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); 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, 111 len); 112 113 return memcpy_from_msg(skb_put(skb, len), info->msg, len); 114 } 115 116 static void virtio_transport_init_hdr(struct sk_buff *skb, 117 struct virtio_vsock_pkt_info *info, 118 size_t payload_len, 119 u32 src_cid, 120 u32 src_port, 121 u32 dst_cid, 122 u32 dst_port) 123 { 124 struct virtio_vsock_hdr *hdr; 125 126 hdr = virtio_vsock_hdr(skb); 127 hdr->type = cpu_to_le16(info->type); 128 hdr->op = cpu_to_le16(info->op); 129 hdr->src_cid = cpu_to_le64(src_cid); 130 hdr->dst_cid = cpu_to_le64(dst_cid); 131 hdr->src_port = cpu_to_le32(src_port); 132 hdr->dst_port = cpu_to_le32(dst_port); 133 hdr->flags = cpu_to_le32(info->flags); 134 hdr->len = cpu_to_le32(payload_len); 135 hdr->buf_alloc = cpu_to_le32(0); 136 hdr->fwd_cnt = cpu_to_le32(0); 137 } 138 139 static void virtio_transport_copy_nonlinear_skb(const struct sk_buff *skb, 140 void *dst, 141 size_t len) 142 { 143 struct iov_iter iov_iter = { 0 }; 144 struct kvec kvec; 145 size_t to_copy; 146 147 kvec.iov_base = dst; 148 kvec.iov_len = len; 149 150 iov_iter.iter_type = ITER_KVEC; 151 iov_iter.kvec = &kvec; 152 iov_iter.nr_segs = 1; 153 154 to_copy = min_t(size_t, len, skb->len); 155 156 skb_copy_datagram_iter(skb, VIRTIO_VSOCK_SKB_CB(skb)->offset, 157 &iov_iter, to_copy); 158 } 159 160 /* Packet capture */ 161 static struct sk_buff *virtio_transport_build_skb(void *opaque) 162 { 163 struct virtio_vsock_hdr *pkt_hdr; 164 struct sk_buff *pkt = opaque; 165 struct af_vsockmon_hdr *hdr; 166 struct sk_buff *skb; 167 size_t payload_len; 168 169 /* A packet could be split to fit the RX buffer, so we can retrieve 170 * the payload length from the header and the buffer pointer taking 171 * care of the offset in the original packet. 172 */ 173 pkt_hdr = virtio_vsock_hdr(pkt); 174 payload_len = pkt->len; 175 176 skb = alloc_skb(sizeof(*hdr) + sizeof(*pkt_hdr) + payload_len, 177 GFP_ATOMIC); 178 if (!skb) 179 return NULL; 180 181 hdr = skb_put(skb, sizeof(*hdr)); 182 183 /* pkt->hdr is little-endian so no need to byteswap here */ 184 hdr->src_cid = pkt_hdr->src_cid; 185 hdr->src_port = pkt_hdr->src_port; 186 hdr->dst_cid = pkt_hdr->dst_cid; 187 hdr->dst_port = pkt_hdr->dst_port; 188 189 hdr->transport = cpu_to_le16(AF_VSOCK_TRANSPORT_VIRTIO); 190 hdr->len = cpu_to_le16(sizeof(*pkt_hdr)); 191 memset(hdr->reserved, 0, sizeof(hdr->reserved)); 192 193 switch (le16_to_cpu(pkt_hdr->op)) { 194 case VIRTIO_VSOCK_OP_REQUEST: 195 case VIRTIO_VSOCK_OP_RESPONSE: 196 hdr->op = cpu_to_le16(AF_VSOCK_OP_CONNECT); 197 break; 198 case VIRTIO_VSOCK_OP_RST: 199 case VIRTIO_VSOCK_OP_SHUTDOWN: 200 hdr->op = cpu_to_le16(AF_VSOCK_OP_DISCONNECT); 201 break; 202 case VIRTIO_VSOCK_OP_RW: 203 hdr->op = cpu_to_le16(AF_VSOCK_OP_PAYLOAD); 204 break; 205 case VIRTIO_VSOCK_OP_CREDIT_UPDATE: 206 case VIRTIO_VSOCK_OP_CREDIT_REQUEST: 207 hdr->op = cpu_to_le16(AF_VSOCK_OP_CONTROL); 208 break; 209 default: 210 hdr->op = cpu_to_le16(AF_VSOCK_OP_UNKNOWN); 211 break; 212 } 213 214 skb_put_data(skb, pkt_hdr, sizeof(*pkt_hdr)); 215 216 if (payload_len) { 217 if (skb_is_nonlinear(pkt)) { 218 void *data = skb_put(skb, payload_len); 219 220 virtio_transport_copy_nonlinear_skb(pkt, data, payload_len); 221 } else { 222 skb_put_data(skb, pkt->data, payload_len); 223 } 224 } 225 226 return skb; 227 } 228 229 void virtio_transport_deliver_tap_pkt(struct sk_buff *skb) 230 { 231 if (virtio_vsock_skb_tap_delivered(skb)) 232 return; 233 234 vsock_deliver_tap(virtio_transport_build_skb, skb); 235 virtio_vsock_skb_set_tap_delivered(skb); 236 } 237 EXPORT_SYMBOL_GPL(virtio_transport_deliver_tap_pkt); 238 239 static u16 virtio_transport_get_type(struct sock *sk) 240 { 241 if (sk->sk_type == SOCK_STREAM) 242 return VIRTIO_VSOCK_TYPE_STREAM; 243 else 244 return VIRTIO_VSOCK_TYPE_SEQPACKET; 245 } 246 247 /* Returns new sk_buff on success, otherwise returns NULL. */ 248 static struct sk_buff *virtio_transport_alloc_skb(struct virtio_vsock_pkt_info *info, 249 size_t payload_len, 250 bool zcopy, 251 u32 src_cid, 252 u32 src_port, 253 u32 dst_cid, 254 u32 dst_port) 255 { 256 struct vsock_sock *vsk; 257 struct sk_buff *skb; 258 size_t skb_len; 259 260 skb_len = VIRTIO_VSOCK_SKB_HEADROOM; 261 262 if (!zcopy) 263 skb_len += payload_len; 264 265 skb = virtio_vsock_alloc_skb(skb_len, GFP_KERNEL); 266 if (!skb) 267 return NULL; 268 269 virtio_transport_init_hdr(skb, info, payload_len, src_cid, src_port, 270 dst_cid, dst_port); 271 272 vsk = info->vsk; 273 274 /* If 'vsk' != NULL then payload is always present, so we 275 * will never call '__zerocopy_sg_from_iter()' below without 276 * setting skb owner in 'skb_set_owner_w()'. The only case 277 * when 'vsk' == NULL is VIRTIO_VSOCK_OP_RST control message 278 * without payload. 279 */ 280 WARN_ON_ONCE(!(vsk && (info->msg && payload_len)) && zcopy); 281 282 /* Set owner here, because '__zerocopy_sg_from_iter()' uses 283 * owner of skb without check to update 'sk_wmem_alloc'. 284 */ 285 if (vsk) 286 skb_set_owner_w(skb, sk_vsock(vsk)); 287 288 if (info->msg && payload_len > 0) { 289 int err; 290 291 err = virtio_transport_fill_skb(skb, info, payload_len, zcopy); 292 if (err) 293 goto out; 294 295 if (msg_data_left(info->msg) == 0 && 296 info->type == VIRTIO_VSOCK_TYPE_SEQPACKET) { 297 struct virtio_vsock_hdr *hdr = virtio_vsock_hdr(skb); 298 299 hdr->flags |= cpu_to_le32(VIRTIO_VSOCK_SEQ_EOM); 300 301 if (info->msg->msg_flags & MSG_EOR) 302 hdr->flags |= cpu_to_le32(VIRTIO_VSOCK_SEQ_EOR); 303 } 304 } 305 306 if (info->reply) 307 virtio_vsock_skb_set_reply(skb); 308 309 trace_virtio_transport_alloc_pkt(src_cid, src_port, 310 dst_cid, dst_port, 311 payload_len, 312 info->type, 313 info->op, 314 info->flags, 315 zcopy); 316 317 return skb; 318 out: 319 kfree_skb(skb); 320 return NULL; 321 } 322 323 /* This function can only be used on connecting/connected sockets, 324 * since a socket assigned to a transport is required. 