1 // SPDX-License-Identifier: GPL-2.0 2 /* Multipath TCP 3 * 4 * Copyright (c) 2021, Red Hat. 5 */ 6 7 #define pr_fmt(fmt) "MPTCP: " fmt 8 9 #include <linux/kernel.h> 10 #include <linux/module.h> 11 #include <net/sock.h> 12 #include <net/protocol.h> 13 #include <net/tcp.h> 14 #include <net/mptcp.h> 15 #include "protocol.h" 16 17 #define MIN_INFO_OPTLEN_SIZE 16 18 #define MIN_FULL_INFO_OPTLEN_SIZE 40 19 20 static struct sock *__mptcp_tcp_fallback(struct mptcp_sock *msk) 21 { 22 msk_owned_by_me(msk); 23 24 if (likely(!__mptcp_check_fallback(msk))) 25 return NULL; 26 27 return msk->first; 28 } 29 30 static u32 sockopt_seq_reset(const struct sock *sk) 31 { 32 sock_owned_by_me(sk); 33 34 /* Highbits contain state. Allows to distinguish sockopt_seq 35 * of listener and established: 36 * s0 = new_listener() 37 * sockopt(s0) - seq is 1 38 * s1 = accept(s0) - s1 inherits seq 1 if listener sk (s0) 39 * sockopt(s0) - seq increments to 2 on s0 40 * sockopt(s1) // seq increments to 2 on s1 (different option) 41 * new ssk completes join, inherits options from s0 // seq 2 42 * Needs sync from mptcp join logic, but ssk->seq == msk->seq 43 * 44 * Set High order bits to sk_state so ssk->seq == msk->seq test 45 * will fail. 46 */ 47 48 return (u32)sk->sk_state << 24u; 49 } 50 51 static void sockopt_seq_inc(struct mptcp_sock *msk) 52 { 53 u32 seq = (msk->setsockopt_seq + 1) & 0x00ffffff; 54 55 msk->setsockopt_seq = sockopt_seq_reset((struct sock *)msk) + seq; 56 } 57 58 static int mptcp_get_int_option(struct mptcp_sock *msk, sockptr_t optval, 59 unsigned int optlen, int *val) 60 { 61 if (optlen < sizeof(int)) 62 return -EINVAL; 63 64 if (copy_from_sockptr(val, optval, sizeof(*val))) 65 return -EFAULT; 66 67 return 0; 68 } 69 70 static void mptcp_sol_socket_sync_intval(struct mptcp_sock *msk, int optname, int val) 71 { 72 struct mptcp_subflow_context *subflow; 73 struct sock *sk = (struct sock *)msk; 74 75 lock_sock(sk); 76 sockopt_seq_inc(msk); 77 78 mptcp_for_each_subflow(msk, subflow) { 79 struct sock *ssk = mptcp_subflow_tcp_sock(subflow); 80 bool slow = lock_sock_fast(ssk); 81 82 switch (optname) { 83 case SO_DEBUG: 84 sock_valbool_flag(ssk, SOCK_DBG, !!val); 85 break; 86 case SO_KEEPALIVE: 87 if (ssk->sk_prot->keepalive) 88 ssk->sk_prot->keepalive(ssk, !!val); 89 sock_valbool_flag(ssk, SOCK_KEEPOPEN, !!val); 90 break; 91 case SO_PRIORITY: 92 WRITE_ONCE(ssk->sk_priority, val); 93 break; 94 case SO_SNDBUF: 95 case SO_SNDBUFFORCE: 96 ssk->sk_userlocks |= SOCK_SNDBUF_LOCK; 97 WRITE_ONCE(ssk->sk_sndbuf, sk->sk_sndbuf); 98 mptcp_subflow_ctx(ssk)->cached_sndbuf = sk->sk_sndbuf; 99 break; 100 case SO_RCVBUF: 101 case SO_RCVBUFFORCE: 102 ssk->sk_userlocks |= SOCK_RCVBUF_LOCK; 103 WRITE_ONCE(ssk->sk_rcvbuf, sk->sk_rcvbuf); 104 break; 105 case SO_MARK: 106 if (READ_ONCE(ssk->sk_mark) != sk->sk_mark) { 107 WRITE_ONCE(ssk->sk_mark, sk->sk_mark); 108 sk_dst_reset(ssk); 109 } 110 break; 111 case SO_INCOMING_CPU: 112 WRITE_ONCE(ssk->sk_incoming_cpu, val); 113 break; 114 } 115 116 subflow->setsockopt_seq = msk->setsockopt_seq; 117 unlock_sock_fast(ssk, slow); 118 } 119 120 release_sock(sk); 121 } 122 123 static int mptcp_sol_socket_intval(struct mptcp_sock *msk, int optname, int val) 124 { 125 sockptr_t optval = KERNEL_SOCKPTR(&val); 126 struct sock *sk = (struct sock *)msk; 127 int ret; 128 129 ret = sock_setsockopt(sk->sk_socket, SOL_SOCKET, optname, 130 optval, sizeof(val)); 131 if (ret) 132 return ret; 133 134 mptcp_sol_socket_sync_intval(msk, optname, val); 135 return 0; 136 } 137 138 static void mptcp_so_incoming_cpu(struct mptcp_sock *msk, int val) 139 { 140 struct sock *sk = (struct sock *)msk; 141 142 WRITE_ONCE(sk->sk_incoming_cpu, val); 143 144 mptcp_sol_socket_sync_intval(msk, SO_INCOMING_CPU, val); 145 } 146 147 static int mptcp_setsockopt_sol_socket_tstamp(struct mptcp_sock *msk, int optname, int val) 148 { 149 sockptr_t optval = KERNEL_SOCKPTR(&val); 150 struct mptcp_subflow_context *subflow; 151 struct sock *sk = (struct sock *)msk; 152 int ret; 153 154 ret = sock_setsockopt(sk->sk_socket, SOL_SOCKET, optname, 155 optval, sizeof(val)); 156 if (ret) 157 return ret; 158 159 lock_sock(sk); 160 mptcp_for_each_subflow(msk, subflow) { 161 struct sock *ssk = mptcp_subflow_tcp_sock(subflow); 162 bool slow = lock_sock_fast(ssk); 163 164 sock_set_timestamp(sk, optname, !!val); 165 unlock_sock_fast(ssk, slow); 166 } 167 168 release_sock(sk); 169 return 0; 170 } 171 172 static int mptcp_setsockopt_sol_socket_int(struct mptcp_sock *msk, int optname, 173 sockptr_t optval, 174 unsigned int optlen) 175 { 176 int val, ret; 177 178 ret = mptcp_get_int_option(msk, optval, optlen, &val); 179 if (ret) 180 return ret; 181 182 switch (optname) { 183 case SO_KEEPALIVE: 184 case SO_DEBUG: 185 case SO_MARK: 186 case SO_PRIORITY: 187 case SO_SNDBUF: 188 case SO_SNDBUFFORCE: 189 case SO_RCVBUF: 190 case SO_RCVBUFFORCE: 191 return mptcp_sol_socket_intval(msk, optname, val); 192 case SO_INCOMING_CPU: 193 mptcp_so_incoming_cpu(msk, val); 194 return 0; 195 case SO_TIMESTAMP_OLD: 196 case SO_TIMESTAMP_NEW: 197 case SO_TIMESTAMPNS_OLD: 198 case SO_TIMESTAMPNS_NEW: 199 return mptcp_setsockopt_sol_socket_tstamp(msk, optname, val); 200 } 201 202 return -ENOPROTOOPT; 203 } 204 205 static int mptcp_setsockopt_sol_socket_timestamping(struct mptcp_sock *msk, 206 int optname, 207 sockptr_t optval, 208 unsigned int