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_fill_skb(struct sk_buff *skb, 74 struct virtio_vsock_pkt_info *info, 75 size_t len, 76 bool zcopy) 77 { 78 struct msghdr *msg = info->msg; 79 80 if (zcopy) 81 return __zerocopy_sg_from_iter(msg, NULL, skb, 82 &msg->msg_iter, len, NULL); 83 84 virtio_vsock_skb_put(skb, len); 85 return skb_copy_datagram_from_iter_full(skb, 0, &msg->msg_iter, len); 86 } 87 88 static void virtio_transport_init_hdr(struct sk_buff *skb, 89 struct virtio_vsock_pkt_info *info, 90 size_t payload_len, 91 u32 src_cid, 92 u32 src_port, 93 u32 dst_cid, 94 u32 dst_port) 95 { 96 struct virtio_vsock_hdr *hdr; 97 98 hdr = virtio_vsock_hdr(skb); 99 hdr->type = cpu_to_le16(info->type); 100 hdr->op = cpu_to_le16(info->op); 101 hdr->src_cid = cpu_to_le64(src_cid); 102 hdr->dst_cid = cpu_to_le64(dst_cid); 103 hdr->src_port = cpu_to_le32(src_port); 104 hdr->dst_port = cpu_to_le32(dst_port); 105 hdr->flags = cpu_to_le32(info->flags); 106 hdr->len = cpu_to_le32(payload_len); 107 hdr->buf_alloc = cpu_to_le32(0); 108 hdr->fwd_cnt = cpu_to_le32(0); 109 } 110 111 /* Packet capture */ 112 static struct sk_buff *virtio_transport_build_skb(void *opaque) 113 { 114 struct virtio_vsock_hdr *pkt_hdr; 115 struct sk_buff *pkt = opaque; 116 struct af_vsockmon_hdr *hdr; 117 struct sk_buff *skb; 118 size_t payload_len; 119 120 /* A packet could be split to fit the RX buffer, so we use 121 * the payload length from the header, which has been updated 122 * by the sender to reflect the fragment size. 123 */ 124 pkt_hdr = virtio_vsock_hdr(pkt); 125 payload_len = le32_to_cpu(pkt_hdr->len); 126 127 skb = alloc_skb(sizeof(*hdr) + sizeof(*pkt_hdr) + payload_len, 128 GFP_ATOMIC); 129 if (!skb) 130 return NULL; 131 132 hdr = skb_put(skb, sizeof(*hdr)); 133 134 /* pkt->hdr is little-endian so no need to byteswap here */ 135 hdr->src_cid = pkt_hdr->src_cid; 136 hdr->src_port = pkt_hdr->src_port; 137 hdr->dst_cid = pkt_hdr->dst_cid; 138 hdr->dst_port = pkt_hdr->dst_port; 139 140 hdr->transport = cpu_to_le16(AF_VSOCK_TRANSPORT_VIRTIO); 141 hdr->len = cpu_to_le16(sizeof(*pkt_hdr)); 142 memset(hdr->reserved, 0, sizeof(hdr->reserved)); 143 144 switch (le16_to_cpu(pkt_hdr->op)) { 145 case VIRTIO_VSOCK_OP_REQUEST: 146 case VIRTIO_VSOCK_OP_RESPONSE: 147 hdr->op = cpu_to_le16(AF_VSOCK_OP_CONNECT); 148 break; 149 case VIRTIO_VSOCK_OP_RST: 150 case VIRTIO_VSOCK_OP_SHUTDOWN: 151 hdr->op = cpu_to_le16(AF_VSOCK_OP_DISCONNECT); 152 break; 153 case VIRTIO_VSOCK_OP_RW: 154 hdr->op = cpu_to_le16(AF_VSOCK_OP_PAYLOAD); 155 break; 156 case VIRTIO_VSOCK_OP_CREDIT_UPDATE: 157 case VIRTIO_VSOCK_OP_CREDIT_REQUEST: 158 hdr->op = cpu_to_le16(AF_VSOCK_OP_CONTROL); 159 break; 160 default: 161 hdr->op = cpu_to_le16(AF_VSOCK_OP_UNKNOWN); 162 break; 163 } 164 165 skb_put_data(skb, pkt_hdr, sizeof(*pkt_hdr)); 166 167 if (payload_len) { 168 struct iov_iter iov_iter; 169 struct kvec kvec; 170 void *data = skb_put(skb, payload_len); 171 172 kvec.iov_base = data; 173 kvec.iov_len = payload_len; 174 iov_iter_kvec(&iov_iter, ITER_DEST, &kvec, 1, payload_len); 175 176 if (skb_copy_datagram_iter(pkt, VIRTIO_VSOCK_SKB_CB(pkt)->offset, 177 &iov_iter, payload_len)) { 178 kfree_skb(skb); 179 return NULL; 180 } 181 } 182 183 return skb; 184 } 185 186 void virtio_transport_deliver_tap_pkt(struct sk_buff *skb) 187 { 188 if (virtio_vsock_skb_tap_delivered(skb)) 189 return; 190 191 vsock_deliver_tap(virtio_transport_build_skb, skb); 192 virtio_vsock_skb_set_tap_delivered(skb); 193 } 194 EXPORT_SYMBOL_GPL(virtio_transport_deliver_tap_pkt); 195 196 static u16 virtio_transport_get_type(struct sock *sk) 197 { 198 if (sk->sk_type == SOCK_STREAM) 199 return VIRTIO_VSOCK_TYPE_STREAM; 200 else 201 return VIRTIO_VSOCK_TYPE_SEQPACKET; 202 } 203 204 /* Returns new sk_buff on success, otherwise returns NULL. */ 205 static struct sk_buff *virtio_transport_alloc_skb(struct virtio_vsock_pkt_info *info, 206 size_t payload_len, 207 bool zcopy, 208 struct ubuf_info *uarg, 209 u32 src_cid, 210 u32 src_port, 211 u32 dst_cid, 212 u32 dst_port) 213 { 214 struct vsock_sock *vsk; 215 struct sk_buff *skb; 216 size_t skb_len; 217 218 skb_len = VIRTIO_VSOCK_SKB_HEADROOM; 219 220 if (!zcopy) 221 skb_len += payload_len; 222 223 skb = virtio_vsock_alloc_skb(skb_len, GFP_KERNEL); 224 if (!skb) 225 return NULL; 226 227 virtio_transport_init_hdr(skb, info, payload_len, src_cid, src_port, 228 dst_cid, dst_port); 229 230 vsk = info->vsk; 231 232 /* If 'vsk' != NULL then payload is always present, so we 233 * will never call '__zerocopy_sg_from_iter()' below without 234 * setting skb owner in 'skb_set_owner_w()'. The only case 235 * when 'vsk' == NULL is VIRTIO_VSOCK_OP_RST control message 236 * without payload. 237 */ 238 WARN_ON_ONCE(!(vsk && (info->msg && payload_len)) && zcopy); 239 240 /* Set owner here, because '__zerocopy_sg_from_iter()' uses 241 * owner of skb without check to update 'sk_wmem_alloc'. 242 */ 243 if (vsk) 244 skb_set_owner_w(skb, sk_vsock(vsk)); 245 246 if (info->msg && payload_len > 0) { 247 int err; 248 249 /* Bind the zerocopy lifetime before filling frags so error 250 * rollback frees managed fixed-buffer pages through 251 * the uarg-aware path. 252 */ 253 skb_zcopy_set(skb, uarg, NULL); 254 255 err = virtio_transport_fill_skb(skb, info, payload_len, zcopy); 256 if (err) 257 goto out; 258 259 if (msg_data_left(info->msg) == 0 && 260 info->type == VIRTIO_VSOCK_TYPE_SEQPACKET) { 261 struct virtio_vsock_hdr *hdr = virtio_vsock_hdr(skb); 262 263 hdr->flags |= cpu_to_le32(VIRTIO_VSOCK_SEQ_EOM); 264 265 if (info->msg->msg_flags & MSG_EOR) 266 hdr->flags |= cpu_to_le32(VIRTIO_VSOCK_SEQ_EOR); 267 } 268 } 269 270 if (info->reply) 271 virtio_vsock_skb_set_reply(skb); 272 273 trace_virtio_transport_alloc_pkt(src_cid, src_port, 274 dst_cid, dst_port, 275 payload_len, 276 info->type, 277 info->op, 278 info->flags, 279 zcopy); 280 281 return skb; 282 out: 283 kfree_skb(skb); 284 return NULL; 285 } 286 287 /* This function can only be used on connecting/connected sockets, 288 * since a socket assigned to a transport is required. 