325 * 326 * Do not use on listener sockets! 327 */ 328 static int virtio_transport_send_pkt_info(struct vsock_sock *vsk, 329 struct virtio_vsock_pkt_info *info) 330 { 331 u32 max_skb_len = VIRTIO_VSOCK_MAX_PKT_BUF_SIZE; 332 u32 src_cid, src_port, dst_cid, dst_port; 333 const struct virtio_transport *t_ops; 334 struct virtio_vsock_sock *vvs; 335 u32 pkt_len = info->pkt_len; 336 bool can_zcopy = false; 337 u32 rest_len; 338 int ret; 339 340 info->type = virtio_transport_get_type(sk_vsock(vsk)); 341 342 t_ops = virtio_transport_get_ops(vsk); 343 if (unlikely(!t_ops)) 344 return -EFAULT; 345 346 src_cid = t_ops->transport.get_local_cid(); 347 src_port = vsk->local_addr.svm_port; 348 if (!info->remote_cid) { 349 dst_cid = vsk->remote_addr.svm_cid; 350 dst_port = vsk->remote_addr.svm_port; 351 } else { 352 dst_cid = info->remote_cid; 353 dst_port = info->remote_port; 354 } 355 356 vvs = vsk->trans; 357 358 /* virtio_transport_get_credit might return less than pkt_len credit */ 359 pkt_len = virtio_transport_get_credit(vvs, pkt_len); 360 361 /* Do not send zero length OP_RW pkt */ 362 if (pkt_len == 0 && info->op == VIRTIO_VSOCK_OP_RW) 363 return pkt_len; 364 365 if (info->msg) { 366 /* If zerocopy is not enabled by 'setsockopt()', we behave as 367 * there is no MSG_ZEROCOPY flag set. 368 */ 369 if (!sock_flag(sk_vsock(vsk), SOCK_ZEROCOPY)) 370 info->msg->msg_flags &= ~MSG_ZEROCOPY; 371 372 if (info->msg->msg_flags & MSG_ZEROCOPY) 373 can_zcopy = virtio_transport_can_zcopy(t_ops, info, pkt_len); 374 375 if (can_zcopy) 376 max_skb_len = min_t(u32, VIRTIO_VSOCK_MAX_PKT_BUF_SIZE, 377 (MAX_SKB_FRAGS * PAGE_SIZE)); 378 } 379 380 rest_len = pkt_len; 381 382 do { 383 struct sk_buff *skb; 384 size_t skb_len; 385 386 skb_len = min(max_skb_len, rest_len); 387 388 skb = virtio_transport_alloc_skb(info, skb_len, can_zcopy, 389 src_cid, src_port, 390 dst_cid, dst_port); 391 if (!skb) { 392 ret = -ENOMEM; 393 break; 394 } 395 396 /* We process buffer part by part, allocating skb on 397 * each iteration. If this is last skb for this buffer 398 * and MSG_ZEROCOPY mode is in use - we must allocate 399 * completion for the current syscall. 400 */ 401 if (info->msg && info->msg->msg_flags & MSG_ZEROCOPY && 402 skb_len == rest_len && info->op == VIRTIO_VSOCK_OP_RW) { 403 if (virtio_transport_init_zcopy_skb(vsk, skb, 404 info->msg, 405 can_zcopy)) { 406 kfree_skb(skb); 407 ret = -ENOMEM; 408 break; 409 } 410 } 411 412 virtio_transport_inc_tx_pkt(vvs, skb); 413 414 ret = t_ops->send_pkt(skb); 415 if (ret < 0) 416 break; 417 418 /* Both virtio and vhost 'send_pkt()' returns 'skb_len', 419 * but for reliability use 'ret' instead of 'skb_len'. 420 * Also if partial send happens (e.g. 'ret' != 'skb_len') 421 * somehow, we break this loop, but account such returned 422 * value in 'virtio_transport_put_credit()'. 423 */ 424 rest_len -= ret; 425 426 if (WARN_ONCE(ret != skb_len, 427 "'send_pkt()' returns %i, but %zu expected\n", 428 ret, skb_len)) 429 break; 430 } while (rest_len); 431 432 virtio_transport_put_credit(vvs, rest_len); 433 434 /* Return number of bytes, if any data has been sent. */ 435 if (rest_len != pkt_len) 436 ret = pkt_len - rest_len; 437 438 return ret; 439 } 440 441 static bool virtio_transport_inc_rx_pkt(struct virtio_vsock_sock *vvs, 442 u32 len) 443 { 444 if (vvs->rx_bytes + len > vvs->buf_alloc) 445 return false; 446 447 vvs->rx_bytes += len; 448 return true; 449 } 450 451 static void virtio_transport_dec_rx_pkt(struct virtio_vsock_sock *vvs, 452 u32 len) 453 { 454 vvs->rx_bytes -= len; 455 vvs->fwd_cnt += len; 456 } 457 458 void virtio_transport_inc_tx_pkt(struct virtio_vsock_sock *vvs, struct sk_buff *skb) 459 { 460 struct virtio_vsock_hdr *hdr = virtio_vsock_hdr(skb); 461 462 spin_lock_bh(&vvs->rx_lock); 463 vvs->last_fwd_cnt = vvs->fwd_cnt; 464 hdr->fwd_cnt = cpu_to_le32(vvs->fwd_cnt); 465 hdr->buf_alloc = cpu_to_le32(vvs->buf_alloc); 466 spin_unlock_bh(&vvs->rx_lock); 467 } 468 EXPORT_SYMBOL_GPL(virtio_transport_inc_tx_pkt); 469 470 void virtio_transport_consume_skb_sent(struct sk_buff *skb, bool consume) 471 { 472 struct sock *s = skb->sk; 473 474 if (s && skb->len) { 475 struct vsock_sock *vs = vsock_sk(s); 476 struct virtio_vsock_sock *vvs; 477 478 vvs = vs->trans; 479 480 spin_lock_bh(&vvs->tx_lock); 481 vvs->bytes_unsent -= skb->len; 482 spin_unlock_bh(&vvs->tx_lock); 483 } 484 485 if (consume) 486 consume_skb(skb); 487 } 488 EXPORT_SYMBOL_GPL(virtio_transport_consume_skb_sent); 489 490 u32 virtio_transport_get_credit(struct virtio_vsock_sock *vvs, u32 credit) 491 { 492 u32 ret; 493 494 if (!credit) 495 return 0; 496 497 spin_lock_bh(&vvs->tx_lock); 498 ret = vvs->peer_buf_alloc - (vvs->tx_cnt - vvs->peer_fwd_cnt); 499 if (ret > credit) 500 ret = credit; 501 vvs->tx_cnt += ret; 502 vvs->bytes_unsent += ret; 503 spin_unlock_bh(&vvs->tx_lock); 504 505 return ret; 506 } 507 EXPORT_SYMBOL_GPL(virtio_transport_get_credit); 508 509 void virtio_transport_put_credit(struct virtio_vsock_sock *vvs, u32 credit) 510 { 511 if (!credit) 512 return; 513 514 spin_lock_bh(&vvs->tx_lock); 515 vvs->tx_cnt -= credit; 516 vvs->bytes_unsent -= credit; 517 spin_unlock_bh(&vvs->tx_lock); 518 } 519 EXPORT_SYMBOL_GPL(virtio_transport_put_credit); 520 521 static int virtio_transport_send_credit_update(struct vsock_sock *vsk) 522 { 523 struct virtio_vsock_pkt_info info = { 524 .op = VIRTIO_VSOCK_OP_CREDIT_UPDATE, 525 .vsk = vsk, 526 }; 527 528 return virtio_transport_send_pkt_info(vsk, &info); 529 } 530 531 static ssize_t 532 virtio_transport_stream_do_peek(struct vsock_sock *vsk, 533 struct msghdr *msg, 534 size_t len) 535 { 536 struct virtio_vsock_sock *vvs = vsk->trans; 537 struct sk_buff *skb; 538 size_t total = 0; 539 int err; 540 541 spin_lock_bh(&vvs->rx_lock); 542 543 skb_queue_walk(&vvs->rx_queue, skb) { 544 size_t bytes; 545 546 bytes = len - total; 547 if (bytes > skb->len) 548 bytes = skb->len; 549 550 spin_unlock_bh(&vvs->rx_lock); 551 552 /* sk_lock is held by caller so no one else can dequeue. 