1 // SPDX-License-Identifier: GPL-2.0-only 2 /* 3 * common code for virtio vsock 4 * 5 * Copyright (C) 2013-2015 Red Hat, Inc. 6 * Author: Asias He <asias@redhat.com> 7 * Stefan Hajnoczi <stefanha@redhat.com> 8 */ 9 #include <linux/spinlock.h> 10 #include <linux/module.h> 11 #include <linux/sched/signal.h> 12 #include <linux/ctype.h> 13 #include <linux/list.h> 14 #include <linux/virtio_vsock.h> 15 #include <uapi/linux/vsockmon.h> 16 17 #include <net/sock.h> 18 #include <net/af_vsock.h> 19 20 #define CREATE_TRACE_POINTS 21 #include <trace/events/vsock_virtio_transport_common.h> 22 23 /* How long to wait for graceful shutdown of a connection */ 24 #define VSOCK_CLOSE_TIMEOUT (8 * HZ) 25 26 /* Threshold for detecting small packets to copy */ 27 #define GOOD_COPY_LEN 128 28 29 static void virtio_transport_cancel_close_work(struct vsock_sock *vsk, 30 bool cancel_timeout); 31 32 static const struct virtio_transport * 33 virtio_transport_get_ops(struct vsock_sock *vsk) 34 { 35 const struct vsock_transport *t = vsock_core_get_transport(vsk); 36 37 if (WARN_ON(!t)) 38 return NULL; 39 40 return container_of(t, struct virtio_transport, transport); 41 } 42 43 static bool virtio_transport_can_zcopy(const struct virtio_transport *t_ops, 44 struct virtio_vsock_pkt_info *info, 45 size_t pkt_len) 46 { 47 struct iov_iter *iov_iter; 48 49 if (!info->msg) 50 return false; 51 52 iov_iter = &info->msg->msg_iter; 53 54 if (iov_iter->iov_offset) 55 return false; 56 57 /* We can't send whole iov. */ 58 if (iov_iter->count > pkt_len) 59 return false; 60 61 /* Check that transport can send data in zerocopy mode. */ 62 t_ops = virtio_transport_get_ops(info->vsk); 63 64 if (t_ops->can_msgzerocopy) { 65 int pages_to_send = iov_iter_npages(iov_iter, MAX_SKB_FRAGS); 66 67 /* +1 is for packet header. */ 68 return t_ops->can_msgzerocopy(pages_to_send + 1); 69 } 70 71 return true; 72 } 73 74 static int virtio_transport_init_zcopy_skb(struct vsock_sock *vsk, 75 struct sk_buff *skb, 76 struct msghdr *msg, 77 bool zerocopy) 78 { 79 struct ubuf_info *uarg; 80 81 if (msg->msg_ubuf) { 82 uarg = msg->msg_ubuf; 83 net_zcopy_get(uarg); 84 } else { 85 struct iov_iter *iter = &msg->msg_iter; 86 struct ubuf_info_msgzc *uarg_zc; 87 88 uarg = msg_zerocopy_realloc(sk_vsock(vsk), 89 iter->count, 90 NULL, false); 91 if (!uarg) 92 return -1; 93 94 uarg_zc = uarg_to_msgzc(uarg); 95 uarg_zc->zerocopy = zerocopy ? 1 : 0; 96 } 97 98 skb_zcopy_init(skb, uarg); 99 100 return 0; 101 } 102 103 static int virtio_transport_fill_skb(struct sk_buff *skb, 104 struct virtio_vsock_pkt_info *info, 105 size_t len, 106 bool zcopy) 107 { 108 if (zcopy) 109 return __zerocopy_sg_from_iter(info->msg, NULL, skb, 110 &info->msg->msg_iter, len, NULL); 111 112 return memcpy_from_msg(skb_put(skb, len), info->msg, len); 113 } 114 115 static void virtio_transport_init_hdr(struct sk_buff *skb, 116 struct virtio_vsock_pkt_info *info, 117 size_t payload_len, 118 u32 src_cid, 119 u32 src_port, 120 u32 dst_cid, 121 u32 dst_port) 122 { 123 struct virtio_vsock_hdr *hdr; 124 125 hdr = virtio_vsock_hdr(skb); 126 hdr->type = cpu_to_le16(info->type); 127 hdr->op = cpu_to_le16(info->op); 128 hdr->src_cid = cpu_to_le64(src_cid); 129 hdr->dst_cid = cpu_to_le64(dst_cid); 130 hdr->src_port = cpu_to_le32(src_port); 131 hdr->dst_port = cpu_to_le32(dst_port); 132 hdr->flags = cpu_to_le32(info->flags); 133 hdr->len = cpu_to_le32(payload_len); 134 hdr->buf_alloc = cpu_to_le32(0); 135 hdr->fwd_cnt = cpu_to_le32(0); 136 } 137 138 static void virtio_transport_copy_nonlinear_skb(const struct sk_buff *skb, 139 void *dst, 140 size_t len) 141 { 142 struct iov_iter iov_iter = { 0 }; 143 struct kvec kvec; 144 size_t to_copy; 145 146 kvec.iov_base = dst; 147 kvec.iov_len = len; 148 149 iov_iter.iter_type = ITER_KVEC; 150 iov_iter.kvec = &kvec; 151 iov_iter.nr_segs = 1; 152 153 to_copy = min_t(size_t, len, skb->len); 154 155 skb_copy_datagram_iter(skb, VIRTIO_VSOCK_SKB_CB(skb)->offset, 156 &iov_iter, to_copy); 157 } 158 159 /* Packet capture */ 160 static struct sk_buff *virtio_transport_build_skb(void *opaque) 161 { 162 struct virtio_vsock_hdr *pkt_hdr; 163 struct sk_buff *pkt = opaque; 164 struct af_vsockmon_hdr *hdr; 165 struct sk_buff *skb; 166 size_t payload_len; 167 168 /* A packet could be split to fit the RX buffer, so we can retrieve 169 * the payload length from the header and the buffer pointer taking 170 * care of the offset in the original packet. 