1 /* Copyright (C) 2009 Red Hat, Inc. 2 * Author: Michael S. Tsirkin <mst@redhat.com> 3 * 4 * This work is licensed under the terms of the GNU GPL, version 2. 5 * 6 * virtio-net server in host kernel. 7 */ 8 9 #include <linux/compat.h> 10 #include <linux/eventfd.h> 11 #include <linux/vhost.h> 12 #include <linux/virtio_net.h> 13 #include <linux/miscdevice.h> 14 #include <linux/module.h> 15 #include <linux/moduleparam.h> 16 #include <linux/mutex.h> 17 #include <linux/workqueue.h> 18 #include <linux/file.h> 19 #include <linux/slab.h> 20 #include <linux/sched/clock.h> 21 #include <linux/sched/signal.h> 22 #include <linux/vmalloc.h> 23 24 #include <linux/net.h> 25 #include <linux/if_packet.h> 26 #include <linux/if_arp.h> 27 #include <linux/if_tun.h> 28 #include <linux/if_macvlan.h> 29 #include <linux/if_tap.h> 30 #include <linux/if_vlan.h> 31 #include <linux/skb_array.h> 32 #include <linux/skbuff.h> 33 34 #include <net/sock.h> 35 36 #include "vhost.h" 37 38 static int experimental_zcopytx = 1; 39 module_param(experimental_zcopytx, int, 0444); 40 MODULE_PARM_DESC(experimental_zcopytx, "Enable Zero Copy TX;" 41 " 1 -Enable; 0 - Disable"); 42 43 /* Max number of bytes transferred before requeueing the job. 44 * Using this limit prevents one virtqueue from starving others. */ 45 #define VHOST_NET_WEIGHT 0x80000 46 47 /* MAX number of TX used buffers for outstanding zerocopy */ 48 #define VHOST_MAX_PEND 128 49 #define VHOST_GOODCOPY_LEN 256 50 51 /* 52 * For transmit, used buffer len is unused; we override it to track buffer 53 * status internally; used for zerocopy tx only. 54 */ 55 /* Lower device DMA failed */ 56 #define VHOST_DMA_FAILED_LEN ((__force __virtio32)3) 57 /* Lower device DMA done */ 58 #define VHOST_DMA_DONE_LEN ((__force __virtio32)2) 59 /* Lower device DMA in progress */ 60 #define VHOST_DMA_IN_PROGRESS ((__force __virtio32)1) 61 /* Buffer unused */ 62 #define VHOST_DMA_CLEAR_LEN ((__force __virtio32)0) 63 64 #define VHOST_DMA_IS_DONE(len) ((__force u32)(len) >= (__force u32)VHOST_DMA_DONE_LEN) 65 66 enum { 67 VHOST_NET_FEATURES = VHOST_FEATURES | 68 (1ULL << VHOST_NET_F_VIRTIO_NET_HDR) | 69 (1ULL << VIRTIO_NET_F_MRG_RXBUF) | 70 (1ULL << VIRTIO_F_IOMMU_PLATFORM) 71 }; 72 73 enum { 74 VHOST_NET_VQ_RX = 0, 75 VHOST_NET_VQ_TX = 1, 76 VHOST_NET_VQ_MAX = 2, 77 }; 78 79 struct vhost_net_ubuf_ref { 80 /* refcount follows semantics similar to kref: 81 * 0: object is released 82 * 1: no outstanding ubufs 83 * >1: outstanding ubufs 84 */ 85 atomic_t refcount; 86 wait_queue_head_t wait; 87 struct vhost_virtqueue *vq; 88 }; 89 90 #define VHOST_RX_BATCH 64 91 struct vhost_net_buf { 92 struct sk_buff **queue; 93 int tail; 94 int head; 95 }; 96 97 struct vhost_net_virtqueue { 98 struct vhost_virtqueue vq; 99 size_t vhost_hlen; 100 size_t sock_hlen; 101 /* vhost zerocopy support fields below: */ 102 /* last used idx for outstanding DMA zerocopy buffers */ 103 int upend_idx; 104 /* first used idx for DMA done zerocopy buffers */ 105 int done_idx; 106 /* an array of userspace buffers info */ 107 struct ubuf_info *ubuf_info; 108 /* Reference counting for outstanding ubufs. 109 * Protected by vq mutex. Writers must also take device mutex. */ 110 struct vhost_net_ubuf_ref *ubufs; 111 struct skb_array *rx_array; 112 struct vhost_net_buf rxq; 113 }; 114 115 struct vhost_net { 116 struct vhost_dev dev; 117 struct vhost_net_virtqueue vqs[VHOST_NET_VQ_MAX]; 118 struct vhost_poll poll[VHOST_NET_VQ_MAX]; 119 /* Number of TX recently submitted. 