553 * Unlock rx_lock since skb_copy_datagram_iter() may sleep. 554 */ 555 err = skb_copy_datagram_iter(skb, VIRTIO_VSOCK_SKB_CB(skb)->offset, 556 &msg->msg_iter, bytes); 557 if (err) 558 goto out; 559 560 total += bytes; 561 562 spin_lock_bh(&vvs->rx_lock); 563 564 if (total == len) 565 break; 566 } 567 568 spin_unlock_bh(&vvs->rx_lock); 569 570 return total; 571 572 out: 573 if (total) 574 err = total; 575 return err; 576 } 577 578 static ssize_t 579 virtio_transport_stream_do_dequeue(struct vsock_sock *vsk, 580 struct msghdr *msg, 581 size_t len) 582 { 583 struct virtio_vsock_sock *vvs = vsk->trans; 584 size_t bytes, total = 0; 585 struct sk_buff *skb; 586 u32 fwd_cnt_delta; 587 bool low_rx_bytes; 588 int err = -EFAULT; 589 u32 free_space; 590 591 spin_lock_bh(&vvs->rx_lock); 592 593 if (WARN_ONCE(skb_queue_empty(&vvs->rx_queue) && vvs->rx_bytes, 594 "rx_queue is empty, but rx_bytes is non-zero\n")) { 595 spin_unlock_bh(&vvs->rx_lock); 596 return err; 597 } 598 599 while (total < len && !skb_queue_empty(&vvs->rx_queue)) { 600 skb = skb_peek(&vvs->rx_queue); 601 602 bytes = min_t(size_t, len - total, 603 skb->len - VIRTIO_VSOCK_SKB_CB(skb)->offset); 604 605 /* sk_lock is held by caller so no one else can dequeue. 606 * Unlock rx_lock since skb_copy_datagram_iter() may sleep. 607 */ 608 spin_unlock_bh(&vvs->rx_lock); 609 610 err = skb_copy_datagram_iter(skb, 611 VIRTIO_VSOCK_SKB_CB(skb)->offset, 612 &msg->msg_iter, bytes); 613 if (err) 614 goto out; 615 616 spin_lock_bh(&vvs->rx_lock); 617 618 total += bytes; 619 620 VIRTIO_VSOCK_SKB_CB(skb)->offset += bytes; 621 622 if (skb->len == VIRTIO_VSOCK_SKB_CB(skb)->offset) { 623 u32 pkt_len = le32_to_cpu(virtio_vsock_hdr(skb)->len); 624 625 virtio_transport_dec_rx_pkt(vvs, pkt_len); 626 __skb_unlink(skb, &vvs->rx_queue); 627 consume_skb(skb); 628 } 629 } 630 631 fwd_cnt_delta = vvs->fwd_cnt - vvs->last_fwd_cnt; 632 free_space = vvs->buf_alloc - fwd_cnt_delta; 633 low_rx_bytes = (vvs->rx_bytes < 634 sock_rcvlowat(sk_vsock(vsk), 0, INT_MAX)); 635 636 spin_unlock_bh(&vvs->rx_lock); 637 638 /* To reduce the number of credit update messages, 639 * don't update credits as long as lots of space is available. 640 * Note: the limit chosen here is arbitrary. Setting the limit 641 * too high causes extra messages. Too low causes transmitter 642 * stalls. As stalls are in theory more expensive than extra 643 * messages, we set the limit to a high value. TODO: experiment 644 * with different values. Also send credit update message when 645 * number of bytes in rx queue is not enough to wake up reader. 646 */ 647 if (fwd_cnt_delta && 648 (free_space < VIRTIO_VSOCK_MAX_PKT_BUF_SIZE || low_rx_bytes)) 649 virtio_transport_send_credit_update(vsk); 650 651 return total; 652 653 out: 654 if (total) 655 err = total; 656 return err; 657 } 658 659 static ssize_t 660 virtio_transport_seqpacket_do_peek(struct vsock_sock *vsk, 661 struct msghdr *msg) 662 { 663 struct virtio_vsock_sock *vvs = vsk->trans; 664 struct sk_buff *skb; 665 size_t total, len; 666 667 spin_lock_bh(&vvs->rx_lock); 668 669 if (!vvs->msg_count) { 670 spin_unlock_bh(&vvs->rx_lock); 671 return 0; 672 } 673 674 total = 0; 675 len = msg_data_left(msg); 676 677 skb_queue_walk(&vvs->rx_queue, skb) { 678 struct virtio_vsock_hdr *hdr; 679 680 if (total < len) { 681 size_t bytes; 682 int err; 683 684 bytes = len - total; 685 if (bytes > skb->len) 686 bytes = skb->len; 687 688 spin_unlock_bh(&vvs->rx_lock); 689 690 /* sk_lock is held by caller so no one else can dequeue. 691 * Unlock rx_lock since skb_copy_datagram_iter() may sleep. 692 */ 693 err = skb_copy_datagram_iter(skb, VIRTIO_VSOCK_SKB_CB(skb)->offset, 694 &msg->msg_iter, bytes); 695 if (err) 696 return err; 697 698 spin_lock_bh(&vvs->rx_lock); 699 } 700 701 total += skb->len; 702 hdr = virtio_vsock_hdr(skb); 703 704 if (le32_to_cpu(hdr->flags) & VIRTIO_VSOCK_SEQ_EOM) { 705 if (le32_to_cpu(hdr->flags) & VIRTIO_VSOCK_SEQ_EOR) 706 msg->msg_flags |= MSG_EOR; 707 708 break; 709 } 710 } 711 712 spin_unlock_bh(&vvs->rx_lock); 713 714 return total; 715 } 716 717 static int virtio_transport_seqpacket_do_dequeue(struct vsock_sock *vsk, 718 struct msghdr *msg, 719 int flags) 720 { 721 struct virtio_vsock_sock *vvs = vsk->trans; 722 int dequeued_len = 0; 723 size_t user_buf_len = msg_data_left(msg); 724 bool msg_ready = false; 725 struct sk_buff *skb; 726 727 spin_lock_bh(&vvs->rx_lock); 728 729 if (vvs->msg_count == 0) { 730 spin_unlock_bh(&vvs->rx_lock); 731 return 0; 732 } 733 734 while (!msg_ready) { 735 struct virtio_vsock_hdr *hdr; 736 size_t pkt_len; 737 738 skb = __skb_dequeue(&vvs->rx_queue); 739 if (!skb) 740 break; 741 hdr = virtio_vsock_hdr(skb); 742 pkt_len = (size_t)le32_to_cpu(hdr->len); 743 744 if (dequeued_len >= 0) { 745 size_t bytes_to_copy; 746 747 bytes_to_copy = min(user_buf_len, pkt_len); 748 749 if (bytes_to_copy) { 750 int err; 751 752 /* sk_lock is held by caller so no one else can dequeue. 753 * Unlock rx_lock since skb_copy_datagram_iter() may sleep. 