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 kfree_skb(skb); 404 ret = -ENOMEM; 405 break; 406 } 407 } 408 409 virtio_transport_inc_tx_pkt(vvs, skb); 410 411 ret = t_ops->send_pkt(skb); 412 if (ret < 0) 413 break; 414 415 /* Both virtio and vhost 'send_pkt()' returns 'skb_len', 416 * but for reliability use 'ret' instead of 'skb_len'. 417 * Also if partial send happens (e.g. 'ret' != 'skb_len') 418 * somehow, we break this loop, but account such returned 419 * value in 'virtio_transport_put_credit()'. 420 */ 421 rest_len -= ret; 422 423 if (WARN_ONCE(ret != skb_len, 424 "'send_pkt()' returns %i, but %zu expected\n", 425 ret, skb_len)) 426 break; 427 } while (rest_len); 428 429 virtio_transport_put_credit(vvs, rest_len); 430 431 /* Return number of bytes, if any data has been sent. */ 432 if (rest_len != pkt_len) 433 ret = pkt_len - rest_len; 434 435 return ret; 436 } 437 438 static bool virtio_transport_inc_rx_pkt(struct virtio_vsock_sock *vvs, 439 u32 len) 440 { 441 if (vvs->rx_bytes + len > vvs->buf_alloc) 442 return false; 443 444 vvs->rx_bytes += len; 445 return true; 446 } 447 448 static void virtio_transport_dec_rx_pkt(struct virtio_vsock_sock *vvs, 449 u32 len) 450 { 451 vvs->rx_bytes -= len; 452 vvs->fwd_cnt += len; 453 } 454 455 void virtio_transport_inc_tx_pkt(struct virtio_vsock_sock *vvs, struct sk_buff *skb) 456 { 457 struct virtio_vsock_hdr *hdr = virtio_vsock_hdr(skb); 458 459 spin_lock_bh(&vvs->rx_lock); 460 vvs->last_fwd_cnt = vvs->fwd_cnt; 461 hdr->fwd_cnt = cpu_to_le32(vvs->fwd_cnt); 462 hdr->buf_alloc = cpu_to_le32(vvs->buf_alloc); 463 spin_unlock_bh(&vvs->rx_lock); 464 } 465 EXPORT_SYMBOL_GPL(virtio_transport_inc_tx_pkt); 466 467 void virtio_transport_consume_skb_sent(struct sk_buff *skb, bool consume) 468 { 469 struct sock *s = skb->sk; 470 471 if (s && skb->len) { 472 struct vsock_sock *vs = vsock_sk(s); 473 struct virtio_vsock_sock *vvs; 474 475 vvs = vs->trans; 476 477 spin_lock_bh(&vvs->tx_lock); 478 vvs->bytes_unsent -= skb->len; 479 spin_unlock_bh(&vvs->tx_lock); 480 } 481 482 if (consume) 483 consume_skb(skb); 484 } 485 EXPORT_SYMBOL_GPL(virtio_transport_consume_skb_sent); 486 487 u32 virtio_transport_get_credit(struct virtio_vsock_sock *vvs, u32 credit) 488 { 489 u32 ret; 490 491 if (!credit) 492 return 0; 493 494 spin_lock_bh(&vvs->tx_lock); 495 ret = vvs->peer_buf_alloc - (vvs->tx_cnt - vvs->peer_fwd_cnt); 496 if (ret > credit) 497 ret = credit; 498 vvs->tx_cnt += ret; 499 vvs->bytes_unsent += ret; 500 spin_unlock_bh(&vvs->tx_lock); 501 502 return ret; 503 } 504 EXPORT_SYMBOL_GPL(virtio_transport_get_credit); 505 506 void virtio_transport_put_credit(struct virtio_vsock_sock *vvs, u32 credit) 507 { 508 if (!credit) 509 return; 510 511 spin_lock_bh(&vvs->tx_lock); 512 vvs->tx_cnt -= credit; 513 vvs->bytes_unsent -= credit; 514 spin_unlock_bh(&vvs->tx_lock); 515 } 516 EXPORT_SYMBOL_GPL(virtio_transport_put_credit); 517 518 static int virtio_transport_send_credit_update(struct vsock_sock *vsk) 519 { 520 struct virtio_vsock_pkt_info info = { 521 .op = VIRTIO_VSOCK_OP_CREDIT_UPDATE, 522 .vsk = vsk, 523 }; 524 525 return virtio_transport_send_pkt_info(vsk, &info); 526 } 527 528 static ssize_t 529 virtio_transport_stream_do_peek(struct vsock_sock *vsk, 530 struct msghdr *msg, 531 size_t len) 532 { 533 struct virtio_vsock_sock *vvs = vsk->trans; 534 struct sk_buff *skb; 535 size_t total = 0; 536 int err; 537 538 spin_lock_bh(&vvs->rx_lock); 539 540 skb_queue_walk(&vvs->rx_queue, skb) { 541 size_t bytes; 542 543 bytes = len - total; 544 if (bytes > skb->len) 545 bytes = skb->len; 546 547 spin_unlock_bh(&vvs->rx_lock); 548 549 /* sk_lock is held by caller so no one else can dequeue. 550 * Unlock rx_lock since skb_copy_datagram_iter() may sleep. 551 */ 552 err = skb_copy_datagram_iter(skb, VIRTIO_VSOCK_SKB_CB(skb)->offset, 553 &msg->msg_iter, bytes); 554 if (err) 555 goto out; 556 557 total += bytes; 558 559 spin_lock_bh(&vvs->rx_lock); 560 561 if (total == len) 562 break; 563 } 564 565 spin_unlock_bh(&vvs->rx_lock); 566 567 return total; 568 569 out: 570 if (total) 571 err = total; 572 return err; 573 } 574 575 static ssize_t 576 virtio_transport_stream_do_dequeue(struct vsock_sock *vsk, 577 struct msghdr *msg, 578 size_t len) 579 { 580 struct virtio_vsock_sock *vvs = vsk->trans; 581 size_t bytes, total = 0; 582 struct sk_buff *skb; 583 u32 fwd_cnt_delta; 584 bool low_rx_bytes; 585 int err = -EFAULT; 586 u32 free_space; 587 588 spin_lock_bh(&vvs->rx_lock); 589 590 if (WARN_ONCE(skb_queue_empty(&vvs->rx_queue) && vvs->rx_bytes, 591 "rx_queue is empty, but rx_bytes is non-zero\n")) { 592 spin_unlock_bh(&vvs->rx_lock); 593 return err; 594 } 595 596 while (total < len && !skb_queue_empty(&vvs->rx_queue)) { 597 skb = skb_peek(&vvs->rx_queue); 598 599 bytes = min_t(size_t, len - total, 600 skb->len - VIRTIO_VSOCK_SKB_CB(skb)->offset); 601 602 /* sk_lock is held by caller so no one else can dequeue. 