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 virtio_vsock_skb_put(skb, len); 113 return skb_copy_datagram_from_iter(skb, 0, &info->msg->msg_iter, 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->buf_used + len > vvs->buf_alloc) 445 return false; 446 447 vvs->rx_bytes += len; 448 vvs->buf_used += len; 449 return true; 450 } 451 452 static void virtio_transport_dec_rx_pkt(struct virtio_vsock_sock *vvs, 453 u32 bytes_read, u32 bytes_dequeued) 454 { 455 vvs->rx_bytes -= bytes_read; 456 vvs->buf_used -= bytes_dequeued; 457 vvs->fwd_cnt += bytes_dequeued; 458 } 459 460 void virtio_transport_inc_tx_pkt(struct virtio_vsock_sock *vvs, struct sk_buff *skb) 461 { 462 struct virtio_vsock_hdr *hdr = virtio_vsock_hdr(skb); 463 464 spin_lock_bh(&vvs->rx_lock); 465 vvs->last_fwd_cnt = vvs->fwd_cnt; 466 hdr->fwd_cnt = cpu_to_le32(vvs->fwd_cnt); 467 hdr->buf_alloc = cpu_to_le32(vvs->buf_alloc); 468 spin_unlock_bh(&vvs->rx_lock); 469 } 470 EXPORT_SYMBOL_GPL(virtio_transport_inc_tx_pkt); 471 472 void virtio_transport_consume_skb_sent(struct sk_buff *skb, bool consume) 473 { 474 struct sock *s = skb->sk; 475 476 if (s && skb->len) { 477 struct vsock_sock *vs = vsock_sk(s); 478 struct virtio_vsock_sock *vvs; 479 480 vvs = vs->trans; 481 482 spin_lock_bh(&vvs->tx_lock); 483 vvs->bytes_unsent -= skb->len; 484 spin_unlock_bh(&vvs->tx_lock); 485 } 486 487 if (consume) 488 consume_skb(skb); 489 } 490 EXPORT_SYMBOL_GPL(virtio_transport_consume_skb_sent); 491 492 u32 virtio_transport_get_credit(struct virtio_vsock_sock *vvs, u32 credit) 493 { 494 u32 ret; 495 496 if (!credit) 497 return 0; 498 499 spin_lock_bh(&vvs->tx_lock); 500 ret = vvs->peer_buf_alloc - (vvs->tx_cnt - vvs->peer_fwd_cnt); 501 if (ret > credit) 502 ret = credit; 503 vvs->tx_cnt += ret; 504 vvs->bytes_unsent += ret; 505 spin_unlock_bh(&vvs->tx_lock); 506 507 return ret; 508 } 509 EXPORT_SYMBOL_GPL(virtio_transport_get_credit); 510 511 void virtio_transport_put_credit(struct virtio_vsock_sock *vvs, u32 credit) 512 { 513 if (!credit) 514 return; 515 516 spin_lock_bh(&vvs->tx_lock); 517 vvs->tx_cnt -= credit; 518 vvs->bytes_unsent -= credit; 519 spin_unlock_bh(&vvs->tx_lock); 520 } 521 EXPORT_SYMBOL_GPL(virtio_transport_put_credit); 522 523 static int virtio_transport_send_credit_update(struct vsock_sock *vsk) 524 { 525 struct virtio_vsock_pkt_info info = { 526 .op = VIRTIO_VSOCK_OP_CREDIT_UPDATE, 527 .vsk = vsk, 528 }; 529 530 return virtio_transport_send_pkt_info(vsk, &info); 531 } 532 533 static ssize_t 534 virtio_transport_stream_do_peek(struct vsock_sock *vsk, 535 struct msghdr *msg, 536 size_t len) 537 { 538 struct virtio_vsock_sock *vvs = vsk->trans; 539 struct sk_buff *skb; 540 size_t total = 0; 541 int err; 542 543 spin_lock_bh(&vvs->rx_lock); 544 545 skb_queue_walk(&vvs->rx_queue, skb) { 546 size_t bytes; 547 548 bytes = len - total; 549 if (bytes > skb->len) 550 bytes = skb->len; 551 552 spin_unlock_bh(&vvs->rx_lock); 553 554 /* sk_lock is held by caller so no one else can dequeue. 555 * Unlock rx_lock since skb_copy_datagram_iter() may sleep. 556 */ 557 err = skb_copy_datagram_iter(skb, VIRTIO_VSOCK_SKB_CB(skb)->offset, 558 &msg->msg_iter, bytes); 559 if (err) 560 goto out; 561 562 total += bytes; 563 564 spin_lock_bh(&vvs->rx_lock); 565 566 if (total == len) 567 break; 568 } 569 570 spin_unlock_bh(&vvs->rx_lock); 571 572 return total; 573 574 out: 575 if (total) 576 err = total; 577 return err; 578 } 579 580 static ssize_t 581 virtio_transport_stream_do_dequeue(struct vsock_sock *vsk, 582 struct msghdr *msg, 583 size_t len) 584 { 585 struct virtio_vsock_sock *vvs = vsk->trans; 586 struct sk_buff *skb; 587 u32 fwd_cnt_delta; 588 bool low_rx_bytes; 589 int err = -EFAULT; 590 size_t total = 0; 591 u32 free_space; 592 593 spin_lock_bh(&vvs->rx_lock); 594 595 if (WARN_ONCE(skb_queue_empty(&vvs->rx_queue) && vvs->rx_bytes, 596 "rx_queue is empty, but rx_bytes is non-zero\n")) { 597 spin_unlock_bh(&vvs->rx_lock); 598 return err; 599 } 600 601 while (total < len && !skb_queue_empty(&vvs->rx_queue)) { 602 size_t bytes, dequeued = 0; 603 604 skb = skb_peek(&vvs->rx_queue); 605 606 bytes = min_t(size_t, len - total, 607 skb->len - VIRTIO_VSOCK_SKB_CB(skb)->offset); 608 609 /* sk_lock is held by caller so no one else can dequeue. 610 * Unlock rx_lock since skb_copy_datagram_iter() may sleep. 