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 19 static struct sock *__mptcp_tcp_fallback(struct mptcp_sock *msk) 20 { 21 sock_owned_by_me((const struct sock *)msk); 22 23 if (likely(!__mptcp_check_fallback(msk))) 24 return NULL; 25 26 return msk->first; 27 } 28 29 static u32 sockopt_seq_reset(const struct sock *sk) 30 { 31 sock_owned_by_me(sk); 32 33 /* Highbits contain state. Allows to distinguish sockopt_seq 34 * of listener and established: 35 * s0 = new_listener() 36 * sockopt(s0) - seq is 1 37 * s1 = accept(s0) - s1 inherits seq 1 if listener sk (s0) 38 * sockopt(s0) - seq increments to 2 on s0 39 * sockopt(s1) // seq increments to 2 on s1 (different option) 40 * new ssk completes join, inherits options from s0 // seq 2 41 * Needs sync from mptcp join logic, but ssk->seq == msk->seq 42 * 43 * Set High order bits to sk_state so ssk->seq == msk->seq test 44 * will fail. 45 */ 46 47 return (u32)sk->sk_state << 24u; 48 } 49 50 static void sockopt_seq_inc(struct mptcp_sock *msk) 51 { 52 u32 seq = (msk->setsockopt_seq + 1) & 0x00ffffff; 53 54 msk->setsockopt_seq = sockopt_seq_reset((struct sock *)msk) + seq; 55 } 56 57 static int mptcp_get_int_option(struct mptcp_sock *msk, sockptr_t optval, 58 unsigned int optlen, int *val) 59 { 60 if (optlen < sizeof(int)) 61 return -EINVAL; 62 63 if (copy_from_sockptr(val, optval, sizeof(*val))) 64 return -EFAULT; 65 66 return 0; 67 } 68 69 static void mptcp_sol_socket_sync_intval(struct mptcp_sock *msk, int optname, int val) 70 { 71 struct mptcp_subflow_context *subflow; 72 struct sock *sk = (struct sock *)msk; 73 74 lock_sock(sk); 75 sockopt_seq_inc(msk); 76 77 mptcp_for_each_subflow(msk, subflow) { 78 struct sock *ssk = mptcp_subflow_tcp_sock(subflow); 79 bool slow = lock_sock_fast(ssk); 80 81 switch (optname) { 82 case SO_DEBUG: 83 sock_valbool_flag(ssk, SOCK_DBG, !!val); 84 break; 85 case SO_KEEPALIVE: 86 if (ssk->sk_prot->keepalive) 87 ssk->sk_prot->keepalive(ssk, !!val); 88 sock_valbool_flag(ssk, SOCK_KEEPOPEN, !!val); 89 break; 90 case SO_PRIORITY: 91 ssk->sk_priority = val; 92 break; 93 case SO_SNDBUF: 94 case SO_SNDBUFFORCE: 95 ssk->sk_userlocks |= SOCK_SNDBUF_LOCK; 96 WRITE_ONCE(ssk->sk_sndbuf, sk->sk_sndbuf); 97 break; 98 case SO_RCVBUF: 99 case SO_RCVBUFFORCE: 100 ssk->sk_userlocks |= SOCK_RCVBUF_LOCK; 101 WRITE_ONCE(ssk->sk_rcvbuf, sk->sk_rcvbuf); 102 break; 103 case SO_MARK: 104 if (READ_ONCE(ssk->sk_mark) != sk->sk_mark) { 105 ssk->sk_mark = sk->sk_mark; 106 sk_dst_reset(ssk); 107 } 108 break; 109 case SO_INCOMING_CPU: 110 WRITE_ONCE(ssk->sk_incoming_cpu, val); 111 break; 112 } 113 114 subflow->setsockopt_seq = msk->setsockopt_seq; 115 unlock_sock_fast(ssk, slow); 116 } 117 118 release_sock(sk); 119 } 120 121 static int mptcp_sol_socket_intval(struct mptcp_sock *msk, int optname, int val) 122 { 123 sockptr_t optval = KERNEL_SOCKPTR(&val); 124 struct sock *sk = (struct sock *)msk; 125 int ret; 126 127 ret = sock_setsockopt(sk->sk_socket, SOL_SOCKET, optname, 128 optval, sizeof(val)); 129 if (ret) 130 return ret; 131 132 mptcp_sol_socket_sync_intval(msk, optname, val); 133 return 0; 134 } 135 136 static void mptcp_so_incoming_cpu(struct mptcp_sock *msk, int val) 137 { 138 struct sock *sk = (struct sock *)msk; 139 140 WRITE_ONCE(sk->sk_incoming_cpu, val); 141 142 mptcp_sol_socket_sync_intval(msk, SO_INCOMING_CPU, val); 143 } 144 145 static int mptcp_setsockopt_sol_socket_tstamp(struct mptcp_sock *msk, int optname, int val) 146 { 147 sockptr_t optval = KERNEL_SOCKPTR(&val); 148 struct mptcp_subflow_context *subflow; 149 struct sock *sk = (struct sock *)msk; 150 int ret; 151 152 ret = sock_setsockopt(sk->sk_socket, SOL_SOCKET, optname, 153 optval, sizeof(val)); 154 if (ret) 155 return ret; 156 157 lock_sock(sk); 158 mptcp_for_each_subflow(msk, subflow) { 159 struct sock *ssk = mptcp_subflow_tcp_sock(subflow); 160 bool slow = lock_sock_fast(ssk); 161 162 sock_set_timestamp(sk, optname, !!val); 163 unlock_sock_fast(ssk, slow); 