171 */ 172 pkt_hdr = virtio_vsock_hdr(pkt); 173 payload_len = pkt->len; 174 175 skb = alloc_skb(sizeof(*hdr) + sizeof(*pkt_hdr) + payload_len, 176 GFP_ATOMIC); 177 if (!skb) 178 return NULL; 179 180 hdr = skb_put(skb, sizeof(*hdr)); 181 182 /* pkt->hdr is little-endian so no need to byteswap here */ 183 hdr->src_cid = pkt_hdr->src_cid; 184 hdr->src_port = pkt_hdr->src_port; 185 hdr->dst_cid = pkt_hdr->dst_cid; 186 hdr->dst_port = pkt_hdr->dst_port; 187 188 hdr->transport = cpu_to_le16(AF_VSOCK_TRANSPORT_VIRTIO); 189 hdr->len = cpu_to_le16(sizeof(*pkt_hdr)); 190 memset(hdr->reserved, 0, sizeof(hdr->reserved)); 191 192 switch (le16_to_cpu(pkt_hdr->op)) { 193 case VIRTIO_VSOCK_OP_REQUEST: 194 case VIRTIO_VSOCK_OP_RESPONSE: 195 hdr->op = cpu_to_le16(AF_VSOCK_OP_CONNECT); 196 break; 197 case VIRTIO_VSOCK_OP_RST: 198 case VIRTIO_VSOCK_OP_SHUTDOWN: 199 hdr->op = cpu_to_le16(AF_VSOCK_OP_DISCONNECT); 200 break; 201 case VIRTIO_VSOCK_OP_RW: 202 hdr->op = cpu_to_le16(AF_VSOCK_OP_PAYLOAD); 203 break; 204 case VIRTIO_VSOCK_OP_CREDIT_UPDATE: 205 case VIRTIO_VSOCK_OP_CREDIT_REQUEST: 206 hdr->op = cpu_to_le16(AF_VSOCK_OP_CONTROL); 207 break; 208 default: 209 hdr->op = cpu_to_le16(AF_VSOCK_OP_UNKNOWN); 210 break; 211 } 212 213 skb_put_data(skb, pkt_hdr, sizeof(*pkt_hdr)); 214 215 if (payload_len) { 216 if (skb_is_nonlinear(pkt)) { 217 void *data = skb_put(skb, payload_len); 218 219 virtio_transport_copy_nonlinear_skb(pkt, data, payload_len); 220 } else { 221 skb_put_data(skb, pkt->data, payload_len); 222 } 223 } 224 225 return skb; 226 } 227 228 void virtio_transport_deliver_tap_pkt(struct sk_buff *skb) 229 { 230 if (virtio_vsock_skb_tap_delivered(skb)) 231 return; 232 233 vsock_deliver_tap(virtio_transport_build_skb, skb); 234 virtio_vsock_skb_set_tap_delivered(skb); 235 } 236 EXPORT_SYMBOL_GPL(virtio_transport_deliver_tap_pkt); 237 238 static u16 virtio_transport_get_type(struct sock *sk) 239 { 240 if (sk->sk_type == SOCK_STREAM) 241 return VIRTIO_VSOCK_TYPE_STREAM; 242 else 243 return VIRTIO_VSOCK_TYPE_SEQPACKET; 244 } 245 246 /* Returns new sk_buff on success, otherwise returns NULL. */ 247 static struct sk_buff *virtio_transport_alloc_skb(struct virtio_vsock_pkt_info *info, 248 size_t payload_len, 249 bool zcopy, 250 u32 src_cid, 251 u32 src_port, 252 u32 dst_cid, 253 u32 dst_port) 254 { 255 struct vsock_sock *vsk; 256 struct sk_buff *skb; 257 size_t skb_len; 258 259 skb_len = VIRTIO_VSOCK_SKB_HEADROOM; 260 261 if (!zcopy) 262 skb_len += payload_len; 263 264 skb = virtio_vsock_alloc_skb(skb_len, GFP_KERNEL); 265 if (!skb) 266 return NULL; 267 268 virtio_transport_init_hdr(skb, info, payload_len, src_cid, src_port, 269 dst_cid, dst_port); 270 271 vsk = info->vsk; 272 273 /* If 'vsk' != NULL then payload is always present, so we 274 * will never call '__zerocopy_sg_from_iter()' below without 275 * setting skb owner in 'skb_set_owner_w()'. The only case 276 * when 'vsk' == NULL is VIRTIO_VSOCK_OP_RST control message 277 * without payload. 278 */ 279 WARN_ON_ONCE(!(vsk && (info->msg && payload_len)) && zcopy); 280 281 /* Set owner here, because '__zerocopy_sg_from_iter()' uses 282 * owner of skb without check to update 'sk_wmem_alloc'. 283 */ 284 if (vsk) 285 skb_set_owner_w(skb, sk_vsock(vsk)); 286 287 if (info->msg && payload_len > 0) { 288 int err; 289 290 err = virtio_transport_fill_skb(skb, info, payload_len, zcopy); 291 if (err) 292 goto out; 293 294 if (msg_data_left(info->msg) == 0 && 295 info->type == VIRTIO_VSOCK_TYPE_SEQPACKET) { 296 struct virtio_vsock_hdr *hdr = virtio_vsock_hdr(skb); 297 298 hdr->flags |= cpu_to_le32(VIRTIO_VSOCK_SEQ_EOM); 299 300 if (info->msg->msg_flags & MSG_EOR) 301 hdr->flags |= cpu_to_le32(VIRTIO_VSOCK_SEQ_EOR); 302 } 303 } 304 305 if (info->reply) 306 virtio_vsock_skb_set_reply(skb); 307 308 trace_virtio_transport_alloc_pkt(src_cid, src_port, 309 dst_cid, dst_port, 310 payload_len, 311 info->type, 312 info->op, 313 info->flags, 314 zcopy); 315 316 return skb; 317 out: 318 kfree_skb(skb); 319 return NULL; 320 } 321 322 /* This function can only be used on connecting/connected sockets, 323 * since a socket assigned to a transport is required. 324 * 325 * Do not use on listener sockets! 