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