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 if (t_ops->can_msgzerocopy) { 64 int pages_to_send = iov_iter_npages(iov_iter, MAX_SKB_FRAGS); 65 66 /* +1 is for packet header. */ 67 return t_ops->can_msgzerocopy(pages_to_send + 1); 68 } 69 70 return true; 71 } 72 73 static int virtio_transport_init_zcopy_skb(struct vsock_sock *vsk, 74 struct sk_buff *skb, 75 struct msghdr *msg, 76 size_t pkt_len, 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 ubuf_info_msgzc *uarg_zc; 86 87 uarg = msg_zerocopy_realloc(sk_vsock(vsk), 88 pkt_len, NULL, false); 89 if (!uarg) 90 return -1; 91 92 uarg_zc = uarg_to_msgzc(uarg); 93 uarg_zc->zerocopy = zerocopy ? 1 : 0; 94 } 95 96 skb_zcopy_init(skb, uarg); 97 98 return 0; 99 } 100 101 static int virtio_transport_fill_skb(struct sk_buff *skb, 102 struct virtio_vsock_pkt_info *info, 103 size_t len, 104 bool zcopy) 105 { 106 struct msghdr *msg = info->msg; 107 108 if (zcopy) 109 return __zerocopy_sg_from_iter(msg, NULL, skb, 110 &msg->msg_iter, len, NULL); 111 112 virtio_vsock_skb_put(skb, len); 113 return skb_copy_datagram_from_iter_full(skb, 0, &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 * Pass pkt_len because msg iter is already consumed 402 * by virtio_transport_fill_skb(), so iter->count 403 * can not be used for RLIMIT_MEMLOCK pinned-pages 404 * accounting done by msg_zerocopy_realloc(). 405 */ 406 if (info->msg && info->msg->msg_flags & MSG_ZEROCOPY && 407 skb_len == rest_len && info->op == VIRTIO_VSOCK_OP_RW) { 408 if (virtio_transport_init_zcopy_skb(vsk, skb, 409 info->msg, 410 pkt_len, 411 can_zcopy)) { 412 kfree_skb(skb); 413 ret = -ENOMEM; 414 break; 415 } 416 } 417 418 virtio_transport_inc_tx_pkt(vvs, skb); 419 420 ret = t_ops->send_pkt(skb, info->net); 421 if (ret < 0) 422 break; 423 424 /* Both virtio and vhost 'send_pkt()' returns 'skb_len', 425 * but for reliability use 'ret' instead of 'skb_len'. 426 * Also if partial send happens (e.g. 'ret' != 'skb_len') 427 * somehow, we break this loop, but account such returned 428 * value in 'virtio_transport_put_credit()'. 429 */ 430 rest_len -= ret; 431 432 if (WARN_ONCE(ret != skb_len, 433 "'send_pkt()' returns %i, but %zu expected\n", 434 ret, skb_len)) 435 break; 436 } while (rest_len); 437 438 virtio_transport_put_credit(vvs, rest_len); 439 440 /* Return number of bytes, if any data has been sent. */ 441 if (rest_len != pkt_len) 442 ret = pkt_len - rest_len; 443 444 return ret; 445 } 446 447 static bool virtio_transport_inc_rx_pkt(struct virtio_vsock_sock *vvs, 448 u32 len) 449 { 450 if (vvs->buf_used + len > vvs->buf_alloc) 451 return false; 452 453 vvs->rx_bytes += len; 454 vvs->buf_used += len; 455 return true; 456 } 457 458 static void virtio_transport_dec_rx_pkt(struct virtio_vsock_sock *vvs, 459 u32 bytes_read, u32 bytes_dequeued) 460 { 461 vvs->rx_bytes -= bytes_read; 462 vvs->buf_used -= bytes_dequeued; 463 vvs->fwd_cnt += bytes_dequeued; 464 } 465 466 void virtio_transport_inc_tx_pkt(struct virtio_vsock_sock *vvs, struct sk_buff *skb) 467 { 468 struct virtio_vsock_hdr *hdr = virtio_vsock_hdr(skb); 469 470 spin_lock_bh(&vvs->rx_lock); 471 vvs->last_fwd_cnt = vvs->fwd_cnt; 472 hdr->fwd_cnt = cpu_to_le32(vvs->fwd_cnt); 473 hdr->buf_alloc = cpu_to_le32(vvs->buf_alloc); 474 spin_unlock_bh(&vvs->rx_lock); 475 } 476 EXPORT_SYMBOL_GPL(virtio_transport_inc_tx_pkt); 477 478 void virtio_transport_consume_skb_sent(struct sk_buff *skb, bool consume) 479 { 480 struct sock *s = skb->sk; 481 482 if (s && skb->len) { 483 struct vsock_sock *vs = vsock_sk(s); 484 struct virtio_vsock_sock *vvs; 485 486 vvs = vs->trans; 487 488 spin_lock_bh(&vvs->tx_lock); 489 vvs->bytes_unsent -= skb->len; 490 spin_unlock_bh(&vvs->tx_lock); 491 } 492 493 if (consume) 494 consume_skb(skb); 495 } 496 EXPORT_SYMBOL_GPL(virtio_transport_consume_skb_sent); 497 498 u32 virtio_transport_get_credit(struct virtio_vsock_sock *vvs, u32 credit) 499 { 500 u32 ret; 501 502 if (!credit) 503 return 0; 504 505 spin_lock_bh(&vvs->tx_lock); 506 ret = min_t(u32, credit, virtio_transport_has_space(vvs)); 507 vvs->tx_cnt += ret; 508 vvs->bytes_unsent += ret; 509 spin_unlock_bh(&vvs->tx_lock); 510 511 return ret; 512 } 513 EXPORT_SYMBOL_GPL(virtio_transport_get_credit); 514 515 void virtio_transport_put_credit(struct virtio_vsock_sock *vvs, u32 credit) 516 { 517 if (!credit) 518 return; 519 520 spin_lock_bh(&vvs->tx_lock); 521 vvs->tx_cnt -= credit; 522 vvs->bytes_unsent -= credit; 523 spin_unlock_bh(&vvs->tx_lock); 524 } 525 EXPORT_SYMBOL_GPL(virtio_transport_put_credit); 526 527 static int virtio_transport_send_credit_update(struct vsock_sock *vsk) 528 { 529 struct virtio_vsock_pkt_info info = { 530 .op = VIRTIO_VSOCK_OP_CREDIT_UPDATE, 531 .vsk = vsk, 532 .net = sock_net(sk_vsock(vsk)), 533 }; 534 535 return virtio_transport_send_pkt_info(vsk, &info); 536 } 537 538 static ssize_t 539 virtio_transport_stream_do_peek(struct vsock_sock *vsk, 540 struct msghdr *msg, 541 size_t len) 542 { 543 struct virtio_vsock_sock *vvs = vsk->trans; 544 struct sk_buff *skb; 545 size_t total = 0; 546 int err; 547 548 spin_lock_bh(&vvs->rx_lock); 549 550 skb_queue_walk(&vvs->rx_queue, skb) { 551 size_t bytes; 552 553 bytes = min_t(size_t, len - total, 554 skb->len - VIRTIO_VSOCK_SKB_CB(skb)->offset); 555 556 spin_unlock_bh(&vvs->rx_lock); 557 558 /* sk_lock is held by caller so no one else can dequeue. 