164 } 165 166 release_sock(sk); 167 return 0; 168 } 169 170 static int mptcp_setsockopt_sol_socket_int(struct mptcp_sock *msk, int optname, 171 sockptr_t optval, 172 unsigned int optlen) 173 { 174 int val, ret; 175 176 ret = mptcp_get_int_option(msk, optval, optlen, &val); 177 if (ret) 178 return ret; 179 180 switch (optname) { 181 case SO_KEEPALIVE: 182 mptcp_sol_socket_sync_intval(msk, optname, val); 183 return 0; 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 socket *ssock; 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 ssock = __mptcp_nmpc_socket(msk); 304 if (!ssock) { 305 release_sock(sk); 306 return -EINVAL; 307 } 308 309 ret = sock_setsockopt(ssock, SOL_SOCKET, optname, optval, optlen); 310 if (ret == 0) { 311 if (optname == SO_REUSEPORT) 312 sk->sk_reuseport = ssock->sk->sk_reuseport; 313 else if (optname == SO_REUSEADDR) 314 sk->sk_reuse = ssock->sk->sk_reuse; 315 else if (optname == SO_BINDTODEVICE) 316 sk->sk_bound_dev_if = ssock->sk->sk_bound_dev_if; 317 else if (optname == SO_BINDTOIFINDEX) 318 sk->sk_bound_dev_if = ssock->sk->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_BUSY_POLL: 347 case SO_PREFER_BUSY_POLL: 348 case SO_BUSY_POLL_BUDGET: 349 /* No need to copy: only relevant for msk */ 350 return sock_setsockopt(sk->sk_socket, SOL_SOCKET, optname, optval, optlen); 351 case SO_NO_CHECK: 352 case SO_DONTROUTE: 353 case SO_BROADCAST: 354 case SO_BSDCOMPAT: 355 case SO_PASSCRED: 356 case SO_PASSSEC: 357 case SO_RXQ_OVFL: 358 case SO_WIFI_STATUS: 359 case SO_NOFCS: 360 case SO_SELECT_ERR_QUEUE: 361 return 0; 362 } 363 364 /* SO_OOBINLINE is not supported, let's avoid the related mess 365 * SO_ATTACH_FILTER, SO_ATTACH_BPF, SO_ATTACH_REUSEPORT_CBPF, 366 * SO_DETACH_REUSEPORT_BPF, SO_DETACH_FILTER, SO_LOCK_FILTER, 367 * we must be careful with subflows 368 * 369 * SO_ATTACH_REUSEPORT_EBPF is not supported, at it checks 370 * explicitly the sk_protocol field 371 * 372 * SO_PEEK_OFF is unsupported, as it is for plain TCP 373 * SO_MAX_PACING_RATE is unsupported, we must be careful with subflows 374 * SO_CNX_ADVICE is currently unsupported, could possibly be relevant, 375 * but likely needs careful design 376 * 377 * SO_ZEROCOPY is currently unsupported, TODO in sndmsg 378 * SO_TXTIME is currently unsupported 379 */ 380 381 return -EOPNOTSUPP; 382 } 383 384 static int mptcp_setsockopt_v6(struct mptcp_sock *msk, int optname, 385 sockptr_t optval, unsigned int optlen) 386 { 387 struct sock *sk = (struct sock *)msk; 388 int ret = -EOPNOTSUPP; 389 struct socket *ssock; 390 391 switch (optname) { 392 case IPV6_V6ONLY: 393 case IPV6_TRANSPARENT: 394 case IPV6_FREEBIND: 395 lock_sock(sk); 396 ssock = __mptcp_nmpc_socket(msk); 397 if (!ssock) { 398 release_sock(sk); 399 return -EINVAL; 400 } 401 402 ret = tcp_setsockopt(ssock->sk, SOL_IPV6, optname, optval, optlen); 403 if (ret != 0) { 404 release_sock(sk); 405 return ret; 406 } 407 408 sockopt_seq_inc(msk); 409 410 switch (optname) { 411 case IPV6_V6ONLY: 412 sk->sk_ipv6only = ssock->sk->sk_ipv6only; 413 break; 414 case IPV6_TRANSPARENT: 415 inet_sk(sk)->transparent = inet_sk(ssock->sk)->transparent; 416 break; 417 case IPV6_FREEBIND: 418 inet_sk(sk)->freebind = inet_sk(ssock->sk)->freebind; 419 break; 420 } 421 422 release_sock(sk); 423 break; 424 } 425 426 return ret; 427 } 428 429 static bool mptcp_supported_sockopt(int level, int optname) 430 { 431 if (level == SOL_IP) { 432 switch (optname) { 433 /* should work fine */ 434 case IP_FREEBIND: 435 case IP_TRANSPARENT: 436 437 /* the following are control cmsg related */ 438 case IP_PKTINFO: 439 case IP_RECVTTL: 440 case IP_RECVTOS: 441 case IP_RECVOPTS: 442 case IP_RETOPTS: 443 case IP_PASSSEC: 444 case IP_RECVORIGDSTADDR: 445 case IP_CHECKSUM: 446 case IP_RECVFRAGSIZE: 447 448 /* common stuff that need some love */ 449 case IP_TOS: 450 case IP_TTL: 451 case IP_BIND_ADDRESS_NO_PORT: 452 case IP_MTU_DISCOVER: 453 case IP_RECVERR: 