289 * 290 * Do not use on listener sockets! 291 */ 292 static int virtio_transport_send_pkt_info(struct vsock_sock *vsk, 293 struct virtio_vsock_pkt_info *info) 294 { 295 u32 max_skb_len = VIRTIO_VSOCK_MAX_PKT_BUF_SIZE; 296 u32 src_cid, src_port, dst_cid, dst_port; 297 const struct virtio_transport *t_ops; 298 struct virtio_vsock_sock *vvs; 299 struct ubuf_info *uarg = NULL; 300 u32 pkt_len = info->pkt_len; 301 bool can_zcopy = false; 302 bool have_uref = false; 303 u32 rest_len; 304 int ret; 305 306 info->type = virtio_transport_get_type(sk_vsock(vsk)); 307 308 t_ops = virtio_transport_get_ops(vsk); 309 if (unlikely(!t_ops)) 310 return -EFAULT; 311 312 src_cid = t_ops->transport.get_local_cid(); 313 src_port = vsk->local_addr.svm_port; 314 if (!info->remote_cid) { 315 dst_cid = vsk->remote_addr.svm_cid; 316 dst_port = vsk->remote_addr.svm_port; 317 } else { 318 dst_cid = info->remote_cid; 319 dst_port = info->remote_port; 320 } 321 322 vvs = vsk->trans; 323 324 /* virtio_transport_get_credit might return less than pkt_len credit */ 325 pkt_len = virtio_transport_get_credit(vvs, pkt_len); 326 327 /* Do not send zero length OP_RW pkt */ 328 if (pkt_len == 0 && info->op == VIRTIO_VSOCK_OP_RW) 329 return pkt_len; 330 331 if (info->msg) { 332 /* If zerocopy is not enabled by 'setsockopt()', we behave as 333 * there is no MSG_ZEROCOPY flag set. 334 */ 335 if (!sock_flag(sk_vsock(vsk), SOCK_ZEROCOPY)) 336 info->msg->msg_flags &= ~MSG_ZEROCOPY; 337 338 if (info->msg->msg_flags & MSG_ZEROCOPY) 339 can_zcopy = virtio_transport_can_zcopy(t_ops, info, pkt_len); 340 341 if (can_zcopy) 342 max_skb_len = min_t(u32, VIRTIO_VSOCK_MAX_PKT_BUF_SIZE, 343 (MAX_SKB_FRAGS * PAGE_SIZE)); 344 345 if (info->msg->msg_flags & MSG_ZEROCOPY && 346 info->op == VIRTIO_VSOCK_OP_RW) { 347 uarg = info->msg->msg_ubuf; 348 349 if (!uarg) { 350 uarg = msg_zerocopy_realloc(sk_vsock(vsk), 351 pkt_len, NULL, false); 352 if (!uarg) { 353 virtio_transport_put_credit(vvs, pkt_len); 354 return -ENOMEM; 355 } 356 357 if (!can_zcopy) 358 uarg_to_msgzc(uarg)->zerocopy = 0; 359 360 have_uref = true; 361 } 362 } 363 } 364 365 rest_len = pkt_len; 366 367 do { 368 struct sk_buff *skb; 369 size_t skb_len; 370 371 skb_len = min(max_skb_len, rest_len); 372 373 skb = virtio_transport_alloc_skb(info, skb_len, can_zcopy, 374 uarg, 375 src_cid, src_port, 376 dst_cid, dst_port); 377 if (!skb) { 378 ret = -ENOMEM; 379 break; 380 } 381 382 virtio_transport_inc_tx_pkt(vvs, skb); 383 384 ret = t_ops->send_pkt(skb, info->net); 385 if (ret < 0) 386 break; 387 388 /* Both virtio and vhost 'send_pkt()' returns 'skb_len', 389 * but for reliability use 'ret' instead of 'skb_len'. 390 * Also if partial send happens (e.g. 'ret' != 'skb_len') 391 * somehow, we break this loop, but account such returned 392 * value in 'virtio_transport_put_credit()'. 393 */ 394 rest_len -= ret; 395 396 if (WARN_ONCE(ret != skb_len, 397 "'send_pkt()' returns %i, but %zu expected\n", 398 ret, skb_len)) 399 break; 400 } while (rest_len); 401 402 virtio_transport_put_credit(vvs, rest_len); 403 404 /* msg_zerocopy_realloc() initializes the ubuf_info refcnt to 1. 405 * skb_zcopy_set() increases it for each skb, so we can drop that 406 * initial reference to keep it balanced. 407 */ 408 if (have_uref) { 409 if (rest_len == pkt_len) 410 /* No data sent, abort the notification. */ 411 net_zcopy_put_abort(uarg, true); 412 else 413 net_zcopy_put(uarg); 414 } 415 416 /* Return number of bytes, if any data has been sent. */ 417 if (rest_len != pkt_len) 418 ret = pkt_len - rest_len; 419 420 return ret; 421 } 422 423 static bool virtio_transport_inc_rx_pkt(struct virtio_vsock_sock *vvs, 424 u32 len) 425 { 426 u64 skb_overhead = (skb_queue_len(&vvs->rx_queue) + 1) * SKB_TRUESIZE(0); 427 428 /* Allow at most buf_alloc * 2 total budget (payload + overhead), 429 * similar to how SO_RCVBUF is doubled to reserve space for sk_buff 430 * metadata. Check payload against buf_alloc to be sure the other 431 * peer is respecting the credit, and sk_buff overhead to bound 432 * queue growth. 433 */ 434 if ((u64)vvs->buf_used + len > vvs->buf_alloc || 435 skb_overhead > vvs->buf_alloc) 436 return false; 437 438 vvs->rx_bytes += len; 439 vvs->buf_used += len; 440 return true; 441 } 442 443 static void virtio_transport_dec_rx_pkt(struct virtio_vsock_sock *vvs, 444 u32 bytes_read, u32 bytes_dequeued) 445 { 446 vvs->rx_bytes -= bytes_read; 447 vvs->buf_used -= bytes_dequeued; 448 vvs->fwd_cnt += bytes_dequeued; 449 } 450 451 void virtio_transport_inc_tx_pkt(struct virtio_vsock_sock *vvs, struct sk_buff *skb) 452 { 453 struct virtio_vsock_hdr *hdr = virtio_vsock_hdr(skb); 454 455 spin_lock_bh(&vvs->rx_lock); 456 vvs->last_fwd_cnt = vvs->fwd_cnt; 457 hdr->fwd_cnt = cpu_to_le32(vvs->fwd_cnt); 458 hdr->buf_alloc = cpu_to_le32(vvs->buf_alloc); 459 spin_unlock_bh(&vvs->rx_lock); 460 } 461 EXPORT_SYMBOL_GPL(virtio_transport_inc_tx_pkt); 462 463 void virtio_transport_consume_skb_sent(struct sk_buff *skb, bool consume) 464 { 465 struct sock *s = skb->sk; 466 467 if (s && skb->len) { 468 struct vsock_sock *vs = vsock_sk(s); 469 struct virtio_vsock_sock *vvs; 470 471 vvs = vs->trans; 472 473 spin_lock_bh(&vvs->tx_lock); 474 vvs->bytes_unsent -= skb->len; 475 spin_unlock_bh(&vvs->tx_lock); 476 } 477 478 if (consume) 479 consume_skb(skb); 480 } 481 EXPORT_SYMBOL_GPL(virtio_transport_consume_skb_sent); 482 483 u32 virtio_transport_get_credit(struct virtio_vsock_sock *vvs, u32 credit) 484 { 485 u32 ret; 486 487 if (!credit) 488 return 0; 489 490 spin_lock_bh(&vvs->tx_lock); 491 ret = min_t(u32, credit, virtio_transport_has_space(vvs)); 492 vvs->tx_cnt += ret; 493 vvs->bytes_unsent += ret; 494 spin_unlock_bh(&vvs->tx_lock); 495 496 return ret; 497 } 498 EXPORT_SYMBOL_GPL(virtio_transport_get_credit); 499 500 void virtio_transport_put_credit(struct virtio_vsock_sock *vvs, u32 credit) 501 { 502 if (!credit) 503 return; 504 505 spin_lock_bh(&vvs->tx_lock); 506 vvs->tx_cnt -= credit; 507 