611 */ 612 spin_unlock_bh(&vvs->rx_lock); 613 614 err = skb_copy_datagram_iter(skb, 615 VIRTIO_VSOCK_SKB_CB(skb)->offset, 616 &msg->msg_iter, bytes); 617 if (err) 618 goto out; 619 620 spin_lock_bh(&vvs->rx_lock); 621 622 total += bytes; 623 624 VIRTIO_VSOCK_SKB_CB(skb)->offset += bytes; 625 626 if (skb->len == VIRTIO_VSOCK_SKB_CB(skb)->offset) { 627 dequeued = le32_to_cpu(virtio_vsock_hdr(skb)->len); 628 __skb_unlink(skb, &vvs->rx_queue); 629 consume_skb(skb); 630 } 631 632 virtio_transport_dec_rx_pkt(vvs, bytes, dequeued); 633 } 634 635 fwd_cnt_delta = vvs->fwd_cnt - vvs->last_fwd_cnt; 636 free_space = vvs->buf_alloc - fwd_cnt_delta; 637 low_rx_bytes = (vvs->rx_bytes < 638 sock_rcvlowat(sk_vsock(vsk), 0, INT_MAX)); 639 640 spin_unlock_bh(&vvs->rx_lock); 641 642 /* To reduce the number of credit update messages, 643 * don't update credits as long as lots of space is available. 644 * Note: the limit chosen here is arbitrary. Setting the limit 645 * too high causes extra messages. Too low causes transmitter 646 * stalls. As stalls are in theory more expensive than extra 647 * messages, we set the limit to a high value. TODO: experiment 648 * with different values. Also send credit update message when 649 * number of bytes in rx queue is not enough to wake up reader. 650 */ 651 if (fwd_cnt_delta && 652 (free_space < VIRTIO_VSOCK_MAX_PKT_BUF_SIZE || low_rx_bytes)) 653 virtio_transport_send_credit_update(vsk); 654 655 return total; 656 657 out: 658 if (total) 659 err = total; 660 return err; 661 } 662 663 static ssize_t 664 virtio_transport_seqpacket_do_peek(struct vsock_sock *vsk, 665 struct msghdr *msg) 666 { 667 struct virtio_vsock_sock *vvs = vsk->trans; 668 struct sk_buff *skb; 669 size_t total, len; 670 671 spin_lock_bh(&vvs->rx_lock); 672 673 if (!vvs->msg_count) { 674 spin_unlock_bh(&vvs->rx_lock); 675 return 0; 676 } 677 678 total = 0; 679 len = msg_data_left(msg); 680 681 skb_queue_walk(&vvs->rx_queue, skb) { 682 struct virtio_vsock_hdr *hdr; 683 684 if (total < len) { 685 size_t bytes; 686 int err; 687 688 bytes = len - total; 689 if (bytes > skb->len) 690 bytes = skb->len; 691 692 spin_unlock_bh(&vvs->rx_lock); 693 694 /* sk_lock is held by caller so no one else can dequeue. 695 * Unlock rx_lock since skb_copy_datagram_iter() may sleep. 696 */ 697 err = skb_copy_datagram_iter(skb, VIRTIO_VSOCK_SKB_CB(skb)->offset, 698 &msg->msg_iter, bytes); 699 if (err) 700 return err; 701 702 spin_lock_bh(&vvs->rx_lock); 703 } 704 705 total += skb->len; 706 hdr = virtio_vsock_hdr(skb); 707 708 if (le32_to_cpu(hdr->flags) & VIRTIO_VSOCK_SEQ_EOM) { 709 if (le32_to_cpu(hdr->flags) & VIRTIO_VSOCK_SEQ_EOR) 710 msg->msg_flags |= MSG_EOR; 711 712 break; 713 } 714 } 715 716 spin_unlock_bh(&vvs->rx_lock); 717 718 return total; 719 } 720 721 static int virtio_transport_seqpacket_do_dequeue(struct vsock_sock *vsk, 722 struct msghdr *msg, 723 int flags) 724 { 725 struct virtio_vsock_sock *vvs = vsk->trans; 726 int dequeued_len = 0; 727 size_t user_buf_len = msg_data_left(msg); 728 bool msg_ready = false; 729 struct sk_buff *skb; 730 731 spin_lock_bh(&vvs->rx_lock); 732 733 if (vvs->msg_count == 0) { 734 spin_unlock_bh(&vvs->rx_lock); 735 return 0; 736 } 737 738 while (!msg_ready) { 739 struct virtio_vsock_hdr *hdr; 740 size_t pkt_len; 741 742 skb = __skb_dequeue(&vvs->rx_queue); 743 if (!skb) 744 break; 745 hdr = virtio_vsock_hdr(skb); 746 pkt_len = (size_t)le32_to_cpu(hdr->len); 747 748 if (dequeued_len >= 0) { 749 size_t bytes_to_copy; 750 751 bytes_to_copy = min(user_buf_len, pkt_len); 752 753 if (bytes_to_copy) { 754 int err; 755 756 /* sk_lock is held by caller so no one else can dequeue. 757 * Unlock rx_lock since skb_copy_datagram_iter() may sleep. 758 */ 759 spin_unlock_bh(&vvs->rx_lock); 760 761 err = skb_copy_datagram_iter(skb, 0, 762 &msg->msg_iter, 763 bytes_to_copy); 764 if (err) { 765 /* Copy of message failed. Rest of 766 * fragments will be freed without copy. 767 */ 768 dequeued_len = err; 769 } else { 770 user_buf_len -= bytes_to_copy; 771 } 772 773 spin_lock_bh(&vvs->rx_lock); 774 } 775 776 if (dequeued_len >= 0) 777 dequeued_len += pkt_len; 778 } 779 780 if (le32_to_cpu(hdr->flags) & VIRTIO_VSOCK_SEQ_EOM) { 781 msg_ready = true; 782 vvs->msg_count--; 783 784 if (le32_to_cpu(hdr->flags) & VIRTIO_VSOCK_SEQ_EOR) 785 msg->msg_flags |= MSG_EOR; 786 } 787 788 virtio_transport_dec_rx_pkt(vvs, pkt_len, pkt_len); 789 kfree_skb(skb); 790 } 791 792 spin_unlock_bh(&vvs->rx_lock); 793 794 virtio_transport_send_credit_update(vsk); 795 796 return dequeued_len; 797 } 798 799 ssize_t 800 virtio_transport_stream_dequeue(struct vsock_sock *vsk, 801 struct msghdr *msg, 802 size_t len, int flags) 803 { 804 if (flags & MSG_PEEK) 805 return virtio_transport_stream_do_peek(vsk, msg, len); 806 else 807 return virtio_transport_stream_do_dequeue(vsk, msg, len); 808 } 809 