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/mptcp.h> 10 #include "protocol.h" 11 12 #include "mib.h" 13 14 /* path manager command handlers */ 15 16 int mptcp_pm_announce_addr(struct mptcp_sock *msk, 17 const struct mptcp_addr_info *addr, 18 bool echo) 19 { 20 u8 add_addr = READ_ONCE(msk->pm.addr_signal); 21 22 pr_debug("msk=%p, local_id=%d, echo=%d", msk, addr->id, echo); 23 24 lockdep_assert_held(&msk->pm.lock); 25 26 if (add_addr & 27 (echo ? BIT(MPTCP_ADD_ADDR_ECHO) : BIT(MPTCP_ADD_ADDR_SIGNAL))) { 28 MPTCP_INC_STATS(sock_net((struct sock *)msk), 29 echo ? MPTCP_MIB_ECHOADDTXDROP : MPTCP_MIB_ADDADDRTXDROP); 30 return -EINVAL; 31 } 32 33 if (echo) { 34 msk->pm.remote = *addr; 35 add_addr |= BIT(MPTCP_ADD_ADDR_ECHO); 36 } else { 37 msk->pm.local = *addr; 38 add_addr |= BIT(MPTCP_ADD_ADDR_SIGNAL); 39 } 40 WRITE_ONCE(msk->pm.addr_signal, add_addr); 41 return 0; 42 } 43 44 int mptcp_pm_remove_addr(struct mptcp_sock *msk, const struct mptcp_rm_list *rm_list) 45 { 46 u8 rm_addr = READ_ONCE(msk->pm.addr_signal); 47 48 pr_debug("msk=%p, rm_list_nr=%d", msk, rm_list->nr); 49 50 if (rm_addr) { 51 MPTCP_ADD_STATS(sock_net((struct sock *)msk), 52 MPTCP_MIB_RMADDRTXDROP, rm_list->nr); 53 return -EINVAL; 54 } 55 56 msk->pm.rm_list_tx = *rm_list; 57 rm_addr |= BIT(MPTCP_RM_ADDR_SIGNAL); 58 WRITE_ONCE(msk->pm.addr_signal, rm_addr); 59 mptcp_pm_nl_addr_send_ack(msk); 60 return 0; 61 } 62 63 int mptcp_pm_remove_subflow(struct mptcp_sock *msk, const struct mptcp_rm_list *rm_list) 64 { 65 pr_debug("msk=%p, rm_list_nr=%d", msk, rm_list->nr); 66 67 spin_lock_bh(&msk->pm.lock); 68 mptcp_pm_nl_rm_subflow_received(msk, rm_list); 69 spin_unlock_bh(&msk->pm.lock); 70 return 0; 71 } 72 73 /* path manager event handlers */ 74 75 void mptcp_pm_new_connection(struct mptcp_sock *msk, const struct sock *ssk, int server_side) 76 { 77 struct mptcp_pm_data *pm = &msk->pm; 78 79 pr_debug("msk=%p, token=%u side=%d", msk, READ_ONCE(msk->token), server_side); 80 81 WRITE_ONCE(pm->server_side, server_side); 82 mptcp_event(MPTCP_EVENT_CREATED, msk, ssk, GFP_ATOMIC); 83 } 84 85 bool mptcp_pm_allow_new_subflow(struct mptcp_sock *msk) 86 { 87 struct mptcp_pm_data *pm = &msk->pm; 88 unsigned int subflows_max; 89 int ret = 0; 90 91 if (mptcp_pm_is_userspace(msk)) { 92 if (mptcp_userspace_pm_active(msk)) { 93 spin_lock_bh(&pm->lock); 94 pm->subflows++; 95 spin_unlock_bh(&pm->lock); 96 return true; 97 } 98 return false; 99 } 100 101 subflows_max = mptcp_pm_get_subflows_max(msk); 102 103 pr_debug("msk=%p subflows=%d max=%d allow=%d", msk, pm->subflows, 104 subflows_max, READ_ONCE(pm->accept_subflow)); 105 106 /* try to avoid acquiring the lock below */ 107 if (!READ_ONCE(pm->accept_subflow)) 108 return false; 109 110 spin_lock_bh(&pm->lock); 111 if (READ_ONCE(pm->accept_subflow)) { 112 ret = pm->subflows < subflows_max; 113 if (ret && ++pm->subflows == subflows_max) 114 WRITE_ONCE(pm->accept_subflow, false); 115 } 116 spin_unlock_bh(&pm->lock); 117 118 return ret; 119 } 120 121 /* return true if the new status bit is currently cleared, that is, this event 122 * can be server, eventually by an already scheduled work 123 */ 124 static bool mptcp_pm_schedule_work(struct