754 */ 755 spin_unlock_bh(&vvs->rx_lock); 756 757 err = skb_copy_datagram_iter(skb, 0, 758 &msg->msg_iter, 759 bytes_to_copy); 760 if (err) { 761 /* Copy of message failed. Rest of 762 * fragments will be freed without copy. 763 */ 764 dequeued_len = err; 765 } else { 766 user_buf_len -= bytes_to_copy; 767 } 768 769 spin_lock_bh(&vvs->rx_lock); 770 } 771 772 if (dequeued_len >= 0) 773 dequeued_len += pkt_len; 774 } 775 776 if (le32_to_cpu(hdr->flags) & VIRTIO_VSOCK_SEQ_EOM) { 777 msg_ready = true; 778 vvs->msg_count--; 779 780 if (le32_to_cpu(hdr->flags) & VIRTIO_VSOCK_SEQ_EOR) 781 msg->msg_flags |= MSG_EOR; 782 } 783 784 virtio_transport_dec_rx_pkt(vvs, pkt_len); 785 kfree_skb(skb); 786 } 787 788 spin_unlock_bh(&vvs->rx_lock); 789 790 virtio_transport_send_credit_update(vsk); 791 792 return dequeued_len; 793 } 794 795 ssize_t 796 virtio_transport_stream_dequeue(struct vsock_sock *vsk, 797 struct msghdr *msg, 798 size_t len, int flags) 799 { 800 if (flags & MSG_PEEK) 801 return virtio_transport_stream_do_peek(vsk, msg, len); 802 else 803 return virtio_transport_stream_do_dequeue(vsk, msg, len); 804 } 805 EXPORT_SYMBOL_GPL(virtio_transport_stream_dequeue); 806 807 ssize_t 808 virtio_transport_seqpacket_dequeue(struct vsock_sock *vsk, 809 struct msghdr *msg, 810 int flags) 811 { 812 if (flags & MSG_PEEK) 813 return virtio_transport_seqpacket_do_peek(vsk, msg); 814 else 815 return virtio_transport_seqpacket_do_dequeue(vsk, msg, flags); 816 } 817 EXPORT_SYMBOL_GPL(virtio_transport_seqpacket_dequeue); 818 819 int 820 virtio_transport_seqpacket_enqueue(struct vsock_sock *vsk, 821 struct msghdr *msg, 822 size_t len) 823 { 824 struct virtio_vsock_sock *vvs = vsk->trans; 825 826 spin_lock_bh(&vvs->tx_lock); 827 828 if (len > vvs->peer_buf_alloc) { 829 spin_unlock_bh(&vvs->tx_lock); 830 return -EMSGSIZE; 831 } 832 833 spin_unlock_bh(&vvs->tx_lock); 834 835 return virtio_transport_stream_enqueue(vsk, msg, len); 836 } 837 EXPORT_SYMBOL_GPL(virtio_transport_seqpacket_enqueue); 838 839 int 840 virtio_transport_dgram_dequeue(struct vsock_sock *vsk, 841 struct msghdr *msg, 842 size_t len, int flags) 843 { 844 return -EOPNOTSUPP; 845 } 846 EXPORT_SYMBOL_GPL(virtio_transport_dgram_dequeue); 847 848 s64 virtio_transport_stream_has_data(struct vsock_sock *vsk) 849 { 850 struct virtio_vsock_sock *vvs = vsk->trans; 851 s64 bytes; 852 853 spin_lock_bh(&vvs->rx_lock); 854 bytes = vvs->rx_bytes; 855 spin_unlock_bh(&vvs->rx_lock); 856 857 return bytes; 858 } 859 EXPORT_SYMBOL_GPL(virtio_transport_stream_has_data); 860 861 u32 virtio_transport_seqpacket_has_data(struct vsock_sock *vsk) 862 { 863 struct virtio_vsock_sock *vvs = vsk->trans; 864 u32 msg_count; 865 866 spin_lock_bh(&vvs->rx_lock); 867 msg_count = vvs->msg_count; 868 spin_unlock_bh(&vvs->rx_lock); 869 870 return msg_count; 871 } 872 EXPORT_SYMBOL_GPL(virtio_transport_seqpacket_has_data); 873 874 static s64 virtio_transport_has_space(struct vsock_sock *vsk) 875 { 876 struct virtio_vsock_sock *vvs = vsk->trans; 877 s64 bytes; 878 879 bytes = (s64)vvs->peer_buf_alloc - (vvs->tx_cnt - vvs->peer_fwd_cnt); 880 if (bytes < 0) 881 bytes = 0; 882 883 return bytes; 884 } 885 886 s64 virtio_transport_stream_has_space(struct vsock_sock *vsk) 887 { 888 struct virtio_vsock_sock *vvs = vsk->trans; 889 s64 bytes; 890 891 spin_lock_bh(&vvs->tx_lock); 892 bytes = virtio_transport_has_space(vsk); 893 spin_unlock_bh(&vvs->tx_lock); 894 895 return bytes; 896 } 897 EXPORT_SYMBOL_GPL(virtio_transport_stream_has_space); 898 899 int virtio_transport_do_socket_init(struct vsock_sock *vsk, 900 struct vsock_sock *psk) 901 { 902 struct virtio_vsock_sock *vvs; 903 904 vvs = kzalloc(sizeof(*vvs), GFP_KERNEL); 905 if (!vvs) 906 return -ENOMEM; 907 908 vsk->trans = vvs; 909 vvs->vsk = vsk; 910 if (psk && psk->trans) { 911 struct virtio_vsock_sock *ptrans = psk->trans; 912 913 vvs->peer_buf_alloc = ptrans->peer_buf_alloc; 914 } 915 916 if (vsk->buffer_size > VIRTIO_VSOCK_MAX_BUF_SIZE) 917 vsk->buffer_size = VIRTIO_VSOCK_MAX_BUF_SIZE; 918 919 vvs->buf_alloc = vsk->buffer_size; 920 921 spin_lock_init(&vvs->rx_lock); 922 spin_lock_init(&vvs->tx_lock); 923 skb_queue_head_init(&vvs->rx_queue); 924 925 return 0; 926 } 927 EXPORT_SYMBOL_GPL(virtio_transport_do_socket_init); 928 929 /* sk_lock held by the caller */ 930 void virtio_transport_notify_buffer_size(struct vsock_sock *vsk, u64 *val) 931 { 932 struct virtio_vsock_sock *vvs = vsk->trans; 933 934 if (*val > VIRTIO_VSOCK_MAX_BUF_SIZE) 935 *val = VIRTIO_VSOCK_MAX_BUF_SIZE; 936 937 vvs->buf_alloc = *val; 938 939 virtio_transport_send_credit_update(vsk); 940 } 941 EXPORT_SYMBOL_GPL(virtio_transport_notify_buffer_size); 942 943 int 944 virtio_transport_notify_poll_in(struct vsock_sock *vsk, 945 size_t target, 946 bool *data_ready_now) 947 { 948 *data_ready_now = vsock_stream_has_data(vsk) >= target; 949 950 return 0; 951 } 952 EXPORT_SYMBOL_GPL(virtio_transport_notify_poll_in); 953 954 int 955 virtio_transport_notify_poll_out(struct vsock_sock *vsk, 956 size_t target, 957 bool *space_avail_now) 958 { 959 s64 free_space; 960 961 free_space = vsock_stream_has_space(vsk); 962 if (free_space > 0) 963 *space_avail_now = true; 964 else if (free_space == 0) 965 *space_avail_now = false; 966 967 return 0; 968 } 969 EXPORT_SYMBOL_GPL(virtio_transport_notify_poll_out); 970 971 int virtio_transport_notify_recv_init(struct vsock_sock *vsk, 972 size_t target, struct vsock_transport_recv_notify_data *data) 973 { 974 return 0; 975 } 976 EXPORT_SYMBOL_GPL(virtio_transport_notify_recv_init); 977 978 int virtio_transport_notify_recv_pre_block(struct vsock_sock *vsk, 979 size_t target, struct vsock_transport_recv_notify_data *data) 980 { 981 return 0; 982 } 983 EXPORT_SYMBOL_GPL(virtio_transport_notify_recv_pre_block); 984 985 int virtio_transport_notify_recv_pre_dequeue(struct vsock_sock *vsk, 986 size_t target, struct vsock_transport_recv_notify_data *data) 987 { 988 return 0; 989 } 990 