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\n", 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\n", 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 struct timer_list *add_timer = NULL; 343 344 spin_lock_bh(&msk->pm.lock); 345 entry = mptcp_lookup_anno_list_by_saddr(msk, addr); 346 if (entry && (!check_id || entry->addr.id == addr->id)) { 347 entry->retrans_times = ADD_ADDR_RETRANS_MAX; 348 add_timer = &entry->add_timer; 349 } 350 if (!check_id && entry) 351 list_del(&entry->list); 352 spin_unlock_bh(&msk->pm.lock); 353 354 /* no lock, because sk_stop_timer_sync() is calling del_timer_sync() */ 355 if (add_timer) 356 sk_stop_timer_sync(sk, add_timer); 357 358 return entry; 359 } 360 361 bool mptcp_pm_alloc_anno_list(struct mptcp_sock *msk, 362 const struct mptcp_addr_info *addr) 363 { 364 struct mptcp_pm_add_entry *add_entry = NULL; 365 struct sock *sk = (struct sock *)msk; 366 struct net *net = sock_net(sk); 367 368 lockdep_assert_held(&msk->pm.lock); 369 370 add_entry = mptcp_lookup_anno_list_by_saddr(msk, addr); 371 372 if (add_entry) { 373 if (WARN_ON_ONCE(mptcp_pm_is_kernel(msk))) 374 return false; 375 376 sk_reset_timer(sk, &add_entry->add_timer, 377 jiffies + mptcp_get_add_addr_timeout(net)); 378 return true; 379 } 380 381 add_entry = kmalloc(sizeof(*add_entry), GFP_ATOMIC); 382 if (!add_entry) 383 return false; 384 385 list_add(&add_entry->list, &msk->pm.anno_list); 386 387 add_entry->addr = *addr; 388 add_entry->sock = msk; 389 add_entry->retrans_times = 0; 390 391 timer_setup(&add_entry->add_timer, mptcp_pm_add_timer, 0); 392 sk_reset_timer(sk, &add_entry->add_timer, 393 jiffies + mptcp_get_add_addr_timeout(net)); 394 395 return true; 396 } 397 398 void mptcp_pm_free_anno_list(struct mptcp_sock *msk) 399 { 400 struct mptcp_pm_add_entry *entry, *tmp; 401 struct sock *sk = (struct sock *)msk; 402 LIST_HEAD(free_list); 403 404 pr_debug("msk=%p\n", msk); 405 406 spin_lock_bh(&msk->pm.lock); 407 list_splice_init(&msk->pm.anno_list, &free_list); 408 spin_unlock_bh(&msk->pm.lock); 409 410 list_for_each_entry_safe(entry, tmp, &free_list, list) { 411 sk_stop_timer_sync(sk, &entry->add_timer); 412 kfree(entry); 413 } 414 } 415 416 /* Fill all the remote addresses into the array addrs[], 417 * and return the array size. 418 */ 419 static unsigned int fill_remote_addresses_vec(struct mptcp_sock *msk, 420 struct mptcp_addr_info *local, 421 bool fullmesh, 422 struct mptcp_addr_info *addrs) 423 { 424 bool deny_id0 = READ_ONCE(msk->pm.remote_deny_join_id0); 425 struct sock *sk = (struct sock *)msk, *ssk; 426 struct mptcp_subflow_context *subflow; 427 struct mptcp_addr_info remote = { 0 }; 428 unsigned int subflows_max; 429 int i = 0; 430 431 subflows_max = mptcp_pm_get_subflows_max(msk); 432 remote_address((struct sock_common *)sk, &remote); 433 434 /* Non-fullmesh endpoint, fill in the single entry 435 * corresponding to the primary MPC subflow remote address 436 */ 437 if (!fullmesh) { 438 if (deny_id0) 439 return 0; 440 441 if (!mptcp_pm_addr_families_match(sk, local, &remote)) 442 return 0; 443 444 msk->pm.subflows++; 445 addrs[i++] = remote; 446 } else { 447 DECLARE_BITMAP(unavail_id, MPTCP_PM_MAX_ADDR_ID + 1); 448 449 /* Forbid creation of new subflows matching existing 450 * ones, possibly already created by incoming ADD_ADDR 451 */ 452 bitmap_zero(unavail_id, MPTCP_PM_MAX_ADDR_ID + 1); 453 mptcp_for_each_subflow(msk, subflow) 454 if (READ_ONCE(subflow->local_id) == local->id) 455 __set_bit(subflow->remote_id, unavail_id); 456 457 mptcp_for_each_subflow(msk, subflow) { 458 ssk = mptcp_subflow_tcp_sock(subflow); 459 remote_address((struct sock_common *)ssk, &addrs[i]); 460 addrs[i].id = READ_ONCE(subflow->remote_id); 461 if (deny_id0 && !addrs[i].id) 462 continue; 463 464 if (test_bit(addrs[i].id, unavail_id)) 465 continue; 466 467 if (!mptcp_pm_addr_families_match(sk, local, &addrs[i])) 468 continue; 469 470 if (msk->pm.subflows < subflows_max) { 471 /* forbid creating multiple address towards 472 * this id 473 */ 474 __set_bit(addrs[i].id, unavail_id); 475 msk->pm.subflows++; 476 i++; 477 } 478 } 479 } 480 481 return i; 482 } 483 484 static void __mptcp_pm_send_ack(struct mptcp_sock *msk, struct mptcp_subflow_context *subflow, 485 bool prio, bool backup) 486 { 487 struct sock *ssk = mptcp_subflow_tcp_sock(subflow); 488 bool slow; 489 490 pr_debug("send ack for %s\n", 491 prio ? "mp_prio" : (mptcp_pm_should_add_signal(msk) ? "add_addr" : "rm_addr")); 492 493 slow = lock_sock_fast(ssk); 494 if (prio) { 495 subflow->send_mp_prio = 1; 496 subflow->request_bkup = backup; 497 } 498 499 __mptcp_subflow_send_ack(ssk); 500 unlock_sock_fast(ssk, slow); 501 } 502 503 static void mptcp_pm_send_ack(struct mptcp_sock *msk, struct mptcp_subflow_context *subflow, 504 bool prio, bool backup) 505 { 506 spin_unlock_bh(&msk->pm.lock); 507 __mptcp_pm_send_ack(msk, subflow, prio, backup); 508 spin_lock_bh(&msk->pm.lock); 509 } 510 511 static struct mptcp_pm_addr_entry * 512 __lookup_addr_by_id(struct pm_nl_pernet *pernet, unsigned int id) 513 { 514 struct mptcp_pm_addr_entry *entry; 515 516 list_for_each_entry(entry, &pernet->local_addr_list, list) { 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 struct mptcp_pm_addr_entry local; 540 unsigned int add_addr_signal_max; 541 bool signal_and_subflow = false; 542 unsigned int local_addr_max; 543 struct pm_nl_pernet *pernet; 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.addr, &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_addr_info *addrs) 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 msk->pm.subflows++; 689 addrs[i] = entry->addr; 690 691 /* Special case for ID0: set the correct ID */ 692 if (mptcp_addresses_equal(&entry->addr, &mpc_addr, entry->addr.port)) 693 addrs[i].id = 0; 694 695 i++; 696 } 697 } 698 rcu_read_unlock(); 699 700 /* If the array is empty, fill in the single 701 * 'IPADDRANY' local address 702 */ 703 if (!i) { 704 struct mptcp_addr_info local; 705 706 memset(&local, 0, sizeof(local)); 707 local.