EXPORT_SYMBOL_GPL(virtio_transport_stream_dequeue); 810 811 ssize_t 812 virtio_transport_seqpacket_dequeue(struct vsock_sock *vsk, 813 struct msghdr *msg, 814 int flags) 815 { 816 if (flags & MSG_PEEK) 817 return virtio_transport_seqpacket_do_peek(vsk, msg); 818 else 819 return virtio_transport_seqpacket_do_dequeue(vsk, msg, flags); 820 } 821 EXPORT_SYMBOL_GPL(virtio_transport_seqpacket_dequeue); 822 823 int 824 virtio_transport_seqpacket_enqueue(struct vsock_sock *vsk, 825 struct msghdr *msg, 826 size_t len) 827 { 828 struct virtio_vsock_sock *vvs = vsk->trans; 829 830 spin_lock_bh(&vvs->tx_lock); 831 832 if (len > vvs->peer_buf_alloc) { 833 spin_unlock_bh(&vvs->tx_lock); 834 return -EMSGSIZE; 835 } 836 837 spin_unlock_bh(&vvs->tx_lock); 838 839 return virtio_transport_stream_enqueue(vsk, msg, len); 840 } 841 EXPORT_SYMBOL_GPL(virtio_transport_seqpacket_enqueue); 842 843 int 844 virtio_transport_dgram_dequeue(struct vsock_sock *vsk, 845 struct msghdr *msg, 846 size_t len, int flags) 847 { 848 return -EOPNOTSUPP; 849 } 850 EXPORT_SYMBOL_GPL(virtio_transport_dgram_dequeue); 851 852 s64 virtio_transport_stream_has_data(struct vsock_sock *vsk) 853 { 854 struct virtio_vsock_sock *vvs = vsk->trans; 855 s64 bytes; 856 857 spin_lock_bh(&vvs->rx_lock); 858 bytes = vvs->rx_bytes; 859 spin_unlock_bh(&vvs->rx_lock); 860 861 return bytes; 862 } 863 EXPORT_SYMBOL_GPL(virtio_transport_stream_has_data); 864 865 u32 virtio_transport_seqpacket_has_data(struct vsock_sock *vsk) 866 { 867 struct virtio_vsock_sock *vvs = vsk->trans; 868 u32 msg_count; 869 870 spin_lock_bh(&vvs->rx_lock); 871 msg_count = vvs->msg_count; 872 spin_unlock_bh(&vvs->rx_lock); 873 874 return msg_count; 875 } 876 EXPORT_SYMBOL_GPL(virtio_transport_seqpacket_has_data); 877 878 static s64 virtio_transport_has_space(struct vsock_sock *vsk) 879 { 880 struct virtio_vsock_sock *vvs = vsk->trans; 881 s64 bytes; 882 883 bytes = (s64)vvs->peer_buf_alloc - (vvs->tx_cnt - vvs->peer_fwd_cnt); 884 if (bytes < 0) 885 bytes = 0; 886 887 return bytes; 888 } 889 890 s64 virtio_transport_stream_has_space(struct vsock_sock *vsk) 891 { 892 struct virtio_vsock_sock *vvs = vsk->trans; 893 s64 bytes; 894 895 spin_lock_bh(&vvs->tx_lock); 896 bytes = virtio_transport_has_space(vsk); 897 spin_unlock_bh(&vvs->tx_lock); 898 899 return bytes; 900 } 901 EXPORT_SYMBOL_GPL(virtio_transport_stream_has_space); 902 903 int virtio_transport_do_socket_init(struct vsock_sock *vsk, 904 struct vsock_sock *psk) 905 { 906 struct virtio_vsock_sock *vvs; 907 908 vvs = kzalloc(sizeof(*vvs), GFP_KERNEL); 909 if (!vvs) 910 return -ENOMEM; 911 912 vsk->trans = vvs; 913 vvs->vsk = vsk; 914 if (psk && psk->trans) { 915 struct virtio_vsock_sock *ptrans = psk->trans; 916 917 vvs->peer_buf_alloc = ptrans->peer_buf_alloc; 918 } 919 920 if (vsk->buffer_size > VIRTIO_VSOCK_MAX_BUF_SIZE) 921 vsk->buffer_size = VIRTIO_VSOCK_MAX_BUF_SIZE; 922 923 vvs->buf_alloc = vsk->buffer_size; 924 925 spin_lock_init(&vvs->rx_lock); 926 spin_lock_init(&vvs->tx_lock); 927 skb_queue_head_init(&vvs->rx_queue); 928 929 return 0; 930 } 931 EXPORT_SYMBOL_GPL(virtio_transport_do_socket_init); 932 933 /* sk_lock held by the caller */ 934 void virtio_transport_notify_buffer_size(struct vsock_sock *vsk, u64 *val) 935 { 936 struct virtio_vsock_sock *vvs = vsk->trans; 937 938 if (*val > VIRTIO_VSOCK_MAX_BUF_SIZE) 939 *val = VIRTIO_VSOCK_MAX_BUF_SIZE; 940 941 vvs->buf_alloc = *val; 942 943 virtio_transport_send_credit_update(vsk); 944 } 945 EXPORT_SYMBOL_GPL(virtio_transport_notify_buffer_size); 946 947 int 948 virtio_transport_notify_poll_in(struct vsock_sock *vsk, 949 size_t target, 950 bool *data_ready_now) 951 { 952 *data_ready_now = vsock_stream_has_data(vsk) >= target; 953 954 return 0; 955 } 956 EXPORT_SYMBOL_GPL(virtio_transport_notify_poll_in); 957 958 int 959 virtio_transport_notify_poll_out(struct vsock_sock *vsk, 960 size_t target, 961 bool *space_avail_now) 962 { 963 s64 free_space; 964 965 free_space = vsock_stream_has_space(vsk); 966 if (free_space > 0) 967 *space_avail_now = true; 968 else if (free_space == 0) 969 *space_avail_now = false; 970 971 return 0; 972 } 973 EXPORT_SYMBOL_GPL(virtio_transport_notify_poll_out); 974 975 int virtio_transport_notify_recv_init(struct vsock_sock *vsk, 976 size_t target, struct vsock_transport_recv_notify_data *data) 977 { 978 return 0; 979 } 980 EXPORT_SYMBOL_GPL(virtio_transport_notify_recv_init); 981 982 int virtio_transport_notify_recv_pre_block(struct vsock_sock *vsk, 983 size_t target, struct vsock_transport_recv_notify_data *data) 984 { 985 return 0; 986 } 987 EXPORT_SYMBOL_GPL(virtio_transport_notify_recv_pre_block); 