326 */ 327 static int virtio_transport_send_pkt_info(struct vsock_sock *vsk, 328 struct virtio_vsock_pkt_info *info) 329 { 330 u32 max_skb_len = VIRTIO_VSOCK_MAX_PKT_BUF_SIZE; 331 u32 src_cid, src_port, dst_cid, dst_port; 332 const struct virtio_transport *t_ops; 333 struct virtio_vsock_sock *vvs; 334 u32 pkt_len = info->pkt_len; 335 bool can_zcopy = false; 336 u32 rest_len; 337 int ret; 338 339 info->type = virtio_transport_get_type(sk_vsock(vsk)); 340 341 t_ops = virtio_transport_get_ops(vsk); 342 if (unlikely(!t_ops)) 343 return -EFAULT; 344 345 src_cid = t_ops->transport.get_local_cid(); 346 src_port = vsk->local_addr.svm_port; 347 if (!info->remote_cid) { 348 dst_cid = vsk->remote_addr.svm_cid; 349 dst_port = vsk->remote_addr.svm_port; 350 } else { 351 dst_cid = info->remote_cid; 352 dst_port = info->remote_port; 353 } 354 355 vvs = vsk->trans; 356 357 /* virtio_transport_get_credit might return less than pkt_len credit */ 358 pkt_len = virtio_transport_get_credit(vvs, pkt_len); 359 360 /* Do not send zero length OP_RW pkt */ 361 if (pkt_len == 0 && info->op == VIRTIO_VSOCK_OP_RW) 362 return pkt_len; 363 364 if (info->msg) { 365 /* If zerocopy is not enabled by 'setsockopt()', we behave as 366 * there is no MSG_ZEROCOPY flag set. 367 */ 368 if (!sock_flag(sk_vsock(vsk), SOCK_ZEROCOPY)) 369 info->msg->msg_flags &= ~MSG_ZEROCOPY; 370 371 if (info->msg->msg_flags & MSG_ZEROCOPY) 372 can_zcopy = virtio_transport_can_zcopy(t_ops, info, pkt_len); 373 374 if (can_zcopy) 375 max_skb_len = min_t(u32, VIRTIO_VSOCK_MAX_PKT_BUF_SIZE, 376 (MAX_SKB_FRAGS * PAGE_SIZE)); 377 } 378 379 rest_len = pkt_len; 380 381 do { 382 struct sk_buff *skb; 383 size_t skb_len; 384 385 skb_len = min(max_skb_len, rest_len); 386 387 skb = virtio_transport_alloc_skb(info, skb_len, can_zcopy, 388 src_cid, src_port, 389 dst_cid, dst_port); 390 if (!skb) { 391 ret = -ENOMEM; 392 break; 393 } 394 395 /* We process buffer part by part, allocating skb on 396 * each iteration. If this is last skb for this buffer 397 * and MSG_ZEROCOPY mode is in use - we must allocate 398 * completion for the current syscall. 399 */ 400 if (info->msg && info->msg->msg_flags & MSG_ZEROCOPY && 401 skb_len == rest_len && info->op == VIRTIO_VSOCK_OP_RW) { 402 if (virtio_transport_init_zcopy_skb(vsk, skb, 403 info->msg, 404 can_zcopy)) { 405 kfree_skb(skb); 406 ret = -ENOMEM; 407 break; 408 } 409 } 410 411 virtio_transport_inc_tx_pkt(vvs, skb); 412 413 ret = t_ops->send_pkt(skb); 414 if (ret < 0) 415 break; 416 417 /* Both virtio and vhost 'send_pkt()' returns 'skb_len', 418 * but for reliability use 'ret' instead of 'skb_len'. 419 * Also if partial send happens (e.g. 'ret' != 'skb_len') 420 * somehow, we break this loop, but account such returned 421 * value in 'virtio_transport_put_credit()'. 422 */ 423 rest_len -= ret; 424 425 if (WARN_ONCE(ret != skb_len, 426 "'send_pkt()' returns %i, but %zu expected\n", 427 ret, skb_len)) 428 break; 429 } while (rest_len); 430 431 virtio_transport_put_credit(vvs, rest_len); 432 433 /* Return number of bytes, if any data has been sent. */ 434 if (rest_len != pkt_len) 435 ret = pkt_len - rest_len; 436 437 return ret; 438 } 439 440 static bool virtio_transport_inc_rx_pkt(struct virtio_vsock_sock *vvs, 441 u32 len) 442 { 443 if (vvs->buf_used + len > vvs->buf_alloc) 444 return false; 445 446 vvs->rx_bytes += len; 447 vvs->buf_used += len; 448 return true; 449 } 450 451 static void virtio_transport_dec_rx_pkt(struct virtio_vsock_sock *vvs, 452 u32 bytes_read, u32 bytes_dequeued) 453 { 454 vvs->rx_bytes -= bytes_read; 455 vvs->buf_used -= bytes_dequeued; 456 vvs->fwd_cnt += bytes_dequeued; 457 } 458 459 void virtio_transport_inc_tx_pkt(struct virtio_vsock_sock *vvs, struct sk_buff *skb) 460 { 461 struct virtio_vsock_hdr *hdr = virtio_vsock_hdr(skb); 462 463 spin_lock_bh(&vvs->rx_lock); 464 vvs->last_fwd_cnt = vvs->fwd_cnt; 465 hdr->fwd_cnt = cpu_to_le32(vvs->fwd_cnt); 466 hdr->buf_alloc = cpu_to_le32(vvs->buf_alloc); 467 spin_unlock_bh(&vvs->rx_lock); 468 } 469 EXPORT_SYMBOL_GPL(virtio_transport_inc_tx_pkt); 470 471 void virtio_transport_consume_skb_sent(struct sk_buff *skb, bool consume) 472 { 473 struct sock *s = skb->sk; 474 475 if (s && skb->len) { 476 struct vsock_sock *vs = vsock_sk(s); 477 struct virtio_vsock_sock *vvs; 478 479 vvs = vs->trans; 480 481 spin_lock_bh(&vvs->tx_lock); 482 vvs->bytes_unsent -= skb->len; 483 spin_unlock_bh(&vvs->tx_lock); 484 } 485 486 if (consume) 487 consume_skb(skb); 488 } 489 EXPORT_SYMBOL_GPL(virtio_transport_consume_skb_sent); 490 491 u32 virtio_transport_get_credit(struct virtio_vsock_sock *vvs, u32 credit) 492 { 493 u32 ret; 494 495 if (!credit) 496 return 0; 497 498 spin_lock_bh(&vvs->tx_lock); 499 ret = vvs->peer_buf_alloc - (vvs->tx_cnt - vvs->peer_fwd_cnt); 500 if (ret > credit) 501 ret = credit; 502 vvs->tx_cnt += ret; 503 vvs->bytes_unsent += ret; 504 spin_unlock_bh(&vvs->tx_lock); 505 506 return ret; 507 } 508 EXPORT_SYMBOL_GPL(virtio_transport_get_credit); 509 510 void virtio_transport_put_credit(struct virtio_vsock_sock *vvs, u32 credit) 511 { 512 if (!credit) 513 return; 514 515 spin_lock_bh(&vvs->tx_lock); 516 vvs->tx_cnt -= credit; 517 vvs->bytes_unsent -= credit; 518 spin_unlock_bh(&vvs->tx_lock); 519 } 520 EXPORT_SYMBOL_GPL(virtio_transport_put_credit); 521 522 static int virtio_transport_send_credit_update(struct vsock_sock *vsk) 523 { 524 struct virtio_vsock_pkt_info info = { 525 .op = VIRTIO_VSOCK_OP_CREDIT_UPDATE, 526 .vsk = vsk, 527 }; 528 529 return virtio_transport_send_pkt_info(vsk, &info); 530 } 531 532 static ssize_t 533 virtio_transport_stream_do_peek(struct vsock_sock *vsk, 534 struct msghdr *msg, 535 size_t len) 536 { 537 struct virtio_vsock_sock *vvs = vsk->trans; 538 struct sk_buff *skb; 539 size_t total = 0; 540 int err; 541 542 spin_lock_bh(&vvs->rx_lock); 543 544 skb_queue_walk(&vvs->rx_queue, skb) { 545 size_t bytes; 546 547 bytes = len - total; 548 if (bytes > skb->len) 549 bytes = skb->len; 550 551 spin_unlock_bh(&vvs->rx_lock); 552 553 /* sk_lock is held by caller so no one else can dequeue. 554 * Unlock rx_lock since skb_copy_datagram_iter() may sleep. 555 */ 556 err = skb_copy_datagram_iter(skb, VIRTIO_VSOCK_SKB_CB(skb)->offset, 557 &msg->msg_iter, bytes); 558 if (err) 559 goto out; 560 561 total += bytes; 562 563 spin_lock_bh(&vvs->rx_lock); 564 565 if (total == len) 566 break; 567 } 568 569 spin_unlock_bh(&vvs->rx_lock); 570 571 return total; 572 573 out: 574 if (total) 575 err = total; 576 return err; 577 } 578 579 static ssize_t 580 virtio_transport_stream_do_dequeue(struct vsock_sock *vsk, 581 struct msghdr *msg, 582 size_t len) 583 { 584 struct virtio_vsock_sock *vvs = vsk->trans; 585 struct sk_buff *skb; 586 u32 fwd_cnt_delta; 587 bool low_rx_bytes; 588 int err = -EFAULT; 589 size_t total = 0; 590 u32 free_space; 591 592 spin_lock_bh(&vvs->rx_lock); 593 594 if (WARN_ONCE(skb_queue_empty(&vvs->rx_queue) && vvs->rx_bytes, 595 "rx_queue is empty, but rx_bytes is non-zero\n")) { 596 spin_unlock_bh(&vvs->rx_lock); 597 return err; 598 } 599 600 while (total < len && !skb_queue_empty(&vvs->rx_queue)) { 601 size_t bytes, dequeued = 0; 602 603 skb = skb_peek(&vvs->rx_queue); 604 605 bytes = min_t(size_t, len - total, 606 skb->len - VIRTIO_VSOCK_SKB_CB(skb)->offset); 607 608 /* sk_lock is held by caller so no one else can dequeue. 609 * Unlock rx_lock since skb_copy_datagram_iter() may sleep. 610 */ 611 spin_unlock_bh(&vvs->rx_lock); 612 613 err = skb_copy_datagram_iter(skb, 614 VIRTIO_VSOCK_SKB_CB(skb)->offset, 615 &msg->msg_iter, bytes); 616 if (err) 617 goto out; 618 619 spin_lock_bh(&vvs->rx_lock); 620 621 total += bytes; 622 623 VIRTIO_VSOCK_SKB_CB(skb)->offset += bytes; 624 625 if (skb->len == VIRTIO_VSOCK_SKB_CB(skb)->offset) { 626 dequeued = le32_to_cpu(virtio_vsock_hdr(skb)->len); 627 __skb_unlink(skb, &vvs->rx_queue); 628 consume_skb(skb); 629 } 630 631 virtio_transport_dec_rx_pkt(vvs, bytes, dequeued); 632 } 633 634 fwd_cnt_delta = vvs->fwd_cnt - vvs->last_fwd_cnt; 635 free_space = vvs->buf_alloc - fwd_cnt_delta; 636 low_rx_bytes = (vvs->rx_bytes < 637 sock_rcvlowat(sk_vsock(vsk), 0, INT_MAX)); 638 639 spin_unlock_bh(&vvs->rx_lock); 640 641 /* To reduce the number of credit update messages, 642 * don't update credits as long as lots of space is available. 643 * Note: the limit chosen here is arbitrary. Setting the limit 644 * too high causes extra messages. Too low causes transmitter 645 * stalls. As stalls are in theory more expensive than extra 646 * messages, we set the limit to a high value. TODO: experiment 647 * with different values. Also send credit update message when 648 * number of bytes in rx queue is not enough to wake up reader. 649 */ 650 if (fwd_cnt_delta && 651 (free_space < VIRTIO_VSOCK_MAX_PKT_BUF_SIZE || low_rx_bytes)) 652 virtio_transport_send_credit_update(vsk); 653 654 return total; 655 656 out: 657 if (total) 658 err = total; 659 return err; 660 } 661 662 static ssize_t 663 virtio_transport_seqpacket_do_peek(struct vsock_sock *vsk, 664 struct msghdr *msg) 665 { 666 struct virtio_vsock_sock *vvs = vsk->trans; 667 struct sk_buff *skb; 668 size_t total, len; 669 670 spin_lock_bh(&vvs->rx_lock); 671 672 if (!vvs->msg_count) { 673 spin_unlock_bh(&vvs->rx_lock); 674 return 0; 675 } 676 677 total = 0; 678 len = msg_data_left(msg); 679 680 skb_queue_walk(&vvs->rx_queue, skb) { 681 struct virtio_vsock_hdr *hdr; 682 683 if (total < len) { 684 size_t bytes; 685 int err; 686 687 bytes = len - total; 688 if (bytes > skb->len) 689 bytes = skb->len; 690 691 spin_unlock_bh(&vvs->rx_lock); 692 693 /* sk_lock is held by caller so no one else can dequeue. 