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, &local, remote)) 715 return 0; 716 717 msk->pm.subflows++; 718 addrs[i++] = local; 719 } 720 721 return i; 722 } 723 724 static void mptcp_pm_nl_add_addr_received(struct mptcp_sock *msk) 725 { 726 struct mptcp_addr_info addrs[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, addrs); 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, &addrs[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; 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 mptcp_pm_send_ack(msk, subflow, false, false); 798 break; 799 } 800 } 801 } 802 803 int mptcp_pm_nl_mp_prio_send_ack(struct mptcp_sock *msk, 804 struct mptcp_addr_info *addr, 805 struct mptcp_addr_info *rem, 806 u8 bkup) 807 { 808 struct mptcp_subflow_context *subflow; 809 810 pr_debug("bkup=%d\n", bkup); 811 812 mptcp_for_each_subflow(msk, subflow) { 813 struct sock *ssk = mptcp_subflow_tcp_sock(subflow); 814 struct mptcp_addr_info local, remote; 815 816 mptcp_local_address((struct sock_common *)ssk, &local); 817 if (!mptcp_addresses_equal(&local, addr, addr->port)) 818 continue; 819 820 if (rem && rem->family != AF_UNSPEC) { 821 remote_address((struct sock_common *)ssk, &remote); 822 if (!mptcp_addresses_equal(&remote, rem, rem->port)) 823 continue; 824 } 825 826 __mptcp_pm_send_ack(msk, subflow, true, bkup); 827 return 0; 828 } 829 830 return -EINVAL; 831 } 832 833 static void mptcp_pm_nl_rm_addr_or_subflow(struct mptcp_sock *msk, 834 const struct mptcp_rm_list *rm_list, 835 enum linux_mptcp_mib_field rm_type) 836 { 837 struct mptcp_subflow_context *subflow, *tmp; 838 struct sock *sk = (struct sock *)msk; 839 u8 i; 840 841 pr_debug("%s rm_list_nr %d\n", 842 rm_type == MPTCP_MIB_RMADDR ? "address" : "subflow", rm_list->nr); 843 844 msk_owned_by_me(msk); 845 846 if (sk->sk_state == TCP_LISTEN) 847 return; 848 849 if (!rm_list->nr) 850 return; 851 852 if (list_empty(&msk->conn_list)) 853 return; 854 855 for (i = 0; i < rm_list->nr; i++) { 856 u8 rm_id = rm_list->ids[i]; 857 bool removed = false; 858 859 mptcp_for_each_subflow_safe(msk, subflow, tmp) { 860 struct sock *ssk = mptcp_subflow_tcp_sock(subflow); 861 u8 remote_id = READ_ONCE(subflow->remote_id); 862 int how = RCV_SHUTDOWN | SEND_SHUTDOWN; 863 u8 id = subflow_get_local_id(subflow); 864 865 if ((1 << inet_sk_state_load(ssk)) & 866 (TCPF_FIN_WAIT1 | TCPF_FIN_WAIT2 | TCPF_CLOSING | TCPF_CLOSE)) 867 continue; 868 if (rm_type == MPTCP_MIB_RMADDR && remote_id != rm_id) 869 continue; 870 if (rm_type == MPTCP_MIB_RMSUBFLOW && id != rm_id) 871 continue; 872 873 pr_debug(" -> %s rm_list_ids[%d]=%u local_id=%u remote_id=%u mpc_id=%u\n", 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 removed |= subflow->request_join; 879 880 /* the following takes care of updating the subflows counter */ 881 mptcp_close_ssk(sk, ssk, subflow); 882 spin_lock_bh(&msk->pm.lock); 883 884 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\n", 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, bool replace) 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 /* allow callers that only need to look up the local 1013 * addr's id to skip replacement. This allows them to 1014 * avoid calling synchronize_rcu in the packet recv 1015 * path. 1016 */ 1017 if (!replace) { 1018 kfree(entry); 1019 ret = cur->addr.id; 1020 goto out; 1021 } 1022 1023 pernet->addrs--; 1024 entry->addr.id = cur->addr.id; 1025 list_del_rcu(&cur->list); 1026 del_entry = cur; 1027 break; 1028 } 1029 } 1030 1031 if (!entry->addr.id && needs_id) { 1032 find_next: 1033 entry->addr.id = find_next_zero_bit(pernet->id_bitmap, 1034 MPTCP_PM_MAX_ADDR_ID + 1, 1035 pernet->next_id); 1036 if (!entry->addr.id && pernet->next_id != 1) { 1037 pernet->next_id = 1; 1038 goto find_next; 1039 } 1040 } 1041 1042 if (!entry->addr.id && needs_id) 1043 goto out; 1044 1045 __set_bit(entry->addr.id, pernet->id_bitmap); 1046 if (entry->addr.id > pernet->next_id) 1047 pernet->next_id = entry->addr.id; 1048 1049 if (entry->flags & MPTCP_PM_ADDR_FLAG_SIGNAL) { 1050 addr_max = pernet->add_addr_signal_max; 1051 WRITE_ONCE(pernet->add_addr_signal_max, addr_max + 1); 1052 } 1053 if (entry->flags & MPTCP_PM_ADDR_FLAG_SUBFLOW) { 1054 addr_max = pernet->local_addr_max; 1055 WRITE_ONCE(pernet->local_addr_max, addr_max + 1); 1056 } 1057 1058 pernet->addrs++; 1059 if (!entry->addr.port) 1060 list_add_tail_rcu(&entry->list, &pernet->local_addr_list); 1061 else 1062 list_add_rcu(&entry->list, &pernet->local_addr_list); 1063 ret = entry->addr.id; 1064 1065 out: 1066 spin_unlock_bh(&pernet->lock); 1067 1068 /* just replaced an existing entry, free it */ 1069 if (del_entry) { 1070 synchronize_rcu(); 1071 __mptcp_pm_release_addr_entry(del_entry); 1072 } 1073 return ret; 1074 } 1075 1076 static struct lock_class_key mptcp_slock_keys[2]; 1077 static struct lock_class_key mptcp_keys[2]; 1078 1079 static int mptcp_pm_nl_create_listen_socket(struct sock *sk, 1080 struct mptcp_pm_addr_entry *entry) 1081 { 1082 bool is_ipv6 = sk->sk_family == AF_INET6; 1083 int addrlen = sizeof(struct sockaddr_in); 1084 struct sockaddr_storage addr; 1085 struct sock *newsk, *ssk; 1086 int backlog = 1024; 1087 int err; 1088 1089 err = sock_create_kern(sock_net(sk), entry->addr.family, 1090 SOCK_STREAM, IPPROTO_MPTCP, &entry->lsk); 1091 if (err) 1092 return err; 1093 1094 newsk = entry->lsk->sk; 1095 if (!newsk) 1096 return -EINVAL; 1097 1098 /* The subflow socket lock is acquired in a nested to the msk one 1099 * in several places, even by the TCP stack, and this msk is a kernel 1100 * socket: lockdep complains. Instead of propagating the _nested 1101 * modifiers in several places, re-init the lock class for the msk 1102 * socket to an mptcp specific one. 1103 */ 1104 sock_lock_init_class_and_name(newsk, 1105 is_ipv6 ? "mlock-AF_INET6" : "mlock-AF_INET", 1106 &mptcp_slock_keys[is_ipv6], 1107 is_ipv6 ? "msk_lock-AF_INET6" : "msk_lock-AF_INET", 1108 &mptcp_keys[is_ipv6]); 1109 1110 lock_sock(newsk); 1111 ssk = __mptcp_nmpc_sk(mptcp_sk(newsk)); 1112 release_sock(newsk); 1113 if (IS_ERR(ssk)) 1114 return PTR_ERR(ssk); 1115 1116 mptcp_info2sockaddr(&entry->addr, &addr, entry->addr.family); 1117 #if IS_ENABLED(CONFIG_MPTCP_IPV6) 1118 if (entry->addr.family == AF_INET6) 1119 addrlen = sizeof(struct sockaddr_in6); 1120 #endif 1121 if (ssk->sk_family == AF_INET) 1122 err = inet_bind_sk(ssk, (struct sockaddr *)&addr, addrlen); 1123 #if IS_ENABLED(CONFIG_MPTCP_IPV6) 1124 else if (ssk->sk_family == AF_INET6) 1125 err = inet6_bind_sk(ssk, (struct sockaddr *)&addr, addrlen); 1126 #endif 1127 if (err) 1128 return err; 1129 1130 /* We don't use mptcp_set_state() here because it needs to be called 1131 * under the msk socket lock. For the moment, that will not bring 1132 * anything more than only calling inet_sk_state_store(), because the 1133 * old status is known (TCP_CLOSE). 