559 * Unlock rx_lock since skb_copy_datagram_iter() may sleep. 560 */ 561 err = skb_copy_datagram_iter(skb, VIRTIO_VSOCK_SKB_CB(skb)->offset, 562 &msg->msg_iter, bytes); 563 if (err) 564 goto out; 565 566 total += bytes; 567 568 spin_lock_bh(&vvs->rx_lock); 569 570 if (total == len) 571 break; 572 } 573 574 spin_unlock_bh(&vvs->rx_lock); 575 576 return total; 577 578 out: 579 if (total) 580 err = total; 581 return err; 582 } 583 584 static ssize_t 585 virtio_transport_stream_do_dequeue(struct vsock_sock *vsk, 586 struct msghdr *msg, 587 size_t len) 588 { 589 struct virtio_vsock_sock *vvs = vsk->trans; 590 struct sk_buff *skb; 591 u32 fwd_cnt_delta; 592 bool low_rx_bytes; 593 int err = -EFAULT; 594 size_t total = 0; 595 u32 free_space; 596 597 spin_lock_bh(&vvs->rx_lock); 598 599 if (WARN_ONCE(skb_queue_empty(&vvs->rx_queue) && vvs->rx_bytes, 600 "rx_queue is empty, but rx_bytes is non-zero\n")) { 601 spin_unlock_bh(&vvs->rx_lock); 602 return err; 603 } 604 605 while (total < len && !skb_queue_empty(&vvs->rx_queue)) { 606 size_t bytes, dequeued = 0; 607 608 skb = skb_peek(&vvs->rx_queue); 609 610 bytes = min_t(size_t, len - total, 611 skb->len - VIRTIO_VSOCK_SKB_CB(skb)->offset); 612 613 /* sk_lock is held by caller so no one else can dequeue. 614 * Unlock rx_lock since skb_copy_datagram_iter() may sleep. 615 */ 616 spin_unlock_bh(&vvs->rx_lock); 617 618 err = skb_copy_datagram_iter(skb, 619 VIRTIO_VSOCK_SKB_CB(skb)->offset, 620 &msg->msg_iter, bytes); 621 if (err) 622 goto out; 623 624 spin_lock_bh(&vvs->rx_lock); 625 626 total += bytes; 627 628 VIRTIO_VSOCK_SKB_CB(skb)->offset += bytes; 629 630 if (skb->len == VIRTIO_VSOCK_SKB_CB(skb)->offset) { 631 dequeued = le32_to_cpu(virtio_vsock_hdr(skb)->len); 632 __skb_unlink(skb, &vvs->rx_queue); 633 consume_skb(skb); 634 } 635 636 virtio_transport_dec_rx_pkt(vvs, bytes, dequeued); 637 } 638 639 fwd_cnt_delta = vvs->fwd_cnt - vvs->last_fwd_cnt; 640 free_space = vvs->buf_alloc - fwd_cnt_delta; 641 low_rx_bytes = (vvs->rx_bytes < 642 sock_rcvlowat(sk_vsock(vsk), 0, INT_MAX)); 643 644 spin_unlock_bh(&vvs->rx_lock); 645 646 /* To reduce the number of credit update messages, 647 * don't update credits as long as lots of space is available. 648 * Note: the limit chosen here is arbitrary. Setting the limit 649 * too high causes extra messages. Too low causes transmitter 650 * stalls. As stalls are in theory more expensive than extra 651 * messages, we set the limit to a high value. TODO: experiment 652 * with different values. Also send credit update message when 653 * number of bytes in rx queue is not enough to wake up reader. 654 */ 655 if (fwd_cnt_delta && 656 (free_space < VIRTIO_VSOCK_MAX_PKT_BUF_SIZE || low_rx_bytes)) 657 virtio_transport_send_credit_update(vsk); 658 659 return total; 660 661 out: 662 if (total) 663 err = total; 664 return err; 665 } 666 667 static ssize_t 668 virtio_transport_seqpacket_do_peek(struct vsock_sock *vsk, 669 struct msghdr *msg) 670 { 671 struct virtio_vsock_sock *vvs = vsk->trans; 672 struct sk_buff *skb; 673 size_t total, len; 674 675 spin_lock_bh(&vvs->rx_lock); 676 677 if (!vvs->msg_count) { 678 spin_unlock_bh(&vvs->rx_lock); 679 return 0; 680 } 681 682 total = 0; 683 len = msg_data_left(msg); 684 685 skb_queue_walk(&vvs->rx_queue, skb) { 686 struct virtio_vsock_hdr *hdr; 687 688 if (total < len) { 689 size_t bytes; 690 int err; 691 692 bytes = len - total; 693 if (bytes > skb->len) 694 bytes = skb->len; 695 696 spin_unlock_bh(&vvs->rx_lock); 697 698 /* sk_lock is held by caller so no one else can dequeue. 699 * Unlock rx_lock since skb_copy_datagram_iter() may sleep. 700 */ 701 err = skb_copy_datagram_iter(skb, VIRTIO_VSOCK_SKB_CB(skb)->offset, 702 &msg->msg_iter, bytes); 703 if (err) 704 return err; 705 706 spin_lock_bh(&vvs->rx_lock); 707 } 708 709 total += skb->len; 710 hdr = virtio_vsock_hdr(skb); 711 712 if (le32_to_cpu(hdr->flags) & VIRTIO_VSOCK_SEQ_EOM) { 713 if (le32_to_cpu(hdr->flags) & VIRTIO_VSOCK_SEQ_EOR) 714 msg->msg_flags |= MSG_EOR; 715 716 break; 717 } 718 } 719 720 spin_unlock_bh(&vvs->rx_lock); 721 722 return total; 723 } 724 725 static int virtio_transport_seqpacket_do_dequeue(struct vsock_sock *vsk, 726 struct msghdr *msg, 727 int flags) 728 { 729 struct virtio_vsock_sock *vvs = vsk->trans; 730 int dequeued_len = 0; 731 size_t user_buf_len = msg_data_left(msg); 732 bool msg_ready = false; 733 struct sk_buff *skb; 734 735 spin_lock_bh(&vvs->rx_lock); 736 737 if (vvs->msg_count == 0) { 738 spin_unlock_bh(&vvs->rx_lock); 739 return 0; 740 } 741 742 while (!msg_ready) { 743 struct virtio_vsock_hdr *hdr; 744 size_t pkt_len; 745 746 skb = __skb_dequeue(&vvs->rx_queue); 747 if (!skb) 748 break; 749 hdr = virtio_vsock_hdr(skb); 750 pkt_len = (size_t)le32_to_cpu(hdr->len); 751 752 if (dequeued_len >= 0) { 753 size_t bytes_to_copy; 754 755 bytes_to_copy = min(user_buf_len, pkt_len); 756 757 if (bytes_to_copy) { 758 int err; 759 760 /* sk_lock is held by caller so no one else can dequeue. 