1 // SPDX-License-Identifier: GPL-2.0 2 #include <net/tcp.h> 3 #include <net/strparser.h> 4 #include <net/xfrm.h> 5 #include <net/esp.h> 6 #include <net/espintcp.h> 7 #include <linux/skmsg.h> 8 #include <net/inet_common.h> 9 #if IS_ENABLED(CONFIG_IPV6) 10 #include <net/ipv6_stubs.h> 11 #endif 12 13 static void handle_nonesp(struct espintcp_ctx *ctx, struct sk_buff *skb, 14 struct sock *sk) 15 { 16 if (atomic_read(&sk->sk_rmem_alloc) >= sk->sk_rcvbuf || 17 !sk_rmem_schedule(sk, skb, skb->truesize)) { 18 XFRM_INC_STATS(sock_net(sk), LINUX_MIB_XFRMINERROR); 19 kfree_skb(skb); 20 return; 21 } 22 23 skb_set_owner_r(skb, sk); 24 25 memset(skb->cb, 0, sizeof(skb->cb)); 26 skb_queue_tail(&ctx->ike_queue, skb); 27 ctx->saved_data_ready(sk); 28 } 29 30 static void handle_esp(struct sk_buff *skb, struct sock *sk) 31 { 32 struct tcp_skb_cb *tcp_cb = (struct tcp_skb_cb *)skb->cb; 33 34 skb_reset_transport_header(skb); 35 36 /* restore IP CB, we need at least IP6CB->nhoff */ 37 memmove(skb->cb, &tcp_cb->header, sizeof(tcp_cb->header)); 38 39 rcu_read_lock(); 40 skb->dev = dev_get_by_index_rcu(sock_net(sk), skb->skb_iif); 41 local_bh_disable(); 42 #if IS_ENABLED(CONFIG_IPV6) 43 if (sk->sk_family == AF_INET6) 44 ipv6_stub->xfrm6_rcv_encap(skb, IPPROTO_ESP, 0, TCP_ENCAP_ESPINTCP); 45 else 46 #endif 47 xfrm4_rcv_encap(skb, IPPROTO_ESP, 0, TCP_ENCAP_ESPINTCP); 48 local_bh_enable(); 49 rcu_read_unlock(); 50 } 51 52 static void espintcp_rcv(struct strparser *strp, struct sk_buff *skb) 53 { 54 struct espintcp_ctx *ctx = container_of(strp, struct espintcp_ctx, 55 strp); 56 struct strp_msg *rxm = strp_msg(skb); 57 int len = rxm->full_len - 2; 58 u32 nonesp_marker; 59 int err; 60 61 /* keepalive packet? */ 62 if (unlikely(len == 1)) { 63 u8 data; 64 65 err = skb_copy_bits(skb, rxm->offset + 2, &data, 1); 66 if (err < 0) { 67 XFRM_INC_STATS(sock_net(strp->sk), LINUX_MIB_XFRMINHDRERROR); 68 kfree_skb(skb); 69 return; 70 } 71 72 if (data == 0xff) { 73 kfree_skb(skb); 74 return; 75 } 76 } 77 78 /* drop other short messages */ 79 if (unlikely(len <= sizeof(nonesp_marker))) { 80 XFRM_INC_STATS(sock_net(strp->sk), LINUX_MIB_XFRMINHDRERROR); 81 kfree_skb(skb); 82 return; 83 } 84 85 err = skb_copy_bits(skb, rxm->offset + 2, &nonesp_marker, 86 sizeof(nonesp_marker)); 87 if (err < 0) { 88 XFRM_INC_STATS(sock_net(strp->sk), LINUX_MIB_XFRMINHDRERROR); 89 kfree_skb(skb); 90 return; 91 } 92 93 /* remove header, leave non-ESP marker/SPI */ 94 if (!pskb_pull(skb, rxm->offset + 2)) { 95 XFRM_INC_STATS(sock_net(strp->sk), LINUX_MIB_XFRMINERROR); 96 kfree_skb(skb); 97 return; 98 } 99 100 if (pskb_trim(skb, rxm->full_len - 2) != 0) { 101 XFRM_INC_STATS(sock_net(strp->sk), LINUX_MIB_XFRMINERROR); 102 kfree_skb(skb); 103 return; 104 } 105 106 if (nonesp_marker == 0) 107 handle_nonesp(ctx, skb, strp->sk); 108 else 109 handle_esp(skb, strp->sk); 110 } 111 112 static int