1134 */ 1135 inet_sk_state_store(newsk, TCP_LISTEN); 1136 lock_sock(ssk); 1137 WRITE_ONCE(mptcp_subflow_ctx(ssk)->pm_listener, true); 1138 err = __inet_listen_sk(ssk, backlog); 1139 if (!err) 1140 mptcp_event_pm_listener(ssk, MPTCP_EVENT_LISTENER_CREATED); 1141 release_sock(ssk); 1142 return err; 1143 } 1144 1145 int mptcp_pm_nl_get_local_id(struct mptcp_sock *msk, struct mptcp_addr_info *skc) 1146 { 1147 struct mptcp_pm_addr_entry *entry; 1148 struct pm_nl_pernet *pernet; 1149 int ret = -1; 1150 1151 pernet = pm_nl_get_pernet_from_msk(msk); 1152 1153 rcu_read_lock(); 1154 list_for_each_entry_rcu(entry, &pernet->local_addr_list, list) { 1155 if (mptcp_addresses_equal(&entry->addr, skc, entry->addr.port)) { 1156 ret = entry->addr.id; 1157 break; 1158 } 1159 } 1160 rcu_read_unlock(); 1161 if (ret >= 0) 1162 return ret; 1163 1164 /* address not found, add to local list */ 1165 entry = kmalloc(sizeof(*entry), GFP_ATOMIC); 1166 if (!entry) 1167 return -ENOMEM; 1168 1169 entry->addr = *skc; 1170 entry->addr.id = 0; 1171 entry->addr.port = 0; 1172 entry->ifindex = 0; 1173 entry->flags = MPTCP_PM_ADDR_FLAG_IMPLICIT; 1174 entry->lsk = NULL; 1175 ret = mptcp_pm_nl_append_new_local_addr(pernet, entry, true, false); 1176 if (ret < 0) 1177 kfree(entry); 1178 1179 return ret; 1180 } 1181 1182 bool mptcp_pm_nl_is_backup(struct mptcp_sock *msk, struct mptcp_addr_info *skc) 1183 { 1184 struct pm_nl_pernet *pernet = pm_nl_get_pernet_from_msk(msk); 1185 struct mptcp_pm_addr_entry *entry; 1186 bool backup = false; 1187 1188 rcu_read_lock(); 1189 list_for_each_entry_rcu(entry, &pernet->local_addr_list, list) { 1190 if (mptcp_addresses_equal(&entry->addr, skc, entry->addr.port)) { 1191 backup = !!(entry->flags & MPTCP_PM_ADDR_FLAG_BACKUP); 1192 break; 1193 } 1194 } 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_UNS_ADMIN_PERM, 1207 }, 1208 }; 1209 1210 static const struct nla_policy 1211 mptcp_pm_addr_policy[MPTCP_PM_ADDR_ATTR_MAX + 1] = { 1212 [MPTCP_PM_ADDR_ATTR_FAMILY] = { .type = NLA_U16, }, 1213 [MPTCP_PM_ADDR_ATTR_ID] = { .type = NLA_U8, }, 1214 [MPTCP_PM_ADDR_ATTR_ADDR4] = { .type = NLA_U32, }, 1215 [MPTCP_PM_ADDR_ATTR_ADDR6] = 1216 NLA_POLICY_EXACT_LEN(sizeof(struct in6_addr)), 1217 [MPTCP_PM_ADDR_ATTR_PORT] = { .type = NLA_U16 }, 1218 [MPTCP_PM_ADDR_ATTR_FLAGS] = { .type = NLA_U32 }, 1219 [MPTCP_PM_ADDR_ATTR_IF_IDX] = { .type = NLA_S32 }, 1220 }; 1221 1222 static const struct nla_policy mptcp_pm_policy[MPTCP_PM_ATTR_MAX + 1] = { 1223 [MPTCP_PM_ATTR_ADDR] = 1224 NLA_POLICY_NESTED(mptcp_pm_addr_policy), 1225 [MPTCP_PM_ATTR_RCV_ADD_ADDRS] = { .type = NLA_U32, }, 1226 [MPTCP_PM_ATTR_SUBFLOWS] = { .type = NLA_U32, }, 1227 [MPTCP_PM_ATTR_TOKEN] = { .type = NLA_U32, }, 1228 [MPTCP_PM_ATTR_LOC_ID] = { .type = NLA_U8, }, 1229 [MPTCP_PM_ATTR_ADDR_REMOTE] = 1230 NLA_POLICY_NESTED(mptcp_pm_addr_policy), 1231 }; 1232 1233 void mptcp_pm_nl_subflow_chk_stale(const struct mptcp_sock *msk, struct sock *ssk) 1234 { 1235 struct mptcp_subflow_context *iter, *subflow = mptcp_subflow_ctx(ssk); 1236 struct sock *sk = (struct sock *)msk; 1237 unsigned int active_max_loss_cnt; 1238 struct net *net = sock_net(sk); 1239 unsigned int stale_loss_cnt; 1240 bool slow; 1241 1242 stale_loss_cnt = mptcp_stale_loss_cnt(net); 1243 if (subflow->stale || !stale_loss_cnt || subflow->stale_count <= stale_loss_cnt) 1244 return; 1245 1246 /* look for another available subflow not in loss state */ 1247 active_max_loss_cnt = max_t(int, stale_loss_cnt - 1, 1); 1248 mptcp_for_each_subflow(msk, iter) { 1249 if (iter != subflow && mptcp_subflow_active(iter) && 1250 iter->stale_count < active_max_loss_cnt) { 1251 /* we have some alternatives, try to mark this subflow as idle ...*/ 1252 slow = lock_sock_fast(ssk); 1253 if (!tcp_rtx_and_write_queues_empty(ssk)) { 1254 subflow->stale = 1; 1255 __mptcp_retransmit_pending_data(sk); 1256 MPTCP_INC_STATS(net, MPTCP_MIB_SUBFLOWSTALE); 1257 } 1258 unlock_sock_fast(ssk, slow); 1259 1260 /* always try to push the pending data regardless of re-injections: 1261 * we can possibly use backup subflows now, and subflow selection 1262 * is cheap under the msk socket lock 1263 */ 1264 __mptcp_push_pending(sk, 0); 1265 return; 1266 } 1267 } 1268 } 1269 1270 static int mptcp_pm_family_to_addr(int family) 1271 { 1272 #if IS_ENABLED(CONFIG_MPTCP_IPV6) 1273 if (family == AF_INET6) 1274 return MPTCP_PM_ADDR_ATTR_ADDR6; 1275 #endif 1276 return MPTCP_PM_ADDR_ATTR_ADDR4; 1277 } 1278 1279 static int mptcp_pm_parse_pm_addr_attr(struct nlattr *tb[], 1280 const struct nlattr *attr, 1281 struct genl_info *info, 1282 struct mptcp_addr_info *addr, 1283 bool require_family) 1284 { 1285 int err, addr_addr; 1286 1287 if (!attr) { 1288 GENL_SET_ERR_MSG(info, "missing address info"); 1289 return -EINVAL; 1290 } 1291 1292 /* no validation needed - was already done via nested policy */ 1293 err = nla_parse_nested_deprecated(tb, MPTCP_PM_ADDR_ATTR_MAX, attr, 1294 mptcp_pm_addr_policy, info->extack); 1295 if (err) 1296 return err; 1297 1298 if (tb[MPTCP_PM_ADDR_ATTR_ID]) 1299 addr->id = nla_get_u8(tb[MPTCP_PM_ADDR_ATTR_ID]); 1300 1301 if (!tb[MPTCP_PM_ADDR_ATTR_FAMILY]) { 1302 if (!require_family) 1303 return 0; 1304 1305 NL_SET_ERR_MSG_ATTR(info->extack, attr, 1306 "missing family"); 1307 return -EINVAL; 1308 } 1309 1310 addr->family = nla_get_u16(tb[MPTCP_PM_ADDR_ATTR_FAMILY]); 1311 if (addr->family != AF_INET 1312 #if IS_ENABLED(CONFIG_MPTCP_IPV6) 1313 && addr->family != AF_INET6 1314 #endif 1315 ) { 1316 NL_SET_ERR_MSG_ATTR(info->extack, attr, 1317 "unknown address family"); 1318 return -EINVAL; 1319 } 1320 addr_addr = mptcp_pm_family_to_addr(addr->family); 1321 if (!tb[addr_addr]) { 1322 NL_SET_ERR_MSG_ATTR(info->extack, attr, 1323 "missing address data"); 1324 return -EINVAL; 1325 } 1326 1327 #if IS_ENABLED(CONFIG_MPTCP_IPV6) 1328 if (addr->family == AF_INET6) 1329 addr->addr6 = nla_get_in6_addr(tb[addr_addr]); 1330 else 1331 #endif 1332 addr->addr.s_addr = nla_get_in_addr(tb[addr_addr]); 1333 1334 if (tb[MPTCP_PM_ADDR_ATTR_PORT]) 1335 addr->port = htons(nla_get_u16(tb[MPTCP_PM_ADDR_ATTR_PORT])); 1336 1337 return 0; 1338 } 1339 1340 int mptcp_pm_parse_addr(struct nlattr *attr, struct genl_info *info, 1341 struct mptcp_addr_info *addr) 1342 { 1343 struct nlattr *tb[MPTCP_PM_ADDR_ATTR_MAX + 1]; 1344 1345 memset(addr, 0, sizeof(*addr)); 1346 1347 return mptcp_pm_parse_pm_addr_attr(tb, attr, info, addr, true); 1348 } 1349 1350 int mptcp_pm_parse_entry(struct nlattr *attr, struct genl_info *info, 1351 bool require_family, 1352 struct mptcp_pm_addr_entry *entry) 1353 { 1354 struct nlattr *tb[MPTCP_PM_ADDR_ATTR_MAX + 1]; 1355 int err; 1356 1357 memset(entry, 0, sizeof(*entry)); 1358 1359 err = mptcp_pm_parse_pm_addr_attr(tb, attr, info, &entry->addr, require_family); 1360 if (err) 1361 return err; 1362 1363 if (tb[MPTCP_PM_ADDR_ATTR_IF_IDX]) { 1364 u32 val = nla_get_s32(tb[MPTCP_PM_ADDR_ATTR_IF_IDX]); 1365 1366 entry->ifindex = val; 1367 } 1368 1369 if (tb[MPTCP_PM_ADDR_ATTR_FLAGS]) 1370 entry->flags = nla_get_u32(tb[MPTCP_PM_ADDR_ATTR_FLAGS]); 1371 1372 if (tb[MPTCP_PM_ADDR_ATTR_PORT]) 1373 entry->addr.port = htons(nla_get_u16(tb[MPTCP_PM_ADDR_ATTR_PORT])); 1374 1375 return 0; 1376 } 1377 1378 static struct pm_nl_pernet *genl_info_pm_nl(struct genl_info *info) 1379 { 1380 return pm_nl_get_pernet(genl_info_net(info)); 1381 } 1382 1383 static int mptcp_nl_add_subflow_or_signal_addr(struct net *net, 1384 struct mptcp_addr_info *addr) 1385 { 1386 struct mptcp_sock *msk; 1387 long s_slot = 0, s_num = 0; 1388 1389 while ((msk = mptcp_token_iter_next(net, &s_slot, &s_num)) != NULL) { 1390 struct sock *sk = (struct sock *)msk; 1391 struct mptcp_addr_info mpc_addr; 1392 1393 if (!READ_ONCE(msk->fully_established) || 1394 mptcp_pm_is_userspace(msk)) 1395 goto next; 1396 1397 /* if the endp linked to the init sf is re-added with a != ID */ 1398 mptcp_local_address((struct sock_common *)msk, &mpc_addr); 1399 1400 lock_sock(sk); 1401 spin_lock_bh(&msk->pm.lock); 1402 if (mptcp_addresses_equal(addr, &mpc_addr, addr->port)) 1403 msk->mpc_endpoint_id = addr->id; 1404 mptcp_pm_create_subflow_or_signal_addr(msk); 1405 spin_unlock_bh(&msk->pm.lock); 1406 release_sock(sk); 1407 1408 next: 1409 sock_put(sk); 1410 cond_resched(); 1411 } 1412 1413 return 0; 1414 } 1415 1416 static bool mptcp_pm_has_addr_attr_id(const struct nlattr *attr, 1417 struct genl_info *info) 1418 { 1419 struct nlattr *tb[MPTCP_PM_ADDR_ATTR_MAX + 1]; 1420 1421 if (!nla_parse_nested_deprecated(tb, MPTCP_PM_ADDR_ATTR_MAX, attr, 1422 mptcp_pm_addr_policy, info->extack) && 1423 tb[MPTCP_PM_ADDR_ATTR_ID]) 1424 return true; 1425 return false; 1426 } 1427 1428 static int mptcp_nl_cmd_add_addr(struct sk_buff *skb, struct genl_info *info) 1429 { 1430 struct nlattr *attr = info->attrs[MPTCP_PM_ATTR_ADDR]; 1431 struct pm_nl_pernet *pernet = genl_info_pm_nl(info); 1432 struct mptcp_pm_addr_entry addr, *entry; 1433 int ret; 1434 1435 ret = mptcp_pm_parse_entry(attr, info, true, &addr); 1436 if (ret < 0) 1437 return ret; 1438 1439 if (addr.addr.port && !address_use_port(&addr)) { 1440 GENL_SET_ERR_MSG(info, "flags must have signal and not subflow when using port"); 1441 return -EINVAL; 1442 } 1443 1444 if (addr.flags & MPTCP_PM_ADDR_FLAG_SIGNAL && 1445 addr.flags & MPTCP_PM_ADDR_FLAG_FULLMESH) { 1446 GENL_SET_ERR_MSG(info, "flags mustn't have both signal and fullmesh"); 1447 return -EINVAL; 1448 } 1449 1450 if (addr.flags & MPTCP_PM_ADDR_FLAG_IMPLICIT) { 1451 GENL_SET_ERR_MSG(info, "can't create IMPLICIT endpoint"); 1452 return -EINVAL; 1453 } 1454 1455 entry = kzalloc(sizeof(*entry), GFP_KERNEL_ACCOUNT); 1456 if (!entry) { 1457 GENL_SET_ERR_MSG(info, "can't allocate addr"); 1458 return -ENOMEM; 1459 } 1460 1461 *entry = addr; 1462 if (entry->addr.port) { 1463 ret = mptcp_pm_nl_create_listen_socket(skb->sk, entry); 1464 if (ret) { 1465 GENL_SET_ERR_MSG_FMT(info, "create listen socket error: %d", ret); 1466 goto out_free; 1467 } 1468 } 1469 ret = mptcp_pm_nl_append_new_local_addr(pernet, entry, 1470 !mptcp_pm_has_addr_attr_id(attr, info), 1471 true); 1472 if (ret < 0) { 1473 GENL_SET_ERR_MSG_FMT(info, "too many addresses or duplicate one: %d", ret); 1474 goto out_free; 1475 } 1476 1477 mptcp_nl_add_subflow_or_signal_addr(sock_net(skb->sk), &entry->addr); 1478 return 0; 1479 1480 out_free: 1481 __mptcp_pm_release_addr_entry(entry); 1482 return ret; 1483 } 1484 1485 int mptcp_pm_nl_get_flags_and_ifindex_by_id(struct mptcp_sock *msk, unsigned int id, 1486 u8 *flags, int *ifindex) 1487 { 1488 struct mptcp_pm_addr_entry *entry; 1489 struct sock *sk = (struct sock *)msk; 1490 struct net *net = sock_net(sk); 1491 1492 /* No entries with ID 0 */ 1493 if (id == 0) 1494 return 0; 1495 1496 rcu_read_lock(); 1497 entry = __lookup_addr_by_id(pm_nl_get_pernet(net), id); 1498 if (entry) { 1499 *flags = entry->flags; 1500 *ifindex = entry->ifindex; 1501 } 1502 rcu_read_unlock(); 1503 1504 return 0; 1505 } 1506 1507 static bool remove_anno_list_by_saddr(struct mptcp_sock *msk, 1508 const struct mptcp_addr_info *addr) 1509 { 1510 struct mptcp_pm_add_entry *entry; 1511 1512 entry = mptcp_pm_del_add_timer(msk, addr, false); 1513 if (entry) { 1514 kfree(entry); 1515 return true; 1516 } 1517 1518 return false; 1519 } 1520 1521 static u8 mptcp_endp_get_local_id(struct mptcp_sock *msk, 1522 const struct mptcp_addr_info *addr) 1523 { 1524 return msk->mpc_endpoint_id == addr->id ? 0 : addr->id; 1525 } 1526 1527 static bool mptcp_pm_remove_anno_addr(struct mptcp_sock *msk, 1528 const struct mptcp_addr_info *addr, 1529 bool force) 1530 { 1531 struct mptcp_rm_list list = { .nr = 0 }; 1532 bool ret; 1533 1534 list.ids[list.nr++] = mptcp_endp_get_local_id(msk, addr); 1535 1536 ret = remove_anno_list_by_saddr(msk, addr); 1537 if (ret || force) { 1538 spin_lock_bh(&msk->pm.lock); 1539 if (ret) { 1540 __set_bit(addr->id, msk->pm.id_avail_bitmap); 1541 msk->pm.add_addr_signaled--; 1542 } 1543 mptcp_pm_remove_addr(msk, &list); 1544 spin_unlock_bh(&msk->pm.lock); 1545 } 1546 return ret; 1547 } 1548 1549 static void __mark_subflow_endp_available(struct mptcp_sock *msk, u8 id) 1550 { 1551 /* If it was marked as used, and not ID 0, decrement local_addr_used */ 1552 if (!__test_and_set_bit(id ? : msk->mpc_endpoint_id, msk->pm.id_avail_bitmap) && 1553 id && !WARN_ON_ONCE(msk->pm.local_addr_used == 0)) 1554 msk->pm.local_addr_used--; 1555 } 1556 1557 static int mptcp_nl_remove_subflow_and_signal_addr(struct net *net, 1558 const struct mptcp_pm_addr_entry *entry) 1559 { 1560 const struct mptcp_addr_info *addr = &entry->addr; 1561 struct mptcp_rm_list list = { .nr = 1 }; 1562 long s_slot = 0, s_num = 0; 1563 struct mptcp_sock *msk; 1564 1565 pr_debug("remove_id=%d\n", addr->id); 1566 1567 while ((msk = mptcp_token_iter_next(net, &s_slot, &s_num)) != NULL) { 1568 struct sock *sk = (struct sock *)msk; 1569 bool remove_subflow; 1570 1571 if (mptcp_pm_is_userspace(msk)) 1572 goto next; 1573 1574 lock_sock(sk); 1575 remove_subflow = lookup_subflow_by_saddr(&msk->conn_list, addr); 1576 mptcp_pm_remove_anno_addr(msk, addr, remove_subflow && 1577 !(entry->flags & MPTCP_PM_ADDR_FLAG_IMPLICIT)); 1578 1579 list.ids[0] = mptcp_endp_get_local_id(msk, addr); 1580 if (remove_subflow) { 1581 spin_lock_bh(&msk->pm.lock); 1582 mptcp_pm_nl_rm_subflow_received(msk, &list); 1583 spin_unlock_bh(&msk->pm.lock); 1584 } 1585 1586 if (entry->flags & MPTCP_PM_ADDR_FLAG_SUBFLOW) { 1587 spin_lock_bh(&msk->pm.lock); 1588 __mark_subflow_endp_available(msk, list.ids[0]); 1589 spin_unlock_bh(&msk->pm.lock); 1590 } 1591 1592 if (msk->mpc_endpoint_id == entry->addr.id) 1593 msk->mpc_endpoint_id = 0; 1594 release_sock(sk); 1595 1596 next: 1597 sock_put(sk); 1598 cond_resched(); 1599 } 1600 1601 return 0; 1602 } 1603 1604 static int mptcp_nl_remove_id_zero_address(struct net *net, 1605 struct mptcp_addr_info *addr) 1606 { 1607 struct mptcp_rm_list list = { .nr = 0 }; 1608 long s_slot = 0, s_num = 0; 1609 struct mptcp_sock *msk; 1610 1611 list.ids[list.nr++] = 0; 1612 1613 while ((msk = mptcp_token_iter_next(net, &s_slot, &s_num)) != NULL) { 1614 struct sock *sk = (struct sock *)msk; 1615 struct mptcp_addr_info msk_local; 1616 1617 if (list_empty(&msk->conn_list) || mptcp_pm_is_userspace(msk)) 1618 goto next; 1619 1620 mptcp_local_address((struct sock_common *)msk, &msk_local); 1621 if (!mptcp_addresses_equal(&msk_local, addr, addr->port)) 1622 goto next; 1623 1624 lock_sock(sk); 1625 spin_lock_bh(&msk->pm.lock); 1626 mptcp_pm_remove_addr(msk, &list); 1627 mptcp_pm_nl_rm_subflow_received(msk, &list); 1628 __mark_subflow_endp_available(msk, 0); 1629 spin_unlock_bh(&msk->pm.lock); 1630 release_sock(sk); 1631 1632 next: 1633 sock_put(sk); 1634 cond_resched(); 1635 } 1636 1637 return 0; 1638 } 1639 1640 static int mptcp_nl_cmd_del_addr(struct sk_buff *skb, struct genl_info *info) 1641 { 1642 struct nlattr *attr = info->attrs[MPTCP_PM_ATTR_ADDR]; 1643 struct pm_nl_pernet *pernet = genl_info_pm_nl(info); 1644 struct mptcp_pm_addr_entry addr, *entry; 1645 unsigned int addr_max; 1646 int ret; 1647 1648 ret = mptcp_pm_parse_entry(attr, info, false, &addr); 1649 if (ret < 0) 1650 return ret; 1651 1652 /* the zero id address is special: the first address used by the msk 1653 * always gets such an id, so different subflows can have different zero 1654 * id addresses. Additionally zero id is not accounted for in id_bitmap. 