vvs->bytes_unsent -= credit; 508 spin_unlock_bh(&vvs->tx_lock); 509 } 510 EXPORT_SYMBOL_GPL(virtio_transport_put_credit); 511 512 static int virtio_transport_send_credit_update(struct vsock_sock *vsk) 513 { 514 struct virtio_vsock_pkt_info info = { 515 .op = VIRTIO_VSOCK_OP_CREDIT_UPDATE, 516 .vsk = vsk, 517 .net = sock_net(sk_vsock(vsk)), 518 }; 519 520 return virtio_transport_send_pkt_info(vsk, &info); 521 } 522 523 static ssize_t 524 virtio_transport_stream_do_peek(struct vsock_sock *vsk, 525 struct msghdr *msg, 526 size_t len) 527 { 528 struct virtio_vsock_sock *vvs = vsk->trans; 529 struct sk_buff *skb; 530 size_t total = 0; 531 int err; 532 533 spin_lock_bh(&vvs->rx_lock); 534 535 skb_queue_walk(&vvs->rx_queue, skb) { 536 size_t bytes; 537 538 bytes = min_t(size_t, len - total, 539 skb->len - VIRTIO_VSOCK_SKB_CB(skb)->offset); 540 541 spin_unlock_bh(&vvs->rx_lock); 542 543 /* sk_lock is held by caller so no one else can dequeue. 544 * Unlock rx_lock since skb_copy_datagram_iter() may sleep. 545 */ 546 err = skb_copy_datagram_iter(skb, VIRTIO_VSOCK_SKB_CB(skb)->offset, 547 &msg->msg_iter, bytes); 548 if (err) 549 goto out; 550 551 total += bytes; 552 553 spin_lock_bh(&vvs->rx_lock); 554 555 if (total == len) 556 break; 557 } 558 559 spin_unlock_bh(&vvs->rx_lock); 560 561 return total; 562 563 out: 564 if (total) 565 err = total; 566 return err; 567 } 568 569 static ssize_t 570 virtio_transport_stream_do_dequeue(struct vsock_sock *vsk, 571 struct msghdr *msg, 572 size_t len) 573 { 574 struct virtio_vsock_sock *vvs = vsk->trans; 575 struct sk_buff *skb; 576 u32 fwd_cnt_delta; 577 bool low_rx_bytes; 578 int err = -EFAULT; 579 size_t total = 0; 580 u32 free_space; 581 582 spin_lock_bh(&vvs->rx_lock); 583 584 if (WARN_ONCE(skb_queue_empty(&vvs->rx_queue) && vvs->rx_bytes, 585 "rx_queue is empty, but rx_bytes is non-zero\n")) { 586 spin_unlock_bh(&vvs->rx_lock); 587 return err; 588 } 589 590 while (total < len && !skb_queue_empty(&vvs->rx_queue)) { 591 size_t bytes, dequeued = 0; 592 593 skb = skb_peek(&vvs->rx_queue); 594 595 bytes = min_t(size_t, len - total, 596 skb->len - VIRTIO_VSOCK_SKB_CB(skb)->offset); 597 598 /* sk_lock is held by caller so no one else can dequeue. 599 * Unlock rx_lock since skb_copy_datagram_iter() may sleep. 600 */ 601 spin_unlock_bh(&vvs->rx_lock); 602 603 err = skb_copy_datagram_iter(skb, 604 VIRTIO_VSOCK_SKB_CB(skb)->offset, 605 &msg->msg_iter, bytes); 606 if (err) 607 goto out; 608 609 spin_lock_bh(&vvs->rx_lock); 610 611 total += bytes; 612 613 VIRTIO_VSOCK_SKB_CB(skb)->offset += bytes; 614 615 if (skb->len == VIRTIO_VSOCK_SKB_CB(skb)->offset) { 616 dequeued = le32_to_cpu(virtio_vsock_hdr(skb)->len); 617 __skb_unlink(skb, &vvs->rx_queue); 618 consume_skb(skb); 619 } 620 621 virtio_transport_dec_rx_pkt(vvs, bytes, dequeued); 622 } 623 624 fwd_cnt_delta = vvs->fwd_cnt - vvs->last_fwd_cnt; 625 free_space = vvs->buf_alloc - fwd_cnt_delta; 626 low_rx_bytes = (vvs->rx_bytes < 627 sock_rcvlowat(sk_vsock(vsk), 0, INT_MAX)); 628 629 spin_unlock_bh(&vvs->rx_lock); 630 631 /* To reduce the number of credit update messages, 632 * don't update credits as long as lots of space is available. 633 * Note: the limit chosen here is arbitrary. Setting the limit 634 * too high causes extra messages. Too low causes transmitter 635 * stalls. As stalls are in theory more expensive than extra 636 * messages, we set the limit to a high value. TODO: experiment 637 * with different values. Also send credit update message when 638 * number of bytes in rx queue is not enough to wake up reader. 639 */ 640 if (fwd_cnt_delta && 641 (free_space < VIRTIO_VSOCK_MAX_PKT_BUF_SIZE || low_rx_bytes)) 642 virtio_transport_send_credit_update(vsk); 643 644 return total; 645 646 out: 647 if (total) 648 err = total; 649 return err; 650 } 651 652 static ssize_t 653 virtio_transport_seqpacket_do_peek(struct vsock_sock *vsk, 654 struct msghdr *msg) 655 { 656 struct virtio_vsock_sock *vvs = vsk->trans; 657 struct sk_buff *skb; 658 size_t total, len; 659 660 spin_lock_bh(&vvs->rx_lock); 661 662 if (!vvs->msg_count) { 663 spin_unlock_bh(&vvs->rx_lock); 664 return 0; 665 } 666 667 total = 0; 668 len = msg_data_left(msg); 669 670 skb_queue_walk(&vvs->rx_queue, skb) { 671 struct virtio_vsock_hdr *hdr; 672 673 if (total < len) { 674 size_t bytes; 675 int err; 676 677 bytes = len - total; 678 if (bytes > skb->len) 679 bytes = skb->len; 680 681 spin_unlock_bh(&vvs->rx_lock); 682 683 /* sk_lock is held by caller so no one else can dequeue. 684 * Unlock rx_lock since skb_copy_datagram_iter() may sleep. 685 */ 686 err = skb_copy_datagram_iter(skb, VIRTIO_VSOCK_SKB_CB(skb)->offset, 687 &msg->msg_iter, bytes); 688 if (err) 689 return err; 690 691 spin_lock_bh(&vvs->rx_lock); 692 } 693 694 total += skb->len; 695 hdr = virtio_vsock_hdr(skb); 696 697 if (le32_to_cpu(hdr->flags) & VIRTIO_VSOCK_SEQ_EOM) { 698 if (le32_to_cpu(hdr->flags) & VIRTIO_VSOCK_SEQ_EOR) 699 msg->msg_flags |= MSG_EOR; 700 701 break; 702 } 703 } 704 705 spin_unlock_bh(&vvs->rx_lock); 706 707 return total; 708 } 709 710 static int virtio_transport_seqpacket_do_dequeue(struct vsock_sock *vsk, 711 struct msghdr *msg, 712 int flags) 713 { 714 struct virtio_vsock_sock *vvs = vsk->trans; 715 int dequeued_len = 0; 716 size_t user_buf_len = msg_data_left(msg); 717 bool msg_ready = false; 718 struct sk_buff *skb; 719 720 spin_lock_bh(&vvs->rx_lock); 721 722 if (vvs->msg_count == 0) { 723 spin_unlock_bh(&vvs->rx_lock); 724 return 0; 725 } 726 727 while (!msg_ready) { 728 struct virtio_vsock_hdr *hdr; 729 size_t pkt_len; 730 731 skb = __skb_dequeue(&vvs->rx_queue); 732 if (!skb) 733 break; 734 hdr = virtio_vsock_hdr(skb); 735 pkt_len = (size_t)le32_to_cpu(hdr->len); 736 737 if (dequeued_len >= 0) { 738 size_t bytes_to_copy; 739 740 bytes_to_copy = min(user_buf_len, pkt_len); 741 742 if (bytes_to_copy) { 743 int err; 744 745 /* sk_lock is held by caller so no one else can dequeue. 746 * Unlock rx_lock since skb_copy_datagram_iter() may sleep. 