454 455 /* possibly less common may deserve some love */ 456 case IP_MINTTL: 457 458 /* the following is apparently a no-op for plain TCP */ 459 case IP_RECVERR_RFC4884: 460 return true; 461 } 462 463 /* IP_OPTIONS is not supported, needs subflow care */ 464 /* IP_HDRINCL, IP_NODEFRAG are not supported, RAW specific */ 465 /* IP_MULTICAST_TTL, IP_MULTICAST_LOOP, IP_UNICAST_IF, 466 * IP_ADD_MEMBERSHIP, IP_ADD_SOURCE_MEMBERSHIP, IP_DROP_MEMBERSHIP, 467 * IP_DROP_SOURCE_MEMBERSHIP, IP_BLOCK_SOURCE, IP_UNBLOCK_SOURCE, 468 * MCAST_JOIN_GROUP, MCAST_LEAVE_GROUP MCAST_JOIN_SOURCE_GROUP, 469 * MCAST_LEAVE_SOURCE_GROUP, MCAST_BLOCK_SOURCE, MCAST_UNBLOCK_SOURCE, 470 * MCAST_MSFILTER, IP_MULTICAST_ALL are not supported, better not deal 471 * with mcast stuff 472 */ 473 /* IP_IPSEC_POLICY, IP_XFRM_POLICY are nut supported, unrelated here */ 474 return false; 475 } 476 if (level == SOL_IPV6) { 477 switch (optname) { 478 case IPV6_V6ONLY: 479 480 /* the following are control cmsg related */ 481 case IPV6_RECVPKTINFO: 482 case IPV6_2292PKTINFO: 483 case IPV6_RECVHOPLIMIT: 484 case IPV6_2292HOPLIMIT: 485 case IPV6_RECVRTHDR: 486 case IPV6_2292RTHDR: 487 case IPV6_RECVHOPOPTS: 488 case IPV6_2292HOPOPTS: 489 case IPV6_RECVDSTOPTS: 490 case IPV6_2292DSTOPTS: 491 case IPV6_RECVTCLASS: 492 case IPV6_FLOWINFO: 493 case IPV6_RECVPATHMTU: 494 case IPV6_RECVORIGDSTADDR: 495 case IPV6_RECVFRAGSIZE: 496 497 /* the following ones need some love but are quite common */ 498 case IPV6_TCLASS: 499 case IPV6_TRANSPARENT: 500 case IPV6_FREEBIND: 501 case IPV6_PKTINFO: 502 case IPV6_2292PKTOPTIONS: 503 case IPV6_UNICAST_HOPS: 504 case IPV6_MTU_DISCOVER: 505 case IPV6_MTU: 506 case IPV6_RECVERR: 507 case IPV6_FLOWINFO_SEND: 508 case IPV6_FLOWLABEL_MGR: 509 case IPV6_MINHOPCOUNT: 510 case IPV6_DONTFRAG: 511 case IPV6_AUTOFLOWLABEL: 512 513 /* the following one is a no-op for plain TCP */ 514 case IPV6_RECVERR_RFC4884: 515 return true; 516 } 517 518 /* IPV6_HOPOPTS, IPV6_RTHDRDSTOPTS, IPV6_RTHDR, IPV6_DSTOPTS are 519 * not supported 520 */ 521 /* IPV6_MULTICAST_HOPS, IPV6_MULTICAST_LOOP, IPV6_UNICAST_IF, 522 * IPV6_MULTICAST_IF, IPV6_ADDRFORM, 523 * IPV6_ADD_MEMBERSHIP, IPV6_DROP_MEMBERSHIP, IPV6_JOIN_ANYCAST, 524 * IPV6_LEAVE_ANYCAST, IPV6_MULTICAST_ALL, MCAST_JOIN_GROUP, MCAST_LEAVE_GROUP, 525 * MCAST_JOIN_SOURCE_GROUP, MCAST_LEAVE_SOURCE_GROUP, 526 * MCAST_BLOCK_SOURCE, MCAST_UNBLOCK_SOURCE, MCAST_MSFILTER 527 * are not supported better not deal with mcast 528 */ 529 /* IPV6_ROUTER_ALERT, IPV6_ROUTER_ALERT_ISOLATE are not supported, since are evil */ 530 531 /* IPV6_IPSEC_POLICY, IPV6_XFRM_POLICY are not supported */ 532 /* IPV6_ADDR_PREFERENCES is not supported, we must be careful with subflows */ 533 return false; 534 } 535 if (level == SOL_TCP) { 536 switch (optname) { 537 /* the following are no-op or should work just fine */ 538 case TCP_THIN_DUPACK: 539 case TCP_DEFER_ACCEPT: 540 541 /* the following need some love */ 542 case TCP_MAXSEG: 543 case TCP_NODELAY: 544 case TCP_THIN_LINEAR_TIMEOUTS: 545 case TCP_CONGESTION: 546 case TCP_CORK: 547 case TCP_KEEPIDLE: 548 case TCP_KEEPINTVL: 549 case TCP_KEEPCNT: 550 case TCP_SYNCNT: 551 case TCP_SAVE_SYN: 552 case TCP_LINGER2: 553 case TCP_WINDOW_CLAMP: 554 case TCP_QUICKACK: 555 case TCP_USER_TIMEOUT: 556 case TCP_TIMESTAMP: 557 case TCP_NOTSENT_LOWAT: 558 case TCP_TX_DELAY: 559 case TCP_INQ: 560 return true; 561 } 562 563 /* TCP_MD5SIG, TCP_MD5SIG_EXT are not supported, MD5 is not compatible with MPTCP */ 564 565 /* TCP_REPAIR, TCP_REPAIR_QUEUE, TCP_QUEUE_SEQ, TCP_REPAIR_OPTIONS, 566 * TCP_REPAIR_WINDOW are not supported, better avoid this mess 567 */ 568 /* TCP_FASTOPEN_KEY, TCP_FASTOPEN TCP_FASTOPEN_CONNECT, TCP_FASTOPEN_NO_COOKIE, 569 * are not supported fastopen is currently unsupported 570 */ 571 } 572 return false; 573 } 574 575 static int mptcp_setsockopt_sol_tcp_congestion(struct mptcp_sock *msk, sockptr_t optval, 576 unsigned int optlen) 577 { 578 struct mptcp_subflow_context *subflow; 579 struct sock *sk = (struct sock *)msk; 580 char name[TCP_CA_NAME_MAX]; 581 bool cap_net_admin; 582 int ret; 583 584 if (optlen < 1) 585 return -EINVAL; 586 587 ret = strncpy_from_sockptr(name, optval, 588 min_t(long, TCP_CA_NAME_MAX - 1, optlen)); 589 if (ret < 0) 590 return -EFAULT; 591 592 name[ret] = 0; 593 594 cap_net_admin = ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN); 595 596 ret = 0; 597 lock_sock(sk); 598 sockopt_seq_inc(msk); 599 mptcp_for_each_subflow(msk, subflow) { 600 struct sock *ssk = mptcp_subflow_tcp_sock(subflow); 601 int err; 602 603 lock_sock(ssk); 604 err = tcp_set_congestion_control(ssk, name, true, cap_net_admin); 605 if (err < 0 && ret == 0) 606 ret = err; 607 subflow->setsockopt_seq = msk->setsockopt_seq; 608 release_sock(ssk); 609 } 610 611 if (ret == 0) 612 strcpy(msk->ca_name, name); 613 614 release_sock(sk); 615 return ret; 616 } 617 618 static int mptcp_setsockopt_sol_tcp_cork(struct mptcp_sock *msk, sockptr_t optval, 619 unsigned int optlen) 620 { 621 struct mptcp_subflow_context *subflow; 622 struct sock *sk = (struct sock *)msk; 623 int val; 624 625 if (optlen < sizeof(int)) 626 return -EINVAL; 627 628 if (copy_from_sockptr(&val, optval, sizeof(val))) 629 return -EFAULT; 630 631 lock_sock(sk); 632 sockopt_seq_inc(msk); 633 msk->cork = !!val; 634 mptcp_for_each_subflow(msk, subflow) { 635 struct sock *ssk = mptcp_subflow_tcp_sock(subflow); 636 637 lock_sock(ssk); 638 __tcp_sock_set_cork(ssk, !!val); 639 release_sock(ssk); 640 } 641 if (!val) 642 mptcp_check_and_set_pending(sk); 643 release_sock(sk); 644 645 return 0; 646 } 647 648 static int mptcp_setsockopt_sol_tcp_nodelay(struct mptcp_sock *msk, sockptr_t optval, 649 unsigned int optlen) 650 { 651 struct mptcp_subflow_context *subflow; 652 struct sock *sk = (struct sock *)msk; 653 int val; 654 655 if (optlen < sizeof(int)) 656 return -EINVAL; 657 658 if (copy_from_sockptr(&val, optval, sizeof(val))) 659 return -EFAULT; 660 661 lock_sock(sk); 662 sockopt_seq_inc(msk); 663 msk->nodelay = !!val; 664 mptcp_for_each_subflow(msk, subflow) { 665 struct sock *ssk = mptcp_subflow_tcp_sock(subflow); 666 667 lock_sock(ssk); 668 __tcp_sock_set_nodelay(ssk, !!val); 669 release_sock(ssk); 670 } 671 if (val) 672 mptcp_check_and_set_pending(sk); 673 release_sock(sk); 674 675 return 0; 676 } 677 678 static int mptcp_setsockopt_sol_ip_set_transparent(struct mptcp_sock *msk, int optname, 679 sockptr_t optval, unsigned int optlen) 680 { 681 struct sock *sk = (struct sock *)msk; 682 struct inet_sock *issk; 683 struct socket *ssock; 684 int err; 685 686 err = ip_setsockopt(sk, SOL_IP, optname, optval, optlen); 687 if (err != 0) 688 return err; 689 690 lock_sock(sk); 691 692 ssock = __mptcp_nmpc_socket(msk); 693 if (!ssock) { 694 release_sock(sk); 695 return -EINVAL; 696 } 697 698 issk = inet_sk(ssock->sk); 699 700 switch (optname) { 701 case IP_FREEBIND: 702 issk->freebind = inet_sk(sk)->freebind; 703 break; 704 case IP_TRANSPARENT: 705 issk->transparent = inet_sk(sk)->transparent; 706 break; 707 default: 708 release_sock(sk); 709 WARN_ON_ONCE(1); 710 return -EOPNOTSUPP; 711 } 712 713 sockopt_seq_inc(msk); 714 release_sock(sk); 715 return 0; 716 } 717 718 static int mptcp_setsockopt_v4_set_tos(struct mptcp_sock *msk, int optname, 719 sockptr_t optval, unsigned int optlen) 720 { 721 struct mptcp_subflow_context *subflow; 722 struct sock *sk = (struct sock *)msk; 723 int err, val; 724 725 err = ip_setsockopt(sk, SOL_IP, optname, optval, optlen); 726 727 if (err != 0) 728 return err; 729 730 lock_sock(sk); 731 sockopt_seq_inc(msk); 732 val = inet_sk(sk)->tos; 733 mptcp_for_each_subflow(msk, subflow) { 734 struct sock *ssk = mptcp_subflow_tcp_sock(subflow); 735 736 __ip_sock_set_tos(ssk, val); 737 } 738 release_sock(sk); 739 740 return err; 741 } 742 743 static int mptcp_setsockopt_v4(struct mptcp_sock *msk, int optname, 744 sockptr_t optval, unsigned