761 * Unlock rx_lock since skb_copy_datagram_iter() may sleep. 762 */ 763 spin_unlock_bh(&vvs->rx_lock); 764 765 err = skb_copy_datagram_iter(skb, 0, 766 &msg->msg_iter, 767 bytes_to_copy); 768 if (err) { 769 /* Copy of message failed. Rest of 770 * fragments will be freed without copy. 771 */ 772 dequeued_len = err; 773 } else { 774 user_buf_len -= bytes_to_copy; 775 } 776 777 spin_lock_bh(&vvs->rx_lock); 778 } 779 780 if (dequeued_len >= 0) 781 dequeued_len += pkt_len; 782 } 783 784 if (le32_to_cpu(hdr->flags) & VIRTIO_VSOCK_SEQ_EOM) { 785 msg_ready = true; 786 vvs->msg_count--; 787 788 if (le32_to_cpu(hdr->flags) & VIRTIO_VSOCK_SEQ_EOR) 789 msg->msg_flags |= MSG_EOR; 790 } 791 792 virtio_transport_dec_rx_pkt(vvs, pkt_len, pkt_len); 793 kfree_skb(skb); 794 } 795 796 spin_unlock_bh(&vvs->rx_lock); 797 798 virtio_transport_send_credit_update(vsk); 799 800 return dequeued_len; 801 } 802 803 ssize_t 804 virtio_transport_stream_dequeue(struct vsock_sock *vsk, 805 struct msghdr *msg, 806 size_t len, int flags) 807 { 808 if (flags & MSG_PEEK) 809 return virtio_transport_stream_do_peek(vsk, msg, len); 810 else 811 return virtio_transport_stream_do_dequeue(vsk, msg, len); 812 } 813 EXPORT_SYMBOL_GPL(virtio_transport_stream_dequeue); 814 815 ssize_t 816 virtio_transport_seqpacket_dequeue(struct vsock_sock *vsk, 817 struct msghdr *msg, 818 int flags) 819 { 820 if (flags & MSG_PEEK) 821 return virtio_transport_seqpacket_do_peek(vsk, msg); 822 else 823 return virtio_transport_seqpacket_do_dequeue(vsk, msg, flags); 824 } 825 EXPORT_SYMBOL_GPL(virtio_transport_seqpacket_dequeue); 826 827 static u32 virtio_transport_tx_buf_size(struct virtio_vsock_sock *vvs) 828 { 829 /* The peer advertises its receive buffer via peer_buf_alloc, but we 830 * cap it to our local buf_alloc so a remote peer cannot force us to 831 * queue more data than our own buffer configuration allows. 832 */ 833 return min(vvs->peer_buf_alloc, vvs->buf_alloc); 834 } 835 836 int 837 virtio_transport_seqpacket_enqueue(struct vsock_sock *vsk, 838 struct msghdr *msg, 839 size_t len) 840 { 841 struct virtio_vsock_sock *vvs = vsk->trans; 842 843 spin_lock_bh(&vvs->tx_lock); 844 845 if (len > virtio_transport_tx_buf_size(vvs)) { 846 spin_unlock_bh(&vvs->tx_lock); 847 return -EMSGSIZE; 848 } 849 850 spin_unlock_bh(&vvs->tx_lock); 851 852 return virtio_transport_stream_enqueue(vsk, msg, len); 853 } 854 EXPORT_SYMBOL_GPL(virtio_transport_seqpacket_enqueue); 855 856 int 857 virtio_transport_dgram_dequeue(struct vsock_sock *vsk, 858 struct msghdr *msg, 859 size_t len, int flags) 860 { 861 return -EOPNOTSUPP; 862 } 863 EXPORT_SYMBOL_GPL(virtio_transport_dgram_dequeue); 864 865 s64 virtio_transport_stream_has_data(struct vsock_sock *vsk) 866 { 867 struct virtio_vsock_sock *vvs = vsk->trans; 868 s64 bytes; 869 870 spin_lock_bh(&vvs->rx_lock); 871 bytes = vvs->rx_bytes; 872 spin_unlock_bh(&vvs->rx_lock); 873 874 return bytes; 875 } 876 EXPORT_SYMBOL_GPL(virtio_transport_stream_has_data); 877 878 u32 virtio_transport_seqpacket_has_data(struct vsock_sock *vsk) 879 { 880 struct virtio_vsock_sock *vvs = vsk->trans; 881 u32 msg_count; 882 883 spin_lock_bh(&vvs->rx_lock); 884 msg_count = vvs->msg_count; 885 spin_unlock_bh(&vvs->rx_lock); 886 887 return msg_count; 888 } 889 EXPORT_SYMBOL_GPL(virtio_transport_seqpacket_has_data); 890 891 static s64 virtio_transport_has_space(struct virtio_vsock_sock *vvs) 892 { 893 s64 bytes; 894 895 /* Use s64 arithmetic so if the peer shrinks peer_buf_alloc while 896 * we have bytes in flight (tx_cnt - peer_fwd_cnt), the subtraction 897 * does not underflow. 898 */ 899 bytes = (s64)virtio_transport_tx_buf_size(vvs) - 900 (vvs->tx_cnt - vvs->peer_fwd_cnt); 901 if (bytes < 0) 902 bytes = 0; 903 904 return bytes; 905 } 906 907 s64 virtio_transport_stream_has_space(struct vsock_sock *vsk) 908 { 909 struct virtio_vsock_sock *vvs = vsk->trans; 910 s64 bytes; 911 912 spin_lock_bh(&vvs->tx_lock); 913 bytes = virtio_transport_has_space(vvs); 914 spin_unlock_bh(&vvs->tx_lock); 915 916 return bytes; 917 } 918 EXPORT_SYMBOL_GPL(virtio_transport_stream_has_space); 919 920 int virtio_transport_do_socket_init(struct vsock_sock *vsk, 921 struct vsock_sock *psk) 922 { 923 struct virtio_vsock_sock *vvs; 924 925 vvs = kzalloc_obj(*vvs); 926 if (!vvs) 927 return -ENOMEM; 928 929 vsk->trans = vvs; 930 vvs->vsk = vsk; 931 if (psk && psk->trans) { 932 struct virtio_vsock_sock *ptrans = psk->trans; 933 934 vvs->peer_buf_alloc = ptrans->peer_buf_alloc; 935 } 936 937 if (vsk->buffer_size > VIRTIO_VSOCK_MAX_BUF_SIZE) 938 vsk->buffer_size = VIRTIO_VSOCK_MAX_BUF_SIZE; 939 940 vvs->buf_alloc = vsk->buffer_size; 941 942 spin_lock_init(&vvs->rx_lock); 