1 // SPDX-License-Identifier: GPL-2.0 2 /* Multipath TCP 3 * 4 * Copyright (c) 2025, Matthieu Baerts. 5 */ 6 7 #define pr_fmt(fmt) "MPTCP: " fmt 8 9 #include <net/netns/generic.h> 10 11 #include "protocol.h" 12 #include "mib.h" 13 #include "mptcp_pm_gen.h" 14 15 static int pm_nl_pernet_id; 16 17 struct pm_nl_pernet { 18 /* protects pernet updates */ 19 spinlock_t lock; 20 struct list_head endp_list; 21 u8 endpoints; 22 u8 endp_signal_max; 23 u8 endp_subflow_max; 24 u8 endp_laminar_max; 25 u8 endp_fullmesh_max; 26 u8 limit_add_addr_accepted; 27 u8 limit_extra_subflows; 28 u8 next_id; 29 DECLARE_BITMAP(id_bitmap, MPTCP_PM_MAX_ADDR_ID + 1); 30 }; 31 32 #define MPTCP_PM_ADDR_MAX 8 33 34 static struct pm_nl_pernet *pm_nl_get_pernet(const struct net *net) 35 { 36 return net_generic(net, pm_nl_pernet_id); 37 } 38 39 static struct pm_nl_pernet * 40 pm_nl_get_pernet_from_msk(const struct mptcp_sock *msk) 41 { 42 return pm_nl_get_pernet(sock_net((struct sock *)msk)); 43 } 44 45 static struct pm_nl_pernet *genl_info_pm_nl(struct genl_info *info) 46 { 47 return pm_nl_get_pernet(genl_info_net(info)); 48 } 49 50 u8 mptcp_pm_get_endp_signal_max(const struct mptcp_sock *msk) 51 { 52 const struct pm_nl_pernet *pernet = pm_nl_get_pernet_from_msk(msk); 53 54 return READ_ONCE(pernet->endp_signal_max); 55 } 56 EXPORT_SYMBOL_GPL(mptcp_pm_get_endp_signal_max); 57 58 u8 mptcp_pm_get_endp_subflow_max(const struct mptcp_sock *msk) 59 { 60 struct pm_nl_pernet *pernet = pm_nl_get_pernet_from_msk(msk); 61 62 return READ_ONCE(pernet->endp_subflow_max); 63 } 64 EXPORT_SYMBOL_GPL(mptcp_pm_get_endp_subflow_max); 65 66 u8 mptcp_pm_get_endp_laminar_max(const struct mptcp_sock *msk) 67 { 68 struct pm_nl_pernet *pernet = pm_nl_get_pernet_from_msk(msk); 69 70 return READ_ONCE(pernet->endp_laminar_max); 71 } 72 EXPORT_SYMBOL_GPL(mptcp_pm_get_endp_laminar_max); 73 74 u8 mptcp_pm_get_endp_fullmesh_max(const struct mptcp_sock *msk) 75 { 76 struct pm_nl_pernet *pernet = pm_nl_get_pernet_from_msk(msk); 77 78 return READ_ONCE(pernet->endp_fullmesh_max); 79 } 80 EXPORT_SYMBOL_GPL(mptcp_pm_get_endp_fullmesh_max); 81 82 u8 mptcp_pm_get_limit_add_addr_accepted(const struct mptcp_sock *msk) 83 { 84 struct pm_nl_pernet *pernet = pm_nl_get_pernet_from_msk(msk); 85 86 return READ_ONCE(pernet->limit_add_addr_accepted); 87 } 88 EXPORT_SYMBOL_GPL(mptcp_pm_get_limit_add_addr_accepted); 89 90 u8 mptcp_pm_get_limit_extra_subflows(const struct mptcp_sock *msk) 91 { 92 struct pm_nl_pernet *pernet = pm_nl_get_pernet_from_msk(msk); 93 94 return READ_ONCE(pernet->limit_extra_subflows); 95 } 96 EXPORT_SYMBOL_GPL(mptcp_pm_get_limit_extra_subflows); 97 98 static bool lookup_subflow_by_daddr(const struct list_head *list, 99 const struct mptcp_addr_info *daddr) 100 { 101 struct mptcp_subflow_context *subflow; 102 struct mptcp_addr_info cur; 103 104 list_for_each_entry(subflow, list, node) { 105 struct sock *ssk = mptcp_subflow_tcp_sock(subflow); 106 107 if (!((1 << inet_sk_state_load(ssk)) & 108 (TCPF_ESTABLISHED | TCPF_SYN_SENT | TCPF_SYN_RECV))) 109 continue; 110 111 mptcp_remote_address((struct sock_common *)ssk, &cur); 112 if (mptcp_addresses_equal(&cur, daddr, daddr->port)) 113 return true; 114 } 115 116 return false; 117 } 118 119 static bool 120 select_local_address(const struct pm_nl_pernet *pernet, 121 const struct mptcp_sock *msk, 122 struct mptcp_pm_local *new_local) 123 { 124 struct mptcp_pm_addr_entry *entry; 125 bool found = false; 126 127 msk_owned_by_me(msk); 128 129 rcu_read_lock(); 130 list_for_each_entry_rcu(entry, &pernet->endp_list, list) { 131 if (!(entry->flags & MPTCP_PM_ADDR_FLAG_SUBFLOW)) 132 continue; 133 134 if (!test_bit(entry->addr.id, msk->pm.id_avail_bitmap)) 135 continue; 136 137 new_local->addr = entry->addr; 138 new_local->flags = entry->flags; 139 new_local->ifindex = entry->ifindex; 140 found = true; 141 break; 142 } 143 rcu_read_unlock(); 144 145 return found; 146 } 147 148 static bool 149 select_signal_address(struct pm_nl_pernet *pernet, const struct mptcp_sock *msk, 150 struct mptcp_pm_local *new_local) 151 { 152 struct mptcp_pm_addr_entry *entry; 153 bool found = false; 154 155 rcu_read_lock(); 156 /* do not keep any additional per socket state, just signal 157 * the address list in order. 158 * Note: removal from the local address list during the msk life-cycle 159 * can lead to additional addresses not being announced. 160 */ 161 list_for_each_entry_rcu(entry, &pernet->endp_list, list) { 162 if (!test_bit(entry->addr.id, msk->pm.id_avail_bitmap)) 163 continue; 164 165 if (!(entry->flags & MPTCP_PM_ADDR_FLAG_SIGNAL)) 166 continue; 167 168 new_local->addr = entry->addr; 169 new_local->flags = entry->flags; 170 new_local->ifindex = entry->ifindex; 171 found = true; 172 break; 173 } 174 rcu_read_unlock(); 175 176 return found; 177 } 178 179 static unsigned int 180 fill_remote_addr(struct mptcp_sock *msk, struct mptcp_addr_info *local, 181 struct mptcp_addr_info *addrs) 182 { 183 bool deny_id0 = READ_ONCE(msk->pm.remote_deny_join_id0); 184 struct mptcp_addr_info remote = { 0 }; 185 struct sock *sk = (struct sock *)msk; 186 187 if (deny_id0) 188 return 0; 189 190 mptcp_remote_address((struct sock_common *)sk, &remote); 191 192 if (!mptcp_pm_addr_families_match(sk, local, &remote)) 193 return 0; 194 195 msk->pm.extra_subflows++; 196 *addrs = remote; 197 198 return 1; 199 } 200 201 static unsigned int 202 fill_remote_addresses_fullmesh(struct mptcp_sock *msk, 203 struct mptcp_addr_info *local, 204 struct mptcp_addr_info *addrs) 205 { 206 u8 limit_extra_subflows = mptcp_pm_get_limit_extra_subflows(msk); 207 bool deny_id0 = READ_ONCE(msk->pm.remote_deny_join_id0); 208 DECLARE_BITMAP(unavail_id, MPTCP_PM_MAX_ADDR_ID + 1); 209 struct sock *sk = (struct sock *)msk, *ssk; 210 struct mptcp_subflow_context *subflow; 211 int i = 0; 212 213 /* Forbid creation of new subflows matching existing ones, possibly 214 * already created by incoming ADD_ADDR 215 */ 216 bitmap_zero(unavail_id, MPTCP_PM_MAX_ADDR_ID + 1); 217 