EXPORT_SYMBOL_GPL(virtio_transport_notify_recv_pre_dequeue); 991 992 int virtio_transport_notify_recv_post_dequeue(struct vsock_sock *vsk, 993 size_t target, ssize_t copied, bool data_read, 994 struct vsock_transport_recv_notify_data *data) 995 { 996 return 0; 997 } 998 EXPORT_SYMBOL_GPL(virtio_transport_notify_recv_post_dequeue); 999 1000 int virtio_transport_notify_send_init(struct vsock_sock *vsk, 1001 struct vsock_transport_send_notify_data *data) 1002 { 1003 return 0; 1004 } 1005 EXPORT_SYMBOL_GPL(virtio_transport_notify_send_init); 1006 1007 int virtio_transport_notify_send_pre_block(struct vsock_sock *vsk, 1008 struct vsock_transport_send_notify_data *data) 1009 { 1010 return 0; 1011 } 1012 EXPORT_SYMBOL_GPL(virtio_transport_notify_send_pre_block); 1013 1014 int virtio_transport_notify_send_pre_enqueue(struct vsock_sock *vsk, 1015 struct vsock_transport_send_notify_data *data) 1016 { 1017 return 0; 1018 } 1019 EXPORT_SYMBOL_GPL(virtio_transport_notify_send_pre_enqueue); 1020 1021 int virtio_transport_notify_send_post_enqueue(struct vsock_sock *vsk, 1022 ssize_t written, struct vsock_transport_send_notify_data *data) 1023 { 1024 return 0; 1025 } 1026 EXPORT_SYMBOL_GPL(virtio_transport_notify_send_post_enqueue); 1027 1028 u64 virtio_transport_stream_rcvhiwat(struct vsock_sock *vsk) 1029 { 1030 return vsk->buffer_size; 1031 } 1032 EXPORT_SYMBOL_GPL(virtio_transport_stream_rcvhiwat); 1033 1034 bool virtio_transport_stream_is_active(struct vsock_sock *vsk) 1035 { 1036 return true; 1037 } 1038 EXPORT_SYMBOL_GPL(virtio_transport_stream_is_active); 1039 1040 bool virtio_transport_stream_allow(u32 cid, u32 port) 1041 { 1042 return true; 1043 } 1044 EXPORT_SYMBOL_GPL(virtio_transport_stream_allow); 1045 1046 int virtio_transport_dgram_bind(struct vsock_sock *vsk, 1047 struct sockaddr_vm *addr) 1048 { 1049 return -EOPNOTSUPP; 1050 } 1051 EXPORT_SYMBOL_GPL(virtio_transport_dgram_bind); 1052 1053 bool virtio_transport_dgram_allow(u32 cid, u32 port) 1054 { 1055 return false; 1056 } 1057 EXPORT_SYMBOL_GPL(virtio_transport_dgram_allow); 1058 1059 int virtio_transport_connect(struct vsock_sock *vsk) 1060 { 1061 struct virtio_vsock_pkt_info info = { 1062 .op = VIRTIO_VSOCK_OP_REQUEST, 1063 .vsk = vsk, 1064 }; 1065 1066 return virtio_transport_send_pkt_info(vsk, &info); 1067 } 1068 EXPORT_SYMBOL_GPL(virtio_transport_connect); 1069 1070 int virtio_transport_shutdown(struct vsock_sock *vsk, int mode) 1071 { 1072 struct virtio_vsock_pkt_info info = { 1073 .op = VIRTIO_VSOCK_OP_SHUTDOWN, 1074 .flags = (mode & RCV_SHUTDOWN ? 1075 VIRTIO_VSOCK_SHUTDOWN_RCV : 0) | 1076 (mode & SEND_SHUTDOWN ? 1077 VIRTIO_VSOCK_SHUTDOWN_SEND : 0), 1078 .vsk = vsk, 1079 }; 1080 1081 return virtio_transport_send_pkt_info(vsk, &info); 1082 } 1083 EXPORT_SYMBOL_GPL(virtio_transport_shutdown); 1084 1085 int 1086 virtio_transport_dgram_enqueue(struct vsock_sock *vsk, 1087 struct sockaddr_vm *remote_addr, 1088 struct msghdr *msg, 1089 size_t dgram_len) 1090 { 1091 return -EOPNOTSUPP; 1092 } 1093 EXPORT_SYMBOL_GPL(virtio_transport_dgram_enqueue); 1094 1095 ssize_t 1096 virtio_transport_stream_enqueue(struct vsock_sock *vsk, 1097 struct msghdr *msg, 1098 size_t len) 1099 { 1100 struct virtio_vsock_pkt_info info = { 1101 .op = VIRTIO_VSOCK_OP_RW, 1102 .msg = msg, 1103 .pkt_len = len, 1104 .vsk = vsk, 1105 }; 1106 1107 return virtio_transport_send_pkt_info(vsk, &info); 1108 } 1109 EXPORT_SYMBOL_GPL(virtio_transport_stream_enqueue); 1110 1111 void virtio_transport_destruct(struct vsock_sock *vsk) 1112 { 1113 struct virtio_vsock_sock *vvs = vsk->trans; 1114 1115 virtio_transport_cancel_close_work(vsk, true); 1116 1117 kfree(vvs); 1118 vsk->trans = NULL; 1119 } 1120 EXPORT_SYMBOL_GPL(virtio_transport_destruct); 1121 1122 ssize_t virtio_transport_unsent_bytes(struct vsock_sock *vsk) 1123 { 1124 struct virtio_vsock_sock *vvs = vsk->trans; 1125 size_t ret; 1126 1127 spin_lock_bh(&vvs->tx_lock); 1128 ret = vvs->bytes_unsent; 1129 spin_unlock_bh(&vvs->tx_lock); 1130 1131 return ret; 1132 } 1133 EXPORT_SYMBOL_GPL(virtio_transport_unsent_bytes); 1134 1135 static int virtio_transport_reset(struct vsock_sock *vsk, 1136 struct sk_buff *skb) 1137 { 1138 struct virtio_vsock_pkt_info info = { 1139 .op = VIRTIO_VSOCK_OP_RST, 1140 .reply = !!skb, 1141 .vsk = vsk, 1142 }; 1143 1144 /* Send RST only if the original pkt is not a RST pkt */ 1145 if (skb && le16_to_cpu(virtio_vsock_hdr(skb)->op) == VIRTIO_VSOCK_OP_RST) 1146 return 0; 1147 1148 return virtio_transport_send_pkt_info(vsk, &info); 1149 } 1150 1151 /* Normally packets are associated with a socket. There may be no socket if an 1152 * attempt was made to connect to a socket that does not exist. 1153 */ 1154 static int virtio_transport_reset_no_sock(const struct virtio_transport *t, 1155 struct sk_buff *skb) 1156 { 1157 struct virtio_vsock_hdr *hdr = virtio_vsock_hdr(skb); 1158 struct virtio_vsock_pkt_info info = { 1159 .op = VIRTIO_VSOCK_OP_RST, 1160 .type = le16_to_cpu(hdr->type), 1161 .reply = true, 1162 }; 1163 struct sk_buff *reply; 1164 1165 /* Send RST only if the original pkt is not a RST pkt */ 1166 if (le16_to_cpu(hdr->op) == VIRTIO_VSOCK_OP_RST) 1167 return 0; 1168 1169 if (!t) 1170 return -ENOTCONN; 1171 1172 reply = virtio_transport_alloc_skb(&info, 0, false, 1173 le64_to_cpu(hdr->dst_cid), 1174 le32_to_cpu(hdr->dst_port), 1175 le64_to_cpu(hdr->src_cid), 1176 le32_to_cpu(hdr->src_port)); 1177 if (!reply) 1178 return -ENOMEM; 1179 1180 return t->send_pkt(reply); 1181 } 1182 1183 /* This function should be called with sk_lock held and SOCK_DONE set */ 1184 static void virtio_transport_remove_sock(struct vsock_sock *vsk) 1185 { 1186 struct virtio_vsock_sock *vvs = vsk->trans; 1187 1188 /* We don't need to take rx_lock, as the socket is closing