optlen) 209 { 210 struct mptcp_subflow_context *subflow; 211 struct sock *sk = (struct sock *)msk; 212 struct so_timestamping timestamping; 213 int ret; 214 215 if (optlen == sizeof(timestamping)) { 216 if (copy_from_sockptr(×tamping, optval, 217 sizeof(timestamping))) 218 return -EFAULT; 219 } else if (optlen == sizeof(int)) { 220 memset(×tamping, 0, sizeof(timestamping)); 221 222 if (copy_from_sockptr(×tamping.flags, optval, sizeof(int))) 223 return -EFAULT; 224 } else { 225 return -EINVAL; 226 } 227 228 ret = sock_setsockopt(sk->sk_socket, SOL_SOCKET, optname, 229 KERNEL_SOCKPTR(×tamping), 230 sizeof(timestamping)); 231 if (ret) 232 return ret; 233 234 lock_sock(sk); 235 236 mptcp_for_each_subflow(msk, subflow) { 237 struct sock *ssk = mptcp_subflow_tcp_sock(subflow); 238 bool slow = lock_sock_fast(ssk); 239 240 sock_set_timestamping(sk, optname, timestamping); 241 unlock_sock_fast(ssk, slow); 242 } 243 244 release_sock(sk); 245 246 return 0; 247 } 248 249 static int mptcp_setsockopt_sol_socket_linger(struct mptcp_sock *msk, sockptr_t optval, 250 unsigned int optlen) 251 { 252 struct mptcp_subflow_context *subflow; 253 struct sock *sk = (struct sock *)msk; 254 struct linger ling; 255 sockptr_t kopt; 256 int ret; 257 258 if (optlen < sizeof(ling)) 259 return -EINVAL; 260 261 if (copy_from_sockptr(&ling, optval, sizeof(ling))) 262 return -EFAULT; 263 264 kopt = KERNEL_SOCKPTR(&ling); 265 ret = sock_setsockopt(sk->sk_socket, SOL_SOCKET, SO_LINGER, kopt, sizeof(ling)); 266 if (ret) 267 return ret; 268 269 lock_sock(sk); 270 sockopt_seq_inc(msk); 271 mptcp_for_each_subflow(msk, subflow) { 272 struct sock *ssk = mptcp_subflow_tcp_sock(subflow); 273 bool slow = lock_sock_fast(ssk); 274 275 if (!ling.l_onoff) { 276 sock_reset_flag(ssk, SOCK_LINGER); 277 } else { 278 ssk->sk_lingertime = sk->sk_lingertime; 279 sock_set_flag(ssk, SOCK_LINGER); 280 } 281 282 subflow->setsockopt_seq = msk->setsockopt_seq; 283 unlock_sock_fast(ssk, slow); 284 } 285 286 release_sock(sk); 287 return 0; 288 } 289 290 static int mptcp_setsockopt_sol_socket(struct mptcp_sock *msk, int optname, 291 sockptr_t optval, unsigned int optlen) 292 { 293 struct sock *sk = (struct sock *)msk; 294 struct sock *ssk; 295 int ret; 296 297 switch (optname) { 298 case SO_REUSEPORT: 299 case SO_REUSEADDR: 300 case SO_BINDTODEVICE: 301 case SO_BINDTOIFINDEX: 302 lock_sock(sk); 303 ssk = __mptcp_nmpc_sk(msk); 304 if (IS_ERR(ssk)) { 305 release_sock(sk); 306 return PTR_ERR(ssk); 307 } 308 309 ret = sk_setsockopt(ssk, SOL_SOCKET, optname, optval, optlen); 310 if (ret == 0) { 311 if (optname == SO_REUSEPORT) 312 sk->sk_reuseport = ssk->sk_reuseport; 313 else if (optname == SO_REUSEADDR) 314 sk->sk_reuse = ssk->sk_reuse; 315 else if (optname == SO_BINDTODEVICE) 316 sk->sk_bound_dev_if = ssk->sk_bound_dev_if; 317 else if (optname == SO_BINDTOIFINDEX) 318 sk->sk_bound_dev_if = ssk->sk_bound_dev_if; 319 } 320 release_sock(sk); 321 return ret; 322 case SO_KEEPALIVE: 323 case SO_PRIORITY: 324 case SO_SNDBUF: 325 case SO_SNDBUFFORCE: 326 case SO_RCVBUF: 327 case SO_RCVBUFFORCE: 328 case SO_MARK: 329 case SO_INCOMING_CPU: 330 case SO_DEBUG: 331 case SO_TIMESTAMP_OLD: 332 case SO_TIMESTAMP_NEW: 333 case SO_TIMESTAMPNS_OLD: 334 case SO_TIMESTAMPNS_NEW: 335 return mptcp_setsockopt_sol_socket_int(msk, optname, optval, 336 optlen); 337 case SO_TIMESTAMPING_OLD: 338 case SO_TIMESTAMPING_NEW: 339 return mptcp_setsockopt_sol_socket_timestamping(msk, optname, 340 optval, optlen); 341 case SO_LINGER: 342 return mptcp_setsockopt_sol_socket_linger(msk, optval, optlen); 343 case SO_RCVLOWAT: 344 case SO_RCVTIMEO_OLD: 345 case SO_RCVTIMEO_NEW: 346 case SO_SNDTIMEO_OLD: 347 case SO_SNDTIMEO_NEW: 348 case SO_BUSY_POLL: 349 case SO_PREFER_BUSY_POLL: 350 case SO_BUSY_POLL_BUDGET: 351 /* No need to copy: only relevant for msk */ 352 return sock_setsockopt(sk->sk_socket, SOL_SOCKET, optname, optval, optlen); 353 case SO_NO_CHECK: 354 case SO_DONTROUTE: 355 case SO_BROADCAST: 356 case SO_BSDCOMPAT: 357 case SO_PASSCRED: 358 case SO_PASSPIDFD: 359 case SO_PASSSEC: 360 case SO_RXQ_OVFL: 361 case SO_WIFI_STATUS: 362 case SO_NOFCS: 363 case SO_SELECT_ERR_QUEUE: 364 return 0; 365 } 366 367 /* SO_OOBINLINE is not supported, let's avoid the related mess 368 * SO_ATTACH_FILTER, SO_ATTACH_BPF, SO_ATTACH_REUSEPORT_CBPF, 369 * SO_DETACH_REUSEPORT_BPF, SO_DETACH_FILTER, SO_LOCK_FILTER, 370 * we must be careful with subflows 371 * 372 * SO_ATTACH_REUSEPORT_EBPF is not supported, at it checks 373 * explicitly the sk_protocol field 374 * 375 * SO_PEEK_OFF is unsupported, as it is for plain TCP 376 * SO_MAX_PACING_RATE is unsupported, we must be careful with subflows 377 * SO_CNX_ADVICE is currently unsupported, could possibly be relevant, 378 * but likely needs careful design 379 * 380 * SO_ZEROCOPY is currently unsupported, TODO in sndmsg 381 * SO_TXTIME is currently unsupported 382 */ 383 384 return -EOPNOTSUPP; 385 } 386 387 static int mptcp_setsockopt_v6(struct mptcp_sock *msk, int optname, 388 sockptr_t optval, unsigned int optlen) 389 { 390 struct sock *sk = (struct sock *)msk; 391 int ret = -EOPNOTSUPP; 392 struct sock *ssk; 393 394 switch (optname) { 395 case IPV6_V6ONLY: 396 case IPV6_TRANSPARENT: 397 case IPV6_FREEBIND: 398 lock_sock(sk); 399 ssk = __mptcp_nmpc_sk(msk); 400 if (IS_ERR(ssk)) { 401 release_sock(sk); 402 return PTR_ERR(ssk); 