mptcp_for_each_subflow(msk, subflow) 218 if (READ_ONCE(subflow->local_id) == local->id) 219 __set_bit(subflow->remote_id, unavail_id); 220 221 mptcp_for_each_subflow(msk, subflow) { 222 ssk = mptcp_subflow_tcp_sock(subflow); 223 mptcp_remote_address((struct sock_common *)ssk, &addrs[i]); 224 addrs[i].id = READ_ONCE(subflow->remote_id); 225 if (deny_id0 && !addrs[i].id) 226 continue; 227 228 if (test_bit(addrs[i].id, unavail_id)) 229 continue; 230 231 if (!mptcp_pm_addr_families_match(sk, local, &addrs[i])) 232 continue; 233 234 /* forbid creating multiple address towards this id */ 235 __set_bit(addrs[i].id, unavail_id); 236 msk->pm.extra_subflows++; 237 i++; 238 239 if (msk->pm.extra_subflows >= limit_extra_subflows) 240 break; 241 } 242 243 return i; 244 } 245 246 /* Fill all the remote addresses into the array addrs[], 247 * and return the array size. 248 */ 249 static unsigned int 250 fill_remote_addresses_vec(struct mptcp_sock *msk, struct mptcp_addr_info *local, 251 bool fullmesh, struct mptcp_addr_info *addrs) 252 { 253 /* Non-fullmesh: fill in the single entry corresponding to the primary 254 * MPC subflow remote address, and return 1, corresponding to 1 entry. 255 */ 256 if (!fullmesh) 257 return fill_remote_addr(msk, local, addrs); 258 259 /* Fullmesh endpoint: fill all possible remote addresses */ 260 return fill_remote_addresses_fullmesh(msk, local, addrs); 261 } 262 263 static struct mptcp_pm_addr_entry * 264 __lookup_addr_by_id(struct pm_nl_pernet *pernet, unsigned int id) 265 { 266 struct mptcp_pm_addr_entry *entry; 267 268 list_for_each_entry_rcu(entry, &pernet->endp_list, list, 269 lockdep_is_held(&pernet->lock)) { 270 if (entry->addr.id == id) 271 return entry; 272 } 273 return NULL; 274 } 275 276 static struct mptcp_pm_addr_entry * 277 __lookup_addr(struct pm_nl_pernet *pernet, const struct mptcp_addr_info *info) 278 { 279 struct mptcp_pm_addr_entry *entry; 280 281 list_for_each_entry_rcu(entry, &pernet->endp_list, list, 282 lockdep_is_held(&pernet->lock)) { 283 if (mptcp_addresses_equal(&entry->addr, info, entry->addr.port)) 284 return entry; 285 } 286 return NULL; 287 } 288 289 static u8 mptcp_endp_get_local_id(struct mptcp_sock *msk, 290 const struct mptcp_addr_info *addr) 291 { 292 return msk->mpc_endpoint_id == addr->id ? 0 : addr->id; 293 } 294 295 /* Set mpc_endpoint_id, and send MP_PRIO for ID0 if needed */ 296 static void mptcp_mpc_endpoint_setup(struct mptcp_sock *msk) 297 { 298 struct mptcp_subflow_context *subflow; 299 struct mptcp_pm_addr_entry *entry; 300 struct mptcp_addr_info mpc_addr; 301 struct pm_nl_pernet *pernet; 302 bool backup = false; 303 304 /* do lazy endpoint usage accounting for the MPC subflows */ 305 if (likely(msk->pm.status & BIT(MPTCP_PM_MPC_ENDPOINT_ACCOUNTED)) || 306 !msk->first) 307 return; 308 309 subflow = mptcp_subflow_ctx(msk->first); 310 pernet = pm_nl_get_pernet_from_msk(msk); 311 312 mptcp_local_address((struct sock_common *)msk->first, &mpc_addr); 313 rcu_read_lock(); 314 entry = __lookup_addr(pernet, &mpc_addr); 315 if (entry) { 316 __clear_bit(entry->addr.id, msk->pm.id_avail_bitmap); 317 msk->mpc_endpoint_id = entry->addr.id; 318 backup = !!(entry->flags & MPTCP_PM_ADDR_FLAG_BACKUP); 319 } 320 rcu_read_unlock(); 321 322 /* Send MP_PRIO */ 323 if (backup) 324 mptcp_pm_send_ack(msk, subflow, true, backup); 325 326 msk->pm.status |= BIT(MPTCP_PM_MPC_ENDPOINT_ACCOUNTED); 327 } 328 329 static void mptcp_pm_create_subflow_or_signal_addr(struct mptcp_sock *msk) 330 { 331 u8 limit_extra_subflows = mptcp_pm_get_limit_extra_subflows(msk); 332 struct pm_nl_pernet *pernet = pm_nl_get_pernet_from_msk(msk); 333 u8 endp_subflow_max = mptcp_pm_get_endp_subflow_max(msk); 334 u8 endp_signal_max = mptcp_pm_get_endp_signal_max(msk); 335 struct sock *sk = (struct sock *)msk; 336 bool signal_and_subflow = false; 337 struct mptcp_pm_local local; 338 339 mptcp_mpc_endpoint_setup(msk); 340 341 pr_debug("local %d:%d signal %d:%d subflows %d:%d\n", 342 msk->pm.local_addr_used, endp_subflow_max, 343 msk->pm.add_addr_signaled, endp_signal_max, 344 msk->pm.extra_subflows, limit_extra_subflows); 345 346 /* check first for announce */ 347 if (msk->pm.add_addr_signaled < endp_signal_max) { 348 /* due to racing events on both ends we can reach here while 349 * previous add address is still running: if we invoke now 350 * mptcp_pm_announce_addr(), that will fail and the 351 * corresponding id will be marked as used. 352 * Instead let the PM machinery reschedule us when the 353 * current address announce will be completed. 354 */ 355 if (msk->pm.addr_signal & BIT(MPTCP_ADD_ADDR_SIGNAL)) 356 return; 357 358 if (!select_signal_address(pernet, msk, &local)) 359 goto subflow; 360 361 /* If the alloc fails, we are on memory pressure, not worth 362 * continuing, and trying to create subflows. 363 */ 364 if (!mptcp_pm_alloc_anno_list(msk, &local.addr)) 365 return; 366 367 __clear_bit(local.addr.id, msk->pm.id_avail_bitmap); 368 msk->pm.add_addr_signaled++; 369 370 /* Special case for ID0: set the correct ID */ 371 if (local.addr.id == msk->mpc_endpoint_id) 372 local.addr.id = 0; 373 374 mptcp_pm_announce_addr(msk, &local.addr, false); 375 mptcp_pm_addr_send_ack(msk); 376 377 if (local.flags & MPTCP_PM_ADDR_FLAG_SUBFLOW) 378 signal_and_subflow = true; 379 } 380 381 subflow: 382 /* No need to try establishing subflows to remote id0 if not allowed */ 383 if (mptcp_pm_add_addr_c_flag_case(msk)) 384 goto exit; 385 386 /* check if should create a new subflow */ 387 while (msk->pm.local_addr_used < endp_subflow_max && 388 msk->pm.extra_subflows < limit_extra_subflows) { 389 struct mptcp_addr_info addrs[MPTCP_PM_ADDR_MAX]; 390 bool fullmesh; 391 int i, nr; 392 393 if (signal_and_subflow) 394 signal_and_subflow = false; 395 else if (!select_local_address(pernet, msk, &local)) 396 break; 397 398 fullmesh = !!