943 spin_lock_init(&vvs->tx_lock); 944 skb_queue_head_init(&vvs->rx_queue); 945 946 return 0; 947 } 948 EXPORT_SYMBOL_GPL(virtio_transport_do_socket_init); 949 950 /* sk_lock held by the caller */ 951 void virtio_transport_notify_buffer_size(struct vsock_sock *vsk, u64 *val) 952 { 953 struct virtio_vsock_sock *vvs = vsk->trans; 954 955 if (*val > VIRTIO_VSOCK_MAX_BUF_SIZE) 956 *val = VIRTIO_VSOCK_MAX_BUF_SIZE; 957 958 vvs->buf_alloc = *val; 959 960 virtio_transport_send_credit_update(vsk); 961 } 962 EXPORT_SYMBOL_GPL(virtio_transport_notify_buffer_size); 963 964 int 965 virtio_transport_notify_poll_in(struct vsock_sock *vsk, 966 size_t target, 967 bool *data_ready_now) 968 { 969 *data_ready_now = vsock_stream_has_data(vsk) >= target; 970 971 return 0; 972 } 973 EXPORT_SYMBOL_GPL(virtio_transport_notify_poll_in); 974 975 int 976 virtio_transport_notify_poll_out(struct vsock_sock *vsk, 977 size_t target, 978 bool *space_avail_now) 979 { 980 s64 free_space; 981 982 free_space = vsock_stream_has_space(vsk); 983 if (free_space > 0) 984 *space_avail_now = true; 985 else if (free_space == 0) 986 *space_avail_now = false; 987 988 return 0; 989 } 990 EXPORT_SYMBOL_GPL(virtio_transport_notify_poll_out); 991 992 int virtio_transport_notify_recv_init(struct vsock_sock *vsk, 993 size_t target, struct vsock_transport_recv_notify_data *data) 994 { 995 return 0; 996 } 997 EXPORT_SYMBOL_GPL(virtio_transport_notify_recv_init); 998 999 int virtio_transport_notify_recv_pre_block(struct vsock_sock *vsk, 1000 size_t target, struct vsock_transport_recv_notify_data *data) 1001 { 1002 return 0; 1003 } 1004 EXPORT_SYMBOL_GPL(virtio_transport_notify_recv_pre_block); 1005 1006 int virtio_transport_notify_recv_pre_dequeue(struct vsock_sock *vsk, 1007 size_t target, struct vsock_transport_recv_notify_data *data) 1008 { 1009 return 0; 1010 } 1011 EXPORT_SYMBOL_GPL(virtio_transport_notify_recv_pre_dequeue); 1012 1013 int virtio_transport_notify_recv_post_dequeue(struct vsock_sock *vsk, 1014 size_t target, ssize_t copied, bool data_read, 1015 struct vsock_transport_recv_notify_data *data) 1016 { 1017 return 0; 1018 } 1019 EXPORT_SYMBOL_GPL(virtio_transport_notify_recv_post_dequeue); 1020 1021 int virtio_transport_notify_send_init(struct vsock_sock *vsk, 1022 struct vsock_transport_send_notify_data *data) 1023 { 1024 return 0; 1025 } 1026 EXPORT_SYMBOL_GPL(virtio_transport_notify_send_init); 1027 1028 int virtio_transport_notify_send_pre_block(struct vsock_sock *vsk, 1029 struct vsock_transport_send_notify_data *data) 1030 { 1031 return 0; 1032 } 1033 EXPORT_SYMBOL_GPL(virtio_transport_notify_send_pre_block); 1034 1035 int virtio_transport_notify_send_pre_enqueue(struct vsock_sock *vsk, 1036 struct vsock_transport_send_notify_data *data) 1037 { 1038 return 0; 1039 } 1040 EXPORT_SYMBOL_GPL(virtio_transport_notify_send_pre_enqueue); 1041 1042 int virtio_transport_notify_send_post_enqueue(struct vsock_sock *vsk, 1043 ssize_t written, struct vsock_transport_send_notify_data *data) 1044 { 1045 return 0; 1046 } 1047 EXPORT_SYMBOL_GPL(virtio_transport_notify_send_post_enqueue); 1048 1049 u64 virtio_transport_stream_rcvhiwat(struct vsock_sock *vsk) 1050 { 1051 return vsk->buffer_size; 1052 } 1053 EXPORT_SYMBOL_GPL(virtio_transport_stream_rcvhiwat); 1054 1055 bool virtio_transport_stream_is_active(struct vsock_sock *vsk) 1056 { 1057 return true; 1058 } 1059 EXPORT_SYMBOL_GPL(virtio_transport_stream_is_active); 1060 1061 int virtio_transport_dgram_bind(struct vsock_sock *vsk, 1062 struct sockaddr_vm *addr) 1063 { 1064 return -EOPNOTSUPP; 1065 } 1066 EXPORT_SYMBOL_GPL(virtio_transport_dgram_bind); 1067 1068 bool virtio_transport_dgram_allow(struct vsock_sock *vsk, u32 cid, u32 port) 1069 { 1070 return false; 1071 } 1072 EXPORT_SYMBOL_GPL(virtio_transport_dgram_allow); 1073 1074 int virtio_transport_connect(struct vsock_sock *vsk) 1075 { 1076 struct virtio_vsock_pkt_info info = { 1077 .op = VIRTIO_VSOCK_OP_REQUEST, 1078 .vsk = vsk, 1079 .net = sock_net(sk_vsock(vsk)), 1080 }; 1081 1082 return virtio_transport_send_pkt_info(vsk, &info); 1083 } 1084 EXPORT_SYMBOL_GPL(virtio_transport_connect); 1085 1086 int virtio_transport_shutdown(struct vsock_sock *vsk, int mode) 1087 { 1088 struct virtio_vsock_pkt_info info = { 1089 .op = VIRTIO_VSOCK_OP_SHUTDOWN, 1090 .flags = (mode & RCV_SHUTDOWN ? 1091 VIRTIO_VSOCK_SHUTDOWN_RCV : 0) | 1092 (mode & SEND_SHUTDOWN ? 