espintcp_parse(struct strparser *strp, struct sk_buff *skb) 113 { 114 struct strp_msg *rxm = strp_msg(skb); 115 __be16 blen; 116 u16 len; 117 int err; 118 119 if (skb->len < rxm->offset + 2) 120 return 0; 121 122 err = skb_copy_bits(skb, rxm->offset, &blen, sizeof(blen)); 123 if (err < 0) 124 return err; 125 126 len = be16_to_cpu(blen); 127 if (len < 2) 128 return -EINVAL; 129 130 return len; 131 } 132 133 static int espintcp_recvmsg(struct sock *sk, struct msghdr *msg, size_t len, 134 int flags, int *addr_len) 135 { 136 struct espintcp_ctx *ctx = espintcp_getctx(sk); 137 struct sk_buff *skb; 138 int err = 0; 139 int copied; 140 int off = 0; 141 142 skb = __skb_recv_datagram(sk, &ctx->ike_queue, flags, &off, &err); 143 if (!skb) { 144 if (err == -EAGAIN && sk->sk_shutdown & RCV_SHUTDOWN) 145 return 0; 146 return err; 147 } 148 149 copied = len; 150 if (copied > skb->len) 151 copied = skb->len; 152 else if (copied < skb->len) 153 msg->msg_flags |= MSG_TRUNC; 154 155 err = skb_copy_datagram_msg(skb, 0, msg, copied); 156 if (unlikely(err)) { 157 kfree_skb(skb); 158 return err; 159 } 160 161 if (flags & MSG_TRUNC) 162 copied = skb->len; 163 kfree_skb(skb); 164 return copied; 165 } 166 167 int espintcp_queue_out(struct sock *sk, struct sk_buff *skb) 168 { 169 struct espintcp_ctx *ctx = espintcp_getctx(sk); 170 171 if (skb_queue_len(&ctx->out_queue) >= READ_ONCE(netdev_max_backlog)) 172 return -ENOBUFS; 173 174 __skb_queue_tail(&ctx->out_queue, skb); 175 176 return 0; 177 } 178 EXPORT_SYMBOL_GPL(espintcp_queue_out); 179 180 /* espintcp length field is 2B and length includes the length field's size */ 181 #define MAX_ESPINTCP_MSG (((1 << 16) - 1) - 2) 182 183 static int espintcp_sendskb_locked(struct sock *sk, struct espintcp_msg *emsg, 184 int flags) 185 { 186 do { 187 int ret; 188 189 ret = skb_send_sock_locked(sk, emsg->skb, 190 emsg->offset, emsg->len); 191 if (ret < 0) 192 return ret; 193 194 emsg->len -= ret; 195 emsg->offset += ret; 196 } while (emsg->len > 0); 197 198 kfree_skb(emsg->skb); 199 memset(emsg, 0, sizeof(*emsg)); 200 201 return 0; 202 } 203 204 static int espintcp_sendskmsg_locked(struct sock *sk, 205 struct espintcp_msg *emsg, int flags) 206 { 207 struct sk_msg *skmsg = &emsg->skmsg; 208 struct scatterlist *sg; 209 int done = 0; 210 int ret; 211 212 flags |= MSG_SENDPAGE_NOTLAST; 213 sg = &skmsg->sg.data[skmsg->sg.start]; 214 do { 215 size_t size = sg->length - emsg->offset; 216 int offset = sg->offset + emsg->offset; 217 struct page *p; 218 219 emsg->offset = 0; 220 221 if (sg_is_last(sg)) 222 flags &= ~MSG_SENDPAGE_NOTLAST; 223 224 p = sg_page(sg); 225 retry: 226 ret = do_tcp_sendpages(sk, p, offset, size, flags); 227 if (ret < 0) { 228 emsg->offset = offset - sg->offset; 229 skmsg->sg.start += done; 230 return ret; 231 } 232 233 if (ret != size) { 234 offset += ret; 235 size -= ret; 236 goto retry; 237 } 238 239 done++; 240 put_page(p); 241 sk_mem_uncharge(sk, sg->length); 242 sg = sg_next(sg); 243 } while (sg); 244 245 memset(emsg, 0, sizeof(*emsg)); 246 247 return 0; 248 } 249 250 static int espintcp_push_msgs(struct sock *sk, int flags) 251 { 252 struct espintcp_ctx *ctx = espintcp_getctx(sk); 253 struct espintcp_msg *emsg = &ctx->partial; 254 int err; 255 256 if (!emsg->len) 257 return 0; 258 259 if (ctx->tx_running) 260 return -EAGAIN; 261 ctx->tx_running = 1; 262 263 if (emsg->skb) 264 err = espintcp_sendskb_locked(sk, emsg, flags); 265 else 266 err = espintcp_sendskmsg_locked(sk, emsg, flags); 267 if (err == -EAGAIN) { 268 ctx->tx_running = 0; 269 return flags & MSG_DONTWAIT ? -EAGAIN : 0; 270 } 271 if (!err) 272 memset(emsg, 0, sizeof(*emsg)); 273 274 ctx->tx_running = 0; 275 276 return err; 277 } 278 279 int espintcp_push_skb(struct sock *sk, struct sk_buff *skb) 280 { 281 struct espintcp_ctx *ctx = espintcp_getctx(sk); 282 struct espintcp_msg *emsg = &ctx->partial; 283 unsigned int len; 284 int offset; 285 286 if (sk->sk_state != TCP_ESTABLISHED) { 287 kfree_skb(skb); 288 return -ECONNRESET; 289 } 290 291 offset = skb_transport_offset(skb); 292 len = skb->len - offset; 293 294 espintcp_push_msgs(sk, 0); 295 296 if (emsg->len) { 297 kfree_skb(skb); 298 return -ENOBUFS; 299 } 300 301 skb_set_owner_w(skb, sk); 302 303 emsg->offset = offset; 304 emsg->len = len; 305 emsg->skb = skb; 306 307 espintcp_push_msgs(sk, 0); 308 309 return 0; 310 } 311 EXPORT_SYMBOL_GPL(espintcp_push_skb); 312 313 static int espintcp_sendmsg(struct sock *sk, struct msghdr *msg, size_t size) 314 { 315 long timeo = sock_sndtimeo(sk, msg->msg_flags & MSG_DONTWAIT); 316 struct espintcp_ctx *ctx = espintcp_getctx(sk); 317 struct espintcp_msg *emsg = &ctx->partial; 318 struct iov_iter pfx_iter; 319 struct kvec pfx_iov = {}; 320 size_t msglen = size + 2; 321 char buf[2] = {0}; 322 int err, end; 323 324 if (msg->msg_flags & ~MSG_DONTWAIT) 325 return -EOPNOTSUPP; 326 327 if (size > MAX_ESPINTCP_MSG) 328 return -EMSGSIZE; 329 330 if (msg->msg_controllen) 331 return -EOPNOTSUPP; 332 333 lock_sock(sk); 334 335 err = espintcp_push_msgs(sk, msg->msg_flags & MSG_DONTWAIT); 336 if (err < 0) { 337 if (err != -EAGAIN || !(msg->msg_flags & MSG_DONTWAIT)) 338 err = -ENOBUFS; 339 goto unlock; 340 } 341 342 sk_msg_init(&emsg->skmsg); 343 while (1) { 344 /* only -ENOMEM is possible since we don't coalesce */ 345 err = sk_msg_alloc(sk, &emsg->skmsg, msglen, 0); 346 if (!err) 347 break; 348 349 err = sk_stream_wait_memory(sk, &timeo); 350 if (err) 351 goto fail; 352 } 353 354 *((__be16 *)buf) = cpu_to_be16(msglen); 355 pfx_iov.iov_base = buf; 356 pfx_iov.iov_len = sizeof(buf); 357 iov_iter_kvec(&pfx_iter, ITER_SOURCE, &pfx_iov, 1, pfx_iov.iov_len); 358 359 err = sk_msg_memcopy_from_iter(sk, &pfx_iter, &emsg->skmsg, 360 pfx_iov.iov_len); 361 if (err < 0) 362 goto