(local.flags & MPTCP_PM_ADDR_FLAG_FULLMESH); 399 400 __clear_bit(local.addr.id, msk->pm.id_avail_bitmap); 401 402 /* Special case for ID0: set the correct ID */ 403 if (local.addr.id == msk->mpc_endpoint_id) 404 local.addr.id = 0; 405 else /* local_addr_used is not decr for ID 0 */ 406 msk->pm.local_addr_used++; 407 408 nr = fill_remote_addresses_vec(msk, &local.addr, fullmesh, addrs); 409 if (nr == 0) 410 continue; 411 412 spin_unlock_bh(&msk->pm.lock); 413 for (i = 0; i < nr; i++) 414 __mptcp_subflow_connect(sk, &local, &addrs[i]); 415 spin_lock_bh(&msk->pm.lock); 416 } 417 418 exit: 419 mptcp_pm_nl_check_work_pending(msk); 420 } 421 422 static void mptcp_pm_nl_fully_established(struct mptcp_sock *msk) 423 { 424 mptcp_pm_create_subflow_or_signal_addr(msk); 425 } 426 427 static void mptcp_pm_nl_subflow_established(struct mptcp_sock *msk) 428 { 429 mptcp_pm_create_subflow_or_signal_addr(msk); 430 } 431 432 static unsigned int 433 fill_local_addresses_vec_fullmesh(struct mptcp_sock *msk, 434 struct mptcp_addr_info *remote, 435 struct mptcp_pm_local *locals, 436 bool c_flag_case) 437 { 438 u8 limit_extra_subflows = mptcp_pm_get_limit_extra_subflows(msk); 439 struct pm_nl_pernet *pernet = pm_nl_get_pernet_from_msk(msk); 440 struct sock *sk = (struct sock *)msk; 441 struct mptcp_pm_addr_entry *entry; 442 struct mptcp_pm_local *local; 443 int i = 0; 444 445 rcu_read_lock(); 446 list_for_each_entry_rcu(entry, &pernet->endp_list, list) { 447 bool is_id0; 448 449 if (!(entry->flags & MPTCP_PM_ADDR_FLAG_FULLMESH)) 450 continue; 451 452 if (!mptcp_pm_addr_families_match(sk, &entry->addr, remote)) 453 continue; 454 455 local = &locals[i]; 456 local->addr = entry->addr; 457 local->flags = entry->flags; 458 local->ifindex = entry->ifindex; 459 460 is_id0 = local->addr.id == msk->mpc_endpoint_id; 461 462 if (c_flag_case && 463 (entry->flags & MPTCP_PM_ADDR_FLAG_SUBFLOW)) { 464 __clear_bit(local->addr.id, msk->pm.id_avail_bitmap); 465 466 if (!is_id0) 467 msk->pm.local_addr_used++; 468 } 469 470 /* Special case for ID0: set the correct ID */ 471 if (is_id0) 472 local->addr.id = 0; 473 474 msk->pm.extra_subflows++; 475 i++; 476 477 if (msk->pm.extra_subflows >= limit_extra_subflows) 478 break; 479 } 480 rcu_read_unlock(); 481 482 return i; 483 } 484 485 static unsigned int 486 fill_local_laminar_endp(struct mptcp_sock *msk, struct mptcp_addr_info *remote, 487 struct mptcp_pm_local *locals) 488 { 489 struct pm_nl_pernet *pernet = pm_nl_get_pernet_from_msk(msk); 490 DECLARE_BITMAP(unavail_id, MPTCP_PM_MAX_ADDR_ID + 1); 491 struct mptcp_subflow_context *subflow; 492 struct sock *sk = (struct sock *)msk; 493 struct mptcp_pm_addr_entry *entry; 494 struct mptcp_pm_local *local; 495 int found = 0; 496 497 /* Forbid creation of new subflows matching existing ones, possibly 498 * already created by 'subflow' endpoints 499 */ 500 bitmap_zero(unavail_id, MPTCP_PM_MAX_ADDR_ID + 1); 501 mptcp_for_each_subflow(msk, subflow) { 502 struct sock *ssk = mptcp_subflow_tcp_sock(subflow); 503 504 if ((1 << inet_sk_state_load(ssk)) & 505 (TCPF_FIN_WAIT1 | TCPF_FIN_WAIT2 | TCPF_CLOSING | 506 TCPF_CLOSE)) 507 continue; 508 509 __set_bit(subflow_get_local_id(subflow), unavail_id); 510 } 511 512 rcu_read_lock(); 513 list_for_each_entry_rcu(entry, &pernet->endp_list, list) { 514 if (!(entry->flags & MPTCP_PM_ADDR_FLAG_LAMINAR)) 515 continue; 516 517 if (!mptcp_pm_addr_families_match(sk, &entry->addr, remote)) 518 continue; 519 520 if (test_bit(mptcp_endp_get_local_id(msk, &entry->addr), 521 unavail_id)) 522 continue; 523 524 local = &locals[0]; 525 local->addr = entry->addr; 526 local->flags = entry->flags; 527 local->ifindex = entry->ifindex; 528 529 if (entry->flags & MPTCP_PM_ADDR_FLAG_SUBFLOW) { 530 __clear_bit(local->addr.id, msk->pm.id_avail_bitmap); 531 532 if (local->addr.id != msk->mpc_endpoint_id) 533 msk->pm.local_addr_used++; 534 } 535 536 msk->pm.extra_subflows++; 537 found = 1; 538 break; 539 } 540 rcu_read_unlock(); 541 542 return found; 543 } 544 545 static unsigned int 546 fill_local_addresses_vec_c_flag(struct mptcp_sock *msk, 547 struct mptcp_addr_info *remote, 548 struct mptcp_pm_local *locals) 549 { 550 u8 limit_extra_subflows = mptcp_pm_get_limit_extra_subflows(msk); 551 struct pm_nl_pernet *pernet = pm_nl_get_pernet_from_msk(msk); 552 u8 endp_subflow_max = mptcp_pm_get_endp_subflow_max(msk); 553 struct sock *sk = (struct sock *)msk; 554 struct mptcp_pm_local *local; 555 int i = 0; 556 557 while (msk->pm.local_addr_used < endp_subflow_max) { 558 local = &locals[i]; 559 560 if (!select_local_address(pernet, msk, local)) 561 break; 562 563 __clear_bit(local->addr.id, msk->pm.id_avail_bitmap); 564 565 if (!mptcp_pm_addr_families_match(sk, &local->addr, remote)) 566 continue; 567 568 if (local->addr.id == msk->mpc_endpoint_id) 569 continue; 570 571 msk->pm.local_addr_used++; 572 msk->pm.extra_subflows++; 573 i++; 574 575 if (msk->pm.extra_subflows >= limit_extra_subflows) 576 break; 577 } 578 579 return i; 580 } 581 582 static unsigned int 583 fill_local_address_any(struct mptcp_sock *msk, struct mptcp_addr_info *remote, 584 struct mptcp_pm_local *local) 585 { 586 struct sock *sk = (struct sock *)msk; 587 588 memset(local, 0, sizeof(*local)); 589 local->addr.family = 590 #if IS_ENABLED(CONFIG_MPTCP_IPV6) 591 remote->family == AF_INET6 && 592 ipv6_addr_v4mapped(&remote->addr6) ? AF_INET : 593 #endif 594 remote->family; 595 596 if (!mptcp_pm_addr_families_match(sk, &local->addr, remote)) 597 return 0; 598 599 msk->pm.extra_subflows++; 600 601 return 1; 602 } 603 604 /* Fill all the local addresses into the array addrs[], 605 * and return the array size. 606 */ 607 static unsigned int 608 fill_local_addresses_vec(struct mptcp_sock *msk, struct mptcp_addr_info *remote, 609 struct mptcp_pm_local *locals) 610 { 611 bool c_flag_case = remote->id && mptcp_pm_add_addr_c_flag_case(msk); 612 613 /* If there is at least one MPTCP endpoint with a fullmesh flag */ 614 if (mptcp_pm_get_endp_fullmesh_max(msk)) 615 return fill_local_addresses_vec_fullmesh(msk, remote, locals, 616 c_flag_case); 617 618 /* If there is at least one MPTCP endpoint with a laminar flag */ 619 if (mptcp_pm_get_endp_laminar_max(msk)) 