120 * Protected by tx vq lock. */ 121 unsigned tx_packets; 122 /* Number of times zerocopy TX recently failed. 123 * Protected by tx vq lock. */ 124 unsigned tx_zcopy_err; 125 /* Flush in progress. Protected by tx vq lock. */ 126 bool tx_flush; 127 }; 128 129 static unsigned vhost_net_zcopy_mask __read_mostly; 130 131 static void *vhost_net_buf_get_ptr(struct vhost_net_buf *rxq) 132 { 133 if (rxq->tail != rxq->head) 134 return rxq->queue[rxq->head]; 135 else 136 return NULL; 137 } 138 139 static int vhost_net_buf_get_size(struct vhost_net_buf *rxq) 140 { 141 return rxq->tail - rxq->head; 142 } 143 144 static int vhost_net_buf_is_empty(struct vhost_net_buf *rxq) 145 { 146 return rxq->tail == rxq->head; 147 } 148 149 static void *vhost_net_buf_consume(struct vhost_net_buf *rxq) 150 { 151 void *ret = vhost_net_buf_get_ptr(rxq); 152 ++rxq->head; 153 return ret; 154 } 155 156 static int vhost_net_buf_produce(struct vhost_net_virtqueue *nvq) 157 { 158 struct vhost_net_buf *rxq = &nvq->rxq; 159 160 rxq->head = 0; 161 rxq->tail = skb_array_consume_batched(nvq->rx_array, rxq->queue, 162 VHOST_RX_BATCH); 163 return rxq->tail; 164 } 165 166 static void vhost_net_buf_unproduce(struct vhost_net_virtqueue *nvq) 167 { 168 struct vhost_net_buf *rxq = &nvq->rxq; 169 170 if (nvq->rx_array && !vhost_net_buf_is_empty(rxq)) { 171 skb_array_unconsume(nvq->rx_array, rxq->queue + rxq->head, 172 vhost_net_buf_get_size(rxq)); 173 rxq->head = rxq->tail = 0; 174 } 175 } 176 177 static int vhost_net_buf_peek(struct vhost_net_virtqueue *nvq) 178 { 179 struct vhost_net_buf *rxq = &nvq->rxq; 180 181 if (!vhost_net_buf_is_empty(rxq)) 182 goto out; 183 184 if (!vhost_net_buf_produce(nvq)) 185 return 0; 186 187 out: 188 return __skb_array_len_with_tag(vhost_net_buf_get_ptr(rxq)); 189 } 190 191 static void vhost_net_buf_init(struct vhost_net_buf *rxq) 192 { 193 rxq->head = rxq->tail = 0; 194 } 195 196 static void vhost_net_enable_zcopy(int vq) 197 { 198 vhost_net_zcopy_mask |= 0x1 << vq; 199 } 200 201 static struct vhost_net_ubuf_ref * 202 vhost_net_ubuf_alloc(struct vhost_virtqueue *vq, bool zcopy) 203 { 204 struct vhost_net_ubuf_ref *ubufs; 205 /* No zero copy backend? Nothing to count. */ 206 if (!zcopy) 207 return NULL; 208 ubufs = kmalloc(sizeof(*ubufs), GFP_KERNEL); 209 if (!ubufs) 210 return ERR_PTR(-ENOMEM); 211 atomic_set(&ubufs->refcount, 1); 212 init_waitqueue_head(&ubufs->wait); 213 ubufs->vq = vq; 214 return ubufs; 215 } 216 217 static int vhost_net_ubuf_put(struct vhost_net_ubuf_ref *ubufs) 218 { 219 int r = atomic_sub_return(1, &ubufs->refcount); 220 if (unlikely(!r)) 221 wake_up(&ubufs->wait); 222 return r; 223 } 224 225 static void vhost_net_ubuf_put_and_wait(struct vhost_net_ubuf_ref *ubufs) 226 { 227 vhost_net_ubuf_put(ubufs); 228 wait_event(ubufs->wait, !atomic_read(&ubufs->refcount)); 229 } 230 231 static void vhost_net_ubuf_put_wait_and_free(struct vhost_net_ubuf_ref *ubufs) 232 { 233 vhost_net_ubuf_put_and_wait(ubufs); 234 kfree(ubufs); 235 } 236 237 static void vhost_net_clear_ubuf_info(struct vhost_net *n) 238 { 239 int i; 240 241 for (i = 0; i < VHOST_NET_VQ_MAX; ++i) { 242 kfree(n->vqs[i].ubuf_info); 243 n->vqs[i].ubuf_info = NULL; 244 } 245 } 246 247 static int vhost_net_set_ubuf_info(struct vhost_net *n) 248 { 249 bool zcopy; 250 int i; 251 252 for (i = 0; i < VHOST_NET_VQ_MAX; ++i) { 253 zcopy = vhost_net_zcopy_mask & (0x1 << i); 254 if (!zcopy) 255 continue; 256 n->vqs[i].ubuf_info = kmalloc(sizeof(*n->vqs[i].ubuf_info) * 257 UIO_MAXIOV, GFP_KERNEL); 258 if (!n->vqs[i].ubuf_info) 259 goto err; 260 } 261 return 0; 262 263 err: 264 vhost_net_clear_ubuf_info(n); 265 return -ENOMEM; 266 } 267 268 static void vhost_net_vq_reset(struct vhost_net *n) 269 { 270 int i; 271 272 vhost_net_clear_ubuf_info(n); 273 274 for (i = 0; i < VHOST_NET_VQ_MAX; i++) { 275 n->vqs[i].done_idx = 0; 276 n->vqs[i].upend_idx = 0; 277 n->vqs[i].ubufs = NULL; 278 n->vqs[i].vhost_hlen = 0; 279 n->vqs[i].sock_hlen = 0; 280 vhost_net_buf_init(&n->vqs[i].rxq); 281 } 282 283 } 284 285 static void vhost_net_tx_packet(struct vhost_net *net) 286 { 287 ++net->tx_packets; 288 if (net->tx_packets < 1024) 289 return; 290 net->tx_packets = 0; 291 net->tx_zcopy_err = 0; 292 } 293 294 static void vhost_net_tx_err(struct vhost_net *net) 295 { 296 ++net->tx_zcopy_err; 297 } 298 299 static bool vhost_net_tx_select_zcopy(struct vhost_net *net) 300 { 301 /* TX flush waits for outstanding DMAs to be done. 302 * Don't start new DMAs. 303 */ 304 return !net->tx_flush && 305 net->tx_packets / 64 >= net->tx_zcopy_err; 306 } 307 308 static bool vhost_sock_zcopy(struct socket *sock) 309 { 310 return unlikely(experimental_zcopytx) && 311 sock_flag(sock->sk, SOCK_ZEROCOPY); 312 } 313 314 /* In case of DMA done not in order in lower device driver for some reason. 315 * upend_idx is used to track end of used idx, done_idx is used to track head 316 * of used idx. Once lower device DMA done contiguously, we will signal KVM 317 * guest used idx. 318 */ 319 static void vhost_zerocopy_signal_used(struct vhost_net *net, 320 struct vhost_virtqueue *vq) 321 { 322 struct vhost_net_virtqueue *nvq = 323 container_of(vq, struct vhost_net_virtqueue, vq); 324 int i, add; 325 int j = 0; 326 327 for (i = nvq->done_idx; i != nvq->upend_idx; i = (i + 1) % UIO_MAXIOV) { 328 if (vq->heads[i].len == VHOST_DMA_FAILED_LEN) 329 vhost_net_tx_err(net); 330 if (VHOST_DMA_IS_DONE(vq->heads[i].len)) { 331 vq->heads[i].len = VHOST_DMA_CLEAR_LEN; 332 ++j; 333 } else 334 break; 335 } 336 while (j) { 337 add = min(UIO_MAXIOV - nvq->done_idx, j); 338 vhost_add_used_and_signal_n(vq->dev, vq, 339 &vq->heads[nvq->done_idx], add); 340 nvq->done_idx = (nvq->done_idx + add) % UIO_MAXIOV; 341 j -= add; 342 } 343 } 344 345 static void vhost_zerocopy_callback(struct ubuf_info *ubuf, bool success) 346 { 347 struct vhost_net_ubuf_ref *ubufs = ubuf->ctx; 348 struct vhost_virtqueue *vq = ubufs->vq; 349 int cnt; 350 351 rcu_read_lock_bh(); 352 353 /* set len to mark this desc buffers done DMA */ 354 vq->heads[ubuf->desc].len = success ? 355 VHOST_DMA_DONE_LEN : VHOST_DMA_FAILED_LEN; 356 cnt = vhost_net_ubuf_put(ubufs); 357 358 /* 359 * Trigger polling thread if guest stopped submitting new buffers: 360 * in this case, the refcount after decrement will eventually reach 1. 361 * We also trigger polling periodically after each 16 packets 362 * (the value 16 here is more or less arbitrary, it's tuned to trigger 363 * less than 10% of times). 364 */ 365 if (cnt <= 1 || !