1093 VIRTIO_VSOCK_SHUTDOWN_SEND : 0), 1094 .vsk = vsk, 1095 .net = sock_net(sk_vsock(vsk)), 1096 }; 1097 1098 return virtio_transport_send_pkt_info(vsk, &info); 1099 } 1100 EXPORT_SYMBOL_GPL(virtio_transport_shutdown); 1101 1102 int 1103 virtio_transport_dgram_enqueue(struct vsock_sock *vsk, 1104 struct sockaddr_vm *remote_addr, 1105 struct msghdr *msg, 1106 size_t dgram_len) 1107 { 1108 return -EOPNOTSUPP; 1109 } 1110 EXPORT_SYMBOL_GPL(virtio_transport_dgram_enqueue); 1111 1112 ssize_t 1113 virtio_transport_stream_enqueue(struct vsock_sock *vsk, 1114 struct msghdr *msg, 1115 size_t len) 1116 { 1117 struct virtio_vsock_pkt_info info = { 1118 .op = VIRTIO_VSOCK_OP_RW, 1119 .msg = msg, 1120 .pkt_len = len, 1121 .vsk = vsk, 1122 .net = sock_net(sk_vsock(vsk)), 1123 }; 1124 1125 return virtio_transport_send_pkt_info(vsk, &info); 1126 } 1127 EXPORT_SYMBOL_GPL(virtio_transport_stream_enqueue); 1128 1129 void virtio_transport_destruct(struct vsock_sock *vsk) 1130 { 1131 struct virtio_vsock_sock *vvs = vsk->trans; 1132 1133 virtio_transport_cancel_close_work(vsk, true); 1134 1135 kfree(vvs); 1136 vsk->trans = NULL; 1137 } 1138 EXPORT_SYMBOL_GPL(virtio_transport_destruct); 1139 1140 ssize_t virtio_transport_unsent_bytes(struct vsock_sock *vsk) 1141 { 1142 struct virtio_vsock_sock *vvs = vsk->trans; 1143 size_t ret; 1144 1145 spin_lock_bh(&vvs->tx_lock); 1146 ret = vvs->bytes_unsent; 1147 spin_unlock_bh(&vvs->tx_lock); 1148 1149 return ret; 1150 } 1151 EXPORT_SYMBOL_GPL(virtio_transport_unsent_bytes); 1152 1153 static int virtio_transport_reset(struct vsock_sock *vsk, 1154 struct sk_buff *skb) 1155 { 1156 struct virtio_vsock_pkt_info info = { 1157 .op = VIRTIO_VSOCK_OP_RST, 1158 .reply = !!skb, 1159 .vsk = vsk, 1160 .net = sock_net(sk_vsock(vsk)), 1161 }; 1162 1163 /* Send RST only if the original pkt is not a RST pkt */ 1164 if (skb && le16_to_cpu(virtio_vsock_hdr(skb)->op) == VIRTIO_VSOCK_OP_RST) 1165 return 0; 1166 1167 return virtio_transport_send_pkt_info(vsk, &info); 1168 } 1169 1170 /* Normally packets are associated with a socket. There may be no socket if an 1171 * attempt was made to connect to a socket that does not exist. 1172 * 1173 * net refers to the namespace of whoever sent the invalid message. For 1174 * loopback, this is the namespace of the socket. For vhost, this is the 1175 * namespace of the VM (i.e., vhost_vsock). 1176 */ 1177 static int virtio_transport_reset_no_sock(const struct virtio_transport *t, 1178 struct sk_buff *skb, struct net *net) 1179 { 1180 struct virtio_vsock_hdr *hdr = virtio_vsock_hdr(skb); 1181 struct virtio_vsock_pkt_info info = { 1182 .op = VIRTIO_VSOCK_OP_RST, 1183 .type = le16_to_cpu(hdr->type), 1184 .reply = true, 1185 1186 /* Set sk owner to socket we are replying to (may be NULL for 1187 * non-loopback). This keeps a reference to the sock and 1188 * sock_net(sk) until the reply skb is freed. 1189 */ 1190 .vsk = vsock_sk(skb->sk), 1191 1192 /* net is not defined here because we pass it directly to 1193 * t->send_pkt(), instead of relying on 1194 * virtio_transport_send_pkt_info() to pass it. It is not needed 1195 * by virtio_transport_alloc_skb(). 1196 */ 1197 }; 1198 struct sk_buff *reply; 1199 1200 /* Send RST only if the original pkt is not a RST pkt */ 1201 if (le16_to_cpu(hdr->op) == VIRTIO_VSOCK_OP_RST) 1202 return 0; 1203 1204 if (!t) 1205 return -ENOTCONN; 1206 1207 reply = virtio_transport_alloc_skb(&info, 0, false, 1208 le64_to_cpu(hdr->dst_cid), 1209 le32_to_cpu(hdr->dst_port), 1210 le64_to_cpu(hdr->src_cid), 1211 le32_to_cpu(hdr->src_port)); 1212 if (!reply) 1213 return -ENOMEM; 1214 1215 return t->send_pkt(reply, net); 1216 } 1217 1218 /* This function should be called with sk_lock held and SOCK_DONE set */ 1219 static void virtio_transport_remove_sock(struct vsock_sock *vsk) 1220 { 1221 struct virtio_vsock_sock *vvs = vsk->trans; 1222 1223 /* We don't need to take rx_lock, as the socket is closing and we are 1224 * removing it. 1225 */ 1226 __skb_queue_purge(&vvs->rx_queue); 1227 vsock_remove_sock(vsk); 1228 } 1229 1230 static void virtio_transport_cancel_close_work(struct vsock_sock *vsk, 1231 bool cancel_timeout) 1232 { 1233 struct sock *sk = sk_vsock(vsk); 1234 1235 if (vsk->close_work_scheduled && 1236 (!cancel_timeout || cancel_delayed_work(&vsk->close_work))) { 1237 vsk->close_work_scheduled = false; 1238 1239 virtio_transport_remove_sock(vsk); 1240 1241 /* Release refcnt obtained when we scheduled the timeout */ 1242 sock_put(sk); 1243 } 1244 } 1245 1246 static void virtio_transport_do_close(struct vsock_sock *vsk, 1247 bool cancel_timeout) 1248 { 1249 struct sock *sk = sk_vsock(vsk); 1250 1251 sock_set_flag(sk, SOCK_DONE); 1252 vsk->peer_shutdown = SHUTDOWN_MASK; 1253 if (vsock_stream_has_data(vsk) <= 0) 1254 sk->sk_state = TCP_CLOSING; 1255 sk->sk_state_change(sk); 1256 1257 virtio_transport_cancel_close_work(vsk, cancel_timeout); 1258 } 1259 1260 static void virtio_transport_close_timeout(struct work_struct *work) 1261 { 1262 struct vsock_sock *vsk = 1263 container_of(work, struct vsock_sock, close_work.work); 1264 struct sock *sk = sk_vsock(vsk); 1265 1266 sock_hold(sk); 1267 lock_sock(sk); 1268 1269 if (!sock_flag(sk, SOCK_DONE)) { 1270 (void)virtio_transport_reset(vsk, NULL); 1271 1272 virtio_transport_do_close(vsk, false); 1273 } 1274 1275 vsk->close_work_scheduled = false; 1276 1277 release_sock(sk); 1278 sock_put(sk); 1279 } 1280 1281 /* User context, vsk->sk is locked */ 1282 static bool virtio_transport_close(struct vsock_sock *vsk) 1283 { 1284 struct sock *sk = &vsk->sk; 1285 1286 if (!