403 } 404 405 ret = tcp_setsockopt(ssk, SOL_IPV6, optname, optval, optlen); 406 if (ret != 0) { 407 release_sock(sk); 408 return ret; 409 } 410 411 sockopt_seq_inc(msk); 412 413 switch (optname) { 414 case IPV6_V6ONLY: 415 sk->sk_ipv6only = ssk->sk_ipv6only; 416 break; 417 case IPV6_TRANSPARENT: 418 inet_assign_bit(TRANSPARENT, sk, 419 inet_test_bit(TRANSPARENT, ssk)); 420 break; 421 case IPV6_FREEBIND: 422 inet_assign_bit(FREEBIND, sk, 423 inet_test_bit(FREEBIND, ssk)); 424 break; 425 } 426 427 release_sock(sk); 428 break; 429 } 430 431 return ret; 432 } 433 434 static bool mptcp_supported_sockopt(int level, int optname) 435 { 436 if (level == SOL_IP) { 437 switch (optname) { 438 /* should work fine */ 439 case IP_FREEBIND: 440 case IP_TRANSPARENT: 441 case IP_BIND_ADDRESS_NO_PORT: 442 case IP_LOCAL_PORT_RANGE: 443 444 /* the following are control cmsg related */ 445 case IP_PKTINFO: 446 case IP_RECVTTL: 447 case IP_RECVTOS: 448 case IP_RECVOPTS: 449 case IP_RETOPTS: 450 case IP_PASSSEC: 451 case IP_RECVORIGDSTADDR: 452 case IP_CHECKSUM: 453 case IP_RECVFRAGSIZE: 454 455 /* common stuff that need some love */ 456 case IP_TOS: 457 case IP_TTL: 458 case IP_MTU_DISCOVER: 459 case IP_RECVERR: 460 461 /* possibly less common may deserve some love */ 462 case IP_MINTTL: 463 464 /* the following is apparently a no-op for plain TCP */ 465 case IP_RECVERR_RFC4884: 466 return true; 467 } 468 469 /* IP_OPTIONS is not supported, needs subflow care */ 470 /* IP_HDRINCL, IP_NODEFRAG are not supported, RAW specific */ 471 /* IP_MULTICAST_TTL, IP_MULTICAST_LOOP, IP_UNICAST_IF, 472 * IP_ADD_MEMBERSHIP, IP_ADD_SOURCE_MEMBERSHIP, IP_DROP_MEMBERSHIP, 473 * IP_DROP_SOURCE_MEMBERSHIP, IP_BLOCK_SOURCE, IP_UNBLOCK_SOURCE, 474 * MCAST_JOIN_GROUP, MCAST_LEAVE_GROUP MCAST_JOIN_SOURCE_GROUP, 475 * MCAST_LEAVE_SOURCE_GROUP, MCAST_BLOCK_SOURCE, MCAST_UNBLOCK_SOURCE, 476 * MCAST_MSFILTER, IP_MULTICAST_ALL are not supported, better not deal 477 * with mcast stuff 478 */ 479 /* IP_IPSEC_POLICY, IP_XFRM_POLICY are nut supported, unrelated here */ 480 return false; 481 } 482 if (level == SOL_IPV6) { 483 switch (optname) { 484 case IPV6_V6ONLY: 485 486 /* the following are control cmsg related */ 487 case IPV6_RECVPKTINFO: 488 case IPV6_2292PKTINFO: 489 case IPV6_RECVHOPLIMIT: 490 case IPV6_2292HOPLIMIT: 491 case IPV6_RECVRTHDR: 492 case IPV6_2292RTHDR: 493 case IPV6_RECVHOPOPTS: 494 case IPV6_2292HOPOPTS: 495 case IPV6_RECVDSTOPTS: 496 case IPV6_2292DSTOPTS: 497 case IPV6_RECVTCLASS: 498 case IPV6_FLOWINFO: 499 case IPV6_RECVPATHMTU: 500 case IPV6_RECVORIGDSTADDR: 501 case IPV6_RECVFRAGSIZE: 502 503 /* the following ones need some love but are quite common */ 504 case IPV6_TCLASS: 505 case IPV6_TRANSPARENT: 506 case IPV6_FREEBIND: 507 case IPV6_PKTINFO: 508 case IPV6_2292PKTOPTIONS: 509 case IPV6_UNICAST_HOPS: 510 case IPV6_MTU_DISCOVER: 511 case IPV6_MTU: 512 case IPV6_RECVERR: 513 case IPV6_FLOWINFO_SEND: 514 case IPV6_FLOWLABEL_MGR: 515 case IPV6_MINHOPCOUNT: 516 case IPV6_DONTFRAG: 517 case IPV6_AUTOFLOWLABEL: 518 519 /* the following one is a no-op for plain TCP */ 520 case IPV6_RECVERR_RFC4884: 521 return true; 522 } 523 524 /* IPV6_HOPOPTS, IPV6_RTHDRDSTOPTS, IPV6_RTHDR, IPV6_DSTOPTS are 525 * not supported 526 */ 527 /* IPV6_MULTICAST_HOPS, IPV6_MULTICAST_LOOP, IPV6_UNICAST_IF, 528 * IPV6_MULTICAST_IF, IPV6_ADDRFORM, 529 * IPV6_ADD_MEMBERSHIP, IPV6_DROP_MEMBERSHIP, IPV6_JOIN_ANYCAST, 530 * IPV6_LEAVE_ANYCAST, IPV6_MULTICAST_ALL, MCAST_JOIN_GROUP, MCAST_LEAVE_GROUP, 531 * MCAST_JOIN_SOURCE_GROUP, MCAST_LEAVE_SOURCE_GROUP, 532 * MCAST_BLOCK_SOURCE, MCAST_UNBLOCK_SOURCE, MCAST_MSFILTER 533 * are not supported better not deal with mcast 534 */ 535 /* IPV6_ROUTER_ALERT, IPV6_ROUTER_ALERT_ISOLATE are not supported, since are evil */ 536 537 /* IPV6_IPSEC_POLICY, IPV6_XFRM_POLICY are not supported */ 538 /* IPV6_ADDR_PREFERENCES is not supported, we must be careful with subflows */ 539 return false; 540 } 541 if (level == SOL_TCP) { 542 switch (optname) { 543 /* the following are no-op or should work just fine */ 544 case TCP_THIN_DUPACK: 545 case TCP_DEFER_ACCEPT: 546 547 /* the following need some love */ 548 case TCP_MAXSEG: 549 case TCP_NODELAY: 550 case TCP_THIN_LINEAR_TIMEOUTS: 551 case TCP_CONGESTION: 552 case TCP_CORK: 553 case TCP_KEEPIDLE: 554 case TCP_KEEPINTVL: 555 case TCP_KEEPCNT: 556 case TCP_SYNCNT: 557 case TCP_SAVE_SYN: 558 case TCP_LINGER2: 559 case TCP_WINDOW_CLAMP: 560 case TCP_QUICKACK: 561 case TCP_USER_TIMEOUT: 562 case TCP_TIMESTAMP: 563 case TCP_NOTSENT_LOWAT: 564 case TCP_TX_DELAY: 565 case TCP_INQ: 566 case TCP_FASTOPEN: 567 case TCP_FASTOPEN_CONNECT: 568 case TCP_FASTOPEN_KEY: 569 case TCP_FASTOPEN_NO_COOKIE: 570 return true; 571 } 572 573 /* TCP_MD5SIG, TCP_MD5SIG_EXT are not supported, MD5 is not compatible with MPTCP */ 574 575 /* TCP_REPAIR, TCP_REPAIR_QUEUE, TCP_QUEUE_SEQ, TCP_REPAIR_OPTIONS, 576 * TCP_REPAIR_WINDOW are not supported, better avoid this mess 577 */ 578 } 579 return false; 580 } 581 582 static int mptcp_setsockopt_sol_tcp_congestion(struct mptcp_sock *msk, sockptr_t optval, 583 unsigned int optlen) 584 { 585 struct mptcp_subflow_context *subflow; 586 struct sock *sk = (struct sock *)msk; 587 char name[TCP_CA_NAME_MAX]; 588 bool cap_net_admin; 589 int ret; 590 591 if (optlen < 1) 592 return -EINVAL; 593 594 ret = strncpy_from_sockptr(name, optval, 