1 // SPDX-License-Identifier: GPL-2.0-only 2 /* 3 * common code for virtio vsock 4 * 5 * Copyright (C) 2013-2015 Red Hat, Inc. 6 * Author: Asias He <asias@redhat.com> 7 * Stefan Hajnoczi <stefanha@redhat.com> 8 */ 9 #include <linux/spinlock.h> 10 #include <linux/module.h> 11 #include <linux/sched/signal.h> 12 #include <linux/ctype.h> 13 #include <linux/list.h> 14 #include <linux/virtio_vsock.h> 15 #include <uapi/linux/vsockmon.h> 16 17 #include <net/sock.h> 18 #include <net/af_vsock.h> 19 20 #define CREATE_TRACE_POINTS 21 #include <trace/events/vsock_virtio_transport_common.h> 22 23 /* How long to wait for graceful shutdown of a connection */ 24 #define VSOCK_CLOSE_TIMEOUT (8 * HZ) 25 26 /* Threshold for detecting small packets to copy */ 27 #define GOOD_COPY_LEN 128 28 29 static void virtio_transport_cancel_close_work(struct vsock_sock *vsk, 30 bool cancel_timeout); 31 32 static const struct virtio_transport * 33 virtio_transport_get_ops(struct vsock_sock *vsk) 34 { 35 const struct vsock_transport *t = vsock_core_get_transport(vsk); 36 37 if (WARN_ON(!t)) 38 return NULL; 39 40 return container_of(t, struct virtio_transport, transport); 41 } 42 43 static bool virtio_transport_can_zcopy(const struct virtio_transport *t_ops, 44 struct virtio_vsock_pkt_info *info, 45 size_t pkt_len) 46 { 47 struct iov_iter *iov_iter; 48 49 if (!info->msg) 50 return false; 51 52 iov_iter = &info->msg->msg_iter; 53 54 if (iov_iter->iov_offset) 55 return false; 56 57 /* We can't send whole iov. */ 58 if (iov_iter->count > pkt_len) 59 return false; 60 61 /* Check that transport can send data in zerocopy mode. */ 62 t_ops = virtio_transport_get_ops(info->vsk); 63 64 if (t_ops->can_msgzerocopy) { 65 int pages_to_send = iov_iter_npages(iov_iter, MAX_SKB_FRAGS); 66 67 /* +1 is for packet header. */ 68 return t_ops->can_msgzerocopy(pages_to_send + 1); 69 } 70 71 return true; 72 } 73 74 static int virtio_transport_init_zcopy_skb(struct vsock_sock *vsk, 75 struct sk_buff *skb, 76 struct msghdr *msg, 77 bool zerocopy) 78 { 79 struct ubuf_info *uarg; 80 81 if (msg->msg_ubuf) { 82 uarg = msg->msg_ubuf; 83 net_zcopy_get(uarg); 84 } else { 85 struct iov_iter *iter = &msg->msg_iter; 86 struct ubuf_info_msgzc *uarg_zc; 87 88 uarg = msg_zerocopy_realloc(sk_vsock(vsk), 89 iter->count, 90 NULL, false); 91 if (!uarg) 92 return -1; 93 94 uarg_zc = uarg_to_msgzc(uarg); 95 uarg_zc->zerocopy = zerocopy ? 1 : 0; 96 } 97 98 skb_zcopy_init(skb, uarg); 99 100 return 0; 101 } 102 103 static int virtio_transport_fill_skb(struct sk_buff *skb, 104 struct virtio_vsock_pkt_info *info, 105 size_t len, 106 bool zcopy) 107 { 108 struct msghdr *msg = info->msg; 109 110 if (zcopy) 111 return __zerocopy_sg_from_iter(msg, NULL, skb, 112 &msg->msg_iter, len, NULL); 113 114 virtio_vsock_skb_put(skb, len); 115 return skb_copy_datagram_from_iter_full(skb, 0, &msg->msg_iter, len); 116 } 117 118 static void virtio_transport_init_hdr(struct sk_buff *skb, 119 struct virtio_vsock_pkt_info *info, 120 size_t payload_len, 121 u32 src_cid, 122 u32 src_port, 123 u32 dst_cid, 124 u32 dst_port) 125 { 126 struct virtio_vsock_hdr *hdr; 127 128 hdr = virtio_vsock_hdr(skb); 129 hdr->type = cpu_to_le16(info->type); 130 hdr->op = cpu_to_le16(info->op); 131 hdr->src_cid = cpu_to_le64(src_cid); 132 hdr->dst_cid = cpu_to_le64(dst_cid); 133 hdr->src_port = cpu_to_le32(src_port); 134 hdr->dst_port = cpu_to_le32(dst_port); 135 hdr->flags = cpu_to_le32(info->flags); 136 hdr->len = cpu_to_le32(payload_len); 137 hdr->buf_alloc = cpu_to_le32(0); 138 hdr->fwd_cnt = cpu_to_le32(0); 139 } 140 141 static void virtio_transport_copy_nonlinear_skb(const struct sk_buff *skb, 142 void *dst, 143 size_t len) 144 { 145 struct iov_iter iov_iter = { 0 }; 146 struct kvec kvec; 147 size_t to_copy; 148 149 kvec.iov_base = dst; 150 kvec.iov_len = len; 151 152 iov_iter.iter_type = ITER_KVEC; 153 iov_iter.kvec = &kvec; 154 iov_iter.nr_segs = 1; 155 156 to_copy = min_t(size_t, len, skb->len); 157 158 skb_copy_datagram_iter(skb, VIRTIO_VSOCK_SKB_CB(skb)->offset, 159 &iov_iter, to_copy); 160 } 161 162 /* Packet capture */ 163 static struct sk_buff *virtio_transport_build_skb(void *opaque) 164 { 165 struct virtio_vsock_hdr *pkt_hdr; 166 struct sk_buff *pkt = opaque; 167 struct af_vsockmon_hdr *hdr; 168 struct sk_buff *skb; 169 size_t payload_len; 170 171 /* A packet could be split to fit the RX buffer, so we can retrieve 172 * the payload length from the header and the buffer pointer taking 173 * care of the offset in the original packet. 174 */ 175 pkt_hdr = virtio_vsock_hdr(pkt); 176 payload_len = pkt->len; 177 178 skb = alloc_skb(sizeof(*hdr) + sizeof(*pkt_hdr) + payload_len, 179 GFP_ATOMIC); 180 if (!skb) 181 return NULL; 182 183 hdr = skb_put(skb, sizeof(*hdr)); 184 185 /* pkt->hdr is little-endian so no need to byteswap here */ 186 hdr->src_cid = pkt_hdr->src_cid; 187 hdr->src_port = pkt_hdr->src_port; 188 hdr->dst_cid = pkt_hdr->dst_cid; 189 hdr->dst_port = pkt_hdr->dst_port; 190 191 hdr->transport = cpu_to_le16(AF_VSOCK_TRANSPORT_VIRTIO); 192 hdr->len = cpu_to_le16(sizeof(*pkt_hdr)); 193 memset(hdr->reserved, 0, sizeof(hdr->reserved)); 194 195 switch (le16_to_cpu(pkt_hdr->op)) { 196 case VIRTIO_VSOCK_OP_REQUEST: 197 case VIRTIO_VSOCK_OP_RESPONSE: 198 hdr->op = cpu_to_le16(AF_VSOCK_OP_CONNECT); 199 break; 200 case VIRTIO_VSOCK_OP_RST: 201 case VIRTIO_VSOCK_OP_SHUTDOWN: 202 hdr->op = cpu_to_le16(AF_VSOCK_OP_DISCONNECT); 203 break; 204 case VIRTIO_VSOCK_OP_RW: 205 hdr->op = cpu_to_le16(AF_VSOCK_OP_PAYLOAD); 206 break; 207 case VIRTIO_VSOCK_OP_CREDIT_UPDATE: 208 case VIRTIO_VSOCK_OP_CREDIT_REQUEST: 209 hdr->op = cpu_to_le16(AF_VSOCK_OP_CONTROL); 210 break; 211 default: 212 hdr->op = cpu_to_le16(AF_VSOCK_OP_UNKNOWN); 213 break; 214 } 215 216 skb_put_data(skb, pkt_hdr, sizeof(*pkt_hdr)); 217 218 if (payload_len) { 219 if (skb_is_nonlinear(pkt)) { 220 void *data = skb_put(skb, payload_len); 221 222 virtio_transport_copy_nonlinear_skb(pkt, data, payload_len); 223 } else { 224 skb_put_data(skb, pkt->data, payload_len); 225 } 226 } 227 228 return skb; 229 } 230 231 