988 989 int virtio_transport_notify_recv_pre_dequeue(struct vsock_sock *vsk, 990 size_t target, struct vsock_transport_recv_notify_data *data) 991 { 992 return 0; 993 } 994 EXPORT_SYMBOL_GPL(virtio_transport_notify_recv_pre_dequeue); 995 996 int virtio_transport_notify_recv_post_dequeue(struct vsock_sock *vsk, 997 size_t target, ssize_t copied, bool data_read, 998 struct vsock_transport_recv_notify_data *data) 999 { 1000 return 0; 1001 } 1002 EXPORT_SYMBOL_GPL(virtio_transport_notify_recv_post_dequeue); 1003 1004 int virtio_transport_notify_send_init(struct vsock_sock *vsk, 1005 struct vsock_transport_send_notify_data *data) 1006 { 1007 return 0; 1008 } 1009 EXPORT_SYMBOL_GPL(virtio_transport_notify_send_init); 1010 1011 int virtio_transport_notify_send_pre_block(struct vsock_sock *vsk, 1012 struct vsock_transport_send_notify_data *data) 1013 { 1014 return 0; 1015 } 1016 EXPORT_SYMBOL_GPL(virtio_transport_notify_send_pre_block); 1017 1018 int virtio_transport_notify_send_pre_enqueue(struct vsock_sock *vsk, 1019 struct vsock_transport_send_notify_data *data) 1020 { 1021 return 0; 1022 } 1023 EXPORT_SYMBOL_GPL(virtio_transport_notify_send_pre_enqueue); 1024 1025 int virtio_transport_notify_send_post_enqueue(struct vsock_sock *vsk, 1026 ssize_t written, struct vsock_transport_send_notify_data *data) 1027 { 1028 return 0; 1029 } 1030 EXPORT_SYMBOL_GPL(virtio_transport_notify_send_post_enqueue); 1031 1032 u64 virtio_transport_stream_rcvhiwat(struct vsock_sock *vsk) 1033 { 1034 return vsk->buffer_size; 1035 } 1036 EXPORT_SYMBOL_GPL(virtio_transport_stream_rcvhiwat); 1037 1038 bool virtio_transport_stream_is_active(struct vsock_sock *vsk) 1039 { 1040 return true; 1041 } 1042 EXPORT_SYMBOL_GPL(virtio_transport_stream_is_active); 1043 1044 bool virtio_transport_stream_allow(u32 cid, u32 port) 1045 { 1046 return true; 1047 } 1048 EXPORT_SYMBOL_GPL(virtio_transport_stream_allow); 1049 1050 int virtio_transport_dgram_bind(struct vsock_sock *vsk, 1051 struct sockaddr_vm *addr) 1052 { 1053 return -EOPNOTSUPP; 1054 } 1055 EXPORT_SYMBOL_GPL(virtio_transport_dgram_bind); 1056 1057 bool virtio_transport_dgram_allow(u32 cid, u32 port) 1058 { 1059 return false; 1060 } 1061 EXPORT_SYMBOL_GPL(virtio_transport_dgram_allow); 1062 1063 int virtio_transport_connect(struct vsock_sock *vsk) 1064 { 1065 struct virtio_vsock_pkt_info info = { 1066 .op = VIRTIO_VSOCK_OP_REQUEST, 1067 .vsk = vsk, 1068 }; 1069 1070 return virtio_transport_send_pkt_info(vsk, &info); 1071 } 1072 EXPORT_SYMBOL_GPL(virtio_transport_connect); 1073 1074 int virtio_transport_shutdown(struct vsock_sock *vsk, int mode) 1075 { 1076 struct virtio_vsock_pkt_info info = { 1077 .op = VIRTIO_VSOCK_OP_SHUTDOWN, 1078 .flags = (mode & RCV_SHUTDOWN ? 1079 VIRTIO_VSOCK_SHUTDOWN_RCV : 0) | 1080 (mode & SEND_SHUTDOWN ? 1081 VIRTIO_VSOCK_SHUTDOWN_SEND : 0), 1082 .vsk = vsk, 1083 }; 1084 1085 return virtio_transport_send_pkt_info(vsk, &info); 1086 } 1087 EXPORT_SYMBOL_GPL(virtio_transport_shutdown); 1088 1089 int 1090 virtio_transport_dgram_enqueue(struct vsock_sock *vsk, 1091 struct sockaddr_vm *remote_addr, 1092 struct msghdr *msg, 1093 size_t dgram_len) 1094 { 1095 return -EOPNOTSUPP; 1096 } 1097 EXPORT_SYMBOL_GPL(virtio_transport_dgram_enqueue); 1098 1099 ssize_t 1100 virtio_transport_stream_enqueue(struct vsock_sock *vsk, 1101 struct msghdr *msg, 1102 size_t len) 1103 { 1104 struct virtio_vsock_pkt_info info = { 1105 .op = VIRTIO_VSOCK_OP_RW, 1106 .msg = msg, 1107 .pkt_len = len, 1108 .vsk = vsk, 1109 }; 1110 1111 return virtio_transport_send_pkt_info(vsk, &info); 1112 } 1113 EXPORT_SYMBOL_GPL(virtio_transport_stream_enqueue); 1114 1115 void virtio_transport_destruct(struct vsock_sock *vsk) 1116 { 1117 struct virtio_vsock_sock *vvs = vsk->trans; 1118 1119 virtio_transport_cancel_close_work(vsk, true); 1120 1121 kfree(vvs); 1122 vsk->trans = NULL; 1123 } 1124 EXPORT_SYMBOL_GPL(virtio_transport_destruct); 1125 1126 ssize_t virtio_transport_unsent_bytes(struct vsock_sock *vsk) 1127 { 1128 struct virtio_vsock_sock *vvs = vsk->trans; 1129 size_t ret; 1130 1131 spin_lock_bh(&vvs->tx_lock); 1132 ret = vvs->bytes_unsent; 1133 spin_unlock_bh(&vvs->tx_lock); 1134 1135 return ret; 1136 } 1137 EXPORT_SYMBOL_GPL(virtio_transport_unsent_bytes); 1138 1139 static int virtio_transport_reset(struct vsock_sock *vsk, 1140 struct sk_buff *skb) 1141 { 1142 struct virtio_vsock_pkt_info info = { 1143 .op = VIRTIO_VSOCK_OP_RST, 1144 .reply = !!skb, 1145 .vsk = vsk, 1146 }; 1147 1148 /* Send RST only if the original pkt is not a RST pkt */ 1149 if (skb && le16_to_cpu(virtio_vsock_hdr(skb)->op) == VIRTIO_VSOCK_OP_RST) 1150 return 0; 1151 1152 return virtio_transport_send_pkt_info(vsk, &info); 1153 } 1154 1155 /* Normally packets are associated with a socket. There may be no socket if an 1156 * attempt was made to connect to a socket that does not exist. 