595 min_t(long, TCP_CA_NAME_MAX - 1, optlen)); 596 if (ret < 0) 597 return -EFAULT; 598 599 name[ret] = 0; 600 601 cap_net_admin = ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN); 602 603 ret = 0; 604 lock_sock(sk); 605 sockopt_seq_inc(msk); 606 mptcp_for_each_subflow(msk, subflow) { 607 struct sock *ssk = mptcp_subflow_tcp_sock(subflow); 608 int err; 609 610 lock_sock(ssk); 611 err = tcp_set_congestion_control(ssk, name, true, cap_net_admin); 612 if (err < 0 && ret == 0) 613 ret = err; 614 subflow->setsockopt_seq = msk->setsockopt_seq; 615 release_sock(ssk); 616 } 617 618 if (ret == 0) 619 strcpy(msk->ca_name, name); 620 621 release_sock(sk); 622 return ret; 623 } 624 625 static int __mptcp_setsockopt_sol_tcp_cork(struct mptcp_sock *msk, int val) 626 { 627 struct mptcp_subflow_context *subflow; 628 struct sock *sk = (struct sock *)msk; 629 630 sockopt_seq_inc(msk); 631 msk->cork = !!val; 632 mptcp_for_each_subflow(msk, subflow) { 633 struct sock *ssk = mptcp_subflow_tcp_sock(subflow); 634 635 lock_sock(ssk); 636 __tcp_sock_set_cork(ssk, !!val); 637 release_sock(ssk); 638 } 639 if (!val) 640 mptcp_check_and_set_pending(sk); 641 642 return 0; 643 } 644 645 static int __mptcp_setsockopt_sol_tcp_nodelay(struct mptcp_sock *msk, int val) 646 { 647 struct mptcp_subflow_context *subflow; 648 struct sock *sk = (struct sock *)msk; 649 650 sockopt_seq_inc(msk); 651 msk->nodelay = !!val; 652 mptcp_for_each_subflow(msk, subflow) { 653 struct sock *ssk = mptcp_subflow_tcp_sock(subflow); 654 655 lock_sock(ssk); 656 __tcp_sock_set_nodelay(ssk, !!val); 657 release_sock(ssk); 658 } 659 if (val) 660 mptcp_check_and_set_pending(sk); 661 return 0; 662 } 663 664 static int mptcp_setsockopt_sol_ip_set(struct mptcp_sock *msk, int optname, 665 sockptr_t optval, unsigned int optlen) 666 { 667 struct sock *sk = (struct sock *)msk; 668 struct sock *ssk; 669 int err; 670 671 err = ip_setsockopt(sk, SOL_IP, optname, optval, optlen); 672 if (err != 0) 673 return err; 674 675 lock_sock(sk); 676 677 ssk = __mptcp_nmpc_sk(msk); 678 if (IS_ERR(ssk)) { 679 release_sock(sk); 680 return PTR_ERR(ssk); 681 } 682 683 switch (optname) { 684 case IP_FREEBIND: 685 inet_assign_bit(FREEBIND, ssk, inet_test_bit(FREEBIND, sk)); 686 break; 687 case IP_TRANSPARENT: 688 inet_assign_bit(TRANSPARENT, ssk, 689 inet_test_bit(TRANSPARENT, sk)); 690 break; 691 case IP_BIND_ADDRESS_NO_PORT: 692 inet_assign_bit(BIND_ADDRESS_NO_PORT, ssk, 693 inet_test_bit(BIND_ADDRESS_NO_PORT, sk)); 694 break; 695 case IP_LOCAL_PORT_RANGE: 696 WRITE_ONCE(inet_sk(ssk)->local_port_range, 697 READ_ONCE(inet_sk(sk)->local_port_range)); 698 break; 699 default: 700 release_sock(sk); 701 WARN_ON_ONCE(1); 702 return -EOPNOTSUPP; 703 } 704 705 sockopt_seq_inc(msk); 706 release_sock(sk); 707 return 0; 708 } 709 710 static int mptcp_setsockopt_v4_set_tos(struct mptcp_sock *msk, int optname, 711 sockptr_t optval, unsigned int optlen) 712 { 713 struct mptcp_subflow_context *subflow; 714 struct sock *sk = (struct sock *)msk; 715 int err, val; 716 717 err = ip_setsockopt(sk, SOL_IP, optname, optval, optlen); 718 719 if (err != 0) 720 return err; 721 722 lock_sock(sk); 723 sockopt_seq_inc(msk); 724 val = READ_ONCE(inet_sk(sk)->tos); 725 mptcp_for_each_subflow(msk, subflow) { 726 struct sock *ssk = mptcp_subflow_tcp_sock(subflow); 727 bool slow; 728 729 slow = lock_sock_fast(ssk); 730 __ip_sock_set_tos(ssk, val); 731 unlock_sock_fast(ssk, slow); 732 } 733 release_sock(sk); 734 735 return 0; 736 } 737 738 static int mptcp_setsockopt_v4(struct mptcp_sock *msk, int optname, 739 sockptr_t optval, unsigned int optlen) 740 { 741 switch (optname) { 742 case IP_FREEBIND: 743 case IP_TRANSPARENT: 744 case IP_BIND_ADDRESS_NO_PORT: 745 case IP_LOCAL_PORT_RANGE: 746 return mptcp_setsockopt_sol_ip_set(msk, optname, optval, optlen); 747 case IP_TOS: 748 return mptcp_setsockopt_v4_set_tos(msk, optname, optval, optlen); 749 } 750 751 return -EOPNOTSUPP; 752 } 753 754 static int mptcp_setsockopt_first_sf_only(struct mptcp_sock *msk, int level, int optname, 755 sockptr_t optval, unsigned int optlen) 756 { 757 struct sock *sk = (struct sock *)msk; 758 struct sock *ssk; 759 int ret; 760 761 /* Limit to first subflow, before the connection establishment */ 762 lock_sock(sk); 763 ssk = __mptcp_nmpc_sk(msk); 764 if (IS_ERR(ssk)) { 765 ret = PTR_ERR(ssk); 766 goto unlock; 767 } 768 769 ret = tcp_setsockopt(ssk, level, optname, optval, optlen); 770 771 unlock: 772 release_sock(sk); 773 return ret; 774 } 775 776 static int mptcp_setsockopt_sol_tcp(struct mptcp_sock *msk, int optname, 777 sockptr_t optval, unsigned int optlen) 778 { 779 struct sock *sk = (void *)msk; 780 int ret, val; 781 782 switch (optname) { 783 case TCP_ULP: 784 return -EOPNOTSUPP; 785 case TCP_CONGESTION: 786 return mptcp_setsockopt_sol_tcp_congestion(msk, optval, optlen); 787 case TCP_DEFER_ACCEPT: 788 /* See tcp.c: TCP_DEFER_ACCEPT does not fail */ 789 mptcp_setsockopt_first_sf_only(msk, SOL_TCP, optname, optval, optlen); 790 return 0; 791 case TCP_FASTOPEN: 792 case TCP_FASTOPEN_CONNECT: 793 case TCP_FASTOPEN_KEY: 794 case TCP_FASTOPEN_NO_COOKIE: 795 return mptcp_setsockopt_first_sf_only(msk, SOL_TCP, optname, 796 optval, optlen); 797 } 798 799 ret = mptcp_get_int_option(msk, optval, optlen, &val); 800 if (ret) 801 return ret; 802 803 lock_sock(sk); 804 switch (optname) { 805 case TCP_INQ: 