void virtio_transport_deliver_tap_pkt(struct sk_buff *skb) 232 { 233 if (virtio_vsock_skb_tap_delivered(skb)) 234 return; 235 236 vsock_deliver_tap(virtio_transport_build_skb, skb); 237 virtio_vsock_skb_set_tap_delivered(skb); 238 } 239 EXPORT_SYMBOL_GPL(virtio_transport_deliver_tap_pkt); 240 241 static u16 virtio_transport_get_type(struct sock *sk) 242 { 243 if (sk->sk_type == SOCK_STREAM) 244 return VIRTIO_VSOCK_TYPE_STREAM; 245 else 246 return VIRTIO_VSOCK_TYPE_SEQPACKET; 247 } 248 249 /* Returns new sk_buff on success, otherwise returns NULL. */ 250 static struct sk_buff *virtio_transport_alloc_skb(struct virtio_vsock_pkt_info *info, 251 size_t payload_len, 252 bool zcopy, 253 u32 src_cid, 254 u32 src_port, 255 u32 dst_cid, 256 u32 dst_port) 257 { 258 struct vsock_sock *vsk; 259 struct sk_buff *skb; 260 size_t skb_len; 261 262 skb_len = VIRTIO_VSOCK_SKB_HEADROOM; 263 264 if (!zcopy) 265 skb_len += payload_len; 266 267 skb = virtio_vsock_alloc_skb(skb_len, GFP_KERNEL); 268 if (!skb) 269 return NULL; 270 271 virtio_transport_init_hdr(skb, info, payload_len, src_cid, src_port, 272 dst_cid, dst_port); 273 274 vsk = info->vsk; 275 276 /* If 'vsk' != NULL then payload is always present, so we 277 * will never call '__zerocopy_sg_from_iter()' below without 278 * setting skb owner in 'skb_set_owner_w()'. The only case 279 * when 'vsk' == NULL is VIRTIO_VSOCK_OP_RST control message 280 * without payload. 281 */ 282 WARN_ON_ONCE(!(vsk && (info->msg && payload_len)) && zcopy); 283 284 /* Set owner here, because '__zerocopy_sg_from_iter()' uses 285 * owner of skb without check to update 'sk_wmem_alloc'. 286 */ 287 if (vsk) 288 skb_set_owner_w(skb, sk_vsock(vsk)); 289 290 if (info->msg && payload_len > 0) { 291 int err; 292 293 err = virtio_transport_fill_skb(skb, info, payload_len, zcopy); 294 if (err) 295 goto out; 296 297 if (msg_data_left(info->msg) == 0 && 298 info->type == VIRTIO_VSOCK_TYPE_SEQPACKET) { 299 struct virtio_vsock_hdr *hdr = virtio_vsock_hdr(skb); 300 301 hdr->flags |= cpu_to_le32(VIRTIO_VSOCK_SEQ_EOM); 302 303 if (info->msg->msg_flags & MSG_EOR) 304 hdr->flags |= cpu_to_le32(VIRTIO_VSOCK_SEQ_EOR); 305 } 306 } 307 308 if (info->reply) 309 virtio_vsock_skb_set_reply(skb); 310 311 trace_virtio_transport_alloc_pkt(src_cid, src_port, 312 dst_cid, dst_port, 313 payload_len, 314 info->type, 315 info->op, 316 info->flags, 317 zcopy); 318 319 return skb; 320 out: 321 kfree_skb(skb); 322 return NULL; 323 } 324 325 /* This function can only be used on connecting/connected sockets, 326 * since a socket assigned to a transport is required. 327 * 328 * Do not use on listener sockets! 329 */ 330 static int virtio_transport_send_pkt_info(struct vsock_sock *vsk, 331 struct virtio_vsock_pkt_info *info) 332 { 333 u32 max_skb_len = VIRTIO_VSOCK_MAX_PKT_BUF_SIZE; 334 u32 src_cid, src_port, dst_cid, dst_port; 335 const struct virtio_transport *t_ops; 336 struct virtio_vsock_sock *vvs; 337 u32 pkt_len = info->pkt_len; 338 bool can_zcopy = false; 339 u32 rest_len; 340 int ret; 341 342 info->type = virtio_transport_get_type(sk_vsock(vsk)); 343 344 t_ops = virtio_transport_get_ops(vsk); 345 if (unlikely(!t_ops)) 346 return -EFAULT; 347 348 src_cid = t_ops->transport.get_local_cid(); 349 src_port = vsk->local_addr.svm_port; 350 if (!info->remote_cid) { 351 dst_cid = vsk->remote_addr.svm_cid; 352 dst_port = vsk->remote_addr.svm_port; 353 } else { 354 dst_cid = info->remote_cid; 355 dst_port = info->remote_port; 356 } 357 358 vvs = vsk->trans; 359 360 /* virtio_transport_get_credit might return less than pkt_len credit */ 361 pkt_len = virtio_transport_get_credit(vvs, pkt_len); 362 363 /* Do not send zero length OP_RW pkt */ 364 if (pkt_len == 0 && info->op == VIRTIO_VSOCK_OP_RW) 365 return pkt_len; 366 367 if (info->msg) { 368 /* If zerocopy is not enabled by 'setsockopt()', we behave as 369 * there is no MSG_ZEROCOPY flag set. 370 */ 371 if (!sock_flag(sk_vsock(vsk), SOCK_ZEROCOPY)) 372 info->msg->msg_flags &= ~MSG_ZEROCOPY; 373 374 if (info->msg->msg_flags & MSG_ZEROCOPY) 375 can_zcopy = virtio_transport_can_zcopy(t_ops, info, pkt_len); 376 377 if (can_zcopy) 378 max_skb_len = min_t(u32, VIRTIO_VSOCK_MAX_PKT_BUF_SIZE, 379 (MAX_SKB_FRAGS * PAGE_SIZE)); 380 } 381 382 rest_len = pkt_len; 383 384 do { 385 struct sk_buff *skb; 386 size_t skb_len; 387 388 skb_len = min(max_skb_len, rest_len); 389 390 skb = virtio_transport_alloc_skb(info, skb_len, can_zcopy, 391 src_cid, src_port, 392 dst_cid, dst_port); 393 if (!skb) { 394 ret = -ENOMEM; 395 break; 396 } 397 398 /* We process buffer part by part, allocating skb on 399 * each iteration. If this is last skb for this buffer 400 * and MSG_ZEROCOPY mode is in use - we must allocate 401 * completion for the current syscall. 402 */ 403 if (info->msg && info->msg->msg_flags & MSG_ZEROCOPY && 404 skb_len == rest_len && info->op == VIRTIO_VSOCK_OP_RW) { 405 if (virtio_transport_init_zcopy_skb(vsk, skb, 406 info->msg, 407 can_zcopy)) { 408 kfree_skb(skb); 409 ret = -ENOMEM; 410 break; 411 } 412 } 413 414 virtio_transport_inc_tx_pkt(vvs, skb); 415 416 ret = t_ops->send_pkt(skb); 417 if (ret < 0) 418 break; 419 420 /* Both virtio and vhost 'send_pkt()' returns 'skb_len', 421 * but for reliability use 'ret' instead of 'skb_len'. 422 * Also if partial send happens (e.g. 'ret' != 'skb_len') 423 * somehow, we break this loop, but account such returned 424 * value in 'virtio_transport_put_credit()'. 