620 return fill_local_laminar_endp(msk, remote, locals); 621 622 /* Special case: peer sets the C flag, accept one ADD_ADDR if default 623 * limits are used -- accepting no ADD_ADDR -- and use subflow endpoints 624 */ 625 if (c_flag_case) 626 return fill_local_addresses_vec_c_flag(msk, remote, locals); 627 628 /* No special case: fill in the single 'IPADDRANY' local address */ 629 return fill_local_address_any(msk, remote, &locals[0]); 630 } 631 632 static void mptcp_pm_nl_add_addr_received(struct mptcp_sock *msk) 633 { 634 u8 limit_add_addr_accepted = mptcp_pm_get_limit_add_addr_accepted(msk); 635 u8 limit_extra_subflows = mptcp_pm_get_limit_extra_subflows(msk); 636 struct mptcp_pm_local locals[MPTCP_PM_ADDR_MAX]; 637 struct sock *sk = (struct sock *)msk; 638 struct mptcp_addr_info remote; 639 bool sf_created = false; 640 int i, nr; 641 642 pr_debug("accepted %d:%d remote family %d\n", 643 msk->pm.add_addr_accepted, limit_add_addr_accepted, 644 msk->pm.remote.family); 645 646 remote = msk->pm.remote; 647 mptcp_pm_announce_addr(msk, &remote, true); 648 mptcp_pm_addr_send_ack(msk); 649 mptcp_mpc_endpoint_setup(msk); 650 651 if (lookup_subflow_by_daddr(&msk->conn_list, &remote)) 652 return; 653 654 /* pick id 0 port, if none is provided the remote address */ 655 if (!remote.port) 656 remote.port = sk->sk_dport; 657 658 /* connect to the specified remote address, using whatever 659 * local address the routing configuration will pick. 660 */ 661 nr = fill_local_addresses_vec(msk, &remote, locals); 662 if (nr == 0) 663 return; 664 665 spin_unlock_bh(&msk->pm.lock); 666 for (i = 0; i < nr; i++) 667 if (__mptcp_subflow_connect(sk, &locals[i], &remote) == 0) 668 sf_created = true; 669 spin_lock_bh(&msk->pm.lock); 670 671 if (sf_created) { 672 /* add_addr_accepted is not decr for ID 0 */ 673 if (remote.id) 674 msk->pm.add_addr_accepted++; 675 if (msk->pm.add_addr_accepted >= limit_add_addr_accepted || 676 msk->pm.extra_subflows >= limit_extra_subflows) 677 WRITE_ONCE(msk->pm.accept_addr, false); 678 } 679 } 680 681 void mptcp_pm_nl_rm_addr(struct mptcp_sock *msk, u8 rm_id) 682 { 683 if (rm_id && !WARN_ON_ONCE(msk->pm.add_addr_accepted == 0)) { 684 u8 limit_add_addr_accepted = 685 mptcp_pm_get_limit_add_addr_accepted(msk); 686 687 /* Note: if the subflow has been closed before, this 688 * add_addr_accepted counter will not be decremented. 689 */ 690 if (--msk->pm.add_addr_accepted < limit_add_addr_accepted) 691 WRITE_ONCE(msk->pm.accept_addr, true); 692 } 693 } 694 695 static bool address_use_port(struct mptcp_pm_addr_entry *entry) 696 { 697 return (entry->flags & 698 (MPTCP_PM_ADDR_FLAG_SIGNAL | MPTCP_PM_ADDR_FLAG_SUBFLOW)) == 699 MPTCP_PM_ADDR_FLAG_SIGNAL; 700 } 701 702 /* caller must ensure the RCU grace period is already elapsed */ 703 static void __mptcp_pm_release_addr_entry(struct mptcp_pm_addr_entry *entry) 704 { 705 if (entry->lsk) 706 sock_release(entry->lsk); 707 kfree(entry); 708 } 709 710 static int mptcp_pm_nl_append_new_local_addr(struct pm_nl_pernet *pernet, 711 struct mptcp_pm_addr_entry *entry, 712 bool needs_id, bool replace) 713 { 714 struct mptcp_pm_addr_entry *cur, *del_entry = NULL; 715 int ret = -EINVAL; 716 u8 addr_max; 717 718 spin_lock_bh(&pernet->lock); 719 /* to keep the code simple, don't do IDR-like allocation for address ID, 720 * just bail when we exceed limits 721 */ 722 if (pernet->next_id == MPTCP_PM_MAX_ADDR_ID) 723 pernet->next_id = 1; 724 if (pernet->endpoints >= MPTCP_PM_ADDR_MAX) { 725 ret = -ERANGE; 726 goto out; 727 } 728 if (test_bit(entry->addr.id, pernet->id_bitmap)) { 729 ret = -EBUSY; 730 goto out; 731 } 732 733 /* do not insert duplicate address, differentiate on port only 734 * singled addresses 735 */ 736 if (!address_use_port(entry)) 737 entry->addr.port = 0; 738 list_for_each_entry(cur, &pernet->endp_list, list) { 739 if (mptcp_addresses_equal(&cur->addr, &entry->addr, 740 cur->addr.port || entry->addr.port)) { 741 /* allow replacing the exiting endpoint only if such 742 * endpoint is an implicit one and the user-space 743 * did not provide an endpoint id 744 */ 745 if (!(cur->flags & MPTCP_PM_ADDR_FLAG_IMPLICIT)) { 746 ret = -EEXIST; 747 goto out; 748 } 749 if (entry->addr.id) 750 goto out; 751 752 /* allow callers that only need to look up the local 753 * addr's id to skip replacement. This allows them to 754 * avoid calling synchronize_rcu in the packet recv 755 * path. 756 */ 757 if (!replace) { 758 kfree(entry); 759 ret = cur->addr.id; 760 goto out; 761 } 762 763 pernet->endpoints--; 764 entry->addr.id = cur->addr.id; 765 list_del_rcu(&cur->list); 766 del_entry = cur; 767 break; 768 } 769 } 770 771 if (!entry->addr.id && needs_id) { 772 find_next: 773 entry->addr.id = find_next_zero_bit(pernet->id_bitmap, 774 MPTCP_PM_MAX_ADDR_ID + 1, 775 pernet->next_id); 776 if (!entry->addr.id && pernet->next_id != 1) { 777 pernet->next_id = 1; 778 goto find_next; 779 } 780 } 781 782 if (!entry->addr.id && needs_id) 783 goto out; 784 785 __set_bit(entry->addr.id, pernet->id_bitmap); 786 if (entry->addr.id > pernet->next_id) 787 pernet->next_id = entry->addr.id; 788 789 if (entry->flags & MPTCP_PM_ADDR_FLAG_SIGNAL) { 790 addr_max = pernet->endp_signal_max; 791 WRITE_ONCE(pernet->endp_signal_max, addr_max + 1); 792 } 793 if (entry->flags & MPTCP_PM_ADDR_FLAG_SUBFLOW) { 794 addr_max = pernet->endp_subflow_max; 795 WRITE_ONCE(pernet->endp_subflow_max, addr_max + 1); 796 } 797 if (entry->flags & MPTCP_PM_ADDR_FLAG_LAMINAR) { 798 addr_max = pernet->endp_laminar_max; 799 WRITE_ONCE(pernet->endp_laminar_max, addr_max + 1); 800 } 801 if (entry->flags & MPTCP_PM_ADDR_FLAG_FULLMESH) { 802 addr_max = pernet->endp_fullmesh_max; 803 WRITE_ONCE(pernet->endp_fullmesh_max, addr_max + 1); 804 } 805 806 pernet->endpoints++; 807 if (!entry->addr.port) 808 list_add_tail_rcu(&entry->list, &pernet->endp_list); 809 else 810 list_add_rcu(&entry->list, &pernet->endp_list); 811 ret = entry->addr.id; 812 813 out: 814 spin_unlock_bh(&pernet->lock); 815 816 /* just replaced an existing