806 if (val < 0 || val > 1) 807 ret = -EINVAL; 808 else 809 msk->recvmsg_inq = !!val; 810 break; 811 case TCP_NOTSENT_LOWAT: 812 WRITE_ONCE(msk->notsent_lowat, val); 813 mptcp_write_space(sk); 814 break; 815 case TCP_CORK: 816 ret = __mptcp_setsockopt_sol_tcp_cork(msk, val); 817 break; 818 case TCP_NODELAY: 819 ret = __mptcp_setsockopt_sol_tcp_nodelay(msk, val); 820 break; 821 default: 822 ret = -ENOPROTOOPT; 823 } 824 825 release_sock(sk); 826 return ret; 827 } 828 829 int mptcp_setsockopt(struct sock *sk, int level, int optname, 830 sockptr_t optval, unsigned int optlen) 831 { 832 struct mptcp_sock *msk = mptcp_sk(sk); 833 struct sock *ssk; 834 835 pr_debug("msk=%p", msk); 836 837 if (level == SOL_SOCKET) 838 return mptcp_setsockopt_sol_socket(msk, optname, optval, optlen); 839 840 if (!mptcp_supported_sockopt(level, optname)) 841 return -ENOPROTOOPT; 842 843 /* @@ the meaning of setsockopt() when the socket is connected and 844 * there are multiple subflows is not yet defined. It is up to the 845 * MPTCP-level socket to configure the subflows until the subflow 846 * is in TCP fallback, when TCP socket options are passed through 847 * to the one remaining subflow. 848 */ 849 lock_sock(sk); 850 ssk = __mptcp_tcp_fallback(msk); 851 release_sock(sk); 852 if (ssk) 853 return tcp_setsockopt(ssk, level, optname, optval, optlen); 854 855 if (level == SOL_IP) 856 return mptcp_setsockopt_v4(msk, optname, optval, optlen); 857 858 if (level == SOL_IPV6) 859 return mptcp_setsockopt_v6(msk, optname, optval, optlen); 860 861 if (level == SOL_TCP) 862 return mptcp_setsockopt_sol_tcp(msk, optname, optval, optlen); 863 864 return -EOPNOTSUPP; 865 } 866 867 static int mptcp_getsockopt_first_sf_only(struct mptcp_sock *msk, int level, int optname, 868 char __user *optval, int __user *optlen) 869 { 870 struct sock *sk = (struct sock *)msk; 871 struct sock *ssk; 872 int ret; 873 874 lock_sock(sk); 875 ssk = msk->first; 876 if (ssk) { 877 ret = tcp_getsockopt(ssk, level, optname, optval, optlen); 878 goto out; 879 } 880 881 ssk = __mptcp_nmpc_sk(msk); 882 if (IS_ERR(ssk)) { 883 ret = PTR_ERR(ssk); 884 goto out; 885 } 886 887 ret = tcp_getsockopt(ssk, level, optname, optval, optlen); 888 889 out: 890 release_sock(sk); 891 return ret; 892 } 893 894 void mptcp_diag_fill_info(struct mptcp_sock *msk, struct mptcp_info *info) 895 { 896 struct sock *sk = (struct sock *)msk; 897 u32 flags = 0; 898 bool slow; 899 u32 now; 900 901 memset(info, 0, sizeof(*info)); 902 903 info->mptcpi_subflows = READ_ONCE(msk->pm.subflows); 904 info->mptcpi_add_addr_signal = READ_ONCE(msk->pm.add_addr_signaled); 905 info->mptcpi_add_addr_accepted = READ_ONCE(msk->pm.add_addr_accepted); 906 info->mptcpi_local_addr_used = READ_ONCE(msk->pm.local_addr_used); 907 908 if (inet_sk_state_load(sk) == TCP_LISTEN) 909 return; 910 911 /* The following limits only make sense for the in-kernel PM */ 912 if (mptcp_pm_is_kernel(msk)) { 913 info->mptcpi_subflows_max = 914 mptcp_pm_get_subflows_max(msk); 915 info->mptcpi_add_addr_signal_max = 916 mptcp_pm_get_add_addr_signal_max(msk); 917 info->mptcpi_add_addr_accepted_max = 918 mptcp_pm_get_add_addr_accept_max(msk); 919 info->mptcpi_local_addr_max = 920 mptcp_pm_get_local_addr_max(msk); 921 } 922 923 if (__mptcp_check_fallback(msk)) 924 flags |= MPTCP_INFO_FLAG_FALLBACK; 925 if (READ_ONCE(msk->can_ack)) 926 flags |= MPTCP_INFO_FLAG_REMOTE_KEY_RECEIVED; 927 info->mptcpi_flags = flags; 928 929 slow = lock_sock_fast(sk); 930 info->mptcpi_csum_enabled = READ_ONCE(msk->csum_enabled); 931 info->mptcpi_token = msk->token; 932 info->mptcpi_write_seq = msk->write_seq; 933 info->mptcpi_retransmits = inet_csk(sk)->icsk_retransmits; 934 info->mptcpi_bytes_sent = msk->bytes_sent; 935 info->mptcpi_bytes_received = msk->bytes_received; 936 info->mptcpi_bytes_retrans = msk->bytes_retrans; 937 info->mptcpi_subflows_total = info->mptcpi_subflows + 938 __mptcp_has_initial_subflow(msk); 939 now = tcp_jiffies32; 940 info->mptcpi_last_data_sent = jiffies_to_msecs(now - msk->last_data_sent); 941 info->mptcpi_last_data_recv = jiffies_to_msecs(now - msk->last_data_recv); 942 unlock_sock_fast(sk, slow); 943 944 mptcp_data_lock(sk); 945 info->mptcpi_last_ack_recv = jiffies_to_msecs(now - msk->last_ack_recv); 946 info->mptcpi_snd_una = msk->snd_una; 947 info->mptcpi_rcv_nxt = msk->ack_seq; 948 info->mptcpi_bytes_acked = msk->bytes_acked; 949 mptcp_data_unlock(sk); 950 } 951 EXPORT_SYMBOL_GPL(mptcp_diag_fill_info); 952 953 static int mptcp_getsockopt_info(struct mptcp_sock *msk, char __user *optval, int __user *optlen) 954 { 955 struct mptcp_info m_info; 956 int len; 957 958 if (get_user(len, optlen)) 959 return -EFAULT; 960 961 len = min_t(unsigned int, len, sizeof(struct mptcp_info)); 962 963 mptcp_diag_fill_info(msk, &m_info); 964 965 if (put_user(len, optlen)) 966 return -EFAULT; 967 968 if (copy_to_user(optval, &m_info, len)) 969 return -EFAULT; 970 971 return 0; 972 } 973 974 static int mptcp_put_subflow_data(struct mptcp_subflow_data *sfd, 975 char __user *optval, 976 u32 copied, 977 int __user *optlen) 978 { 979 u32 copylen = min_t(u32, sfd->size_subflow_data, sizeof(*sfd)); 980 981 if (copied) 982 copied += sfd->size_subflow_data; 983 else 984 copied = copylen; 985 986 if (put_user(copied, optlen)) 987 return -EFAULT; 988 989 if (copy_to_user(optval, sfd, copylen)) 990 return -EFAULT; 991 992 return 0; 993 } 994 995 static int mptcp_get_subflow_data(struct mptcp_subflow_data *sfd, 996 char __user *optval, 997 int __user *optlen) 998 { 999 int len, copylen; 1000 1001 if (get_user(len, optlen)) 1002 return -EFAULT; 1003 1004 /* if mptcp_subflow_data size is changed, need to adjust 1005 * this function to deal with programs using old version. 