425 */ 426 rest_len -= ret; 427 428 if (WARN_ONCE(ret != skb_len, 429 "'send_pkt()' returns %i, but %zu expected\n", 430 ret, skb_len)) 431 break; 432 } while (rest_len); 433 434 virtio_transport_put_credit(vvs, rest_len); 435 436 /* Return number of bytes, if any data has been sent. */ 437 if (rest_len != pkt_len) 438 ret = pkt_len - rest_len; 439 440 return ret; 441 } 442 443 static bool virtio_transport_inc_rx_pkt(struct virtio_vsock_sock *vvs, 444 u32 len) 445 { 446 if (vvs->buf_used + len > vvs->buf_alloc) 447 return false; 448 449 vvs->rx_bytes += len; 450 vvs->buf_used += len; 451 return true; 452 } 453 454 static void virtio_transport_dec_rx_pkt(struct virtio_vsock_sock *vvs, 455 u32 bytes_read, u32 bytes_dequeued) 456 { 457 vvs->rx_bytes -= bytes_read; 458 vvs->buf_used -= bytes_dequeued; 459 vvs->fwd_cnt += bytes_dequeued; 460 } 461 462 void virtio_transport_inc_tx_pkt(struct virtio_vsock_sock *vvs, struct sk_buff *skb) 463 { 464 struct virtio_vsock_hdr *hdr = virtio_vsock_hdr(skb); 465 466 spin_lock_bh(&vvs->rx_lock); 467 vvs->last_fwd_cnt = vvs->fwd_cnt; 468 hdr->fwd_cnt = cpu_to_le32(vvs->fwd_cnt); 469 hdr->buf_alloc = cpu_to_le32(vvs->buf_alloc); 470 spin_unlock_bh(&vvs->rx_lock); 471 } 472 EXPORT_SYMBOL_GPL(virtio_transport_inc_tx_pkt); 473 474 void virtio_transport_consume_skb_sent(struct sk_buff *skb, bool consume) 475 { 476 struct sock *s = skb->sk; 477 478 if (s && skb->len) { 479 struct vsock_sock *vs = vsock_sk(s); 480 struct virtio_vsock_sock *vvs; 481 482 vvs = vs->trans; 483 484 spin_lock_bh(&vvs->tx_lock); 485 vvs->bytes_unsent -= skb->len; 486 spin_unlock_bh(&vvs->tx_lock); 487 } 488 489 if (consume) 490 consume_skb(skb); 491 } 492 EXPORT_SYMBOL_GPL(virtio_transport_consume_skb_sent); 493 494 u32 virtio_transport_get_credit(struct virtio_vsock_sock *vvs, u32 credit) 495 { 496 u32 ret; 497 498 if (!credit) 499 return 0; 500 501 spin_lock_bh(&vvs->tx_lock); 502 ret = vvs->peer_buf_alloc - (vvs->tx_cnt - vvs->peer_fwd_cnt); 503 if (ret > credit) 504 ret = credit; 505 vvs->tx_cnt += ret; 506 vvs->bytes_unsent += ret; 507 spin_unlock_bh(&vvs->tx_lock); 508 509 return ret; 510 } 511 EXPORT_SYMBOL_GPL(virtio_transport_get_credit); 512 513 void virtio_transport_put_credit(struct virtio_vsock_sock *vvs, u32 credit) 514 { 515 if (!credit) 516 return; 517 518 spin_lock_bh(&vvs->tx_lock); 519 vvs->tx_cnt -= credit; 520 vvs->bytes_unsent -= credit; 521 spin_unlock_bh(&vvs->tx_lock); 522 } 523 EXPORT_SYMBOL_GPL(virtio_transport_put_credit); 524 525 static int virtio_transport_send_credit_update(struct vsock_sock *vsk) 526 { 527 struct virtio_vsock_pkt_info info = { 528 .op = VIRTIO_VSOCK_OP_CREDIT_UPDATE, 529 .vsk = vsk, 530 }; 531 532 return virtio_transport_send_pkt_info(vsk, &info); 533 } 534 535 static ssize_t 536 virtio_transport_stream_do_peek(struct vsock_sock *vsk, 537 struct msghdr *msg, 538 size_t len) 539 { 540 struct virtio_vsock_sock *vvs = vsk->trans; 541 struct sk_buff *skb; 542 size_t total = 0; 543 int err; 544 545 spin_lock_bh(&vvs->rx_lock); 546 547 skb_queue_walk(&vvs->rx_queue, skb) { 548 size_t bytes; 549 550 bytes = len - total; 551 if (bytes > skb->len) 552 bytes = skb->len; 553 554 spin_unlock_bh(&vvs->rx_lock); 555 556 /* sk_lock is held by caller so no one else can dequeue. 557 * Unlock rx_lock since skb_copy_datagram_iter() may sleep. 558 */ 559 err = skb_copy_datagram_iter(skb, VIRTIO_VSOCK_SKB_CB(skb)->offset, 560 &msg->msg_iter, bytes); 561 if (err) 562 goto out; 563 564 total += bytes; 565 566 spin_lock_bh(&vvs->rx_lock); 567 568 if (total == len) 569 break; 570 } 571 572 spin_unlock_bh(&vvs->rx_lock); 573 574 return total; 575 576 out: 577 if (total) 578 err = total; 579 return err; 580 } 581 582 static ssize_t 583 virtio_transport_stream_do_dequeue(struct vsock_sock *vsk, 584 struct msghdr *msg, 585 size_t len) 586 { 587 struct virtio_vsock_sock *vvs = vsk->trans; 588 struct sk_buff *skb; 589 u32 fwd_cnt_delta; 590 bool low_rx_bytes; 591 int err = -EFAULT; 592 size_t total = 0; 593 u32 free_space; 594 595 spin_lock_bh(&vvs->rx_lock); 596 597 if (WARN_ONCE(skb_queue_empty(&vvs->rx_queue) && vvs->rx_bytes, 598 "rx_queue is empty, but rx_bytes is non-zero\n")) { 599 spin_unlock_bh(&vvs->rx_lock); 600 return err; 601 } 602 603 while (total < len && !skb_queue_empty(&vvs->rx_queue)) { 604 size_t bytes, dequeued = 0; 605 606 skb = skb_peek(&vvs->rx_queue); 607 608 bytes = min_t(size_t, len - total, 609 skb->len - VIRTIO_VSOCK_SKB_CB(skb)->offset); 610 611 /* sk_lock is held by caller so no one else can dequeue. 612 * Unlock rx_lock since skb_copy_datagram_iter() may sleep. 613 */ 614 spin_unlock_bh(&vvs->rx_lock); 615 616 err = skb_copy_datagram_iter(skb, 617 VIRTIO_VSOCK_SKB_CB(skb)->offset, 618 &msg->msg_iter, bytes); 619 if (err) 620 goto out; 621 622 spin_lock_bh(&vvs->rx_lock); 623 624 total += bytes; 625 626 VIRTIO_VSOCK_SKB_CB(skb)->offset += bytes; 627 628 if (skb->len == VIRTIO_VSOCK_SKB_CB(skb)->offset) { 629 dequeued = le32_to_cpu(virtio_vsock_hdr(skb)->len); 630 __skb_unlink(skb, &vvs->rx_queue); 631 consume_skb(skb); 632 } 633 634 virtio_transport_dec_rx_pkt(vvs, bytes, dequeued); 635 } 636 637 fwd_cnt_delta = vvs->fwd_cnt - vvs->last_fwd_cnt; 638 free_space = vvs->buf_alloc - fwd_cnt_delta; 639 low_rx_bytes = (vvs->rx_bytes < 640 sock_rcvlowat(sk_vsock(vsk), 0, INT_MAX)); 641 642 spin_unlock_bh(&vvs->rx_lock); 643 644 /* To reduce the number of credit update messages, 645 * don't update credits as long as lots of space is available. 646 * Note: the limit chosen here is arbitrary. Setting the limit 647 * too high causes extra messages. Too low causes transmitter 648 * stalls. As stalls are in theory more expensive than extra 649 * messages, we set the limit to a high value. TODO: experiment 650 * with different values. Also send credit update message when 651 * number of bytes in rx queue is not enough to wake up reader. 652 */ 653 if (fwd_cnt_delta && 654 (free_space < VIRTIO_VSOCK_MAX_PKT_BUF_SIZE || low_rx_bytes)) 655 virtio_transport_send_credit_update(vsk); 656 657 return total; 658 659 out: 660 if (total) 661 err = total; 662 return err; 663 } 664 665 static ssize_t 666 virtio_transport_seqpacket_do_peek(struct vsock_sock *vsk, 667 struct msghdr *msg) 668 { 669 struct virtio_vsock_sock *vvs = vsk->trans; 670 struct sk_buff *skb; 671 size_t total, len; 672 673 spin_lock_bh(&vvs->rx_lock); 674 675 if (!vvs->msg_count) { 676 spin_unlock_bh(&vvs->rx_lock); 677 return 0; 678 } 679 680 total = 0; 681 len = msg_data_left(msg); 682 683 skb_queue_walk(&vvs->rx_queue, skb) { 684 struct virtio_vsock_hdr *hdr; 685 686 if (total < len) { 687 size_t bytes; 688 int err; 689 690 bytes = len - total; 691 if (bytes > skb->len) 692 bytes = skb->len; 693 694 spin_unlock_bh(&vvs->rx_lock); 695 696 /* sk_lock is held by caller so no one else can dequeue. 