603 * Unlock rx_lock since skb_copy_datagram_iter() may sleep. 604 */ 605 spin_unlock_bh(&vvs->rx_lock); 606 607 err = skb_copy_datagram_iter(skb, 608 VIRTIO_VSOCK_SKB_CB(skb)->offset, 609 &msg->msg_iter, bytes); 610 if (err) 611 goto out; 612 613 spin_lock_bh(&vvs->rx_lock); 614 615 total += bytes; 616 617 VIRTIO_VSOCK_SKB_CB(skb)->offset += bytes; 618 619 if (skb->len == VIRTIO_VSOCK_SKB_CB(skb)->offset) { 620 u32 pkt_len = le32_to_cpu(virtio_vsock_hdr(skb)->len); 621 622 virtio_transport_dec_rx_pkt(vvs, pkt_len); 623 __skb_unlink(skb, &vvs->rx_queue); 624 consume_skb(skb); 625 } 626 } 627 628 fwd_cnt_delta = vvs->fwd_cnt - vvs->last_fwd_cnt; 629 free_space = vvs->buf_alloc - fwd_cnt_delta; 630 low_rx_bytes = (vvs->rx_bytes < 631 sock_rcvlowat(sk_vsock(vsk), 0, INT_MAX)); 632 633 spin_unlock_bh(&vvs->rx_lock); 634 635 /* To reduce the number of credit update messages, 636 * don't update credits as long as lots of space is available. 637 * Note: the limit chosen here is arbitrary. Setting the limit 638 * too high causes extra messages. Too low causes transmitter 639 * stalls. As stalls are in theory more expensive than extra 640 * messages, we set the limit to a high value. TODO: experiment 641 * with different values. Also send credit update message when 642 * number of bytes in rx queue is not enough to wake up reader. 643 */ 644 if (fwd_cnt_delta && 645 (free_space < VIRTIO_VSOCK_MAX_PKT_BUF_SIZE || low_rx_bytes)) 646 virtio_transport_send_credit_update(vsk); 647 648 return total; 649 650 out: 651 if (total) 652 err = total; 653 return err; 654 } 655 656 static ssize_t 657 virtio_transport_seqpacket_do_peek(struct vsock_sock *vsk, 658 struct msghdr *msg) 659 { 660 struct virtio_vsock_sock *vvs = vsk->trans; 661 struct sk_buff *skb; 662 size_t total, len; 663 664 spin_lock_bh(&vvs->rx_lock); 665 666 if (!vvs->msg_count) { 667 spin_unlock_bh(&vvs->rx_lock); 668 return 0; 669 } 670 671 total = 0; 672 len = msg_data_left(msg); 673 674 skb_queue_walk(&vvs->rx_queue, skb) { 675 struct virtio_vsock_hdr *hdr; 676 677 if (total < len) { 678 size_t bytes; 679 int err; 680 681 bytes = len - total; 682 if (bytes > skb->len) 683 bytes = skb->len; 684 685 spin_unlock_bh(&vvs->rx_lock); 686 687 /* sk_lock is held by caller so no one else can dequeue. 688 * Unlock rx_lock since skb_copy_datagram_iter() may sleep. 689 */ 690 err = skb_copy_datagram_iter(skb, VIRTIO_VSOCK_SKB_CB(skb)->offset, 691 &msg->msg_iter, bytes); 692 if (err) 693 return err; 694 695 spin_lock_bh(&vvs->rx_lock); 696 } 697 698 total += skb->len; 699 hdr = virtio_vsock_hdr(skb); 700 701 if (le32_to_cpu(hdr->flags) & VIRTIO_VSOCK_SEQ_EOM) { 702 if (le32_to_cpu(hdr->flags) & VIRTIO_VSOCK_SEQ_EOR) 703 msg->msg_flags |= MSG_EOR; 704 705 break; 706 } 707 } 708 709 spin_unlock_bh(&vvs->rx_lock); 710 711 return total; 712 } 713 714 static int virtio_transport_seqpacket_do_dequeue(struct vsock_sock *vsk, 715 struct msghdr *msg, 716 int flags) 717 { 718 struct virtio_vsock_sock *vvs = vsk->trans; 719 int dequeued_len = 0; 720 size_t user_buf_len = msg_data_left(msg); 721 bool msg_ready = false; 722 struct sk_buff *skb; 723 724 spin_lock_bh(&vvs->rx_lock); 725 726 if (vvs->msg_count == 0) { 727 spin_unlock_bh(&vvs->rx_lock); 728 return 0; 729 } 730 731 while (!msg_ready) { 732 struct virtio_vsock_hdr *hdr; 733 size_t pkt_len; 734 735 skb = __skb_dequeue(&vvs->rx_queue); 736 if (!skb) 737 break; 738 hdr = virtio_vsock_hdr(skb); 739 pkt_len = (size_t)le32_to_cpu(hdr->len); 740 741 if (dequeued_len >= 0) { 742 size_t bytes_to_copy; 743 744 bytes_to_copy = min(user_buf_len, pkt_len); 745 746 if (bytes_to_copy) { 747 int err; 748 749 /* sk_lock is held by caller so no one else can dequeue. 750 * Unlock rx_lock since skb_copy_datagram_iter() may sleep. 