1157 */ 1158 static int virtio_transport_reset_no_sock(const struct virtio_transport *t, 1159 struct sk_buff *skb) 1160 { 1161 struct virtio_vsock_hdr *hdr = virtio_vsock_hdr(skb); 1162 struct virtio_vsock_pkt_info info = { 1163 .op = VIRTIO_VSOCK_OP_RST, 1164 .type = le16_to_cpu(hdr->type), 1165 .reply = true, 1166 }; 1167 struct sk_buff *reply; 1168 1169 /* Send RST only if the original pkt is not a RST pkt */ 1170 if (le16_to_cpu(hdr->op) == VIRTIO_VSOCK_OP_RST) 1171 return 0; 1172 1173 if (!t) 1174 return -ENOTCONN; 1175 1176 reply = virtio_transport_alloc_skb(&info, 0, false, 1177 le64_to_cpu(hdr->dst_cid), 1178 le32_to_cpu(hdr->dst_port), 1179 le64_to_cpu(hdr->src_cid), 1180 le32_to_cpu(hdr->src_port)); 1181 if (!reply) 1182 return -ENOMEM; 1183 1184 return t->send_pkt(reply); 1185 } 1186 1187 /* This function should be called with sk_lock held and SOCK_DONE set */ 1188 static void virtio_transport_remove_sock(struct vsock_sock *vsk) 1189 { 1190 struct virtio_vsock_sock *vvs = vsk->trans; 1191 1192 /* We don't need to take rx_lock, as the socket is closing and we are 1193 * removing it. 1194 */ 1195 __skb_queue_purge(&vvs->rx_queue); 1196 vsock_remove_sock(vsk); 1197 } 1198 1199 static void virtio_transport_cancel_close_work(struct vsock_sock *vsk, 1200 bool cancel_timeout) 1201 { 1202 struct sock *sk = sk_vsock(vsk); 1203 1204 if (vsk->close_work_scheduled && 1205 (!cancel_timeout || cancel_delayed_work(&vsk->close_work))) { 1206 vsk->close_work_scheduled = false; 1207 1208 virtio_transport_remove_sock(vsk); 1209 1210 /* Release refcnt obtained when we scheduled the timeout */ 1211 sock_put(sk); 1212 } 1213 } 1214 1215 static void virtio_transport_do_close(struct vsock_sock *vsk, 1216 bool cancel_timeout) 1217 { 1218 struct sock *sk = sk_vsock(vsk); 1219 1220 sock_set_flag(sk, SOCK_DONE); 1221 vsk->peer_shutdown = SHUTDOWN_MASK; 1222 if (vsock_stream_has_data(vsk) <= 0) 1223 sk->sk_state = TCP_CLOSING; 1224 sk->sk_state_change(sk); 1225 1226 virtio_transport_cancel_close_work(vsk, cancel_timeout); 1227 } 1228 1229 static void virtio_transport_close_timeout(struct work_struct *work) 1230 { 1231 struct vsock_sock *vsk = 1232 container_of(work, struct vsock_sock, close_work.work); 1233 struct sock *sk = sk_vsock(vsk); 1234 1235 sock_hold(sk); 1236 lock_sock(sk); 1237 1238 if (!sock_flag(sk, SOCK_DONE)) { 1239 (void)virtio_transport_reset(vsk, NULL); 1240 1241 virtio_transport_do_close(vsk, false); 1242 } 1243 1244 vsk->close_work_scheduled = false; 1245 1246 release_sock(sk); 1247 sock_put(sk); 1248 } 1249 1250 /* User context, vsk->sk is locked */ 1251 static bool virtio_transport_close(struct vsock_sock *vsk) 1252 { 1253 struct sock *sk = &vsk->sk; 1254 1255 if (!(sk->sk_state == TCP_ESTABLISHED || 1256 sk->sk_state == TCP_CLOSING)) 1257 return true; 1258 1259 /* Already received SHUTDOWN from peer, reply with RST */ 1260 if ((vsk->peer_shutdown & SHUTDOWN_MASK) == SHUTDOWN_MASK) { 1261 (void)virtio_transport_reset(vsk, NULL); 1262 return true; 1263 } 1264 1265 if ((sk->sk_shutdown & SHUTDOWN_MASK) != SHUTDOWN_MASK) 1266 (void)virtio_transport_shutdown(vsk, SHUTDOWN_MASK); 1267 1268 if (!(current->flags & PF_EXITING)) 1269 vsock_linger(sk); 1270 1271 if (sock_flag(sk, SOCK_DONE)) { 1272 return true; 1273 } 1274 1275 sock_hold(sk); 1276 INIT_DELAYED_WORK(&vsk->close_work, 1277 virtio_transport_close_timeout); 1278 vsk->close_work_scheduled = true; 1279 schedule_delayed_work(&vsk->close_work, VSOCK_CLOSE_TIMEOUT); 1280 return false; 1281 } 1282 1283 void virtio_transport_release(struct vsock_sock *vsk) 1284 { 1285 struct sock *sk = &vsk->sk; 1286 bool remove_sock = true; 1287 1288 if (sk->sk_type == SOCK_STREAM || sk->sk_type == SOCK_SEQPACKET) 1289 remove_sock = virtio_transport_close(vsk); 1290 1291 if (remove_sock) { 1292 sock_set_flag(sk, SOCK_DONE); 1293 virtio_transport_remove_sock(vsk); 1294 } 1295 } 1296 EXPORT_SYMBOL_GPL(virtio_transport_release); 1297 1298 static int 1299 virtio_transport_recv_connecting(struct sock *sk, 1300 struct sk_buff *skb) 1301 { 1302 struct virtio_vsock_hdr *hdr = virtio_vsock_hdr(skb); 1303 struct vsock_sock *vsk = vsock_sk(sk); 1304 int skerr; 1305 int err; 1306 1307 switch (le16_to_cpu(hdr->op)) { 1308 case VIRTIO_VSOCK_OP_RESPONSE: 1309 sk->sk_state = TCP_ESTABLISHED; 1310 sk->sk_socket->state = SS_CONNECTED; 1311 vsock_insert_connected(vsk); 1312 sk->sk_state_change(sk); 1313 break; 1314 case VIRTIO_VSOCK_OP_INVALID: 1315 break; 1316 case VIRTIO_VSOCK_OP_RST: 1317 skerr = ECONNRESET; 1318 err = 0; 1319 goto destroy; 1320 default: 1321 skerr = EPROTO; 1322 err = -EINVAL; 1323 goto destroy; 1324 } 1325 return 0; 1326 1327 destroy: 1328 virtio_transport_reset(vsk, skb); 1329 sk->sk_state = TCP_CLOSE; 1330 sk->sk_err = skerr; 1331 sk_error_report(sk); 1332 return err; 1333 } 1334 1335 static void 1336 virtio_transport_recv_enqueue(struct vsock_sock *vsk, 1337 struct sk_buff *skb) 1338 { 1339 struct virtio_vsock_sock *vvs = vsk->trans; 1340 bool can_enqueue, free_pkt = false; 1341 struct virtio_vsock_hdr *hdr; 1342 u32 len; 1343 1344 hdr = virtio_vsock_hdr(skb); 1345 len = le32_to_cpu(hdr->len); 1346 1347 spin_lock_bh(&vvs->rx_lock); 1348 1349 can_enqueue = virtio_transport_inc_rx_pkt(vvs, len); 1350 if (!can_enqueue) { 1351 free_pkt = true; 1352 goto out; 1353 } 1354 1355 if (le32_to_cpu(hdr->flags) & VIRTIO_VSOCK_SEQ_EOM) 1356 vvs->msg_count++; 1357 1358 /* Try to copy small packets into the buffer of last packet queued, 1359 * to avoid wasting memory queueing the entire buffer with a small 1360 * payload. 1361 */ 1362 if (len <= GOOD_COPY_LEN && !skb_queue_empty(&vvs->rx_queue)) { 1363 struct virtio_vsock_hdr *last_hdr; 1364 struct sk_buff *last_skb; 1365 1366 last_skb = skb_peek_tail(&vvs->rx_queue); 1367 last_hdr = virtio_vsock_hdr(last_skb); 1368 1369 /* If there is space in the last packet queued, we copy the 1370 * new packet in its buffer. We avoid this if the last packet 1371 * queued has VIRTIO_VSOCK_SEQ_EOM set, because this is 1372 * delimiter of SEQPACKET message, so 'pkt' is the first packet 1373 * of a new message. 1374 */ 1375 if (skb->len < skb_tailroom(last_skb) && 1376 !(le32_to_cpu(last_hdr->flags) & VIRTIO_VSOCK_SEQ_EOM)) { 1377 memcpy(skb_put(last_skb, skb->len), skb->data, skb->len); 1378 free_pkt = true; 1379 last_hdr->flags |= hdr->flags; 1380 le32_add_cpu(&last_hdr->len, len); 1381 goto out; 1382 } 1383 } 1384 1385 __skb_queue_tail(&vvs->rx_queue, skb); 1386 1387 out: 1388 spin_unlock_bh(&vvs->rx_lock); 1389 if (free_pkt) 1390 kfree_skb(skb); 1391 } 1392 1393 static int 1394 virtio_transport_recv_connected(struct sock *sk, 1395 struct sk_buff *skb) 1396 { 1397 struct virtio_vsock_hdr *hdr = virtio_vsock_hdr(skb); 1398 struct vsock_sock *vsk = vsock_sk(sk); 1399 int err = 0; 1400 1401 switch (le16_to_cpu(hdr->op)) { 1402 case VIRTIO_VSOCK_OP_RW: 1403 virtio_transport_recv_enqueue(vsk, skb); 1404 vsock_data_ready(sk); 1405 return err; 1406 case VIRTIO_VSOCK_OP_CREDIT_REQUEST: 1407 virtio_transport_send_credit_update(vsk); 1408 break; 1409 case VIRTIO_VSOCK_OP_CREDIT_UPDATE: 1410 sk->sk_write_space(sk); 1411 break; 1412 case VIRTIO_VSOCK_OP_SHUTDOWN: 1413 if (le32_to_cpu(hdr->flags) & VIRTIO_VSOCK_SHUTDOWN_RCV) 1414 vsk->peer_shutdown |= RCV_SHUTDOWN; 1415 if (le32_to_cpu(hdr->flags) & VIRTIO_VSOCK_SHUTDOWN_SEND) 1416 vsk->peer_shutdown |= SEND_SHUTDOWN; 1417 if (vsk->peer_shutdown == SHUTDOWN_MASK) { 1418 if (vsock_stream_has_data(vsk) <= 0 && !sock_flag(sk, SOCK_DONE)) { 1419 (void)virtio_transport_reset(vsk, NULL); 1420 virtio_transport_do_close(vsk, true); 1421 } 1422 /* Remove this socket anyway because the remote peer sent 1423 * the shutdown. This way a new connection will succeed 1424 * if the remote peer uses the same source port, 1425 * even if the old socket is still unreleased, but now disconnected. 1426 */ 1427 vsock_remove_sock(vsk); 1428 } 1429 if (le32_to_cpu(virtio_vsock_hdr(skb)->flags)) 1430 sk->sk_state_change(sk); 1431 break; 1432 case VIRTIO_VSOCK_OP_RST: 1433 virtio_transport_do_close(vsk, true); 1434 break; 1435 default: 1436 err = -EINVAL; 1437 break; 1438 } 1439 1440 kfree_skb(skb); 1441 return err; 1442 } 1443 1444 static void 1445 virtio_transport_recv_disconnecting(struct sock *sk, 1446 struct sk_buff *skb) 1447 { 1448 struct virtio_vsock_hdr *hdr = virtio_vsock_hdr(skb); 1449 struct vsock_sock *vsk = vsock_sk(sk); 1450 1451 if (le16_to_cpu(hdr->op) == VIRTIO_VSOCK_OP_RST) 1452 virtio_transport_do_close(vsk, true); 1453 } 1454 1455 static int 1456 virtio_transport_send_response(struct vsock_sock *vsk, 1457 struct sk_buff *skb) 1458 { 1459 struct virtio_vsock_hdr *hdr = virtio_vsock_hdr(skb); 1460 struct virtio_vsock_pkt_info info = { 1461 .op = VIRTIO_VSOCK_OP_RESPONSE, 1462 .remote_cid = le64_to_cpu(hdr->src_cid), 1463 .remote_port = le32_to_cpu(hdr->src_port), 1464 .reply = true, 1465 .vsk = vsk, 1466 }; 1467 1468 return virtio_transport_send_pkt_info(vsk, &info); 