1006 */ 1007 BUILD_BUG_ON(sizeof(*sfd) != MIN_INFO_OPTLEN_SIZE); 1008 1009 if (len < MIN_INFO_OPTLEN_SIZE) 1010 return -EINVAL; 1011 1012 memset(sfd, 0, sizeof(*sfd)); 1013 1014 copylen = min_t(unsigned int, len, sizeof(*sfd)); 1015 if (copy_from_user(sfd, optval, copylen)) 1016 return -EFAULT; 1017 1018 /* size_subflow_data is u32, but len is signed */ 1019 if (sfd->size_subflow_data > INT_MAX || 1020 sfd->size_user > INT_MAX) 1021 return -EINVAL; 1022 1023 if (sfd->size_subflow_data < MIN_INFO_OPTLEN_SIZE || 1024 sfd->size_subflow_data > len) 1025 return -EINVAL; 1026 1027 if (sfd->num_subflows || sfd->size_kernel) 1028 return -EINVAL; 1029 1030 return len - sfd->size_subflow_data; 1031 } 1032 1033 static int mptcp_getsockopt_tcpinfo(struct mptcp_sock *msk, char __user *optval, 1034 int __user *optlen) 1035 { 1036 struct mptcp_subflow_context *subflow; 1037 struct sock *sk = (struct sock *)msk; 1038 unsigned int sfcount = 0, copied = 0; 1039 struct mptcp_subflow_data sfd; 1040 char __user *infoptr; 1041 int len; 1042 1043 len = mptcp_get_subflow_data(&sfd, optval, optlen); 1044 if (len < 0) 1045 return len; 1046 1047 sfd.size_kernel = sizeof(struct tcp_info); 1048 sfd.size_user = min_t(unsigned int, sfd.size_user, 1049 sizeof(struct tcp_info)); 1050 1051 infoptr = optval + sfd.size_subflow_data; 1052 1053 lock_sock(sk); 1054 1055 mptcp_for_each_subflow(msk, subflow) { 1056 struct sock *ssk = mptcp_subflow_tcp_sock(subflow); 1057 1058 ++sfcount; 1059 1060 if (len && len >= sfd.size_user) { 1061 struct tcp_info info; 1062 1063 tcp_get_info(ssk, &info); 1064 1065 if (copy_to_user(infoptr, &info, sfd.size_user)) { 1066 release_sock(sk); 1067 return -EFAULT; 1068 } 1069 1070 infoptr += sfd.size_user; 1071 copied += sfd.size_user; 1072 len -= sfd.size_user; 1073 } 1074 } 1075 1076 release_sock(sk); 1077 1078 sfd.num_subflows = sfcount; 1079 1080 if (mptcp_put_subflow_data(&sfd, optval, copied, optlen)) 1081 return -EFAULT; 1082 1083 return 0; 1084 } 1085 1086 static void mptcp_get_sub_addrs(const struct sock *sk, struct mptcp_subflow_addrs *a) 1087 { 1088 const struct inet_sock *inet = inet_sk(sk); 1089 1090 memset(a, 0, sizeof(*a)); 1091 1092 if (sk->sk_family == AF_INET) { 1093 a->sin_local.sin_family = AF_INET; 1094 a->sin_local.sin_port = inet->inet_sport; 1095 a->sin_local.sin_addr.s_addr = inet->inet_rcv_saddr; 1096 1097 if (!a->sin_local.sin_addr.s_addr) 1098 a->sin_local.sin_addr.s_addr = inet->inet_saddr; 1099 1100 a->sin_remote.sin_family = AF_INET; 1101 a->sin_remote.sin_port = inet->inet_dport; 1102 a->sin_remote.sin_addr.s_addr = inet->inet_daddr; 1103 #if IS_ENABLED(CONFIG_IPV6) 1104 } else if (sk->sk_family == AF_INET6) { 1105 const struct ipv6_pinfo *np = inet6_sk(sk); 1106 1107 if (WARN_ON_ONCE(!np)) 1108 return; 1109 1110 a->sin6_local.sin6_family = AF_INET6; 1111 a->sin6_local.sin6_port = inet->inet_sport; 1112 1113 if (ipv6_addr_any(&sk->sk_v6_rcv_saddr)) 1114 a->sin6_local.sin6_addr = np->saddr; 1115 else 1116 a->sin6_local.sin6_addr = sk->sk_v6_rcv_saddr; 1117 1118 a->sin6_remote.sin6_family = AF_INET6; 1119 a->sin6_remote.sin6_port = inet->inet_dport; 1120 a->sin6_remote.sin6_addr = sk->sk_v6_daddr; 1121 #endif 1122 } 1123 } 1124 1125 static int mptcp_getsockopt_subflow_addrs(struct mptcp_sock *msk, char __user *optval, 1126 int __user *optlen) 1127 { 1128 struct mptcp_subflow_context *subflow; 1129 struct sock *sk = (struct sock *)msk; 1130 unsigned int sfcount = 0, copied = 0; 1131 struct mptcp_subflow_data sfd; 1132 char __user *addrptr; 1133 int len; 1134 1135 len = mptcp_get_subflow_data(&sfd, optval, optlen); 1136 if (len < 0) 1137 return len; 1138 1139 sfd.size_kernel = sizeof(struct mptcp_subflow_addrs); 1140 sfd.size_user = min_t(unsigned int, sfd.size_user, 1141 sizeof(struct mptcp_subflow_addrs)); 1142 1143 addrptr = optval + sfd.size_subflow_data; 1144 1145 lock_sock(sk); 1146 1147 mptcp_for_each_subflow(msk, subflow) { 1148 struct sock *ssk = mptcp_subflow_tcp_sock(subflow); 1149 1150 ++sfcount; 1151 1152 if (len && len >= sfd.size_user) { 1153 struct mptcp_subflow_addrs a; 1154 1155 mptcp_get_sub_addrs(ssk, &a); 1156 1157 if (copy_to_user(addrptr, &a, sfd.size_user)) { 1158 release_sock(sk); 1159 return -EFAULT; 1160 } 1161 1162 addrptr += sfd.size_user; 1163 copied += sfd.size_user; 1164 len -= sfd.size_user; 1165 } 1166 } 1167 1168 release_sock(sk); 1169 1170 sfd.num_subflows = sfcount; 1171 1172 if (mptcp_put_subflow_data(&sfd, optval, copied, optlen)) 1173 return -EFAULT; 1174 1175 return 0; 1176 } 1177 1178 static int mptcp_get_full_info(struct mptcp_full_info *mfi, 1179 char __user *optval, 1180 int __user *optlen) 1181 { 1182 int len; 1183 1184 BUILD_BUG_ON(offsetof(struct mptcp_full_info, mptcp_info) != 1185 MIN_FULL_INFO_OPTLEN_SIZE); 1186 1187 if (get_user(len, optlen)) 1188 return -EFAULT; 1189 1190 if (len < MIN_FULL_INFO_OPTLEN_SIZE) 1191 return -EINVAL; 1192 1193 memset(mfi, 0, sizeof(*mfi)); 1194 if (copy_from_user(mfi, optval, MIN_FULL_INFO_OPTLEN_SIZE)) 