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 static s64 virtio_transport_has_space(struct virtio_vsock_sock *vvs); 32 33 static const struct virtio_transport * 34 virtio_transport_get_ops(struct vsock_sock *vsk) 35 { 36 const struct vsock_transport *t = vsock_core_get_transport(vsk); 37 38 if (WARN_ON(!t)) 39 return NULL; 40 41 return container_of(t, struct virtio_transport, transport); 42 } 43 44 static bool virtio_transport_can_zcopy(const struct virtio_transport *t_ops, 45 struct virtio_vsock_pkt_info *info, 46 size_t pkt_len) 47 { 48 struct iov_iter *iov_iter; 49 50 if (!info->msg) 51 return false; 52 53 iov_iter = &info->msg->msg_iter; 54 55 if (iov_iter->iov_offset) 56 return false; 57 58 /* We can't send whole iov. */ 59 if (iov_iter->count > pkt_len) 60 return false; 61 62 /* Check that transport can send data in zerocopy mode. */ 63 t_ops = virtio_transport_get_ops(info->vsk); 64 65 if (t_ops->can_msgzerocopy) { 66 int pages_to_send = iov_iter_npages(iov_iter, MAX_SKB_FRAGS); 67 68 /* +1 is for packet header. */ 69 return t_ops->can_msgzerocopy(pages_to_send + 1); 70 } 71 72 return true; 73 } 74 75 static int virtio_transport_init_zcopy_skb(struct vsock_sock *vsk, 76 struct sk_buff *skb, 77 struct msghdr *msg, 78 bool zerocopy) 79 { 80 struct ubuf_info *uarg; 81 82 if (msg->msg_ubuf) { 83 uarg = msg->msg_ubuf; 84 net_zcopy_get(uarg); 85 } else { 86 struct iov_iter *iter = &msg->msg_iter; 87 struct ubuf_info_msgzc *uarg_zc; 88 89 uarg = msg_zerocopy_realloc(sk_vsock(vsk), 90 iter->count, 91 NULL, false); 92 if (!uarg) 93 return -1; 94 95 uarg_zc = uarg_to_msgzc(uarg); 96 uarg_zc->zerocopy = zerocopy ? 1 : 0; 97 } 98 99 skb_zcopy_init(skb, uarg); 100 101 return 0; 102 } 103 104 static int virtio_transport_fill_skb(struct sk_buff *skb, 105 struct virtio_vsock_pkt_info *info, 106 size_t len, 107 bool zcopy) 108 { 109 struct msghdr *msg = info->msg; 110 111 if (zcopy) 112 return __zerocopy_sg_from_iter(msg, NULL, skb, 113 &msg->msg_iter, len, NULL); 114 115 virtio_vsock_skb_put(skb, len); 116 return skb_copy_datagram_from_iter_full(skb, 0, &msg->msg_iter, len); 117 } 118 119 static void virtio_transport_init_hdr(struct sk_buff *skb, 120 struct virtio_vsock_pkt_info *info, 121 size_t payload_len, 122 u32 src_cid, 123 u32 src_port, 124 u32 dst_cid, 125 u32 dst_port) 126 { 127 struct virtio_vsock_hdr *hdr; 128 129 hdr = virtio_vsock_hdr(skb); 130 hdr->type = cpu_to_le16(info->type); 131 hdr->op = cpu_to_le16(info->op); 132 hdr->src_cid = cpu_to_le64(src_cid); 133 hdr->dst_cid = cpu_to_le64(dst_cid); 134 hdr->src_port = cpu_to_le32(src_port); 135 hdr->dst_port = cpu_to_le32(dst_port); 136 hdr->flags = cpu_to_le32(info->flags); 137 hdr->len = cpu_to_le32(payload_len); 138 hdr->buf_alloc = cpu_to_le32(0); 139 hdr->fwd_cnt = cpu_to_le32(0); 140 } 141 142 static void virtio_transport_copy_nonlinear_skb(const struct sk_buff *skb, 143 void *dst, 144 size_t len) 145 { 146 struct iov_iter iov_iter = { 0 }; 147 struct kvec kvec; 148 size_t to_copy; 149 150 kvec.iov_base = dst; 151 kvec.iov_len = len; 152 153 iov_iter.iter_type = ITER_KVEC; 154 iov_iter.kvec = &kvec; 155 iov_iter.nr_segs = 1; 156 157 to_copy = min_t(size_t, len, skb->len); 158 159 skb_copy_datagram_iter(skb, VIRTIO_VSOCK_SKB_CB(skb)->offset, 160 &iov_iter, to_copy); 161 } 162 163 /* Packet capture */ 164 static struct sk_buff *virtio_transport_build_skb(void *opaque) 165 { 166 struct virtio_vsock_hdr *pkt_hdr; 167 struct sk_buff *pkt = opaque; 168 struct af_vsockmon_hdr *hdr; 169 struct sk_buff *skb; 170 size_t payload_len; 171 172 /* A packet could be split to fit the RX buffer, so we can retrieve 173 * the payload length from the header and the buffer pointer taking 174 * care of the offset in the original packet. 175 */ 176 pkt_hdr = virtio_vsock_hdr(pkt); 177 payload_len = pkt->len; 178 179 skb = alloc_skb(sizeof(*hdr) + sizeof(*pkt_hdr) + payload_len, 180 GFP_ATOMIC); 181 if (!skb) 182 return NULL; 183 184 hdr = skb_put(skb, sizeof(*hdr)); 185 186 /* pkt->hdr is little-endian so no need to byteswap here */ 187 hdr->src_cid = pkt_hdr->src_cid; 188 hdr->src_port = pkt_hdr->src_port; 189 hdr->dst_cid = pkt_hdr->dst_cid; 190 hdr->dst_port = pkt_hdr->dst_port; 191 192 hdr->transport = cpu_to_le16(AF_VSOCK_TRANSPORT_VIRTIO); 193 hdr->len = cpu_to_le16(sizeof(*pkt_hdr)); 194 memset(hdr->reserved, 0, sizeof(hdr->reserved)); 195 196 switch (le16_to_cpu(pkt_hdr->op)) { 197 case VIRTIO_VSOCK_OP_REQUEST: 198 case VIRTIO_VSOCK_OP_RESPONSE: 199 hdr->op = cpu_to_le16(AF_VSOCK_OP_CONNECT); 200 break; 201 case VIRTIO_VSOCK_OP_RST: 202 case VIRTIO_VSOCK_OP_SHUTDOWN: 203 hdr->op = cpu_to_le16(AF_VSOCK_OP_DISCONNECT); 204 break; 205 case VIRTIO_VSOCK_OP_RW: 206 hdr->op = cpu_to_le16(AF_VSOCK_OP_PAYLOAD); 207 break; 208 case VIRTIO_VSOCK_OP_CREDIT_UPDATE: 209 case VIRTIO_VSOCK_OP_CREDIT_REQUEST: 210 hdr->op = cpu_to_le16(AF_VSOCK_OP_CONTROL); 211 break; 212 default: 213 hdr->op = cpu_to_le16(AF_VSOCK_OP_UNKNOWN); 214 break; 215 } 216 217 skb_put_data(skb, pkt_hdr, sizeof(*pkt_hdr)); 218 219 if (payload_len) { 220 if (skb_is_nonlinear(pkt)) { 221 void *data = skb_put(skb, payload_len); 222 223 virtio_transport_copy_nonlinear_skb(pkt, data, payload_len); 224 } else { 225 skb_put_data(skb, pkt->data, payload_len); 226 } 227 } 228 229 return skb; 230 } 231 232 void virtio_transport_deliver_tap_pkt(struct sk_buff *skb) 233 { 234 if (virtio_vsock_skb_tap_delivered(skb)) 235 return; 236 237 vsock_deliver_tap(virtio_transport_build_skb, skb); 238 virtio_vsock_skb_set_tap_delivered(skb); 239 } 240 EXPORT_SYMBOL_GPL(virtio_transport_deliver_tap_pkt); 241 242 static u16 virtio_transport_get_type(struct sock *sk) 243 { 244 if (sk->sk_type == SOCK_STREAM) 245 return VIRTIO_VSOCK_TYPE_STREAM; 246 else 247 return VIRTIO_VSOCK_TYPE_SEQPACKET; 248 } 249 250 /* Returns new sk_buff on success, otherwise returns NULL. */ 251 static struct sk_buff *virtio_transport_alloc_skb(struct virtio_vsock_pkt_info *info, 252 size_t payload_len, 253 bool zcopy, 254 u32 src_cid, 255 u32 src_port, 256 u32 dst_cid, 257 u32 dst_port) 258 { 259 struct vsock_sock *vsk; 260 struct sk_buff *skb; 261 size_t skb_len; 262 263 skb_len = VIRTIO_VSOCK_SKB_HEADROOM; 264 265 if (!zcopy) 266 skb_len += payload_len; 267 268 skb = virtio_vsock_alloc_skb(skb_len, GFP_KERNEL); 269 if (!skb) 270 return NULL; 271 272 virtio_transport_init_hdr(skb, info, payload_len, src_cid, src_port, 273 dst_cid, dst_port); 274 275 vsk = info->vsk; 276 277 /* If 'vsk' != NULL then payload is always present, so we 278 * will never call '__zerocopy_sg_from_iter()' below without 279 * setting skb owner in 'skb_set_owner_w()'. The only case 280 * when 'vsk' == NULL is VIRTIO_VSOCK_OP_RST control message 281 * without payload. 