mptcp_sock *msk, 125 enum mptcp_pm_status new_status) 126 { 127 pr_debug("msk=%p status=%x new=%lx", msk, msk->pm.status, 128 BIT(new_status)); 129 if (msk->pm.status & BIT(new_status)) 130 return false; 131 132 msk->pm.status |= BIT(new_status); 133 mptcp_schedule_work((struct sock *)msk); 134 return true; 135 } 136 137 void mptcp_pm_fully_established(struct mptcp_sock *msk, const struct sock *ssk) 138 { 139 struct mptcp_pm_data *pm = &msk->pm; 140 bool announce = false; 141 142 pr_debug("msk=%p", msk); 143 144 spin_lock_bh(&pm->lock); 145 146 /* mptcp_pm_fully_established() can be invoked by multiple 147 * racing paths - accept() and check_fully_established() 148 * be sure to serve this event only once. 149 */ 150 if (READ_ONCE(pm->work_pending) && 151 !(msk->pm.status & BIT(MPTCP_PM_ALREADY_ESTABLISHED))) 152 mptcp_pm_schedule_work(msk, MPTCP_PM_ESTABLISHED); 153 154 if ((msk->pm.status & BIT(MPTCP_PM_ALREADY_ESTABLISHED)) == 0) 155 announce = true; 156 157 msk->pm.status |= BIT(MPTCP_PM_ALREADY_ESTABLISHED); 158 spin_unlock_bh(&pm->lock); 159 160 if (announce) 161 mptcp_event(MPTCP_EVENT_ESTABLISHED, msk, ssk, GFP_ATOMIC); 162 } 163 164 void mptcp_pm_connection_closed(struct mptcp_sock *msk) 165 { 166 pr_debug("msk=%p", msk); 167 } 168 169 void mptcp_pm_subflow_established(struct mptcp_sock *msk) 170 { 171 struct mptcp_pm_data *pm = &msk->pm; 172 173 pr_debug("msk=%p", msk); 174 175 if (!READ_ONCE(pm->work_pending)) 176 return; 177 178 spin_lock_bh(&pm->lock); 179 180 if (READ_ONCE(pm->work_pending)) 181 mptcp_pm_schedule_work(msk, MPTCP_PM_SUBFLOW_ESTABLISHED); 182 183 spin_unlock_bh(&pm->lock); 184 } 185 186 void mptcp_pm_subflow_check_next(struct mptcp_sock *msk, 187 const struct mptcp_subflow_context *subflow) 188 { 189 struct mptcp_pm_data *pm = &msk->pm; 190 bool update_subflows; 191 192 update_subflows = subflow->request_join || subflow->mp_join; 193 if (mptcp_pm_is_userspace(msk)) { 194 if (update_subflows) { 195 spin_lock_bh(&pm->lock); 196 pm->subflows--; 197 spin_unlock_bh(&pm->lock); 198 } 199 return; 200 } 201 202 if (!READ_ONCE(pm->work_pending) && !update_subflows) 203 return; 204 205 spin_lock_bh(&pm->lock); 206 if (update_subflows) 207 __mptcp_pm_close_subflow(msk); 208 209 /* Even if this subflow is not really established, tell the PM to try 210 * to pick the next ones, if possible. 211 */ 212 if (mptcp_pm_nl_check_work_pending(msk)) 213 mptcp_pm_schedule_work(msk, MPTCP_PM_SUBFLOW_ESTABLISHED); 214 215 spin_unlock_bh(&pm->lock); 216 } 217 218 void mptcp_pm_add_addr_received(const struct sock *ssk, 219 const struct mptcp_addr_info *addr) 220 { 221 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk); 222 struct mptcp_sock *msk = mptcp_sk(subflow->conn); 223 struct mptcp_pm_data *pm = &msk->pm; 224 225 pr_debug("msk=%p remote_id=%d accept=%d", msk, addr->id, 226 READ_ONCE(pm->accept_addr)); 227 228 mptcp_event_addr_announced(ssk, addr); 229 230 spin_lock_bh(&pm->lock); 231 232 if (mptcp_pm_is_userspace(msk)) { 233 if (mptcp_userspace_pm_active(msk)) { 234 mptcp_pm_announce_addr(msk, addr, true); 235 mptcp_pm_add_addr_send_ack(msk); 236 } else { 237 __MPTCP_INC_STATS(sock_net((struct sock *)msk), MPTCP_MIB_ADDADDRDROP); 