1195 return -EFAULT; 1196 1197 if (mfi->size_tcpinfo_kernel || 1198 mfi->size_sfinfo_kernel || 1199 mfi->num_subflows) 1200 return -EINVAL; 1201 1202 if (mfi->size_sfinfo_user > INT_MAX || 1203 mfi->size_tcpinfo_user > INT_MAX) 1204 return -EINVAL; 1205 1206 return len - MIN_FULL_INFO_OPTLEN_SIZE; 1207 } 1208 1209 static int mptcp_put_full_info(struct mptcp_full_info *mfi, 1210 char __user *optval, 1211 u32 copylen, 1212 int __user *optlen) 1213 { 1214 copylen += MIN_FULL_INFO_OPTLEN_SIZE; 1215 if (put_user(copylen, optlen)) 1216 return -EFAULT; 1217 1218 if (copy_to_user(optval, mfi, copylen)) 1219 return -EFAULT; 1220 return 0; 1221 } 1222 1223 static int mptcp_getsockopt_full_info(struct mptcp_sock *msk, char __user *optval, 1224 int __user *optlen) 1225 { 1226 unsigned int sfcount = 0, copylen = 0; 1227 struct mptcp_subflow_context *subflow; 1228 struct sock *sk = (struct sock *)msk; 1229 void __user *tcpinfoptr, *sfinfoptr; 1230 struct mptcp_full_info mfi; 1231 int len; 1232 1233 len = mptcp_get_full_info(&mfi, optval, optlen); 1234 if (len < 0) 1235 return len; 1236 1237 /* don't bother filling the mptcp info if there is not enough 1238 * user-space-provided storage 1239 */ 1240 if (len > 0) { 1241 mptcp_diag_fill_info(msk, &mfi.mptcp_info); 1242 copylen += min_t(unsigned int, len, sizeof(struct mptcp_info)); 1243 } 1244 1245 mfi.size_tcpinfo_kernel = sizeof(struct tcp_info); 1246 mfi.size_tcpinfo_user = min_t(unsigned int, mfi.size_tcpinfo_user, 1247 sizeof(struct tcp_info)); 1248 sfinfoptr = u64_to_user_ptr(mfi.subflow_info); 1249 mfi.size_sfinfo_kernel = sizeof(struct mptcp_subflow_info); 1250 mfi.size_sfinfo_user = min_t(unsigned int, mfi.size_sfinfo_user, 1251 sizeof(struct mptcp_subflow_info)); 1252 tcpinfoptr = u64_to_user_ptr(mfi.tcp_info); 1253 1254 lock_sock(sk); 1255 mptcp_for_each_subflow(msk, subflow) { 1256 struct sock *ssk = mptcp_subflow_tcp_sock(subflow); 1257 struct mptcp_subflow_info sfinfo; 1258 struct tcp_info tcp_info; 1259 1260 if (sfcount++ >= mfi.size_arrays_user) 1261 continue; 1262 1263 /* fetch addr/tcp_info only if the user space buffers 1264 * are wide enough 1265 */ 1266 memset(&sfinfo, 0, sizeof(sfinfo)); 1267 sfinfo.id = subflow->subflow_id; 1268 if (mfi.size_sfinfo_user > 1269 offsetof(struct mptcp_subflow_info, addrs)) 1270 mptcp_get_sub_addrs(ssk, &sfinfo.addrs); 1271 if (copy_to_user(sfinfoptr, &sfinfo, mfi.size_sfinfo_user)) 1272 goto fail_release; 1273 1274 if (mfi.size_tcpinfo_user) { 1275 tcp_get_info(ssk, &tcp_info); 1276 if (copy_to_user(tcpinfoptr, &tcp_info, 1277 mfi.size_tcpinfo_user)) 1278 goto fail_release; 1279 } 1280 1281 tcpinfoptr += mfi.size_tcpinfo_user; 1282 sfinfoptr += mfi.size_sfinfo_user; 1283 } 1284 release_sock(sk); 1285 1286 mfi.num_subflows = sfcount; 1287 if (mptcp_put_full_info(&mfi, optval, copylen, optlen)) 1288 return -EFAULT; 1289 1290 return 0; 1291 1292 fail_release: 1293 release_sock(sk); 1294 return -EFAULT; 1295 } 1296 1297 static int mptcp_put_int_option(struct mptcp_sock *msk, char __user *optval, 1298 int __user *optlen, int val) 1299 { 1300 int len; 1301 1302 if (get_user(len, optlen)) 1303 return -EFAULT; 1304 if (len < 0) 1305 return -EINVAL; 1306 1307 if (len < sizeof(int) && len > 0 && val >= 0 && val <= 255) { 1308 unsigned char ucval = (unsigned char)val; 1309 1310 len = 1; 1311 if (put_user(len, optlen)) 1312 return -EFAULT; 1313 if (copy_to_user(optval, &ucval, 1)) 1314 return -EFAULT; 1315 } else { 1316 len = min_t(unsigned int, len, sizeof(int)); 1317 if (put_user(len, optlen)) 1318 return -EFAULT; 1319 if (copy_to_user(optval, &val, len)) 1320 return -EFAULT; 1321 } 1322 1323 return 0; 1324 } 1325 1326 static int mptcp_getsockopt_sol_tcp(struct mptcp_sock *msk, int optname, 1327 char __user *optval, int __user *optlen) 1328 { 1329 switch (optname) { 1330 case TCP_ULP: 1331 case TCP_CONGESTION: 1332 case TCP_INFO: 1333 case TCP_CC_INFO: 1334 case TCP_DEFER_ACCEPT: 1335 case TCP_FASTOPEN: 1336 case TCP_FASTOPEN_CONNECT: 1337 case TCP_FASTOPEN_KEY: 1338 case TCP_FASTOPEN_NO_COOKIE: 1339 return mptcp_getsockopt_first_sf_only(msk, SOL_TCP, optname, 1340 optval, optlen); 1341 case TCP_INQ: 1342 return mptcp_put_int_option(msk, optval, optlen, msk->recvmsg_inq); 1343 case TCP_CORK: 1344 return mptcp_put_int_option(msk, optval, optlen, msk->cork); 1345 case TCP_NODELAY: 1346 return mptcp_put_int_option(msk, optval, optlen, msk->nodelay); 1347 case TCP_NOTSENT_LOWAT: 1348 return mptcp_put_int_option(msk, optval, optlen, msk->notsent_lowat); 1349 case TCP_IS_MPTCP: 1350 return mptcp_put_int_option(msk, optval, optlen, 1); 1351 } 1352 return -EOPNOTSUPP; 1353 } 1354 1355 static int mptcp_getsockopt_v4(struct mptcp_sock *msk, int optname, 1356 char __user *optval, int __user *optlen) 1357 { 1358 struct sock *sk = (void *)msk; 1359 1360 switch (optname) { 1361 case IP_TOS: 1362 return mptcp_put_int_option(msk, optval, optlen, READ_ONCE(inet_sk(sk)->tos)); 1363 case IP_BIND_ADDRESS_NO_PORT: 1364 return mptcp_put_int_option(msk, optval, optlen, 1365 inet_test_bit(BIND_ADDRESS_NO_PORT, sk)); 1366 case IP_LOCAL_PORT_RANGE: 1367 return mptcp_put_int_option(msk, optval, optlen, 1368 READ_ONCE(inet_sk(sk)->local_port_range)); 1369 } 1370 1371 return -EOPNOTSUPP; 1372 } 1373 1374 static int mptcp_getsockopt_sol_mptcp(struct mptcp_sock *msk, int optname, 1375 char __user *optval, int __user *optlen) 1376 { 1377 switch (optname) { 1378 case MPTCP_INFO: 1379 return mptcp_getsockopt_info(msk, optval, optlen); 1380 case MPTCP_FULL_INFO: 1381 return mptcp_getsockopt_full_info(msk, optval, optlen); 1382 case MPTCP_TCPINFO: 1383 return mptcp_getsockopt_tcpinfo(msk, optval, optlen); 1384 case MPTCP_SUBFLOW_ADDRS: 1385 return mptcp_getsockopt_subflow_addrs(msk, optval, optlen); 1386 } 1387 1388 return -EOPNOTSUPP; 1389 } 1390 1391 int mptcp_getsockopt(struct sock *sk, int level, int optname, 1392 char __user *optval, int __user *option) 1393 { 1394 struct mptcp_sock *msk = mptcp_sk(sk); 1395 struct sock *ssk; 1396 1397 pr_debug("msk=%p", msk); 1398 1399 /* @@ the meaning of setsockopt() when the socket is connected and 1400 * there are multiple subflows is not yet defined. It is up to the 1401 * MPTCP-level socket to configure the subflows until the subflow 1402 * is in TCP fallback, when socket options are passed through 1403 * to the one remaining subflow. 1404 */ 1405 lock_sock(sk); 1406 ssk = __mptcp_tcp_fallback(msk); 1407 release_sock(sk); 1408 if (ssk) 1409 return tcp_getsockopt(ssk, level, optname, optval, option); 1410 1411 if (level == SOL_IP) 1412 return mptcp_getsockopt_v4(msk, optname, optval, option); 1413 if (level == SOL_TCP) 1414 return mptcp_getsockopt_sol_tcp(msk, optname, optval, option); 1415 if (level == SOL_MPTCP) 1416 return mptcp_getsockopt_sol_mptcp(msk, optname, optval, option); 1417 return -EOPNOTSUPP; 1418 } 1419 1420 static void sync_socket_options(struct mptcp_sock *msk, struct sock *ssk) 1421 { 1422 static const unsigned int tx_rx_locks = SOCK_RCVBUF_LOCK | SOCK_SNDBUF_LOCK; 1423 struct sock *sk = (struct sock *)msk; 1424 1425 if (ssk->sk_prot->keepalive) { 1426 if (sock_flag(sk, SOCK_KEEPOPEN)) 1427 ssk->sk_prot->keepalive(ssk, 1); 1428 else 1429 ssk->sk_prot->keepalive(ssk, 0); 1430 } 1431 1432 ssk->sk_priority = sk->sk_priority; 1433 ssk->sk_bound_dev_if = sk->sk_bound_dev_if; 1434 ssk->sk_incoming_cpu = sk->sk_incoming_cpu; 1435 ssk->sk_ipv6only = sk->sk_ipv6only; 1436 __ip_sock_set_tos(ssk, inet_sk(sk)->tos); 1437 1438 if (sk->sk_userlocks & tx_rx_locks) { 1439 ssk->sk_userlocks |= sk->sk_userlocks & tx_rx_locks; 1440 if (sk->sk_userlocks & SOCK_SNDBUF_LOCK) { 1441 WRITE_ONCE(ssk->sk_sndbuf, sk->sk_sndbuf); 1442 mptcp_subflow_ctx(ssk)->cached_sndbuf = sk->sk_sndbuf; 1443 } 1444 if (sk->sk_userlocks & SOCK_RCVBUF_LOCK) 1445 WRITE_ONCE(ssk->sk_rcvbuf, sk->sk_rcvbuf); 1446 } 1447 1448 if (sock_flag(sk, SOCK_LINGER)) { 1449 ssk->sk_lingertime = sk->sk_lingertime; 1450 sock_set_flag(ssk, SOCK_LINGER); 1451 } else { 1452 sock_reset_flag(ssk, SOCK_LINGER); 1453 } 1454 1455 if (sk->sk_mark != ssk->sk_mark) { 1456 ssk->sk_mark = sk->sk_mark; 1457 sk_dst_reset(ssk); 1458 } 1459 1460 sock_valbool_flag(ssk, SOCK_DBG, sock_flag(sk, SOCK_DBG)); 1461 1462 if (inet_csk(sk)->icsk_ca_ops != inet_csk(ssk)->icsk_ca_ops) 1463 tcp_set_congestion_control(ssk, msk->ca_name, false, true); 1464 __tcp_sock_set_cork(ssk, !!msk->cork); 1465 __tcp_sock_set_nodelay(ssk, !!msk->nodelay); 1466 1467 inet_assign_bit(TRANSPARENT, ssk, inet_test_bit(TRANSPARENT, sk)); 1468 inet_assign_bit(FREEBIND, ssk, inet_test_bit(FREEBIND, sk)); 1469 inet_assign_bit(BIND_ADDRESS_NO_PORT, ssk, inet_test_bit(BIND_ADDRESS_NO_PORT, sk)); 1470 WRITE_ONCE(inet_sk(ssk)->local_port_range, READ_ONCE(inet_sk(sk)->local_port_range)); 1471 } 1472 1473 void mptcp_sockopt_sync_locked(struct mptcp_sock *msk, struct sock *ssk) 1474 { 1475 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk); 1476 1477 msk_owned_by_me(msk); 1478 1479 ssk->sk_rcvlowat = 0; 1480 1481 /* subflows must ignore any latency-related settings: will not affect 1482 * the user-space - only the msk is relevant - but will foul the 1483 * mptcp scheduler 1484 */ 1485 tcp_sk(ssk)->notsent_lowat = UINT_MAX; 1486 1487 if (READ_ONCE(subflow->setsockopt_seq) != msk->setsockopt_seq) { 1488 sync_socket_options(msk, ssk); 1489 1490 subflow->setsockopt_seq = msk->setsockopt_seq; 1491 } 1492 } 1493 1494 /* unfortunately this is different enough from the tcp version so 1495 * that we can't factor it out 1496 */ 1497 int mptcp_set_rcvlowat(struct sock *sk, int val) 1498 { 1499 struct mptcp_subflow_context *subflow; 1500 int space, cap; 1501 1502 /* bpf can land here with a wrong sk type */ 1503 if (sk->sk_protocol == IPPROTO_TCP) 1504 return -EINVAL; 1505 1506 if (sk->sk_userlocks & SOCK_RCVBUF_LOCK) 1507 cap = sk->sk_rcvbuf >> 1; 1508 else 1509 cap = READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_rmem[2]) >> 1; 1510 val = min(val, cap); 1511 WRITE_ONCE(sk->sk_rcvlowat, val ? : 1); 1512 1513 /* Check if we need to signal EPOLLIN right now */ 1514 if (mptcp_epollin_ready(sk)) 1515 sk->sk_data_ready(sk); 1516 1517 if (sk->sk_userlocks & SOCK_RCVBUF_LOCK) 1518 return 0; 1519 1520 space = __tcp_space_from_win(mptcp_sk(sk)->scaling_ratio, val); 1521 if (space <= sk->sk_rcvbuf) 1522 return 0; 1523 1524 /* propagate the rcvbuf changes to all the subflows */ 1525 WRITE_ONCE(sk->sk_rcvbuf, space); 1526 mptcp_for_each_subflow(mptcp_sk(sk), subflow) { 1527 struct sock *ssk = mptcp_subflow_tcp_sock(subflow); 1528 bool slow; 1529 1530 slow = lock_sock_fast(ssk); 1531 WRITE_ONCE(ssk->sk_rcvbuf, space); 1532 WRITE_ONCE(tcp_sk(ssk)->window_clamp, val); 1533 unlock_sock_fast(ssk, slow); 1534 } 1535 return 0; 1536 } 1537