694 * Unlock rx_lock since skb_copy_datagram_iter() may sleep. 695 */ 696 err = skb_copy_datagram_iter(skb, VIRTIO_VSOCK_SKB_CB(skb)->offset, 697 &msg->msg_iter, bytes); 698 if (err) 699 return err; 700 701 spin_lock_bh(&vvs->rx_lock); 702 } 703 704 total += skb->len; 705 hdr = virtio_vsock_hdr(skb); 706 707 if (le32_to_cpu(hdr->flags) & VIRTIO_VSOCK_SEQ_EOM) { 708 if (le32_to_cpu(hdr->flags) & VIRTIO_VSOCK_SEQ_EOR) 709 msg->msg_flags |= MSG_EOR; 710 711 break; 712 } 713 } 714 715 spin_unlock_bh(&vvs->rx_lock); 716 717 return total; 718 } 719 720 static int virtio_transport_seqpacket_do_dequeue(struct vsock_sock *vsk, 721 struct msghdr *msg, 722 int flags) 723 { 724 struct virtio_vsock_sock *vvs = vsk->trans; 725 int dequeued_len = 0; 726 size_t user_buf_len = msg_data_left(msg); 727 bool msg_ready = false; 728 struct sk_buff *skb; 729 730 spin_lock_bh(&vvs->rx_lock); 731 732 if (vvs->msg_count == 0) { 733 spin_unlock_bh(&vvs->rx_lock); 734 return 0; 735 } 736 737 while (!msg_ready) { 738 struct virtio_vsock_hdr *hdr; 739 size_t pkt_len; 740 741 skb = __skb_dequeue(&vvs->rx_queue); 742 if (!skb) 743 break; 744 hdr = virtio_vsock_hdr(skb); 745 pkt_len = (size_t)le32_to_cpu(hdr->len); 746 747 if (dequeued_len >= 0) { 748 size_t bytes_to_copy; 749 750 bytes_to_copy = min(user_buf_len, pkt_len); 751 752 if (bytes_to_copy) { 753 int err; 754 755 /* sk_lock is held by caller so no one else can dequeue. 756 * Unlock rx_lock since skb_copy_datagram_iter() may sleep. 757 */ 758 spin_unlock_bh(&vvs->rx_lock); 759 760 err = skb_copy_datagram_iter(skb, 0, 761 &msg->msg_iter, 762 bytes_to_copy); 763 if (err) { 764 /* Copy of message failed. Rest of 765 * fragments will be freed without copy. 766 */ 767 dequeued_len = err; 768 } else { 769 user_buf_len -= bytes_to_copy; 770 } 771 772 spin_lock_bh(&vvs->rx_lock); 773 } 774 775 if (dequeued_len >= 0) 776 dequeued_len += pkt_len; 777 } 778 779 if (le32_to_cpu(hdr->flags) & VIRTIO_VSOCK_SEQ_EOM) { 780 msg_ready = true; 781 vvs->msg_count--; 782 783 if (le32_to_cpu(hdr->flags) & VIRTIO_VSOCK_SEQ_EOR) 784 msg->msg_flags |= MSG_EOR; 785 } 786 787 virtio_transport_dec_rx_pkt(vvs, pkt_len, pkt_len); 788 kfree_skb(skb); 789 } 790 791 spin_unlock_bh(&vvs->rx_lock); 792 793 virtio_transport_send_credit_update(vsk); 794 795 return dequeued_len; 796 } 797 798 ssize_t 799 virtio_transport_stream_dequeue(struct vsock_sock *vsk, 800 struct msghdr *msg, 801 size_t len, int flags) 802 { 803 if (flags & MSG_PEEK) 804 return virtio_transport_stream_do_peek(vsk, msg, len); 805 else 806 return virtio_transport_stream_do_dequeue(vsk, msg, len); 807 } 808 EXPORT_SYMBOL_GPL(virtio_transport_stream_dequeue); 809 810 ssize_t 811 virtio_transport_seqpacket_dequeue(struct vsock_sock *vsk, 812 struct msghdr *msg, 813 int flags) 814 { 815 if (flags & MSG_PEEK) 816 return virtio_transport_seqpacket_do_peek(vsk, msg); 817 else 818 return virtio_transport_seqpacket_do_dequeue(vsk, msg, flags); 819 } 820 EXPORT_SYMBOL_GPL(virtio_transport_seqpacket_dequeue); 821 822 int 823 virtio_transport_seqpacket_enqueue(struct vsock_sock *vsk, 824 struct msghdr *msg, 825 size_t len) 826 { 827 struct virtio_vsock_sock *vvs = vsk->trans; 828 829 spin_lock_bh(&vvs->tx_lock); 830 831 if (len > vvs->peer_buf_alloc) { 832 spin_unlock_bh(&vvs->tx_lock); 833 return -EMSGSIZE; 834 } 835 836 spin_unlock_bh(&vvs->tx_lock); 837 838 return virtio_transport_stream_enqueue(vsk, msg, len); 839 } 840 EXPORT_SYMBOL_GPL(virtio_transport_seqpacket_enqueue); 841 842 int 843 virtio_transport_dgram_dequeue(struct vsock_sock *vsk, 844 struct msghdr *msg, 845 size_t len, int flags) 846 { 847 return -EOPNOTSUPP; 848 } 849 EXPORT_SYMBOL_GPL(virtio_transport_dgram_dequeue); 850 851 s64 virtio_transport_stream_has_data(struct vsock_sock *vsk) 852 { 853 struct virtio_vsock_sock *vvs = vsk->trans; 854 s64 bytes; 855 856 spin_lock_bh(&vvs->rx_lock); 857 bytes = vvs->rx_bytes; 858 spin_unlock_bh(&vvs->rx_lock); 859 860 return bytes; 861 } 862 EXPORT_SYMBOL_GPL(virtio_transport_stream_has_data); 863 864 u32 virtio_transport_seqpacket_has_data(struct vsock_sock *vsk) 865 { 866 struct virtio_vsock_sock *vvs = vsk->trans; 867 u32 msg_count; 868 869 spin_lock_bh(&vvs->rx_lock); 870 msg_count = vvs->msg_count; 871 spin_unlock_bh(&vvs->rx_lock); 872 873 return msg_count; 874 } 875 EXPORT_SYMBOL_GPL(virtio_transport_seqpacket_has_data); 876 877 static s64 virtio_transport_has_space(struct vsock_sock *vsk) 878 { 879 struct virtio_vsock_sock *vvs = vsk->trans; 880 s64 bytes; 881 882 bytes = (s64)vvs->peer_buf_alloc - (vvs->tx_cnt - vvs->peer_fwd_cnt); 883 if (bytes < 0) 884 bytes = 0; 885 886 return bytes; 887 } 888 889 s64 virtio_transport_stream_has_space(struct vsock_sock *vsk) 890 { 891 struct virtio_vsock_sock *vvs = vsk->trans; 892 s64 bytes; 893 894 spin_lock_bh(&vvs->tx_lock); 895 bytes = virtio_transport_has_space(vsk); 896 spin_unlock_bh(&vvs->tx_lock); 897 898 return bytes; 899 } 900 EXPORT_SYMBOL_GPL(virtio_transport_stream_has_space); 901 902 int virtio_transport_do_socket_init(struct vsock_sock *vsk, 903 struct vsock_sock *psk) 904 { 905 struct virtio_vsock_sock *vvs; 906 907 vvs = kzalloc(sizeof(*vvs), GFP_KERNEL); 908 if (!vvs) 909 return -ENOMEM; 910 911 vsk->trans = vvs; 912 vvs->vsk = vsk; 913 if (psk && psk->trans) { 914 struct virtio_vsock_sock *ptrans = psk->trans; 915 916 vvs->peer_buf_alloc = ptrans->peer_buf_alloc; 917 } 918 919 if (vsk->buffer_size > VIRTIO_VSOCK_MAX_BUF_SIZE) 920 vsk->buffer_size = VIRTIO_VSOCK_MAX_BUF_SIZE; 921 922 vvs->buf_alloc = vsk->buffer_size; 923 924 spin_lock_init(&vvs->rx_lock); 925 spin_lock_init(&vvs->tx_lock); 926 skb_queue_head_init(&vvs->rx_queue); 927 928 return 0; 929 } 930 EXPORT_SYMBOL_GPL(virtio_transport_do_socket_init); 931 932 /* sk_lock held by the caller */ 933 void virtio_transport_notify_buffer_size(struct vsock_sock *vsk, u64 *val) 934 { 935 struct virtio_vsock_sock *vvs = vsk->trans; 936 937 if (*val > VIRTIO_VSOCK_MAX_BUF_SIZE) 938 *val = VIRTIO_VSOCK_MAX_BUF_SIZE; 939 940 vvs->buf_alloc = *val; 941 942 virtio_transport_send_credit_update(vsk); 943 } 944 EXPORT_SYMBOL_GPL(virtio_transport_notify_buffer_size); 945 946 int 947 virtio_transport_notify_poll_in(struct vsock_sock *vsk, 948 size_t target, 949 bool *data_ready_now) 950 { 951 *data_ready_now = vsock_stream_has_data(vsk) >= target; 952 953 return 0; 954 } 955 EXPORT_SYMBOL_GPL(virtio_transport_notify_poll_in); 956 957 int 958 virtio_transport_notify_poll_out(struct vsock_sock *vsk, 959 size_t target, 960 bool *space_avail_now) 961 { 962 s64 free_space; 963 964 free_space = vsock_stream_has_space(vsk); 965 if (free_space > 0) 966 *space_avail_now = true; 967 else if (free_space == 0) 968 *space_avail_now = false; 969 970 return 0; 971 } 972 EXPORT_SYMBOL_GPL(virtio_transport_notify_poll_out); 973 974 int virtio_transport_notify_recv_init(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_init); 980 981 int virtio_transport_notify_recv_pre_block(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_block); 987 988 int virtio_transport_notify_recv_pre_dequeue(struct vsock_sock *vsk, 989 size_t target, struct vsock_transport_recv_notify_data *data) 990 { 991 return 0; 992 } 993 EXPORT_SYMBOL_GPL(virtio_transport_notify_recv_pre_dequeue); 994 995 int virtio_transport_notify_recv_post_dequeue(struct vsock_sock *vsk, 996 size_t target, ssize_t copied, bool data_read, 997 struct vsock_transport_recv_notify_data *data) 998 { 999 return 0; 1000 } 1001 EXPORT_SYMBOL_GPL(virtio_transport_notify_recv_post_dequeue); 1002 1003 int virtio_transport_notify_send_init(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_init); 1009 1010 int virtio_transport_notify_send_pre_block(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_block); 1016 1017 int virtio_transport_notify_send_pre_enqueue(struct vsock_sock *vsk, 1018 struct vsock_transport_send_notify_data *data) 1019 { 1020 return 0; 1021 } 1022 EXPORT_SYMBOL_GPL(virtio_transport_notify_send_pre_enqueue); 1023 1024 int virtio_transport_notify_send_post_enqueue(struct vsock_sock *vsk, 1025 ssize_t written, struct vsock_transport_send_notify_data *data) 1026 { 1027 return 0; 1028 } 1029 EXPORT_SYMBOL_GPL(virtio_transport_notify_send_post_enqueue); 1030 1031 u64 virtio_transport_stream_rcvhiwat(struct vsock_sock *vsk) 1032 { 1033 return vsk->buffer_size; 1034 } 1035 EXPORT_SYMBOL_GPL(virtio_transport_stream_rcvhiwat); 1036 1037 bool virtio_transport_stream_is_active(struct vsock_sock *vsk) 1038 { 1039 return true; 1040 } 1041 EXPORT_SYMBOL_GPL(virtio_transport_stream_is_active); 1042 1043 bool virtio_transport_stream_allow(u32 cid, u32 port) 1044 { 1045 return true; 1046 } 1047 EXPORT_SYMBOL_GPL(virtio_transport_stream_allow); 1048 1049 int virtio_transport_dgram_bind(struct vsock_sock *vsk, 1050 struct sockaddr_vm *addr) 1051 { 1052 return -EOPNOTSUPP; 1053 } 1054 EXPORT_SYMBOL_GPL(virtio_transport_dgram_bind); 1055 1056 bool virtio_transport_dgram_allow(u32 cid, u32 port) 1057 { 1058 return false; 1059 } 1060 EXPORT_SYMBOL_GPL(virtio_transport_dgram_allow); 1061 1062 int virtio_transport_connect(struct vsock_sock *vsk) 1063 { 1064 struct virtio_vsock_pkt_info info = { 1065 .op = VIRTIO_VSOCK_OP_REQUEST, 1066 .vsk = vsk, 1067 }; 1068 1069 return virtio_transport_send_pkt_info(vsk, &info); 1070 } 1071 EXPORT_SYMBOL_GPL(virtio_transport_connect); 1072 1073 int virtio_transport_shutdown(struct vsock_sock *vsk, int mode) 1074 { 1075 struct virtio_vsock_pkt_info info = { 1076 .op = VIRTIO_VSOCK_OP_SHUTDOWN, 1077 .flags = (mode & RCV_SHUTDOWN ? 