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, gfp_t gfp) 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); 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 = (ssk->sk_state == TCP_CLOSE) && 185 (subflow->request_join || subflow->mp_join) && 186 mptcp_pm_is_kernel(msk); 187 if (!READ_ONCE(pm->work_pending) && !update_subflows) 188 return; 189 190 spin_lock_bh(&pm->lock); 191 if (update_subflows) 192 pm->subflows--; 193 194 /* Even if this subflow is not really established, tell the PM to try 195 * to pick the next ones, if possible. 196 */ 197 if (mptcp_pm_nl_check_work_pending(msk)) 198 mptcp_pm_schedule_work(msk, MPTCP_PM_SUBFLOW_ESTABLISHED); 199 200 spin_unlock_bh(&pm->lock); 201 } 202 203 void mptcp_pm_add_addr_received(const struct sock *ssk, 204 const struct mptcp_addr_info *addr) 205 { 206 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk); 207 struct mptcp_sock *msk = mptcp_sk(subflow->conn); 208 struct mptcp_pm_data *pm = &msk->pm; 209 210 pr_debug("msk=%p remote_id=%d accept=%d", msk, addr->id, 211 READ_ONCE(pm->accept_addr)); 212 213 mptcp_event_addr_announced(ssk, addr); 214 215 spin_lock_bh(&pm->lock); 216 217 if (mptcp_pm_is_userspace(msk)) { 218 if (mptcp_userspace_pm_active(msk)) { 219 mptcp_pm_announce_addr(msk, addr, true); 220 mptcp_pm_add_addr_send_ack(msk); 221 } else { 222 __MPTCP_INC_STATS(sock_net((struct sock *)msk), MPTCP_MIB_ADDADDRDROP); 223 } 224 } else if (!READ_ONCE(pm->accept_addr)) { 225 mptcp_pm_announce_addr(msk, addr, true); 226 mptcp_pm_add_addr_send_ack(msk); 227 } else if (mptcp_pm_schedule_work(msk, MPTCP_PM_ADD_ADDR_RECEIVED)) { 228 pm->remote = *addr; 229 } else { 230 __MPTCP_INC_STATS(sock_net((struct sock *)msk), MPTCP_MIB_ADDADDRDROP); 231 } 232 233 spin_unlock_bh(&pm->lock); 234 } 235 236 void mptcp_pm_add_addr_echoed(struct mptcp_sock *msk, 237 const struct mptcp_addr_info *addr) 238 { 239 struct mptcp_pm_data *pm = &msk->pm; 240 241 pr_debug("msk=%p", msk); 242 243 spin_lock_bh(&pm->lock); 244 245 if (mptcp_lookup_anno_list_by_saddr(msk, addr) && READ_ONCE(pm->work_pending)) 246 mptcp_pm_schedule_work(msk, MPTCP_PM_SUBFLOW_ESTABLISHED); 247 248 spin_unlock_bh(&pm->lock); 249 } 250 251 void mptcp_pm_add_addr_send_ack(struct mptcp_sock *msk) 252 { 253 if (!mptcp_pm_should_add_signal(msk)) 254 return; 255 256 mptcp_pm_schedule_work(msk, MPTCP_PM_ADD_ADDR_SEND_ACK); 257 } 258 259 void mptcp_pm_rm_addr_received(struct mptcp_sock *msk, 260 const struct mptcp_rm_list *rm_list) 261 { 262 struct mptcp_pm_data *pm = &msk->pm; 263 u8 i; 264 265 pr_debug("msk=%p remote_ids_nr=%d", msk, rm_list->nr); 266 267 for (i = 0; i < rm_list->nr; i++) 268 mptcp_event_addr_removed(msk, rm_list->ids[i]); 269 270 spin_lock_bh(&pm->lock); 271 if (mptcp_pm_schedule_work(msk, MPTCP_PM_RM_ADDR_RECEIVED)) 272 pm->rm_list_rx = *rm_list; 273 else 274 __MPTCP_INC_STATS(sock_net((struct sock *)msk), MPTCP_MIB_RMADDRDROP); 275 spin_unlock_bh(&pm->lock); 276 } 277 278 void mptcp_pm_mp_prio_received(struct