1 // SPDX-License-Identifier: GPL-2.0 2 /* Copyright (c) 2017 - 2018 Covalent IO, Inc. http://covalent.io */ 3 4 #include <linux/skmsg.h> 5 #include <linux/filter.h> 6 #include <linux/bpf.h> 7 #include <linux/init.h> 8 #include <linux/wait.h> 9 #include <linux/util_macros.h> 10 11 #include <net/inet_common.h> 12 #include <net/tls.h> 13 #include <asm/ioctls.h> 14 15 void tcp_eat_skb(struct sock *sk, struct sk_buff *skb) 16 { 17 struct tcp_sock *tcp; 18 int copied; 19 20 if (!skb || !skb->len || !sk_is_tcp(sk)) 21 return; 22 23 if (skb_bpf_strparser(skb)) 24 return; 25 26 tcp = tcp_sk(sk); 27 copied = tcp->copied_seq + skb->len; 28 WRITE_ONCE(tcp->copied_seq, copied); 29 tcp_rcv_space_adjust(sk); 30 __tcp_cleanup_rbuf(sk, skb->len); 31 } 32 33 static int bpf_tcp_ingress(struct sock *sk, struct sk_psock *psock, 34 struct sk_msg *msg, u32 apply_bytes) 35 { 36 bool apply = apply_bytes; 37 struct scatterlist *sge; 38 u32 size, copied = 0; 39 struct sk_msg *tmp; 40 int i, ret = 0; 41 42 tmp = kzalloc_obj(*tmp, __GFP_NOWARN | GFP_KERNEL); 43 if (unlikely(!tmp)) 44 return -ENOMEM; 45 46 lock_sock(sk); 47 tmp->sg.start = msg->sg.start; 48 i = msg->sg.start; 49 do { 50 sge = sk_msg_elem(msg, i); 51 size = (apply && apply_bytes < sge->length) ? 52 apply_bytes : sge->length; 53 if (!__sk_rmem_schedule(sk, size, false)) { 54 if (!copied) 55 ret = -ENOMEM; 56 break; 57 } 58 59 sk_mem_charge(sk, size); 60 atomic_add(size, &sk->sk_rmem_alloc); 61 sk_msg_xfer(tmp, msg, i, size); 62 copied += size; 63 if (sge->length) 64 get_page(sk_msg_page(tmp, i)); 65 sk_msg_iter_var_next(i); 66 tmp->sg.end = i; 67 if (apply) { 68 apply_bytes -= size; 69 if (!apply_bytes) { 70 if (sge->length) 71 sk_msg_iter_var_prev(i); 72 break; 73 } 74 } 75 } while (i != msg->sg.end); 76 77 if (!ret) { 78 msg->sg.start = i; 79 if (!sk_psock_queue_msg(psock, tmp)) 80 atomic_sub(copied, &sk->sk_rmem_alloc); 81 sk_psock_data_ready(sk, psock); 82 } else { 83 sk_msg_free(sk, tmp); 84 kfree(tmp); 85 } 86 87 release_sock(sk); 88 return ret; 89 } 90 91 static int tcp_bpf_push(struct sock *sk, struct sk_msg *msg, u32 apply_bytes, 92 int flags, bool uncharge) 93 { 94 struct msghdr msghdr = {}; 95 bool apply = apply_bytes; 96 struct scatterlist *sge; 97 struct page *page; 98 int size, ret = 0; 99 u32 off; 100 101 while (1) { 102 struct bio_vec bvec; 103 bool has_tx_ulp; 104 105 sge = sk_msg_elem(msg, msg->sg.start); 106 size = (apply && apply_bytes < sge->length) ? 