697 * Unlock rx_lock since skb_copy_datagram_iter() may sleep. 698 */ 699 err = skb_copy_datagram_iter(skb, VIRTIO_VSOCK_SKB_CB(skb)->offset, 700 &msg->msg_iter, bytes); 701 if (err) 702 return err; 703 704 spin_lock_bh(&vvs->rx_lock); 705 } 706 707 total += skb->len; 708 hdr = virtio_vsock_hdr(skb); 709 710 if (le32_to_cpu(hdr->flags) & VIRTIO_VSOCK_SEQ_EOM) { 711 if (le32_to_cpu(hdr->flags) & VIRTIO_VSOCK_SEQ_EOR) 712 msg->msg_flags |= MSG_EOR; 713 714 break; 715 } 716 } 717 718 spin_unlock_bh(&vvs->rx_lock); 719 720 return total; 721 } 722 723 static int virtio_transport_seqpacket_do_dequeue(struct vsock_sock *vsk, 724 struct msghdr *msg, 725 int flags) 726 { 727 struct virtio_vsock_sock *vvs = vsk->trans; 728 int dequeued_len = 0; 729 size_t user_buf_len = msg_data_left(msg); 730 bool msg_ready = false; 731 struct sk_buff *skb; 732 733 spin_lock_bh(&vvs->rx_lock); 734 735 if (vvs->msg_count == 0) { 736 spin_unlock_bh(&vvs->rx_lock); 737 return 0; 738 } 739 740 while (!msg_ready) { 741 struct virtio_vsock_hdr *hdr; 742 size_t pkt_len; 743 744 skb = __skb_dequeue(&vvs->rx_queue); 745 if (!skb) 746 break; 747 hdr = virtio_vsock_hdr(skb); 748 pkt_len = (size_t)le32_to_cpu(hdr->len); 749 750 if (dequeued_len >= 0) { 751 size_t bytes_to_copy; 752 753 bytes_to_copy = min(user_buf_len, pkt_len); 754 755 if (bytes_to_copy) { 756 int err; 757 758 /* sk_lock is held by caller so no one else can dequeue. 759 * Unlock rx_lock since skb_copy_datagram_iter() may sleep. 760 */ 761 spin_unlock_bh(&vvs->rx_lock); 762 763 err = skb_copy_datagram_iter(skb, 0, 764 &msg->msg_iter, 765 bytes_to_copy); 766 if (err) { 767 /* Copy of message failed. Rest of 768 * fragments will be freed without copy. 769 */ 770 dequeued_len = err; 771 } else { 772 user_buf_len -= bytes_to_copy; 773 } 774 775 spin_lock_bh(&vvs->rx_lock); 776 } 777 778 if (dequeued_len >= 0) 779 dequeued_len += pkt_len; 780 } 781 782 if (le32_to_cpu(hdr->flags) & VIRTIO_VSOCK_SEQ_EOM) { 783 msg_ready = true; 784 vvs->msg_count--; 785 786 if (le32_to_cpu(hdr->flags) & VIRTIO_VSOCK_SEQ_EOR) 787 msg->msg_flags |= MSG_EOR; 788 } 789 790 virtio_transport_dec_rx_pkt(vvs, pkt_len, pkt_len); 791 kfree_skb(skb); 792 } 793 794 spin_unlock_bh(&vvs->rx_lock); 795 796 virtio_transport_send_credit_update(vsk); 797 798 return dequeued_len; 799 } 800 801 ssize_t 802 virtio_transport_stream_dequeue(struct vsock_sock *vsk, 803 struct msghdr *msg, 804 size_t len, int flags) 805 { 806 if (flags & MSG_PEEK) 807 return virtio_transport_stream_do_peek(vsk, msg, len); 808 else 809 return virtio_transport_stream_do_dequeue(vsk, msg, len); 810 } 811 EXPORT_SYMBOL_GPL(virtio_transport_stream_dequeue); 812 813 ssize_t 814 virtio_transport_seqpacket_dequeue(struct vsock_sock *vsk, 815 struct msghdr *msg, 816 int flags) 817 { 818 if (flags & MSG_PEEK) 819 return virtio_transport_seqpacket_do_peek(vsk, msg); 820 else 821 return virtio_transport_seqpacket_do_dequeue(vsk, msg, flags); 822 } 823 EXPORT_SYMBOL_GPL(virtio_transport_seqpacket_dequeue); 824 825 int 826 virtio_transport_seqpacket_enqueue(struct vsock_sock *vsk, 827 struct msghdr *msg, 828 size_t len) 829 { 830 struct virtio_vsock_sock *vvs = vsk->trans; 831 832 spin_lock_bh(&vvs->tx_lock); 833 834 if (len > vvs->peer_buf_alloc) { 835 spin_unlock_bh(&vvs->tx_lock); 836 return -EMSGSIZE; 837 } 838 839 spin_unlock_bh(&vvs->tx_lock); 840 841 return virtio_transport_stream_enqueue(vsk, msg, len); 842 } 843 EXPORT_SYMBOL_GPL(virtio_transport_seqpacket_enqueue); 844 845 int 846 virtio_transport_dgram_dequeue(struct vsock_sock *vsk, 847 struct msghdr *msg, 848 size_t len, int flags) 849 { 850 return -EOPNOTSUPP; 851 } 852 EXPORT_SYMBOL_GPL(virtio_transport_dgram_dequeue); 853 854 s64 virtio_transport_stream_has_data(struct vsock_sock *vsk) 855 { 856 struct virtio_vsock_sock *vvs = vsk->trans; 857 s64 bytes; 858 859 spin_lock_bh(&vvs->rx_lock); 860 bytes = vvs->rx_bytes; 861 spin_unlock_bh(&vvs->rx_lock); 862 863 return bytes; 864 } 865 EXPORT_SYMBOL_GPL(virtio_transport_stream_has_data); 866 867 u32 virtio_transport_seqpacket_has_data(struct vsock_sock *vsk) 868 { 869 struct virtio_vsock_sock *vvs = vsk->trans; 870 u32 msg_count; 871 872 spin_lock_bh(&vvs->rx_lock); 873 msg_count = vvs->msg_count; 874 spin_unlock_bh(&vvs->rx_lock); 875 876 return msg_count; 877 } 878 EXPORT_SYMBOL_GPL(virtio_transport_seqpacket_has_data); 879 880 static s64 virtio_transport_has_space(struct vsock_sock *vsk) 881 { 882 struct virtio_vsock_sock *vvs = vsk->trans; 883 s64 bytes; 884 885 bytes = (s64)vvs->peer_buf_alloc - (vvs->tx_cnt - vvs->peer_fwd_cnt); 886 if (bytes < 0) 887 bytes = 0; 888 889 return bytes; 890 } 891 892 s64 virtio_transport_stream_has_space(struct vsock_sock *vsk) 893 { 894 struct virtio_vsock_sock *vvs = vsk->trans; 895 s64 bytes; 896 897 spin_lock_bh(&vvs->tx_lock); 898 bytes = virtio_transport_has_space(vsk); 899 spin_unlock_bh(&vvs->tx_lock); 900 901 return bytes; 902 } 903 EXPORT_SYMBOL_GPL(virtio_transport_stream_has_space); 904 905 int virtio_transport_do_socket_init(struct vsock_sock *vsk, 906 struct vsock_sock *psk) 907 { 908 struct virtio_vsock_sock *vvs; 909 910 vvs = kzalloc(sizeof(*vvs), GFP_KERNEL); 911 if (!vvs) 912 return -ENOMEM; 913 914 vsk->trans = vvs; 915 vvs->vsk = vsk; 916 if (psk && psk->trans) { 917 struct virtio_vsock_sock *ptrans = psk->trans; 918 919 vvs->peer_buf_alloc = ptrans->peer_buf_alloc; 920 } 921 922 if (vsk->buffer_size > VIRTIO_VSOCK_MAX_BUF_SIZE) 923 vsk->buffer_size = VIRTIO_VSOCK_MAX_BUF_SIZE; 924 925 vvs->buf_alloc = vsk->buffer_size; 926 927 spin_lock_init(&vvs->rx_lock); 928 spin_lock_init(&vvs->tx_lock); 929 skb_queue_head_init(&vvs->rx_queue); 930 931 return 0; 932 } 933 EXPORT_SYMBOL_GPL(virtio_transport_do_socket_init); 934 935 /* sk_lock held by the caller */ 936 void virtio_transport_notify_buffer_size(struct vsock_sock *vsk, u64 *val) 937 { 938 struct virtio_vsock_sock *vvs = vsk->trans; 939 940 if (*val > VIRTIO_VSOCK_MAX_BUF_SIZE) 