747 */ 748 spin_unlock_bh(&vvs->rx_lock); 749 750 err = skb_copy_datagram_iter(skb, 0, 751 &msg->msg_iter, 752 bytes_to_copy); 753 if (err) { 754 /* Copy of message failed. Rest of 755 * fragments will be freed without copy. 756 */ 757 dequeued_len = err; 758 } else { 759 user_buf_len -= bytes_to_copy; 760 } 761 762 spin_lock_bh(&vvs->rx_lock); 763 } 764 765 if (dequeued_len >= 0) 766 dequeued_len += pkt_len; 767 } 768 769 if (le32_to_cpu(hdr->flags) & VIRTIO_VSOCK_SEQ_EOM) { 770 msg_ready = true; 771 vvs->msg_count--; 772 773 if (le32_to_cpu(hdr->flags) & VIRTIO_VSOCK_SEQ_EOR) 774 msg->msg_flags |= MSG_EOR; 775 } 776 777 virtio_transport_dec_rx_pkt(vvs, pkt_len, pkt_len); 778 kfree_skb(skb); 779 } 780 781 spin_unlock_bh(&vvs->rx_lock); 782 783 virtio_transport_send_credit_update(vsk); 784 785 return dequeued_len; 786 } 787 788 ssize_t 789 virtio_transport_stream_dequeue(struct vsock_sock *vsk, 790 struct msghdr *msg, 791 size_t len, int flags) 792 { 793 if (flags & MSG_PEEK) 794 return virtio_transport_stream_do_peek(vsk, msg, len); 795 else 796 return virtio_transport_stream_do_dequeue(vsk, msg, len); 797 } 798 EXPORT_SYMBOL_GPL(virtio_transport_stream_dequeue); 799 800 ssize_t 801 virtio_transport_seqpacket_dequeue(struct vsock_sock *vsk, 802 struct msghdr *msg, 803 int flags) 804 { 805 if (flags & MSG_PEEK) 806 return virtio_transport_seqpacket_do_peek(vsk, msg); 807 else 808 return virtio_transport_seqpacket_do_dequeue(vsk, msg, flags); 809 } 810 EXPORT_SYMBOL_GPL(virtio_transport_seqpacket_dequeue); 811 812 static u32 virtio_transport_tx_buf_size(struct virtio_vsock_sock *vvs) 813 { 814 /* The peer advertises its receive buffer via peer_buf_alloc, but we 815 * cap it to our local buf_alloc so a remote peer cannot force us to 816 * queue more data than our own buffer configuration allows. 817 */ 818 return min(vvs->peer_buf_alloc, vvs->buf_alloc); 819 } 820 821 int 822 virtio_transport_seqpacket_enqueue(struct vsock_sock *vsk, 823 struct msghdr *msg, 824 size_t len) 825 { 826 struct virtio_vsock_sock *vvs = vsk->trans; 827 828 spin_lock_bh(&vvs->tx_lock); 829 830 if (len > virtio_transport_tx_buf_size(vvs)) { 831 spin_unlock_bh(&vvs->tx_lock); 832 return -EMSGSIZE; 833 } 834 835 spin_unlock_bh(&vvs->tx_lock); 836 837 return virtio_transport_stream_enqueue(vsk, msg, len); 838 } 839 EXPORT_SYMBOL_GPL(virtio_transport_seqpacket_enqueue); 840 841 int 842 virtio_transport_dgram_dequeue(struct vsock_sock *vsk, 843 struct msghdr *msg, 844 size_t len, int flags) 845 { 846 return -EOPNOTSUPP; 847 } 848 EXPORT_SYMBOL_GPL(virtio_transport_dgram_dequeue); 849 850 s64 virtio_transport_stream_has_data(struct vsock_sock *vsk) 851 { 852 struct virtio_vsock_sock *vvs = vsk->trans; 853 s64 bytes; 854 855 spin_lock_bh(&vvs->rx_lock); 856 bytes = vvs->rx_bytes; 857 spin_unlock_bh(&vvs->rx_lock); 858 859 return bytes; 860 } 861 EXPORT_SYMBOL_GPL(virtio_transport_stream_has_data); 862 863 u32 virtio_transport_seqpacket_has_data(struct vsock_sock *vsk) 864 { 865 struct virtio_vsock_sock *vvs = vsk->trans; 866 u32 msg_count; 867 868 spin_lock_bh(&vvs->rx_lock); 869 msg_count = vvs->msg_count; 870 spin_unlock_bh(&vvs->rx_lock); 871 872 return msg_count; 873 } 874 EXPORT_SYMBOL_GPL(virtio_transport_seqpacket_has_data); 875 876 static s64 virtio_transport_has_space(struct virtio_vsock_sock *vvs) 877 { 878 s64 bytes; 879 880 /* Use s64 arithmetic so if the peer shrinks peer_buf_alloc while 881 * we have bytes in flight (tx_cnt - peer_fwd_cnt), the subtraction 882 * does not underflow. 883 */ 884 bytes = (s64)virtio_transport_tx_buf_size(vvs) - 885 (vvs->tx_cnt - vvs->peer_fwd_cnt); 886 if (bytes < 0) 887 bytes = 0; 888 889 return bytes; 890 } 891 892 s64 virtio_transport_stream_has_space(struct vsock_sock *vsk) 893 { 894 struct virtio_vsock_sock *vvs = vsk->trans; 895 s64 bytes; 896 897 spin_lock_bh(&vvs->tx_lock); 898 bytes = virtio_transport_has_space(vvs); 899 spin_unlock_bh(&vvs->tx_lock); 900 901 return bytes; 902 } 903 EXPORT_SYMBOL_GPL(virtio_transport_stream_has_space); 904 905 int virtio_transport_do_socket_init(struct vsock_sock *vsk, 906 struct vsock_sock *psk) 907 { 908 struct virtio_vsock_sock *vvs; 909 910 vvs = kzalloc_obj(*vvs); 911 if (!vvs) 912 return -ENOMEM; 913 914 vsk->trans = vvs; 915 vvs->vsk = vsk; 916 if (psk && psk->trans) { 917 struct virtio_vsock_sock *ptrans = psk->trans; 918 919 vvs->peer_buf_alloc = ptrans->peer_buf_alloc; 920 } 921 922 if (vsk->buffer_size > VIRTIO_VSOCK_MAX_BUF_SIZE) 923 vsk->buffer_size = VIRTIO_VSOCK_MAX_BUF_SIZE; 924 925 vvs->buf_alloc = vsk->buffer_size; 926 927 spin_lock_init(&vvs->rx_lock); 928 spin_lock_init(&vvs->tx_lock); 929 skb_queue_head_init(&vvs->rx_queue); 930 931 return 0; 932 } 933 EXPORT_SYMBOL_GPL(virtio_transport_do_socket_init); 934 935 /* sk_lock held by the caller */ 936 void virtio_transport_notify_buffer_size(struct vsock_sock *vsk, u64 *val) 937 { 938 struct virtio_vsock_sock *vvs = vsk->trans; 939 940 if (*val > VIRTIO_VSOCK_MAX_BUF_SIZE) 941 *val = VIRTIO_VSOCK_MAX_BUF_SIZE; 942 943 vvs->buf_alloc = *val; 944 945 virtio_transport_send_credit_update(vsk); 946 } 947 EXPORT_SYMBOL_GPL(virtio_transport_notify_buffer_size); 948 949 int 950 virtio_transport_notify_poll_in(struct vsock_sock *vsk, 951 size_t target, 952 bool *data_ready_now) 953 { 954 *data_ready_now = vsock_stream_has_data(vsk) >= target; 955 956 return 0; 957 } 958 EXPORT_SYMBOL_GPL(virtio_transport_notify_poll_in); 959 960 int 961 virtio_transport_notify_poll_out(struct vsock_sock *vsk, 962 size_t target, 963 bool *space_avail_now) 964 { 965 s64 free_space; 966 967 free_space = vsock_stream_has_space(vsk); 968 if (free_space > 0) 969 *space_avail_now = true; 970 else if (free_space == 0) 971 *space_avail_now = false; 972 973 return 0; 974 } 975 EXPORT_SYMBOL_GPL(virtio_transport_notify_poll_out); 976 977 int virtio_transport_notify_recv_init(struct vsock_sock *vsk, 978 size_t target, struct vsock_transport_recv_notify_data *data) 979 { 980 return 0; 981 } 982 EXPORT_SYMBOL_GPL(virtio_transport_notify_recv_init); 