238 } 239 } else if (!READ_ONCE(pm->accept_addr)) { 240 mptcp_pm_announce_addr(msk, addr, true); 241 mptcp_pm_add_addr_send_ack(msk); 242 } else if (mptcp_pm_schedule_work(msk, MPTCP_PM_ADD_ADDR_RECEIVED)) { 243 pm->remote = *addr; 244 } else { 245 __MPTCP_INC_STATS(sock_net((struct sock *)msk), MPTCP_MIB_ADDADDRDROP); 246 } 247 248 spin_unlock_bh(&pm->lock); 249 } 250 251 void mptcp_pm_add_addr_echoed(struct mptcp_sock *msk, 252 const struct mptcp_addr_info *addr) 253 { 254 struct mptcp_pm_data *pm = &msk->pm; 255 256 pr_debug("msk=%p", msk); 257 258 spin_lock_bh(&pm->lock); 259 260 if (mptcp_lookup_anno_list_by_saddr(msk, addr) && READ_ONCE(pm->work_pending)) 261 mptcp_pm_schedule_work(msk, MPTCP_PM_SUBFLOW_ESTABLISHED); 262 263 spin_unlock_bh(&pm->lock); 264 } 265 266 void mptcp_pm_add_addr_send_ack(struct mptcp_sock *msk) 267 { 268 if (!mptcp_pm_should_add_signal(msk)) 269 return; 270 271 mptcp_pm_schedule_work(msk, MPTCP_PM_ADD_ADDR_SEND_ACK); 272 } 273 274 void mptcp_pm_rm_addr_received(struct mptcp_sock *msk, 275 const struct mptcp_rm_list *rm_list) 276 { 277 struct mptcp_pm_data *pm = &msk->pm; 278 u8 i; 279 280 pr_debug("msk=%p remote_ids_nr=%d", msk, rm_list->nr); 281 282 for (i = 0; i < rm_list->nr; i++) 283 mptcp_event_addr_removed(msk, rm_list->ids[i]); 284 285 spin_lock_bh(&pm->lock); 286 if (mptcp_pm_schedule_work(msk, MPTCP_PM_RM_ADDR_RECEIVED)) 287 pm->rm_list_rx = *rm_list; 288 else 289 __MPTCP_INC_STATS(sock_net((struct sock *)msk), MPTCP_MIB_RMADDRDROP); 290 spin_unlock_bh(&pm->lock); 291 } 292 293 void mptcp_pm_mp_prio_received(struct sock *ssk, u8 bkup) 294 { 295 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk); 296 struct sock *sk = subflow->conn; 297 struct mptcp_sock *msk; 298 299 pr_debug("subflow->backup=%d, bkup=%d\n", subflow->backup, bkup); 300 msk = mptcp_sk(sk); 301 if (subflow->backup != bkup) 302 subflow->backup = bkup; 303 304 mptcp_event(MPTCP_EVENT_SUB_PRIORITY, msk, ssk, GFP_ATOMIC); 305 } 306 307 void mptcp_pm_mp_fail_received(struct sock *sk, u64 fail_seq) 308 { 309 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk); 310 struct mptcp_sock *msk = mptcp_sk(subflow->conn); 311 312 pr_debug("fail_seq=%llu", fail_seq); 313 314 if (!READ_ONCE(msk->allow_infinite_fallback)) 315 return; 316 317 if (!subflow->fail_tout) { 318 pr_debug("send MP_FAIL response and infinite map"); 319 320 subflow->send_mp_fail = 1; 321 subflow->send_infinite_map = 1; 322 tcp_send_ack(sk); 323 } else { 324 pr_debug("MP_FAIL response received"); 325 WRITE_ONCE(subflow->fail_tout, 0); 326 } 327 } 328 329 /* path manager helpers */ 330 331 bool mptcp_pm_add_addr_signal(struct mptcp_sock *msk, const struct sk_buff *skb, 332 unsigned int opt_size, unsigned int remaining, 333 struct mptcp_addr_info *addr, bool *echo, 334 bool *drop_other_suboptions) 335 { 336 int ret = false; 337 u8 add_addr; 338 u8 family; 339 bool port; 340 341 spin_lock_bh(&msk->pm.lock); 342 343 /* double check after the lock is acquired */ 344 if (!mptcp_pm_should_add_signal(msk)) 345 goto out_unlock; 346 347 /* always drop every other options for pure ack ADD_ADDR; this is a 348 * plain dup-ack from TCP perspective. The other MPTCP-relevant info, 349 * if any, will be carried by the 'original' TCP ack 350 */ 351 if (skb && skb_is_tcp_pure_ack(skb)) { 352 remaining += opt_size; 353 *drop_other_suboptions = true; 354 } 355 356 *echo = mptcp_pm_should_add_signal_echo(msk); 357 port = !!