282 */ 283 WARN_ON_ONCE(!(vsk && (info->msg && payload_len)) && zcopy); 284 285 /* Set owner here, because '__zerocopy_sg_from_iter()' uses 286 * owner of skb without check to update 'sk_wmem_alloc'. 287 */ 288 if (vsk) 289 skb_set_owner_w(skb, sk_vsock(vsk)); 290 291 if (info->msg && payload_len > 0) { 292 int err; 293 294 err = virtio_transport_fill_skb(skb, info, payload_len, zcopy); 295 if (err) 296 goto out; 297 298 if (msg_data_left(info->msg) == 0 && 299 info->type == VIRTIO_VSOCK_TYPE_SEQPACKET) { 300 struct virtio_vsock_hdr *hdr = virtio_vsock_hdr(skb); 301 302 hdr->flags |= cpu_to_le32(VIRTIO_VSOCK_SEQ_EOM); 303 304 if (info->msg->msg_flags & MSG_EOR) 305 hdr->flags |= cpu_to_le32(VIRTIO_VSOCK_SEQ_EOR); 306 } 307 } 308 309 if (info->reply) 310 virtio_vsock_skb_set_reply(skb); 311 312 trace_virtio_transport_alloc_pkt(src_cid, src_port, 313 dst_cid, dst_port, 314 payload_len, 315 info->type, 316 info->op, 317 info->flags, 318 zcopy); 319 320 return skb; 321 out: 322 kfree_skb(skb); 323 return NULL; 324 } 325 326 /* This function can only be used on connecting/connected sockets, 327 * since a socket assigned to a transport is required. 328 * 329 * Do not use on listener sockets! 330 */ 331 static int virtio_transport_send_pkt_info(struct vsock_sock *vsk, 332 struct virtio_vsock_pkt_info *info) 333 { 334 u32 max_skb_len = VIRTIO_VSOCK_MAX_PKT_BUF_SIZE; 335 u32 src_cid, src_port, dst_cid, dst_port; 336 const struct virtio_transport *t_ops; 337 struct virtio_vsock_sock *vvs; 338 u32 pkt_len = info->pkt_len; 339 bool can_zcopy = false; 340 u32 rest_len; 341 int ret; 342 343 info->type = virtio_transport_get_type(sk_vsock(vsk)); 344 345 t_ops = virtio_transport_get_ops(vsk); 346 if (unlikely(!t_ops)) 347 return -EFAULT; 348 349 src_cid = t_ops->transport.get_local_cid(); 350 src_port = vsk->local_addr.svm_port; 351 if (!info->remote_cid) { 352 dst_cid = vsk->remote_addr.svm_cid; 353 dst_port = vsk->remote_addr.svm_port; 354 } else { 355 dst_cid = info->remote_cid; 356 dst_port = info->remote_port; 357 } 358 359 vvs = vsk->trans; 360 361 /* virtio_transport_get_credit might return less than pkt_len credit */ 362 pkt_len = virtio_transport_get_credit(vvs, pkt_len); 363 364 /* Do not send zero length OP_RW pkt */ 365 if (pkt_len == 0 && info->op == VIRTIO_VSOCK_OP_RW) 366 return pkt_len; 367 368 if (info->msg) { 369 /* If zerocopy is not enabled by 'setsockopt()', we behave as 370 * there is no MSG_ZEROCOPY flag set. 371 */ 372 if (!sock_flag(sk_vsock(vsk), SOCK_ZEROCOPY)) 373 info->msg->msg_flags &= ~MSG_ZEROCOPY; 374 375 if (info->msg->msg_flags & MSG_ZEROCOPY) 376 can_zcopy = virtio_transport_can_zcopy(t_ops, info, pkt_len); 377 378 if (can_zcopy) 379 max_skb_len = min_t(u32, VIRTIO_VSOCK_MAX_PKT_BUF_SIZE, 380 (MAX_SKB_FRAGS * PAGE_SIZE)); 381 } 382 383 rest_len = pkt_len; 384 385 do { 386 struct sk_buff *skb; 387 size_t skb_len; 388 389 skb_len = min(max_skb_len, rest_len); 390 391 skb = virtio_transport_alloc_skb(info, skb_len, can_zcopy, 392 src_cid, src_port, 393 dst_cid, dst_port); 394 if (!skb) { 395 ret = -ENOMEM; 396 break; 397 } 398 399 /* We process buffer part by part, allocating skb on 400 * each iteration. If this is last skb for this buffer 401 * and MSG_ZEROCOPY mode is in use - we must allocate 402 * completion for the current syscall. 403 */ 404 if (info->msg && info->msg->msg_flags & MSG_ZEROCOPY && 405 skb_len == rest_len && info->op == VIRTIO_VSOCK_OP_RW) { 406 if (virtio_transport_init_zcopy_skb(vsk, skb, 407 info->msg, 408 can_zcopy)) { 409 kfree_skb(skb); 410 ret = -ENOMEM; 411 break; 412 } 413 } 414 415 virtio_transport_inc_tx_pkt(vvs, skb); 416 417 ret = t_ops->send_pkt(skb, info->net); 418 if (ret < 0) 419 break; 420 421 /* Both virtio and vhost 'send_pkt()' returns 'skb_len', 422 * but for reliability use 'ret' instead of 'skb_len'. 423 * Also if partial send happens (e.g. 'ret' != 'skb_len') 424 * somehow, we break this loop, but account such returned 425 * value in 'virtio_transport_put_credit()'. 426 */ 427 rest_len -= ret; 428 429 if (WARN_ONCE(ret != skb_len, 430 "'send_pkt()' returns %i, but %zu expected\n", 431 ret, skb_len)) 432 break; 433 } while (rest_len); 434 435 virtio_transport_put_credit(vvs, rest_len); 436 437 /* Return number of bytes, if any data has been sent. */ 438 if (rest_len != pkt_len) 439 ret = pkt_len - rest_len; 440 441 return ret; 442 } 443 444 static bool virtio_transport_inc_rx_pkt(struct virtio_vsock_sock *vvs, 445 u32 len) 446 { 447 if (vvs->buf_used + len > vvs->buf_alloc) 448 return false; 449 450 vvs->rx_bytes += len; 451 vvs->buf_used += len; 452 return true; 453 } 454 455 static void virtio_transport_dec_rx_pkt(struct virtio_vsock_sock *vvs, 456 u32 bytes_read, u32 bytes_dequeued) 457 { 458 vvs->rx_bytes -= bytes_read; 459 vvs->buf_used -= bytes_dequeued; 460 vvs->fwd_cnt += bytes_dequeued; 461 } 462 463 void virtio_transport_inc_tx_pkt(struct virtio_vsock_sock *vvs, struct sk_buff *skb) 464 { 465 struct virtio_vsock_hdr *hdr = virtio_vsock_hdr(skb); 466 467 spin_lock_bh(&vvs->rx_lock); 468 vvs->last_fwd_cnt = vvs->fwd_cnt; 469 hdr->fwd_cnt = cpu_to_le32(vvs->fwd_cnt); 470 hdr->buf_alloc = cpu_to_le32(vvs->buf_alloc); 471 spin_unlock_bh(&vvs->rx_lock); 472 } 473 EXPORT_SYMBOL_GPL(virtio_transport_inc_tx_pkt); 474 475 void virtio_transport_consume_skb_sent(struct sk_buff *skb, bool consume) 476 { 477 struct sock *s = skb->sk; 478 479 if (s && skb->len) { 480 struct vsock_sock *vs = vsock_sk(s); 481 struct virtio_vsock_sock *vvs; 482 483 vvs = vs->trans; 484 485 spin_lock_bh(&vvs->tx_lock); 486 vvs->bytes_unsent -= skb->len; 487 spin_unlock_bh(&vvs->tx_lock); 488 } 489 490 if (consume) 491 consume_skb(skb); 492 } 493 EXPORT_SYMBOL_GPL(virtio_transport_consume_skb_sent); 494 495 u32 virtio_transport_get_credit(struct virtio_vsock_sock *vvs, u32 credit) 496 { 497 u32 ret; 498 499 if (!credit) 500 return 0; 501 502 spin_lock_bh(&vvs->tx_lock); 503 ret = min_t(u32, credit, virtio_transport_has_space(vvs)); 504 vvs->tx_cnt += ret; 505 vvs->bytes_unsent += ret; 506 spin_unlock_bh(&vvs->tx_lock); 507 508 return ret; 509 } 510 EXPORT_SYMBOL_GPL(virtio_transport_get_credit); 511 512 void virtio_transport_put_credit(struct virtio_vsock_sock *vvs, u32 credit) 513 { 514 if (!credit) 515 return; 516 517 spin_lock_bh(&vvs->tx_lock); 518 vvs->tx_cnt -= credit; 519 vvs->bytes_unsent -= credit; 520 spin_unlock_bh(&vvs->tx_lock); 521 } 522 EXPORT_SYMBOL_GPL(virtio_transport_put_credit); 523 524 static int virtio_transport_send_credit_update(struct vsock_sock *vsk) 525 { 526 struct virtio_vsock_pkt_info info = { 527 .op = VIRTIO_VSOCK_OP_CREDIT_UPDATE, 528 .vsk = vsk, 529 .net = sock_net(sk_vsock(vsk)), 530 }; 531 532 return virtio_transport_send_pkt_info(vsk, &info); 533 } 534 535 static ssize_t 536 virtio_transport_stream_do_peek(struct vsock_sock *vsk, 537 struct msghdr *msg, 538 size_t len) 539 { 540 struct virtio_vsock_sock *vvs = vsk->trans; 541 struct sk_buff *skb; 542 size_t total = 0; 543 int err; 544 545 spin_lock_bh(&vvs->rx_lock); 546 547 skb_queue_walk(&vvs->rx_queue, skb) { 548 size_t bytes; 549 550 bytes = len - total; 551 if (bytes > skb->len) 552 bytes = skb->len; 553 554 spin_unlock_bh(&vvs->rx_lock); 555 556 /* sk_lock is held by caller so no one else can dequeue. 557 * Unlock rx_lock since skb_copy_datagram_iter() may sleep. 558 */ 559 err = skb_copy_datagram_iter(skb, VIRTIO_VSOCK_SKB_CB(skb)->offset, 560 &msg->msg_iter, bytes); 561 if (err) 562 goto out; 563 564 total += bytes; 565 566 spin_lock_bh(&vvs->rx_lock); 567 568 if (total == len) 569 break; 570 } 571 572 spin_unlock_bh(&vvs->rx_lock); 573 574 return total; 575 576 out: 577 if (total) 578 err = total; 579 return err; 580 } 581 582 static ssize_t 583 virtio_transport_stream_do_dequeue(struct vsock_sock *vsk, 584 struct msghdr *msg, 585 size_t len) 586 { 587 struct virtio_vsock_sock *vvs = vsk->trans; 588 struct sk_buff *skb; 589 u32 fwd_cnt_delta; 590 bool low_rx_bytes; 591 int err = -EFAULT; 592 size_t total = 0; 593 u32 free_space; 594 595 spin_lock_bh(&vvs->rx_lock); 596 597 if (WARN_ONCE(skb_queue_empty(&vvs->rx_queue) && vvs->rx_bytes, 598 "rx_queue is empty, but rx_bytes is non-zero\n")) { 599 spin_unlock_bh(&vvs->rx_lock); 600 return err; 601 } 602 603 while (total < len && !skb_queue_empty(&vvs->rx_queue)) { 604 size_t bytes, dequeued = 0; 605 606 skb = skb_peek(&vvs->rx_queue); 607 608 bytes = min_t(size_t, len - total, 609 skb->len - VIRTIO_VSOCK_SKB_CB(skb)->offset); 610 611 /* sk_lock is held by caller so no one else can dequeue. 612 * Unlock rx_lock since skb_copy_datagram_iter() may sleep. 613 */ 614 spin_unlock_bh(&vvs->rx_lock); 615 616 err = skb_copy_datagram_iter(skb, 617 VIRTIO_VSOCK_SKB_CB(skb)->offset, 618 &msg->msg_iter, bytes); 619 if (err) 620 goto out; 621 622 spin_lock_bh(&vvs->rx_lock); 623 624 total += bytes; 625 626 VIRTIO_VSOCK_SKB_CB(skb)->offset += bytes; 627 628 if (skb->len == VIRTIO_VSOCK_SKB_CB(skb)->offset) { 629 dequeued = le32_to_cpu(virtio_vsock_hdr(skb)->len); 630 __skb_unlink(skb, &vvs->rx_queue); 631 consume_skb(skb); 632 } 633 634 virtio_transport_dec_rx_pkt(vvs, bytes, dequeued); 635 } 636 637 fwd_cnt_delta = vvs->fwd_cnt - vvs->last_fwd_cnt; 638 free_space = vvs->buf_alloc - fwd_cnt_delta; 639 low_rx_bytes = (vvs->rx_bytes < 640 sock_rcvlowat(sk_vsock(vsk), 0, INT_MAX)); 641 642 spin_unlock_bh(&vvs->rx_lock); 643 644 /* To reduce the number of credit update messages, 645 * don't update credits as long as lots of space is available. 646 * Note: the limit chosen here is arbitrary. Setting the limit 647 * too high causes extra messages. Too low causes transmitter 648 * stalls. As stalls are in theory more expensive than extra 649 * messages, we set the limit to a high value. TODO: experiment 650 * with different values. Also send credit update message when 651 * number of bytes in rx queue is not enough to wake up reader. 652 */ 653 if (fwd_cnt_delta && 654 (free_space < VIRTIO_VSOCK_MAX_PKT_BUF_SIZE || low_rx_bytes)) 655 virtio_transport_send_credit_update(vsk); 656 657 return total; 658 659 out: 660 if (total) 661 err = total; 662 return err; 663 } 664 665 static ssize_t 666 virtio_transport_seqpacket_do_peek(struct vsock_sock *vsk, 667 struct msghdr *msg) 668 { 669 struct virtio_vsock_sock *vvs = vsk->trans; 670 struct sk_buff *skb; 671 size_t total, len; 672 673 spin_lock_bh(&vvs->rx_lock); 674 675 if (!vvs->msg_count) { 676 spin_unlock_bh(&vvs->rx_lock); 677 return 0; 678 } 679 680 total = 0; 681 len = msg_data_left(msg); 682 683 skb_queue_walk(&vvs->rx_queue, skb) { 684 struct virtio_vsock_hdr *hdr; 685 686 if (total < len) { 687 size_t bytes; 688 int err; 689 690 bytes = len - total; 691 if (bytes > skb->len) 692 bytes = skb->len; 693 694 spin_unlock_bh(&vvs->rx_lock); 695 696 /* sk_lock is held by caller so no one else can dequeue. 697 * Unlock rx_lock since skb_copy_datagram_iter() may sleep. 698 */ 699 err = skb_copy_datagram_iter(skb, VIRTIO_VSOCK_SKB_CB(skb)->offset, 700 &msg->msg_iter, bytes); 701 if (err) 702 return err; 703 704 spin_lock_bh(&vvs->rx_lock); 705 } 706 707 total += skb->len; 708 hdr = virtio_vsock_hdr(skb); 709 710 if (le32_to_cpu(hdr->flags) & VIRTIO_VSOCK_SEQ_EOM) { 711 if (le32_to_cpu(hdr->flags) & VIRTIO_VSOCK_SEQ_EOR) 712 msg->msg_flags |= MSG_EOR; 713 714 break; 715 } 716 } 717 718 spin_unlock_bh(&vvs->rx_lock); 719 720 return total; 721 } 722 723 static int virtio_transport_seqpacket_do_dequeue(struct vsock_sock *vsk, 724 struct msghdr *msg, 725 int flags) 726 { 727 struct virtio_vsock_sock *vvs = vsk->trans; 728 int dequeued_len = 0; 729 size_t user_buf_len = msg_data_left(msg); 730 bool msg_ready = false; 731 struct sk_buff *skb; 732 733 spin_lock_bh(&vvs->rx_lock); 734 735 if (vvs->msg_count == 0) { 736 spin_unlock_bh(&vvs->rx_lock); 737 return 0; 738 } 739 740 while (!msg_ready) { 741 struct virtio_vsock_hdr *hdr; 742 size_t pkt_len; 743 744 skb = __skb_dequeue(&vvs->rx_queue); 745 if (!skb) 746 break; 747 hdr = virtio_vsock_hdr(skb); 748 pkt_len = (size_t)le32_to_cpu(hdr->len); 749 750 if (dequeued_len >= 0) { 751 size_t bytes_to_copy; 752 753 bytes_to_copy = min(user_buf_len, pkt_len); 754 755 if (bytes_to_copy) { 756 int err; 757 758 /* sk_lock is held by caller so no one else can dequeue. 