1078 VIRTIO_VSOCK_SHUTDOWN_RCV : 0) | 1079 (mode & SEND_SHUTDOWN ? 1080 VIRTIO_VSOCK_SHUTDOWN_SEND : 0), 1081 .vsk = vsk, 1082 }; 1083 1084 return virtio_transport_send_pkt_info(vsk, &info); 1085 } 1086 EXPORT_SYMBOL_GPL(virtio_transport_shutdown); 1087 1088 int 1089 virtio_transport_dgram_enqueue(struct vsock_sock *vsk, 1090 struct sockaddr_vm *remote_addr, 1091 struct msghdr *msg, 1092 size_t dgram_len) 1093 { 1094 return -EOPNOTSUPP; 1095 } 1096 EXPORT_SYMBOL_GPL(virtio_transport_dgram_enqueue); 1097 1098 ssize_t 1099 virtio_transport_stream_enqueue(struct vsock_sock *vsk, 1100 struct msghdr *msg, 1101 size_t len) 1102 { 1103 struct virtio_vsock_pkt_info info = { 1104 .op = VIRTIO_VSOCK_OP_RW, 1105 .msg = msg, 1106 .pkt_len = len, 1107 .vsk = vsk, 1108 }; 1109 1110 return virtio_transport_send_pkt_info(vsk, &info); 1111 } 1112 EXPORT_SYMBOL_GPL(virtio_transport_stream_enqueue); 1113 1114 void virtio_transport_destruct(struct vsock_sock *vsk) 1115 { 1116 struct virtio_vsock_sock *vvs = vsk->trans; 1117 1118 virtio_transport_cancel_close_work(vsk, true); 1119 1120 kfree(vvs); 1121 vsk->trans = NULL; 1122 } 1123 EXPORT_SYMBOL_GPL(virtio_transport_destruct); 1124 1125 ssize_t virtio_transport_unsent_bytes(struct vsock_sock *vsk) 1126 { 1127 struct virtio_vsock_sock *vvs = vsk->trans; 1128 size_t ret; 1129 1130 spin_lock_bh(&vvs->tx_lock); 1131 ret = vvs->bytes_unsent; 1132 spin_unlock_bh(&vvs->tx_lock); 1133 1134 return ret; 1135 } 1136 EXPORT_SYMBOL_GPL(virtio_transport_unsent_bytes); 1137 1138 static int virtio_transport_reset(struct vsock_sock *vsk, 1139 struct sk_buff *skb) 1140 { 1141 struct virtio_vsock_pkt_info info = { 1142 .op = VIRTIO_VSOCK_OP_RST, 1143 .reply = !!skb, 1144 .vsk = vsk, 1145 }; 1146 1147 /* Send RST only if the original pkt is not a RST pkt */ 1148 if (skb && le16_to_cpu(virtio_vsock_hdr(skb)->op) == VIRTIO_VSOCK_OP_RST) 1149 return 0; 1150 1151 return virtio_transport_send_pkt_info(vsk, &info); 1152 } 1153 1154 /* Normally packets are associated with a socket. There may be no socket if an 1155 * attempt was made to connect to a socket that does not exist. 1156 */ 1157 static int virtio_transport_reset_no_sock(const struct virtio_transport *t, 1158 struct sk_buff *skb) 1159 { 1160 struct virtio_vsock_hdr *hdr = virtio_vsock_hdr(skb); 1161 struct virtio_vsock_pkt_info info = { 1162 .op = VIRTIO_VSOCK_OP_RST, 1163 .type = le16_to_cpu(hdr->type), 1164 .reply = true, 1165 }; 1166 struct sk_buff *reply; 1167 1168 /* Send RST only if the original pkt is not a RST pkt */ 1169 if (le16_to_cpu(hdr->op) == VIRTIO_VSOCK_OP_RST) 1170 return 0; 1171 1172 if (!t) 1173 return -ENOTCONN; 1174 1175 reply = virtio_transport_alloc_skb(&info, 0, false, 1176 le64_to_cpu(hdr->dst_cid), 1177 le32_to_cpu(hdr->dst_port), 1178 le64_to_cpu(hdr->src_cid), 1179 le32_to_cpu(hdr->src_port)); 1180 if (!reply) 1181 return -ENOMEM; 1182 1183 return t->send_pkt(reply); 1184 } 1185 1186 /* This function should be called with sk_lock held and SOCK_DONE set */ 1187 static void virtio_transport_remove_sock(struct vsock_sock *vsk) 1188 { 1189 struct virtio_vsock_sock *vvs = vsk->trans; 1190 1191 /* We don't need to take rx_lock, as the socket is closing and we are 1192 * removing it. 1193 */ 1194 __skb_queue_purge(&vvs->rx_queue); 1195 vsock_remove_sock(vsk); 1196 } 1197 1198 static void virtio_transport_cancel_close_work(struct vsock_sock *vsk, 1199 bool cancel_timeout) 1200 { 1201 struct sock *sk = sk_vsock(vsk); 1202 1203 if (vsk->close_work_scheduled && 1204 (!cancel_timeout || cancel_delayed_work(&vsk->close_work))) { 1205 vsk->close_work_scheduled = false; 1206 1207 virtio_transport_remove_sock(vsk); 1208 1209 /* Release refcnt obtained when we scheduled the timeout */ 1210 sock_put(sk); 1211 } 1212 } 1213 1214 static void virtio_transport_do_close(struct vsock_sock *vsk, 1215 bool cancel_timeout) 1216 { 1217 struct sock *sk = sk_vsock(vsk); 1218 1219 sock_set_flag(sk, SOCK_DONE); 1220 vsk->peer_shutdown = SHUTDOWN_MASK; 1221 if (vsock_stream_has_data(vsk) <= 0) 1222 sk->sk_state = TCP_CLOSING; 1223 sk->sk_state_change(sk); 1224 1225 virtio_transport_cancel_close_work(vsk, cancel_timeout); 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 (!