int optlen) 745 { 746 switch (optname) { 747 case IP_FREEBIND: 748 case IP_TRANSPARENT: 749 return mptcp_setsockopt_sol_ip_set_transparent(msk, optname, optval, optlen); 750 case IP_TOS: 751 return mptcp_setsockopt_v4_set_tos(msk, optname, optval, optlen); 752 } 753 754 return -EOPNOTSUPP; 755 } 756 757 static int mptcp_setsockopt_sol_tcp(struct mptcp_sock *msk, int optname, 758 sockptr_t optval, unsigned int optlen) 759 { 760 struct sock *sk = (void *)msk; 761 int ret, val; 762 763 switch (optname) { 764 case TCP_INQ: 765 ret = mptcp_get_int_option(msk, optval, optlen, &val); 766 if (ret) 767 return ret; 768 if (val < 0 || val > 1) 769 return -EINVAL; 770 771 lock_sock(sk); 772 msk->recvmsg_inq = !!val; 773 release_sock(sk); 774 return 0; 775 case TCP_ULP: 776 return -EOPNOTSUPP; 777 case TCP_CONGESTION: 778 return mptcp_setsockopt_sol_tcp_congestion(msk, optval, optlen); 779 case TCP_CORK: 780 return mptcp_setsockopt_sol_tcp_cork(msk, optval, optlen); 781 case TCP_NODELAY: 782 return mptcp_setsockopt_sol_tcp_nodelay(msk, optval, optlen); 783 } 784 785 return -EOPNOTSUPP; 786 } 787 788 int mptcp_setsockopt(struct sock *sk, int level, int optname, 789 sockptr_t optval, unsigned int optlen) 790 { 791 struct mptcp_sock *msk = mptcp_sk(sk); 792 struct sock *ssk; 793 794 pr_debug("msk=%p", msk); 795 796 if (level == SOL_SOCKET) 797 return mptcp_setsockopt_sol_socket(msk, optname, optval, optlen); 798 799 if (!mptcp_supported_sockopt(level, optname)) 800 return -ENOPROTOOPT; 801 802 /* @@ the meaning of setsockopt() when the socket is connected and 803 * there are multiple subflows is not yet defined. It is up to the 804 * MPTCP-level socket to configure the subflows until the subflow 805 * is in TCP fallback, when TCP socket options are passed through 806 * to the one remaining subflow. 807 */ 808 lock_sock(sk); 809 ssk = __mptcp_tcp_fallback(msk); 810 release_sock(sk); 811 if (ssk) 812 return tcp_setsockopt(ssk, level, optname, optval, optlen); 813 814 if (level == SOL_IP) 815 return mptcp_setsockopt_v4(msk, optname, optval, optlen); 816 817 if (level == SOL_IPV6) 818 return mptcp_setsockopt_v6(msk, optname, optval, optlen); 819 820 if (level == SOL_TCP) 821 return mptcp_setsockopt_sol_tcp(msk, optname, optval, optlen); 822 823 return -EOPNOTSUPP; 824 } 825 826 static int mptcp_getsockopt_first_sf_only(struct mptcp_sock *msk, int level, int optname, 827 char __user *optval, int __user *optlen) 828 { 829 struct sock *sk = (struct sock *)msk; 830 struct socket *ssock; 831 int ret = -EINVAL; 832 struct sock *ssk; 833 834 lock_sock(sk); 835 ssk = msk->first; 836 if (ssk) { 837 ret = tcp_getsockopt(ssk, level, optname, optval, optlen); 838 goto out; 839 } 840 841 ssock = __mptcp_nmpc_socket(msk); 842 if (!ssock) 843 goto out; 844 845 ret = tcp_getsockopt(ssock->sk, level, optname, optval, optlen); 846 847 out: 848 release_sock(sk); 849 return ret; 850 } 851 852 void mptcp_diag_fill_info(struct mptcp_sock *msk, struct mptcp_info *info) 853 { 854 struct sock *sk = &msk->sk.icsk_inet.sk; 855 u32 flags = 0; 856 bool slow; 857 u8 val; 858 859 memset(info, 0, sizeof(*info)); 860 861 slow = lock_sock_fast(sk); 862 863 info->mptcpi_subflows = READ_ONCE(msk->pm.subflows); 864 info->mptcpi_add_addr_signal = READ_ONCE(msk->pm.add_addr_signaled); 865 info->mptcpi_add_addr_accepted = READ_ONCE(msk->pm.add_addr_accepted); 866 info->mptcpi_local_addr_used = READ_ONCE(msk->pm.local_addr_used); 867 info->mptcpi_subflows_max = mptcp_pm_get_subflows_max(msk); 868 val = mptcp_pm_get_add_addr_signal_max(msk); 869 info->mptcpi_add_addr_signal_max = val; 870 val = mptcp_pm_get_add_addr_accept_max(msk); 871 info->mptcpi_add_addr_accepted_max = val; 872 info->mptcpi_local_addr_max = mptcp_pm_get_local_addr_max(msk); 873 if (test_bit(MPTCP_FALLBACK_DONE, &msk->flags)) 874 flags |= MPTCP_INFO_FLAG_FALLBACK; 875 if (READ_ONCE(msk->can_ack)) 876 flags |= MPTCP_INFO_FLAG_REMOTE_KEY_RECEIVED; 877 