and we are 1189 * removing it. 1190 */ 1191 __skb_queue_purge(&vvs->rx_queue); 1192 vsock_remove_sock(vsk); 1193 } 1194 1195 static void virtio_transport_wait_close(struct sock *sk, long timeout) 1196 { 1197 if (timeout) { 1198 DEFINE_WAIT_FUNC(wait, woken_wake_function); 1199 1200 add_wait_queue(sk_sleep(sk), &wait); 1201 1202 do { 1203 if (sk_wait_event(sk, &timeout, 1204 sock_flag(sk, SOCK_DONE), &wait)) 1205 break; 1206 } while (!signal_pending(current) && timeout); 1207 1208 remove_wait_queue(sk_sleep(sk), &wait); 1209 } 1210 } 1211 1212 static void virtio_transport_cancel_close_work(struct vsock_sock *vsk, 1213 bool cancel_timeout) 1214 { 1215 struct sock *sk = sk_vsock(vsk); 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_do_close(struct vsock_sock *vsk, 1229 bool cancel_timeout) 1230 { 1231 struct sock *sk = sk_vsock(vsk); 1232 1233 sock_set_flag(sk, SOCK_DONE); 1234 vsk->peer_shutdown = SHUTDOWN_MASK; 1235 if (vsock_stream_has_data(vsk) <= 0) 1236 sk->sk_state = TCP_CLOSING; 1237 sk->sk_state_change(sk); 1238 1239 virtio_transport_cancel_close_work(vsk, cancel_timeout); 1240 } 1241 1242 static void virtio_transport_close_timeout(struct work_struct *work) 1243 { 1244 struct vsock_sock *vsk = 1245 container_of(work, struct vsock_sock, close_work.work); 1246 struct sock *sk = sk_vsock(vsk); 1247 1248 sock_hold(sk); 1249 lock_sock(sk); 1250 1251 if (!sock_flag(sk, SOCK_DONE)) { 1252 (void)virtio_transport_reset(vsk, NULL); 1253 1254 virtio_transport_do_close(vsk, false); 1255 } 1256 1257 vsk->close_work_scheduled = false; 1258 1259 release_sock(sk); 1260 sock_put(sk); 1261 } 1262 1263 /* User context, vsk->sk is locked */ 1264 static bool virtio_transport_close(struct vsock_sock *vsk) 1265 { 1266 struct sock *sk = &vsk->sk; 1267 1268 if (!(sk->sk_state == TCP_ESTABLISHED || 1269 sk->sk_state == TCP_CLOSING)) 1270 return true; 1271 1272 /* Already received SHUTDOWN from peer, reply with RST */ 1273 if ((vsk->peer_shutdown & SHUTDOWN_MASK) == SHUTDOWN_MASK) { 1274 (void)virtio_transport_reset(vsk, NULL); 1275 return true; 1276 } 1277 1278 if ((sk->sk_shutdown & SHUTDOWN_MASK) != SHUTDOWN_MASK) 1279 (void)virtio_transport_shutdown(vsk, SHUTDOWN_MASK); 1280 1281 if (sock_flag(sk, SOCK_LINGER) && !(current->flags & PF_EXITING)) 1282 virtio_transport_wait_close(sk, sk->sk_lingertime); 1283 1284 if (sock_flag(sk, SOCK_DONE)) { 1285 return true; 1286 } 1287 1288 sock_hold(sk); 1289 INIT_DELAYED_WORK(&vsk->close_work, 1290 virtio_transport_close_timeout); 1291 vsk->close_work_scheduled = true; 1292 schedule_delayed_work(&vsk->close_work, VSOCK_CLOSE_TIMEOUT); 1293 return false; 1294 } 1295 1296 void virtio_transport_release(struct vsock_sock *vsk) 1297 { 1298 struct sock *sk = &vsk->sk; 1299 bool remove_sock = true; 1300 1301 if (sk->sk_type == SOCK_STREAM || sk->sk_type == SOCK_SEQPACKET) 1302 remove_sock = virtio_transport_close(vsk); 1303 1304 if (remove_sock) { 1305 sock_set_flag(sk, SOCK_DONE); 1306 virtio_transport_remove_sock(vsk); 1307 } 1308 } 1309 EXPORT_SYMBOL_GPL(virtio_transport_release); 1310 1311 static int 1312 virtio_transport_recv_connecting(struct sock *sk, 1313 struct sk_buff *skb) 1314 { 1315 struct virtio_vsock_hdr *hdr = virtio_vsock_hdr(skb); 1316 struct vsock_sock *vsk = vsock_sk(sk); 1317 int skerr; 1318 int err; 1319 1320 switch (le16_to_cpu(hdr->op)) { 1321 case VIRTIO_VSOCK_OP_RESPONSE: 1322 sk->sk_state = TCP_ESTABLISHED; 1323 sk->sk_socket->state = SS_CONNECTED; 1324 vsock_insert_connected(vsk); 1325 sk->sk_state_change(sk); 1326 break; 1327 case VIRTIO_VSOCK_OP_INVALID: 1328 break; 1329 case VIRTIO_VSOCK_OP_RST: 1330 skerr = ECONNRESET; 1331 err = 0; 1332 goto destroy; 1333 default: 1334 skerr = EPROTO; 1335 err = -EINVAL; 1336 goto destroy; 1337 } 1338 return 0; 1339 1340 destroy: 1341 virtio_transport_reset(vsk, skb); 1342 sk->sk_state = TCP_CLOSE; 1343 sk->sk_err = skerr; 1344 sk_error_report(sk); 1345 return err; 1346 } 1347 1348 static void 1349 virtio_transport_recv_enqueue(struct vsock_sock *vsk, 1350 struct sk_buff *skb) 1351 { 1352 struct virtio_vsock_sock *vvs = vsk->trans; 1353 bool can_enqueue, free_pkt = false; 1354 struct virtio_vsock_hdr *hdr; 1355 u32 len; 1356 1357 hdr = virtio_vsock_hdr(skb); 1358 len = le32_to_cpu(hdr->len); 1359 1360 spin_lock_bh(&vvs->rx_lock); 1361 1362 can_enqueue = virtio_transport_inc_rx_pkt(vvs, len); 1363 if (!can_enqueue) { 1364 free_pkt = true; 1365 goto out; 1366 } 1367 1368 if (le32_to_cpu(hdr->flags) & VIRTIO_VSOCK_SEQ_EOM) 1369 vvs->msg_count++; 1370 1371 /* Try to copy small packets into the buffer of last packet queued, 1372 * to avoid wasting memory queueing the entire buffer with a small 1373 * payload. 1374 */ 1375 if (len <= GOOD_COPY_LEN && !skb_queue_empty(&vvs->rx_queue)) { 1376 struct virtio_vsock_hdr *last_hdr; 1377 struct sk_buff *last_skb; 1378 1379 last_skb = skb_peek_tail(&vvs->rx_queue); 1380 last_hdr = virtio_vsock_hdr(last_skb); 1381 1382 /* If there is space in the last packet queued, we copy the 1383 * new packet in its buffer. We avoid this if the last packet 1384 * queued has VIRTIO_VSOCK_SEQ_EOM set, because this is 1385 * delimiter of SEQPACKET message, so 'pkt' is the first packet 1386 * of a new message. 1387 */ 1388 if (skb->len < skb_tailroom(last_skb) && 1389 !