(current->flags & PF_EXITING)) 1268 vsock_linger(sk); 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 or assigned to another transport before 1631 * lock_sock (note: listener sockets are not assigned to any transport) 1632 */ 1633 if (sock_flag(sk, SOCK_DONE) || 1634 (sk->sk_state != TCP_LISTEN && vsk->transport != &t->transport)) { 1635 (void)virtio_transport_reset_no_sock(t, skb); 1636 release_sock(sk); 1637 sock_put(sk); 1638 goto free_pkt; 1639 } 1640 1641 space_available = virtio_transport_space_update(sk, skb); 1642 1643 /* Update CID in case it has changed after a transport reset event */ 1644 if (vsk->local_addr.svm_cid != VMADDR_CID_ANY) 1645 vsk->local_addr.svm_cid = dst.svm_cid; 1646 1647 if (space_available) 1648 sk->sk_write_space(sk); 1649 1650 switch (sk->sk_state) { 1651 case TCP_LISTEN: 1652 virtio_transport_recv_listen(sk, skb, t); 1653 kfree_skb(skb); 1654 break; 1655 case TCP_SYN_SENT: 1656 virtio_transport_recv_connecting(sk, skb); 1657 kfree_skb(skb); 1658 break; 1659 case TCP_ESTABLISHED: 1660 virtio_transport_recv_connected(sk, skb); 1661 break; 1662 case TCP_CLOSING: 1663 virtio_transport_recv_disconnecting(sk, skb); 1664 kfree_skb(skb); 1665 break; 1666 default: 1667 (void)virtio_transport_reset_no_sock(t, skb); 1668 kfree_skb(skb); 1669 break; 1670 } 1671 1672 release_sock(sk); 1673 1674 /* Release refcnt obtained when we fetched this socket out of the 1675 * bound or connected list. 1676 */ 1677 sock_put(sk); 1678 return; 1679 1680 free_pkt: 1681 kfree_skb(skb); 1682 } 1683 EXPORT_SYMBOL_GPL(virtio_transport_recv_pkt); 1684 1685 /* Remove skbs found in a queue that have a vsk that matches. 1686 * 1687 * Each skb is freed. 1688 * 1689 * Returns the count of skbs that were reply packets. 1690 */ 1691 int virtio_transport_purge_skbs(void *vsk, struct sk_buff_head *queue) 1692 { 1693 struct sk_buff_head freeme; 1694 struct sk_buff *skb, *tmp; 1695 int cnt = 0; 1696 1697 skb_queue_head_init(&freeme); 1698 1699 spin_lock_bh(&queue->lock); 1700 skb_queue_walk_safe(queue, skb, tmp) { 1701 if (vsock_sk(skb->sk) != vsk) 1702 continue; 1703 1704 __skb_unlink(skb, queue); 1705 __skb_queue_tail(&freeme, skb); 1706 1707 if (virtio_vsock_skb_reply(skb)) 1708 cnt++; 1709 } 1710 spin_unlock_bh(&queue->lock); 1711 1712 __skb_queue_purge(&freeme); 1713 1714 return cnt; 1715 } 1716 EXPORT_SYMBOL_GPL(virtio_transport_purge_skbs); 1717 1718 int virtio_transport_read_skb(struct vsock_sock *vsk, skb_read_actor_t recv_actor) 1719 { 1720 struct virtio_vsock_sock *vvs = vsk->trans; 1721 struct sock *sk = sk_vsock(vsk); 1722 struct virtio_vsock_hdr *hdr; 1723 struct sk_buff *skb; 1724 u32 pkt_len; 1725 int off = 0; 1726 int err; 1727 1728 spin_lock_bh(&vvs->rx_lock); 1729 /* Use __skb_recv_datagram() for race-free handling of the receive. It 1730 * works for types other than dgrams. 1731 */ 1732 skb = __skb_recv_datagram(sk, &vvs->rx_queue, MSG_DONTWAIT, &off, &err); 1733 if (!skb) { 1734 spin_unlock_bh(&vvs->rx_lock); 1735 return err; 1736 } 1737 1738 hdr = virtio_vsock_hdr(skb); 1739 if (le32_to_cpu(hdr->flags) & VIRTIO_VSOCK_SEQ_EOM) 1740 vvs->msg_count--; 1741 1742 pkt_len = le32_to_cpu(hdr->len); 1743 virtio_transport_dec_rx_pkt(vvs, pkt_len, pkt_len); 1744 spin_unlock_bh(&vvs->rx_lock); 1745 1746 virtio_transport_send_credit_update(vsk); 1747 1748 return recv_actor(sk, skb); 1749 } 1750 EXPORT_SYMBOL_GPL(virtio_transport_read_skb); 1751 1752 int virtio_transport_notify_set_rcvlowat(struct vsock_sock *vsk, int val) 1753 { 1754 struct virtio_vsock_sock *vvs = vsk->trans; 1755 bool send_update; 1756 1757 spin_lock_bh(&vvs->rx_lock); 1758 1759 /* If number of available bytes is less than new SO_RCVLOWAT value, 1760 * kick sender to send more data, because sender may sleep in its 1761 * 'send()' syscall waiting for enough space at our side. Also 1762 * don't send credit update when peer already knows actual value - 1763 * such transmission will be useless. 1764 */ 1765 send_update = (vvs->rx_bytes < val) && 1766 (vvs->fwd_cnt != vvs->last_fwd_cnt); 1767 1768 spin_unlock_bh(&vvs->rx_lock); 1769 1770 if (send_update) { 1771 int err; 1772 1773 err = virtio_transport_send_credit_update(vsk); 1774 if (err < 0) 1775 return err; 1776 } 1777 1778 return 0; 1779 } 1780 EXPORT_SYMBOL_GPL(virtio_transport_notify_set_rcvlowat); 1781 1782 MODULE_LICENSE("GPL v2"); 1783 MODULE_AUTHOR("Asias He"); 1784 MODULE_DESCRIPTION("common code for virtio vsock"); 1785