(sk->sk_state == TCP_ESTABLISHED || 1287 sk->sk_state == TCP_CLOSING)) 1288 return true; 1289 1290 /* Already received SHUTDOWN from peer, reply with RST */ 1291 if ((vsk->peer_shutdown & SHUTDOWN_MASK) == SHUTDOWN_MASK) { 1292 (void)virtio_transport_reset(vsk, NULL); 1293 return true; 1294 } 1295 1296 if ((sk->sk_shutdown & SHUTDOWN_MASK) != SHUTDOWN_MASK) 1297 (void)virtio_transport_shutdown(vsk, SHUTDOWN_MASK); 1298 1299 if (!(current->flags & PF_EXITING)) 1300 vsock_linger(sk); 1301 1302 if (sock_flag(sk, SOCK_DONE)) { 1303 return true; 1304 } 1305 1306 sock_hold(sk); 1307 INIT_DELAYED_WORK(&vsk->close_work, 1308 virtio_transport_close_timeout); 1309 vsk->close_work_scheduled = true; 1310 schedule_delayed_work(&vsk->close_work, VSOCK_CLOSE_TIMEOUT); 1311 return false; 1312 } 1313 1314 void virtio_transport_release(struct vsock_sock *vsk) 1315 { 1316 struct sock *sk = &vsk->sk; 1317 bool remove_sock = true; 1318 1319 if (sk->sk_type == SOCK_STREAM || sk->sk_type == SOCK_SEQPACKET) 1320 remove_sock = virtio_transport_close(vsk); 1321 1322 if (remove_sock) { 1323 sock_set_flag(sk, SOCK_DONE); 1324 virtio_transport_remove_sock(vsk); 1325 } 1326 } 1327 EXPORT_SYMBOL_GPL(virtio_transport_release); 1328 1329 static int 1330 virtio_transport_recv_connecting(struct sock *sk, 1331 struct sk_buff *skb) 1332 { 1333 struct virtio_vsock_hdr *hdr = virtio_vsock_hdr(skb); 1334 struct vsock_sock *vsk = vsock_sk(sk); 1335 int skerr; 1336 int err; 1337 1338 switch (le16_to_cpu(hdr->op)) { 1339 case VIRTIO_VSOCK_OP_RESPONSE: 1340 sk->sk_state = TCP_ESTABLISHED; 1341 sk->sk_socket->state = SS_CONNECTED; 1342 vsock_insert_connected(vsk); 1343 sk->sk_state_change(sk); 1344 break; 1345 case VIRTIO_VSOCK_OP_INVALID: 1346 break; 1347 case VIRTIO_VSOCK_OP_RST: 1348 skerr = ECONNRESET; 1349 err = 0; 1350 goto destroy; 1351 default: 1352 skerr = EPROTO; 1353 err = -EINVAL; 1354 goto destroy; 1355 } 1356 return 0; 1357 1358 destroy: 1359 virtio_transport_reset(vsk, skb); 1360 sk->sk_state = TCP_CLOSE; 1361 sk->sk_err = skerr; 1362 sk_error_report(sk); 1363 return err; 1364 } 1365 1366 static void 1367 virtio_transport_recv_enqueue(struct vsock_sock *vsk, 1368 struct sk_buff *skb) 1369 { 1370 struct virtio_vsock_sock *vvs = vsk->trans; 1371 bool can_enqueue, free_pkt = false; 1372 struct virtio_vsock_hdr *hdr; 1373 u32 len; 1374 1375 hdr = virtio_vsock_hdr(skb); 1376 len = le32_to_cpu(hdr->len); 1377 1378 spin_lock_bh(&vvs->rx_lock); 1379 1380 can_enqueue = virtio_transport_inc_rx_pkt(vvs, len); 1381 if (!can_enqueue) { 1382 free_pkt = true; 1383 goto out; 1384 } 1385 1386 if (le32_to_cpu(hdr->flags) & VIRTIO_VSOCK_SEQ_EOM) 1387 vvs->msg_count++; 1388 1389 /* Try to copy small packets into the buffer of last packet queued, 1390 * to avoid wasting memory queueing the entire buffer with a small 1391 * payload. Skip non-linear (e.g. zerocopy) skbs; these carry payload 1392 * in skb_shinfo. 1393 */ 1394 if (len <= GOOD_COPY_LEN && !skb_queue_empty(&vvs->rx_queue) && 1395 !skb_is_nonlinear(skb)) { 1396 struct virtio_vsock_hdr *last_hdr; 1397 struct sk_buff *last_skb; 1398 1399 last_skb = skb_peek_tail(&vvs->rx_queue); 1400 last_hdr = virtio_vsock_hdr(last_skb); 1401 1402 /* If there is space in the last packet queued, we copy the 1403 * new packet in its buffer. We avoid this if the last packet 1404 * queued has VIRTIO_VSOCK_SEQ_EOM set, because this is 1405 * delimiter of SEQPACKET message, so 'pkt' is the first packet 1406 * of a new message. 1407 */ 1408 if (skb->len < skb_tailroom(last_skb) && 1409 !(le32_to_cpu(last_hdr->flags) & VIRTIO_VSOCK_SEQ_EOM)) { 1410 memcpy(skb_put(last_skb, skb->len), skb->data, skb->len); 1411 free_pkt = true; 1412 last_hdr->flags |= hdr->flags; 1413 le32_add_cpu(&last_hdr->len, len); 1414 goto out; 1415 } 1416 } 1417 1418 __skb_queue_tail(&vvs->rx_queue, skb); 1419 1420 out: 1421 spin_unlock_bh(&vvs->rx_lock); 1422 if (free_pkt) 1423 kfree_skb(skb); 1424 } 1425 1426 static int 1427 virtio_transport_recv_connected(struct sock *sk, 1428 struct sk_buff *skb) 1429 { 1430 struct virtio_vsock_hdr *hdr = virtio_vsock_hdr(skb); 1431 struct vsock_sock *vsk = vsock_sk(sk); 1432 int err = 0; 1433 1434 switch (le16_to_cpu(hdr->op)) { 1435 case VIRTIO_VSOCK_OP_RW: 1436 virtio_transport_recv_enqueue(vsk, skb); 1437 vsock_data_ready(sk); 1438 return err; 1439 case VIRTIO_VSOCK_OP_CREDIT_REQUEST: 1440 virtio_transport_send_credit_update(vsk); 1441 break; 1442 case VIRTIO_VSOCK_OP_CREDIT_UPDATE: 1443 sk->sk_write_space(sk); 1444 break; 1445 case VIRTIO_VSOCK_OP_SHUTDOWN: 1446 if (le32_to_cpu(hdr->flags) & VIRTIO_VSOCK_SHUTDOWN_RCV) 1447 vsk->peer_shutdown |= RCV_SHUTDOWN; 1448 if (le32_to_cpu(hdr->flags) & VIRTIO_VSOCK_SHUTDOWN_SEND) 1449 vsk->peer_shutdown |= SEND_SHUTDOWN; 1450 if (vsk->peer_shutdown == SHUTDOWN_MASK) { 1451 if (vsock_stream_has_data(vsk) <= 0 && !sock_flag(sk, SOCK_DONE)) { 1452 (void)virtio_transport_reset(vsk, NULL); 1453 virtio_transport_do_close(vsk, true); 1454 } 1455 /* Remove this socket anyway because the remote peer sent 1456 * the shutdown. This way a new connection will succeed 1457 * if the remote peer uses the same source port, 1458 * even if the old socket is still unreleased, but now disconnected. 