759 * Unlock rx_lock since skb_copy_datagram_iter() may sleep. 760 */ 761 spin_unlock_bh(&vvs->rx_lock); 762 763 err = skb_copy_datagram_iter(skb, 0, 764 &msg->msg_iter, 765 bytes_to_copy); 766 if (err) { 767 /* Copy of message failed. Rest of 768 * fragments will be freed without copy. 769 */ 770 dequeued_len = err; 771 } else { 772 user_buf_len -= bytes_to_copy; 773 } 774 775 spin_lock_bh(&vvs->rx_lock); 776 } 777 778 if (dequeued_len >= 0) 779 dequeued_len += pkt_len; 780 } 781 782 if (le32_to_cpu(hdr->flags) & VIRTIO_VSOCK_SEQ_EOM) { 783 msg_ready = true; 784 vvs->msg_count--; 785 786 if (le32_to_cpu(hdr->flags) & VIRTIO_VSOCK_SEQ_EOR) 787 msg->msg_flags |= MSG_EOR; 788 } 789 790 virtio_transport_dec_rx_pkt(vvs, pkt_len, pkt_len); 791 kfree_skb(skb); 792 } 793 794 spin_unlock_bh(&vvs->rx_lock); 795 796 virtio_transport_send_credit_update(vsk); 797 798 return dequeued_len; 799 } 800 801 ssize_t 802 virtio_transport_stream_dequeue(struct vsock_sock *vsk, 803 struct msghdr *msg, 804 size_t len, int flags) 805 { 806 if (flags & MSG_PEEK) 807 return virtio_transport_stream_do_peek(vsk, msg, len); 808 else 809 return virtio_transport_stream_do_dequeue(vsk, msg, len); 810 } 811 EXPORT_SYMBOL_GPL(virtio_transport_stream_dequeue); 812 813 ssize_t 814 virtio_transport_seqpacket_dequeue(struct vsock_sock *vsk, 815 struct msghdr *msg, 816 int flags) 817 { 818 if (flags & MSG_PEEK) 819 return virtio_transport_seqpacket_do_peek(vsk, msg); 820 else 821 return virtio_transport_seqpacket_do_dequeue(vsk, msg, flags); 822 } 823 EXPORT_SYMBOL_GPL(virtio_transport_seqpacket_dequeue); 824 825 static u32 virtio_transport_tx_buf_size(struct virtio_vsock_sock *vvs) 826 { 827 /* The peer advertises its receive buffer via peer_buf_alloc, but we 828 * cap it to our local buf_alloc so a remote peer cannot force us to 829 * queue more data than our own buffer configuration allows. 830 */ 831 return min(vvs->peer_buf_alloc, vvs->buf_alloc); 832 } 833 834 int 835 virtio_transport_seqpacket_enqueue(struct vsock_sock *vsk, 836 struct msghdr *msg, 837 size_t len) 838 { 839 struct virtio_vsock_sock *vvs = vsk->trans; 840 841 spin_lock_bh(&vvs->tx_lock); 842 843 if (len > virtio_transport_tx_buf_size(vvs)) { 844 spin_unlock_bh(&vvs->tx_lock); 845 return -EMSGSIZE; 846 } 847 848 spin_unlock_bh(&vvs->tx_lock); 849 850 return virtio_transport_stream_enqueue(vsk, msg, len); 851 } 852 EXPORT_SYMBOL_GPL(virtio_transport_seqpacket_enqueue); 853 854 int 855 virtio_transport_dgram_dequeue(struct vsock_sock *vsk, 856 struct msghdr *msg, 857 size_t len, int flags) 858 { 859 return -EOPNOTSUPP; 860 } 861 EXPORT_SYMBOL_GPL(virtio_transport_dgram_dequeue); 862 863 s64 virtio_transport_stream_has_data(struct vsock_sock *vsk) 864 { 865 struct virtio_vsock_sock *vvs = vsk->trans; 866 s64 bytes; 867 868 spin_lock_bh(&vvs->rx_lock); 869 bytes = vvs->rx_bytes; 870 spin_unlock_bh(&vvs->rx_lock); 871 872 return bytes; 873 } 874 EXPORT_SYMBOL_GPL(virtio_transport_stream_has_data); 875 876 u32 virtio_transport_seqpacket_has_data(struct vsock_sock *vsk) 877 { 878 struct virtio_vsock_sock *vvs = vsk->trans; 879 u32 msg_count; 880 881 spin_lock_bh(&vvs->rx_lock); 882 msg_count = vvs->msg_count; 883 spin_unlock_bh(&vvs->rx_lock); 884 885 return msg_count; 886 } 887 EXPORT_SYMBOL_GPL(virtio_transport_seqpacket_has_data); 888 889 static s64 virtio_transport_has_space(struct virtio_vsock_sock *vvs) 890 { 891 s64 bytes; 892 893 /* Use s64 arithmetic so if the peer shrinks peer_buf_alloc while 894 * we have bytes in flight (tx_cnt - peer_fwd_cnt), the subtraction 895 * does not underflow. 896 */ 897 bytes = (s64)virtio_transport_tx_buf_size(vvs) - 898 (vvs->tx_cnt - vvs->peer_fwd_cnt); 899 if (bytes < 0) 900 bytes = 0; 901 902 return bytes; 903 } 904 905 s64 virtio_transport_stream_has_space(struct vsock_sock *vsk) 906 { 907 struct virtio_vsock_sock *vvs = vsk->trans; 908 s64 bytes; 909 910 spin_lock_bh(&vvs->tx_lock); 911 bytes = virtio_transport_has_space(vvs); 912 spin_unlock_bh(&vvs->tx_lock); 913 914 return bytes; 915 } 916 EXPORT_SYMBOL_GPL(virtio_transport_stream_has_space); 917 918 int virtio_transport_do_socket_init(struct vsock_sock *vsk, 919 struct vsock_sock *psk) 920 { 921 struct virtio_vsock_sock *vvs; 922 923 vvs = kzalloc(sizeof(*vvs), GFP_KERNEL); 924 if (!vvs) 925 return -ENOMEM; 926 927 vsk->trans = vvs; 928 vvs->vsk = vsk; 929 if (psk && psk->trans) { 930 struct virtio_vsock_sock *ptrans = psk->trans; 931 932 vvs->peer_buf_alloc = ptrans->peer_buf_alloc; 933 } 934 935 if (vsk->buffer_size > VIRTIO_VSOCK_MAX_BUF_SIZE) 936 vsk->buffer_size = VIRTIO_VSOCK_MAX_BUF_SIZE; 937 938 vvs->buf_alloc = vsk->buffer_size; 939 940 spin_lock_init(&vvs->rx_lock); 941 spin_lock_init(&vvs->tx_lock); 942 skb_queue_head_init(&vvs->rx_queue); 943 944 return 0; 945 } 946 EXPORT_SYMBOL_GPL(virtio_transport_do_socket_init); 947 948 /* sk_lock held by the caller */ 949 void virtio_transport_notify_buffer_size(struct vsock_sock *vsk, u64 *val) 950 { 951 struct virtio_vsock_sock *vvs = vsk->trans; 952 953 if (*val > VIRTIO_VSOCK_MAX_BUF_SIZE) 954 *val = VIRTIO_VSOCK_MAX_BUF_SIZE; 955 956 vvs->buf_alloc = *val; 957 958 virtio_transport_send_credit_update(vsk); 959 } 960 EXPORT_SYMBOL_GPL(virtio_transport_notify_buffer_size); 961 962 int 963 virtio_transport_notify_poll_in(struct vsock_sock *vsk, 964 size_t target, 965 bool *data_ready_now) 966 { 967 *data_ready_now = vsock_stream_has_data(vsk) >= target; 968 969 return 0; 970 } 971 EXPORT_SYMBOL_GPL(virtio_transport_notify_poll_in); 972 973 int 974 virtio_transport_notify_poll_out(struct vsock_sock *vsk, 975 size_t target, 976 bool *space_avail_now) 977 { 978 s64 free_space; 979 980 