(cnt % 16)) 366 vhost_poll_queue(&vq->poll); 367 368 rcu_read_unlock_bh(); 369 } 370 371 static inline unsigned long busy_clock(void) 372 { 373 return local_clock() >> 10; 374 } 375 376 static bool vhost_can_busy_poll(struct vhost_dev *dev, 377 unsigned long endtime) 378 { 379 return likely(!need_resched()) && 380 likely(!time_after(busy_clock(), endtime)) && 381 likely(!signal_pending(current)) && 382 !vhost_has_work(dev); 383 } 384 385 static void vhost_net_disable_vq(struct vhost_net *n, 386 struct vhost_virtqueue *vq) 387 { 388 struct vhost_net_virtqueue *nvq = 389 container_of(vq, struct vhost_net_virtqueue, vq); 390 struct vhost_poll *poll = n->poll + (nvq - n->vqs); 391 if (!vq->private_data) 392 return; 393 vhost_poll_stop(poll); 394 } 395 396 static int vhost_net_enable_vq(struct vhost_net *n, 397 struct vhost_virtqueue *vq) 398 { 399 struct vhost_net_virtqueue *nvq = 400 container_of(vq, struct vhost_net_virtqueue, vq); 401 struct vhost_poll *poll = n->poll + (nvq - n->vqs); 402 struct socket *sock; 403 404 sock = vq->private_data; 405 if (!sock) 406 return 0; 407 408 return vhost_poll_start(poll, sock->file); 409 } 410 411 static int vhost_net_tx_get_vq_desc(struct vhost_net *net, 412 struct vhost_virtqueue *vq, 413 struct iovec iov[], unsigned int iov_size, 414 unsigned int *out_num, unsigned int *in_num) 415 { 416 unsigned long uninitialized_var(endtime); 417 int r = vhost_get_vq_desc(vq, vq->iov, ARRAY_SIZE(vq->iov), 418 out_num, in_num, NULL, NULL); 419 420 if (r == vq->num && vq->busyloop_timeout) { 421 preempt_disable(); 422 endtime = busy_clock() + vq->busyloop_timeout; 423 while (vhost_can_busy_poll(vq->dev, endtime) && 424 vhost_vq_avail_empty(vq->dev, vq)) 425 cpu_relax(); 426 preempt_enable(); 427 r = vhost_get_vq_desc(vq, vq->iov, ARRAY_SIZE(vq->iov), 428 out_num, in_num, NULL, NULL); 429 } 430 431 return r; 432 } 433 434 static bool vhost_exceeds_maxpend(struct vhost_net *net) 435 { 436 struct vhost_net_virtqueue *nvq = &net->vqs[VHOST_NET_VQ_TX]; 437 struct vhost_virtqueue *vq = &nvq->vq; 438 439 return (nvq->upend_idx + vq->num - VHOST_MAX_PEND) % UIO_MAXIOV 440 == nvq->done_idx; 441 } 442 443 /* Expects to be always run from workqueue - which acts as 444 * read-size critical section for our kind of RCU. */ 445 static void handle_tx(struct vhost_net *net) 446 { 447 struct vhost_net_virtqueue *nvq = &net->vqs[VHOST_NET_VQ_TX]; 448 struct vhost_virtqueue *vq = &nvq->vq; 449 unsigned out, in; 450 int head; 451 struct msghdr msg = { 452 .msg_name = NULL, 453 .msg_namelen = 0, 454 .msg_control = NULL, 455 .msg_controllen = 0, 456 .msg_flags = MSG_DONTWAIT, 457 }; 458 size_t len, total_len = 0; 459 int err; 460 size_t hdr_size; 461 struct socket *sock; 462 struct vhost_net_ubuf_ref *uninitialized_var(ubufs); 463 bool zcopy, zcopy_used; 464 465 mutex_lock(&vq->mutex); 466 sock = vq->private_data; 467 if (!sock) 468 goto out; 469 470 if (!vq_iotlb_prefetch(vq)) 471 goto out; 472 473 vhost_disable_notify(&net->dev, vq); 474 475 hdr_size = nvq->vhost_hlen; 476 zcopy = nvq->ubufs; 477 478 for (;;) { 479 /* Release DMAs done buffers first */ 480 if (zcopy) 481 vhost_zerocopy_signal_used(net, vq); 482 483 /* If more outstanding DMAs, queue the work. 484 * Handle upend_idx wrap around 485 */ 486 if (unlikely(vhost_exceeds_maxpend(net))) 487 break; 488 489 head = vhost_net_tx_get_vq_desc(net, vq, vq->iov, 490 ARRAY_SIZE(vq->iov), 491 &out, &in); 492 /* On error, stop handling until the next kick. */ 493 if (unlikely(head < 0)) 494 break; 495 /* Nothing new? Wait for eventfd to tell us they refilled. */ 496 if (head == vq->num) { 497 if (unlikely(vhost_enable_notify(&net->dev, vq))) { 498 vhost_disable_notify(&net->dev, vq); 499 continue; 500 } 501 break; 502 } 503 if (in) { 504 vq_err(vq, "Unexpected descriptor format for TX: " 505 "out %d, int %d\n", out, in); 506 break; 507 } 508 /* Skip header. TODO: support TSO. */ 509 len = iov_length(vq->iov, out); 510 iov_iter_init(&msg.msg_iter, WRITE, vq->iov, out, len); 511 iov_iter_advance(&msg.msg_iter, hdr_size); 512 /* Sanity check */ 513 if (!msg_data_left(&msg)) { 514 vq_err(vq, "Unexpected header len for TX: " 515 "%zd expected %zd\n", 516 len, hdr_size); 517 break; 518 } 519 len = msg_data_left(&msg); 520 521 zcopy_used = zcopy && len >= VHOST_GOODCOPY_LEN 522 && (nvq->upend_idx + 1) % UIO_MAXIOV != 523 nvq->done_idx 524 && vhost_net_tx_select_zcopy(net); 525 526 /* use msg_control to pass vhost zerocopy ubuf info to skb */ 527 if (zcopy_used) { 528 struct ubuf_info *ubuf; 529 ubuf = nvq->ubuf_info + nvq->upend_idx; 530 531 vq->heads[nvq->upend_idx].id = cpu_to_vhost32(vq, head); 532 vq->heads[nvq->upend_idx].len = VHOST_DMA_IN_PROGRESS; 533 ubuf->callback = vhost_zerocopy_callback; 534 ubuf->ctx = nvq->ubufs; 535 ubuf->desc = nvq->upend_idx; 536 atomic_set(&ubuf->refcnt, 1); 537 msg.msg_control = ubuf; 538 msg.msg_controllen = sizeof(ubuf); 539 ubufs = nvq->ubufs; 540 atomic_inc(&ubufs->refcount); 541 nvq->upend_idx = (nvq->upend_idx + 1) % UIO_MAXIOV; 542 } else { 543 msg.msg_control = NULL; 544 ubufs = NULL; 545 } 546 547 total_len += len; 548 if (total_len < VHOST_NET_WEIGHT && 549 !vhost_vq_avail_empty(&net->dev, vq) && 550 likely(!vhost_exceeds_maxpend(net))) { 551 msg.msg_flags |= MSG_MORE; 552 } else { 553 msg.msg_flags &= ~MSG_MORE; 554 } 555 556 /* TODO: Check specific error and bomb out unless ENOBUFS? */ 557 err = sock->ops->sendmsg(sock, &msg, len); 558 if (unlikely(err < 0)) { 559 if (zcopy_used) { 560 vhost_net_ubuf_put(ubufs); 561 nvq->upend_idx = ((unsigned)nvq->upend_idx - 1) 562 % UIO_MAXIOV; 563 } 564 vhost_discard_vq_desc(vq, 1); 565 break; 566 } 567 if (err != len) 568 pr_debug("Truncated TX packet: " 569 " len %d != %zd\n", err, len); 570 if (!zcopy_used) 571 vhost_add_used_and_signal(&net->dev, vq, head, 0); 572 else 573 vhost_zerocopy_signal_used(net, vq); 574 vhost_net_tx_packet(net); 575 if (unlikely(total_len >= VHOST_NET_WEIGHT)) { 576 vhost_poll_queue(&vq->poll); 577 break; 578 } 579 } 580 out: 581 mutex_unlock(&vq->mutex); 582 } 583 584 static int peek_head_len(struct vhost_net_virtqueue *rvq, struct sock *sk) 585 { 586 struct sk_buff *head; 587 int len = 0; 588 unsigned long flags; 589 590 if (rvq->rx_array) 591 return vhost_net_buf_peek(rvq); 592 593 spin_lock_irqsave(&sk->sk_receive_queue.lock, flags); 594 head = skb_peek(&sk->sk_receive_queue); 595 if (likely(head)) { 596 len = head->len; 597 if (skb_vlan_tag_present(head)) 598 len += VLAN_HLEN; 599 } 600 601 spin_unlock_irqrestore(&sk->sk_receive_queue.lock, flags); 602 return len; 603 } 604 605 static int sk_has_rx_data(struct sock *sk) 606 { 607 struct socket *sock = sk->sk_socket; 608 609 if (sock->ops->peek_len) 610 return sock->ops->peek_len(sock); 611 612 return skb_queue_empty(&sk->sk_receive_queue); 613 } 614 615 static int vhost_net_rx_peek_head_len(struct vhost_net *net, struct sock *sk) 616 { 617 struct vhost_net_virtqueue *rvq = &net->vqs[VHOST_NET_VQ_RX]; 618 struct vhost_net_virtqueue *nvq = &net->vqs[VHOST_NET_VQ_TX]; 619 struct vhost_virtqueue *vq = &nvq->vq; 620 unsigned long uninitialized_var(endtime); 621 int len = peek_head_len(rvq, sk); 622 