1459 */ 1460 vsock_remove_sock(vsk); 1461 } 1462 if (le32_to_cpu(virtio_vsock_hdr(skb)->flags)) 1463 sk->sk_state_change(sk); 1464 break; 1465 case VIRTIO_VSOCK_OP_RST: 1466 virtio_transport_do_close(vsk, true); 1467 break; 1468 default: 1469 err = -EINVAL; 1470 break; 1471 } 1472 1473 kfree_skb(skb); 1474 return err; 1475 } 1476 1477 static void 1478 virtio_transport_recv_disconnecting(struct sock *sk, 1479 struct sk_buff *skb) 1480 { 1481 struct virtio_vsock_hdr *hdr = virtio_vsock_hdr(skb); 1482 struct vsock_sock *vsk = vsock_sk(sk); 1483 1484 if (le16_to_cpu(hdr->op) == VIRTIO_VSOCK_OP_RST) 1485 virtio_transport_do_close(vsk, true); 1486 } 1487 1488 static int 1489 virtio_transport_send_response(struct vsock_sock *vsk, 1490 struct sk_buff *skb) 1491 { 1492 struct virtio_vsock_hdr *hdr = virtio_vsock_hdr(skb); 1493 struct virtio_vsock_pkt_info info = { 1494 .op = VIRTIO_VSOCK_OP_RESPONSE, 1495 .remote_cid = le64_to_cpu(hdr->src_cid), 1496 .remote_port = le32_to_cpu(hdr->src_port), 1497 .reply = true, 1498 .vsk = vsk, 1499 .net = sock_net(sk_vsock(vsk)), 1500 }; 1501 1502 return virtio_transport_send_pkt_info(vsk, &info); 1503 } 1504 1505 static bool virtio_transport_space_update(struct sock *sk, 1506 struct sk_buff *skb) 1507 { 1508 struct virtio_vsock_hdr *hdr = virtio_vsock_hdr(skb); 1509 struct vsock_sock *vsk = vsock_sk(sk); 1510 struct virtio_vsock_sock *vvs = vsk->trans; 1511 bool space_available; 1512 1513 /* Listener sockets are not associated with any transport, so we are 1514 * not able to take the state to see if there is space available in the 1515 * remote peer, but since they are only used to receive requests, we 1516 * can assume that there is always space available in the other peer. 1517 */ 1518 if (!vvs) 1519 return true; 1520 1521 /* buf_alloc and fwd_cnt is always included in the hdr */ 1522 spin_lock_bh(&vvs->tx_lock); 1523 vvs->peer_buf_alloc = le32_to_cpu(hdr->buf_alloc); 1524 vvs->peer_fwd_cnt = le32_to_cpu(hdr->fwd_cnt); 1525 space_available = virtio_transport_has_space(vvs); 1526 spin_unlock_bh(&vvs->tx_lock); 1527 return space_available; 1528 } 1529 1530 /* Handle server socket */ 1531 static int 1532 virtio_transport_recv_listen(struct sock *sk, struct sk_buff *skb, 1533 struct virtio_transport *t) 1534 { 1535 struct virtio_vsock_hdr *hdr = virtio_vsock_hdr(skb); 1536 struct vsock_sock *vsk = vsock_sk(sk); 1537 struct vsock_sock *vchild; 1538 struct sock *child; 1539 int ret; 1540 1541 if (le16_to_cpu(hdr->op) != VIRTIO_VSOCK_OP_REQUEST) { 1542 virtio_transport_reset_no_sock(t, skb, sock_net(sk)); 1543 return -EINVAL; 1544 } 1545 1546 if (sk_acceptq_is_full(sk)) { 1547 virtio_transport_reset_no_sock(t, skb, sock_net(sk)); 1548 return -ENOMEM; 1549 } 1550 1551 /* __vsock_release() might have already flushed accept_queue. 1552 * Subsequent enqueues would lead to a memory leak. 1553 */ 1554 if (sk->sk_shutdown == SHUTDOWN_MASK) { 1555 virtio_transport_reset_no_sock(t, skb, sock_net(sk)); 1556 return -ESHUTDOWN; 1557 } 1558 1559 child = vsock_create_connected(sk); 1560 if (!child) { 1561 virtio_transport_reset_no_sock(t, skb, sock_net(sk)); 1562 return -ENOMEM; 1563 } 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 sk_acceptq_added(sk); 1587 if (virtio_transport_space_update(child, skb)) 1588 child->sk_write_space(child); 1589 1590 vsock_insert_connected(vchild); 1591 vsock_enqueue_accept(sk, child); 1592 virtio_transport_send_response(vchild, skb); 1593 1594 release_sock(child); 1595 1596 sk->sk_data_ready(sk); 1597 return 0; 1598 } 1599 1600 static bool virtio_transport_valid_type(u16 type) 1601 { 1602 return (type == VIRTIO_VSOCK_TYPE_STREAM) || 1603 (type == VIRTIO_VSOCK_TYPE_SEQPACKET); 1604 } 1605 1606 /* We are under the virtio-vsock's vsock->rx_lock or vhost-vsock's vq->mutex 1607 * lock. 1608 */ 1609 void virtio_transport_recv_pkt(struct virtio_transport *t, 1610 struct sk_buff *skb, struct net *net) 1611 { 1612 struct virtio_vsock_hdr *hdr = virtio_vsock_hdr(skb); 1613 struct sockaddr_vm src, dst; 1614 struct vsock_sock *vsk; 1615 struct sock *sk; 1616 bool space_available; 1617 1618 vsock_addr_init(&src, le64_to_cpu(hdr->src_cid), 1619 le32_to_cpu(hdr->src_port)); 1620 vsock_addr_init(&dst, le64_to_cpu(hdr->dst_cid), 1621 le32_to_cpu(hdr->dst_port)); 1622 1623 trace_virtio_transport_recv_pkt(src.svm_cid, src.svm_port, 1624 dst.svm_cid, dst.svm_port, 1625 le32_to_cpu(hdr->len), 1626 le16_to_cpu(hdr->type), 1627 le16_to_cpu(hdr->op), 1628 le32_to_cpu(hdr->flags), 1629 le32_to_cpu(hdr->buf_alloc), 1630 le32_to_cpu(hdr->fwd_cnt)); 1631 1632 if (!virtio_transport_valid_type(le16_to_cpu(hdr->type))) { 1633 (void)virtio_transport_reset_no_sock(t, skb, net); 1634 goto free_pkt; 1635 } 1636 1637 /* The socket must be in connected or bound table 1638 * otherwise send reset