info->mptcpi_flags = flags; 878 info->mptcpi_token = READ_ONCE(msk->token); 879 info->mptcpi_write_seq = READ_ONCE(msk->write_seq); 880 info->mptcpi_snd_una = READ_ONCE(msk->snd_una); 881 info->mptcpi_rcv_nxt = READ_ONCE(msk->ack_seq); 882 info->mptcpi_csum_enabled = READ_ONCE(msk->csum_enabled); 883 884 unlock_sock_fast(sk, slow); 885 } 886 EXPORT_SYMBOL_GPL(mptcp_diag_fill_info); 887 888 static int mptcp_getsockopt_info(struct mptcp_sock *msk, char __user *optval, int __user *optlen) 889 { 890 struct mptcp_info m_info; 891 int len; 892 893 if (get_user(len, optlen)) 894 return -EFAULT; 895 896 len = min_t(unsigned int, len, sizeof(struct mptcp_info)); 897 898 mptcp_diag_fill_info(msk, &m_info); 899 900 if (put_user(len, optlen)) 901 return -EFAULT; 902 903 if (copy_to_user(optval, &m_info, len)) 904 return -EFAULT; 905 906 return 0; 907 } 908 909 static int mptcp_put_subflow_data(struct mptcp_subflow_data *sfd, 910 char __user *optval, 911 u32 copied, 912 int __user *optlen) 913 { 914 u32 copylen = min_t(u32, sfd->size_subflow_data, sizeof(*sfd)); 915 916 if (copied) 917 copied += sfd->size_subflow_data; 918 else 919 copied = copylen; 920 921 if (put_user(copied, optlen)) 922 return -EFAULT; 923 924 if (copy_to_user(optval, sfd, copylen)) 925 return -EFAULT; 926 927 return 0; 928 } 929 930 static int mptcp_get_subflow_data(struct mptcp_subflow_data *sfd, 931 char __user *optval, int __user *optlen) 932 { 933 int len, copylen; 934 935 if (get_user(len, optlen)) 936 return -EFAULT; 937 938 /* if mptcp_subflow_data size is changed, need to adjust 939 * this function to deal with programs using old version. 940 */ 941 BUILD_BUG_ON(sizeof(*sfd) != MIN_INFO_OPTLEN_SIZE); 942 943 if (len < MIN_INFO_OPTLEN_SIZE) 944 return -EINVAL; 945 946 memset(sfd, 0, sizeof(*sfd)); 947 948 copylen = min_t(unsigned int, len, sizeof(*sfd)); 949 if (copy_from_user(sfd, optval, copylen)) 950 return -EFAULT; 951 952 /* size_subflow_data is u32, but len is signed */ 953 if (sfd->size_subflow_data > INT_MAX || 954 sfd->size_user > INT_MAX) 955 return -EINVAL; 956 957 if (sfd->size_subflow_data < MIN_INFO_OPTLEN_SIZE || 958 sfd->size_subflow_data > len) 959 return -EINVAL; 960 961 if (sfd->num_subflows || sfd->size_kernel) 962 return -EINVAL; 963 964 return len - sfd->size_subflow_data; 965 } 966 967 static int mptcp_getsockopt_tcpinfo(struct mptcp_sock *msk, char __user *optval, 968 int __user *optlen) 969 { 970 struct mptcp_subflow_context *subflow; 971 struct sock *sk = &msk->sk.icsk_inet.sk; 972 unsigned int sfcount = 0, copied = 0; 973 struct mptcp_subflow_data sfd; 974 char __user *infoptr; 975 int len; 976 977 len = mptcp_get_subflow_data(&sfd, optval, optlen); 978 if (len < 0) 979 return len; 980 981 sfd.size_kernel = sizeof(struct tcp_info); 982 sfd.size_user = min_t(unsigned int, sfd.size_user, 983 sizeof(struct tcp_info)); 984 985 infoptr = optval + sfd.size_subflow_data; 986 987 lock_sock(sk); 988 989 mptcp_for_each_subflow(msk, subflow) { 990 struct sock *ssk = mptcp_subflow_tcp_sock(subflow); 991 992 ++sfcount; 993 994 if (len && len >= sfd.size_user) { 995 struct tcp_info info; 996 997 tcp_get_info(ssk, &info); 998 999 if (copy_to_user(infoptr, &info, sfd.size_user)) { 1000 release_sock(sk); 1001 return -EFAULT; 1002 } 1003 1004 infoptr += sfd.size_user; 1005 copied += sfd.size_user; 1006 len -= sfd.size_user; 1007 } 1008 } 1009 1010 release_sock(sk); 1011 1012 sfd.num_subflows = sfcount; 1013 1014 if (mptcp_put_subflow_data(&sfd, optval, copied, optlen)) 1015 return -EFAULT; 1016 1017 return 0; 1018 } 1019 1020 static void mptcp_get_sub_addrs(const struct sock *sk, struct mptcp_subflow_addrs *a) 1021 { 1022 struct inet_sock *inet = inet_sk(sk); 1023 1024 memset(a, 0, sizeof(*a)); 1025 1026 if (sk->sk_family == AF_INET) { 1027 a->sin_local.sin_family = AF_INET; 1028 a->sin_local.sin_port = inet->inet_sport; 1029 a->sin_local.sin_addr.s_addr = inet->inet_rcv_saddr; 1030 1031 if (!a->sin_local.sin_addr.s_addr) 1032 a->sin_local.sin_addr.s_addr = inet->inet_saddr; 1033 1034 a->sin_remote.sin_family = AF_INET; 1035 a->sin_remote.sin_port = inet->inet_dport; 1036 a->sin_remote.sin_addr.s_addr = inet->inet_daddr; 1037 #if IS_ENABLED(CONFIG_IPV6) 1038 } else if (sk->sk_family == AF_INET6) { 1039 const struct ipv6_pinfo *np = inet6_sk(sk); 1040 1041 if (WARN_ON_ONCE(!np)) 1042 return; 1043 1044 a->sin6_local.sin6_family = AF_INET6; 1045 a->sin6_local.sin6_port = inet->inet_sport; 1046 1047 if (ipv6_addr_any(&sk->sk_v6_rcv_saddr)) 1048 a->sin6_local.sin6_addr = np->saddr; 1049 else 1050 a->sin6_local.sin6_addr = sk->sk_v6_rcv_saddr; 1051 1052 a->sin6_remote.sin6_family = AF_INET6; 1053 a->sin6_remote.sin6_port = inet->inet_dport; 1054 a->sin6_remote.sin6_addr = sk->sk_v6_daddr; 1055 #endif 1056 } 1057 } 1058 1059 static int mptcp_getsockopt_subflow_addrs(struct mptcp_sock *msk, char __user *optval, 1060 int __user *optlen) 1061 { 1062 struct sock *sk = &msk->sk.icsk_inet.sk; 1063 struct mptcp_subflow_context *subflow; 1064 unsigned int sfcount = 0, copied = 0; 1065 struct mptcp_subflow_data sfd; 1066 char __user *addrptr; 1067 int len; 1068 1069 len = mptcp_get_subflow_data(&sfd, optval, optlen); 1070 if (len < 0) 1071 return len; 1072 1073 sfd.size_kernel = sizeof(struct mptcp_subflow_addrs); 1074 sfd.size_user = min_t(unsigned int, sfd.size_user, 1075 sizeof(struct mptcp_subflow_addrs)); 1076 1077 addrptr = optval + sfd.size_subflow_data; 1078 1079 lock_sock(sk); 1080 1081 mptcp_for_each_subflow(msk, subflow) { 1082 struct sock *ssk = mptcp_subflow_tcp_sock(subflow); 1083 1084 ++sfcount; 1085 1086 if (len && len >= sfd.size_user) { 1087 struct mptcp_subflow_addrs a; 1088 1089 mptcp_get_sub_addrs(ssk, &a); 1090 1091 if (copy_to_user(addrptr, &a, sfd.size_user)) { 1092 release_sock(sk); 1093 return -EFAULT; 1094 } 1095 1096 addrptr += sfd.size_user; 1097 copied += sfd.size_user; 1098 len -= sfd.size_user; 1099 } 1100 } 1101 1102 release_sock(sk); 1103 1104 sfd.num_subflows = sfcount; 1105 1106 if (mptcp_put_subflow_data(&sfd, optval, copied, optlen)) 1107 return -EFAULT; 1108 1109 return 0; 1110 } 1111 1112 static int mptcp_put_int_option(struct mptcp_sock *msk, char __user *optval, 1113 int __user *optlen, int val) 1114 { 1115 int len; 1116 1117 if (get_user(len, optlen)) 1118 return -EFAULT; 1119 if (len < 0) 1120 return -EINVAL; 1121 1122 if (len < sizeof(int) && len > 0 && val >= 0 && val <= 255) { 1123 unsigned char ucval = (unsigned char)val; 1124 1125 len = 1; 1126 if (put_user(len, optlen)) 1127 return -EFAULT; 1128 if (copy_to_user(optval, &ucval, 1)) 1129 return -EFAULT; 1130 } else { 1131 len = min_t(unsigned int, len, sizeof(int)); 1132 if (put_user(len, optlen)) 1133 return -EFAULT; 1134 if (copy_to_user(optval, &val, len)) 1135 return -EFAULT; 1136 } 1137 1138 return 0; 1139 } 1140 1141 static int mptcp_getsockopt_sol_tcp(struct mptcp_sock *msk, int optname, 1142 char __user *optval, int __user *optlen) 1143 { 1144 switch (optname) { 1145 case TCP_ULP: 1146 case TCP_CONGESTION: 1147 case TCP_INFO: 1148 case TCP_CC_INFO: 1149 return mptcp_getsockopt_first_sf_only(msk, SOL_TCP, optname, 1150 optval, optlen); 1151 case TCP_INQ: 1152 return mptcp_put_int_option(msk, optval, optlen, msk->recvmsg_inq); 1153 case TCP_CORK: 1154 return mptcp_put_int_option(msk, optval, optlen, msk->cork); 1155 case TCP_NODELAY: 1156 return mptcp_put_int_option(msk, optval, optlen, msk->nodelay); 1157 } 1158 return -EOPNOTSUPP; 1159 } 1160 1161 static int mptcp_getsockopt_v4(struct mptcp_sock *msk, int optname, 1162 char __user *optval, int __user *optlen) 1163 { 1164 struct sock *sk = (void *)msk; 1165 1166 switch (optname) { 1167 case IP_TOS: 1168 return mptcp_put_int_option(msk, optval, optlen, inet_sk(sk)->tos); 1169 } 1170 1171 return -EOPNOTSUPP; 1172 } 1173 1174 static int mptcp_getsockopt_sol_mptcp(struct mptcp_sock *msk, int optname, 1175 char __user *optval, int __user *optlen) 1176 { 1177 switch (optname) { 1178 case MPTCP_INFO: 1179 return mptcp_getsockopt_info(msk, optval, optlen); 1180 case MPTCP_TCPINFO: 1181 return mptcp_getsockopt_tcpinfo(msk, optval, optlen); 1182 case MPTCP_SUBFLOW_ADDRS: 1183 return mptcp_getsockopt_subflow_addrs(msk, optval, optlen); 1184 } 1185 1186 return -EOPNOTSUPP; 1187 } 1188 1189 int mptcp_getsockopt(struct sock *sk, int level, int optname, 1190 char __user *optval, int __user *option) 1191 { 1192 struct mptcp_sock *msk = mptcp_sk(sk); 1193 struct sock *ssk; 1194 1195 pr_debug("msk=%p", msk); 1196 1197 /* @@ the meaning of setsockopt() when the socket is connected and 1198 * there are multiple subflows is not yet defined. It is up to the 1199 * MPTCP-level socket to configure the subflows until the subflow 1200 * is in TCP fallback, when socket options are passed through 1201 * to the one remaining subflow. 1202 */ 1203 lock_sock(sk); 1204 ssk = __mptcp_tcp_fallback(msk); 1205 release_sock(sk); 1206 if (ssk) 1207 return tcp_getsockopt(ssk, level, optname, optval, option); 1208 1209 if (level == SOL_IP) 1210 return mptcp_getsockopt_v4(msk, optname, optval, option); 1211 if (level == SOL_TCP) 1212 return mptcp_getsockopt_sol_tcp(msk, optname, optval, option); 1213 if (level == SOL_MPTCP) 1214 return mptcp_getsockopt_sol_mptcp(msk, optname, optval, option); 1215 return -EOPNOTSUPP; 1216 } 1217 1218 static void sync_socket_options(struct mptcp_sock *msk, struct sock *ssk) 1219 { 1220 static const unsigned int tx_rx_locks = SOCK_RCVBUF_LOCK | SOCK_SNDBUF_LOCK; 1221 struct sock *sk = (struct sock *)msk; 1222 1223 if (ssk->sk_prot->keepalive) { 1224 if (sock_flag(sk, SOCK_KEEPOPEN)) 1225 ssk->sk_prot->keepalive(ssk, 1); 1226 else 1227 ssk->sk_prot->keepalive(ssk, 0); 1228 } 1229 1230 ssk->sk_priority = sk->sk_priority; 1231 ssk->sk_bound_dev_if = sk->sk_bound_dev_if; 1232 ssk->sk_incoming_cpu = sk->sk_incoming_cpu; 1233 __ip_sock_set_tos(ssk, inet_sk(sk)->tos); 1234 1235 if (sk->sk_userlocks & tx_rx_locks) { 1236 ssk->sk_userlocks |= sk->sk_userlocks & tx_rx_locks; 1237 if (sk->sk_userlocks & SOCK_SNDBUF_LOCK) 1238 WRITE_ONCE(ssk->sk_sndbuf, sk->sk_sndbuf); 1239 if (sk->sk_userlocks & SOCK_RCVBUF_LOCK) 1240 WRITE_ONCE(ssk->sk_rcvbuf, sk->sk_rcvbuf); 1241 } 1242 1243 if (sock_flag(sk, SOCK_LINGER)) { 1244 ssk->sk_lingertime = sk->sk_lingertime; 1245 sock_set_flag(ssk, SOCK_LINGER); 1246 } else { 1247 sock_reset_flag(ssk, SOCK_LINGER); 1248 } 1249 1250 if (sk->sk_mark != ssk->sk_mark) { 1251 ssk->sk_mark = sk->sk_mark; 1252 sk_dst_reset(ssk); 1253 } 1254 1255 sock_valbool_flag(ssk, SOCK_DBG, sock_flag(sk, SOCK_DBG)); 1256 1257 if (inet_csk(sk)->icsk_ca_ops != inet_csk(ssk)->icsk_ca_ops) 1258 tcp_set_congestion_control(ssk, msk->ca_name, false, true); 1259 __tcp_sock_set_cork(ssk, !!msk->cork); 1260 __tcp_sock_set_nodelay(ssk, !!msk->nodelay); 1261 1262 inet_sk(ssk)->transparent = inet_sk(sk)->transparent; 1263 inet_sk(ssk)->freebind = inet_sk(sk)->freebind; 1264 } 1265 1266 static void __mptcp_sockopt_sync(struct mptcp_sock *msk, struct sock *ssk) 1267 { 1268 bool slow = lock_sock_fast(ssk); 1269 1270 sync_socket_options(msk, ssk); 1271 1272 unlock_sock_fast(ssk, slow); 1273 } 1274 1275 void mptcp_sockopt_sync(struct mptcp_sock *msk, struct sock *ssk) 1276 { 1277 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk); 1278 1279 msk_owned_by_me(msk); 1280 1281 if (READ_ONCE(subflow->setsockopt_seq) != msk->setsockopt_seq) { 1282 __mptcp_sockopt_sync(msk, ssk); 1283 1284 subflow->setsockopt_seq = msk->setsockopt_seq; 1285 } 1286 } 1287 1288 void mptcp_sockopt_sync_locked(struct mptcp_sock *msk, struct sock *ssk) 1289 { 1290 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk); 1291 1292 msk_owned_by_me(msk); 1293 1294 if (READ_ONCE(subflow->setsockopt_seq) != msk->setsockopt_seq) { 1295 sync_socket_options(msk, ssk); 1296 1297 subflow->setsockopt_seq = msk->setsockopt_seq; 1298 } 1299 } 1300