623 if (!len && vq->busyloop_timeout) { 624 /* Both tx vq and rx socket were polled here */ 625 mutex_lock(&vq->mutex); 626 vhost_disable_notify(&net->dev, vq); 627 628 preempt_disable(); 629 endtime = busy_clock() + vq->busyloop_timeout; 630 631 while (vhost_can_busy_poll(&net->dev, endtime) && 632 !sk_has_rx_data(sk) && 633 vhost_vq_avail_empty(&net->dev, vq)) 634 cpu_relax(); 635 636 preempt_enable(); 637 638 if (vhost_enable_notify(&net->dev, vq)) 639 vhost_poll_queue(&vq->poll); 640 mutex_unlock(&vq->mutex); 641 642 len = peek_head_len(rvq, sk); 643 } 644 645 return len; 646 } 647 648 /* This is a multi-buffer version of vhost_get_desc, that works if 649 * vq has read descriptors only. 650 * @vq - the relevant virtqueue 651 * @datalen - data length we'll be reading 652 * @iovcount - returned count of io vectors we fill 653 * @log - vhost log 654 * @log_num - log offset 655 * @quota - headcount quota, 1 for big buffer 656 * returns number of buffer heads allocated, negative on error 657 */ 658 static int get_rx_bufs(struct vhost_virtqueue *vq, 659 struct vring_used_elem *heads, 660 int datalen, 661 unsigned *iovcount, 662 struct vhost_log *log, 663 unsigned *log_num, 664 unsigned int quota) 665 { 666 unsigned int out, in; 667 int seg = 0; 668 int headcount = 0; 669 unsigned d; 670 int r, nlogs = 0; 671 /* len is always initialized before use since we are always called with 672 * datalen > 0. 673 */ 674 u32 uninitialized_var(len); 675 676 while (datalen > 0 && headcount < quota) { 677 if (unlikely(seg >= UIO_MAXIOV)) { 678 r = -ENOBUFS; 679 goto err; 680 } 681 r = vhost_get_vq_desc(vq, vq->iov + seg, 682 ARRAY_SIZE(vq->iov) - seg, &out, 683 &in, log, log_num); 684 if (unlikely(r < 0)) 685 goto err; 686 687 d = r; 688 if (d == vq->num) { 689 r = 0; 690 goto err; 691 } 692 if (unlikely(out || in <= 0)) { 693 vq_err(vq, "unexpected descriptor format for RX: " 694 "out %d, in %d\n", out, in); 695 r = -EINVAL; 696 goto err; 697 } 698 if (unlikely(log)) { 699 nlogs += *log_num; 700 log += *log_num; 701 } 702 heads[headcount].id = cpu_to_vhost32(vq, d); 703 len = iov_length(vq->iov + seg, in); 704 heads[headcount].len = cpu_to_vhost32(vq, len); 705 datalen -= len; 706 ++headcount; 707 seg += in; 708 } 709 heads[headcount - 1].len = cpu_to_vhost32(vq, len + datalen); 710 *iovcount = seg; 711 if (unlikely(log)) 712 *log_num = nlogs; 713 714 /* Detect overrun */ 715 if (unlikely(datalen > 0)) { 716 r = UIO_MAXIOV + 1; 717 goto err; 718 } 719 return headcount; 720 err: 721 vhost_discard_vq_desc(vq, headcount); 722 return r; 723 } 724 725 /* Expects to be always run from workqueue - which acts as 726 * read-size critical section for our kind of RCU. */ 727 static void handle_rx(struct vhost_net *net) 728 { 729 struct vhost_net_virtqueue *nvq = &net->vqs[VHOST_NET_VQ_RX]; 730 struct vhost_virtqueue *vq = &nvq->vq; 731 unsigned uninitialized_var(in), log; 732 struct vhost_log *vq_log; 733 struct msghdr msg = { 734 .msg_name = NULL, 735 .msg_namelen = 0, 736 .msg_control = NULL, /* FIXME: get and handle RX aux data. */ 737 .msg_controllen = 0, 738 .msg_flags = MSG_DONTWAIT, 739 }; 740 struct virtio_net_hdr hdr = { 741 .flags = 0, 742 .gso_type = VIRTIO_NET_HDR_GSO_NONE 743 }; 744 size_t total_len = 0; 745 int err, mergeable; 746 s16 headcount; 747 size_t vhost_hlen, sock_hlen; 748 size_t vhost_len, sock_len; 749 struct socket *sock; 750 struct iov_iter fixup; 751 __virtio16 num_buffers; 752 753 mutex_lock(&vq->mutex); 754 sock = vq->private_data; 755 if (!sock) 756 goto out; 757 758 if (!vq_iotlb_prefetch(vq)) 759 goto out; 760 761 vhost_disable_notify(&net->dev, vq); 762 vhost_net_disable_vq(net, vq); 763 764 vhost_hlen = nvq->vhost_hlen; 765 sock_hlen = nvq->sock_hlen; 766 767 vq_log = unlikely(vhost_has_feature(vq, VHOST_F_LOG_ALL)) ? 