1469 } 1470 1471 static bool virtio_transport_space_update(struct sock *sk, 1472 struct sk_buff *skb) 1473 { 1474 struct virtio_vsock_hdr *hdr = virtio_vsock_hdr(skb); 1475 struct vsock_sock *vsk = vsock_sk(sk); 1476 struct virtio_vsock_sock *vvs = vsk->trans; 1477 bool space_available; 1478 1479 /* Listener sockets are not associated with any transport, so we are 1480 * not able to take the state to see if there is space available in the 1481 * remote peer, but since they are only used to receive requests, we 1482 * can assume that there is always space available in the other peer. 1483 */ 1484 if (!vvs) 1485 return true; 1486 1487 /* buf_alloc and fwd_cnt is always included in the hdr */ 1488 spin_lock_bh(&vvs->tx_lock); 1489 vvs->peer_buf_alloc = le32_to_cpu(hdr->buf_alloc); 1490 vvs->peer_fwd_cnt = le32_to_cpu(hdr->fwd_cnt); 1491 space_available = virtio_transport_has_space(vsk); 1492 spin_unlock_bh(&vvs->tx_lock); 1493 return space_available; 1494 } 1495 1496 /* Handle server socket */ 1497 static int 1498 virtio_transport_recv_listen(struct sock *sk, struct sk_buff *skb, 1499 struct virtio_transport *t) 1500 { 1501 struct virtio_vsock_hdr *hdr = virtio_vsock_hdr(skb); 1502 struct vsock_sock *vsk = vsock_sk(sk); 1503 struct vsock_sock *vchild; 1504 struct sock *child; 1505 int ret; 1506 1507 if (le16_to_cpu(hdr->op) != VIRTIO_VSOCK_OP_REQUEST) { 1508 virtio_transport_reset_no_sock(t, skb); 1509 return -EINVAL; 1510 } 1511 1512 if (sk_acceptq_is_full(sk)) { 1513 virtio_transport_reset_no_sock(t, skb); 1514 return -ENOMEM; 1515 } 1516 1517 /* __vsock_release() might have already flushed accept_queue. 1518 * Subsequent enqueues would lead to a memory leak. 1519 */ 1520 if (sk->sk_shutdown == SHUTDOWN_MASK) { 1521 virtio_transport_reset_no_sock(t, skb); 1522 return -ESHUTDOWN; 1523 } 1524 1525 child = vsock_create_connected(sk); 1526 if (!child) { 1527 virtio_transport_reset_no_sock(t, skb); 1528 return -ENOMEM; 1529 } 1530 1531 sk_acceptq_added(sk); 1532 1533 lock_sock_nested(child, SINGLE_DEPTH_NESTING); 1534 1535 child->sk_state = TCP_ESTABLISHED; 1536 1537 vchild = vsock_sk(child); 1538 vsock_addr_init(&vchild->local_addr, le64_to_cpu(hdr->dst_cid), 1539 le32_to_cpu(hdr->dst_port)); 1540 vsock_addr_init(&vchild->remote_addr, le64_to_cpu(hdr->src_cid), 1541 le32_to_cpu(hdr->src_port)); 1542 1543 ret = vsock_assign_transport(vchild, vsk); 1544 /* Transport assigned (looking at remote_addr) must be the same 1545 * where we received the request. 1546 */ 1547 if (ret || vchild->transport != &t->transport) { 1548 release_sock(child); 1549 virtio_transport_reset_no_sock(t, skb); 1550 sock_put(child); 1551 return ret; 1552 } 1553 1554 if (virtio_transport_space_update(child, skb)) 1555 child->sk_write_space(child); 1556 1557 vsock_insert_connected(vchild); 1558 vsock_enqueue_accept(sk, child); 1559 virtio_transport_send_response(vchild, skb); 1560 1561 release_sock(child); 1562 1563 sk->sk_data_ready(sk); 1564 return 0; 1565 } 1566 1567 static bool virtio_transport_valid_type(u16 type) 1568 { 1569 return (type == VIRTIO_VSOCK_TYPE_STREAM) || 1570 (type == VIRTIO_VSOCK_TYPE_SEQPACKET); 1571 } 1572 1573 /* We are under the virtio-vsock's vsock->rx_lock or vhost-vsock's vq->mutex 1574 * lock. 1575 */ 1576 void virtio_transport_recv_pkt(struct virtio_transport *t, 1577 struct sk_buff *skb) 1578 { 1579 struct virtio_vsock_hdr *hdr = virtio_vsock_hdr(skb); 1580 struct sockaddr_vm src, dst; 1581 struct vsock_sock *vsk; 1582 struct sock *sk; 1583 bool space_available; 1584 1585 vsock_addr_init(&src, le64_to_cpu(hdr->src_cid), 1586 le32_to_cpu(hdr->src_port)); 1587 vsock_addr_init(&dst, le64_to_cpu(hdr->dst_cid), 1588 le32_to_cpu(hdr->dst_port)); 1589 1590 trace_virtio_transport_recv_pkt(src.svm_cid, src.svm_port, 1591 dst.svm_cid, dst.svm_port, 1592 le32_to_cpu(hdr->len), 1593 le16_to_cpu(hdr->type), 1594 le16_to_cpu(hdr->op), 1595 le32_to_cpu(hdr->flags), 1596 le32_to_cpu(hdr->buf_alloc), 1597 le32_to_cpu(hdr->fwd_cnt)); 1598 1599 if (!virtio_transport_valid_type(le16_to_cpu(hdr->type))) { 1600 (void)virtio_transport_reset_no_sock(t, skb); 1601 goto free_pkt; 1602 } 1603 1604 /* The socket must be in connected or bound table 1605 * otherwise send reset back 1606 */ 1607 sk = vsock_find_connected_socket(&src, &dst); 1608 if (!sk) { 1609 sk = vsock_find_bound_socket(&dst); 1610 if (!sk) { 1611 (void)virtio_transport_reset_no_sock(t, skb); 1612 goto free_pkt; 1613 } 1614 } 1615 1616 if (virtio_transport_get_type(sk) != le16_to_cpu(hdr->type)) { 1617 (void)virtio_transport_reset_no_sock(t, skb); 1618 sock_put(sk); 