941 *val = VIRTIO_VSOCK_MAX_BUF_SIZE; 942 943 vvs->buf_alloc = *val; 944 945 virtio_transport_send_credit_update(vsk); 946 } 947 EXPORT_SYMBOL_GPL(virtio_transport_notify_buffer_size); 948 949 int 950 virtio_transport_notify_poll_in(struct vsock_sock *vsk, 951 size_t target, 952 bool *data_ready_now) 953 { 954 *data_ready_now = vsock_stream_has_data(vsk) >= target; 955 956 return 0; 957 } 958 EXPORT_SYMBOL_GPL(virtio_transport_notify_poll_in); 959 960 int 961 virtio_transport_notify_poll_out(struct vsock_sock *vsk, 962 size_t target, 963 bool *space_avail_now) 964 { 965 s64 free_space; 966 967 free_space = vsock_stream_has_space(vsk); 968 if (free_space > 0) 969 *space_avail_now = true; 970 else if (free_space == 0) 971 *space_avail_now = false; 972 973 return 0; 974 } 975 EXPORT_SYMBOL_GPL(virtio_transport_notify_poll_out); 976 977 int virtio_transport_notify_recv_init(struct vsock_sock *vsk, 978 size_t target, struct vsock_transport_recv_notify_data *data) 979 { 980 return 0; 981 } 982 EXPORT_SYMBOL_GPL(virtio_transport_notify_recv_init); 983 984 int virtio_transport_notify_recv_pre_block(struct vsock_sock *vsk, 985 size_t target, struct vsock_transport_recv_notify_data *data) 986 { 987 return 0; 988 } 989 EXPORT_SYMBOL_GPL(virtio_transport_notify_recv_pre_block); 990 991 int virtio_transport_notify_recv_pre_dequeue(struct vsock_sock *vsk, 992 size_t target, struct vsock_transport_recv_notify_data *data) 993 { 994 return 0; 995 } 996 EXPORT_SYMBOL_GPL(virtio_transport_notify_recv_pre_dequeue); 997 998 int virtio_transport_notify_recv_post_dequeue(struct vsock_sock *vsk, 999 size_t target, ssize_t copied, bool data_read, 1000 struct vsock_transport_recv_notify_data *data) 1001 { 1002 return 0; 1003 } 1004 EXPORT_SYMBOL_GPL(virtio_transport_notify_recv_post_dequeue); 1005 1006 int virtio_transport_notify_send_init(struct vsock_sock *vsk, 1007 struct vsock_transport_send_notify_data *data) 1008 { 1009 return 0; 1010 } 1011 EXPORT_SYMBOL_GPL(virtio_transport_notify_send_init); 1012 1013 int virtio_transport_notify_send_pre_block(struct vsock_sock *vsk, 1014 struct vsock_transport_send_notify_data *data) 1015 { 1016 return 0; 1017 } 1018 EXPORT_SYMBOL_GPL(virtio_transport_notify_send_pre_block); 1019 1020 int virtio_transport_notify_send_pre_enqueue(struct vsock_sock *vsk, 1021 struct vsock_transport_send_notify_data *data) 1022 { 1023 return 0; 1024 } 1025 EXPORT_SYMBOL_GPL(virtio_transport_notify_send_pre_enqueue); 1026 1027 int virtio_transport_notify_send_post_enqueue(struct vsock_sock *vsk, 1028 ssize_t written, struct vsock_transport_send_notify_data *data) 1029 { 1030 return 0; 1031 } 1032 EXPORT_SYMBOL_GPL(virtio_transport_notify_send_post_enqueue); 1033 1034 u64 virtio_transport_stream_rcvhiwat(struct vsock_sock *vsk) 1035 { 1036 return vsk->buffer_size; 1037 } 1038 EXPORT_SYMBOL_GPL(virtio_transport_stream_rcvhiwat); 1039 1040 bool virtio_transport_stream_is_active(struct vsock_sock *vsk) 1041 { 1042 return true; 1043 } 1044 EXPORT_SYMBOL_GPL(virtio_transport_stream_is_active); 1045 1046 bool virtio_transport_stream_allow(u32 cid, u32 port) 1047 { 1048 return true; 1049 } 1050 EXPORT_SYMBOL_GPL(virtio_transport_stream_allow); 1051 1052 int virtio_transport_dgram_bind(struct vsock_sock *vsk, 1053 struct sockaddr_vm *addr) 1054 { 1055 return -EOPNOTSUPP; 1056 } 1057 EXPORT_SYMBOL_GPL(virtio_transport_dgram_bind); 1058 1059 bool virtio_transport_dgram_allow(u32 cid, u32 port) 1060 { 1061 return false; 1062 } 1063 EXPORT_SYMBOL_GPL(virtio_transport_dgram_allow); 1064 1065 int virtio_transport_connect(struct vsock_sock *vsk) 1066 { 1067 struct virtio_vsock_pkt_info info = { 1068 .op = VIRTIO_VSOCK_OP_REQUEST, 1069 .vsk = vsk, 1070 }; 1071 1072 return virtio_transport_send_pkt_info(vsk, &info); 1073 } 1074 EXPORT_SYMBOL_GPL(virtio_transport_connect); 1075 1076 int virtio_transport_shutdown(struct vsock_sock *vsk, int mode) 1077 { 1078 struct virtio_vsock_pkt_info info = { 1079 .op = VIRTIO_VSOCK_OP_SHUTDOWN, 1080 .flags = (mode & RCV_SHUTDOWN ? 1081 VIRTIO_VSOCK_SHUTDOWN_RCV : 0) | 1082 (mode & SEND_SHUTDOWN ? 1083 VIRTIO_VSOCK_SHUTDOWN_SEND : 0), 1084 .vsk = vsk, 1085 }; 1086 1087 return virtio_transport_send_pkt_info(vsk, &info); 1088 } 1089 EXPORT_SYMBOL_GPL(virtio_transport_shutdown); 1090 1091 int 1092 virtio_transport_dgram_enqueue(struct vsock_sock *vsk, 1093 struct sockaddr_vm *remote_addr, 1094 struct msghdr *msg, 1095 size_t dgram_len) 1096 { 1097 return -EOPNOTSUPP; 1098 } 1099 EXPORT_SYMBOL_GPL(virtio_transport_dgram_enqueue); 1100 1101 ssize_t 1102 virtio_transport_stream_enqueue(struct vsock_sock *vsk, 1103 struct msghdr *msg, 1104 size_t len) 1105 { 1106 struct virtio_vsock_pkt_info info = { 1107 .op = VIRTIO_VSOCK_OP_RW, 1108 .msg = msg, 1109 .pkt_len = len, 1110 .vsk = vsk, 1111 }; 1112 1113 return virtio_transport_send_pkt_info(vsk, &info); 1114 } 1115 EXPORT_SYMBOL_GPL(virtio_transport_stream_enqueue); 1116 1117 void virtio_transport_destruct(struct vsock_sock *vsk) 1118 { 1119 struct virtio_vsock_sock *vvs = vsk->trans; 1120 1121 virtio_transport_cancel_close_work(vsk, true); 1122 1123 kfree(vvs); 1124 vsk->trans = NULL; 1125 } 1126 EXPORT_SYMBOL_GPL(virtio_transport_destruct); 1127 1128 ssize_t virtio_transport_unsent_bytes(struct vsock_sock *vsk) 1129 { 1130 struct virtio_vsock_sock *vvs = vsk->trans; 1131 size_t ret; 1132 1133 spin_lock_bh(&vvs->tx_lock); 1134 ret = vvs->bytes_unsent; 1135 spin_unlock_bh(&vvs->tx_lock); 1136 1137 return ret; 1138 } 1139 EXPORT_SYMBOL_GPL(virtio_transport_unsent_bytes); 1140 1141 static int virtio_transport_reset(struct vsock_sock *vsk, 1142 struct sk_buff *skb) 1143 { 1144 struct virtio_vsock_pkt_info info = { 1145 .op = VIRTIO_VSOCK_OP_RST, 1146 .reply = !!skb, 1147 .vsk = vsk, 1148 }; 1149 1150 /* Send RST only if the original pkt is not a RST pkt */ 1151 if (skb && le16_to_cpu(virtio_vsock_hdr(skb)->op) == VIRTIO_VSOCK_OP_RST) 1152 return 0; 1153 1154 return virtio_transport_send_pkt_info(vsk, &info); 1155 } 1156 1157 /* Normally packets are associated with a socket. There may be no socket if an 1158 * attempt was made to connect to a socket that does not exist. 