entry, free it */ 817 if (del_entry) { 818 synchronize_rcu(); 819 __mptcp_pm_release_addr_entry(del_entry); 820 } 821 return ret; 822 } 823 824 static struct lock_class_key mptcp_slock_keys[2]; 825 static struct lock_class_key mptcp_keys[2]; 826 827 static int mptcp_pm_nl_create_listen_socket(struct sock *sk, 828 struct mptcp_pm_addr_entry *entry) 829 { 830 bool is_ipv6 = sk->sk_family == AF_INET6; 831 int addrlen = sizeof(struct sockaddr_in); 832 struct sockaddr_storage addr; 833 struct sock *newsk, *ssk; 834 int backlog = 1024; 835 int err; 836 837 err = sock_create_kern(sock_net(sk), entry->addr.family, 838 SOCK_STREAM, IPPROTO_MPTCP, &entry->lsk); 839 if (err) 840 return err; 841 842 newsk = entry->lsk->sk; 843 if (!newsk) 844 return -EINVAL; 845 846 /* The subflow socket lock is acquired in a nested to the msk one 847 * in several places, even by the TCP stack, and this msk is a kernel 848 * socket: lockdep complains. Instead of propagating the _nested 849 * modifiers in several places, re-init the lock class for the msk 850 * socket to an mptcp specific one. 851 */ 852 sock_lock_init_class_and_name(newsk, 853 is_ipv6 ? "mlock-AF_INET6" : "mlock-AF_INET", 854 &mptcp_slock_keys[is_ipv6], 855 is_ipv6 ? "msk_lock-AF_INET6" : "msk_lock-AF_INET", 856 &mptcp_keys[is_ipv6]); 857 858 lock_sock(newsk); 859 ssk = __mptcp_nmpc_sk(mptcp_sk(newsk)); 860 release_sock(newsk); 861 if (IS_ERR(ssk)) 862 return PTR_ERR(ssk); 863 864 mptcp_info2sockaddr(&entry->addr, &addr, entry->addr.family); 865 #if IS_ENABLED(CONFIG_MPTCP_IPV6) 866 if (entry->addr.family == AF_INET6) 867 addrlen = sizeof(struct sockaddr_in6); 868 #endif 869 if (ssk->sk_family == AF_INET) 870 err = inet_bind_sk(ssk, (struct sockaddr_unsized *)&addr, addrlen); 871 #if IS_ENABLED(CONFIG_MPTCP_IPV6) 872 else if (ssk->sk_family == AF_INET6) 873 err = inet6_bind_sk(ssk, (struct sockaddr_unsized *)&addr, addrlen); 874 #endif 875 if (err) 876 return err; 877 878 /* We don't use mptcp_set_state() here because it needs to be called 879 * under the msk socket lock. For the moment, that will not bring 880 * anything more than only calling inet_sk_state_store(), because the 881 * old status is known (TCP_CLOSE). 882 */ 883 inet_sk_state_store(newsk, TCP_LISTEN); 884 lock_sock(ssk); 885 WRITE_ONCE(mptcp_subflow_ctx(ssk)->pm_listener, true); 886 err = __inet_listen_sk(ssk, backlog); 887 if (!err) 888 mptcp_event_pm_listener(ssk, MPTCP_EVENT_LISTENER_CREATED); 889 release_sock(ssk); 890 return err; 891 } 892 893 int mptcp_pm_nl_get_local_id(struct mptcp_sock *msk, 894 struct mptcp_pm_addr_entry *skc) 895 { 896 struct mptcp_pm_addr_entry *entry; 897 struct pm_nl_pernet *pernet; 898 int ret; 899 900 pernet = pm_nl_get_pernet_from_msk(msk); 901 902 rcu_read_lock(); 903 entry = __lookup_addr(pernet, &skc->addr); 904 ret = entry ? entry->addr.id : -1; 905 rcu_read_unlock(); 906 if (ret >= 0) 907 return ret; 908 909 /* address not found, add to local list */ 910 entry = kmemdup(skc, sizeof(*skc), GFP_ATOMIC); 911 if (!entry) 912 return -ENOMEM; 913 914 entry->addr.port = 0; 915 ret = mptcp_pm_nl_append_new_local_addr(pernet, entry, true, false); 916 if (ret < 0) 917 kfree(entry); 918 919 return ret; 920 } 921 922 bool mptcp_pm_nl_is_backup(struct mptcp_sock *msk, struct mptcp_addr_info *skc) 923 { 924 struct pm_nl_pernet *pernet = pm_nl_get_pernet_from_msk(msk); 925 struct mptcp_pm_addr_entry *entry; 926 bool backup; 927 928 rcu_read_lock(); 929 entry = __lookup_addr(pernet, skc); 930 backup = entry && !!(entry->flags & MPTCP_PM_ADDR_FLAG_BACKUP); 931 rcu_read_unlock(); 932 933 return backup; 934 } 935 936 static int mptcp_nl_add_subflow_or_signal_addr(struct net *net, 937 struct mptcp_addr_info *addr) 938 { 939 struct mptcp_sock *msk; 940 long s_slot = 0, s_num = 0; 941 942 while ((msk = mptcp_token_iter_next(net, &s_slot, &s_num)) != NULL) { 943 struct sock *sk = (struct sock *)msk; 944 struct mptcp_addr_info mpc_addr; 945 946 if (!READ_ONCE(msk->fully_established) || 947 mptcp_pm_is_userspace(msk)) 948 goto next; 949 950 /* if the endp linked to the init sf is re-added with a != ID */ 951 mptcp_local_address((struct sock_common *)msk, &mpc_addr); 952 953 lock_sock(sk); 954 spin_lock_bh(&msk->pm.lock); 955 if (mptcp_addresses_equal(addr, &mpc_addr, addr->port)) 956 msk->mpc_endpoint_id = addr->id; 957 mptcp_pm_create_subflow_or_signal_addr(msk); 958 spin_unlock_bh(&msk->pm.lock); 959 release_sock(sk); 960 961 next: 962 sock_put(sk); 963 cond_resched(); 964 } 965 966 return 0; 967 } 968 969 static bool mptcp_pm_has_addr_attr_id(const struct nlattr *attr, 970 struct genl_info *info) 971 { 972 struct nlattr *tb[MPTCP_PM_ADDR_ATTR_MAX + 1]; 973 974 if (!nla_parse_nested_deprecated(tb, MPTCP_PM_ADDR_ATTR_MAX, attr, 975 mptcp_pm_address_nl_policy, info->extack) && 976 tb[MPTCP_PM_ADDR_ATTR_ID]) 977 return true; 978 return false; 979 } 980 981 /* Add an MPTCP endpoint */ 982 int mptcp_pm_nl_add_addr_doit(struct sk_buff *skb, struct genl_info *info) 983 { 984 struct pm_nl_pernet *pernet = genl_info_pm_nl(info); 985 struct mptcp_pm_addr_entry addr, *entry; 986 struct nlattr *attr; 987 int ret; 988 989 if (GENL_REQ_ATTR_CHECK(info, MPTCP_PM_ENDPOINT_ADDR)) 990 return -EINVAL; 991 992 attr = info->attrs[MPTCP_PM_ENDPOINT_ADDR]; 993 ret = mptcp_pm_parse_entry(attr, info, true, &addr); 994 if (ret < 0) 995 return ret; 996 997 if (addr.addr.port && !address_use_port(&addr)) { 998 NL_SET_ERR_MSG_ATTR(info->extack, attr, 999 "flags must have signal and not subflow when using port"); 1000 return -EINVAL; 1001 } 1002 1003 if (addr.flags & MPTCP_PM_ADDR_FLAG_SIGNAL && 1004 addr.flags & MPTCP_PM_ADDR_FLAG_FULLMESH) { 1005 NL_SET_ERR_MSG_ATTR(info->extack, attr, 1006 "flags mustn't have both signal and fullmesh"); 1007 return -EINVAL; 1008 } 1009 1010 if (addr.flags & MPTCP_PM_ADDR_FLAG_IMPLICIT) { 1011 NL_SET_ERR_MSG_ATTR(info->extack, attr, 1012 "can't create IMPLICIT endpoint"); 1013 return -EINVAL; 1014 } 1015 1016 entry = kmemdup(&addr, sizeof(addr), GFP_KERNEL_ACCOUNT); 