768 vq->log : NULL; 769 mergeable = vhost_has_feature(vq, VIRTIO_NET_F_MRG_RXBUF); 770 771 while ((sock_len = vhost_net_rx_peek_head_len(net, sock->sk))) { 772 sock_len += sock_hlen; 773 vhost_len = sock_len + vhost_hlen; 774 headcount = get_rx_bufs(vq, vq->heads, vhost_len, 775 &in, vq_log, &log, 776 likely(mergeable) ? UIO_MAXIOV : 1); 777 /* On error, stop handling until the next kick. */ 778 if (unlikely(headcount < 0)) 779 goto out; 780 if (nvq->rx_array) 781 msg.msg_control = vhost_net_buf_consume(&nvq->rxq); 782 /* On overrun, truncate and discard */ 783 if (unlikely(headcount > UIO_MAXIOV)) { 784 iov_iter_init(&msg.msg_iter, READ, vq->iov, 1, 1); 785 err = sock->ops->recvmsg(sock, &msg, 786 1, MSG_DONTWAIT | MSG_TRUNC); 787 pr_debug("Discarded rx packet: len %zd\n", sock_len); 788 continue; 789 } 790 /* OK, now we need to know about added descriptors. */ 791 if (!headcount) { 792 if (unlikely(vhost_enable_notify(&net->dev, vq))) { 793 /* They have slipped one in as we were 794 * doing that: check again. */ 795 vhost_disable_notify(&net->dev, vq); 796 continue; 797 } 798 /* Nothing new? Wait for eventfd to tell us 799 * they refilled. */ 800 goto out; 801 } 802 /* We don't need to be notified again. */ 803 iov_iter_init(&msg.msg_iter, READ, vq->iov, in, vhost_len); 804 fixup = msg.msg_iter; 805 if (unlikely((vhost_hlen))) { 806 /* We will supply the header ourselves 807 * TODO: support TSO. 808 */ 809 iov_iter_advance(&msg.msg_iter, vhost_hlen); 810 } 811 err = sock->ops->recvmsg(sock, &msg, 812 sock_len, MSG_DONTWAIT | MSG_TRUNC); 813 /* Userspace might have consumed the packet meanwhile: 814 * it's not supposed to do this usually, but might be hard 815 * to prevent. Discard data we got (if any) and keep going. */ 816 if (unlikely(err != sock_len)) { 817 pr_debug("Discarded rx packet: " 818 " len %d, expected %zd\n", err, sock_len); 819 vhost_discard_vq_desc(vq, headcount); 820 continue; 821 } 822 /* Supply virtio_net_hdr if VHOST_NET_F_VIRTIO_NET_HDR */ 823 if (unlikely(vhost_hlen)) { 824 if (copy_to_iter(&hdr, sizeof(hdr), 825 &fixup) != sizeof(hdr)) { 826 vq_err(vq, "Unable to write vnet_hdr " 827 "at addr %p\n", vq->iov->iov_base); 828 goto out; 829 } 830 } else { 831 /* Header came from socket; we'll need to patch 832 * ->num_buffers over if VIRTIO_NET_F_MRG_RXBUF 833 */ 834 iov_iter_advance(&fixup, sizeof(hdr)); 835 } 836 /* TODO: Should check and handle checksum. */ 837 838 num_buffers = cpu_to_vhost16(vq, headcount); 839 if (likely(mergeable) && 840 copy_to_iter(&num_buffers, sizeof num_buffers, 841 &fixup) != sizeof num_buffers) { 842 vq_err(vq, "Failed num_buffers write"); 843 vhost_discard_vq_desc(vq, headcount); 844 goto out; 845 } 846 vhost_add_used_and_signal_n(&net->dev, vq, vq->heads, 847 headcount); 848 if (unlikely(vq_log)) 849 vhost_log_write(vq, vq_log, log, vhost_len); 850 total_len += vhost_len; 851 if (unlikely(total_len >= VHOST_NET_WEIGHT)) { 852 vhost_poll_queue(&vq->poll); 853 goto out; 854 } 855 } 856 vhost_net_enable_vq(net, vq); 857 out: 858 