1655 * Let's use an 'mptcp_rm_list' instead of the common remove code. 1656 */ 1657 if (addr.addr.id == 0) 1658 return mptcp_nl_remove_id_zero_address(sock_net(skb->sk), &addr.addr); 1659 1660 spin_lock_bh(&pernet->lock); 1661 entry = __lookup_addr_by_id(pernet, addr.addr.id); 1662 if (!entry) { 1663 GENL_SET_ERR_MSG(info, "address not found"); 1664 spin_unlock_bh(&pernet->lock); 1665 return -EINVAL; 1666 } 1667 if (entry->flags & MPTCP_PM_ADDR_FLAG_SIGNAL) { 1668 addr_max = pernet->add_addr_signal_max; 1669 WRITE_ONCE(pernet->add_addr_signal_max, addr_max - 1); 1670 } 1671 if (entry->flags & MPTCP_PM_ADDR_FLAG_SUBFLOW) { 1672 addr_max = pernet->local_addr_max; 1673 WRITE_ONCE(pernet->local_addr_max, addr_max - 1); 1674 } 1675 1676 pernet->addrs--; 1677 list_del_rcu(&entry->list); 1678 __clear_bit(entry->addr.id, pernet->id_bitmap); 1679 spin_unlock_bh(&pernet->lock); 1680 1681 mptcp_nl_remove_subflow_and_signal_addr(sock_net(skb->sk), entry); 1682 synchronize_rcu(); 1683 __mptcp_pm_release_addr_entry(entry); 1684 1685 return ret; 1686 } 1687 1688 /* Called from the userspace PM only */ 1689 void mptcp_pm_remove_addrs(struct mptcp_sock *msk, struct list_head *rm_list) 1690 { 1691 struct mptcp_rm_list alist = { .nr = 0 }; 1692 struct mptcp_pm_addr_entry *entry; 1693 int anno_nr = 0; 1694 1695 list_for_each_entry(entry, rm_list, list) { 1696 if (alist.nr >= MPTCP_RM_IDS_MAX) 1697 break; 1698 1699 /* only delete if either announced or matching a subflow */ 1700 if (remove_anno_list_by_saddr(msk, &entry->addr)) 1701 anno_nr++; 1702 else if (!lookup_subflow_by_saddr(&msk->conn_list, 1703 &entry->addr)) 1704 continue; 1705 1706 alist.ids[alist.nr++] = entry->addr.id; 1707 } 1708 1709 if (alist.nr) { 1710 spin_lock_bh(&msk->pm.lock); 1711 msk->pm.add_addr_signaled -= anno_nr; 1712 mptcp_pm_remove_addr(msk, &alist); 1713 spin_unlock_bh(&msk->pm.lock); 1714 } 1715 } 1716 1717 /* Called from the in-kernel PM only */ 1718 static void mptcp_pm_remove_addrs_and_subflows(struct mptcp_sock *msk, 1719 struct list_head *rm_list) 1720 { 1721 struct mptcp_rm_list alist = { .nr = 0 }, slist = { .nr = 0 }; 1722 struct mptcp_pm_addr_entry *entry; 1723 1724 list_for_each_entry(entry, rm_list, list) { 1725 if (slist.nr < MPTCP_RM_IDS_MAX && 1726 lookup_subflow_by_saddr(&msk->conn_list, &entry->addr)) 1727 slist.ids[slist.nr++] = mptcp_endp_get_local_id(msk, &entry->addr); 1728 1729 if (alist.nr < MPTCP_RM_IDS_MAX && 1730 remove_anno_list_by_saddr(msk, &entry->addr)) 1731 alist.ids[alist.nr++] = mptcp_endp_get_local_id(msk, &entry->addr); 1732 } 1733 1734 spin_lock_bh(&msk->pm.lock); 1735 if (alist.nr) { 1736 msk->pm.add_addr_signaled -= alist.nr; 1737 mptcp_pm_remove_addr(msk, &alist); 1738 } 1739 if (slist.nr) 1740 mptcp_pm_nl_rm_subflow_received(msk, &slist); 1741 /* Reset counters: maybe some subflows have been removed before */ 1742 bitmap_fill(msk->pm.id_avail_bitmap, MPTCP_PM_MAX_ADDR_ID + 1); 1743 msk->pm.local_addr_used = 0; 1744 spin_unlock_bh(&msk->pm.lock); 1745 } 1746 1747 static void mptcp_nl_remove_addrs_list(struct net *net, 1748 struct list_head *rm_list) 1749 { 1750 long s_slot = 0, s_num = 0; 1751 struct mptcp_sock *msk; 1752 1753 if (list_empty(rm_list)) 1754 return; 1755 1756 while ((msk = mptcp_token_iter_next(net, &s_slot, &s_num)) != NULL) { 1757 struct sock *sk = (struct sock *)msk; 1758 1759 if (!mptcp_pm_is_userspace(msk)) { 1760 lock_sock(sk); 1761 mptcp_pm_remove_addrs_and_subflows(msk, rm_list); 1762 release_sock(sk); 1763 } 1764 1765 sock_put(sk); 1766 cond_resched(); 1767 } 1768 } 1769 1770 /* caller must ensure the RCU grace period is already elapsed */ 1771 static void __flush_addrs(struct list_head *list) 1772 { 1773 while (!list_empty(list)) { 1774 struct mptcp_pm_addr_entry *cur; 1775 1776 cur = list_entry(list->next, 1777 struct mptcp_pm_addr_entry, list); 1778 list_del_rcu(&cur->list); 1779 __mptcp_pm_release_addr_entry(cur); 1780 } 1781 } 1782 1783 static void __reset_counters(struct pm_nl_pernet *pernet) 1784 { 1785 WRITE_ONCE(pernet->add_addr_signal_max, 0); 1786 WRITE_ONCE(pernet->add_addr_accept_max, 0); 1787 WRITE_ONCE(pernet->local_addr_max, 0); 1788 pernet->addrs = 0; 1789 } 1790 1791 static int mptcp_nl_cmd_flush_addrs(struct sk_buff *skb, struct genl_info *info) 1792 { 1793 struct pm_nl_pernet *pernet = genl_info_pm_nl(info); 1794 LIST_HEAD(free_list); 1795 1796 spin_lock_bh(&pernet->lock); 1797 list_splice_init(&pernet->local_addr_list, &free_list); 1798 __reset_counters(pernet); 1799 pernet->next_id = 1; 1800 bitmap_zero(pernet->id_bitmap, MPTCP_PM_MAX_ADDR_ID + 1); 1801 spin_unlock_bh(&pernet->lock); 1802 mptcp_nl_remove_addrs_list(sock_net(skb->sk), &free_list); 1803 synchronize_rcu(); 1804 __flush_addrs(&free_list); 1805 return 0; 1806 } 1807 1808 static int mptcp_nl_fill_addr(struct sk_buff *skb, 1809 struct mptcp_pm_addr_entry *entry) 1810 { 1811 struct mptcp_addr_info *addr = &entry->addr; 1812 struct nlattr *attr; 1813 1814 attr = nla_nest_start(skb, MPTCP_PM_ATTR_ADDR); 1815 if (!attr) 1816 return -EMSGSIZE; 1817 1818 if (nla_put_u16(skb, MPTCP_PM_ADDR_ATTR_FAMILY, addr->family)) 1819 goto nla_put_failure; 1820 if (nla_put_u16(skb, MPTCP_PM_ADDR_ATTR_PORT, ntohs(addr->port))) 1821 goto nla_put_failure; 1822 if (nla_put_u8(skb, MPTCP_PM_ADDR_ATTR_ID, addr->id)) 1823 goto nla_put_failure; 1824 if (nla_put_u32(skb, MPTCP_PM_ADDR_ATTR_FLAGS, entry->flags)) 1825 goto nla_put_failure; 1826 if (entry->ifindex && 1827 nla_put_s32(skb, MPTCP_PM_ADDR_ATTR_IF_IDX, entry->ifindex)) 1828 goto nla_put_failure; 1829 1830 if (addr->family == AF_INET && 1831 nla_put_in_addr(skb, MPTCP_PM_ADDR_ATTR_ADDR4, 1832 addr->addr.s_addr)) 1833 goto nla_put_failure; 1834 #if IS_ENABLED(CONFIG_MPTCP_IPV6) 1835 else if (addr->family == AF_INET6 && 1836 