1 // SPDX-License-Identifier: GPL-2.0 2 /* Multipath TCP 3 * 4 * Copyright (c) 2019, Intel Corporation. 5 */ 6 #define pr_fmt(fmt) "MPTCP: " fmt 7 8 #include <linux/kernel.h> 9 #include <net/tcp.h> 10 #include <net/mptcp.h> 11 #include "protocol.h" 12 13 #include "mib.h" 14 15 /* path manager command handlers */ 16 17 int mptcp_pm_announce_addr(struct mptcp_sock *msk, 18 const struct mptcp_addr_info *addr, 19 bool echo) 20 { 21 u8 add_addr = READ_ONCE(msk->pm.addr_signal); 22 23 pr_debug("msk=%p, local_id=%d, echo=%d", msk, addr->id, echo); 24 25 lockdep_assert_held(&msk->pm.lock); 26 27 if (add_addr & 28 (echo ? BIT(MPTCP_ADD_ADDR_ECHO) : BIT(MPTCP_ADD_ADDR_SIGNAL))) { 29 MPTCP_INC_STATS(sock_net((struct sock *)msk), 30 echo ? MPTCP_MIB_ECHOADDTXDROP : MPTCP_MIB_ADDADDRTXDROP); 31 return -EINVAL; 32 } 33 34 if (echo) { 35 msk->pm.remote = *addr; 36 add_addr |= BIT(MPTCP_ADD_ADDR_ECHO); 37 } else { 38 msk->pm.local = *addr; 39 add_addr |= BIT(MPTCP_ADD_ADDR_SIGNAL); 40 } 41 WRITE_ONCE(msk->pm.addr_signal, add_addr); 42 return 0; 43 } 44 45 int mptcp_pm_remove_addr(struct mptcp_sock *msk, const struct mptcp_rm_list *rm_list) 46 { 47 u8 rm_addr = READ_ONCE(msk->pm.addr_signal); 48 49 pr_debug("msk=%p, rm_list_nr=%d", msk, rm_list->nr); 50 51 if (rm_addr) { 52 MPTCP_ADD_STATS(sock_net((struct sock *)msk), 53 MPTCP_MIB_RMADDRTXDROP, rm_list->nr); 54 return -EINVAL; 55 } 56 57 msk->pm.rm_list_tx = *rm_list; 58 rm_addr |= BIT(MPTCP_RM_ADDR_SIGNAL); 59 WRITE_ONCE(msk->pm.addr_signal, rm_addr); 60 mptcp_pm_nl_addr_send_ack(msk); 61 return 0; 62 } 63 64 int mptcp_pm_remove_subflow(struct mptcp_sock *msk, const struct mptcp_rm_list *rm_list) 65 { 66 pr_debug("msk=%p, rm_list_nr=%d", msk, rm_list->nr); 67 68 spin_lock_bh(&msk->pm.lock); 69 mptcp_pm_nl_rm_subflow_received(msk, rm_list); 70 spin_unlock_bh(&msk->pm.lock); 71 return 0; 72 } 73 74 /* path manager event handlers */ 75 76 void mptcp_pm_new_connection(struct mptcp_sock *msk, const struct sock *ssk, int server_side) 77 { 78 struct mptcp_pm_data *pm = &msk->pm; 79 80 pr_debug("msk=%p, token=%u side=%d", msk, msk->token, server_side); 81 82 WRITE_ONCE(pm->server_side, server_side); 83 mptcp_event(MPTCP_EVENT_CREATED, msk, ssk, GFP_ATOMIC); 84 } 85 86 bool mptcp_pm_allow_new_subflow(struct mptcp_sock *msk) 87 { 88 struct mptcp_pm_data *pm = &msk->pm; 89 unsigned int subflows_max; 90 int ret = 0; 91 92 if (mptcp_pm_is_userspace(msk)) { 93 if (mptcp_userspace_pm_active(msk)) { 94 spin_lock_bh(&pm->lock); 95 pm->subflows++; 96 spin_unlock_bh(&pm->lock); 97 return true; 98 } 99 return false; 100 } 101 102 subflows_max = mptcp_pm_get_subflows_max(msk); 103 104 pr_debug("msk=%p subflows=%d max=%d allow=%d", msk, pm->subflows, 105 subflows_max, READ_ONCE(pm->accept_subflow)); 106 107 /* try to avoid acquiring the lock below */ 108 if (!READ_ONCE(pm->accept_subflow)) 109 return false; 110 111 spin_lock_bh(&pm->lock); 112 if (READ_ONCE(pm->accept_subflow)) { 113 ret = pm->subflows < subflows_max; 114 if (ret && ++pm->subflows == subflows_max) 115 WRITE_ONCE(pm->accept_subflow, false); 116 } 117 spin_unlock_bh(&pm->lock); 118 119 return ret; 120 } 121 122 /* return true if the new