107 apply_bytes : sge->length; 108 off = sge->offset; 109 page = sg_page(sge); 110 111 tcp_rate_check_app_limited(sk); 112 retry: 113 msghdr.msg_flags = flags | MSG_SPLICE_PAGES; 114 has_tx_ulp = tls_sw_has_ctx_tx(sk); 115 if (has_tx_ulp) 116 msghdr.msg_flags |= MSG_SENDPAGE_NOPOLICY; 117 118 if (size < sge->length && msg->sg.start != msg->sg.end) 119 msghdr.msg_flags |= MSG_MORE; 120 121 bvec_set_page(&bvec, page, size, off); 122 iov_iter_bvec(&msghdr.msg_iter, ITER_SOURCE, &bvec, 1, size); 123 ret = tcp_sendmsg_locked(sk, &msghdr, size); 124 if (ret <= 0) 125 return ret; 126 127 if (apply) 128 apply_bytes -= ret; 129 msg->sg.size -= ret; 130 sge->offset += ret; 131 sge->length -= ret; 132 if (uncharge) 133 sk_mem_uncharge(sk, ret); 134 if (ret != size) { 135 size -= ret; 136 off += ret; 137 goto retry; 138 } 139 if (!sge->length) { 140 put_page(page); 141 sk_msg_iter_next(msg, start); 142 sg_init_table(sge, 1); 143 if (msg->sg.start == msg->sg.end) 144 break; 145 } 146 if (apply && !apply_bytes) 147 break; 148 } 149 150 return 0; 151 } 152 153 static int tcp_bpf_push_locked(struct sock *sk, struct sk_msg *msg, 154 u32 apply_bytes, int flags, bool uncharge) 155 { 156 int ret; 157 158 lock_sock(sk); 159 ret = tcp_bpf_push(sk, msg, apply_bytes, flags, uncharge); 160 release_sock(sk); 161 return ret; 162 } 163 164 int tcp_bpf_sendmsg_redir(struct sock *sk, bool ingress, 165 struct sk_msg *msg, u32 bytes, int flags) 166 { 167 struct sk_psock *psock = sk_psock_get(sk); 168 int ret; 169 170 if (unlikely(!psock)) 171 return -EPIPE; 172 173 ret = ingress ? bpf_tcp_ingress(sk, psock, msg, bytes) : 174 tcp_bpf_push_locked(sk, msg, bytes, flags, false); 175 sk_psock_put(sk, psock); 176 return ret; 177 } 178 EXPORT_SYMBOL_GPL(tcp_bpf_sendmsg_redir); 179 180 #ifdef CONFIG_BPF_SYSCALL 181 static int tcp_msg_wait_data(struct sock *sk, struct sk_psock *psock, 182 long timeo) 183 { 184 DEFINE_WAIT_FUNC(wait, woken_wake_function); 185 int ret = 0; 186 187 if (sk->sk_shutdown & RCV_SHUTDOWN) 188 return 1; 189 190 if (!timeo) 191 return ret; 192 193 add_wait_queue(sk_sleep(sk), &wait); 194 sk_set_bit(SOCKWQ_ASYNC_WAITDATA, sk); 195 ret = sk_wait_event(sk, &timeo, 196 !list_empty(&psock->ingress_msg) || 197 !skb_queue_empty_lockless(&sk->sk_receive_queue), &wait); 198 sk_clear_bit(SOCKWQ_ASYNC_WAITDATA, sk); 199 remove_wait_queue(sk_sleep(sk), &wait); 200 return ret; 201 } 202 203 static bool is_next_msg_fin(struct sk_psock *psock) 204 { 205 struct scatterlist *sge; 206 struct sk_msg *msg_rx; 207 int i; 208 209 msg_rx = sk_psock_peek_msg(psock); 210 i = msg_rx->sg.start; 211 sge = sk_msg_elem(msg_rx, i); 212 if (!sge->length) { 213 struct sk_buff *skb = msg_rx->skb; 214 215 if (skb && TCP_SKB_CB(skb)->tcp_flags & TCPHDR_FIN) 216 return true; 217 } 218 return false; 219 } 220 221 static int tcp_bpf_recvmsg_parser(struct sock *sk, 222 struct msghdr *msg, 223 size_t len, 224 int flags) 225 { 226 int peek = flags & MSG_PEEK; 227 struct sk_psock *psock; 228 struct tcp_sock *tcp; 229 int copied_from_self = 0; 230 int copied = 0; 231 u32 seq; 232 233 if (unlikely(flags & MSG_ERRQUEUE)) 234 return inet_recv_error(sk, msg, len); 235 236 if (!len) 237 return 0; 238 239 psock = sk_psock_get(sk); 240 if (unlikely(!psock)) 241 return tcp_recvmsg(sk, msg, len, flags); 242 243 lock_sock(sk); 244 tcp = tcp_sk(sk); 245 seq = tcp->copied_seq; 246 /* We may have received data on the sk_receive_queue pre-accept and 247 * then we can not use read_skb in this context because we haven't 248 * assigned a sk_socket yet so have no link to the ops. The work-around 249 * is to check the sk_receive_queue and in these cases read skbs off 250 * queue again. The read_skb hook is not running at this point because 251 * of lock_sock so we avoid having multiple runners in read_skb. 252 */ 253 if (unlikely(!skb_queue_empty(&sk->sk_receive_queue))) { 254 tcp_data_ready(sk); 255 /* This handles the ENOMEM errors if we both receive data 256 * pre accept and are already under memory pressure. At least 257 * let user know to retry. 258 */ 259 if (unlikely(!skb_queue_empty(&sk->sk_receive_queue))) { 260 copied = -EAGAIN; 261 goto out; 262 } 263 } 264 265 msg_bytes_ready: 266 copied = __sk_msg_recvmsg(sk, psock, msg, len, flags, &copied_from_self); 267 /* The typical case for EFAULT is the socket was gracefully 268 * shutdown with a FIN pkt. So check here the other case is 269 * some error on copy_page_to_iter which would be unexpected. 270 * On fin return correct return code to zero. 271 */ 272 if (copied == -EFAULT) { 273 bool is_fin = is_next_msg_fin(psock); 274 275 if (is_fin) { 276 copied = 0; 277 seq++; 278 goto out; 279 } 280 } 281 seq += copied_from_self; 282 if (!copied) { 283 long timeo; 284 int data; 285 286 if (sock_flag(sk, SOCK_DONE)) 287 goto out; 288 289 if (sk->sk_err) { 290 copied = sock_error(sk); 291 goto out; 292 } 293 294 if (sk->sk_shutdown & RCV_SHUTDOWN) 295 goto out; 296 297 if (sk->sk_state == TCP_CLOSE) { 298 copied = -ENOTCONN; 299 goto out; 300 } 301 302 timeo = sock_rcvtimeo(sk, flags & MSG_DONTWAIT); 303 if (!timeo) { 304 copied = -EAGAIN; 305 goto out; 306 } 307 308 if (signal_pending(current)) { 309 copied = sock_intr_errno(timeo); 310 goto out; 311 } 312 313 data = tcp_msg_wait_data(sk, psock, timeo); 314 if (data < 0) { 315 copied = data; 316 goto unlock; 317 } 318 if (data && !sk_psock_queue_empty(psock)) 319 goto msg_bytes_ready; 320 copied = -EAGAIN; 321 } 322 out: 323 if (!peek) 324 WRITE_ONCE(tcp->copied_seq, seq); 325 tcp_rcv_space_adjust(sk); 326 if (copied > 0) 327 __tcp_cleanup_rbuf(sk, copied); 328 329 unlock: 330 release_sock(sk); 331 sk_psock_put(sk, psock); 332 return copied; 333 } 334 335 static int tcp_bpf_ioctl(struct sock *sk, int cmd, int *karg) 336 { 337 bool slow; 338 339 if (cmd != SIOCINQ) 340 return tcp_ioctl(sk, cmd, karg); 341 342 /* works similar as tcp_ioctl */ 343 if (sk->sk_state == TCP_LISTEN) 344 return -EINVAL; 345 346 slow = lock_sock_fast(sk); 347 *karg = sk_psock_msg_inq(sk); 348 unlock_sock_fast(sk, slow); 349 350 return 0; 351 } 352 353 static int tcp_bpf_recvmsg(struct sock *sk, struct msghdr *msg, size_t len, 354 int flags) 355 { 356 struct sk_psock *psock; 357 int copied, ret; 358 359 if (unlikely(flags & MSG_ERRQUEUE)) 360 return inet_recv_error(sk, msg, len); 361 362 if (!len) 363 return 0; 364 365 psock = sk_psock_get(sk); 366 if (unlikely(!psock)) 367 return tcp_recvmsg(sk, msg, len, flags); 368 if (!skb_queue_empty(&sk->sk_receive_queue) && 369 sk_psock_queue_empty(psock)) { 370 sk_psock_put(sk, psock); 371 return tcp_recvmsg(sk, msg, len, flags); 372 } 373 lock_sock(sk); 374 msg_bytes_ready: 375 copied = sk_msg_recvmsg(sk, psock, msg, len, flags); 376 if (!copied) { 377 long timeo; 378 int data; 379 380 timeo = sock_rcvtimeo(sk, flags & MSG_DONTWAIT); 381 data = tcp_msg_wait_data(sk, psock, timeo); 382 if (data < 0) { 383 ret = data; 384 goto unlock; 385 } 386 if (data) { 387 if (!sk_psock_queue_empty(psock)) 388 goto msg_bytes_ready; 389 release_sock(sk); 390 sk_psock_put(sk, psock); 391 return tcp_recvmsg(sk, msg, len, flags); 392 } 393 copied = -EAGAIN; 394 } 395 ret = copied; 396 397 unlock: 398 release_sock(sk); 399 sk_psock_put(sk, psock); 400 return ret; 401 } 402 403 static int tcp_bpf_send_verdict(struct sock *sk, struct sk_psock *psock, 404 struct sk_msg *msg, int *copied, int flags) 405 { 406 bool cork = false, enospc = sk_msg_full(msg), redir_ingress; 407 struct sock *sk_redir; 408 u32 tosend, origsize, sent, delta = 0; 409 u32 eval; 410 int ret; 411 412 more_data: 413 if (psock->eval == __SK_NONE) { 414 /* Track delta in msg size to add/subtract it on SK_DROP from 415 * returned to user copied size. This ensures user doesn't 416 * get a positive return code with msg_cut_data and SK_DROP 417 * verdict. 418 */ 419 delta = msg->sg.size; 420 psock->eval = sk_psock_msg_verdict(sk, psock, msg); 421 delta -= msg->sg.size; 422 } 423 424 if (msg->cork_bytes && 425 msg->cork_bytes > msg->sg.size && !enospc) { 426 psock->cork_bytes = msg->cork_bytes - msg->sg.size; 427 if (!psock->cork) { 428 psock->cork = kzalloc_obj(*psock->cork, 429 GFP_ATOMIC | __GFP_NOWARN); 430 if (!psock->cork) { 431 sk_msg_free(sk, msg); 432 *copied = 0; 433 return -ENOMEM; 434 } 435 } 436 memcpy(psock->cork, msg, sizeof(*msg)); 437 return 0; 438 } 439 440 tosend = msg->sg.size; 441 if (psock->apply_bytes && psock->apply_bytes < tosend) 442 tosend = psock->apply_bytes; 443 eval = __SK_NONE; 444 445 switch (psock->eval) { 446 case __SK_PASS: 447 ret = tcp_bpf_push(sk, msg, tosend, flags, true); 448 if (unlikely(ret)) { 449 *copied -= sk_msg_free(sk, msg); 450 break; 451 } 452 sk_msg_apply_bytes(psock, tosend); 453 break; 454 case __SK_REDIRECT: 455 redir_ingress = psock->redir_ingress; 456 sk_redir = psock->sk_redir; 457 