nla_put_in6_addr(skb, MPTCP_PM_ADDR_ATTR_ADDR6, &addr->addr6)) 1837 goto nla_put_failure; 1838 #endif 1839 nla_nest_end(skb, attr); 1840 return 0; 1841 1842 nla_put_failure: 1843 nla_nest_cancel(skb, attr); 1844 return -EMSGSIZE; 1845 } 1846 1847 static int mptcp_nl_cmd_get_addr(struct sk_buff *skb, struct genl_info *info) 1848 { 1849 struct nlattr *attr = info->attrs[MPTCP_PM_ATTR_ADDR]; 1850 struct pm_nl_pernet *pernet = genl_info_pm_nl(info); 1851 struct mptcp_pm_addr_entry addr, *entry; 1852 struct sk_buff *msg; 1853 void *reply; 1854 int ret; 1855 1856 ret = mptcp_pm_parse_entry(attr, info, false, &addr); 1857 if (ret < 0) 1858 return ret; 1859 1860 msg = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL); 1861 if (!msg) 1862 return -ENOMEM; 1863 1864 reply = genlmsg_put_reply(msg, info, &mptcp_genl_family, 0, 1865 info->genlhdr->cmd); 1866 if (!reply) { 1867 GENL_SET_ERR_MSG(info, "not enough space in Netlink message"); 1868 ret = -EMSGSIZE; 1869 goto fail; 1870 } 1871 1872 spin_lock_bh(&pernet->lock); 1873 entry = __lookup_addr_by_id(pernet, addr.addr.id); 1874 if (!entry) { 1875 GENL_SET_ERR_MSG(info, "address not found"); 1876 ret = -EINVAL; 1877 goto unlock_fail; 1878 } 1879 1880 ret = mptcp_nl_fill_addr(msg, entry); 1881 if (ret) 1882 goto unlock_fail; 1883 1884 genlmsg_end(msg, reply); 1885 ret = genlmsg_reply(msg, info); 1886 spin_unlock_bh(&pernet->lock); 1887 return ret; 1888 1889 unlock_fail: 1890 spin_unlock_bh(&pernet->lock); 1891 1892 fail: 1893 nlmsg_free(msg); 1894 return ret; 1895 } 1896 1897 static int mptcp_nl_cmd_dump_addrs(struct sk_buff *msg, 1898 struct netlink_callback *cb) 1899 { 1900 struct net *net = sock_net(msg->sk); 1901 struct mptcp_pm_addr_entry *entry; 1902 struct pm_nl_pernet *pernet; 1903 int id = cb->args[0]; 1904 void *hdr; 1905 int i; 1906 1907 pernet = pm_nl_get_pernet(net); 1908 1909 spin_lock_bh(&pernet->lock); 1910 for (i = id; i < MPTCP_PM_MAX_ADDR_ID + 1; i++) { 1911 if (test_bit(i, pernet->id_bitmap)) { 1912 entry = __lookup_addr_by_id(pernet, i); 1913 if (!entry) 1914 break; 1915 1916 if (entry->addr.id <= id) 1917 continue; 1918 1919 hdr = genlmsg_put(msg, NETLINK_CB(cb->skb).portid, 1920 cb->nlh->nlmsg_seq, &mptcp_genl_family, 1921 NLM_F_MULTI, MPTCP_PM_CMD_GET_ADDR); 1922 if (!hdr) 1923 break; 1924 1925 if (mptcp_nl_fill_addr(msg, entry) < 0) { 1926 genlmsg_cancel(msg, hdr); 1927 break; 1928 } 1929 1930 id = entry->addr.id; 1931 genlmsg_end(msg, hdr); 1932 } 1933 } 1934 spin_unlock_bh(&pernet->lock); 1935 1936 cb->args[0] = id; 1937 return msg->len; 1938 } 1939 1940 static int parse_limit(struct genl_info *info, int id, unsigned int *limit) 1941 { 1942 struct nlattr *attr = info->attrs[id]; 1943 1944 if (!attr) 1945 return 0; 1946 1947 *limit = nla_get_u32(attr); 1948 if (*limit > MPTCP_PM_ADDR_MAX) { 1949 GENL_SET_ERR_MSG(info, "limit greater than maximum"); 1950 return -EINVAL; 1951 } 1952 return 0; 1953 } 1954 1955 static int 1956 mptcp_nl_cmd_set_limits(struct sk_buff *skb, struct genl_info *info) 1957 { 1958 struct pm_nl_pernet *pernet = genl_info_pm_nl(info); 1959 unsigned int rcv_addrs, subflows; 1960 int ret; 1961 1962 spin_lock_bh(&pernet->lock); 1963 rcv_addrs = pernet->add_addr_accept_max; 1964 ret = parse_limit(info, MPTCP_PM_ATTR_RCV_ADD_ADDRS, &rcv_addrs); 1965 if (ret) 1966 goto unlock; 1967 1968 subflows = pernet->subflows_max; 1969 ret = parse_limit(info, MPTCP_PM_ATTR_SUBFLOWS, &subflows); 1970 if (ret) 1971 goto unlock; 1972 1973 WRITE_ONCE(pernet->add_addr_accept_max, rcv_addrs); 1974 WRITE_ONCE(pernet->subflows_max, subflows); 1975 1976 unlock: 1977 spin_unlock_bh(&pernet->lock); 1978 return ret; 1979 } 1980 1981 static int 1982 mptcp_nl_cmd_get_limits(struct sk_buff *skb, struct genl_info *info) 1983 { 1984 struct pm_nl_pernet *pernet = genl_info_pm_nl(info); 1985 struct sk_buff *msg; 1986 void *reply; 1987 1988 msg = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL); 1989 if (!msg) 1990 return -ENOMEM; 1991 1992 reply = genlmsg_put_reply(msg, info, &mptcp_genl_family, 0, 1993 MPTCP_PM_CMD_GET_LIMITS); 1994 if (!reply) 1995 goto fail; 1996 1997 if (nla_put_u32(msg, MPTCP_PM_ATTR_RCV_ADD_ADDRS, 1998 READ_ONCE(pernet->add_addr_accept_max))) 1999 goto fail; 2000 2001 if (nla_put_u32(msg, MPTCP_PM_ATTR_SUBFLOWS, 2002 READ_ONCE(pernet->subflows_max))) 2003 goto fail; 2004 2005 genlmsg_end(msg, reply); 2006 return genlmsg_reply(msg, info); 2007 2008 fail: 2009 GENL_SET_ERR_MSG(info, "not enough space in Netlink message"); 2010 nlmsg_free(msg); 2011 return -EMSGSIZE; 2012 } 2013 2014 static void mptcp_pm_nl_fullmesh(struct mptcp_sock *msk, 2015 struct mptcp_addr_info *addr) 2016 { 2017 struct mptcp_rm_list list = { .nr = 0 }; 2018 2019 list.ids[list.nr++] = mptcp_endp_get_local_id(msk, addr); 2020 2021 spin_lock_bh(&msk->pm.lock); 2022 mptcp_pm_nl_rm_subflow_received(msk, &list); 2023 __mark_subflow_endp_available(msk, list.ids[0]); 2024 mptcp_pm_create_subflow_or_signal_addr(msk); 2025 spin_unlock_bh(&msk->pm.lock); 2026 } 2027 2028 static int mptcp_nl_set_flags(struct net *net, 2029 struct mptcp_addr_info *addr, 2030 u8 bkup, u8 changed) 2031 { 2032 long s_slot = 0, s_num = 0; 2033 struct mptcp_sock *msk; 2034 int ret = -EINVAL; 2035 2036 while ((msk = mptcp_token_iter_next(net, &s_slot, &s_num)) != NULL) { 2037 struct sock *sk = (struct sock *)msk; 2038 2039 if (list_empty(&msk->conn_list) || mptcp_pm_is_userspace(msk)) 2040 goto next; 2041 2042 lock_sock(sk); 2043 if (changed & MPTCP_PM_ADDR_FLAG_BACKUP) 2044 ret = mptcp_pm_nl_mp_prio_send_ack(msk, addr, NULL, bkup); 2045 if (changed & MPTCP_PM_ADDR_FLAG_FULLMESH) 2046 mptcp_pm_nl_fullmesh(msk, addr); 2047 release_sock(sk); 2048 2049 next: 2050 sock_put(sk); 2051 cond_resched(); 2052 } 2053 2054 return ret; 2055 } 2056 2057 int mptcp_pm_nl_set_flags(struct net *net, struct mptcp_pm_addr_entry *addr, u8 bkup) 2058 { 2059 struct pm_nl_pernet *pernet = pm_nl_get_pernet(net); 2060 u8 changed, mask = MPTCP_PM_ADDR_FLAG_BACKUP | 2061 MPTCP_PM_ADDR_FLAG_FULLMESH; 2062 struct mptcp_pm_addr_entry *entry; 2063 u8 lookup_by_id = 0; 2064 2065 if (addr->addr.family == AF_UNSPEC) { 2066 lookup_by_id = 1; 2067 if (!addr->addr.id) 2068 return -EOPNOTSUPP; 2069 } 2070 2071 spin_lock_bh(&pernet->lock); 2072 entry = lookup_by_id ? __lookup_addr_by_id(pernet, addr->addr.id) : 2073 __lookup_addr(pernet, &addr->addr); 2074 if (!entry) { 2075 spin_unlock_bh(&pernet->lock); 2076 return -EINVAL; 2077 } 2078 if ((addr->flags & MPTCP_PM_ADDR_FLAG_FULLMESH) && 2079 (entry->flags & (MPTCP_PM_ADDR_FLAG_SIGNAL | 2080 MPTCP_PM_ADDR_FLAG_IMPLICIT))) { 2081 spin_unlock_bh(&pernet->lock); 2082 return -EINVAL; 2083 } 2084 2085 changed = (addr->flags ^ entry->flags) & mask; 2086 entry->flags = (entry->flags & ~mask) | (addr->flags & mask); 2087 *addr = *entry; 2088 spin_unlock_bh(&pernet->lock); 