1619 goto free_pkt; 1620 } 1621 1622 if (!skb_set_owner_sk_safe(skb, sk)) { 1623 WARN_ONCE(1, "receiving vsock socket has sk_refcnt == 0\n"); 1624 goto free_pkt; 1625 } 1626 1627 vsk = vsock_sk(sk); 1628 1629 lock_sock(sk); 1630 1631 /* Check if sk has been closed or assigned to another transport before 1632 * lock_sock (note: listener sockets are not assigned to any transport) 1633 */ 1634 if (sock_flag(sk, SOCK_DONE) || 1635 (sk->sk_state != TCP_LISTEN && vsk->transport != &t->transport)) { 1636 (void)virtio_transport_reset_no_sock(t, skb); 1637 release_sock(sk); 1638 sock_put(sk); 1639 goto free_pkt; 1640 } 1641 1642 space_available = virtio_transport_space_update(sk, skb); 1643 1644 /* Update CID in case it has changed after a transport reset event */ 1645 if (vsk->local_addr.svm_cid != VMADDR_CID_ANY) 1646 vsk->local_addr.svm_cid = dst.svm_cid; 1647 1648 if (space_available) 1649 sk->sk_write_space(sk); 1650 1651 switch (sk->sk_state) { 1652 case TCP_LISTEN: 1653 virtio_transport_recv_listen(sk, skb, t); 1654 kfree_skb(skb); 1655 break; 1656 case TCP_SYN_SENT: 1657 virtio_transport_recv_connecting(sk, skb); 1658 kfree_skb(skb); 1659 break; 1660 case TCP_ESTABLISHED: 1661 virtio_transport_recv_connected(sk, skb); 1662 break; 1663 case TCP_CLOSING: 1664 virtio_transport_recv_disconnecting(sk, skb); 1665 kfree_skb(skb); 1666 break; 1667 default: 1668 (void)virtio_transport_reset_no_sock(t, skb); 1669 kfree_skb(skb); 1670 break; 1671 } 1672 1673 release_sock(sk); 1674 1675 /* Release refcnt obtained when we fetched this socket out of the 1676 * bound or connected list. 1677 */ 1678 sock_put(sk); 1679 return; 1680 1681 free_pkt: 1682 kfree_skb(skb); 1683 } 1684 EXPORT_SYMBOL_GPL(virtio_transport_recv_pkt); 1685 1686 /* Remove skbs found in a queue that have a vsk that matches. 1687 * 1688 * Each skb is freed. 1689 * 1690 * Returns the count of skbs that were reply packets. 1691 */ 1692 int virtio_transport_purge_skbs(void *vsk, struct sk_buff_head *queue) 1693 { 1694 struct sk_buff_head freeme; 1695 struct sk_buff *skb, *tmp; 1696 int cnt = 0; 1697 1698 skb_queue_head_init(&freeme); 1699 1700 spin_lock_bh(&queue->lock); 1701 skb_queue_walk_safe(queue, skb, tmp) { 1702 if (vsock_sk(skb->sk) != vsk) 1703 continue; 1704 1705 __skb_unlink(skb, queue); 1706 __skb_queue_tail(&freeme, skb); 1707 1708 if (virtio_vsock_skb_reply(skb)) 1709 cnt++; 1710 } 1711 spin_unlock_bh(&queue->lock); 1712 1713 __skb_queue_purge(&freeme); 1714 1715 return cnt; 1716 } 1717 EXPORT_SYMBOL_GPL(virtio_transport_purge_skbs); 1718 1719 int virtio_transport_read_skb(struct vsock_sock *vsk, skb_read_actor_t recv_actor) 1720 { 1721 struct virtio_vsock_sock *vvs = vsk->trans; 1722 struct sock *sk = sk_vsock(vsk); 1723 struct virtio_vsock_hdr *hdr; 1724 struct sk_buff *skb; 1725 u32 pkt_len; 1726 int off = 0; 1727 int err; 1728 1729 spin_lock_bh(&vvs->rx_lock); 1730 /* Use __skb_recv_datagram() for race-free handling of the receive. It 1731 * works for types other than dgrams. 1732 */ 1733 skb = __skb_recv_datagram(sk, &vvs->rx_queue, MSG_DONTWAIT, &off, &err); 1734 if (!skb) { 1735 spin_unlock_bh(&vvs->rx_lock); 1736 return err; 1737 } 1738 1739 hdr = virtio_vsock_hdr(skb); 1740 if (le32_to_cpu(hdr->flags) & VIRTIO_VSOCK_SEQ_EOM) 1741 vvs->msg_count--; 1742 1743 pkt_len = le32_to_cpu(hdr->len); 1744 virtio_transport_dec_rx_pkt(vvs, pkt_len, pkt_len); 1745 spin_unlock_bh(&vvs->rx_lock); 1746 1747 virtio_transport_send_credit_update(vsk); 1748 1749 return recv_actor(sk, skb); 1750 } 1751 EXPORT_SYMBOL_GPL(virtio_transport_read_skb); 1752 1753 int virtio_transport_notify_set_rcvlowat(struct vsock_sock *vsk, int val) 1754 { 1755 struct virtio_vsock_sock *vvs = vsk->trans; 1756 bool send_update; 1757 1758 spin_lock_bh(&vvs->rx_lock); 1759 1760 /* If number of available bytes is less than new SO_RCVLOWAT value, 1761 * kick sender to send more data, because sender may sleep in its 1762 * 'send()' syscall waiting for enough space at our side. Also 1763 * don't send credit update when peer already knows actual value - 1764 * such transmission will be useless. 1765 */ 1766 send_update = (vvs->rx_bytes < val) && 1767 (vvs->fwd_cnt != vvs->last_fwd_cnt); 1768 1769 spin_unlock_bh(&vvs->rx_lock); 1770 1771 if (send_update) { 1772 int err; 1773 1774 err = virtio_transport_send_credit_update(vsk); 1775 if (err < 0) 1776 return err; 1777 } 1778 1779 return 0; 1780 } 1781 EXPORT_SYMBOL_GPL(virtio_transport_notify_set_rcvlowat); 1782 1783 MODULE_LICENSE("GPL v2"); 1784 MODULE_AUTHOR("Asias He"); 1785 MODULE_DESCRIPTION("common code for virtio vsock"); 1786