751 */ 752 spin_unlock_bh(&vvs->rx_lock); 753 754 err = skb_copy_datagram_iter(skb, 0, 755 &msg->msg_iter, 756 bytes_to_copy); 757 if (err) { 758 /* Copy of message failed. Rest of 759 * fragments will be freed without copy. 760 */ 761 dequeued_len = err; 762 } else { 763 user_buf_len -= bytes_to_copy; 764 } 765 766 spin_lock_bh(&vvs->rx_lock); 767 } 768 769 if (dequeued_len >= 0) 770 dequeued_len += pkt_len; 771 } 772 773 if (le32_to_cpu(hdr->flags) & VIRTIO_VSOCK_SEQ_EOM) { 774 msg_ready = true; 775 vvs->msg_count--; 776 777 if (le32_to_cpu(hdr->flags) & VIRTIO_VSOCK_SEQ_EOR) 778 msg->msg_flags |= MSG_EOR; 779 } 780 781 virtio_transport_dec_rx_pkt(vvs, pkt_len); 782 kfree_skb(skb); 783 } 784 785 spin_unlock_bh(&vvs->rx_lock); 786 787 virtio_transport_send_credit_update(vsk); 788 789 return dequeued_len; 790 } 791 792 ssize_t 793 virtio_transport_stream_dequeue(struct vsock_sock *vsk, 794 struct msghdr *msg, 795 size_t len, int flags) 796 { 797 if (flags & MSG_PEEK) 798 return virtio_transport_stream_do_peek(vsk, msg, len); 799 else 800 return virtio_transport_stream_do_dequeue(vsk, msg, len); 801 } 802 EXPORT_SYMBOL_GPL(virtio_transport_stream_dequeue); 803 804 ssize_t 805 virtio_transport_seqpacket_dequeue(struct vsock_sock *vsk, 806 struct msghdr *msg, 807 int flags) 808 { 809 if (flags & MSG_PEEK) 810 return virtio_transport_seqpacket_do_peek(vsk, msg); 811 else 812 return virtio_transport_seqpacket_do_dequeue(vsk, msg, flags); 813 } 814 EXPORT_SYMBOL_GPL(virtio_transport_seqpacket_dequeue); 815 816 int 817 virtio_transport_seqpacket_enqueue(struct vsock_sock *vsk, 818 struct msghdr *msg, 819 size_t len) 820 { 821 struct virtio_vsock_sock *vvs = vsk->trans; 822 823 spin_lock_bh(&vvs->tx_lock); 824 825 if (len > vvs->peer_buf_alloc) { 826 spin_unlock_bh(&vvs->tx_lock); 827 return -EMSGSIZE; 828 } 829 830 spin_unlock_bh(&vvs->tx_lock); 831 832 return virtio_transport_stream_enqueue(vsk, msg, len); 833 } 834 EXPORT_SYMBOL_GPL(virtio_transport_seqpacket_enqueue); 835 836 int 837 virtio_transport_dgram_dequeue(struct vsock_sock *vsk, 838 struct msghdr *msg, 839 size_t len, int flags) 840 { 841 return -EOPNOTSUPP; 842 } 843 EXPORT_SYMBOL_GPL(virtio_transport_dgram_dequeue); 844 845 s64 virtio_transport_stream_has_data(struct vsock_sock *vsk) 846 { 847 struct virtio_vsock_sock *vvs = vsk->trans; 848 s64 bytes; 849 850 spin_lock_bh(&vvs->rx_lock); 851 bytes = vvs->rx_bytes; 852 spin_unlock_bh(&vvs->rx_lock); 853 854 return bytes; 855 } 856 EXPORT_SYMBOL_GPL(virtio_transport_stream_has_data); 857 858 u32 virtio_transport_seqpacket_has_data(struct vsock_sock *vsk) 859 { 860 struct virtio_vsock_sock *vvs = vsk->trans; 861 u32 msg_count; 862 863 spin_lock_bh(&vvs->rx_lock); 864 msg_count = vvs->msg_count; 865 spin_unlock_bh(&vvs->rx_lock); 866 867 return msg_count; 868 } 869 EXPORT_SYMBOL_GPL(virtio_transport_seqpacket_has_data); 870 871 static s64 virtio_transport_has_space(struct vsock_sock *vsk) 872 { 873 struct virtio_vsock_sock *vvs = vsk->trans; 874 s64 bytes; 875 876 bytes = (s64)vvs->peer_buf_alloc - (vvs->tx_cnt - vvs->peer_fwd_cnt); 877 if (bytes < 0) 878 bytes = 0; 879 880 return bytes; 881 } 882 883 s64 virtio_transport_stream_has_space(struct vsock_sock *vsk) 884 { 885 struct virtio_vsock_sock *vvs = vsk->trans; 886 s64 bytes; 887 888 spin_lock_bh(&vvs->tx_lock); 889 bytes = virtio_transport_has_space(vsk); 890 spin_unlock_bh(&vvs->tx_lock); 891 892 return bytes; 893 } 894 EXPORT_SYMBOL_GPL(virtio_transport_stream_has_space); 895 896 int virtio_transport_do_socket_init(struct vsock_sock *vsk, 897 struct vsock_sock *psk) 898 { 899 struct virtio_vsock_sock *vvs; 900 901 vvs = kzalloc(sizeof(*vvs), GFP_KERNEL); 902 if (!vvs) 903 return -ENOMEM; 904 905 vsk->trans = vvs; 906 vvs->vsk = vsk; 907 if (psk && psk->trans) { 908 struct virtio_vsock_sock *ptrans = psk->trans; 909 910 vvs->peer_buf_alloc = ptrans->peer_buf_alloc; 911 } 912 913 if (vsk->buffer_size > VIRTIO_VSOCK_MAX_BUF_SIZE) 914 vsk->buffer_size = VIRTIO_VSOCK_MAX_BUF_SIZE; 915 916 vvs->buf_alloc = vsk->buffer_size; 917 918 spin_lock_init(&vvs->rx_lock); 919 spin_lock_init(&vvs->tx_lock); 920 skb_queue_head_init(&vvs->rx_queue); 921 922 return 0; 923 } 924 EXPORT_SYMBOL_GPL(virtio_transport_do_socket_init); 925 926 /* sk_lock held by the caller */ 927 void virtio_transport_notify_buffer_size(struct vsock_sock *vsk, u64 *val) 928 { 929 struct virtio_vsock_sock *vvs = vsk->trans; 930 931 if (*val > VIRTIO_VSOCK_MAX_BUF_SIZE) 932 *val = VIRTIO_VSOCK_MAX_BUF_SIZE; 933 934 vvs->buf_alloc = *val; 935 936 virtio_transport_send_credit_update(vsk); 937 } 938 EXPORT_SYMBOL_GPL(virtio_transport_notify_buffer_size); 939 940 int 941 virtio_transport_notify_poll_in(struct vsock_sock *vsk, 942 size_t target, 943 bool *data_ready_now) 944 { 945 *data_ready_now = vsock_stream_has_data(vsk) >= target; 946 947 return 0; 948 } 949 EXPORT_SYMBOL_GPL(virtio_transport_notify_poll_in); 950 951 int 952 virtio_transport_notify_poll_out(struct vsock_sock *vsk, 953 size_t target, 954 bool *space_avail_now) 955 { 956 s64 free_space; 957 958 free_space = vsock_stream_has_space(vsk); 959 if (free_space > 0) 960 *space_avail_now = true; 961 else if (free_space == 0) 962 *space_avail_now = false; 963 964 return 0; 965 } 966 EXPORT_SYMBOL_GPL(virtio_transport_notify_poll_out); 967 968 int