2089 2090 mptcp_nl_set_flags(net, &addr->addr, bkup, changed); 2091 return 0; 2092 } 2093 2094 static int mptcp_nl_cmd_set_flags(struct sk_buff *skb, struct genl_info *info) 2095 { 2096 struct mptcp_pm_addr_entry remote = { .addr = { .family = AF_UNSPEC }, }; 2097 struct mptcp_pm_addr_entry addr = { .addr = { .family = AF_UNSPEC }, }; 2098 struct nlattr *attr_rem = info->attrs[MPTCP_PM_ATTR_ADDR_REMOTE]; 2099 struct nlattr *token = info->attrs[MPTCP_PM_ATTR_TOKEN]; 2100 struct nlattr *attr = info->attrs[MPTCP_PM_ATTR_ADDR]; 2101 struct net *net = sock_net(skb->sk); 2102 u8 bkup = 0; 2103 int ret; 2104 2105 ret = mptcp_pm_parse_entry(attr, info, false, &addr); 2106 if (ret < 0) 2107 return ret; 2108 2109 if (attr_rem) { 2110 ret = mptcp_pm_parse_entry(attr_rem, info, false, &remote); 2111 if (ret < 0) 2112 return ret; 2113 } 2114 2115 if (addr.flags & MPTCP_PM_ADDR_FLAG_BACKUP) 2116 bkup = 1; 2117 2118 return mptcp_pm_set_flags(net, token, &addr, &remote, bkup); 2119 } 2120 2121 static void mptcp_nl_mcast_send(struct net *net, struct sk_buff *nlskb, gfp_t gfp) 2122 { 2123 genlmsg_multicast_netns(&mptcp_genl_family, net, 2124 nlskb, 0, MPTCP_PM_EV_GRP_OFFSET, gfp); 2125 } 2126 2127 bool mptcp_userspace_pm_active(const struct mptcp_sock *msk) 2128 { 2129 return genl_has_listeners(&mptcp_genl_family, 2130 sock_net((const struct sock *)msk), 2131 MPTCP_PM_EV_GRP_OFFSET); 2132 } 2133 2134 static int mptcp_event_add_subflow(struct sk_buff *skb, const struct sock *ssk) 2135 { 2136 const struct inet_sock *issk = inet_sk(ssk); 2137 const struct mptcp_subflow_context *sf; 2138 2139 if (nla_put_u16(skb, MPTCP_ATTR_FAMILY, ssk->sk_family)) 2140 return -EMSGSIZE; 2141 2142 switch (ssk->sk_family) { 2143 case AF_INET: 2144 if (nla_put_in_addr(skb, MPTCP_ATTR_SADDR4, issk->inet_saddr)) 2145 return -EMSGSIZE; 2146 if (nla_put_in_addr(skb, MPTCP_ATTR_DADDR4, issk->inet_daddr)) 2147 return -EMSGSIZE; 2148 break; 2149 #if IS_ENABLED(CONFIG_MPTCP_IPV6) 2150 case AF_INET6: { 2151 const struct ipv6_pinfo *np = inet6_sk(ssk); 2152 2153 if (nla_put_in6_addr(skb, MPTCP_ATTR_SADDR6, &np->saddr)) 2154 return -EMSGSIZE; 2155 if (nla_put_in6_addr(skb, MPTCP_ATTR_DADDR6, &ssk->sk_v6_daddr)) 2156 return -EMSGSIZE; 2157 break; 2158 } 2159 #endif 2160 default: 2161 WARN_ON_ONCE(1); 2162 return -EMSGSIZE; 2163 } 2164 2165 if (nla_put_be16(skb, MPTCP_ATTR_SPORT, issk->inet_sport)) 2166 return -EMSGSIZE; 2167 if (nla_put_be16(skb, MPTCP_ATTR_DPORT, issk->inet_dport)) 2168 return -EMSGSIZE; 2169 2170 sf = mptcp_subflow_ctx(ssk); 2171 if (WARN_ON_ONCE(!sf)) 2172 return -EINVAL; 2173 2174 if (nla_put_u8(skb, MPTCP_ATTR_LOC_ID, subflow_get_local_id(sf))) 2175 return -EMSGSIZE; 2176 2177 if (nla_put_u8(skb, MPTCP_ATTR_REM_ID, sf->remote_id)) 2178 return -EMSGSIZE; 2179 2180 return 0; 2181 } 2182 2183 static int mptcp_event_put_token_and_ssk(struct sk_buff *skb, 2184 const struct mptcp_sock *msk, 2185 const struct sock *ssk) 2186 { 2187 const struct sock *sk = (const struct sock *)msk; 2188 const struct mptcp_subflow_context *sf; 2189 u8 sk_err; 2190 2191 if (nla_put_u32(skb, MPTCP_ATTR_TOKEN, msk->token)) 2192 return -EMSGSIZE; 2193 2194 if (mptcp_event_add_subflow(skb, ssk)) 2195 return -EMSGSIZE; 2196 2197 sf = mptcp_subflow_ctx(ssk); 2198 if (WARN_ON_ONCE(!sf)) 2199 return -EINVAL; 2200 2201 if (nla_put_u8(skb, MPTCP_ATTR_BACKUP, sf->backup)) 2202 return -EMSGSIZE; 2203 2204 if (ssk->sk_bound_dev_if && 2205 nla_put_s32(skb, MPTCP_ATTR_IF_IDX, ssk->sk_bound_dev_if)) 2206 return -EMSGSIZE; 2207 2208 sk_err = READ_ONCE(ssk->sk_err); 2209 if (sk_err && sk->sk_state == TCP_ESTABLISHED && 2210 nla_put_u8(skb, MPTCP_ATTR_ERROR, sk_err)) 2211 return -EMSGSIZE; 2212 2213 return 0; 2214 } 2215 2216 static int mptcp_event_sub_established(struct sk_buff *skb, 2217 const struct mptcp_sock *msk, 2218 const struct sock *ssk) 2219 { 2220 return mptcp_event_put_token_and_ssk(skb, msk, ssk); 2221 } 2222 2223 static int mptcp_event_sub_closed(struct sk_buff *skb, 2224 const struct mptcp_sock *msk, 2225 const struct sock *ssk) 2226 { 2227 const struct mptcp_subflow_context *sf; 2228 2229 if (mptcp_event_put_token_and_ssk(skb, msk, ssk)) 2230 return -EMSGSIZE; 2231 2232 sf = mptcp_subflow_ctx(ssk); 2233 if (!sf->reset_seen) 2234 return 0; 2235 2236 if (nla_put_u32(skb, MPTCP_ATTR_RESET_REASON, sf->reset_reason)) 2237 return -EMSGSIZE; 2238 2239 if (nla_put_u32(skb, MPTCP_ATTR_RESET_FLAGS, sf->reset_transient)) 2240 return -EMSGSIZE; 2241 2242 return 0; 2243 } 2244 2245 static int mptcp_event_created(struct sk_buff *skb, 2246 const struct mptcp_sock *msk, 2247 const struct sock *ssk) 2248 { 2249 int err = nla_put_u32(skb, MPTCP_ATTR_TOKEN, msk->token); 2250 2251 if (err) 2252 return err; 2253 2254 if (nla_put_u8(skb, MPTCP_ATTR_SERVER_SIDE, READ_ONCE(msk->pm.server_side))) 2255 return -EMSGSIZE; 2256 2257 return mptcp_event_add_subflow(skb, ssk); 2258 } 2259 2260 void mptcp_event_addr_removed(const struct mptcp_sock *msk, uint8_t id) 2261 { 2262 struct net *net = sock_net((const struct sock *)msk); 2263 struct nlmsghdr *nlh; 2264 struct sk_buff *skb; 2265 2266 if (!genl_has_listeners(&mptcp_genl_family, net, MPTCP_PM_EV_GRP_OFFSET)) 2267 return; 2268 2269 skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_ATOMIC); 2270 if (!skb) 2271 return; 2272 2273 nlh = genlmsg_put(skb, 0, 0, &mptcp_genl_family, 0, MPTCP_EVENT_REMOVED); 2274 if (!nlh) 2275 goto nla_put_failure; 2276 2277 if (nla_put_u32(skb, MPTCP_ATTR_TOKEN, msk->token)) 2278 goto nla_put_failure; 2279 2280 if (nla_put_u8(skb, MPTCP_ATTR_REM_ID, id)) 2281 goto nla_put_failure; 2282 2283 genlmsg_end(skb, nlh); 2284 mptcp_nl_mcast_send(net, skb, GFP_ATOMIC); 2285 return; 2286 2287 nla_put_failure: 2288 nlmsg_free(skb); 2289 } 2290 2291 void mptcp_event_addr_announced(const struct sock *ssk, 2292 const struct mptcp_addr_info *info) 2293 { 2294 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk); 2295 struct mptcp_sock *msk = mptcp_sk(subflow->conn); 2296 struct net *net = sock_net(ssk); 2297 struct nlmsghdr *nlh; 2298 struct sk_buff *skb; 2299 2300 if (!genl_has_listeners(&mptcp_genl_family, net, MPTCP_PM_EV_GRP_OFFSET)) 2301 return; 2302 2303 skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_ATOMIC); 2304 if (!skb) 2305 return; 2306 2307 nlh = genlmsg_put(skb, 0, 0, &mptcp_genl_family, 0, 2308 MPTCP_EVENT_ANNOUNCED); 2309 if (!nlh) 2310 goto nla_put_failure; 2311 2312 if (nla_put_u32(skb, MPTCP_ATTR_TOKEN, msk->token)) 2313 goto nla_put_failure; 2314 2315 if (nla_put_u8(skb, MPTCP_ATTR_REM_ID, info->id)) 2316 goto nla_put_failure; 2317 2318 if (nla_put_be16(skb, MPTCP_ATTR_DPORT, 2319 info->port == 0 ? 