fail; 363 364 err = sk_msg_memcopy_from_iter(sk, &msg->msg_iter, &emsg->skmsg, size); 365 if (err < 0) 366 goto fail; 367 368 end = emsg->skmsg.sg.end; 369 emsg->len = size; 370 sk_msg_iter_var_prev(end); 371 sg_mark_end(sk_msg_elem(&emsg->skmsg, end)); 372 373 tcp_rate_check_app_limited(sk); 374 375 err = espintcp_push_msgs(sk, msg->msg_flags & MSG_DONTWAIT); 376 /* this message could be partially sent, keep it */ 377 378 release_sock(sk); 379 380 return size; 381 382 fail: 383 sk_msg_free(sk, &emsg->skmsg); 384 memset(emsg, 0, sizeof(*emsg)); 385 unlock: 386 release_sock(sk); 387 return err; 388 } 389 390 static struct proto espintcp_prot __ro_after_init; 391 static struct proto_ops espintcp_ops __ro_after_init; 392 static struct proto espintcp6_prot; 393 static struct proto_ops espintcp6_ops; 394 static DEFINE_MUTEX(tcpv6_prot_mutex); 395 396 static void espintcp_data_ready(struct sock *sk) 397 { 398 struct espintcp_ctx *ctx = espintcp_getctx(sk); 399 400 strp_data_ready(&ctx->strp); 401 } 402 403 static void espintcp_tx_work(struct work_struct *work) 404 { 405 struct espintcp_ctx *ctx = container_of(work, 406 struct espintcp_ctx, work); 407 struct sock *sk = ctx->strp.sk; 408 409 lock_sock(sk); 410 if (!ctx->tx_running) 411 espintcp_push_msgs(sk, 0); 412 release_sock(sk); 413 } 414 415 static void espintcp_write_space(struct sock *sk) 416 { 417 struct espintcp_ctx *ctx = espintcp_getctx(sk); 418 419 schedule_work(&ctx->work); 420 ctx->saved_write_space(sk); 421 } 422 423 static void espintcp_destruct(struct sock *sk) 424 { 425 struct espintcp_ctx *ctx = espintcp_getctx(sk); 426 427 ctx->saved_destruct(sk); 428 kfree(ctx); 429 } 430 431 bool tcp_is_ulp_esp(struct sock *sk) 432 { 433 return sk->sk_prot == &espintcp_prot || sk->sk_prot == &espintcp6_prot; 434 } 435 EXPORT_SYMBOL_GPL(tcp_is_ulp_esp); 436 437 static void build_protos(struct proto *espintcp_prot, 438 struct proto_ops *espintcp_ops, 439 const struct proto *orig_prot, 440 const struct proto_ops *orig_ops); 441 static int espintcp_init_sk(struct sock *sk) 442 { 443 struct inet_connection_sock *icsk = inet_csk(sk); 444 struct strp_callbacks cb = { 445 .rcv_msg = espintcp_rcv, 446 .parse_msg = espintcp_parse, 447 }; 448 struct espintcp_ctx *ctx; 449 int err; 450 451 /* sockmap is not compatible with espintcp */ 452 if (sk->sk_user_data) 453 return -EBUSY; 454 455 ctx = kzalloc(sizeof(*ctx), GFP_KERNEL); 456 if (!ctx) 457 return -ENOMEM; 458 459 err = strp_init(&ctx->strp, sk, &cb); 460 if (err) 461 goto free; 462 463 __sk_dst_reset(sk); 464 465 strp_check_rcv(&ctx->strp); 466 skb_queue_head_init(&ctx->ike_queue); 467 skb_queue_head_init(&ctx->out_queue); 468 469 if (sk->sk_family == AF_INET) { 470 sk->sk_prot = &espintcp_prot; 471 sk->sk_socket->ops = &espintcp_ops; 472 } else { 473 mutex_lock(&tcpv6_prot_mutex); 474 if (!espintcp6_prot.recvmsg) 475 