virtio_transport_notify_recv_init(struct vsock_sock *vsk, 969 size_t target, struct vsock_transport_recv_notify_data *data) 970 { 971 return 0; 972 } 973 EXPORT_SYMBOL_GPL(virtio_transport_notify_recv_init); 974 975 int virtio_transport_notify_recv_pre_block(struct vsock_sock *vsk, 976 size_t target, struct vsock_transport_recv_notify_data *data) 977 { 978 return 0; 979 } 980 EXPORT_SYMBOL_GPL(virtio_transport_notify_recv_pre_block); 981 982 int virtio_transport_notify_recv_pre_dequeue(struct vsock_sock *vsk, 983 size_t target, struct vsock_transport_recv_notify_data *data) 984 { 985 return 0; 986 } 987 EXPORT_SYMBOL_GPL(virtio_transport_notify_recv_pre_dequeue); 988 989 int virtio_transport_notify_recv_post_dequeue(struct vsock_sock *vsk, 990 size_t target, ssize_t copied, bool data_read, 991 struct vsock_transport_recv_notify_data *data) 992 { 993 return 0; 994 } 995 EXPORT_SYMBOL_GPL(virtio_transport_notify_recv_post_dequeue); 996 997 int virtio_transport_notify_send_init(struct vsock_sock *vsk, 998 struct vsock_transport_send_notify_data *data) 999 { 1000 return 0; 1001 } 1002 EXPORT_SYMBOL_GPL(virtio_transport_notify_send_init); 1003 1004 int virtio_transport_notify_send_pre_block(struct vsock_sock *vsk, 1005 struct vsock_transport_send_notify_data *data) 1006 { 1007 return 0; 1008 } 1009 EXPORT_SYMBOL_GPL(virtio_transport_notify_send_pre_block); 1010 1011 int virtio_transport_notify_send_pre_enqueue(struct vsock_sock *vsk, 1012 struct vsock_transport_send_notify_data *data) 1013 { 1014 return 0; 1015 } 1016 EXPORT_SYMBOL_GPL(virtio_transport_notify_send_pre_enqueue); 1017 1018 int virtio_transport_notify_send_post_enqueue(struct vsock_sock *vsk, 1019 ssize_t written, struct vsock_transport_send_notify_data *data) 1020 { 1021 return 0; 1022 } 1023 EXPORT_SYMBOL_GPL(virtio_transport_notify_send_post_enqueue); 1024 1025 u64 virtio_transport_stream_rcvhiwat(struct vsock_sock *vsk) 1026 { 1027 return vsk->buffer_size; 1028 } 1029 EXPORT_SYMBOL_GPL(virtio_transport_stream_rcvhiwat); 1030 1031 bool virtio_transport_stream_is_active(struct vsock_sock *vsk) 1032 { 1033 return true; 1034 } 1035 EXPORT_SYMBOL_GPL(virtio_transport_stream_is_active); 1036 1037 bool virtio_transport_stream_allow(u32 cid, u32 port) 1038 { 1039 return true; 1040 } 1041 EXPORT_SYMBOL_GPL(virtio_transport_stream_allow); 1042 1043 int virtio_transport_dgram_bind(struct vsock_sock *vsk, 1044 struct sockaddr_vm *addr) 1045 { 1046 return -EOPNOTSUPP; 1047 } 1048 EXPORT_SYMBOL_GPL(virtio_transport_dgram_bind); 1049 1050 bool virtio_transport_dgram_allow(u32 cid, u32 port) 1051 { 1052 return false; 1053 } 1054 EXPORT_SYMBOL_GPL(virtio_transport_dgram_allow); 1055 1056 int virtio_transport_connect(struct vsock_sock *vsk) 1057 { 1058 struct virtio_vsock_pkt_info info = { 1059 .op = VIRTIO_VSOCK_OP_REQUEST, 1060 .vsk = vsk, 1061 }; 1062 1063 return virtio_transport_send_pkt_info(vsk, &info); 1064 } 1065 EXPORT_SYMBOL_GPL(virtio_transport_connect); 1066 1067 int virtio_transport_shutdown(struct vsock_sock *vsk, int mode) 1068 { 1069 struct virtio_vsock_pkt_info info = { 1070 .op = VIRTIO_VSOCK_OP_SHUTDOWN, 1071 .flags = (mode & RCV_SHUTDOWN ? 1072 VIRTIO_VSOCK_SHUTDOWN_RCV : 0) | 1073 (mode & SEND_SHUTDOWN ? 1074 VIRTIO_VSOCK_SHUTDOWN_SEND : 0), 1075 .vsk = vsk, 1076 }; 1077 1078 return virtio_transport_send_pkt_info(vsk, &info); 1079 } 1080 EXPORT_SYMBOL_GPL(virtio_transport_shutdown); 1081 1082 int 1083 virtio_transport_dgram_enqueue(struct vsock_sock *vsk, 1084 struct sockaddr_vm *remote_addr, 1085 struct msghdr *msg, 1086 size_t dgram_len) 1087 { 1088 return -EOPNOTSUPP; 1089 } 1090 EXPORT_SYMBOL_GPL(virtio_transport_dgram_enqueue); 1091 1092 ssize_t 1093 virtio_transport_stream_enqueue(struct vsock_sock *vsk, 1094 struct msghdr *msg, 1095 size_t len) 1096 { 1097 struct virtio_vsock_pkt_info info = { 1098 .op = VIRTIO_VSOCK_OP_RW, 1099 .msg = msg, 1100 .pkt_len = len, 1101 .vsk = vsk, 1102 }; 1103 1104 return virtio_transport_send_pkt_info(vsk, &info); 1105 } 1106 EXPORT_SYMBOL_GPL(virtio_transport_stream_enqueue); 1107 1108 void virtio_transport_destruct(struct vsock_sock *vsk) 1109 { 1110 struct virtio_vsock_sock *vvs = vsk->trans; 1111 1112 kfree(vvs); 1113 } 1114 EXPORT_SYMBOL_GPL(virtio_transport_destruct); 1115 1116 ssize_t virtio_transport_unsent_bytes(struct vsock_sock *vsk) 1117 { 1118 struct virtio_vsock_sock *vvs = vsk->trans; 1119 size_t ret; 1120 1121 spin_lock_bh(&vvs->tx_lock); 1122 ret = vvs->bytes_unsent; 1123 spin_unlock_bh(&vvs->tx_lock); 1124 1125 return ret; 1126 } 1127 EXPORT_SYMBOL_GPL(virtio_transport_unsent_bytes); 1128 1129 static int virtio_transport_reset(struct vsock_sock *vsk, 1130 struct sk_buff *skb) 1131 { 1132 struct virtio_vsock_pkt_info info = { 1133 .op = VIRTIO_VSOCK_OP_RST, 1134 .reply = !!skb, 1135 .vsk = vsk, 1136 }; 1137 1138 /* Send RST only if the original pkt is not a RST pkt */ 1139 if (skb && le16_to_cpu(virtio_vsock_hdr(skb)->op) == VIRTIO_VSOCK_OP_RST) 1140 return 0; 1141 1142 return virtio_transport_send_pkt_info(vsk, &info); 1143 } 1144 1145 /* Normally packets are associated with a socket. There may be no socket if an 1146 * attempt was made to connect to a socket that does not exist. 