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