(*echo ? msk->pm.remote.port : msk->pm.local.port); 358 359 family = *echo ? msk->pm.remote.family : msk->pm.local.family; 360 if (remaining < mptcp_add_addr_len(family, *echo, port)) 361 goto out_unlock; 362 363 if (*echo) { 364 *addr = msk->pm.remote; 365 add_addr = msk->pm.addr_signal & ~BIT(MPTCP_ADD_ADDR_ECHO); 366 } else { 367 *addr = msk->pm.local; 368 add_addr = msk->pm.addr_signal & ~BIT(MPTCP_ADD_ADDR_SIGNAL); 369 } 370 WRITE_ONCE(msk->pm.addr_signal, add_addr); 371 ret = true; 372 373 out_unlock: 374 spin_unlock_bh(&msk->pm.lock); 375 return ret; 376 } 377 378 bool mptcp_pm_rm_addr_signal(struct mptcp_sock *msk, unsigned int remaining, 379 struct mptcp_rm_list *rm_list) 380 { 381 int ret = false, len; 382 u8 rm_addr; 383 384 spin_lock_bh(&msk->pm.lock); 385 386 /* double check after the lock is acquired */ 387 if (!mptcp_pm_should_rm_signal(msk)) 388 goto out_unlock; 389 390 rm_addr = msk->pm.addr_signal & ~BIT(MPTCP_RM_ADDR_SIGNAL); 391 len = mptcp_rm_addr_len(&msk->pm.rm_list_tx); 392 if (len < 0) { 393 WRITE_ONCE(msk->pm.addr_signal, rm_addr); 394 goto out_unlock; 395 } 396 if (remaining < len) 397 goto out_unlock; 398 399 *rm_list = msk->pm.rm_list_tx; 400 WRITE_ONCE(msk->pm.addr_signal, rm_addr); 401 ret = true; 402 403 out_unlock: 404 spin_unlock_bh(&msk->pm.lock); 405 return ret; 406 } 407 408 int mptcp_pm_get_local_id(struct mptcp_sock *msk, struct sock_common *skc) 409 { 410 struct mptcp_addr_info skc_local; 411 struct mptcp_addr_info msk_local; 412 413 if (WARN_ON_ONCE(!msk)) 414 return -1; 415 416 /* The 0 ID mapping is defined by the first subflow, copied into the msk 417 * addr 418 */ 419 mptcp_local_address((struct sock_common *)msk, &msk_local); 420 mptcp_local_address((struct sock_common *)skc, &skc_local); 421 if (mptcp_addresses_equal(&msk_local, &skc_local, false)) 422 return 0; 423 424 if (mptcp_pm_is_userspace(msk)) 425 return mptcp_userspace_pm_get_local_id(msk, &skc_local); 426 return mptcp_pm_nl_get_local_id(msk, &skc_local); 427 } 428 429 int mptcp_pm_get_flags_and_ifindex_by_id(struct mptcp_sock *msk, unsigned int id, 430 u8 *flags, int *ifindex) 431 { 432 *flags = 0; 433 *ifindex = 0; 434 435 if (!id) 436 return 0; 437 438 if (mptcp_pm_is_userspace(msk)) 439 return mptcp_userspace_pm_get_flags_and_ifindex_by_id(msk, id, flags, ifindex); 440 return mptcp_pm_nl_get_flags_and_ifindex_by_id(msk, id, flags, ifindex); 441 } 442 443 int mptcp_pm_get_addr(struct sk_buff *skb, struct genl_info *info) 444 { 445 if (info->attrs[MPTCP_PM_ATTR_TOKEN]) 446 return mptcp_userspace_pm_get_addr(skb, info); 447 return mptcp_pm_nl_get_addr(skb, info); 448 } 449 450 int mptcp_pm_dump_addr(struct sk_buff *msg, struct netlink_callback *cb) 451 { 452 const struct genl_info *info = genl_info_dump(cb); 453 454 if (info->attrs[MPTCP_PM_ATTR_TOKEN]) 455 return mptcp_userspace_pm_dump_addr(msg, cb); 456 return mptcp_pm_nl_dump_addr(msg, cb); 457 } 458 459 int mptcp_pm_set_flags(struct