build_protos(&espintcp6_prot, &espintcp6_ops, sk->sk_prot, sk->sk_socket->ops); 476 mutex_unlock(&tcpv6_prot_mutex); 477 478 sk->sk_prot = &espintcp6_prot; 479 sk->sk_socket->ops = &espintcp6_ops; 480 } 481 ctx->saved_data_ready = sk->sk_data_ready; 482 ctx->saved_write_space = sk->sk_write_space; 483 ctx->saved_destruct = sk->sk_destruct; 484 sk->sk_data_ready = espintcp_data_ready; 485 sk->sk_write_space = espintcp_write_space; 486 sk->sk_destruct = espintcp_destruct; 487 rcu_assign_pointer(icsk->icsk_ulp_data, ctx); 488 INIT_WORK(&ctx->work, espintcp_tx_work); 489 490 /* avoid using task_frag */ 491 sk->sk_allocation = GFP_ATOMIC; 492 sk->sk_use_task_frag = false; 493 494 return 0; 495 496 free: 497 kfree(ctx); 498 return err; 499 } 500 501 static void espintcp_release(struct sock *sk) 502 { 503 struct espintcp_ctx *ctx = espintcp_getctx(sk); 504 struct sk_buff_head queue; 505 struct sk_buff *skb; 506 507 __skb_queue_head_init(&queue); 508 skb_queue_splice_init(&ctx->out_queue, &queue); 509 510 while ((skb = __skb_dequeue(&queue))) 511 espintcp_push_skb(sk, skb); 512 513 tcp_release_cb(sk); 514 } 515 516 static void espintcp_close(struct sock *sk, long timeout) 517 { 518 struct espintcp_ctx *ctx = espintcp_getctx(sk); 519 struct espintcp_msg *emsg = &ctx->partial; 520 521 strp_stop(&ctx->strp); 522 523 sk->sk_prot = &tcp_prot; 524 barrier(); 525 526 cancel_work_sync(&ctx->work); 527 strp_done(&ctx->strp); 528 529 skb_queue_purge(&ctx->out_queue); 530 skb_queue_purge(&ctx->ike_queue); 531 532 if (emsg->len) { 533 if (emsg->skb) 534 kfree_skb(emsg->skb); 535 else 536 sk_msg_free(sk, &emsg->skmsg); 537 } 538 539 tcp_close(sk, timeout); 540 } 541 542 static __poll_t espintcp_poll(struct file *file, struct socket *sock, 543 poll_table *wait) 544 { 545 __poll_t mask = datagram_poll(file, sock, wait); 546 struct sock *sk = sock->sk; 547 struct espintcp_ctx *ctx = espintcp_getctx(sk); 548 549 if (!skb_queue_empty(&ctx->ike_queue)) 550 mask |= EPOLLIN | EPOLLRDNORM; 551 552 return mask; 553 } 554 555 static void build_protos(struct proto *espintcp_prot, 556 struct proto_ops *espintcp_ops, 557 const struct proto *orig_prot, 558 const struct proto_ops *orig_ops) 559 { 560 memcpy(espintcp_prot, orig_prot, sizeof(struct proto)); 561 memcpy(espintcp_ops, orig_ops, sizeof(struct proto_ops)); 562 espintcp_prot->sendmsg = espintcp_sendmsg; 563 espintcp_prot->recvmsg = espintcp_recvmsg; 564 espintcp_prot->close = espintcp_close; 565 espintcp_prot->release_cb = espintcp_release; 566 espintcp_ops->poll = espintcp_poll; 567 } 568 569 static struct tcp_ulp_ops espintcp_ulp __read_mostly = { 570 .name = "espintcp", 571 .owner = THIS_MODULE, 572 .init = espintcp_init_sk, 573 }; 574 575 void __init espintcp_init(void) 576 { 577 build_protos(&espintcp_prot, &espintcp_ops, &tcp_prot, &inet_stream_ops); 578 579 tcp_register_ulp(&espintcp_ulp); 580 } 581