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