1147 */ 1148 static int virtio_transport_reset_no_sock(const struct virtio_transport *t, 1149 struct sk_buff *skb) 1150 { 1151 struct virtio_vsock_hdr *hdr = virtio_vsock_hdr(skb); 1152 struct virtio_vsock_pkt_info info = { 1153 .op = VIRTIO_VSOCK_OP_RST, 1154 .type = le16_to_cpu(hdr->type), 1155 .reply = true, 1156 }; 1157 struct sk_buff *reply; 1158 1159 /* Send RST only if the original pkt is not a RST pkt */ 1160 if (le16_to_cpu(hdr->op) == VIRTIO_VSOCK_OP_RST) 1161 return 0; 1162 1163 if (!t) 1164 return -ENOTCONN; 1165 1166 reply = virtio_transport_alloc_skb(&info, 0, false, 1167 le64_to_cpu(hdr->dst_cid), 1168 le32_to_cpu(hdr->dst_port), 1169 le64_to_cpu(hdr->src_cid), 1170 le32_to_cpu(hdr->src_port)); 1171 if (!reply) 1172 return -ENOMEM; 1173 1174 return t->send_pkt(reply); 1175 } 1176 1177 /* This function should be called with sk_lock held and SOCK_DONE set */ 1178 static void virtio_transport_remove_sock(struct vsock_sock *vsk) 1179 { 1180 struct virtio_vsock_sock *vvs = vsk->trans; 1181 1182 /* We don't need to take rx_lock, as the socket is closing and we are 1183 * removing it. 1184 */ 1185 __skb_queue_purge(&vvs->rx_queue); 1186 vsock_remove_sock(vsk); 1187 } 1188 1189 static void virtio_transport_wait_close(struct sock *sk, long timeout) 1190 { 1191 if (timeout) { 1192 DEFINE_WAIT_FUNC(wait, woken_wake_function); 1193 1194 add_wait_queue(sk_sleep(sk), &wait); 1195 1196 do { 1197 if (sk_wait_event(sk, &timeout, 1198 sock_flag(sk, SOCK_DONE), &wait)) 1199 break; 1200 } while (!signal_pending(current) && timeout); 1201 1202 remove_wait_queue(sk_sleep(sk), &wait); 1203 } 1204 } 1205 1206 static void virtio_transport_do_close(struct vsock_sock *vsk, 1207 bool cancel_timeout) 1208 { 1209 struct sock *sk = sk_vsock(vsk); 1210 1211 sock_set_flag(sk, SOCK_DONE); 1212 vsk->peer_shutdown = SHUTDOWN_MASK; 1213 if (vsock_stream_has_data(vsk) <= 0) 1214 sk->sk_state = TCP_CLOSING; 1215 sk->sk_state_change(sk); 1216 1217 if (vsk->close_work_scheduled && 1218 (!cancel_timeout || cancel_delayed_work(&vsk->close_work))) { 1219 vsk->close_work_scheduled = false; 1220 1221 virtio_transport_remove_sock(vsk); 1222 1223 /* Release refcnt obtained when we scheduled the timeout */ 1224 sock_put(sk); 1225 } 1226 } 1227 1228 static void virtio_transport_close_timeout(struct work_struct *work) 1229 { 1230 struct vsock_sock *vsk = 1231 container_of(work, struct vsock_sock, close_work.work); 1232 struct sock *sk = sk_vsock(vsk); 1233 1234 sock_hold(sk); 1235 lock_sock(sk); 1236 1237 if (!sock_flag(sk, SOCK_DONE)) { 1238 (void)virtio_transport_reset(vsk, NULL); 1239 1240 virtio_transport_do_close(vsk, false); 1241 } 1242 1243 vsk->close_work_scheduled = false; 1244 1245 release_sock(sk); 1246 sock_put(sk); 1247 } 1248 1249 /* User context, vsk->sk is locked */ 1250 static bool virtio_transport_close(struct vsock_sock *vsk) 1251 { 1252 struct sock *sk = &vsk->sk; 1253 1254 if (!(sk->sk_state == TCP_ESTABLISHED || 1255 sk->sk_state == TCP_CLOSING)) 1256 return true; 1257 1258 /* Already received SHUTDOWN from peer, reply with RST */ 1259 if ((vsk->peer_shutdown & SHUTDOWN_MASK) == SHUTDOWN_MASK) { 1260 (void)virtio_transport_reset(vsk, NULL); 1261 return true; 1262 } 1263 1264 if ((sk->sk_shutdown & SHUTDOWN_MASK) != SHUTDOWN_MASK) 1265 (void)virtio_transport_shutdown(vsk, SHUTDOWN_MASK); 1266 1267 if (sock_flag(sk, SOCK_LINGER) && !(current->flags & PF_EXITING)) 1268 virtio_transport_wait_close(sk, sk->sk_lingertime); 1269 1270 if (sock_flag(sk, SOCK_DONE)) { 1271 return true; 1272 } 1273 1274 sock_hold(sk); 1275 INIT_DELAYED_WORK(&vsk->close_work, 1276 virtio_transport_close_timeout); 1277 vsk->close_work_scheduled = true; 1278 schedule_delayed_work(&vsk->close_work, VSOCK_CLOSE_TIMEOUT); 1279 return false; 1280 } 1281 1282 void virtio_transport_release(struct vsock_sock *vsk) 1283 { 1284 struct sock *sk = &vsk->sk; 1285 bool remove_sock = true; 1286 1287 if (sk->sk_type == SOCK_STREAM || sk->sk_type == SOCK_SEQPACKET) 1288 remove_sock = virtio_transport_close(vsk); 1289 1290 if (remove_sock) { 1291 sock_set_flag(sk, SOCK_DONE); 1292 virtio_transport_remove_sock(vsk); 1293 } 