sock *ssk, u8 bkup) 279 { 280 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk); 281 struct sock *sk = subflow->conn; 282 struct mptcp_sock *msk; 283 284 pr_debug("subflow->backup=%d, bkup=%d\n", subflow->backup, bkup); 285 msk = mptcp_sk(sk); 286 if (subflow->backup != bkup) { 287 subflow->backup = bkup; 288 mptcp_data_lock(sk); 289 if (!sock_owned_by_user(sk)) 290 msk->last_snd = NULL; 291 else 292 __set_bit(MPTCP_RESET_SCHEDULER, &msk->cb_flags); 293 mptcp_data_unlock(sk); 294 } 295 296 mptcp_event(MPTCP_EVENT_SUB_PRIORITY, msk, ssk, GFP_ATOMIC); 297 } 298 299 void mptcp_pm_mp_fail_received(struct sock *sk, u64 fail_seq) 300 { 301 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk); 302 struct mptcp_sock *msk = mptcp_sk(subflow->conn); 303 struct sock *s = (struct sock *)msk; 304 305 pr_debug("fail_seq=%llu", fail_seq); 306 307 if (mptcp_has_another_subflow(sk) || !READ_ONCE(msk->allow_infinite_fallback)) 308 return; 309 310 if (!READ_ONCE(subflow->mp_fail_response_expect)) { 311 pr_debug("send MP_FAIL response and infinite map"); 312 313 subflow->send_mp_fail = 1; 314 MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_MPFAILTX); 315 subflow->send_infinite_map = 1; 316 } else if (s && inet_sk_state_load(s) != TCP_CLOSE) { 317 pr_debug("MP_FAIL response received"); 318 319 mptcp_data_lock(s); 320 if (inet_sk_state_load(s) != TCP_CLOSE) 321 sk_stop_timer(s, &s->sk_timer); 322 mptcp_data_unlock(s); 323 } 324 } 325 326 /* path manager helpers */ 327 328 bool mptcp_pm_add_addr_signal(struct mptcp_sock *msk, const struct sk_buff *skb, 329 unsigned int opt_size, unsigned int remaining, 330 struct mptcp_addr_info *addr, bool *echo, 331 bool *drop_other_suboptions) 332 { 333 int ret = false; 334 u8 add_addr; 335 u8 family; 336 bool port; 337 338 spin_lock_bh(&msk->pm.lock); 339 340 /* double check after the lock is acquired */ 341 if (!mptcp_pm_should_add_signal(msk)) 342 goto out_unlock; 343 344 /* always drop every other options for pure ack ADD_ADDR; this is a 345 * plain dup-ack from TCP perspective. The other MPTCP-relevant info, 346 * if any, will be carried by the 'original' TCP ack 347 */ 348 if (skb && skb_is_tcp_pure_ack(skb)) { 349 remaining += opt_size; 350 *drop_other_suboptions = true; 351 } 352 353 *echo = mptcp_pm_should_add_signal_echo(msk); 354 port = !!(*echo ? msk->pm.remote.port : msk->pm.local.port); 355 356 family = *echo ? msk->pm.remote.family : msk->pm.local.family; 357 if (remaining < mptcp_add_addr_len(family, *echo, port)) 358 goto out_unlock; 359 360 if (*echo) { 361 *addr = msk->pm.remote; 362 add_addr = msk->pm.addr_signal & ~BIT(MPTCP_ADD_ADDR_ECHO); 363 } else { 364 *addr = msk->pm.local; 365 add_addr = msk->pm.addr_signal & ~BIT(MPTCP_ADD_ADDR_SIGNAL); 366 } 367 WRITE_ONCE(msk->pm.addr_signal, add_addr); 368 ret = true; 369 370 out_unlock: 371 spin_unlock_bh(&msk->pm.lock); 372 return ret; 373 } 374 375 bool mptcp_pm_rm_addr_signal(struct mptcp_sock *msk, unsigned int remaining, 