status bit is currently cleared, that is, this event 123 * can be server, eventually by an already scheduled work 124 */ 125 static bool mptcp_pm_schedule_work(struct mptcp_sock *msk, 126 enum mptcp_pm_status new_status) 127 { 128 pr_debug("msk=%p status=%x new=%lx", msk, msk->pm.status, 129 BIT(new_status)); 130 if (msk->pm.status & BIT(new_status)) 131 return false; 132 133 msk->pm.status |= BIT(new_status); 134 mptcp_schedule_work((struct sock *)msk); 135 return true; 136 } 137 138 void mptcp_pm_fully_established(struct mptcp_sock *msk, const struct sock *ssk) 139 { 140 struct mptcp_pm_data *pm = &msk->pm; 141 bool announce = false; 142 143 pr_debug("msk=%p", msk); 144 145 spin_lock_bh(&pm->lock); 146 147 /* mptcp_pm_fully_established() can be invoked by multiple 148 * racing paths - accept() and check_fully_established() 149 * be sure to serve this event only once. 150 */ 151 if (READ_ONCE(pm->work_pending) && 152 !(msk->pm.status & BIT(MPTCP_PM_ALREADY_ESTABLISHED))) 153 mptcp_pm_schedule_work(msk, MPTCP_PM_ESTABLISHED); 154 155 if ((msk->pm.status & BIT(MPTCP_PM_ALREADY_ESTABLISHED)) == 0) 156 announce = true; 157 158 msk->pm.status |= BIT(MPTCP_PM_ALREADY_ESTABLISHED); 159 spin_unlock_bh(&pm->lock); 160 161 if (announce) 162 mptcp_event(MPTCP_EVENT_ESTABLISHED, msk, ssk, GFP_ATOMIC); 163 } 164 165 void mptcp_pm_connection_closed(struct mptcp_sock *msk) 166 { 167 pr_debug("msk=%p", msk); 168 } 169 170 void mptcp_pm_subflow_established(struct mptcp_sock *msk) 171 { 172 struct mptcp_pm_data *pm = &msk->pm; 173 174 pr_debug("msk=%p", msk); 175 176 if (!READ_ONCE(pm->work_pending)) 177 return; 178 179 spin_lock_bh(&pm->lock); 180 181 if (READ_ONCE(pm->work_pending)) 182 mptcp_pm_schedule_work(msk, MPTCP_PM_SUBFLOW_ESTABLISHED); 183 184 spin_unlock_bh(&pm->lock); 185 } 186 187 void mptcp_pm_subflow_check_next(struct mptcp_sock *msk, const struct sock *ssk, 188 const struct mptcp_subflow_context *subflow) 189 { 190 struct mptcp_pm_data *pm = &msk->pm; 191 bool update_subflows; 192 193 update_subflows = subflow->request_join || subflow->mp_join; 194 if (mptcp_pm_is_userspace(msk)) { 195 if (update_subflows) { 196 spin_lock_bh(&pm->lock); 197 pm->subflows--; 198 spin_unlock_bh(&pm->lock); 199 } 200 return; 201 } 202 203 if (!READ_ONCE(pm->work_pending) && !update_subflows) 204 return; 205 206 spin_lock_bh(&pm->lock); 207 if (update_subflows) 208 __mptcp_pm_close_subflow(msk); 209 210 /* Even if this subflow is not really established, tell the PM to try 211 * to pick the next ones, if possible. 212 */ 213 if (mptcp_pm_nl_check_work_pending(msk)) 214 mptcp_pm_schedule_work(msk, MPTCP_PM_SUBFLOW_ESTABLISHED); 215 216 spin_unlock_bh(&pm->lock); 217 } 218 219 void mptcp_pm_add_addr_received(const struct sock *ssk, 220 const struct mptcp_addr_info *addr) 221 { 222 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk); 223 struct mptcp_sock *msk = mptcp_sk(subflow->conn); 224 struct mptcp_pm_data *pm = &msk->pm; 225 226 pr_debug("msk=%p remote_id=%d accept=%d", msk, addr->id, 227 READ_ONCE(pm->accept_addr)); 228 229 mptcp_event_addr_announced(ssk, addr); 230 231 spin_lock_bh(&pm->lock); 232 233 if (mptcp_pm_is_userspace(msk)) { 