sk_msg_apply_bytes(psock, tosend); 458 if (!psock->apply_bytes) { 459 /* Clean up before releasing the sock lock. */ 460 eval = psock->eval; 461 psock->eval = __SK_NONE; 462 psock->sk_redir = NULL; 463 } 464 if (psock->cork) { 465 cork = true; 466 psock->cork = NULL; 467 } 468 release_sock(sk); 469 470 origsize = msg->sg.size; 471 ret = tcp_bpf_sendmsg_redir(sk_redir, redir_ingress, 472 msg, tosend, flags); 473 sent = origsize - msg->sg.size; 474 475 if (eval == __SK_REDIRECT) 476 sock_put(sk_redir); 477 478 lock_sock(sk); 479 sk_mem_uncharge(sk, sent); 480 if (unlikely(ret < 0)) { 481 int free = sk_msg_free(sk, msg); 482 483 if (!cork) 484 *copied -= free; 485 } 486 if (cork) { 487 sk_msg_free(sk, msg); 488 kfree(msg); 489 msg = NULL; 490 ret = 0; 491 } 492 break; 493 case __SK_DROP: 494 default: 495 sk_msg_free(sk, msg); 496 sk_msg_apply_bytes(psock, tosend); 497 *copied -= (tosend + delta); 498 return -EACCES; 499 } 500 501 if (likely(!ret)) { 502 if (!psock->apply_bytes) { 503 psock->eval = __SK_NONE; 504 if (psock->sk_redir) { 505 sock_put(psock->sk_redir); 506 psock->sk_redir = NULL; 507 } 508 } 509 if (msg && 510 msg->sg.data[msg->sg.start].page_link && 511 msg->sg.data[msg->sg.start].length) 512 goto more_data; 513 } 514 return ret; 515 } 516 517 static int tcp_bpf_sendmsg(struct sock *sk, struct msghdr *msg, size_t size) 518 { 519 struct sk_msg tmp, *msg_tx = NULL; 520 int copied = 0, err = 0, ret = 0; 521 struct sk_psock *psock; 522 long timeo; 523 int flags; 524 525 /* Don't let internal flags through */ 526 flags = (msg->msg_flags & ~MSG_SENDPAGE_DECRYPTED); 527 flags |= MSG_NO_SHARED_FRAGS; 528 529 psock = sk_psock_get(sk); 530 if (unlikely(!psock)) 531 return tcp_sendmsg(sk, msg, size); 532 533 lock_sock(sk); 534 timeo = sock_sndtimeo(sk, msg->msg_flags & MSG_DONTWAIT); 535 while (msg_data_left(msg)) { 536 bool enospc = false; 537 u32 copy, osize; 538 539 if (sk->sk_err) { 540 err = -sk->sk_err; 541 goto out_err; 542 } 543 544 copy = msg_data_left(msg); 545 if (!sk_stream_memory_free(sk)) 546 goto wait_for_sndbuf; 547 if (psock->cork) { 548 msg_tx = psock->cork; 549 } else { 550 msg_tx = &tmp; 551 sk_msg_init(msg_tx); 552 } 553 554 osize = msg_tx->sg.size; 555 err = sk_msg_alloc(sk, msg_tx, msg_tx->sg.size + copy, msg_tx->sg.end - 1); 556 if (err) { 557 if (err != -ENOSPC) 558 goto wait_for_memory; 559 enospc = true; 560 copy = msg_tx->sg.size - osize; 561 } 562 563 ret = sk_msg_memcopy_from_iter(sk, &msg->msg_iter, msg_tx, 564 copy); 565 if (ret < 0) { 566 sk_msg_trim(sk, msg_tx, osize); 567 goto out_err; 568 } 569 570 copied += ret; 571 if (psock->cork_bytes) { 572 if (size > psock->cork_bytes) 573 psock->cork_bytes = 0; 574 else 575 psock->cork_bytes -= size; 576 if (psock->cork_bytes && !enospc) 577 goto out_err; 578 /* All cork bytes are