983 984 int virtio_transport_notify_recv_pre_block(struct vsock_sock *vsk, 985 size_t target, struct vsock_transport_recv_notify_data *data) 986 { 987 return 0; 988 } 989 EXPORT_SYMBOL_GPL(virtio_transport_notify_recv_pre_block); 990 991 int virtio_transport_notify_recv_pre_dequeue(struct vsock_sock *vsk, 992 size_t target, struct vsock_transport_recv_notify_data *data) 993 { 994 return 0; 995 } 996 EXPORT_SYMBOL_GPL(virtio_transport_notify_recv_pre_dequeue); 997 998 int virtio_transport_notify_recv_post_dequeue(struct vsock_sock *vsk, 999 size_t target, ssize_t copied, bool data_read, 1000 struct vsock_transport_recv_notify_data *data) 1001 { 1002 return 0; 1003 } 1004 EXPORT_SYMBOL_GPL(virtio_transport_notify_recv_post_dequeue); 1005 1006 int virtio_transport_notify_send_init(struct vsock_sock *vsk, 1007 struct vsock_transport_send_notify_data *data) 1008 { 1009 return 0; 1010 } 1011 EXPORT_SYMBOL_GPL(virtio_transport_notify_send_init); 1012 1013 int virtio_transport_notify_send_pre_block(struct vsock_sock *vsk, 1014 struct vsock_transport_send_notify_data *data) 1015 { 1016 return 0; 1017 } 1018 EXPORT_SYMBOL_GPL(virtio_transport_notify_send_pre_block); 1019 1020 int virtio_transport_notify_send_pre_enqueue(struct vsock_sock *vsk, 1021 struct vsock_transport_send_notify_data *data) 1022 { 1023 return 0; 1024 } 1025 EXPORT_SYMBOL_GPL(virtio_transport_notify_send_pre_enqueue); 1026 1027 int virtio_transport_notify_send_post_enqueue(struct vsock_sock *vsk, 1028 ssize_t written, struct vsock_transport_send_notify_data *data) 1029 { 1030 return 0; 1031 } 1032 EXPORT_SYMBOL_GPL(virtio_transport_notify_send_post_enqueue); 1033 1034 u64 virtio_transport_stream_rcvhiwat(struct vsock_sock *vsk) 1035 { 1036 return vsk->buffer_size; 1037 } 1038 EXPORT_SYMBOL_GPL(virtio_transport_stream_rcvhiwat); 1039 1040 bool virtio_transport_stream_is_active(struct vsock_sock *vsk) 1041 { 1042 return true; 1043 } 1044 EXPORT_SYMBOL_GPL(virtio_transport_stream_is_active); 1045 1046 int virtio_transport_dgram_bind(struct vsock_sock *vsk, 1047 struct sockaddr_vm *addr) 1048 { 1049 return -EOPNOTSUPP; 1050 } 1051 EXPORT_SYMBOL_GPL(virtio_transport_dgram_bind); 1052 1053 bool virtio_transport_dgram_allow(struct vsock_sock *vsk, u32 cid, u32 port) 1054 { 1055 return false; 1056 } 1057 EXPORT_SYMBOL_GPL(virtio_transport_dgram_allow); 1058 1059 int virtio_transport_connect(struct vsock_sock *vsk) 1060 { 1061 struct virtio_vsock_pkt_info info = { 1062 .op = VIRTIO_VSOCK_OP_REQUEST, 1063 .vsk = vsk, 1064 .net = sock_net(sk_vsock(vsk)), 1065 }; 1066 1067 return virtio_transport_send_pkt_info(vsk, &info); 1068 } 1069 EXPORT_SYMBOL_GPL(virtio_transport_connect); 1070 1071 int virtio_transport_shutdown(struct vsock_sock *vsk, int mode) 1072 { 1073 struct virtio_vsock_pkt_info info = { 1074 .op = VIRTIO_VSOCK_OP_SHUTDOWN, 1075 .flags = (mode & RCV_SHUTDOWN ? 1076 VIRTIO_VSOCK_SHUTDOWN_RCV : 0) | 1077 (mode & SEND_SHUTDOWN ? 1078 VIRTIO_VSOCK_SHUTDOWN_SEND : 0), 1079 .vsk = vsk, 1080 .net = sock_net(sk_vsock(vsk)), 1081 }; 1082 1083 return virtio_transport_send_pkt_info(vsk, &info); 1084 } 1085 EXPORT_SYMBOL_GPL(virtio_transport_shutdown); 1086 1087 int 1088 virtio_transport_dgram_enqueue(struct vsock_sock *vsk, 1089 struct sockaddr_vm *remote_addr, 1090 struct msghdr *msg, 1091 size_t dgram_len) 1092 { 1093 return -EOPNOTSUPP; 1094 } 1095 EXPORT_SYMBOL_GPL(virtio_transport_dgram_enqueue); 1096 1097 ssize_t 1098 virtio_transport_stream_enqueue(struct vsock_sock *vsk, 1099 struct msghdr *msg, 1100 size_t len) 1101 { 1102 struct virtio_vsock_pkt_info info = { 1103 .op = VIRTIO_VSOCK_OP_RW, 1104 .msg = msg, 1105 .pkt_len = len, 1106 .vsk = vsk, 1107 .net = sock_net(sk_vsock(vsk)), 1108 }; 1109 1110 return virtio_transport_send_pkt_info(vsk, &info); 1111 } 1112 EXPORT_SYMBOL_GPL(virtio_transport_stream_enqueue); 1113 1114 void virtio_transport_destruct(struct vsock_sock *vsk) 1115 { 1116 struct virtio_vsock_sock *vvs = vsk->trans; 1117 1118 virtio_transport_cancel_close_work(vsk, true); 1119 1120 kfree(vvs); 1121 vsk->trans = NULL; 1122 } 1123 EXPORT_SYMBOL_GPL(virtio_transport_destruct); 1124 1125 ssize_t virtio_transport_unsent_bytes(struct vsock_sock *vsk) 1126 { 1127 struct virtio_vsock_sock *vvs = vsk->trans; 1128 size_t ret; 1129 1130 spin_lock_bh(&vvs->tx_lock); 1131 ret = vvs->bytes_unsent; 1132 spin_unlock_bh(&vvs->tx_lock); 1133 1134 return ret; 1135 } 1136 EXPORT_SYMBOL_GPL(virtio_transport_unsent_bytes); 1137 1138 static int virtio_transport_reset(struct vsock_sock *vsk, 1139 struct sk_buff *skb) 1140 { 1141 struct virtio_vsock_pkt_info info = { 1142 .op = VIRTIO_VSOCK_OP_RST, 1143 .reply = !!skb, 1144 .vsk = vsk, 1145 .net = sock_net(sk_vsock(vsk)), 1146 }; 1147 1148 /* Send RST only if the original pkt is not a RST pkt */ 1149 if (skb && le16_to_cpu(virtio_vsock_hdr(skb)->op) == VIRTIO_VSOCK_OP_RST) 1150 return 0; 1151 1152 return virtio_transport_send_pkt_info(vsk, &info); 1153 } 1154 1155 /* Normally packets are associated with a socket. There may be no socket if an 1156 * attempt was made to connect to a socket that does not exist. 1157 * 1158 * net refers to the namespace of whoever sent the invalid message. For 1159 * loopback, this is the namespace of the socket. For vhost, this is the 1160 * namespace of the VM (i.e., vhost_vsock). 1161 */ 1162 static int virtio_transport_reset_no_sock(const struct virtio_transport *t, 1163 struct sk_buff *skb, struct net *net) 1164 { 1165 struct virtio_vsock_hdr *hdr = virtio_vsock_hdr(skb); 1166 struct virtio_vsock_pkt_info info = { 1167 .op = VIRTIO_VSOCK_OP_RST, 1168 .type = le16_to_cpu(hdr->type), 1169 .reply = true, 1170 1171 /* Set sk owner to socket we are replying to (may be NULL for 1172 * non-loopback). This keeps a reference to the sock and 1173 * sock_net(sk) until the reply skb is freed. 1174 */ 1175 .vsk = vsock_sk(skb->sk), 1176 1177 /* net is not defined here because we pass it directly to 1178 * t->send_pkt(), instead of relying on 1179 * virtio_transport_send_pkt_info() to pass it. It is not needed 1180 * by virtio_transport_alloc_skb(). 