1159 */ 1160 static int virtio_transport_reset_no_sock(const struct virtio_transport *t, 1161 struct sk_buff *skb) 1162 { 1163 struct virtio_vsock_hdr *hdr = virtio_vsock_hdr(skb); 1164 struct virtio_vsock_pkt_info info = { 1165 .op = VIRTIO_VSOCK_OP_RST, 1166 .type = le16_to_cpu(hdr->type), 1167 .reply = true, 1168 }; 1169 struct sk_buff *reply; 1170 1171 /* Send RST only if the original pkt is not a RST pkt */ 1172 if (le16_to_cpu(hdr->op) == VIRTIO_VSOCK_OP_RST) 1173 return 0; 1174 1175 if (!t) 1176 return -ENOTCONN; 1177 1178 reply = virtio_transport_alloc_skb(&info, 0, false, 1179 le64_to_cpu(hdr->dst_cid), 1180 le32_to_cpu(hdr->dst_port), 1181 le64_to_cpu(hdr->src_cid), 1182 le32_to_cpu(hdr->src_port)); 1183 if (!reply) 1184 return -ENOMEM; 1185 1186 return t->send_pkt(reply); 1187 } 1188 1189 /* This function should be called with sk_lock held and SOCK_DONE set */ 1190 static void virtio_transport_remove_sock(struct vsock_sock *vsk) 1191 { 1192 struct virtio_vsock_sock *vvs = vsk->trans; 1193 1194 /* We don't need to take rx_lock, as the socket is closing and we are 1195 * removing it. 1196 */ 1197 __skb_queue_purge(&vvs->rx_queue); 1198 vsock_remove_sock(vsk); 1199 } 1200 1201 static void virtio_transport_cancel_close_work(struct vsock_sock *vsk, 1202 bool cancel_timeout) 1203 { 1204 struct sock *sk = sk_vsock(vsk); 1205 1206 if (vsk->close_work_scheduled && 1207 (!cancel_timeout || cancel_delayed_work(&vsk->close_work))) { 1208 vsk->close_work_scheduled = false; 1209 1210 virtio_transport_remove_sock(vsk); 1211 1212 /* Release refcnt obtained when we scheduled the timeout */ 1213 sock_put(sk); 1214 } 1215 } 1216 1217 static void virtio_transport_do_close(struct vsock_sock *vsk, 1218 bool cancel_timeout) 1219 { 1220 struct sock *sk = sk_vsock(vsk); 1221 1222 sock_set_flag(sk, SOCK_DONE); 1223 vsk->peer_shutdown = SHUTDOWN_MASK; 1224 if (vsock_stream_has_data(vsk) <= 0) 1225 sk->sk_state = TCP_CLOSING; 1226 sk->sk_state_change(sk); 1227 1228 virtio_transport_cancel_close_work(vsk, cancel_timeout); 1229 } 1230 1231 static void virtio_transport_close_timeout(struct work_struct *work) 1232 { 1233 struct vsock_sock *vsk = 1234 container_of(work, struct vsock_sock, close_work.work); 1235 struct sock *sk = sk_vsock(vsk); 1236 1237 sock_hold(sk); 1238 lock_sock(sk); 1239 1240 if (!sock_flag(sk, SOCK_DONE)) { 1241 (void)virtio_transport_reset(vsk, NULL); 1242 1243 virtio_transport_do_close(vsk, false); 1244 } 1245 1246 vsk->close_work_scheduled = false; 1247 1248 release_sock(sk); 1249 sock_put(sk); 1250 } 1251 1252 /* User context, vsk->sk is locked */ 1253 static bool virtio_transport_close(struct vsock_sock *vsk) 1254 { 1255 struct sock *sk = &vsk->sk; 1256 1257 if (!(sk->sk_state == TCP_ESTABLISHED || 1258 sk->sk_state == TCP_CLOSING)) 1259 return true; 1260 1261 /* Already received SHUTDOWN from peer, reply with RST */ 1262 if ((vsk->peer_shutdown & SHUTDOWN_MASK) == SHUTDOWN_MASK) { 1263 (void)virtio_transport_reset(vsk, NULL); 1264 return true; 1265 } 1266 1267 if ((sk->sk_shutdown & SHUTDOWN_MASK) != SHUTDOWN_MASK) 1268 (void)virtio_transport_shutdown(vsk, SHUTDOWN_MASK); 1269 1270 if (!(current->flags & PF_EXITING)) 1271 vsock_linger(sk); 1272 1273 if (sock_flag(sk, SOCK_DONE)) { 1274 return true; 1275 } 1276 1277 sock_hold(sk); 1278 INIT_DELAYED_WORK(&vsk->close_work, 1279 virtio_transport_close_timeout); 1280 vsk->close_work_scheduled = true; 1281 schedule_delayed_work(&vsk->close_work, VSOCK_CLOSE_TIMEOUT); 1282 return false; 1283 } 1284 1285 void virtio_transport_release(struct vsock_sock *vsk) 1286 { 1287 struct sock *sk = &vsk->sk; 1288 bool remove_sock = true; 1289 1290 if (sk->sk_type == SOCK_STREAM || sk->sk_type == SOCK_SEQPACKET) 1291 remove_sock = virtio_transport_close(vsk); 1292 1293 if (remove_sock) { 1294 sock_set_flag(sk, SOCK_DONE); 1295 virtio_transport_remove_sock(vsk); 1296 } 1297 } 1298 EXPORT_SYMBOL_GPL(virtio_transport_release); 1299 1300 static int 1301 virtio_transport_recv_connecting(struct sock *sk, 1302 struct sk_buff *skb) 1303 { 1304 struct virtio_vsock_hdr *hdr = virtio_vsock_hdr(skb); 1305 struct vsock_sock *vsk = vsock_sk(sk); 1306 int skerr; 1307 int err; 1308 1309 switch (le16_to_cpu(hdr->op)) { 1310 case VIRTIO_VSOCK_OP_RESPONSE: 1311 sk->sk_state = TCP_ESTABLISHED; 1312 sk->sk_socket->state = SS_CONNECTED; 1313 vsock_insert_connected(vsk); 1314 sk->sk_state_change(sk); 1315 break; 1316 case VIRTIO_VSOCK_OP_INVALID: 1317 break; 1318 case VIRTIO_VSOCK_OP_RST: 1319 skerr = ECONNRESET; 1320 err = 0; 1321 goto destroy; 1322 default: 1323 skerr = EPROTO; 1324 err = -EINVAL; 1325 goto destroy; 1326 } 1327 return 0; 1328 1329 destroy: 1330 virtio_transport_reset(vsk, skb); 1331 sk->sk_state = TCP_CLOSE; 1332 sk->sk_err = skerr; 1333 sk_error_report(sk); 1334 return err; 1335 } 1336 1337 static void 1338 virtio_transport_recv_enqueue(struct vsock_sock *vsk, 1339 struct sk_buff *skb) 1340 { 1341 struct virtio_vsock_sock *vvs = vsk->trans; 1342 bool can_enqueue, free_pkt = false; 1343 struct virtio_vsock_hdr *hdr; 1344 u32 len; 1345 1346 hdr = virtio_vsock_hdr(skb); 1347 len = le32_to_cpu(hdr->len); 1348 1349 spin_lock_bh(&vvs->rx_lock); 1350 1351 can_enqueue = virtio_transport_inc_rx_pkt(vvs, len); 1352 if (!can_enqueue) { 1353 free_pkt = true; 1354 goto out; 1355 } 1356 1357 if (le32_to_cpu(hdr->flags) & VIRTIO_VSOCK_SEQ_EOM) 1358 vvs->msg_count++; 1359 1360 /* Try to copy small packets into the buffer of last packet queued, 1361 * to avoid wasting memory queueing the entire buffer with a small 1362 * payload. 1363 */ 1364 if (len <= GOOD_COPY_LEN && !skb_queue_empty(&vvs->rx_queue)) { 1365 struct virtio_vsock_hdr *last_hdr; 1366 struct sk_buff *last_skb; 1367 1368 last_skb = skb_peek_tail(&vvs->rx_queue); 1369 last_hdr = virtio_vsock_hdr(last_skb); 1370 1371 /* If there is space in the last packet queued, we copy the 1372 * new packet in its buffer. We avoid this if the last packet 1373 * queued has VIRTIO_VSOCK_SEQ_EOM set, because this is 1374 * delimiter of SEQPACKET message, so 'pkt' is the first packet 1375 * of a new message. 