376 struct mptcp_rm_list *rm_list) 377 { 378 int ret = false, len; 379 u8 rm_addr; 380 381 spin_lock_bh(&msk->pm.lock); 382 383 /* double check after the lock is acquired */ 384 if (!mptcp_pm_should_rm_signal(msk)) 385 goto out_unlock; 386 387 rm_addr = msk->pm.addr_signal & ~BIT(MPTCP_RM_ADDR_SIGNAL); 388 len = mptcp_rm_addr_len(&msk->pm.rm_list_tx); 389 if (len < 0) { 390 WRITE_ONCE(msk->pm.addr_signal, rm_addr); 391 goto out_unlock; 392 } 393 if (remaining < len) 394 goto out_unlock; 395 396 *rm_list = msk->pm.rm_list_tx; 397 WRITE_ONCE(msk->pm.addr_signal, rm_addr); 398 ret = true; 399 400 out_unlock: 401 spin_unlock_bh(&msk->pm.lock); 402 return ret; 403 } 404 405 int mptcp_pm_get_local_id(struct mptcp_sock *msk, struct sock_common *skc) 406 { 407 return mptcp_pm_nl_get_local_id(msk, skc); 408 } 409 410 void mptcp_pm_subflow_chk_stale(const struct mptcp_sock *msk, struct sock *ssk) 411 { 412 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk); 413 u32 rcv_tstamp = READ_ONCE(tcp_sk(ssk)->rcv_tstamp); 414 415 /* keep track of rtx periods with no progress */ 416 if (!subflow->stale_count) { 417 subflow->stale_rcv_tstamp = rcv_tstamp; 418 subflow->stale_count++; 419 } else if (subflow->stale_rcv_tstamp == rcv_tstamp) { 420 if (subflow->stale_count < U8_MAX) 421 subflow->stale_count++; 422 mptcp_pm_nl_subflow_chk_stale(msk, ssk); 423 } else { 424 subflow->stale_count = 0; 425 mptcp_subflow_set_active(subflow); 426 } 427 } 428 429 void mptcp_pm_data_reset(struct mptcp_sock *msk) 430 { 431 u8 pm_type = mptcp_get_pm_type(sock_net((struct sock *)msk)); 432 struct mptcp_pm_data *pm = &msk->pm; 433 434 pm->add_addr_signaled = 0; 435 pm->add_addr_accepted = 0; 436 pm->local_addr_used = 0; 437 pm->subflows = 0; 438 pm->rm_list_tx.nr = 0; 439 pm->rm_list_rx.nr = 0; 440 WRITE_ONCE(pm->pm_type, pm_type); 441 442 if (pm_type == MPTCP_PM_TYPE_KERNEL) { 443 bool subflows_allowed = !!mptcp_pm_get_subflows_max(msk); 444 445 /* pm->work_pending must be only be set to 'true' when 446 * pm->pm_type is set to MPTCP_PM_TYPE_KERNEL 447 */ 448 WRITE_ONCE(pm->work_pending, 449 (!!mptcp_pm_get_local_addr_max(msk) && 450 subflows_allowed) || 451 !!mptcp_pm_get_add_addr_signal_max(msk)); 452 WRITE_ONCE(pm->accept_addr, 453 !!mptcp_pm_get_add_addr_accept_max(msk) && 454 subflows_allowed); 455 WRITE_ONCE(pm->accept_subflow, subflows_allowed); 456 } else { 457 WRITE_ONCE(pm->work_pending, 0); 458 WRITE_ONCE(pm->accept_addr, 0); 459 WRITE_ONCE(pm->accept_subflow, 0); 460 } 461 462 WRITE_ONCE(pm->addr_signal, 0); 463 WRITE_ONCE(pm->remote_deny_join_id0, false); 464 pm->status = 0; 465 bitmap_fill(msk->pm.id_avail_bitmap, MPTCP_PM_MAX_ADDR_ID + 1); 466 } 467 468 void mptcp_pm_data_init(struct mptcp_sock *msk) 469 { 470 spin_lock_init(&msk->pm.lock); 471 INIT_LIST_HEAD(&msk->pm.anno_list); 472 INIT_LIST_HEAD(&msk->pm.userspace_pm_local_addr_list); 473 mptcp_pm_data_reset(msk); 474 } 475 476 void __init mptcp_pm_init(void) 477 { 478 mptcp_pm_nl_init(); 479 } 480