1017 if (!entry) { 1018 GENL_SET_ERR_MSG(info, "can't allocate addr"); 1019 return -ENOMEM; 1020 } 1021 1022 if (entry->addr.port) { 1023 ret = mptcp_pm_nl_create_listen_socket(skb->sk, entry); 1024 if (ret) { 1025 GENL_SET_ERR_MSG_FMT(info, "create listen socket error: %d", ret); 1026 goto out_free; 1027 } 1028 } 1029 ret = mptcp_pm_nl_append_new_local_addr(pernet, entry, 1030 !mptcp_pm_has_addr_attr_id(attr, info), 1031 true); 1032 if (ret < 0) { 1033 GENL_SET_ERR_MSG_FMT(info, "too many addresses or duplicate one: %d", ret); 1034 goto out_free; 1035 } 1036 1037 mptcp_nl_add_subflow_or_signal_addr(sock_net(skb->sk), &entry->addr); 1038 return 0; 1039 1040 out_free: 1041 __mptcp_pm_release_addr_entry(entry); 1042 return ret; 1043 } 1044 1045 static bool mptcp_pm_remove_anno_addr(struct mptcp_sock *msk, 1046 const struct mptcp_addr_info *addr, 1047 bool force) 1048 { 1049 struct mptcp_rm_list list = { .nr = 0 }; 1050 bool ret; 1051 1052 list.ids[list.nr++] = mptcp_endp_get_local_id(msk, addr); 1053 1054 ret = mptcp_remove_anno_list_by_saddr(msk, addr); 1055 if (ret || force) { 1056 spin_lock_bh(&msk->pm.lock); 1057 if (ret) { 1058 __set_bit(addr->id, msk->pm.id_avail_bitmap); 1059 msk->pm.add_addr_signaled--; 1060 } 1061 mptcp_pm_remove_addr(msk, &list); 1062 spin_unlock_bh(&msk->pm.lock); 1063 } 1064 return ret; 1065 } 1066 1067 static void __mark_subflow_endp_available(struct mptcp_sock *msk, u8 id) 1068 { 1069 /* If it was marked as used, and not ID 0, decrement local_addr_used */ 1070 if (!__test_and_set_bit(id ? : msk->mpc_endpoint_id, msk->pm.id_avail_bitmap) && 1071 id && !WARN_ON_ONCE(msk->pm.local_addr_used == 0)) 1072 msk->pm.local_addr_used--; 1073 } 1074 1075 static int mptcp_nl_remove_subflow_and_signal_addr(struct net *net, 1076 const struct mptcp_pm_addr_entry *entry) 1077 { 1078 const struct mptcp_addr_info *addr = &entry->addr; 1079 struct mptcp_rm_list list = { .nr = 1 }; 1080 long s_slot = 0, s_num = 0; 1081 struct mptcp_sock *msk; 1082 1083 pr_debug("remove_id=%d\n", addr->id); 1084 1085 while ((msk = mptcp_token_iter_next(net, &s_slot, &s_num)) != NULL) { 1086 struct sock *sk = (struct sock *)msk; 1087 bool remove_subflow; 1088 1089 if (mptcp_pm_is_userspace(msk)) 1090 goto next; 1091 1092 lock_sock(sk); 1093 remove_subflow = mptcp_lookup_subflow_by_saddr(&msk->conn_list, addr); 1094 mptcp_pm_remove_anno_addr(msk, addr, remove_subflow && 1095 !(entry->flags & MPTCP_PM_ADDR_FLAG_IMPLICIT)); 1096 1097 list.ids[0] = mptcp_endp_get_local_id(msk, addr); 1098 if (remove_subflow) { 1099 spin_lock_bh(&msk->pm.lock); 1100 mptcp_pm_rm_subflow(msk, &list); 1101 spin_unlock_bh(&msk->pm.lock); 1102 } 1103 1104 if (entry->flags & MPTCP_PM_ADDR_FLAG_SUBFLOW) { 1105 spin_lock_bh(&msk->pm.lock); 1106 __mark_subflow_endp_available(msk, list.ids[0]); 1107 spin_unlock_bh(&msk->pm.lock); 1108 } 1109 1110 if (msk->mpc_endpoint_id == entry->addr.id) 1111 msk->mpc_endpoint_id = 0; 1112 release_sock(sk); 1113 1114 next: 1115 sock_put(sk); 1116 cond_resched(); 1117 } 1118 1119 return 0; 1120 } 1121 1122 static int mptcp_nl_remove_id_zero_address(struct net *net, 1123 struct mptcp_addr_info *addr) 1124 { 1125 struct mptcp_rm_list list = { .nr = 0 }; 1126 long s_slot = 0, s_num = 0; 1127 struct mptcp_sock *msk; 1128 1129 list.ids[list.nr++] = 0; 1130 1131 while ((msk = mptcp_token_iter_next(net, &s_slot, &s_num)) != NULL) { 1132 struct sock *sk = (struct sock *)msk; 1133 struct mptcp_addr_info msk_local; 1134 1135 if (list_empty(&msk->conn_list) || mptcp_pm_is_userspace(msk)) 1136 goto next; 1137 1138 mptcp_local_address((struct sock_common *)msk, &msk_local); 1139 if (!mptcp_addresses_equal(&msk_local, addr, addr->port)) 1140 goto next; 1141 1142 lock_sock(sk); 1143 spin_lock_bh(&msk->pm.lock); 1144 mptcp_pm_remove_addr(msk, &list); 1145 mptcp_pm_rm_subflow(msk, &list); 1146 __mark_subflow_endp_available(msk, 0); 1147 spin_unlock_bh(&msk->pm.lock); 1148 release_sock(sk); 1149 1150 next: 1151 sock_put(sk); 1152 cond_resched(); 1153 } 1154 1155 return 0; 1156 } 1157 1158 /* Remove an MPTCP endpoint */ 1159 int mptcp_pm_nl_del_addr_doit(struct sk_buff *skb, struct genl_info *info) 1160 { 1161 struct pm_nl_pernet *pernet = genl_info_pm_nl(info); 1162 struct mptcp_pm_addr_entry addr, *entry; 1163 struct nlattr *attr; 1164 u8 addr_max; 1165 int ret; 1166 1167 if (GENL_REQ_ATTR_CHECK(info, MPTCP_PM_ENDPOINT_ADDR)) 1168 return -EINVAL; 1169 1170 attr = info->attrs[MPTCP_PM_ENDPOINT_ADDR]; 1171 ret = mptcp_pm_parse_entry(attr, info, false, &addr); 1172 if (ret < 0) 1173 return ret; 1174 1175 /* the zero id address is special: the first address used by the msk 1176 * always gets such an id, so different subflows can have different zero 1177 * id addresses. Additionally zero id is not accounted for in id_bitmap. 1178 * Let's use an 'mptcp_rm_list' instead of the common remove code. 1179 */ 1180 if (addr.addr.id == 0) 1181 return mptcp_nl_remove_id_zero_address(sock_net(skb->sk), &addr.addr); 1182 1183 spin_lock_bh(&pernet->lock); 1184 entry = __lookup_addr_by_id(pernet, addr.addr.id); 1185 if (!entry) { 1186 NL_SET_ERR_MSG_ATTR(info->extack, attr, "address not found"); 1187 spin_unlock_bh(&pernet->lock); 1188 return -EINVAL; 1189 } 1190 if (entry->flags & MPTCP_PM_ADDR_FLAG_SIGNAL) { 1191 addr_max = pernet->endp_signal_max; 1192 WRITE_ONCE(pernet->endp_signal_max, addr_max - 1); 1193 } 1194 if (entry->flags & MPTCP_PM_ADDR_FLAG_SUBFLOW) { 1195 addr_max = pernet->endp_subflow_max; 1196 WRITE_ONCE(pernet->endp_subflow_max, addr_max - 1); 1197 } 1198 if (entry->flags & MPTCP_PM_ADDR_FLAG_LAMINAR) { 1199 addr_max = pernet->endp_laminar_max; 1200 WRITE_ONCE(pernet->endp_laminar_max, addr_max - 1); 1201 } 1202 if (entry->flags & MPTCP_PM_ADDR_FLAG_FULLMESH) { 1203 addr_max = pernet->endp_fullmesh_max; 1204 WRITE_ONCE(pernet->endp_fullmesh_max, addr_max - 1); 1205 } 1206 1207 pernet->endpoints--; 1208 list_del_rcu(&entry->list); 1209 __clear_bit(entry->addr.id, pernet->id_bitmap); 1210 spin_unlock_bh(&pernet->lock); 