1181 */ 1182 }; 1183 struct sk_buff *reply; 1184 1185 /* Send RST only if the original pkt is not a RST pkt */ 1186 if (le16_to_cpu(hdr->op) == VIRTIO_VSOCK_OP_RST) 1187 return 0; 1188 1189 if (!t) 1190 return -ENOTCONN; 1191 1192 reply = virtio_transport_alloc_skb(&info, 0, false, NULL, 1193 le64_to_cpu(hdr->dst_cid), 1194 le32_to_cpu(hdr->dst_port), 1195 le64_to_cpu(hdr->src_cid), 1196 le32_to_cpu(hdr->src_port)); 1197 if (!reply) 1198 return -ENOMEM; 1199 1200 return t->send_pkt(reply, net); 1201 } 1202 1203 /* This function should be called with sk_lock held and SOCK_DONE set */ 1204 static void virtio_transport_remove_sock(struct vsock_sock *vsk) 1205 { 1206 struct virtio_vsock_sock *vvs = vsk->trans; 1207 1208 /* We don't need to take rx_lock, as the socket is closing and we are 1209 * removing it. 1210 */ 1211 __skb_queue_purge(&vvs->rx_queue); 1212 vsock_remove_sock(vsk); 1213 } 1214 1215 static void virtio_transport_cancel_close_work(struct vsock_sock *vsk, 1216 bool cancel_timeout) 1217 { 1218 struct sock *sk = sk_vsock(vsk); 1219 1220 if (vsk->close_work_scheduled && 1221 (!cancel_timeout || cancel_delayed_work(&vsk->close_work))) { 1222 vsk->close_work_scheduled = false; 1223 1224 virtio_transport_remove_sock(vsk); 1225 1226 /* Release refcnt obtained when we scheduled the timeout */ 1227 sock_put(sk); 1228 } 1229 } 1230 1231 static void virtio_transport_do_close(struct vsock_sock *vsk, 1232 bool cancel_timeout) 1233 { 1234 struct sock *sk = sk_vsock(vsk); 1235 1236 sock_set_flag(sk, SOCK_DONE); 1237 WRITE_ONCE(vsk->peer_shutdown, SHUTDOWN_MASK); 1238 if (vsock_stream_has_data(vsk) <= 0) 1239 sk->sk_state = TCP_CLOSING; 1240 sk->sk_state_change(sk); 1241 1242 virtio_transport_cancel_close_work(vsk, cancel_timeout); 1243 } 1244 1245 static void virtio_transport_close_timeout(struct work_struct *work) 1246 { 1247 struct vsock_sock *vsk = 1248 container_of(work, struct vsock_sock, close_work.work); 1249 struct sock *sk = sk_vsock(vsk); 1250 1251 sock_hold(sk); 1252 lock_sock(sk); 1253 1254 if (!sock_flag(sk, SOCK_DONE)) { 1255 (void)virtio_transport_reset(vsk, NULL); 1256 1257 virtio_transport_do_close(vsk, false); 1258 } 1259 1260 vsk->close_work_scheduled = false; 1261 1262 release_sock(sk); 1263 sock_put(sk); 1264 } 1265 1266 /* User context, vsk->sk is locked */ 1267 static bool virtio_transport_close(struct vsock_sock *vsk) 1268 { 1269 struct sock *sk = &vsk->sk; 1270 1271 if (!(sk->sk_state == TCP_ESTABLISHED || 1272 sk->sk_state == TCP_CLOSING)) 1273 return true; 1274 1275 /* Already received SHUTDOWN from peer, reply with RST */ 1276 if ((vsk->peer_shutdown & SHUTDOWN_MASK) == SHUTDOWN_MASK) { 1277 (void)virtio_transport_reset(vsk, NULL); 1278 return true; 1279 } 1280 1281 if ((sk->sk_shutdown & SHUTDOWN_MASK) != SHUTDOWN_MASK) 1282 (void)virtio_transport_shutdown(vsk, SHUTDOWN_MASK); 1283 1284 if (!(current->flags & PF_EXITING)) 1285 vsock_linger(sk); 1286 1287 if (sock_flag(sk, SOCK_DONE)) { 1288 return true; 1289 } 1290 1291 sock_hold(sk); 1292 INIT_DELAYED_WORK(&vsk->close_work, 1293 virtio_transport_close_timeout); 1294 vsk->close_work_scheduled = true; 1295 schedule_delayed_work(&vsk->close_work, VSOCK_CLOSE_TIMEOUT); 1296 return false; 1297 } 1298 1299 void virtio_transport_release(struct vsock_sock *vsk) 1300 { 1301 struct sock *sk = &vsk->sk; 1302 bool remove_sock = true; 1303 1304 if (sk->sk_type == SOCK_STREAM || sk->sk_type == SOCK_SEQPACKET) 1305 remove_sock = virtio_transport_close(vsk); 1306 1307 if (remove_sock) { 1308 sock_set_flag(sk, SOCK_DONE); 1309 virtio_transport_remove_sock(vsk); 1310 } 1311 } 1312 EXPORT_SYMBOL_GPL(virtio_transport_release); 1313 1314 static int 1315 virtio_transport_recv_connecting(struct sock *sk, 1316 struct sk_buff *skb) 1317 { 1318 struct virtio_vsock_hdr *hdr = virtio_vsock_hdr(skb); 1319 struct vsock_sock *vsk = vsock_sk(sk); 1320 int skerr; 1321 int err; 1322 1323 switch (le16_to_cpu(hdr->op)) { 1324 case VIRTIO_VSOCK_OP_RESPONSE: 1325 sk->sk_state = TCP_ESTABLISHED; 1326 sk->sk_socket->state = SS_CONNECTED; 1327 vsock_insert_connected(vsk); 1328 sk->sk_state_change(sk); 1329 break; 1330 case VIRTIO_VSOCK_OP_INVALID: 1331 break; 1332 case VIRTIO_VSOCK_OP_RST: 1333 skerr = ECONNRESET; 1334 err = 0; 1335 goto destroy; 1336 default: 1337 skerr = EPROTO; 1338 err = -EINVAL; 1339 goto destroy; 1340 } 1341 return 0; 1342 1343 destroy: 1344 virtio_transport_reset(vsk, skb); 1345 sk->sk_state = TCP_CLOSE; 1346 sk->sk_err = skerr; 1347 sk_error_report(sk); 1348 return err; 1349 } 1350 1351 static bool 1352 virtio_transport_recv_enqueue(struct vsock_sock *vsk, 1353 struct sk_buff *skb) 1354 { 1355 struct virtio_vsock_sock *vvs = vsk->trans; 1356 bool can_enqueue, free_pkt = false; 1357 struct virtio_vsock_hdr *hdr; 1358 u32 len; 1359 1360 hdr = virtio_vsock_hdr(skb); 1361 len = le32_to_cpu(hdr->len); 1362 1363 spin_lock_bh(&vvs->rx_lock); 1364 1365 can_enqueue = virtio_transport_inc_rx_pkt(vvs, len); 1366 if (!can_enqueue) 1367 goto out; 1368 1369 if (le32_to_cpu(hdr->flags) & VIRTIO_VSOCK_SEQ_EOM) 1370 vvs->msg_count++; 1371 1372 /* Try to copy small packets into the buffer of last packet queued, 1373 * to avoid wasting memory queueing the entire buffer with a small 1374 * payload. Skip non-linear (e.g. zerocopy) skbs; these carry payload 1375 * in skb_shinfo. 1376 */ 1377 if (len <= GOOD_COPY_LEN && !skb_queue_empty(&vvs->rx_queue) && 1378 !skb_is_nonlinear(skb)) { 1379 struct virtio_vsock_hdr *last_hdr; 1380 struct sk_buff *last_skb; 1381 1382 last_skb = skb_peek_tail(&vvs->rx_queue); 1383 last_hdr = virtio_vsock_hdr(last_skb); 1384 1385 /* If there is space in the last packet queued, we copy the 1386 * new packet in its buffer. We avoid this if the last packet 1387 * queued has VIRTIO_VSOCK_SEQ_EOM set, because this is 1388 * delimiter of SEQPACKET message, so 'pkt' is the first packet 1389 * of a new message. 1390 */ 1391 if (skb->len < skb_tailroom(last_skb) && 1392 !