(le32_to_cpu(last_hdr->flags) & VIRTIO_VSOCK_SEQ_EOM)) { 1390 memcpy(skb_put(last_skb, skb->len), skb->data, skb->len); 1391 free_pkt = true; 1392 last_hdr->flags |= hdr->flags; 1393 le32_add_cpu(&last_hdr->len, len); 1394 goto out; 1395 } 1396 } 1397 1398 __skb_queue_tail(&vvs->rx_queue, skb); 1399 1400 out: 1401 spin_unlock_bh(&vvs->rx_lock); 1402 if (free_pkt) 1403 kfree_skb(skb); 1404 } 1405 1406 static int 1407 virtio_transport_recv_connected(struct sock *sk, 1408 struct sk_buff *skb) 1409 { 1410 struct virtio_vsock_hdr *hdr = virtio_vsock_hdr(skb); 1411 struct vsock_sock *vsk = vsock_sk(sk); 1412 int err = 0; 1413 1414 switch (le16_to_cpu(hdr->op)) { 1415 case VIRTIO_VSOCK_OP_RW: 1416 virtio_transport_recv_enqueue(vsk, skb); 1417 vsock_data_ready(sk); 1418 return err; 1419 case VIRTIO_VSOCK_OP_CREDIT_REQUEST: 1420 virtio_transport_send_credit_update(vsk); 1421 break; 1422 case VIRTIO_VSOCK_OP_CREDIT_UPDATE: 1423 sk->sk_write_space(sk); 1424 break; 1425 case VIRTIO_VSOCK_OP_SHUTDOWN: 1426 if (le32_to_cpu(hdr->flags) & VIRTIO_VSOCK_SHUTDOWN_RCV) 1427 vsk->peer_shutdown |= RCV_SHUTDOWN; 1428 if (le32_to_cpu(hdr->flags) & VIRTIO_VSOCK_SHUTDOWN_SEND) 1429 vsk->peer_shutdown |= SEND_SHUTDOWN; 1430 if (vsk->peer_shutdown == SHUTDOWN_MASK) { 1431 if (vsock_stream_has_data(vsk) <= 0 && !sock_flag(sk, SOCK_DONE)) { 1432 (void)virtio_transport_reset(vsk, NULL); 1433 virtio_transport_do_close(vsk, true); 1434 } 1435 /* Remove this socket anyway because the remote peer sent 1436 * the shutdown. This way a new connection will succeed 1437 * if the remote peer uses the same source port, 1438 * even if the old socket is still unreleased, but now disconnected. 1439 */ 1440 vsock_remove_sock(vsk); 1441 } 1442 if (le32_to_cpu(virtio_vsock_hdr(skb)->flags)) 1443 sk->sk_state_change(sk); 1444 break; 1445 case VIRTIO_VSOCK_OP_RST: 1446 virtio_transport_do_close(vsk, true); 1447 break; 1448 default: 1449 err = -EINVAL; 1450 break; 1451 } 1452 1453 kfree_skb(skb); 1454 return err; 1455 } 1456 1457 static void 1458 virtio_transport_recv_disconnecting(struct sock *sk, 1459 struct sk_buff *skb) 1460 { 1461 struct virtio_vsock_hdr *hdr = virtio_vsock_hdr(skb); 1462 struct vsock_sock *vsk = vsock_sk(sk); 1463 1464 if (le16_to_cpu(hdr->op) == VIRTIO_VSOCK_OP_RST) 1465 virtio_transport_do_close(vsk, true); 1466 } 1467 1468 static int 1469 virtio_transport_send_response(struct vsock_sock *vsk, 1470 struct sk_buff *skb) 1471 { 1472 struct virtio_vsock_hdr *hdr = virtio_vsock_hdr(skb); 1473 struct virtio_vsock_pkt_info info = { 1474 .op = VIRTIO_VSOCK_OP_RESPONSE, 1475 .remote_cid = le64_to_cpu(hdr->src_cid), 1476 .remote_port = le32_to_cpu(hdr->src_port), 1477 .reply = true, 1478 .vsk = vsk, 1479 }; 1480 1481 return virtio_transport_send_pkt_info(vsk, &info); 1482 } 1483 1484 static bool virtio_transport_space_update(struct sock *sk, 1485 struct sk_buff *skb) 1486 { 1487 struct virtio_vsock_hdr *hdr = virtio_vsock_hdr(skb); 1488 struct vsock_sock *vsk = vsock_sk(sk); 1489 struct virtio_vsock_sock *vvs = vsk->trans; 1490 bool space_available; 1491 1492 /* Listener sockets are not associated with any transport, so we are 1493 * not able to take the state to see if there is space available in the 1494 * remote peer, but since they are only used to receive requests, we 1495 * can assume that there is always space available in the other peer. 1496 */ 1497 if (!vvs) 1498 return true; 1499 1500 /* buf_alloc and fwd_cnt is always included in the hdr */ 1501 spin_lock_bh(&vvs->tx_lock); 1502 vvs->peer_buf_alloc = le32_to_cpu(hdr->buf_alloc); 1503 vvs->peer_fwd_cnt = le32_to_cpu(hdr->fwd_cnt); 1504 space_available = virtio_transport_has_space(vsk); 1505 spin_unlock_bh(&vvs->tx_lock); 1506 return space_available; 1507 } 1508 1509 /* Handle server socket */ 1510 static int 1511 virtio_transport_recv_listen(struct sock *sk, struct sk_buff *skb, 1512 struct virtio_transport *t) 1513 { 1514 struct virtio_vsock_hdr *hdr = virtio_vsock_hdr(skb); 1515 struct vsock_sock *vsk = vsock_sk(sk); 1516 struct vsock_sock *vchild; 1517 struct sock *child; 1518 int ret; 1519 1520 if (le16_to_cpu(hdr->op) != VIRTIO_VSOCK_OP_REQUEST) { 1521 virtio_transport_reset_no_sock(t, skb); 1522 return -EINVAL; 1523 } 1524 1525 if (sk_acceptq_is_full(sk)) { 1526 virtio_transport_reset_no_sock(t, skb); 1527 return -ENOMEM; 1528 } 1529 1530 /* __vsock_release() might have already flushed accept_queue. 1531 * Subsequent enqueues would lead to a memory leak. 1532 */ 1533 if (sk->sk_shutdown == SHUTDOWN_MASK) { 1534 virtio_transport_reset_no_sock(t, skb); 1535 return -ESHUTDOWN; 1536 } 1537 1538 child = vsock_create_connected(sk); 1539 if (!child) { 1540 virtio_transport_reset_no_sock(t, skb); 1541 return -ENOMEM; 1542 } 1543 1544 sk_acceptq_added(sk); 1545 1546 lock_sock_nested(child, SINGLE_DEPTH_NESTING); 1547 1548 child->sk_state = TCP_ESTABLISHED; 1549 1550 vchild = vsock_sk(child); 1551 vsock_addr_init(&vchild->local_addr, le64_to_cpu(hdr->dst_cid), 1552 le32_to_cpu(hdr->dst_port)); 1553 vsock_addr_init(&vchild->remote_addr, le64_to_cpu(hdr->src_cid), 1554 le32_to_cpu(hdr->src_port)); 1555 1556 ret = vsock_assign_transport(vchild, vsk); 1557 /* Transport assigned (looking at remote_addr) must be the same 1558 * where we received the request. 1559 */ 1560 if (ret || vchild->transport != &t->transport) { 1561 release_sock(child); 1562 virtio_transport_reset_no_sock(t, skb); 1563 sock_put(child); 1564 return ret; 1565 } 1566 1567 if (virtio_transport_space_update(child, skb)) 1568 child->sk_write_space(child); 1569 1570 vsock_insert_connected(vchild); 1571 vsock_enqueue_accept(sk, child); 1572 virtio_transport_send_response(vchild, skb); 1573 1574 release_sock(child); 1575 1576 sk->sk_data_ready(sk); 1577 return 0; 1578 } 1579 1580 static bool virtio_transport_valid_type(u16 type) 1581 { 1582 return (type == VIRTIO_VSOCK_TYPE_STREAM) || 1583 (type == VIRTIO_VSOCK_TYPE_SEQPACKET); 1584 } 1585 1586 /* We are under the virtio-vsock's vsock->rx_lock or vhost-vsock's vq->mutex 1587 * lock. 1588 */ 1589 void virtio_transport_recv_pkt(struct virtio_transport *t, 1590 struct sk_buff *skb) 1591 { 1592 struct virtio_vsock_hdr *hdr = virtio_vsock_hdr(skb); 1593 struct sockaddr_vm src, dst; 1594 struct vsock_sock *vsk; 1595 struct sock *sk; 1596 bool space_available; 1597 1598 vsock_addr_init(&src, le64_to_cpu(hdr->src_cid), 1599 le32_to_cpu(hdr->src_port)); 1600 vsock_addr_init(&dst, le64_to_cpu(hdr->dst_cid), 1601 le32_to_cpu(hdr->dst_port)); 1602 1603 trace_virtio_transport_recv_pkt(src.svm_cid, src.svm_port, 1604 dst.svm_cid, dst.svm_port, 1605 le32_to_cpu(hdr->len), 1606 le16_to_cpu(hdr->type), 1607 le16_to_cpu(hdr->op), 1608 le32_to_cpu(hdr->flags), 1609 le32_to_cpu(hdr->buf_alloc), 1610 le32_to_cpu(hdr->fwd_cnt)); 1611 1612 if (!virtio_transport_valid_type(le16_to_cpu(hdr->type))) { 1613 (void)virtio_transport_reset_no_sock(t, skb); 1614 goto free_pkt; 1615 } 1616 1617 /* The socket must be in connected or bound table 1618 * otherwise send reset back 1619 */ 1620 sk = vsock_find_connected_socket(&src, &dst); 1621 if (!sk) { 1622 sk = vsock_find_bound_socket(&dst); 1623 if (!sk) { 1624 (void)virtio_transport_reset_no_sock(t, skb); 1625 goto free_pkt; 1626 } 1627 } 1628 1629 if (virtio_transport_get_type(sk) != le16_to_cpu(hdr->type)) { 1630 (void)virtio_transport_reset_no_sock(t, skb); 1631 sock_put(sk); 1632 goto free_pkt; 1633 } 1634 1635 if (!skb_set_owner_sk_safe(skb, sk)) { 1636 WARN_ONCE(1, "receiving vsock socket has sk_refcnt == 0\n"); 1637 goto free_pkt; 1638 } 1639 1640 vsk = vsock_sk(sk); 1641 1642 lock_sock(sk); 1643 1644 /* Check if sk has been closed or assigned to another transport before 1645 * lock_sock (note: listener sockets are not assigned to any transport) 1646 */ 1647 if (sock_flag(sk, SOCK_DONE) || 1648 (sk->sk_state != TCP_LISTEN && vsk->transport != &t->transport)) { 1649 (void)virtio_transport_reset_no_sock(t, skb); 1650 release_sock(sk); 1651 sock_put(sk); 1652 goto free_pkt; 1653 } 1654 1655 space_available = virtio_transport_space_update(sk, skb); 1656 1657 /* Update CID in case it has changed after a transport reset event */ 1658 if (vsk->local_addr.svm_cid != VMADDR_CID_ANY) 1659 vsk->local_addr.svm_cid = dst.svm_cid; 1660 1661 if (space_available) 1662 sk->sk_write_space(sk); 1663 1664 switch (sk->sk_state) { 1665 case TCP_LISTEN: 1666 virtio_transport_recv_listen(sk, skb, t); 1667 kfree_skb(skb); 1668 break; 1669 case TCP_SYN_SENT: 1670 virtio_transport_recv_connecting(sk, skb); 1671 kfree_skb(skb); 1672 break; 1673 case TCP_ESTABLISHED: 1674 virtio_transport_recv_connected(sk, skb); 1675 break; 1676 case TCP_CLOSING: 1677 virtio_transport_recv_disconnecting(sk, skb); 1678 kfree_skb(skb); 1679 break; 1680 default: 1681 (void)virtio_transport_reset_no_sock(t, skb); 1682 kfree_skb(skb); 1683 break; 1684 } 1685 1686 release_sock(sk); 1687 1688 /* Release refcnt obtained when we fetched this socket out of the 1689 * bound or connected list. 1690 */ 1691 sock_put(sk); 1692 return; 1693 1694 free_pkt: 1695 kfree_skb(skb); 1696 } 1697 EXPORT_SYMBOL_GPL(virtio_transport_recv_pkt); 1698 1699 /* Remove skbs found in a queue that have a vsk that matches. 1700 * 1701 * Each skb is freed. 1702 * 1703 * Returns the count of skbs that were reply packets. 1704 */ 1705 int virtio_transport_purge_skbs(void *vsk, struct sk_buff_head *queue) 1706 { 1707 struct sk_buff_head freeme; 1708 struct sk_buff *skb, *tmp; 1709 int cnt = 0; 1710 1711 skb_queue_head_init(&freeme); 1712 1713 spin_lock_bh(&queue->lock); 1714 skb_queue_walk_safe(queue, skb, tmp) { 1715 if (vsock_sk(skb->sk) != vsk) 1716 continue; 1717 1718 __skb_unlink(skb, queue); 1719 __skb_queue_tail(&freeme, skb); 1720 1721 if (virtio_vsock_skb_reply(skb)) 1722 cnt++; 1723 } 1724 spin_unlock_bh(&queue->lock); 1725 1726 __skb_queue_purge(&freeme); 1727 1728 return cnt; 1729 } 1730 EXPORT_SYMBOL_GPL(virtio_transport_purge_skbs); 1731 1732 int virtio_transport_read_skb(struct vsock_sock *vsk, skb_read_actor_t recv_actor) 1733 { 1734 struct virtio_vsock_sock *vvs = vsk->trans; 1735 struct sock *sk = sk_vsock(vsk); 1736 struct virtio_vsock_hdr *hdr; 1737 struct sk_buff *skb; 1738 int off = 0; 1739 int err; 1740 1741 spin_lock_bh(&vvs->rx_lock); 1742 /* Use __skb_recv_datagram() for race-free handling of the receive. It 1743 * works for types other than dgrams. 1744 */ 1745 skb = __skb_recv_datagram(sk, &vvs->rx_queue, MSG_DONTWAIT, &off, &err); 1746 if (!skb) { 1747 spin_unlock_bh(&vvs->rx_lock); 1748 return err; 1749 } 1750 1751 hdr = virtio_vsock_hdr(skb); 1752 if (le32_to_cpu(hdr->flags) & VIRTIO_VSOCK_SEQ_EOM) 1753 vvs->msg_count--; 1754 1755 virtio_transport_dec_rx_pkt(vvs, le32_to_cpu(hdr->len)); 1756 spin_unlock_bh(&vvs->rx_lock); 1757 1758 virtio_transport_send_credit_update(vsk); 1759 1760 return recv_actor(sk, skb); 1761 } 1762 EXPORT_SYMBOL_GPL(virtio_transport_read_skb); 1763 1764 int virtio_transport_notify_set_rcvlowat(struct vsock_sock *vsk, int val) 1765 { 1766 struct virtio_vsock_sock *vvs = vsk->trans; 1767 bool send_update; 1768 1769 spin_lock_bh(&vvs->rx_lock); 1770 1771 /* If number of available bytes is less than new SO_RCVLOWAT value, 1772 * kick sender to send more data, because sender may sleep in its 1773 * 'send()' syscall waiting for enough space at our side. Also 1774 * don't send credit update when peer already knows actual value - 1775 * such transmission will be useless. 1776 */ 1777 send_update = (vvs->rx_bytes < val) && 1778 (vvs->fwd_cnt != vvs->last_fwd_cnt); 1779 1780 spin_unlock_bh(&vvs->rx_lock); 1781 1782 if (send_update) { 1783 int err; 1784 1785 err = virtio_transport_send_credit_update(vsk); 1786 if (err < 0) 1787 return err; 1788 } 1789 1790 return 0; 1791 } 1792 EXPORT_SYMBOL_GPL(virtio_transport_notify_set_rcvlowat); 1793 1794 MODULE_LICENSE("GPL v2"); 1795 MODULE_AUTHOR("Asias He"); 1796 MODULE_DESCRIPTION("common code for virtio vsock"); 1797