234 if (mptcp_userspace_pm_active(msk)) { 235 mptcp_pm_announce_addr(msk, addr, true); 236 mptcp_pm_add_addr_send_ack(msk); 237 } else { 238 __MPTCP_INC_STATS(sock_net((struct sock *)msk), MPTCP_MIB_ADDADDRDROP); 239 } 240 } else if (!READ_ONCE(pm->accept_addr)) { 241 mptcp_pm_announce_addr(msk, addr, true); 242 mptcp_pm_add_addr_send_ack(msk); 243 } else if (mptcp_pm_schedule_work(msk, MPTCP_PM_ADD_ADDR_RECEIVED)) { 244 pm->remote = *addr; 245 } else { 246 __MPTCP_INC_STATS(sock_net((struct sock *)msk), MPTCP_MIB_ADDADDRDROP); 247 } 248 249 spin_unlock_bh(&pm->lock); 250 } 251 252 void mptcp_pm_add_addr_echoed(struct mptcp_sock *msk, 253 const struct mptcp_addr_info *addr) 254 { 255 struct mptcp_pm_data *pm = &msk->pm; 256 257 pr_debug("msk=%p", msk); 258 259 spin_lock_bh(&pm->lock); 260 261 if (mptcp_lookup_anno_list_by_saddr(msk, addr) && READ_ONCE(pm->work_pending)) 262 mptcp_pm_schedule_work(msk, MPTCP_PM_SUBFLOW_ESTABLISHED); 263 264 spin_unlock_bh(&pm->lock); 265 } 266 267 void mptcp_pm_add_addr_send_ack(struct mptcp_sock *msk) 268 { 269 if (!mptcp_pm_should_add_signal(msk)) 270 return; 271 272 mptcp_pm_schedule_work(msk, MPTCP_PM_ADD_ADDR_SEND_ACK); 273 } 274 275 void mptcp_pm_rm_addr_received(struct mptcp_sock *msk, 276 const struct mptcp_rm_list *rm_list) 277 { 278 struct mptcp_pm_data *pm = &msk->pm; 279 u8 i; 280 281 pr_debug("msk=%p remote_ids_nr=%d", msk, rm_list->nr); 282 283 for (i = 0; i < rm_list->nr; i++) 284 mptcp_event_addr_removed(msk, rm_list->ids[i]); 285 286 spin_lock_bh(&pm->lock); 287 if (mptcp_pm_schedule_work(msk, MPTCP_PM_RM_ADDR_RECEIVED)) 288 pm->rm_list_rx = *rm_list; 289 else 290 __MPTCP_INC_STATS(sock_net((struct sock *)msk), MPTCP_MIB_RMADDRDROP); 291 spin_unlock_bh(&pm->lock); 292 } 293 294 void mptcp_pm_mp_prio_received(struct sock *ssk, u8 bkup) 295 { 296 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk); 297 struct sock *sk = subflow->conn; 298 struct mptcp_sock *msk; 299 300 pr_debug("subflow->backup=%d, bkup=%d\n", subflow->backup, bkup); 301 msk = mptcp_sk(sk); 302 if (subflow->backup != bkup) { 303 subflow->backup = bkup; 304 mptcp_data_lock(sk); 305 if (!sock_owned_by_user(sk)) 306 msk->last_snd = NULL; 307 else 308 __set_bit(MPTCP_RESET_SCHEDULER, &msk->cb_flags); 309 mptcp_data_unlock(sk); 310 } 311 312 mptcp_event(MPTCP_EVENT_SUB_PRIORITY, msk, ssk, GFP_ATOMIC); 313 } 314 315 void mptcp_pm_mp_fail_received(struct sock *sk, u64 fail_seq) 316 { 317 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk); 318 struct mptcp_sock *msk = mptcp_sk(subflow->conn); 319 320 pr_debug("fail_seq=%llu", fail_seq); 321 322 if (!READ_ONCE(msk->allow_infinite_fallback)) 323 return; 324 325 if (!subflow->fail_tout) { 326 pr_debug("send MP_FAIL response and infinite map"); 327 328 subflow->send_mp_fail = 1; 329 subflow->send_infinite_map = 1; 330 tcp_send_ack(sk); 331 } else { 332 pr_debug("MP_FAIL response received"); 333 WRITE_ONCE(subflow->fail_tout, 0); 334 } 335 } 336 337 /* path manager helpers */ 338 339 bool mptcp_pm_add_addr_signal(struct mptcp_sock *msk, const struct sk_buff *skb, 340 unsigned int opt_size, unsigned int remaining, 341 struct mptcp_addr_info *addr, bool *echo, 342 bool *drop_other_suboptions) 343 { 344 int ret = false; 345 u8 add_addr; 346 u8 family; 347 bool port; 348 349 spin_lock_bh(&msk->pm.lock); 350 351 /* double check after the lock is acquired */ 352 if (!mptcp_pm_should_add_signal(msk)) 353 goto out_unlock; 354 355 /* always drop every other options for pure ack ADD_ADDR; this is a 356 * plain dup-ack from TCP perspective. The other MPTCP-relevant info, 357 * if any, will be carried by the 'original' TCP ack 358 */ 359 if (skb && skb_is_tcp_pure_ack(skb)) { 360 remaining += opt_size; 361 *drop_other_suboptions = true; 362 } 363 364 *echo = mptcp_pm_should_add_signal_echo(msk); 365 port = !!(*echo ? msk->pm.remote.port : msk->pm.local.port); 366 367 family = *echo ? msk->pm.remote.family : msk->pm.local.family; 368 if (remaining < mptcp_add_addr_len(family, *echo, port)) 369 goto out_unlock; 370 371 if (*echo) { 372 *addr = msk->pm.remote; 373 add_addr = msk->pm.addr_signal & ~BIT(MPTCP_ADD_ADDR_ECHO); 374 } else { 375 *addr = msk->pm.local; 376 add_addr = msk->pm.addr_signal & ~BIT(MPTCP_ADD_ADDR_SIGNAL); 377 } 378 WRITE_ONCE(msk->pm.addr_signal, add_addr); 379 ret = true; 380 381 out_unlock: 382 spin_unlock_bh(&msk->pm.lock); 383 return ret; 384 } 385 386 bool mptcp_pm_rm_addr_signal(struct mptcp_sock *msk, unsigned int remaining, 387 struct mptcp_rm_list *rm_list) 388 { 389 int ret = false, len; 390 u8 rm_addr; 391 392 spin_lock_bh(&msk->pm.lock); 393 394 /* double check after the lock is acquired */ 395 if (!mptcp_pm_should_rm_signal(msk)) 396 goto out_unlock; 397 398 rm_addr = msk->pm.addr_signal & ~BIT(MPTCP_RM_ADDR_SIGNAL); 399 len = mptcp_rm_addr_len(&msk->pm.rm_list_tx); 400 if (len < 0) { 401 WRITE_ONCE(msk->pm.addr_signal, rm_addr); 402 goto out_unlock; 403 } 404 if (remaining < len) 405 goto out_unlock; 406 407 *rm_list = msk->pm.rm_list_tx; 408 WRITE_ONCE(msk->pm.addr_signal, rm_addr); 409 ret = true; 410 411 out_unlock: 412 spin_unlock_bh(&msk->pm.lock); 413 return ret; 414 } 415 416 int mptcp_pm_get_local_id(struct mptcp_sock *msk, struct sock_common *skc) 417 { 418 struct mptcp_addr_info skc_local; 419 struct mptcp_addr_info msk_local; 420 421 if (WARN_ON_ONCE(!msk)) 422 return -1; 423 424 /* The 0 ID mapping is defined by the first subflow, copied into the msk 425 * addr 426 */ 427 mptcp_local_address((struct sock_common *)msk, &msk_local); 428 mptcp_local_address((struct sock_common *)skc, &skc_local); 429 if (mptcp_addresses_equal(&msk_local, &skc_local, false)) 430 return 0; 431 432 if (mptcp_pm_is_userspace(msk)) 433 return mptcp_userspace_pm_get_local_id(msk, &skc_local); 434 return mptcp_pm_nl_get_local_id(msk, &skc_local); 435 } 436 437 int mptcp_pm_get_flags_and_ifindex_by_id(struct mptcp_sock *msk, unsigned int id, 438 u8 *flags, int *ifindex) 439 { 440 *flags = 0; 441 *ifindex = 0; 442 443 if (!id) 444 return 0; 445 446 if (mptcp_pm_is_userspace(msk)) 447 return mptcp_userspace_pm_get_flags_and_ifindex_by_id(msk, id, flags, ifindex); 448 return mptcp_pm_nl_get_flags_and_ifindex_by_id(msk, id, flags, ifindex); 449 } 450 451 int mptcp_pm_set_flags(struct net *net, struct nlattr *token, 452 struct mptcp_pm_addr_entry *loc, 453 struct