(le32_to_cpu(last_hdr->flags) & VIRTIO_VSOCK_SEQ_EOM)) { 1393 memcpy(skb_put(last_skb, skb->len), skb->data, skb->len); 1394 free_pkt = true; 1395 last_hdr->flags |= hdr->flags; 1396 le32_add_cpu(&last_hdr->len, len); 1397 goto out; 1398 } 1399 } 1400 1401 __skb_queue_tail(&vvs->rx_queue, skb); 1402 1403 out: 1404 spin_unlock_bh(&vvs->rx_lock); 1405 if (free_pkt) 1406 kfree_skb(skb); 1407 1408 return can_enqueue; 1409 } 1410 1411 static int 1412 virtio_transport_recv_connected(struct sock *sk, 1413 struct sk_buff *skb) 1414 { 1415 struct virtio_vsock_hdr *hdr = virtio_vsock_hdr(skb); 1416 struct vsock_sock *vsk = vsock_sk(sk); 1417 int err = 0; 1418 1419 switch (le16_to_cpu(hdr->op)) { 1420 case VIRTIO_VSOCK_OP_RW: 1421 if (!virtio_transport_recv_enqueue(vsk, skb)) { 1422 /* There is no more space to queue the packet, so let's 1423 * close the connection; otherwise, we'll lose data. 1424 */ 1425 (void)virtio_transport_reset(vsk, skb); 1426 virtio_transport_do_close(vsk, true); 1427 sk->sk_err = ENOBUFS; 1428 sk_error_report(sk); 1429 vsock_remove_sock(vsk); 1430 break; 1431 } 1432 vsock_data_ready(sk); 1433 return err; 1434 case VIRTIO_VSOCK_OP_CREDIT_REQUEST: 1435 virtio_transport_send_credit_update(vsk); 1436 break; 1437 case VIRTIO_VSOCK_OP_CREDIT_UPDATE: 1438 sk->sk_write_space(sk); 1439 break; 1440 case VIRTIO_VSOCK_OP_SHUTDOWN: { 1441 u32 peer_shutdown = READ_ONCE(vsk->peer_shutdown); 1442 1443 if (le32_to_cpu(hdr->flags) & VIRTIO_VSOCK_SHUTDOWN_RCV) 1444 peer_shutdown |= RCV_SHUTDOWN; 1445 if (le32_to_cpu(hdr->flags) & VIRTIO_VSOCK_SHUTDOWN_SEND) 1446 peer_shutdown |= SEND_SHUTDOWN; 1447 WRITE_ONCE(vsk->peer_shutdown, peer_shutdown); 1448 if (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 } 1464 case VIRTIO_VSOCK_OP_RST: 1465 virtio_transport_do_close(vsk, true); 1466 break; 1467 default: 1468 err = -EINVAL; 1469 break; 1470 } 1471 1472 kfree_skb(skb); 1473 return err; 1474 } 1475 1476 static void 1477 virtio_transport_recv_disconnecting(struct sock *sk, 1478 struct sk_buff *skb) 1479 { 1480 struct virtio_vsock_hdr *hdr = virtio_vsock_hdr(skb); 1481 struct vsock_sock *vsk = vsock_sk(sk); 1482 1483 if (le16_to_cpu(hdr->op) == VIRTIO_VSOCK_OP_RST) 1484 virtio_transport_do_close(vsk, true); 1485 } 1486 1487 static int 1488 virtio_transport_send_response(struct vsock_sock *vsk, 1489 struct sk_buff *skb) 1490 { 1491 struct virtio_vsock_hdr *hdr = virtio_vsock_hdr(skb); 1492 struct virtio_vsock_pkt_info info = { 1493 .op = VIRTIO_VSOCK_OP_RESPONSE, 1494 .remote_cid = le64_to_cpu(hdr->src_cid), 1495 .remote_port = le32_to_cpu(hdr->src_port), 1496 .reply = true, 1497 .vsk = vsk, 1498 .net = sock_net(sk_vsock(vsk)), 1499 }; 1500 1501 return virtio_transport_send_pkt_info(vsk, &info); 1502 } 1503 1504 static bool virtio_transport_space_update(struct sock *sk, 1505 struct sk_buff *skb) 1506 { 1507 struct virtio_vsock_hdr *hdr = virtio_vsock_hdr(skb); 1508 struct vsock_sock *vsk = vsock_sk(sk); 1509 struct virtio_vsock_sock *vvs = vsk->trans; 1510 bool space_available; 1511 1512 /* Listener sockets are not associated with any transport, so we are 1513 * not able to take the state to see if there is space available in the 1514 * remote peer, but since they are only used to receive requests, we 1515 * can assume that there is always space available in the other peer. 1516 */ 1517 if (!vvs) 1518 return true; 1519 1520 /* buf_alloc and fwd_cnt is always included in the hdr */ 1521 spin_lock_bh(&vvs->tx_lock); 1522 vvs->peer_buf_alloc = le32_to_cpu(hdr->buf_alloc); 1523 vvs->peer_fwd_cnt = le32_to_cpu(hdr->fwd_cnt); 1524 space_available = virtio_transport_has_space(vvs); 1525 spin_unlock_bh(&vvs->tx_lock); 1526 return space_available; 1527 } 1528 1529 /* Handle server socket */ 1530 static int 1531 virtio_transport_recv_listen(struct sock *sk, struct sk_buff *skb, 1532 struct virtio_transport *t) 1533 { 1534 struct virtio_vsock_hdr *hdr = virtio_vsock_hdr(skb); 1535 struct vsock_sock *vsk = vsock_sk(sk); 1536 struct vsock_sock *vchild; 1537 struct sock *child; 1538 int ret; 1539 1540 if (le16_to_cpu(hdr->op) != VIRTIO_VSOCK_OP_REQUEST) { 1541 virtio_transport_reset_no_sock(t, skb, sock_net(sk)); 1542 return -EINVAL; 1543 } 1544 1545 if (sk_acceptq_is_full(sk)) { 1546 virtio_transport_reset_no_sock(t, skb, sock_net(sk)); 1547 return -ENOMEM; 1548 } 1549 1550 /* __vsock_release() might have already flushed accept_queue. 1551 * Subsequent enqueues would lead to a memory leak. 1552 */ 1553 if (sk->sk_shutdown == SHUTDOWN_MASK) { 1554 virtio_transport_reset_no_sock(t, skb, sock_net(sk)); 1555 return -ESHUTDOWN; 1556 } 1557 1558 child = vsock_create_connected(sk); 1559 if (!child) { 1560 virtio_transport_reset_no_sock(t, skb, sock_net(sk)); 1561 return -ENOMEM; 1562 } 1563 1564 lock_sock_nested(child, SINGLE_DEPTH_NESTING); 1565 1566 child->sk_state = TCP_ESTABLISHED; 1567 1568 vchild = vsock_sk(child); 1569 vsock_addr_init(&vchild->local_addr, le64_to_cpu(hdr->dst_cid), 1570 le32_to_cpu(hdr->dst_port)); 1571 vsock_addr_init(&vchild->remote_addr, le64_to_cpu(hdr->src_cid), 1572 le32_to_cpu(hdr->src_port)); 1573 1574 ret = vsock_assign_transport(vchild, vsk); 1575 /* Transport assigned (looking at remote_addr) must be the same 1576 * where we received the request. 