1211 1212 mptcp_nl_remove_subflow_and_signal_addr(sock_net(skb->sk), entry); 1213 synchronize_rcu(); 1214 __mptcp_pm_release_addr_entry(entry); 1215 1216 return ret; 1217 } 1218 1219 static void mptcp_pm_flush_addrs_and_subflows(struct mptcp_sock *msk, 1220 struct list_head *rm_list) 1221 { 1222 struct mptcp_rm_list alist = { .nr = 0 }, slist = { .nr = 0 }; 1223 struct mptcp_pm_addr_entry *entry; 1224 1225 list_for_each_entry(entry, rm_list, list) { 1226 if (slist.nr < MPTCP_RM_IDS_MAX && 1227 mptcp_lookup_subflow_by_saddr(&msk->conn_list, &entry->addr)) 1228 slist.ids[slist.nr++] = mptcp_endp_get_local_id(msk, &entry->addr); 1229 1230 if (alist.nr < MPTCP_RM_IDS_MAX && 1231 mptcp_remove_anno_list_by_saddr(msk, &entry->addr)) 1232 alist.ids[alist.nr++] = mptcp_endp_get_local_id(msk, &entry->addr); 1233 } 1234 1235 spin_lock_bh(&msk->pm.lock); 1236 if (alist.nr) { 1237 msk->pm.add_addr_signaled -= alist.nr; 1238 mptcp_pm_remove_addr(msk, &alist); 1239 } 1240 if (slist.nr) 1241 mptcp_pm_rm_subflow(msk, &slist); 1242 /* Reset counters: maybe some subflows have been removed before */ 1243 bitmap_fill(msk->pm.id_avail_bitmap, MPTCP_PM_MAX_ADDR_ID + 1); 1244 msk->pm.local_addr_used = 0; 1245 spin_unlock_bh(&msk->pm.lock); 1246 } 1247 1248 static void mptcp_nl_flush_addrs_list(struct net *net, 1249 struct list_head *rm_list) 1250 { 1251 long s_slot = 0, s_num = 0; 1252 struct mptcp_sock *msk; 1253 1254 if (list_empty(rm_list)) 1255 return; 1256 1257 while ((msk = mptcp_token_iter_next(net, &s_slot, &s_num)) != NULL) { 1258 struct sock *sk = (struct sock *)msk; 1259 1260 if (!mptcp_pm_is_userspace(msk)) { 1261 lock_sock(sk); 1262 mptcp_pm_flush_addrs_and_subflows(msk, rm_list); 1263 release_sock(sk); 1264 } 1265 1266 sock_put(sk); 1267 cond_resched(); 1268 } 1269 } 1270 1271 /* caller must ensure the RCU grace period is already elapsed */ 1272 static void __flush_addrs(struct list_head *list) 1273 { 1274 while (!list_empty(list)) { 1275 struct mptcp_pm_addr_entry *cur; 1276 1277 cur = list_entry(list->next, 1278 struct mptcp_pm_addr_entry, list); 1279 list_del_rcu(&cur->list); 1280 __mptcp_pm_release_addr_entry(cur); 1281 } 1282 } 1283 1284 static void __reset_counters(struct pm_nl_pernet *pernet) 1285 { 1286 WRITE_ONCE(pernet->endp_signal_max, 0); 1287 WRITE_ONCE(pernet->endp_subflow_max, 0); 1288 WRITE_ONCE(pernet->endp_laminar_max, 0); 1289 pernet->endpoints = 0; 1290 } 1291 1292 int mptcp_pm_nl_flush_addrs_doit(struct sk_buff *skb, struct genl_info *info) 1293 { 1294 struct pm_nl_pernet *pernet = genl_info_pm_nl(info); 1295 LIST_HEAD(free_list); 1296 1297 spin_lock_bh(&pernet->lock); 1298 list_splice_init(&pernet->endp_list, &free_list); 1299 __reset_counters(pernet); 1300 pernet->next_id = 1; 1301 bitmap_zero(pernet->id_bitmap, MPTCP_PM_MAX_ADDR_ID + 1); 1302 spin_unlock_bh(&pernet->lock); 1303 mptcp_nl_flush_addrs_list(sock_net(skb->sk), &free_list); 1304 synchronize_rcu(); 1305 __flush_addrs(&free_list); 1306 return 0; 1307 } 1308 1309 int mptcp_pm_nl_get_addr(u8 id, struct mptcp_pm_addr_entry *addr, 1310 struct genl_info *info) 1311 { 1312 struct pm_nl_pernet *pernet = genl_info_pm_nl(info); 1313 struct mptcp_pm_addr_entry *entry; 1314 int ret = -EINVAL; 1315 1316 rcu_read_lock(); 1317 entry = __lookup_addr_by_id(pernet, id); 1318 if (entry) { 1319 *addr = *entry; 1320 ret = 0; 1321 } 1322 rcu_read_unlock(); 1323 1324 return ret; 1325 } 1326 1327 int mptcp_pm_nl_dump_addr(struct sk_buff *msg, 1328 struct netlink_callback *cb) 1329 { 1330 struct net *net = sock_net(msg->sk); 1331 struct mptcp_pm_addr_entry *entry; 1332 struct pm_nl_pernet *pernet; 1333 int id = cb->args[0]; 1334 int i; 1335 1336 pernet = pm_nl_get_pernet(net); 1337 1338 rcu_read_lock(); 1339 for (i = id; i < MPTCP_PM_MAX_ADDR_ID + 1; i++) { 1340 if (test_bit(i, pernet->id_bitmap)) { 1341 entry = __lookup_addr_by_id(pernet, i); 1342 if (!entry) 1343 break; 1344 1345 if (entry->addr.id <= id) 1346 continue; 1347 1348 if (mptcp_pm_genl_fill_addr(msg, cb, entry) < 0) 1349 break; 1350 1351 id = entry->addr.id; 1352 } 1353 } 1354 rcu_read_unlock(); 1355 1356 cb->args[0] = id; 1357 return msg->len; 1358 } 1359 1360 static int parse_limit(struct genl_info *info, int id, unsigned int *limit) 1361 { 1362 struct nlattr *attr = info->attrs[id]; 1363 1364 if (!attr) 1365 return 0; 1366 1367 *limit = nla_get_u32(attr); 1368 if (*limit > MPTCP_PM_ADDR_MAX) { 1369 NL_SET_ERR_MSG_ATTR_FMT(info->extack, attr, 1370 "limit greater than maximum (%u)", 1371 MPTCP_PM_ADDR_MAX); 1372 return -EINVAL; 1373 } 1374 return 0; 1375 } 1376 1377 int mptcp_pm_nl_set_limits_doit(struct sk_buff *skb, struct genl_info *info) 1378 { 1379 struct pm_nl_pernet *pernet = genl_info_pm_nl(info); 1380 unsigned int rcv_addrs, subflows; 1381 int ret; 1382 1383 spin_lock_bh(&pernet->lock); 1384 rcv_addrs = pernet->limit_add_addr_accepted; 1385 ret = parse_limit(info, MPTCP_PM_ATTR_RCV_ADD_ADDRS, &rcv_addrs); 1386 if (ret) 1387 goto unlock; 1388 1389 subflows = pernet->limit_extra_subflows; 1390 ret = parse_limit(info, MPTCP_PM_ATTR_SUBFLOWS, &subflows); 1391 if (ret) 1392 goto unlock; 1393 1394 WRITE_ONCE(pernet->limit_add_addr_accepted, rcv_addrs); 1395 WRITE_ONCE(pernet->limit_extra_subflows, subflows); 1396 1397 unlock: 1398 spin_unlock_bh(&pernet->lock); 1399 return ret; 1400 } 1401 1402 int mptcp_pm_nl_get_limits_doit(struct sk_buff *skb, struct genl_info *info) 1403 { 1404 struct pm_nl_pernet *pernet = genl_info_pm_nl(info); 1405 struct sk_buff *msg; 1406 void *reply; 1407 1408 msg = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL); 1409 if (!msg) 1410 return -ENOMEM; 1411 1412 reply = genlmsg_put_reply(msg, info, &mptcp_genl_family, 0, 1413 MPTCP_PM_CMD_GET_LIMITS); 1414 if (!reply) 1415 goto fail; 1416 1417 if (nla_put_u32(msg, MPTCP_PM_ATTR_RCV_ADD_ADDRS, 1418 READ_ONCE(pernet->limit_add_addr_accepted))) 1419 goto fail; 1420 1421 if (nla_put_u32(msg, MPTCP_PM_ATTR_SUBFLOWS, 1422 READ_ONCE(pernet->limit_extra_subflows))) 1423 goto fail; 1424 1425 genlmsg_end(msg, reply); 1426 return genlmsg_reply(msg, info); 1427 1428 fail: 1429 GENL_SET_ERR_MSG(info, "not enough space in Netlink message"); 1430 nlmsg_free(msg); 1431 return -EMSGSIZE; 1432 } 1433 1434 static void mptcp_pm_nl_fullmesh(struct mptcp_sock *msk, 1435 struct mptcp_addr_info *addr) 1436 { 1437 struct mptcp_rm_list list = { .nr = 0 }; 1438 1439 list.ids[list.nr++] = mptcp_endp_get_local_id(msk, addr); 1440 1441 spin_lock_bh(&msk->pm.lock); 1442 mptcp_pm_rm_subflow(msk, &list); 1443 __mark_subflow_endp_available(msk, list.ids[0]); 1444 mptcp_pm_create_subflow_or_signal_addr(msk); 1445 spin_unlock_bh(&msk->pm.lock); 1446 } 1447 1448 static void mptcp_pm_nl_set_flags_all(struct net *net, 1449 struct mptcp_pm_addr_entry *local, 1450 u8 changed) 1451 { 1452 u8 is_subflow = !!