1294 } 1295 EXPORT_SYMBOL_GPL(virtio_transport_release); 1296 1297 static int 1298 virtio_transport_recv_connecting(struct sock *sk, 1299 struct sk_buff *skb) 1300 { 1301 struct virtio_vsock_hdr *hdr = virtio_vsock_hdr(skb); 1302 struct vsock_sock *vsk = vsock_sk(sk); 1303 int skerr; 1304 int err; 1305 1306 switch (le16_to_cpu(hdr->op)) { 1307 case VIRTIO_VSOCK_OP_RESPONSE: 1308 sk->sk_state = TCP_ESTABLISHED; 1309 sk->sk_socket->state = SS_CONNECTED; 1310 vsock_insert_connected(vsk); 1311 sk->sk_state_change(sk); 1312 break; 1313 case VIRTIO_VSOCK_OP_INVALID: 1314 break; 1315 case VIRTIO_VSOCK_OP_RST: 1316 skerr = ECONNRESET; 1317 err = 0; 1318 goto destroy; 1319 default: 1320 skerr = EPROTO; 1321 err = -EINVAL; 1322 goto destroy; 1323 } 1324 return 0; 1325 1326 destroy: 1327 virtio_transport_reset(vsk, skb); 1328 sk->sk_state = TCP_CLOSE; 1329 sk->sk_err = skerr; 1330 sk_error_report(sk); 1331 return err; 1332 } 1333 1334 static void 1335 virtio_transport_recv_enqueue(struct vsock_sock *vsk, 1336 struct sk_buff *skb) 1337 { 1338 struct virtio_vsock_sock *vvs = vsk->trans; 1339 bool can_enqueue, free_pkt = false; 1340 struct virtio_vsock_hdr *hdr; 1341 u32 len; 1342 1343 hdr = virtio_vsock_hdr(skb); 1344 len = le32_to_cpu(hdr->len); 1345 1346 spin_lock_bh(&vvs->rx_lock); 1347 1348 can_enqueue = virtio_transport_inc_rx_pkt(vvs, len); 1349 if (!can_enqueue) { 1350 free_pkt = true; 1351 goto out; 1352 } 1353 1354 if (le32_to_cpu(hdr->flags) & VIRTIO_VSOCK_SEQ_EOM) 1355 vvs->msg_count++; 1356 1357 /* Try to copy small packets into the buffer of last packet queued, 1358 * to avoid wasting memory queueing the entire buffer with a small 1359 * payload. 1360 */ 1361 if (len <= GOOD_COPY_LEN && !skb_queue_empty(&vvs->rx_queue)) { 1362 struct virtio_vsock_hdr *last_hdr; 1363 struct sk_buff *last_skb; 1364 1365 last_skb = skb_peek_tail(&vvs->rx_queue); 1366 last_hdr = virtio_vsock_hdr(last_skb); 1367 1368 /* If there is space in the last packet queued, we copy the 1369 * new packet in its buffer. We avoid this if the last packet 1370 * queued has VIRTIO_VSOCK_SEQ_EOM set, because this is 1371 * delimiter of SEQPACKET message, so 'pkt' is the first packet 1372 * of a new message. 1373 */ 1374 if (skb->len < skb_tailroom(last_skb) && 1375 !(le32_to_cpu(last_hdr->flags) & VIRTIO_VSOCK_SEQ_EOM)) { 1376 memcpy(skb_put(last_skb, skb->len), skb->data, skb->len); 1377 free_pkt = true; 1378 last_hdr->flags |= hdr->flags; 1379 le32_add_cpu(&last_hdr->len, len); 1380 goto out; 1381 } 1382 } 1383 1384 __skb_queue_tail(&vvs->rx_queue, skb); 1385 1386 out: 1387 spin_unlock_bh(&vvs->rx_lock); 1388 if (free_pkt) 1389 kfree_skb(skb); 1390 } 1391 1392 static int 1393 virtio_transport_recv_connected(struct sock *sk, 1394 struct sk_buff *skb) 1395 { 1396 struct virtio_vsock_hdr *hdr = virtio_vsock_hdr(skb); 1397 struct vsock_sock *vsk = vsock_sk(sk); 1398 int err = 0; 1399 1400 switch (le16_to_cpu(hdr->op)) { 1401 case VIRTIO_VSOCK_OP_RW: 1402 virtio_transport_recv_enqueue(vsk, skb); 1403 vsock_data_ready(sk); 1404 return err; 1405 case VIRTIO_VSOCK_OP_CREDIT_REQUEST: 1406 virtio_transport_send_credit_update(vsk); 1407 break; 1408 case VIRTIO_VSOCK_OP_CREDIT_UPDATE: 1409 sk->sk_write_space(sk); 1410 break; 1411 case VIRTIO_VSOCK_OP_SHUTDOWN: 1412 if (le32_to_cpu(hdr->flags) & VIRTIO_VSOCK_SHUTDOWN_RCV) 1413 vsk->peer_shutdown |= RCV_SHUTDOWN; 1414 if (le32_to_cpu(hdr->flags) & VIRTIO_VSOCK_SHUTDOWN_SEND) 1415 vsk->peer_shutdown |= SEND_SHUTDOWN; 1416 if (vsk->peer_shutdown == SHUTDOWN_MASK) { 1417 if (vsock_stream_has_data(vsk) <= 0 && !sock_flag(sk, SOCK_DONE)) { 1418 (void)virtio_transport_reset(vsk, NULL); 1419 virtio_transport_do_close(vsk, true); 1420 } 1421 /* Remove this socket anyway because the remote peer sent 1422 * the shutdown. This way a new connection will succeed 1423 * if the remote peer uses the same source port, 1424 * even if the old socket is still unreleased, but now disconnected. 1425 */ 1426 vsock_remove_sock(vsk); 1427 } 1428 if (le32_to_cpu(virtio_vsock_hdr(skb)->flags)) 1429 sk->sk_state_change(sk); 1430 break; 1431 case VIRTIO_VSOCK_OP_RST: 1432 virtio_transport_do_close(vsk, true); 1433 break; 1434 default: 1435 err = -EINVAL; 1436 break; 1437 } 1438 1439 kfree_skb(skb); 1440 return err; 1441 } 1442 1443 static void 1444 virtio_transport_recv_disconnecting(struct sock *sk, 1445 struct sk_buff *skb) 1446 { 1447 struct virtio_vsock_hdr *hdr = virtio_vsock_hdr(skb); 1448 struct vsock_sock *vsk = vsock_sk(sk); 1449 1450 if (le16_to_cpu(hdr->op) == VIRTIO_VSOCK_OP_RST) 1451 virtio_transport_do_close(vsk, true); 1452 } 1453 1454 static int 1455 virtio_transport_send_response(struct