 } 2383 2384 void mptcp_event(enum mptcp_event_type type, const struct mptcp_sock *msk, 2385 const struct sock *ssk, gfp_t gfp) 2386 { 2387 struct net *net = sock_net((const struct sock *)msk); 2388 struct nlmsghdr *nlh; 2389 struct sk_buff *skb; 2390 2391 if (!genl_has_listeners(&mptcp_genl_family, net, MPTCP_PM_EV_GRP_OFFSET)) 2392 return; 2393 2394 skb = nlmsg_new(NLMSG_DEFAULT_SIZE, gfp); 2395 if (!skb) 2396 return; 2397 2398 nlh = genlmsg_put(skb, 0, 0, &mptcp_genl_family, 0, type); 2399 if (!nlh) 2400 goto nla_put_failure; 2401 2402 switch (type) { 2403 case MPTCP_EVENT_UNSPEC: 2404 WARN_ON_ONCE(1); 2405 break; 2406 case MPTCP_EVENT_CREATED: 2407 case MPTCP_EVENT_ESTABLISHED: 2408 if (mptcp_event_created(skb, msk, ssk) < 0) 2409 goto nla_put_failure; 2410 break; 2411 case MPTCP_EVENT_CLOSED: 2412 if (nla_put_u32(skb, MPTCP_ATTR_TOKEN, msk->token) < 0) 2413 goto nla_put_failure; 2414 break; 2415 case MPTCP_EVENT_ANNOUNCED: 2416 case MPTCP_EVENT_REMOVED: 2417 /* call mptcp_event_addr_announced()/removed instead */ 2418 WARN_ON_ONCE(1); 2419 break; 2420 case MPTCP_EVENT_SUB_ESTABLISHED: 2421 case MPTCP_EVENT_SUB_PRIORITY: 2422 if (mptcp_event_sub_established(skb, msk, ssk) < 0) 2423 goto nla_put_failure; 2424 break; 2425 case MPTCP_EVENT_SUB_CLOSED: 2426 if (mptcp_event_sub_closed(skb, msk, ssk) < 0) 2427 goto nla_put_failure; 2428 break; 2429 case MPTCP_EVENT_LISTENER_CREATED: 2430 case MPTCP_EVENT_LISTENER_CLOSED: 2431 break; 2432 } 2433 2434 genlmsg_end(skb, nlh); 2435 mptcp_nl_mcast_send(net, skb, gfp); 2436 return; 2437 2438 nla_put_failure: 2439 nlmsg_free(skb); 2440 } 2441 2442 static const struct genl_small_ops mptcp_pm_ops[] = { 2443 { 2444 .cmd = MPTCP_PM_CMD_ADD_ADDR, 2445 .doit = mptcp_nl_cmd_add_addr, 2446 .flags = GENL_UNS_ADMIN_PERM, 2447 }, 2448 { 2449 .cmd = MPTCP_PM_CMD_DEL_ADDR, 2450 .doit = mptcp_nl_cmd_del_addr, 2451 .flags = GENL_UNS_ADMIN_PERM, 2452 }, 2453 { 2454 .cmd = MPTCP_PM_CMD_FLUSH_ADDRS, 2455 .doit = mptcp_nl_cmd_flush_addrs, 2456 .flags = GENL_UNS_ADMIN_PERM, 2457 }, 2458 { 2459 .cmd = MPTCP_PM_CMD_GET_ADDR, 2460 .doit = mptcp_nl_cmd_get_addr, 2461 .dumpit = mptcp_nl_cmd_dump_addrs, 2462 }, 2463 { 2464 .cmd = MPTCP_PM_CMD_SET_LIMITS, 2465 .doit = mptcp_nl_cmd_set_limits, 2466 .flags = GENL_UNS_ADMIN_PERM, 2467 }, 2468 { 2469 .cmd = MPTCP_PM_CMD_GET_LIMITS, 2470 .doit = mptcp_nl_cmd_get_limits, 2471 }, 2472 { 2473 .cmd = MPTCP_PM_CMD_SET_FLAGS, 2474 .doit = mptcp_nl_cmd_set_flags, 2475 .flags = GENL_UNS_ADMIN_PERM, 2476 }, 2477 { 2478 .cmd = MPTCP_PM_CMD_ANNOUNCE, 2479 .doit = mptcp_nl_cmd_announce, 2480 .flags = GENL_UNS_ADMIN_PERM, 2481 }, 2482 { 2483 .cmd = MPTCP_PM_CMD_REMOVE, 2484 .doit = mptcp_nl_cmd_remove, 2485 .flags = GENL_UNS_ADMIN_PERM, 2486 }, 2487 { 2488 .cmd = MPTCP_PM_CMD_SUBFLOW_CREATE, 2489 .doit = mptcp_nl_cmd_sf_create, 2490 .flags = GENL_UNS_ADMIN_PERM, 2491 }, 2492 { 2493 .cmd = MPTCP_PM_CMD_SUBFLOW_DESTROY, 2494 .doit = mptcp_nl_cmd_sf_destroy, 2495 .flags = GENL_UNS_ADMIN_PERM, 2496 }, 2497 }; 2498 2499 static struct genl_family mptcp_genl_family __ro_after_init = { 2500 .name = MPTCP_PM_NAME, 2501 .version = MPTCP_PM_VER, 2502 .maxattr = MPTCP_PM_ATTR_MAX, 2503 .policy = mptcp_pm_policy, 2504 .netnsok = true, 2505 .module = THIS_MODULE, 2506 .small_ops = mptcp_pm_ops, 2507 .n_small_ops = ARRAY_SIZE(mptcp_pm_ops), 2508 .resv_start_op = MPTCP_PM_CMD_SUBFLOW_DESTROY + 1, 2509 .mcgrps = mptcp_pm_mcgrps, 2510 .n_mcgrps = ARRAY_SIZE(mptcp_pm_mcgrps), 2511 }; 2512 2513 static int __net_init pm_nl_init_net(struct net *net) 2514 { 2515 struct pm_nl_pernet *pernet = pm_nl_get_pernet(net); 2516 2517 INIT_LIST_HEAD_RCU(&pernet->local_addr_list); 2518 2519 /* Cit. 2 subflows ought to be enough for anybody. */ 2520 pernet->subflows_max = 2; 2521 pernet->next_id = 1; 2522 pernet->stale_loss_cnt = 4; 2523 spin_lock_init(&pernet->lock); 2524 2525 /* No need to initialize other pernet fields, the struct is zeroed at 2526 * allocation time. 2527 */ 2528 2529 return 0; 2530 } 2531 2532 static void __net_exit pm_nl_exit_net(struct list_head *net_list) 2533 { 2534 struct net *net; 2535 2536 list_for_each_entry(net, net_list, exit_list) { 2537 struct pm_nl_pernet *pernet = pm_nl_get_pernet(net); 2538 2539 /* net is removed from namespace list, can't race with 2540 * other modifiers, also netns core already waited for a 2541 * RCU grace period. 2542 */ 2543 __flush_addrs(&pernet->local_addr_list); 2544 } 2545 } 2546 2547 static struct pernet_operations mptcp_pm_pernet_ops = { 2548 .init = pm_nl_init_net, 2549 .exit_batch = pm_nl_exit_net, 2550 .id = &pm_nl_pernet_id, 2551 .size = sizeof(struct pm_nl_pernet), 2552 }; 2553 2554 void __init mptcp_pm_nl_init(void) 2555 { 2556 if (register_pernet_subsys(&mptcp_pm_pernet_ops) < 0) 2557 panic("Failed to register MPTCP PM pernet subsystem.\n"); 2558 2559 if (genl_register_family(&mptcp_genl_family)) 2560 panic("Failed to register MPTCP PM netlink family\n"); 2561 } 2562