1376 */ 1377 if (skb->len < skb_tailroom(last_skb) && 1378 !(le32_to_cpu(last_hdr->flags) & VIRTIO_VSOCK_SEQ_EOM)) { 1379 memcpy(skb_put(last_skb, skb->len), skb->data, skb->len); 1380 free_pkt = true; 1381 last_hdr->flags |= hdr->flags; 1382 le32_add_cpu(&last_hdr->len, len); 1383 goto out; 1384 } 1385 } 1386 1387 __skb_queue_tail(&vvs->rx_queue, skb); 1388 1389 out: 1390 spin_unlock_bh(&vvs->rx_lock); 1391 if (free_pkt) 1392 kfree_skb(skb); 1393 } 1394 1395 static int 1396 virtio_transport_recv_connected(struct sock *sk, 1397 struct sk_buff *skb) 1398 { 1399 struct virtio_vsock_hdr *hdr = virtio_vsock_hdr(skb); 1400 struct vsock_sock *vsk = vsock_sk(sk); 1401 int err = 0; 1402 1403 switch (le16_to_cpu(hdr->op)) { 1404 case VIRTIO_VSOCK_OP_RW: 1405 virtio_transport_recv_enqueue(vsk, skb); 1406 vsock_data_ready(sk); 1407 return err; 1408 case VIRTIO_VSOCK_OP_CREDIT_REQUEST: 1409 virtio_transport_send_credit_update(vsk); 1410 break; 1411 case VIRTIO_VSOCK_OP_CREDIT_UPDATE: 1412 sk->sk_write_space(sk); 1413 break; 1414 case VIRTIO_VSOCK_OP_SHUTDOWN: 1415 if (le32_to_cpu(hdr->flags) & VIRTIO_VSOCK_SHUTDOWN_RCV) 1416 vsk->peer_shutdown |= RCV_SHUTDOWN; 1417 if (le32_to_cpu(hdr->flags) & VIRTIO_VSOCK_SHUTDOWN_SEND) 1418 vsk->peer_shutdown |= SEND_SHUTDOWN; 1419 if (vsk->peer_shutdown == SHUTDOWN_MASK) { 1420 if (vsock_stream_has_data(vsk) <= 0 && !sock_flag(sk, SOCK_DONE)) { 1421 (void)virtio_transport_reset(vsk, NULL); 1422 virtio_transport_do_close(vsk, true); 1423 } 1424 /* Remove this socket anyway because the remote peer sent 1425 * the shutdown. This way a new connection will succeed 1426 * if the remote peer uses the same source port, 1427 * even if the old socket is still unreleased, but now disconnected. 1428 */ 1429 vsock_remove_sock(vsk); 1430 } 1431 if (le32_to_cpu(virtio_vsock_hdr(skb)->flags)) 1432 sk->sk_state_change(sk); 1433 break; 1434 case VIRTIO_VSOCK_OP_RST: 1435 virtio_transport_do_close(vsk, true); 1436 break; 1437 default: 1438 err = -EINVAL; 1439 break; 1440 } 1441 1442 kfree_skb(skb); 1443 return err; 1444 } 1445 1446 static void 1447 virtio_transport_recv_disconnecting(struct sock *sk, 1448 struct sk_buff *skb) 1449 { 1450 struct virtio_vsock_hdr *hdr = virtio_vsock_hdr(skb); 1451 struct vsock_sock *vsk = vsock_sk(sk); 1452 1453 if (le16_to_cpu(hdr->op) == VIRTIO_VSOCK_OP_RST) 1454 virtio_transport_do_close(vsk, true); 1455 } 1456 1457 static int 1458 virtio_transport_send_response(struct vsock_sock *vsk, 1459 struct sk_buff *skb) 1460 { 1461 struct virtio_vsock_hdr *hdr = virtio_vsock_hdr(skb); 1462 struct virtio_vsock_pkt_info info = { 1463 .op = VIRTIO_VSOCK_OP_RESPONSE, 1464 .remote_cid = le64_to_cpu(hdr->src_cid), 1465 .remote_port = le32_to_cpu(hdr->src_port), 1466 .reply = true, 1467 .vsk = vsk, 1468 }; 1469 1470 return virtio_transport_send_pkt_info(vsk, &info); 1471 } 1472 1473 static bool virtio_transport_space_update(struct sock *sk, 1474 struct sk_buff *skb) 1475 { 1476 struct virtio_vsock_hdr *hdr = virtio_vsock_hdr(skb); 1477 struct vsock_sock *vsk = vsock_sk(sk); 1478 struct virtio_vsock_sock *vvs = vsk->trans; 1479 bool space_available; 1480 1481 /* Listener sockets are not associated with any transport, so we are 1482 * not able to take the state to see if there is space available in the 1483 * remote peer, but since they are only used to receive requests, we 1484 * can assume that there is always space available in the other peer. 1485 */ 1486 if (!vvs) 1487 return true; 1488 1489 /* buf_alloc and fwd_cnt is always included in the hdr */ 1490 spin_lock_bh(&vvs->tx_lock); 1491 vvs->peer_buf_alloc = le32_to_cpu(hdr->buf_alloc); 1492 vvs->peer_fwd_cnt = le32_to_cpu(hdr->fwd_cnt); 1493 space_available = virtio_transport_has_space(vsk); 1494 spin_unlock_bh(&vvs->tx_lock); 1495 return space_available; 1496 } 1497 1498 /* Handle server socket */ 1499 static int 1500 virtio_transport_recv_listen(struct sock *sk, struct sk_buff *skb, 1501 struct virtio_transport *t) 1502 { 1503 struct virtio_vsock_hdr *hdr = virtio_vsock_hdr(skb); 1504 struct vsock_sock *vsk = vsock_sk(sk); 1505 struct vsock_sock *vchild; 1506 struct sock *child; 1507 int ret; 1508 1509 if (le16_to_cpu(hdr->op) != VIRTIO_VSOCK_OP_REQUEST) { 1510 virtio_transport_reset_no_sock(t, skb); 1511 return -EINVAL; 1512 } 1513 1514 if (sk_acceptq_is_full(sk)) { 1515 virtio_transport_reset_no_sock(t, skb); 1516 return -ENOMEM; 1517 } 1518 1519 /* __vsock_release() might have already flushed accept_queue. 1520 * Subsequent enqueues would lead to a memory leak. 1521 */ 1522 if (sk->sk_shutdown == SHUTDOWN_MASK) { 1523 virtio_transport_reset_no_sock(t, skb); 1524 return -ESHUTDOWN; 1525 } 1526 1527 child = vsock_create_connected(sk); 1528 if (!child) { 1529 virtio_transport_reset_no_sock(t, skb); 1530 return -ENOMEM; 1531 } 1532 1533 sk_acceptq_added(sk); 1534 1535 lock_sock_nested(child, SINGLE_DEPTH_NESTING); 1536 1537 child->sk_state = TCP_ESTABLISHED; 1538 1539 vchild = vsock_sk(child); 1540 vsock_addr_init(&vchild->local_addr, le64_to_cpu(hdr->dst_cid), 1541 le32_to_cpu(hdr->dst_port)); 1542 vsock_addr_init(&vchild->remote_addr, le64_to_cpu(hdr->src_cid), 1543 le32_to_cpu(hdr->src_port)); 1544 1545 ret = vsock_assign_transport(vchild, vsk); 1546 /* Transport assigned (looking at remote_addr) must be the same 1547 * where we received the request. 