vsock_sock *vsk, 1456 struct sk_buff *skb) 1457 { 1458 struct virtio_vsock_hdr *hdr = virtio_vsock_hdr(skb); 1459 struct virtio_vsock_pkt_info info = { 1460 .op = VIRTIO_VSOCK_OP_RESPONSE, 1461 .remote_cid = le64_to_cpu(hdr->src_cid), 1462 .remote_port = le32_to_cpu(hdr->src_port), 1463 .reply = true, 1464 .vsk = vsk, 1465 }; 1466 1467 return virtio_transport_send_pkt_info(vsk, &info); 1468 } 1469 1470 static bool virtio_transport_space_update(struct sock *sk, 1471 struct sk_buff *skb) 1472 { 1473 struct virtio_vsock_hdr *hdr = virtio_vsock_hdr(skb); 1474 struct vsock_sock *vsk = vsock_sk(sk); 1475 struct virtio_vsock_sock *vvs = vsk->trans; 1476 bool space_available; 1477 1478 /* Listener sockets are not associated with any transport, so we are 1479 * not able to take the state to see if there is space available in the 1480 * remote peer, but since they are only used to receive requests, we 1481 * can assume that there is always space available in the other peer. 1482 */ 1483 if (!vvs) 1484 return true; 1485 1486 /* buf_alloc and fwd_cnt is always included in the hdr */ 1487 spin_lock_bh(&vvs->tx_lock); 1488 vvs->peer_buf_alloc = le32_to_cpu(hdr->buf_alloc); 1489 vvs->peer_fwd_cnt = le32_to_cpu(hdr->fwd_cnt); 1490 space_available = virtio_transport_has_space(vsk); 1491 spin_unlock_bh(&vvs->tx_lock); 1492 return space_available; 1493 } 1494 1495 /* Handle server socket */ 1496 static int 1497 virtio_transport_recv_listen(struct sock *sk, struct sk_buff *skb, 1498 struct virtio_transport *t) 1499 { 1500 struct virtio_vsock_hdr *hdr = virtio_vsock_hdr(skb); 1501 struct vsock_sock *vsk = vsock_sk(sk); 1502 struct vsock_sock *vchild; 1503 struct sock *child; 1504 int ret; 1505 1506 if (le16_to_cpu(hdr->op) != VIRTIO_VSOCK_OP_REQUEST) { 1507 virtio_transport_reset_no_sock(t, skb); 1508 return -EINVAL; 1509 } 1510 1511 if (sk_acceptq_is_full(sk)) { 1512 virtio_transport_reset_no_sock(t, skb); 1513 return -ENOMEM; 1514 } 1515 1516 /* __vsock_release() might have already flushed accept_queue. 1517 * Subsequent enqueues would lead to a memory leak. 1518 */ 1519 if (sk->sk_shutdown == SHUTDOWN_MASK) { 1520 virtio_transport_reset_no_sock(t, skb); 1521 return -ESHUTDOWN; 1522 } 1523 1524 child = vsock_create_connected(sk); 1525 if (!child) { 1526 virtio_transport_reset_no_sock(t, skb); 1527 return -ENOMEM; 1528 } 1529 1530 sk_acceptq_added(sk); 1531 1532 lock_sock_nested(child, SINGLE_DEPTH_NESTING); 1533 1534 child->sk_state = TCP_ESTABLISHED; 1535 1536 vchild = vsock_sk(child); 1537 vsock_addr_init(&vchild->local_addr, le64_to_cpu(hdr->dst_cid), 1538 le32_to_cpu(hdr->dst_port)); 1539 vsock_addr_init(&vchild->remote_addr, le64_to_cpu(hdr->src_cid), 1540 le32_to_cpu(hdr->src_port)); 1541 1542 ret = vsock_assign_transport(vchild, vsk); 1543 /* Transport assigned (looking at remote_addr) must be the same 1544 * where we received the request. 1545 */ 1546 if (ret || vchild->transport != &t->transport) { 1547 release_sock(child); 1548 virtio_transport_reset_no_sock(t, skb); 1549 sock_put(child); 1550 return ret; 1551 } 1552 1553 if (virtio_transport_space_update(child, skb)) 1554 child->sk_write_space(child); 1555 1556 vsock_insert_connected(vchild); 1557 vsock_enqueue_accept(sk, child); 1558 virtio_transport_send_response(vchild, skb); 1559 1560 release_sock(child); 1561 1562 sk->sk_data_ready(sk); 1563 return 0; 1564 } 1565 1566 static bool virtio_transport_valid_type(u16 type) 1567 { 1568 return (type == VIRTIO_VSOCK_TYPE_STREAM) || 1569 (type == VIRTIO_VSOCK_TYPE_SEQPACKET); 1570 } 1571 1572 /* We are under the virtio-vsock's vsock->rx_lock or vhost-vsock's vq->mutex 1573 * lock. 1574 */ 1575 void virtio_transport_recv_pkt(struct virtio_transport *t, 1576 struct sk_buff *skb) 1577 { 1578 struct virtio_vsock_hdr *hdr = virtio_vsock_hdr(skb); 1579 struct sockaddr_vm src, dst; 1580 struct vsock_sock *vsk; 1581 struct sock *sk; 1582 bool space_available; 1583 1584 vsock_addr_init(&src, le64_to_cpu(hdr->src_cid), 1585 le32_to_cpu(hdr->src_port)); 1586 vsock_addr_init(&dst, le64_to_cpu(hdr->dst_cid), 1587 le32_to_cpu(hdr->dst_port)); 1588 1589 trace_virtio_transport_recv_pkt(src.svm_cid, src.svm_port, 1590 dst.svm_cid, dst.svm_port, 1591 le32_to_cpu(hdr->len), 1592 le16_to_cpu(hdr->type), 1593 le16_to_cpu(hdr->op), 1594 le32_to_cpu(hdr->flags), 1595 le32_to_cpu(hdr->buf_alloc), 1596 le32_to_cpu(hdr->fwd_cnt)); 1597 1598 if (!virtio_transport_valid_type(le16_to_cpu(hdr->type))) { 1599 (void)virtio_transport_reset_no_sock(t, skb); 1600 goto free_pkt; 1601 } 1602 1603 /* The socket must be in connected or bound table 1604 * otherwise send reset back 1605 */ 1606 sk = vsock_find_connected_socket(&src, &dst); 