accounted, rerun the prog. */ 579 psock->eval = __SK_NONE; 580 psock->cork_bytes = 0; 581 } 582 583 err = tcp_bpf_send_verdict(sk, psock, msg_tx, &copied, flags); 584 if (unlikely(err < 0)) 585 goto out_err; 586 continue; 587 wait_for_sndbuf: 588 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags); 589 wait_for_memory: 590 err = sk_stream_wait_memory(sk, &timeo); 591 if (err) { 592 if (msg_tx && msg_tx != psock->cork) 593 sk_msg_free(sk, msg_tx); 594 goto out_err; 595 } 596 } 597 out_err: 598 if (err < 0) 599 err = sk_stream_error(sk, msg->msg_flags, err); 600 release_sock(sk); 601 sk_psock_put(sk, psock); 602 return copied > 0 ? copied : err; 603 } 604 605 enum { 606 TCP_BPF_IPV4, 607 TCP_BPF_IPV6, 608 TCP_BPF_NUM_PROTS, 609 }; 610 611 enum { 612 TCP_BPF_BASE, 613 TCP_BPF_TX, 614 TCP_BPF_RX, 615 TCP_BPF_TXRX, 616 TCP_BPF_NUM_CFGS, 617 }; 618 619 static struct proto *tcpv6_prot_saved __read_mostly; 620 static DEFINE_SPINLOCK(tcpv6_prot_lock); 621 static struct proto tcp_bpf_prots[TCP_BPF_NUM_PROTS][TCP_BPF_NUM_CFGS]; 622 623 static void tcp_bpf_rebuild_protos(struct proto prot[TCP_BPF_NUM_CFGS], 624 struct proto *base) 625 { 626 prot[TCP_BPF_BASE] = *base; 627 prot[TCP_BPF_BASE].destroy = sock_map_destroy; 628 prot[TCP_BPF_BASE].close = sock_map_close; 629 prot[TCP_BPF_BASE].recvmsg = tcp_bpf_recvmsg; 630 prot[TCP_BPF_BASE].sock_is_readable = sk_msg_is_readable; 631 prot[TCP_BPF_BASE].ioctl = tcp_bpf_ioctl; 632 633 prot[TCP_BPF_TX] = prot[TCP_BPF_BASE]; 634 prot[TCP_BPF_TX].sendmsg = tcp_bpf_sendmsg; 635 636 prot[TCP_BPF_RX] = prot[TCP_BPF_BASE]; 637 prot[TCP_BPF_RX].recvmsg = tcp_bpf_recvmsg_parser; 638 639 prot[TCP_BPF_TXRX] = prot[TCP_BPF_TX]; 640 prot[TCP_BPF_TXRX].recvmsg = tcp_bpf_recvmsg_parser; 641 } 642 643 static void tcp_bpf_check_v6_needs_rebuild(struct proto *ops) 644 { 645 if (unlikely(ops != smp_load_acquire(&tcpv6_prot_saved))) { 646 spin_lock_bh(&tcpv6_prot_lock); 647 if (likely(ops != tcpv6_prot_saved)) { 648 tcp_bpf_rebuild_protos(tcp_bpf_prots[TCP_BPF_IPV6], ops); 649 smp_store_release(&tcpv6_prot_saved, ops); 650 } 651 spin_unlock_bh(&tcpv6_prot_lock); 652 } 653 } 654 655 static int __init tcp_bpf_v4_build_proto(void) 656 { 657 tcp_bpf_rebuild_protos(tcp_bpf_prots[TCP_BPF_IPV4], &tcp_prot); 658 return 0; 659 } 660 late_initcall(tcp_bpf_v4_build_proto); 661 662 static int tcp_bpf_assert_proto_ops(struct proto *ops) 663 { 664 /* In order to avoid retpoline, we make assumptions when we call 665 * into ops if e.g. a psock is not present. Make sure they are 666 * indeed valid assumptions. 667 */ 668 return ops->recvmsg == tcp_recvmsg && 669 ops->sendmsg == tcp_sendmsg ? 