1548 */ 1549 if (ret || vchild->transport != &t->transport) { 1550 release_sock(child); 1551 virtio_transport_reset_no_sock(t, skb); 1552 sock_put(child); 1553 return ret; 1554 } 1555 1556 if (virtio_transport_space_update(child, skb)) 1557 child->sk_write_space(child); 1558 1559 vsock_insert_connected(vchild); 1560 vsock_enqueue_accept(sk, child); 1561 virtio_transport_send_response(vchild, skb); 1562 1563 release_sock(child); 1564 1565 sk->sk_data_ready(sk); 1566 return 0; 1567 } 1568 1569 static bool virtio_transport_valid_type(u16 type) 1570 { 1571 return (type == VIRTIO_VSOCK_TYPE_STREAM) || 1572 (type == VIRTIO_VSOCK_TYPE_SEQPACKET); 1573 } 1574 1575 /* We are under the virtio-vsock's vsock->rx_lock or vhost-vsock's vq->mutex 1576 * lock. 1577 */ 1578 void virtio_transport_recv_pkt(struct virtio_transport *t, 1579 struct sk_buff *skb) 1580 { 1581 struct virtio_vsock_hdr *hdr = virtio_vsock_hdr(skb); 1582 struct sockaddr_vm src, dst; 1583 struct vsock_sock *vsk; 1584 struct sock *sk; 1585 bool space_available; 1586 1587 vsock_addr_init(&src, le64_to_cpu(hdr->src_cid), 1588 le32_to_cpu(hdr->src_port)); 1589 vsock_addr_init(&dst, le64_to_cpu(hdr->dst_cid), 1590 le32_to_cpu(hdr->dst_port)); 1591 1592 trace_virtio_transport_recv_pkt(src.svm_cid, src.svm_port, 1593 dst.svm_cid, dst.svm_port, 1594 le32_to_cpu(hdr->len), 1595 le16_to_cpu(hdr->type), 1596 le16_to_cpu(hdr->op), 1597 le32_to_cpu(hdr->flags), 1598 le32_to_cpu(hdr->buf_alloc), 1599 le32_to_cpu(hdr->fwd_cnt)); 1600 1601 if (!virtio_transport_valid_type(le16_to_cpu(hdr->type))) { 1602 (void)virtio_transport_reset_no_sock(t, skb); 1603 goto free_pkt; 1604 } 1605 1606 /* The socket must be in connected or bound table 1607 * otherwise send reset back 1608 */ 1609 sk = vsock_find_connected_socket(&src, &dst); 1610 if (!sk) { 1611 sk = vsock_find_bound_socket(&dst); 1612 if (!sk) { 1613 (void)virtio_transport_reset_no_sock(t, skb); 1614 goto free_pkt; 1615 } 1616 } 1617 1618 if (virtio_transport_get_type(sk) != le16_to_cpu(hdr->type)) { 1619 (void)virtio_transport_reset_no_sock(t, skb); 1620 sock_put(sk); 1621 goto free_pkt; 1622 } 1623 1624 if (!skb_set_owner_sk_safe(skb, sk)) { 1625 WARN_ONCE(1, "receiving vsock socket has sk_refcnt == 0\n"); 1626 goto free_pkt; 1627 } 1628 1629 vsk = vsock_sk(sk); 1630 1631 lock_sock(sk); 1632 1633 /* Check if sk has been closed or assigned to another transport before 1634 * lock_sock (note: listener sockets are not assigned to any transport) 1635 */ 1636 if (sock_flag(sk, SOCK_DONE) || 1637 (sk->sk_state != TCP_LISTEN && vsk->transport != &t->transport)) { 1638 (void)virtio_transport_reset_no_sock(t, skb); 1639 release_sock(sk); 1640 sock_put(sk); 1641 goto free_pkt; 1642 } 1643 1644 space_available = virtio_transport_space_update(sk, skb); 1645 1646 /* Update CID in case it has changed after a transport reset event */ 1647 if (vsk->local_addr.svm_cid != VMADDR_CID_ANY) 1648 vsk->local_addr.svm_cid = dst.svm_cid; 1649 1650 if (space_available) 1651 sk->sk_write_space(sk); 1652 1653 switch (sk->sk_state) { 1654 case TCP_LISTEN: 1655 virtio_transport_recv_listen(sk, skb, t); 1656 kfree_skb(skb); 1657 break; 1658 case TCP_SYN_SENT: 1659 virtio_transport_recv_connecting(sk, skb); 1660 kfree_skb(skb); 1661 break; 1662 case TCP_ESTABLISHED: 1663 virtio_transport_recv_connected(sk, skb); 1664 break; 1665 case TCP_CLOSING: 1666 virtio_transport_recv_disconnecting(sk, skb); 1667 kfree_skb(skb); 1668 break; 1669 default: 1670 (void)virtio_transport_reset_no_sock(t, skb); 1671 kfree_skb(skb); 1672 break; 1673 } 1674 1675 release_sock(sk); 1676 1677 /* Release refcnt obtained when we fetched this socket out of the 1678 * bound or connected list. 1679 */ 1680 sock_put(sk); 1681 return; 1682 1683 free_pkt: 1684 kfree_skb(skb); 1685 } 1686 EXPORT_SYMBOL_GPL(virtio_transport_recv_pkt); 1687 1688 /* Remove skbs found in a queue that have a vsk that matches. 1689 * 1690 * Each skb is freed. 1691 * 1692 * Returns the count of skbs that were reply packets. 1693 */ 1694 int virtio_transport_purge_skbs(void *vsk, struct sk_buff_head *queue) 1695 { 1696 struct sk_buff_head freeme; 1697 struct sk_buff *skb, *tmp; 1698 int cnt = 0; 1699 1700 skb_queue_head_init(&freeme); 1701 1702 spin_lock_bh(&queue->lock); 1703 skb_queue_walk_safe(queue, skb, tmp) { 1704 if (vsock_sk(skb->sk) != vsk) 1705 continue; 1706 1707 __skb_unlink(skb, queue); 1708 __skb_queue_tail(&freeme, skb); 1709 1710 if (virtio_vsock_skb_reply(skb)) 1711 cnt++; 1712 } 1713 spin_unlock_bh(&queue->lock); 1714 1715 __skb_queue_purge(&freeme); 1716 1717 return cnt; 1718 } 1719 EXPORT_SYMBOL_GPL(virtio_transport_purge_skbs); 1720 1721 int virtio_transport_read_skb(struct vsock_sock *vsk, skb_read_actor_t recv_actor) 1722 { 1723 struct virtio_vsock_sock *vvs = vsk->trans; 1724 struct sock *sk = sk_vsock(vsk); 1725 struct virtio_vsock_hdr *hdr; 1726 struct sk_buff *skb; 1727 u32 pkt_len; 1728 int off = 0; 1729 int err; 1730 1731 spin_lock_bh(&vvs->rx_lock); 1732 /* Use __skb_recv_datagram() for race-free handling of the receive. It 1733 * works for types other than dgrams. 1734 */ 1735 skb = __skb_recv_datagram(sk, &vvs->rx_queue, MSG_DONTWAIT, &off, &err); 1736 if (!skb) { 1737 spin_unlock_bh(&vvs->rx_lock); 1738 return err; 1739 } 1740 1741 hdr = virtio_vsock_hdr(skb); 1742 if (le32_to_cpu(hdr->flags) & VIRTIO_VSOCK_SEQ_EOM) 1743 vvs->msg_count--; 1744 1745 pkt_len = le32_to_cpu(hdr->len); 1746 virtio_transport_dec_rx_pkt(vvs, pkt_len, pkt_len); 1747 spin_unlock_bh(&vvs->rx_lock); 1748 1749 virtio_transport_send_credit_update(vsk); 1750 1751 return recv_actor(sk, skb); 1752 } 1753 EXPORT_SYMBOL_GPL(virtio_transport_read_skb); 1754 1755 int virtio_transport_notify_set_rcvlowat(struct vsock_sock *vsk, int val) 1756 { 1757 struct virtio_vsock_sock *vvs = vsk->trans; 1758 bool send_update; 1759 1760 spin_lock_bh(&vvs->rx_lock); 1761 1762 /* If number of available bytes is less than new SO_RCVLOWAT value, 1763 * kick sender to send more data, because sender may sleep in its 1764 * 'send()' syscall waiting for enough space at our side. Also 1765 * don't send credit update when peer already knows actual value - 1766 * such transmission will be useless. 1767 */ 1768 send_update = (vvs->rx_bytes < val) && 1769 (vvs->fwd_cnt != vvs->last_fwd_cnt); 1770 1771 spin_unlock_bh(&vvs->rx_lock); 1772 1773 if (send_update) { 1774 int err; 1775 1776 err = virtio_transport_send_credit_update(vsk); 1777 if (err < 0) 1778 return err; 1779 } 1780 1781 return 0; 1782 } 1783 EXPORT_SYMBOL_GPL(virtio_transport_notify_set_rcvlowat); 1784 1785 MODULE_LICENSE("GPL v2"); 1786 MODULE_AUTHOR("Asias He"); 1787 MODULE_DESCRIPTION("common code for virtio vsock"); 1788