1577 */ 1578 if (ret || vchild->transport != &t->transport) { 1579 release_sock(child); 1580 virtio_transport_reset_no_sock(t, skb, sock_net(sk)); 1581 sock_put(child); 1582 return ret; 1583 } 1584 1585 if (virtio_transport_space_update(child, skb)) 1586 child->sk_write_space(child); 1587 1588 vsock_insert_connected(vchild); 1589 vsock_enqueue_accept(sk, child); 1590 virtio_transport_send_response(vchild, skb); 1591 1592 release_sock(child); 1593 1594 sk->sk_data_ready(sk); 1595 return 0; 1596 } 1597 1598 static bool virtio_transport_valid_type(u16 type) 1599 { 1600 return (type == VIRTIO_VSOCK_TYPE_STREAM) || 1601 (type == VIRTIO_VSOCK_TYPE_SEQPACKET); 1602 } 1603 1604 /* We are under the virtio-vsock's vsock->rx_lock or vhost-vsock's vq->mutex 1605 * lock. 1606 */ 1607 void virtio_transport_recv_pkt(struct virtio_transport *t, 1608 struct sk_buff *skb, struct net *net) 1609 { 1610 struct virtio_vsock_hdr *hdr = virtio_vsock_hdr(skb); 1611 struct sockaddr_vm src, dst; 1612 struct vsock_sock *vsk; 1613 struct sock *sk; 1614 bool space_available; 1615 1616 vsock_addr_init(&src, le64_to_cpu(hdr->src_cid), 1617 le32_to_cpu(hdr->src_port)); 1618 vsock_addr_init(&dst, le64_to_cpu(hdr->dst_cid), 1619 le32_to_cpu(hdr->dst_port)); 1620 1621 trace_virtio_transport_recv_pkt(src.svm_cid, src.svm_port, 1622 dst.svm_cid, dst.svm_port, 1623 le32_to_cpu(hdr->len), 1624 le16_to_cpu(hdr->type), 1625 le16_to_cpu(hdr->op), 1626 le32_to_cpu(hdr->flags), 1627 le32_to_cpu(hdr->buf_alloc), 1628 le32_to_cpu(hdr->fwd_cnt)); 1629 1630 if (!virtio_transport_valid_type(le16_to_cpu(hdr->type))) { 1631 (void)virtio_transport_reset_no_sock(t, skb, net); 1632 goto free_pkt; 1633 } 1634 1635 /* The socket must be in connected or bound table 1636 * otherwise send reset back 1637 */ 1638 sk = vsock_find_connected_socket_net(&src, &dst, net); 1639 if (!sk) { 1640 sk = vsock_find_bound_socket_net(&dst, net); 1641 if (!sk) { 1642 (void)virtio_transport_reset_no_sock(t, skb, net); 1643 goto free_pkt; 1644 } 1645 } 1646 1647 if (virtio_transport_get_type(sk) != le16_to_cpu(hdr->type)) { 1648 (void)virtio_transport_reset_no_sock(t, skb, net); 1649 sock_put(sk); 1650 goto free_pkt; 1651 } 1652 1653 if (!skb_set_owner_sk_safe(skb, sk)) { 1654 WARN_ONCE(1, "receiving vsock socket has sk_refcnt == 0\n"); 1655 goto free_pkt; 1656 } 1657 1658 vsk = vsock_sk(sk); 1659 1660 lock_sock(sk); 1661 1662 /* Check if sk has been closed or assigned to another transport before 1663 * lock_sock (note: listener sockets are not assigned to any transport) 1664 */ 1665 if (sock_flag(sk, SOCK_DONE) || 1666 (sk->sk_state != TCP_LISTEN && vsk->transport != &t->transport)) { 1667 (void)virtio_transport_reset_no_sock(t, skb, net); 1668 release_sock(sk); 1669 sock_put(sk); 1670 goto free_pkt; 1671 } 1672 1673 space_available = virtio_transport_space_update(sk, skb); 1674 1675 /* Update CID in case it has changed after a transport reset event */ 1676 if (vsk->local_addr.svm_cid != VMADDR_CID_ANY) 1677 vsk->local_addr.svm_cid = dst.svm_cid; 1678 1679 if (space_available) 1680 sk->sk_write_space(sk); 1681 1682 switch (sk->sk_state) { 1683 case TCP_LISTEN: 1684 virtio_transport_recv_listen(sk, skb, t); 1685 kfree_skb(skb); 1686 break; 1687 case TCP_SYN_SENT: 1688 virtio_transport_recv_connecting(sk, skb); 1689 kfree_skb(skb); 1690 break; 1691 case TCP_ESTABLISHED: 1692 virtio_transport_recv_connected(sk, skb); 1693 break; 1694 case TCP_CLOSING: 1695 virtio_transport_recv_disconnecting(sk, skb); 1696 kfree_skb(skb); 1697 break; 1698 default: 1699 (void)virtio_transport_reset_no_sock(t, skb, net); 1700 kfree_skb(skb); 1701 break; 1702 } 1703 1704 release_sock(sk); 1705 1706 /* Release refcnt obtained when we fetched this socket out of the 1707 * bound or connected list. 1708 */ 1709 sock_put(sk); 1710 return; 1711 1712 free_pkt: 1713 kfree_skb(skb); 1714 } 1715 EXPORT_SYMBOL_GPL(virtio_transport_recv_pkt); 1716 1717 /* Remove skbs found in a queue that have a vsk that matches. 1718 * 1719 * Each skb is freed. 1720 * 1721 * Returns the count of skbs that were reply packets. 1722 */ 1723 int virtio_transport_purge_skbs(void *vsk, struct sk_buff_head *queue) 1724 { 1725 struct sk_buff_head freeme; 1726 struct sk_buff *skb, *tmp; 1727 int cnt = 0; 1728 1729 skb_queue_head_init(&freeme); 1730 1731 spin_lock_bh(&queue->lock); 1732 skb_queue_walk_safe(queue, skb, tmp) { 1733 if (vsock_sk(skb->sk) != vsk) 1734 continue; 1735 1736 __skb_unlink(skb, queue); 1737 __skb_queue_tail(&freeme, skb); 1738 1739 if (virtio_vsock_skb_reply(skb)) 1740 cnt++; 1741 } 1742 spin_unlock_bh(&queue->lock); 1743 1744 __skb_queue_purge(&freeme); 1745 1746 return cnt; 1747 } 1748 EXPORT_SYMBOL_GPL(virtio_transport_purge_skbs); 1749 1750 int virtio_transport_read_skb(struct vsock_sock *vsk, skb_read_actor_t recv_actor) 1751 { 1752 struct virtio_vsock_sock *vvs = vsk->trans; 1753 struct sock *sk = sk_vsock(vsk); 1754 struct virtio_vsock_hdr *hdr; 1755 struct sk_buff *skb; 1756 u32 pkt_len; 1757 int off = 0; 1758 int err; 1759 1760 spin_lock_bh(&vvs->rx_lock); 1761 /* Use __skb_recv_datagram() for race-free handling of the receive. It 1762 * works for types other than dgrams. 1763 */ 1764 skb = __skb_recv_datagram(sk, &vvs->rx_queue, MSG_DONTWAIT, &off, &err); 1765 if (!skb) { 1766 spin_unlock_bh(&vvs->rx_lock); 1767 return err; 1768 } 1769 1770 hdr = virtio_vsock_hdr(skb); 1771 if (le32_to_cpu(hdr->flags) & VIRTIO_VSOCK_SEQ_EOM) 1772 vvs->msg_count--; 1773 1774 pkt_len = le32_to_cpu(hdr->len); 1775 virtio_transport_dec_rx_pkt(vvs, pkt_len, pkt_len); 1776 spin_unlock_bh(&vvs->rx_lock); 1777 1778 virtio_transport_send_credit_update(vsk); 1779 1780 return recv_actor(sk, skb); 1781 } 1782 EXPORT_SYMBOL_GPL(virtio_transport_read_skb); 1783 1784 int virtio_transport_notify_set_rcvlowat(struct vsock_sock *vsk, int val) 1785 { 1786 struct virtio_vsock_sock *vvs = vsk->trans; 1787 bool send_update; 1788 1789 spin_lock_bh(&vvs->rx_lock); 1790 1791 /* If number of available bytes is less than new SO_RCVLOWAT value, 1792 * kick sender to send more data, because sender may sleep in its 1793 * 'send()' syscall waiting for enough space at our side. Also 1794 * don't send credit update when peer already knows actual value - 1795 * such transmission will be useless. 1796 */ 1797 send_update = (vvs->rx_bytes < val) && 1798 (vvs->fwd_cnt != vvs->last_fwd_cnt); 1799 1800 spin_unlock_bh(&vvs->rx_lock); 1801 1802 if (send_update) { 1803 int err; 1804 1805 err = virtio_transport_send_credit_update(vsk); 1806 if (err < 0) 1807 return err; 1808 } 1809 1810 return 0; 1811 } 1812 EXPORT_SYMBOL_GPL(virtio_transport_notify_set_rcvlowat); 1813 1814 MODULE_LICENSE("GPL v2"); 1815 MODULE_AUTHOR("Asias He"); 1816 MODULE_DESCRIPTION("common code for virtio vsock"); 1817