(local->flags & MPTCP_PM_ADDR_FLAG_SUBFLOW); 1453 u8 bkup = !!(local->flags & MPTCP_PM_ADDR_FLAG_BACKUP); 1454 long s_slot = 0, s_num = 0; 1455 struct mptcp_sock *msk; 1456 1457 if (changed == MPTCP_PM_ADDR_FLAG_FULLMESH && !is_subflow) 1458 return; 1459 1460 while ((msk = mptcp_token_iter_next(net, &s_slot, &s_num)) != NULL) { 1461 struct sock *sk = (struct sock *)msk; 1462 1463 if (list_empty(&msk->conn_list) || mptcp_pm_is_userspace(msk)) 1464 goto next; 1465 1466 lock_sock(sk); 1467 if (changed & MPTCP_PM_ADDR_FLAG_BACKUP) 1468 mptcp_pm_mp_prio_send_ack(msk, &local->addr, NULL, bkup); 1469 /* Subflows will only be recreated if the SUBFLOW flag is set */ 1470 if (is_subflow && (changed & MPTCP_PM_ADDR_FLAG_FULLMESH)) 1471 mptcp_pm_nl_fullmesh(msk, &local->addr); 1472 release_sock(sk); 1473 1474 next: 1475 sock_put(sk); 1476 cond_resched(); 1477 } 1478 } 1479 1480 int mptcp_pm_nl_set_flags(struct mptcp_pm_addr_entry *local, 1481 struct genl_info *info) 1482 { 1483 struct nlattr *attr = info->attrs[MPTCP_PM_ATTR_ADDR]; 1484 u8 changed, mask = MPTCP_PM_ADDR_FLAG_BACKUP | 1485 MPTCP_PM_ADDR_FLAG_FULLMESH; 1486 struct net *net = genl_info_net(info); 1487 struct mptcp_pm_addr_entry *entry; 1488 struct pm_nl_pernet *pernet; 1489 u8 lookup_by_id = 0; 1490 1491 pernet = pm_nl_get_pernet(net); 1492 1493 if (local->addr.family == AF_UNSPEC) { 1494 lookup_by_id = 1; 1495 if (!local->addr.id) { 1496 NL_SET_ERR_MSG_ATTR(info->extack, attr, 1497 "missing address ID"); 1498 return -EOPNOTSUPP; 1499 } 1500 } 1501 1502 spin_lock_bh(&pernet->lock); 1503 entry = lookup_by_id ? __lookup_addr_by_id(pernet, local->addr.id) : 1504 __lookup_addr(pernet, &local->addr); 1505 if (!entry) { 1506 spin_unlock_bh(&pernet->lock); 1507 NL_SET_ERR_MSG_ATTR(info->extack, attr, "address not found"); 1508 return -EINVAL; 1509 } 1510 if ((local->flags & MPTCP_PM_ADDR_FLAG_FULLMESH) && 1511 (entry->flags & (MPTCP_PM_ADDR_FLAG_SIGNAL | 1512 MPTCP_PM_ADDR_FLAG_IMPLICIT))) { 1513 spin_unlock_bh(&pernet->lock); 1514 NL_SET_ERR_MSG_ATTR(info->extack, attr, "invalid addr flags"); 1515 return -EINVAL; 1516 } 1517 1518 changed = (local->flags ^ entry->flags) & mask; 1519 entry->flags = (entry->flags & ~mask) | (local->flags & mask); 1520 *local = *entry; 1521 1522 if (changed & MPTCP_PM_ADDR_FLAG_FULLMESH) { 1523 u8 addr_max = pernet->endp_fullmesh_max; 1524 1525 if (entry->flags & MPTCP_PM_ADDR_FLAG_FULLMESH) 1526 addr_max++; 1527 else 1528 addr_max--; 1529 1530 WRITE_ONCE(pernet->endp_fullmesh_max, addr_max); 1531 } 1532 1533 spin_unlock_bh(&pernet->lock); 1534 1535 mptcp_pm_nl_set_flags_all(net, local, changed); 1536 return 0; 1537 } 1538 1539 bool mptcp_pm_nl_check_work_pending(struct mptcp_sock *msk) 1540 { 1541 struct pm_nl_pernet *pernet = pm_nl_get_pernet_from_msk(msk); 1542 1543 if (msk->pm.extra_subflows == mptcp_pm_get_limit_extra_subflows(msk) || 1544 (find_next_and_bit(pernet->id_bitmap, msk->pm.id_avail_bitmap, 1545 MPTCP_PM_MAX_ADDR_ID + 1, 0) == MPTCP_PM_MAX_ADDR_ID + 1)) { 1546 WRITE_ONCE(msk->pm.work_pending, false); 1547 return false; 1548 } 1549 return true; 1550 } 1551 1552 /* Called under PM lock */ 1553 void __mptcp_pm_kernel_worker(struct mptcp_sock *msk) 1554 { 1555 struct mptcp_pm_data *pm = &msk->pm; 1556 1557 if (pm->status & BIT(MPTCP_PM_ADD_ADDR_RECEIVED)) { 1558 pm->status &= ~BIT(MPTCP_PM_ADD_ADDR_RECEIVED); 1559 mptcp_pm_nl_add_addr_received(msk); 1560 } 1561 if (pm->status & BIT(MPTCP_PM_ESTABLISHED)) { 1562 pm->status &= ~BIT(MPTCP_PM_ESTABLISHED); 1563 mptcp_pm_nl_fully_established(msk); 1564 } 1565 if (pm->status & BIT(MPTCP_PM_SUBFLOW_ESTABLISHED)) { 1566 pm->status &= ~BIT(MPTCP_PM_SUBFLOW_ESTABLISHED); 1567 mptcp_pm_nl_subflow_established(msk); 1568 } 1569 } 1570 1571 static int __net_init pm_nl_init_net(struct net *net) 1572 { 1573 struct pm_nl_pernet *pernet = pm_nl_get_pernet(net); 1574 1575 INIT_LIST_HEAD_RCU(&pernet->endp_list); 1576 1577 /* Cit. 2 subflows ought to be enough for anybody. */ 1578 pernet->limit_extra_subflows = 2; 1579 pernet->next_id = 1; 1580 spin_lock_init(&pernet->lock); 1581 1582 /* No need to initialize other pernet fields, the struct is zeroed at 1583 * allocation time. 1584 */ 1585 1586 return 0; 1587 } 1588 1589 static void __net_exit pm_nl_exit_net(struct list_head *net_list) 1590 { 1591 struct net *net; 1592 1593 list_for_each_entry(net, net_list, exit_list) { 1594 struct pm_nl_pernet *pernet = pm_nl_get_pernet(net); 1595 1596 /* net is removed from namespace list, can't race with 1597 * other modifiers, also netns core already waited for a 1598 * RCU grace period. 1599 */ 1600 __flush_addrs(&pernet->endp_list); 1601 } 1602 } 1603 1604 static struct pernet_operations mptcp_pm_pernet_ops = { 1605 .init = pm_nl_init_net, 1606 .exit_batch = pm_nl_exit_net, 1607 .id = &pm_nl_pernet_id, 1608 .size = sizeof(struct pm_nl_pernet), 1609 }; 1610 1611 struct mptcp_pm_ops mptcp_pm_kernel = { 1612 .name = "kernel", 1613 .owner = THIS_MODULE, 1614 }; 1615 1616 void __init mptcp_pm_kernel_register(void) 1617 { 1618 if (register_pernet_subsys(&mptcp_pm_pernet_ops) < 0) 1619 panic("Failed to register MPTCP PM pernet subsystem.\n"); 1620 1621 mptcp_pm_register(&mptcp_pm_kernel); 1622 } 1623