mptcp_pm_addr_entry *rem, u8 bkup) 454 { 455 if (token) 456 return mptcp_userspace_pm_set_flags(net, token, loc, rem, bkup); 457 return mptcp_pm_nl_set_flags(net, loc, bkup); 458 } 459 460 void mptcp_pm_subflow_chk_stale(const struct mptcp_sock *msk, struct sock *ssk) 461 { 462 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk); 463 u32 rcv_tstamp = READ_ONCE(tcp_sk(ssk)->rcv_tstamp); 464 465 /* keep track of rtx periods with no progress */ 466 if (!subflow->stale_count) { 467 subflow->stale_rcv_tstamp = rcv_tstamp; 468 subflow->stale_count++; 469 } else if (subflow->stale_rcv_tstamp == rcv_tstamp) { 470 if (subflow->stale_count < U8_MAX) 471 subflow->stale_count++; 472 mptcp_pm_nl_subflow_chk_stale(msk, ssk); 473 } else { 474 subflow->stale_count = 0; 475 mptcp_subflow_set_active(subflow); 476 } 477 } 478 479 /* if sk is ipv4 or ipv6_only allows only same-family local and remote addresses, 480 * otherwise allow any matching local/remote pair 481 */ 482 bool mptcp_pm_addr_families_match(const struct sock *sk, 483 const struct mptcp_addr_info *loc, 484 const struct mptcp_addr_info *rem) 485 { 486 bool mptcp_is_v4 = sk->sk_family == AF_INET; 487 488 #if IS_ENABLED(CONFIG_MPTCP_IPV6) 489 bool loc_is_v4 = loc->family == AF_INET || ipv6_addr_v4mapped(&loc->addr6); 490 bool rem_is_v4 = rem->family == AF_INET || ipv6_addr_v4mapped(&rem->addr6); 491 492 if (mptcp_is_v4) 493 return loc_is_v4 && rem_is_v4; 494 495 if (ipv6_only_sock(sk)) 496 return !loc_is_v4 && !rem_is_v4; 497 498 return loc_is_v4 == rem_is_v4; 499 #else 500 return mptcp_is_v4 && loc->family == AF_INET && rem->family == AF_INET; 501 #endif 502 } 503 504 void mptcp_pm_data_reset(struct mptcp_sock *msk) 505 { 506 u8 pm_type = mptcp_get_pm_type(sock_net((struct sock *)msk)); 507 struct mptcp_pm_data *pm = &msk->pm; 508 509 pm->add_addr_signaled = 0; 510 pm->add_addr_accepted = 0; 511 pm->local_addr_used = 0; 512 pm->subflows = 0; 513 pm->rm_list_tx.nr = 0; 514 pm->rm_list_rx.nr = 0; 515 WRITE_ONCE(pm->pm_type, pm_type); 516 517 if (pm_type == MPTCP_PM_TYPE_KERNEL) { 518 bool subflows_allowed = !!mptcp_pm_get_subflows_max(msk); 519 520 /* pm->work_pending must be only be set to 'true' when 521 * pm->pm_type is set to MPTCP_PM_TYPE_KERNEL 522 */ 523 WRITE_ONCE(pm->work_pending, 524 (!!mptcp_pm_get_local_addr_max(msk) && 525 subflows_allowed) || 526 !!mptcp_pm_get_add_addr_signal_max(msk)); 527 WRITE_ONCE(pm->accept_addr, 528 !!mptcp_pm_get_add_addr_accept_max(msk) && 529 subflows_allowed); 530 WRITE_ONCE(pm->accept_subflow, subflows_allowed); 531 } else { 532 WRITE_ONCE(pm->work_pending, 0); 533 WRITE_ONCE(pm->accept_addr, 0); 534 WRITE_ONCE(pm->accept_subflow, 0); 535 } 536 537 WRITE_ONCE(pm->addr_signal, 0); 538 WRITE_ONCE(pm->remote_deny_join_id0, false); 539 pm->status = 0; 540 bitmap_fill(msk->pm.id_avail_bitmap, MPTCP_PM_MAX_ADDR_ID + 1); 541 } 542 543 void mptcp_pm_data_init(struct mptcp_sock *msk) 544 { 545 spin_lock_init(&msk->pm.lock); 546 INIT_LIST_HEAD(&msk->pm.anno_list); 547 INIT_LIST_HEAD(&msk->pm.userspace_pm_local_addr_list); 548 mptcp_pm_data_reset(msk); 549 } 550 551 void __init mptcp_pm_init(void) 552 { 553 mptcp_pm_nl_init(); 554 } 555