2320 inet_sk(ssk)->inet_dport : 2321 info->port)) 2322 goto nla_put_failure; 2323 2324 switch (info->family) { 2325 case AF_INET: 2326 if (nla_put_in_addr(skb, MPTCP_ATTR_DADDR4, info->addr.s_addr)) 2327 goto nla_put_failure; 2328 break; 2329 #if IS_ENABLED(CONFIG_MPTCP_IPV6) 2330 case AF_INET6: 2331 if (nla_put_in6_addr(skb, MPTCP_ATTR_DADDR6, &info->addr6)) 2332 goto nla_put_failure; 2333 break; 2334 #endif 2335 default: 2336 WARN_ON_ONCE(1); 2337 goto nla_put_failure; 2338 } 2339 2340 genlmsg_end(skb, nlh); 2341 mptcp_nl_mcast_send(net, skb, GFP_ATOMIC); 2342 return; 2343 2344 nla_put_failure: 2345 nlmsg_free(skb); 2346 } 2347 2348 void mptcp_event_pm_listener(const struct sock *ssk, 2349 enum mptcp_event_type event) 2350 { 2351 const struct inet_sock *issk = inet_sk(ssk); 2352 struct net *net = sock_net(ssk); 2353 struct nlmsghdr *nlh; 2354 struct sk_buff *skb; 2355 2356 if (!genl_has_listeners(&mptcp_genl_family, net, MPTCP_PM_EV_GRP_OFFSET)) 2357 return; 2358 2359 skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL); 2360 if (!skb) 2361 return; 2362 2363 nlh = genlmsg_put(skb, 0, 0, &mptcp_genl_family, 0, event); 2364 if (!nlh) 2365 goto nla_put_failure; 2366 2367 if (nla_put_u16(skb, MPTCP_ATTR_FAMILY, ssk->sk_family)) 2368 goto nla_put_failure; 2369 2370 if (nla_put_be16(skb, MPTCP_ATTR_SPORT, issk->inet_sport)) 2371 goto nla_put_failure; 2372 2373 switch (ssk->sk_family) { 2374 case AF_INET: 2375 if (nla_put_in_addr(skb, MPTCP_ATTR_SADDR4, issk->inet_saddr)) 2376 goto nla_put_failure; 2377 break; 2378 #if IS_ENABLED(CONFIG_MPTCP_IPV6) 2379 case AF_INET6: { 2380 const struct ipv6_pinfo *np = inet6_sk(ssk); 2381 2382 if (nla_put_in6_addr(skb, MPTCP_ATTR_SADDR6, &np->saddr)) 2383 goto nla_put_failure; 2384 break; 2385 } 2386 #endif 2387 default: 2388 WARN_ON_ONCE(1); 2389 goto nla_put_failure; 2390 } 2391 2392 genlmsg_end(skb, nlh); 2393 mptcp_nl_mcast_send(net, skb, GFP_KERNEL); 2394 return; 2395 2396 nla_put_failure: 2397 nlmsg_free(skb); 2398 } 2399 2400 void mptcp_event(enum mptcp_event_type type, const struct mptcp_sock *msk, 2401 const struct sock *ssk, gfp_t gfp) 2402 { 2403 struct net *net = sock_net((const struct sock *)msk); 2404 struct nlmsghdr *nlh; 2405 struct sk_buff *skb; 2406 2407 if (!genl_has_listeners(&mptcp_genl_family, net, MPTCP_PM_EV_GRP_OFFSET)) 2408 return; 2409 2410 skb = nlmsg_new(NLMSG_DEFAULT_SIZE, gfp); 2411 if (!skb) 2412 return; 2413 2414 nlh = genlmsg_put(skb, 0, 0, &mptcp_genl_family, 0, type); 2415 if (!nlh) 2416 goto nla_put_failure; 2417 2418 switch (type) { 2419 case MPTCP_EVENT_UNSPEC: 2420 WARN_ON_ONCE(1); 2421 break; 2422 case MPTCP_EVENT_CREATED: 2423 case MPTCP_EVENT_ESTABLISHED: 2424 if (mptcp_event_created(skb, msk, ssk) < 0) 2425 goto nla_put_failure; 2426 break; 2427 case MPTCP_EVENT_CLOSED: 2428 if (nla_put_u32(skb, MPTCP_ATTR_TOKEN, msk->token) < 0) 2429 goto nla_put_failure; 2430 break; 2431 case MPTCP_EVENT_ANNOUNCED: 2432 case MPTCP_EVENT_REMOVED: 2433 /* call mptcp_event_addr_announced()/removed instead */ 2434 WARN_ON_ONCE(1); 2435 break; 2436 case MPTCP_EVENT_SUB_ESTABLISHED: 2437 case MPTCP_EVENT_SUB_PRIORITY: 2438 if (mptcp_event_sub_established(skb, msk, ssk) < 0) 2439 goto nla_put_failure; 2440 break; 2441 case MPTCP_EVENT_SUB_CLOSED: 2442 if (mptcp_event_sub_closed(skb, msk, ssk) < 0) 2443 goto nla_put_failure; 2444 break; 2445 case MPTCP_EVENT_LISTENER_CREATED: 2446 case MPTCP_EVENT_LISTENER_CLOSED: 2447 break; 2448 } 2449 2450 genlmsg_end(skb, nlh); 2451 mptcp_nl_mcast_send(net, skb, gfp); 2452 return; 2453 2454 nla_put_failure: 2455 nlmsg_free(skb); 2456 } 2457 2458 static const struct genl_small_ops mptcp_pm_ops[] = { 2459 { 2460 .cmd = MPTCP_PM_CMD_ADD_ADDR, 2461 .doit = mptcp_nl_cmd_add_addr, 2462 .flags = GENL_UNS_ADMIN_PERM, 2463 }, 2464 { 2465 .cmd = MPTCP_PM_CMD_DEL_ADDR, 2466 .doit = mptcp_nl_cmd_del_addr, 2467 .flags = GENL_UNS_ADMIN_PERM, 2468 }, 2469 { 2470 .cmd = MPTCP_PM_CMD_FLUSH_ADDRS, 2471 .doit = mptcp_nl_cmd_flush_addrs, 2472 .flags = GENL_UNS_ADMIN_PERM, 2473 }, 2474 { 2475 .cmd = MPTCP_PM_CMD_GET_ADDR, 2476 .doit = mptcp_nl_cmd_get_addr, 2477 .dumpit = mptcp_nl_cmd_dump_addrs, 2478 }, 2479 { 2480 .cmd = MPTCP_PM_CMD_SET_LIMITS, 2481 .doit = mptcp_nl_cmd_set_limits, 2482 .flags = GENL_UNS_ADMIN_PERM, 2483 }, 2484 { 2485 .cmd = MPTCP_PM_CMD_GET_LIMITS, 2486 .doit = mptcp_nl_cmd_get_limits, 2487 }, 2488 { 2489 .cmd = MPTCP_PM_CMD_SET_FLAGS, 2490 .doit = mptcp_nl_cmd_set_flags, 2491 .flags = GENL_UNS_ADMIN_PERM, 2492 }, 2493 { 2494 .cmd = MPTCP_PM_CMD_ANNOUNCE, 2495 .doit = mptcp_nl_cmd_announce, 2496 .flags = GENL_UNS_ADMIN_PERM, 2497 }, 2498 { 2499 .cmd = MPTCP_PM_CMD_REMOVE, 2500 .doit = mptcp_nl_cmd_remove, 2501 .flags = GENL_UNS_ADMIN_PERM, 2502 }, 2503 { 2504 .cmd = MPTCP_PM_CMD_SUBFLOW_CREATE, 2505 .doit = mptcp_nl_cmd_sf_create, 2506 .flags = GENL_UNS_ADMIN_PERM, 2507 }, 2508 { 2509 .cmd = MPTCP_PM_CMD_SUBFLOW_DESTROY, 2510 .doit = mptcp_nl_cmd_sf_destroy, 2511 .flags = GENL_UNS_ADMIN_PERM, 2512 }, 2513 }; 2514 2515 static struct genl_family mptcp_genl_family __ro_after_init = { 2516 .name = MPTCP_PM_NAME, 2517 .version = MPTCP_PM_VER, 2518 .maxattr = MPTCP_PM_ATTR_MAX, 2519 .policy = mptcp_pm_policy, 2520 .netnsok = true, 2521 .module = THIS_MODULE, 2522 .small_ops = mptcp_pm_ops, 2523 .n_small_ops = ARRAY_SIZE(mptcp_pm_ops), 2524 .resv_start_op = MPTCP_PM_CMD_SUBFLOW_DESTROY + 1, 2525 .mcgrps = mptcp_pm_mcgrps, 2526 .n_mcgrps = ARRAY_SIZE(mptcp_pm_mcgrps), 2527 }; 2528 2529 static int __net_init pm_nl_init_net(struct net *net) 2530 { 2531 struct pm_nl_pernet *pernet = pm_nl_get_pernet(net); 2532 2533 INIT_LIST_HEAD_RCU(&pernet->local_addr_list); 2534 2535 /* Cit. 2 subflows ought to be enough for anybody. */ 2536 pernet->subflows_max = 2; 2537 pernet->next_id = 1; 2538 pernet->stale_loss_cnt = 4; 2539 spin_lock_init(&pernet->lock); 2540 2541 /* No need to initialize other pernet fields, the struct is zeroed at 2542 * allocation time. 2543 */ 2544 2545 return 0; 2546 } 2547 2548 static void __net_exit pm_nl_exit_net(struct list_head *net_list) 2549 { 2550 struct net *net; 2551 2552 list_for_each_entry(net, net_list, exit_list) { 2553 struct pm_nl_pernet *pernet = pm_nl_get_pernet(net); 2554 2555 /* net is removed from namespace list, can't race with 2556 * other modifiers, also netns core already waited for a 2557 * RCU grace period. 2558 */ 2559 __flush_addrs(&pernet->local_addr_list); 2560 } 2561 } 2562 2563 static struct pernet_operations mptcp_pm_pernet_ops = { 2564 .init = pm_nl_init_net, 2565 .exit_batch = pm_nl_exit_net, 2566 .id = &pm_nl_pernet_id, 2567 .size = sizeof(struct pm_nl_pernet), 2568 }; 2569 2570 void __init mptcp_pm_nl_init(void) 2571 { 2572 if (register_pernet_subsys(&mptcp_pm_pernet_ops) < 0) 2573 panic("Failed to register MPTCP PM pernet subsystem.\n"); 2574 2575 if (genl_register_family(&mptcp_genl_family)) 2576 panic("Failed to register MPTCP PM netlink family\n"); 2577 } 2578