back 1639 */ 1640 sk = vsock_find_connected_socket_net(&src, &dst, net); 1641 if (!sk) { 1642 sk = vsock_find_bound_socket_net(&dst, net); 1643 if (!sk) { 1644 (void)virtio_transport_reset_no_sock(t, skb, net); 1645 goto free_pkt; 1646 } 1647 } 1648 1649 if (virtio_transport_get_type(sk) != le16_to_cpu(hdr->type)) { 1650 (void)virtio_transport_reset_no_sock(t, skb, net); 1651 sock_put(sk); 1652 goto free_pkt; 1653 } 1654 1655 if (!skb_set_owner_sk_safe(skb, sk)) { 1656 WARN_ONCE(1, "receiving vsock socket has sk_refcnt == 0\n"); 1657 goto free_pkt; 1658 } 1659 1660 vsk = vsock_sk(sk); 1661 1662 lock_sock(sk); 1663 1664 /* Check if sk has been closed or assigned to another transport before 1665 * lock_sock (note: listener sockets are not assigned to any transport) 1666 */ 1667 if (sock_flag(sk, SOCK_DONE) || 1668 (sk->sk_state != TCP_LISTEN && vsk->transport != &t->transport)) { 1669 (void)virtio_transport_reset_no_sock(t, skb, net); 1670 release_sock(sk); 1671 sock_put(sk); 1672 goto free_pkt; 1673 } 1674 1675 space_available = virtio_transport_space_update(sk, skb); 1676 1677 /* Update CID in case it has changed after a transport reset event */ 1678 if (vsk->local_addr.svm_cid != VMADDR_CID_ANY) 1679 vsk->local_addr.svm_cid = dst.svm_cid; 1680 1681 if (space_available) 1682 sk->sk_write_space(sk); 1683 1684 switch (sk->sk_state) { 1685 case TCP_LISTEN: 1686 virtio_transport_recv_listen(sk, skb, t); 1687 kfree_skb(skb); 1688 break; 1689 case TCP_SYN_SENT: 1690 virtio_transport_recv_connecting(sk, skb); 1691 kfree_skb(skb); 1692 break; 1693 case TCP_ESTABLISHED: 1694 virtio_transport_recv_connected(sk, skb); 1695 break; 1696 case TCP_CLOSING: 1697 virtio_transport_recv_disconnecting(sk, skb); 1698 kfree_skb(skb); 1699 break; 1700 default: 1701 (void)virtio_transport_reset_no_sock(t, skb, net); 1702 kfree_skb(skb); 1703 break; 1704 } 1705 1706 release_sock(sk); 1707 1708 /* Release refcnt obtained when we fetched this socket out of the 1709 * bound or connected list. 1710 */ 1711 sock_put(sk); 1712 return; 1713 1714 free_pkt: 1715 kfree_skb(skb); 1716 } 1717 EXPORT_SYMBOL_GPL(virtio_transport_recv_pkt); 1718 1719 /* Remove skbs found in a queue that have a vsk that matches. 1720 * 1721 * Each skb is freed. 1722 * 1723 * Returns the count of skbs that were reply packets. 1724 */ 1725 int virtio_transport_purge_skbs(void *vsk, struct sk_buff_head *queue) 1726 { 1727 struct sk_buff_head freeme; 1728 struct sk_buff *skb, *tmp; 1729 int cnt = 0; 1730 1731 skb_queue_head_init(&freeme); 1732 1733 spin_lock_bh(&queue->lock); 1734 skb_queue_walk_safe(queue, skb, tmp) { 1735 if (vsock_sk(skb->sk) != vsk) 1736 continue; 1737 1738 __skb_unlink(skb, queue); 1739 __skb_queue_tail(&freeme, skb); 1740 1741 if (virtio_vsock_skb_reply(skb)) 1742 cnt++; 1743 } 1744 spin_unlock_bh(&queue->lock); 1745 1746 __skb_queue_purge(&freeme); 1747 1748 return cnt; 1749 } 1750 EXPORT_SYMBOL_GPL(virtio_transport_purge_skbs); 1751 1752 int virtio_transport_read_skb(struct vsock_sock *vsk, skb_read_actor_t recv_actor) 1753 { 1754 struct virtio_vsock_sock *vvs = vsk->trans; 1755 struct sock *sk = sk_vsock(vsk); 1756 struct virtio_vsock_hdr *hdr; 1757 struct sk_buff *skb; 1758 u32 pkt_len; 1759 int off = 0; 1760 int err; 1761 1762 spin_lock_bh(&vvs->rx_lock); 1763 /* Use __skb_recv_datagram() for race-free handling of the receive. It 1764 * works for types other than dgrams. 1765 */ 1766 skb = __skb_recv_datagram(sk, &vvs->rx_queue, MSG_DONTWAIT, &off, &err); 1767 if (!skb) { 1768 spin_unlock_bh(&vvs->rx_lock); 1769 return err; 1770 } 1771 1772 hdr = virtio_vsock_hdr(skb); 1773 if (le32_to_cpu(hdr->flags) & VIRTIO_VSOCK_SEQ_EOM) 1774 vvs->msg_count--; 1775 1776 pkt_len = le32_to_cpu(hdr->len); 1777 virtio_transport_dec_rx_pkt(vvs, pkt_len, pkt_len); 1778 spin_unlock_bh(&vvs->rx_lock); 1779 1780 virtio_transport_send_credit_update(vsk); 1781 1782 return recv_actor(sk, skb); 1783 } 1784 EXPORT_SYMBOL_GPL(virtio_transport_read_skb); 1785 1786 int virtio_transport_notify_set_rcvlowat(struct vsock_sock *vsk, int val) 1787 { 1788 struct virtio_vsock_sock *vvs = vsk->trans; 1789 bool send_update; 1790 1791 spin_lock_bh(&vvs->rx_lock); 1792 1793 /* If number of available bytes is less than new SO_RCVLOWAT value, 1794 * kick sender to send more data, because sender may sleep in its 1795 * 'send()' syscall waiting for enough space at our side. Also 1796 * don't send credit update when peer already knows actual value - 1797 * such transmission will be useless. 1798 */ 1799 send_update = (vvs->rx_bytes < val) && 1800 (vvs->fwd_cnt != vvs->last_fwd_cnt); 1801 1802 spin_unlock_bh(&vvs->rx_lock); 1803 1804 if (send_update) { 1805 int err; 1806 1807 err = virtio_transport_send_credit_update(vsk); 1808 if (err < 0) 1809 return err; 1810 } 1811 1812 return 0; 1813 } 1814 EXPORT_SYMBOL_GPL(virtio_transport_notify_set_rcvlowat); 1815 1816 MODULE_LICENSE("GPL v2"); 1817 MODULE_AUTHOR("Asias He"); 1818 MODULE_DESCRIPTION("common code for virtio vsock"); 1819