free_space = vsock_stream_has_space(vsk); 981 if (free_space > 0) 982 *space_avail_now = true; 983 else if (free_space == 0) 984 *space_avail_now = false; 985 986 return 0; 987 } 988 EXPORT_SYMBOL_GPL(virtio_transport_notify_poll_out); 989 990 int virtio_transport_notify_recv_init(struct vsock_sock *vsk, 991 size_t target, struct vsock_transport_recv_notify_data *data) 992 { 993 return 0; 994 } 995 EXPORT_SYMBOL_GPL(virtio_transport_notify_recv_init); 996 997 int virtio_transport_notify_recv_pre_block(struct vsock_sock *vsk, 998 size_t target, struct vsock_transport_recv_notify_data *data) 999 { 1000 return 0; 1001 } 1002 EXPORT_SYMBOL_GPL(virtio_transport_notify_recv_pre_block); 1003 1004 int virtio_transport_notify_recv_pre_dequeue(struct vsock_sock *vsk, 1005 size_t target, struct vsock_transport_recv_notify_data *data) 1006 { 1007 return 0; 1008 } 1009 EXPORT_SYMBOL_GPL(virtio_transport_notify_recv_pre_dequeue); 1010 1011 int virtio_transport_notify_recv_post_dequeue(struct vsock_sock *vsk, 1012 size_t target, ssize_t copied, bool data_read, 1013 struct vsock_transport_recv_notify_data *data) 1014 { 1015 return 0; 1016 } 1017 EXPORT_SYMBOL_GPL(virtio_transport_notify_recv_post_dequeue); 1018 1019 int virtio_transport_notify_send_init(struct vsock_sock *vsk, 1020 struct vsock_transport_send_notify_data *data) 1021 { 1022 return 0; 1023 } 1024 EXPORT_SYMBOL_GPL(virtio_transport_notify_send_init); 1025 1026 int virtio_transport_notify_send_pre_block(struct vsock_sock *vsk, 1027 struct vsock_transport_send_notify_data *data) 1028 { 1029 return 0; 1030 } 1031 EXPORT_SYMBOL_GPL(virtio_transport_notify_send_pre_block); 1032 1033 int virtio_transport_notify_send_pre_enqueue(struct vsock_sock *vsk, 1034 struct vsock_transport_send_notify_data *data) 1035 { 1036 return 0; 1037 } 1038 EXPORT_SYMBOL_GPL(virtio_transport_notify_send_pre_enqueue); 1039 1040 int virtio_transport_notify_send_post_enqueue(struct vsock_sock *vsk, 1041 ssize_t written, struct vsock_transport_send_notify_data *data) 1042 { 1043 return 0; 1044 } 1045 EXPORT_SYMBOL_GPL(virtio_transport_notify_send_post_enqueue); 1046 1047 u64 virtio_transport_stream_rcvhiwat(struct vsock_sock *vsk) 1048 { 1049 return vsk->buffer_size; 1050 } 1051 EXPORT_SYMBOL_GPL(virtio_transport_stream_rcvhiwat); 1052 1053 bool virtio_transport_stream_is_active(struct vsock_sock *vsk) 1054 { 1055 return true; 1056 } 1057 EXPORT_SYMBOL_GPL(virtio_transport_stream_is_active); 1058 1059 int virtio_transport_dgram_bind(struct vsock_sock *vsk, 1060 struct sockaddr_vm *addr) 1061 { 1062 return -EOPNOTSUPP; 1063 } 1064 EXPORT_SYMBOL_GPL(virtio_transport_dgram_bind); 1065 1066 bool virtio_transport_dgram_allow(struct vsock_sock *vsk, u32 cid, u32 port) 1067 { 1068 return false; 1069 } 1070 EXPORT_SYMBOL_GPL(virtio_transport_dgram_allow); 1071 1072 int virtio_transport_connect(struct vsock_sock *vsk) 1073 { 1074 struct virtio_vsock_pkt_info info = { 1075 .op = VIRTIO_VSOCK_OP_REQUEST, 1076 .vsk = vsk, 1077 .net = sock_net(sk_vsock(vsk)), 1078 }; 1079 1080 return virtio_transport_send_pkt_info(vsk, &info); 1081 } 1082 EXPORT_SYMBOL_GPL(virtio_transport_connect); 1083 1084 int virtio_transport_shutdown(struct vsock_sock *vsk, int mode) 1085 { 1086 struct virtio_vsock_pkt_info info = { 1087 .op = VIRTIO_VSOCK_OP_SHUTDOWN, 1088 .flags = (mode & RCV_SHUTDOWN ? 1089 VIRTIO_VSOCK_SHUTDOWN_RCV : 0) | 1090 (mode & SEND_SHUTDOWN ? 1091 VIRTIO_VSOCK_SHUTDOWN_SEND : 0), 1092 .vsk = vsk, 1093 .net = sock_net(sk_vsock(vsk)), 1094 }; 1095 1096 return virtio_transport_send_pkt_info(vsk, &info); 1097 } 1098 EXPORT_SYMBOL_GPL(virtio_transport_shutdown); 1099 1100 int 1101 virtio_transport_dgram_enqueue(struct vsock_sock *vsk, 1102 struct sockaddr_vm *remote_addr, 1103 struct msghdr *msg, 1104 size_t dgram_len) 1105 { 1106 return -EOPNOTSUPP; 1107 } 1108 EXPORT_SYMBOL_GPL(virtio_transport_dgram_enqueue); 1109 1110 ssize_t 1111 virtio_transport_stream_enqueue(struct vsock_sock *vsk, 1112 struct msghdr *msg, 1113 size_t len) 1114 { 1115 struct virtio_vsock_pkt_info info = { 1116 .op = VIRTIO_VSOCK_OP_RW, 1117 .msg = msg, 1118 .pkt_len = len, 1119 .vsk = vsk, 1120 .net = sock_net(sk_vsock(vsk)), 1121 }; 1122 1123 return virtio_transport_send_pkt_info(vsk, &info); 1124 } 1125 EXPORT_SYMBOL_GPL(virtio_transport_stream_enqueue); 1126 1127 void virtio_transport_destruct(struct vsock_sock *vsk) 1128 { 1129 struct virtio_vsock_sock *vvs = vsk->trans; 1130 1131 virtio_transport_cancel_close_work(vsk, true); 1132 1133 kfree(vvs); 1134 vsk->trans = NULL; 1135 } 1136 EXPORT_SYMBOL_GPL(virtio_transport_destruct); 1137 1138 ssize_t virtio_transport_unsent_bytes(struct vsock_sock *vsk) 1139 { 1140 struct virtio_vsock_sock *vvs = vsk->trans; 1141 size_t ret; 1142 1143 spin_lock_bh(&vvs->tx_lock); 1144 ret = vvs->bytes_unsent; 1145 spin_unlock_bh(&vvs->tx_lock); 1146 1147 return ret; 1148 } 1149 EXPORT_SYMBOL_GPL(virtio_transport_unsent_bytes); 1150 1151 static int virtio_transport_reset(struct vsock_sock *vsk, 1152 struct sk_buff *skb) 1153 { 1154 struct virtio_vsock_pkt_info info = { 1155 .op = VIRTIO_VSOCK_OP_RST, 1156 .reply = !!skb, 1157 .vsk = vsk, 1158 .net = sock_net(sk_vsock(vsk)), 1159 }; 1160 1161 /* Send RST only if the original pkt is not a RST pkt */ 1162 if (skb && le16_to_cpu(virtio_vsock_hdr(skb)->op) == VIRTIO_VSOCK_OP_RST) 1163 return 0; 1164 1165 return virtio_transport_send_pkt_info(vsk, &info); 1166 } 1167 1168 /* Normally packets are associated with a socket. There may be no socket if an 1169 * attempt was made to connect to a socket that does not exist. 1170 * 1171 * net refers to the namespace of whoever sent the invalid message. For 1172 * loopback, this is the namespace of the socket. For vhost, this is the 1173 * namespace of the VM (i.e., vhost_vsock). 1174 */ 1175 static int virtio_transport_reset_no_sock(const struct virtio_transport *t, 1176 struct sk_buff *skb, struct net *net) 1177 { 1178 struct virtio_vsock_hdr *hdr = virtio_vsock_hdr(skb); 1179 struct virtio_vsock_pkt_info info = { 1180 .op = VIRTIO_VSOCK_OP_RST, 1181 .type = le16_to_cpu(hdr->type), 1182 .reply = true, 1183 1184 /* Set sk owner to socket we are replying to (may be NULL for 1185 * non-loopback). This keeps a reference to the sock and 1186 * sock_net(sk) until the reply skb is freed. 