1607 if (!sk) { 1608 sk = vsock_find_bound_socket(&dst); 1609 if (!sk) { 1610 (void)virtio_transport_reset_no_sock(t, skb); 1611 goto free_pkt; 1612 } 1613 } 1614 1615 if (virtio_transport_get_type(sk) != le16_to_cpu(hdr->type)) { 1616 (void)virtio_transport_reset_no_sock(t, skb); 1617 sock_put(sk); 1618 goto free_pkt; 1619 } 1620 1621 if (!skb_set_owner_sk_safe(skb, sk)) { 1622 WARN_ONCE(1, "receiving vsock socket has sk_refcnt == 0\n"); 1623 goto free_pkt; 1624 } 1625 1626 vsk = vsock_sk(sk); 1627 1628 lock_sock(sk); 1629 1630 /* Check if sk has been closed before lock_sock */ 1631 if (sock_flag(sk, SOCK_DONE)) { 1632 (void)virtio_transport_reset_no_sock(t, skb); 1633 release_sock(sk); 1634 sock_put(sk); 1635 goto free_pkt; 1636 } 1637 1638 space_available = virtio_transport_space_update(sk, skb); 1639 1640 /* Update CID in case it has changed after a transport reset event */ 1641 if (vsk->local_addr.svm_cid != VMADDR_CID_ANY) 1642 vsk->local_addr.svm_cid = dst.svm_cid; 1643 1644 if (space_available) 1645 sk->sk_write_space(sk); 1646 1647 switch (sk->sk_state) { 1648 case TCP_LISTEN: 1649 virtio_transport_recv_listen(sk, skb, t); 1650 kfree_skb(skb); 1651 break; 1652 case TCP_SYN_SENT: 1653 virtio_transport_recv_connecting(sk, skb); 1654 kfree_skb(skb); 1655 break; 1656 case TCP_ESTABLISHED: 1657 virtio_transport_recv_connected(sk, skb); 1658 break; 1659 case TCP_CLOSING: 1660 virtio_transport_recv_disconnecting(sk, skb); 1661 kfree_skb(skb); 1662 break; 1663 default: 1664 (void)virtio_transport_reset_no_sock(t, skb); 1665 kfree_skb(skb); 1666 break; 1667 } 1668 1669 release_sock(sk); 1670 1671 /* Release refcnt obtained when we fetched this socket out of the 1672 * bound or connected list. 1673 */ 1674 sock_put(sk); 1675 return; 1676 1677 free_pkt: 1678 kfree_skb(skb); 1679 } 1680 EXPORT_SYMBOL_GPL(virtio_transport_recv_pkt); 1681 1682 /* Remove skbs found in a queue that have a vsk that matches. 1683 * 1684 * Each skb is freed. 1685 * 1686 * Returns the count of skbs that were reply packets. 1687 */ 1688 int virtio_transport_purge_skbs(void *vsk, struct sk_buff_head *queue) 1689 { 1690 struct sk_buff_head freeme; 1691 struct sk_buff *skb, *tmp; 1692 int cnt = 0; 1693 1694 skb_queue_head_init(&freeme); 1695 1696 spin_lock_bh(&queue->lock); 1697 skb_queue_walk_safe(queue, skb, tmp) { 1698 if (vsock_sk(skb->sk) != vsk) 1699 continue; 1700 1701 __skb_unlink(skb, queue); 1702 __skb_queue_tail(&freeme, skb); 1703 1704 if (virtio_vsock_skb_reply(skb)) 1705 cnt++; 1706 } 1707 spin_unlock_bh(&queue->lock); 1708 1709 __skb_queue_purge(&freeme); 1710 1711 return cnt; 1712 } 1713 EXPORT_SYMBOL_GPL(virtio_transport_purge_skbs); 1714 1715 int virtio_transport_read_skb(struct vsock_sock *vsk, skb_read_actor_t recv_actor) 1716 { 1717 struct virtio_vsock_sock *vvs = vsk->trans; 1718 struct sock *sk = sk_vsock(vsk); 1719 struct virtio_vsock_hdr *hdr; 1720 struct sk_buff *skb; 1721 int off = 0; 1722 int err; 1723 1724 spin_lock_bh(&vvs->rx_lock); 1725 /* Use __skb_recv_datagram() for race-free handling of the receive. It 1726 * works for types other than dgrams. 1727 */ 1728 skb = __skb_recv_datagram(sk, &vvs->rx_queue, MSG_DONTWAIT, &off, &err); 1729 if (!skb) { 1730 spin_unlock_bh(&vvs->rx_lock); 1731 return err; 1732 } 1733 1734 hdr = virtio_vsock_hdr(skb); 1735 if (le32_to_cpu(hdr->flags) & VIRTIO_VSOCK_SEQ_EOM) 1736 vvs->msg_count--; 1737 1738 virtio_transport_dec_rx_pkt(vvs, le32_to_cpu(hdr->len)); 1739 spin_unlock_bh(&vvs->rx_lock); 1740 1741 virtio_transport_send_credit_update(vsk); 1742 1743 return recv_actor(sk, skb); 1744 } 1745 EXPORT_SYMBOL_GPL(virtio_transport_read_skb); 1746 1747 int virtio_transport_notify_set_rcvlowat(struct vsock_sock *vsk, int val) 1748 { 1749 struct virtio_vsock_sock *vvs = vsk->trans; 1750 bool send_update; 1751 1752 spin_lock_bh(&vvs->rx_lock); 1753 1754 /* If number of available bytes is less than new SO_RCVLOWAT value, 1755 * kick sender to send more data, because sender may sleep in its 1756 * 'send()' syscall waiting for enough space at our side. Also 1757 * don't send credit update when peer already knows actual value - 1758 * such transmission will be useless. 1759 */ 1760 send_update = (vvs->rx_bytes < val) && 1761 (vvs->fwd_cnt != vvs->last_fwd_cnt); 1762 1763 spin_unlock_bh(&vvs->rx_lock); 1764 1765 if (send_update) { 1766 int err; 1767 1768 err = virtio_transport_send_credit_update(vsk); 1769 if (err < 0) 1770 return err; 1771 } 1772 1773 return 0; 1774 } 1775 EXPORT_SYMBOL_GPL(virtio_transport_notify_set_rcvlowat); 1776 1777 MODULE_LICENSE("GPL v2"); 1778 MODULE_AUTHOR("Asias He"); 1779 MODULE_DESCRIPTION("common code for virtio vsock"); 1780