sk_buff *skb, struct genl_info *info) 460 { 461 if (info->attrs[MPTCP_PM_ATTR_TOKEN]) 462 return mptcp_userspace_pm_set_flags(skb, info); 463 return mptcp_pm_nl_set_flags(skb, info); 464 } 465 466 void mptcp_pm_subflow_chk_stale(const struct mptcp_sock *msk, struct sock *ssk) 467 { 468 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk); 469 u32 rcv_tstamp = READ_ONCE(tcp_sk(ssk)->rcv_tstamp); 470 471 /* keep track of rtx periods with no progress */ 472 if (!subflow->stale_count) { 473 subflow->stale_rcv_tstamp = rcv_tstamp; 474 subflow->stale_count++; 475 } else if (subflow->stale_rcv_tstamp == rcv_tstamp) { 476 if (subflow->stale_count < U8_MAX) 477 subflow->stale_count++; 478 mptcp_pm_nl_subflow_chk_stale(msk, ssk); 479 } else { 480 subflow->stale_count = 0; 481 mptcp_subflow_set_active(subflow); 482 } 483 } 484 485 /* if sk is ipv4 or ipv6_only allows only same-family local and remote addresses, 486 * otherwise allow any matching local/remote pair 487 */ 488 bool mptcp_pm_addr_families_match(const struct sock *sk, 489 const struct mptcp_addr_info *loc, 490 const struct mptcp_addr_info *rem) 491 { 492 bool mptcp_is_v4 = sk->sk_family == AF_INET; 493 494 #if IS_ENABLED(CONFIG_MPTCP_IPV6) 495 bool loc_is_v4 = loc->family == AF_INET || ipv6_addr_v4mapped(&loc->addr6); 496 bool rem_is_v4 = rem->family == AF_INET || ipv6_addr_v4mapped(&rem->addr6); 497 498 if (mptcp_is_v4) 499 return loc_is_v4 && rem_is_v4; 500 501 if (ipv6_only_sock(sk)) 502 return !loc_is_v4 && !rem_is_v4; 503 504 return loc_is_v4 == rem_is_v4; 505 #else 506 return mptcp_is_v4 && loc->family == AF_INET && rem->family == AF_INET; 507 #endif 508 } 509 510 void mptcp_pm_data_reset(struct mptcp_sock *msk) 511 { 512 u8 pm_type = mptcp_get_pm_type(sock_net((struct sock *)msk)); 513 struct mptcp_pm_data *pm = &msk->pm; 514 515 pm->add_addr_signaled = 0; 516 pm->add_addr_accepted = 0; 517 pm->local_addr_used = 0; 518 pm->subflows = 0; 519 pm->rm_list_tx.nr = 0; 520 pm->rm_list_rx.nr = 0; 521 WRITE_ONCE(pm->pm_type, pm_type); 522 523 if (pm_type == MPTCP_PM_TYPE_KERNEL) { 524 bool subflows_allowed = !!mptcp_pm_get_subflows_max(msk); 525 526 /* pm->work_pending must be only be set to 'true' when 527 * pm->pm_type is set to MPTCP_PM_TYPE_KERNEL 528 */ 529 WRITE_ONCE(pm->work_pending, 530 (!!mptcp_pm_get_local_addr_max(msk) && 531 subflows_allowed) || 532 !!mptcp_pm_get_add_addr_signal_max(msk)); 533 WRITE_ONCE(pm->accept_addr, 534 !!mptcp_pm_get_add_addr_accept_max(msk) && 535 subflows_allowed); 536 WRITE_ONCE(pm->accept_subflow, subflows_allowed); 537 } else { 538 WRITE_ONCE(pm->work_pending, 0); 539 WRITE_ONCE(pm->accept_addr, 0); 540 WRITE_ONCE(pm->accept_subflow, 0); 541 } 542 543 WRITE_ONCE(pm->addr_signal, 0); 544 WRITE_ONCE(pm->remote_deny_join_id0, false); 545 pm->status = 0; 546 bitmap_fill(msk->pm.id_avail_bitmap, MPTCP_PM_MAX_ADDR_ID + 1); 547 } 548 549 void mptcp_pm_data_init(struct mptcp_sock *msk) 550 { 551 spin_lock_init(&msk->pm.lock); 552 INIT_LIST_HEAD(&msk->pm.anno_list); 553 INIT_LIST_HEAD(&msk->pm.userspace_pm_local_addr_list); 554 mptcp_pm_data_reset(msk); 555 } 556 557 void __init mptcp_pm_init(void) 558 { 559 mptcp_pm_nl_init(); 560 } 561