0 : -ENOTSUPP; 670 } 671 672 #if IS_ENABLED(CONFIG_BPF_STREAM_PARSER) 673 int tcp_bpf_strp_read_sock(struct strparser *strp, read_descriptor_t *desc, 674 sk_read_actor_t recv_actor) 675 { 676 struct sock *sk = strp->sk; 677 struct sk_psock *psock; 678 struct tcp_sock *tp; 679 int copied = 0; 680 681 tp = tcp_sk(sk); 682 rcu_read_lock(); 683 psock = sk_psock(sk); 684 if (WARN_ON_ONCE(!psock)) { 685 desc->error = -EINVAL; 686 goto out; 687 } 688 689 psock->ingress_bytes = 0; 690 copied = tcp_read_sock_noack(sk, desc, recv_actor, true, 691 &psock->copied_seq); 692 if (copied < 0) 693 goto out; 694 /* recv_actor may redirect skb to another socket (SK_REDIRECT) or 695 * just put skb into ingress queue of current socket (SK_PASS). 696 * For SK_REDIRECT, we need to ack the frame immediately but for 697 * SK_PASS, we want to delay the ack until tcp_bpf_recvmsg_parser(). 698 */ 699 tp->copied_seq = psock->copied_seq - psock->ingress_bytes; 700 tcp_rcv_space_adjust(sk); 701 __tcp_cleanup_rbuf(sk, copied - psock->ingress_bytes); 702 out: 703 rcu_read_unlock(); 704 return copied; 705 } 706 #endif /* CONFIG_BPF_STREAM_PARSER */ 707 708 int tcp_bpf_update_proto(struct sock *sk, struct sk_psock *psock, bool restore) 709 { 710 int family = sk->sk_family == AF_INET6 ? TCP_BPF_IPV6 : TCP_BPF_IPV4; 711 int config = psock->progs.msg_parser ? TCP_BPF_TX : TCP_BPF_BASE; 712 713 if (psock->progs.stream_verdict || psock->progs.skb_verdict) { 714 config = (config == TCP_BPF_TX) ? TCP_BPF_TXRX : TCP_BPF_RX; 715 } 716 717 if (restore) { 718 if (inet_csk_has_ulp(sk)) { 719 /* TLS does not have an unhash proto in SW cases, 720 * but we need to ensure we stop using the sock_map 721 * unhash routine because the associated psock is being 722 * removed. So use the original unhash handler. 723 */ 724 WRITE_ONCE(sk->sk_prot->unhash, psock->saved_unhash); 725 tcp_update_ulp(sk, psock->sk_proto, psock->saved_write_space); 726 } else { 727 WRITE_ONCE(sk->sk_write_space, psock->saved_write_space); 728 /* Pairs with lockless read in sk_clone_lock() */ 729 sock_replace_proto(sk, psock->sk_proto); 730 } 731 return 0; 732 } 733 734 if (sk->sk_family == AF_INET6) { 735 if (tcp_bpf_assert_proto_ops(psock->sk_proto)) 736 return -EINVAL; 737 738 tcp_bpf_check_v6_needs_rebuild(psock->sk_proto); 739 } 740 741 /* Pairs with lockless read in sk_clone_lock() */ 742 sock_replace_proto(sk, &tcp_bpf_prots[family][config]); 743 return 0; 744 } 745 EXPORT_SYMBOL_GPL(tcp_bpf_update_proto); 746 747 /* If a child got cloned from a listening socket that had tcp_bpf 748 * protocol callbacks installed, we need to restore the callbacks to 749 * the default ones because the child does not inherit the psock state 750 * that tcp_bpf callbacks expect. 751 */ 752 void tcp_bpf_clone(const struct sock *sk, struct sock *newsk) 753 { 754 struct proto *prot = newsk->sk_prot; 755 756 if (is_insidevar(prot, tcp_bpf_prots)) 757 newsk->sk_prot = sk->sk_prot_creator; 758 } 759 #endif /* CONFIG_BPF_SYSCALL */ 760