1187 */ 1188 .vsk = vsock_sk(skb->sk), 1189 1190 /* net is not defined here because we pass it directly to 1191 * t->send_pkt(), instead of relying on 1192 * virtio_transport_send_pkt_info() to pass it. It is not needed 1193 * by virtio_transport_alloc_skb(). 1194 */ 1195 }; 1196 struct sk_buff *reply; 1197 1198 /* Send RST only if the original pkt is not a RST pkt */ 1199 if (le16_to_cpu(hdr->op) == VIRTIO_VSOCK_OP_RST) 1200 return 0; 1201 1202 if (!t) 1203 return -ENOTCONN; 1204 1205 reply = virtio_transport_alloc_skb(&info, 0, false, 1206 le64_to_cpu(hdr->dst_cid), 1207 le32_to_cpu(hdr->dst_port), 1208 le64_to_cpu(hdr->src_cid), 1209 le32_to_cpu(hdr->src_port)); 1210 if (!reply) 1211 return -ENOMEM; 1212 1213 return t->send_pkt(reply, net); 1214 } 1215 1216 /* This function should be called with sk_lock held and SOCK_DONE set */ 1217 static void virtio_transport_remove_sock(struct vsock_sock *vsk) 1218 { 1219 struct virtio_vsock_sock *vvs = vsk->trans; 1220 1221 /* We don't need to take rx_lock, as the socket is closing and we are 1222 * removing it. 1223 */ 1224 __skb_queue_purge(&vvs->rx_queue); 1225 vsock_remove_sock(vsk); 1226 } 1227 1228 static void virtio_transport_cancel_close_work(struct vsock_sock *vsk, 1229 bool cancel_timeout) 1230 { 1231 struct sock *sk = sk_vsock(vsk); 1232 1233 if (vsk->close_work_scheduled && 1234 (!cancel_timeout || cancel_delayed_work(&vsk->close_work))) { 1235 vsk->close_work_scheduled = false; 1236 1237 virtio_transport_remove_sock(vsk); 1238 1239 /* Release refcnt obtained when we scheduled the timeout */ 1240 sock_put(sk); 1241 } 1242 } 1243 1244 static void virtio_transport_do_close(struct vsock_sock *vsk, 1245 bool cancel_timeout) 1246 { 1247 struct sock *sk = sk_vsock(vsk); 1248 1249 sock_set_flag(sk, SOCK_DONE); 1250 vsk->peer_shutdown = SHUTDOWN_MASK; 1251 if (vsock_stream_has_data(vsk) <= 0) 1252 sk->sk_state = TCP_CLOSING; 1253 sk->sk_state_change(sk); 1254 1255 virtio_transport_cancel_close_work(vsk, cancel_timeout); 1256 } 1257 1258 static void virtio_transport_close_timeout(struct work_struct *work) 1259 { 1260 struct vsock_sock *vsk = 1261 container_of(work, struct vsock_sock, close_work.work); 1262 struct sock *sk = sk_vsock(vsk); 1263 1264 sock_hold(sk); 1265 lock_sock(sk); 1266 1267 if (!sock_flag(sk, SOCK_DONE)) { 1268 (void)virtio_transport_reset(vsk, NULL); 1269 1270 virtio_transport_do_close(vsk, false); 1271 } 1272 1273 vsk->close_work_scheduled = false; 1274 1275 release_sock(sk); 1276 sock_put(sk); 1277 } 1278 1279 /* User context, vsk->sk is locked */ 1280 static bool virtio_transport_close(struct vsock_sock *vsk) 1281 { 1282 struct sock *sk = &vsk->sk; 1283 1284 if (!(sk->sk_state == TCP_ESTABLISHED || 1285 sk->sk_state == TCP_CLOSING)) 1286 return true; 1287 1288 /* Already received SHUTDOWN from peer, reply with RST */ 1289 if ((vsk->peer_shutdown & SHUTDOWN_MASK) == SHUTDOWN_MASK) { 1290 (void)virtio_transport_reset(vsk, NULL); 1291 return true; 1292 } 1293 1294 if ((sk->sk_shutdown & SHUTDOWN_MASK) != SHUTDOWN_MASK) 1295 (void)virtio_transport_shutdown(vsk, SHUTDOWN_MASK); 1296 1297 if (!(current->flags & PF_EXITING)) 1298 vsock_linger(sk); 1299 1300 if (sock_flag(sk, SOCK_DONE)) { 1301 return true; 1302 } 1303 1304 sock_hold(sk); 1305 INIT_DELAYED_WORK(&vsk->close_work, 1306 virtio_transport_close_timeout); 1307 vsk->close_work_scheduled = true; 1308 schedule_delayed_work(&vsk->close_work, VSOCK_CLOSE_TIMEOUT); 1309 return false; 1310 } 1311 1312 void virtio_transport_release(struct vsock_sock *vsk) 1313 { 1314 struct sock *sk = &vsk->sk; 1315 bool remove_sock = true; 1316 1317 if (sk->sk_type == SOCK_STREAM || sk->sk_type == SOCK_SEQPACKET) 1318 remove_sock = virtio_transport_close(vsk); 1319 1320 if (remove_sock) { 1321 sock_set_flag(sk, SOCK_DONE); 1322 virtio_transport_remove_sock(vsk); 1323 } 1324 } 1325 EXPORT_SYMBOL_GPL(virtio_transport_release); 1326 1327 static int 1328 virtio_transport_recv_connecting(struct sock *sk, 1329 struct sk_buff *skb) 1330 { 1331 struct virtio_vsock_hdr *hdr = virtio_vsock_hdr(skb); 1332 struct vsock_sock *vsk = vsock_sk(sk); 1333 int skerr; 1334 int err; 1335 1336 switch (le16_to_cpu(hdr->op)) { 1337 case VIRTIO_VSOCK_OP_RESPONSE: 1338 sk->sk_state = TCP_ESTABLISHED; 1339 sk->sk_socket->state = SS_CONNECTED; 1340 vsock_insert_connected(vsk); 1341 sk->sk_state_change(sk); 1342 break; 1343 case VIRTIO_VSOCK_OP_INVALID: 1344 break; 1345 case VIRTIO_VSOCK_OP_RST: 1346 skerr = ECONNRESET; 1347 err = 0; 1348 goto destroy; 1349 default: 1350 skerr = EPROTO; 1351 err = -EINVAL; 1352 goto destroy; 1353 } 1354 return 0; 1355 1356 destroy: 1357 virtio_transport_reset(vsk, skb); 1358 sk->sk_state = TCP_CLOSE; 1359 sk->sk_err = skerr; 1360 sk_error_report(sk); 1361 return err; 1362 } 1363 1364 static void 1365 virtio_transport_recv_enqueue(struct vsock_sock *vsk, 1366 struct sk_buff *skb) 1367 { 1368 struct virtio_vsock_sock *vvs = vsk->trans; 1369 bool can_enqueue, free_pkt = false; 1370 struct virtio_vsock_hdr *hdr; 1371 u32 len; 1372 1373 hdr = virtio_vsock_hdr(skb); 1374 len = le32_to_cpu(hdr->len); 1375 1376 spin_lock_bh(&vvs->rx_lock); 1377 1378 can_enqueue = virtio_transport_inc_rx_pkt(vvs, len); 1379 if (!can_enqueue) { 1380 free_pkt = true; 1381 goto out; 1382 } 1383 1384 if (le32_to_cpu(hdr->flags) & VIRTIO_VSOCK_SEQ_EOM) 1385 vvs->msg_count++; 1386 1387 /* Try to copy small packets into the buffer of last packet queued, 1388 * to avoid wasting memory queueing the entire buffer with a small 1389 * payload. Skip non-linear (e.g. zerocopy) skbs; these carry payload 1390 * in skb_shinfo. 1391 */ 1392 if (len <= GOOD_COPY_LEN && !skb_queue_empty(&vvs->rx_queue) && 1393 !skb_is_nonlinear(skb)) { 1394 struct virtio_vsock_hdr *last_hdr; 1395 struct sk_buff *last_skb; 1396 1397 last_skb = skb_peek_tail(&vvs->rx_queue); 1398 last_hdr = virtio_vsock_hdr(last_skb); 1399 1400 /* If there is space in the last packet queued, we copy the 1401 * new packet in its buffer. We avoid this if the last packet 1402 * queued has VIRTIO_VSOCK_SEQ_EOM set, because this is 1403 * delimiter of SEQPACKET message, so 'pkt' is the first packet 1404 * of a new message. 1405 */ 1406 if (skb->len < skb_tailroom(last_skb) && 1407 !(le32_to_cpu(last_hdr->flags) & VIRTIO_VSOCK_SEQ_EOM)) { 1408 memcpy(skb_put(last_skb, skb->len), skb->data, skb->len); 1409 free_pkt = true; 1410 last_hdr->flags |= hdr->flags; 1411 le32_add_cpu(&last_hdr->len, len); 1412 goto out; 1413 } 1414 } 1415 1416 __skb_queue_tail(&vvs->rx_queue, skb); 1417 1418 out: 1419 spin_unlock_bh(&vvs->rx_lock); 1420 if (free_pkt) 1421 kfree_skb(skb); 1422 } 1423 1424 static int 1425 virtio_transport_recv_connected(struct sock *sk, 1426 struct sk_buff *skb) 1427 { 1428 struct virtio_vsock_hdr *hdr = virtio_vsock_hdr(skb); 1429 struct vsock_sock *vsk = vsock_sk(sk); 1430 int err = 0; 1431 1432 switch (le16_to_cpu(hdr->op)) { 1433 case VIRTIO_VSOCK_OP_RW: 1434 virtio_transport_recv_enqueue(vsk, skb); 1435 vsock_data_ready(sk); 1436 return err; 1437 case VIRTIO_VSOCK_OP_CREDIT_REQUEST: 1438 virtio_transport_send_credit_update(vsk); 1439 break; 1440 case VIRTIO_VSOCK_OP_CREDIT_UPDATE: 1441 sk->sk_write_space(sk); 1442 break; 1443 case VIRTIO_VSOCK_OP_SHUTDOWN: 1444 if (le32_to_cpu(hdr->flags) & VIRTIO_VSOCK_SHUTDOWN_RCV) 1445 vsk->peer_shutdown |= RCV_SHUTDOWN; 1446 if (le32_to_cpu(hdr->flags) & VIRTIO_VSOCK_SHUTDOWN_SEND) 1447 vsk->peer_shutdown |= SEND_SHUTDOWN; 1448 if (vsk->peer_shutdown == SHUTDOWN_MASK) { 1449 if (vsock_stream_has_data(vsk) <= 0 && !sock_flag(sk, SOCK_DONE)) { 1450 (void)virtio_transport_reset(vsk, NULL); 1451 virtio_transport_do_close(vsk, true); 1452 } 1453 /* Remove this socket anyway because the remote peer sent 1454 * the shutdown. This way a new connection will succeed 1455 * if the remote peer uses the same source port, 1456 * even if the old socket is still unreleased, but now disconnected. 1457 */ 1458 vsock_remove_sock(vsk); 1459 } 1460 if (le32_to_cpu(virtio_vsock_hdr(skb)->flags)) 1461 sk->sk_state_change(sk); 1462 break; 1463 case VIRTIO_VSOCK_OP_RST: 1464 virtio_transport_do_close(vsk, true); 1465 break; 1466 default: 1467 err = -EINVAL; 1468 break; 1469 } 1470 1471 kfree_skb(skb); 1472 return err; 1473 } 1474 1475 static void 1476 virtio_transport_recv_disconnecting(struct sock *sk, 1477 struct sk_buff *skb) 1478 { 1479 struct virtio_vsock_hdr *hdr = virtio_vsock_hdr(skb); 1480 struct vsock_sock *vsk = vsock_sk(sk); 1481 1482 if (le16_to_cpu(hdr->op) == VIRTIO_VSOCK_OP_RST) 1483 virtio_transport_do_close(vsk, true); 1484 } 1485 1486 static int 1487 virtio_transport_send_response(struct vsock_sock *vsk, 1488 struct sk_buff *skb) 1489 { 1490 struct virtio_vsock_hdr *hdr = virtio_vsock_hdr(skb); 1491 struct virtio_vsock_pkt_info info = { 1492 .op = VIRTIO_VSOCK_OP_RESPONSE, 1493 .remote_cid = le64_to_cpu(hdr->src_cid), 1494 .remote_port = le32_to_cpu(hdr->src_port), 1495 .reply = true, 1496 .vsk = vsk, 1497 .net = sock_net(sk_vsock(vsk)), 1498 }; 1499 1500 return virtio_transport_send_pkt_info(vsk, &info); 1501 } 1502 1503 static bool virtio_transport_space_update(struct sock *sk, 1504 struct sk_buff *skb) 1505 { 1506 struct virtio_vsock_hdr *hdr = virtio_vsock_hdr(skb); 1507 struct vsock_sock *vsk = vsock_sk(sk); 1508 struct virtio_vsock_sock *vvs = vsk->trans; 1509 bool space_available; 1510 1511 /* Listener sockets are not associated with any transport, so we are 1512 * not able to take the state to see if there is space available in the 1513 * remote peer, but since they are only used to receive requests, we 1514 * can assume that there is always space available in the other peer. 1515 */ 1516 if (!vvs) 1517 return true; 1518 1519 /* buf_alloc and fwd_cnt is always included in the hdr */ 1520 spin_lock_bh(&vvs->tx_lock); 1521 vvs->peer_buf_alloc = le32_to_cpu(hdr->buf_alloc); 1522 vvs->peer_fwd_cnt = le32_to_cpu(hdr->fwd_cnt); 1523 space_available = virtio_transport_has_space(vvs); 1524 spin_unlock_bh(&vvs->tx_lock); 1525 return space_available; 1526 } 1527 1528 /* Handle server socket */ 1529 static int 1530 virtio_transport_recv_listen(struct sock *sk, struct sk_buff *skb, 1531 struct virtio_transport *t) 1532 { 1533 struct virtio_vsock_hdr *hdr = virtio_vsock_hdr(skb); 1534 struct vsock_sock *vsk = vsock_sk(sk); 1535 struct vsock_sock *vchild; 1536 struct sock *child; 1537 int ret; 1538 1539 if (le16_to_cpu(hdr->op) != VIRTIO_VSOCK_OP_REQUEST) { 1540 virtio_transport_reset_no_sock(t, skb, sock_net(sk)); 1541 return -EINVAL; 1542 } 1543 1544 if (sk_acceptq_is_full(sk)) { 1545 virtio_transport_reset_no_sock(t, skb, sock_net(sk)); 1546 return -ENOMEM; 1547 } 1548 1549 /* __vsock_release() might have already flushed accept_queue. 1550 * Subsequent enqueues would lead to a memory leak. 1551 */ 1552 if (sk->sk_shutdown == SHUTDOWN_MASK) { 1553 virtio_transport_reset_no_sock(t, skb, sock_net(sk)); 1554 return -ESHUTDOWN; 1555 } 1556 1557 child = vsock_create_connected(sk); 1558 if (!child) { 1559 virtio_transport_reset_no_sock(t, skb, sock_net(sk)); 1560 return -ENOMEM; 1561 } 1562 1563 sk_acceptq_added(sk); 1564 1565 lock_sock_nested(child, SINGLE_DEPTH_NESTING); 1566 1567 child->sk_state = TCP_ESTABLISHED; 1568 1569 vchild = vsock_sk(child); 1570 vsock_addr_init(&vchild->local_addr, le64_to_cpu(hdr->dst_cid), 1571 le32_to_cpu(hdr->dst_port)); 1572 vsock_addr_init(&vchild->remote_addr, le64_to_cpu(hdr->src_cid), 1573 le32_to_cpu(hdr->src_port)); 1574 1575 ret = vsock_assign_transport(vchild, vsk); 1576 /* Transport assigned (looking at remote_addr) must be the same 1577 * where we received the request. 1578 */ 1579 if (ret || vchild->transport != &t->transport) { 1580 release_sock(child); 1581 virtio_transport_reset_no_sock(t, skb, sock_net(sk)); 1582 sock_put(child); 1583 return ret; 1584 } 1585 1586 if (virtio_transport_space_update(child, skb)) 1587 child->sk_write_space(child); 1588 1589 vsock_insert_connected(vchild); 1590 vsock_enqueue_accept(sk, child); 1591 virtio_transport_send_response(vchild, skb); 1592 1593 release_sock(child); 1594 1595 sk->sk_data_ready(sk); 1596 return 0; 1597 } 1598 1599 static bool virtio_transport_valid_type(u16 type) 1600 { 1601 return (type == VIRTIO_VSOCK_TYPE_STREAM) || 1602 (type == VIRTIO_VSOCK_TYPE_SEQPACKET); 1603 } 1604 1605 /* We are under the virtio-vsock's vsock->rx_lock or vhost-vsock's vq->mutex 1606 * lock. 1607 */ 1608 void virtio_transport_recv_pkt(struct virtio_transport *t, 1609 struct sk_buff *skb, struct net *net) 1610 { 1611 struct virtio_vsock_hdr *hdr = virtio_vsock_hdr(skb); 1612 struct sockaddr_vm src, dst; 1613 struct vsock_sock *vsk; 1614 struct sock *sk; 1615 bool space_available; 1616 1617 vsock_addr_init(&src, le64_to_cpu(hdr->src_cid), 1618 le32_to_cpu(hdr->src_port)); 1619 vsock_addr_init(&dst, le64_to_cpu(hdr->dst_cid), 1620 le32_to_cpu(hdr->dst_port)); 1621 1622 trace_virtio_transport_recv_pkt(src.svm_cid, src.svm_port, 1623 dst.svm_cid, dst.svm_port, 1624 le32_to_cpu(hdr->len), 1625 le16_to_cpu(hdr->type), 1626 le16_to_cpu(hdr->op), 1627 le32_to_cpu(hdr->flags), 1628 le32_to_cpu(hdr->buf_alloc), 1629 le32_to_cpu(hdr->fwd_cnt)); 1630 1631 if (!virtio_transport_valid_type(le16_to_cpu(hdr->type))) { 1632 (void)virtio_transport_reset_no_sock(t, skb, net); 1633 goto free_pkt; 1634 } 1635 1636 /* The socket must be in connected or bound table 1637 * otherwise send reset back 1638 */ 1639 sk = vsock_find_connected_socket_net(&src, &dst, net); 1640 if (!sk) { 1641 sk = vsock_find_bound_socket_net(&dst, net); 1642 if (!sk) { 1643 (void)virtio_transport_reset_no_sock(t, skb, net); 1644 goto free_pkt; 1645 } 1646 } 1647 1648 if (virtio_transport_get_type(sk) != le16_to_cpu(hdr->type)) { 1649 (void)virtio_transport_reset_no_sock(t, skb, net); 1650 sock_put(sk); 1651 goto free_pkt; 1652 } 1653 1654 if (!skb_set_owner_sk_safe(skb, sk)) { 1655 WARN_ONCE(1, "receiving vsock socket has sk_refcnt == 0\n"); 1656 goto free_pkt; 1657 } 1658 1659 vsk = vsock_sk(sk); 1660 1661 lock_sock(sk); 1662 1663 /* Check if sk has been closed or assigned to another transport before 1664 * lock_sock (note: listener sockets are not assigned to any transport) 1665 */ 1666 if (sock_flag(sk, SOCK_DONE) || 1667 (sk->sk_state != TCP_LISTEN && vsk->transport != &t->transport)) { 1668 (void)virtio_transport_reset_no_sock(t, skb, net); 1669 release_sock(sk); 1670 sock_put(sk); 1671 goto free_pkt; 1672 } 1673 1674 space_available = virtio_transport_space_update(sk, skb); 1675 1676 /* Update CID in case it has changed after a transport reset event */ 1677 if (vsk->local_addr.svm_cid != VMADDR_CID_ANY) 1678 vsk->local_addr.svm_cid = dst.svm_cid; 1679 1680 if (space_available) 1681 sk->sk_write_space(sk); 1682 1683 switch (sk->sk_state) { 1684 case TCP_LISTEN: 1685 virtio_transport_recv_listen(sk, skb, t); 1686 kfree_skb(skb); 1687 break; 1688 case TCP_SYN_SENT: 1689 virtio_transport_recv_connecting(sk, skb); 1690 kfree_skb(skb); 1691 break; 1692 case TCP_ESTABLISHED: 1693 virtio_transport_recv_connected(sk, skb); 1694 break; 1695 case TCP_CLOSING: 1696 virtio_transport_recv_disconnecting(sk, skb); 1697 kfree_skb(skb); 1698 break; 1699 default: 1700 (void)virtio_transport_reset_no_sock(t, skb, net); 1701 kfree_skb(skb); 1702 break; 1703 } 1704 1705 release_sock(sk); 1706 1707 /* Release refcnt obtained when we fetched this socket out of the 1708 * bound or connected list. 1709 */ 1710 sock_put(sk); 1711 return; 1712 1713 free_pkt: 1714 kfree_skb(skb); 1715 } 1716 EXPORT_SYMBOL_GPL(virtio_transport_recv_pkt); 1717 1718 /* Remove skbs found in a queue that have a vsk that matches. 1719 * 1720 * Each skb is freed. 1721 * 1722 * Returns the count of skbs that were reply packets. 1723 */ 1724 int virtio_transport_purge_skbs(void *vsk, struct sk_buff_head *queue) 1725 { 1726 struct sk_buff_head freeme; 1727 struct sk_buff *skb, *tmp; 1728 int cnt = 0; 1729 1730 skb_queue_head_init(&freeme); 1731 1732 spin_lock_bh(&queue->lock); 1733 skb_queue_walk_safe(queue, skb, tmp) { 1734 if (vsock_sk(skb->sk) != vsk) 1735 continue; 1736 1737 __skb_unlink(skb, queue); 1738 __skb_queue_tail(&freeme, skb); 1739 1740 if (virtio_vsock_skb_reply(skb)) 1741 cnt++; 1742 } 1743 spin_unlock_bh(&queue->lock); 1744 1745 __skb_queue_purge(&freeme); 1746 1747 return cnt; 1748 } 1749 EXPORT_SYMBOL_GPL(virtio_transport_purge_skbs); 1750 1751 int virtio_transport_read_skb(struct vsock_sock *vsk, skb_read_actor_t recv_actor) 1752 { 1753 struct virtio_vsock_sock *vvs = vsk->trans; 1754 struct sock *sk = sk_vsock(vsk); 1755 struct virtio_vsock_hdr *hdr; 1756 struct sk_buff *skb; 1757 u32 pkt_len; 1758 int off = 0; 1759 int err; 1760 1761 spin_lock_bh(&vvs->rx_lock); 1762 /* Use __skb_recv_datagram() for race-free handling of the receive. It 1763 * works for types other than dgrams. 1764 */ 1765 skb = __skb_recv_datagram(sk, &vvs->rx_queue, MSG_DONTWAIT, &off, &err); 1766 if (!skb) { 1767 spin_unlock_bh(&vvs->rx_lock); 1768 return err; 1769 } 1770 1771 hdr = virtio_vsock_hdr(skb); 1772 if (le32_to_cpu(hdr->flags) & VIRTIO_VSOCK_SEQ_EOM) 1773 vvs->msg_count--; 1774 1775 pkt_len = le32_to_cpu(hdr->len); 1776 virtio_transport_dec_rx_pkt(vvs, pkt_len, pkt_len); 1777 spin_unlock_bh(&vvs->rx_lock); 1778 1779 virtio_transport_send_credit_update(vsk); 1780 1781 return recv_actor(sk, skb); 1782 } 1783 EXPORT_SYMBOL_GPL(virtio_transport_read_skb); 1784 1785 int virtio_transport_notify_set_rcvlowat(struct vsock_sock *vsk, int val) 1786 { 1787 struct virtio_vsock_sock *vvs = vsk->trans; 1788 bool send_update; 1789 1790 spin_lock_bh(&vvs->rx_lock); 1791 1792 /* If number of available bytes is less than new SO_RCVLOWAT value, 1793 * kick sender to send more data, because sender may sleep in its 1794 * 'send()' syscall waiting for enough space at our side. Also 1795 * don't send credit update when peer already knows actual value - 1796 * such transmission will be useless. 1797 */ 1798 send_update = (vvs->rx_bytes < val) && 1799 (vvs->fwd_cnt != vvs->last_fwd_cnt); 1800 1801 spin_unlock_bh(&vvs->rx_lock); 1802 1803 if (send_update) { 1804 int err; 1805 1806 err = virtio_transport_send_credit_update(vsk); 1807 if (err < 0) 1808 return err; 1809 } 1810 1811 return 0; 1812 } 1813 EXPORT_SYMBOL_GPL(virtio_transport_notify_set_rcvlowat); 1814 1815 MODULE_LICENSE("GPL v2"); 1816 MODULE_AUTHOR("Asias He"); 1817 MODULE_DESCRIPTION("common code for virtio vsock"); 1818