mutex_unlock(&vq->mutex); 859 } 860 861 static void handle_tx_kick(struct vhost_work *work) 862 { 863 struct vhost_virtqueue *vq = container_of(work, struct vhost_virtqueue, 864 poll.work); 865 struct vhost_net *net = container_of(vq->dev, struct vhost_net, dev); 866 867 handle_tx(net); 868 } 869 870 static void handle_rx_kick(struct vhost_work *work) 871 { 872 struct vhost_virtqueue *vq = container_of(work, struct vhost_virtqueue, 873 poll.work); 874 struct vhost_net *net = container_of(vq->dev, struct vhost_net, dev); 875 876 handle_rx(net); 877 } 878 879 static void handle_tx_net(struct vhost_work *work) 880 { 881 struct vhost_net *net = container_of(work, struct vhost_net, 882 poll[VHOST_NET_VQ_TX].work); 883 handle_tx(net); 884 } 885 886 static void handle_rx_net(struct vhost_work *work) 887 { 888 struct vhost_net *net = container_of(work, struct vhost_net, 889 poll[VHOST_NET_VQ_RX].work); 890 handle_rx(net); 891 } 892 893 static int vhost_net_open(struct inode *inode, struct file *f) 894 { 895 struct vhost_net *n; 896 struct vhost_dev *dev; 897 struct vhost_virtqueue **vqs; 898 struct sk_buff **queue; 899 int i; 900 901 n = kvmalloc(sizeof *n, GFP_KERNEL | __GFP_RETRY_MAYFAIL); 902 if (!n) 903 return -ENOMEM; 904 vqs = kmalloc(VHOST_NET_VQ_MAX * sizeof(*vqs), GFP_KERNEL); 905 if (!vqs) { 906 kvfree(n); 907 return -ENOMEM; 908 } 909 910 queue = kmalloc_array(VHOST_RX_BATCH, sizeof(struct sk_buff *), 911 GFP_KERNEL); 912 if (!queue) { 913 kfree(vqs); 914 kvfree(n); 915 return -ENOMEM; 916 } 917 n->vqs[VHOST_NET_VQ_RX].rxq.queue = queue; 918 919 dev = &n->dev; 920 vqs[VHOST_NET_VQ_TX] = &n->vqs[VHOST_NET_VQ_TX].vq; 921 vqs[VHOST_NET_VQ_RX] = &n->vqs[VHOST_NET_VQ_RX].vq; 922 n->vqs[VHOST_NET_VQ_TX].vq.handle_kick = handle_tx_kick; 923 n->vqs[VHOST_NET_VQ_RX].vq.handle_kick = handle_rx_kick; 924 for (i = 0; i < VHOST_NET_VQ_MAX; i++) { 925 n->vqs[i].ubufs = NULL; 926 n->vqs[i].ubuf_info = NULL; 927 n->vqs[i].upend_idx = 0; 928 n->vqs[i].done_idx = 0; 929 n->vqs[i].vhost_hlen = 0; 930 n->vqs[i].sock_hlen = 0; 931 vhost_net_buf_init(&n->vqs[i].rxq); 932 } 933 vhost_dev_init(dev, vqs, VHOST_NET_VQ_MAX); 934 935 vhost_poll_init(n->poll + VHOST_NET_VQ_TX, handle_tx_net, POLLOUT, dev); 936 vhost_poll_init(n->poll + VHOST_NET_VQ_RX, handle_rx_net, POLLIN, dev); 937 938 f->private_data = n; 939 940 return 0; 941 } 942 943 static struct socket *vhost_net_stop_vq(struct vhost_net *n, 944 struct vhost_virtqueue *vq) 945 { 946 struct socket *sock; 947 struct vhost_net_virtqueue *nvq = 948 container_of(vq, struct vhost_net_virtqueue, vq); 949 950 mutex_lock(&vq->mutex); 951 sock = vq->private_data; 952 vhost_net_disable_vq(n, vq); 953 vq->private_data = NULL; 954 vhost_net_buf_unproduce(nvq); 955 mutex_unlock(&vq->mutex); 956 return sock; 957 } 958 959 static void vhost_net_stop(struct vhost_net *n, struct socket **tx_sock, 960 struct socket **rx_sock) 961 { 962 *tx_sock = vhost_net_stop_vq(n, &n->vqs[VHOST_NET_VQ_TX].vq); 963 *rx_sock = vhost_net_stop_vq(n, &n->vqs[VHOST_NET_VQ_RX].vq); 964 } 965 966 static void vhost_net_flush_vq(struct vhost_net *n, int index) 967 { 968 vhost_poll_flush(n->poll + index); 969 vhost_poll_flush(&n->vqs[index].vq.poll); 970 } 971 972 static void vhost_net_flush(struct vhost_net *n) 973 { 974 vhost_net_flush_vq(n, VHOST_NET_VQ_TX); 975 vhost_net_flush_vq(n, VHOST_NET_VQ_RX); 976 if (n->vqs[VHOST_NET_VQ_TX].ubufs) { 977 mutex_lock(&n->vqs[VHOST_NET_VQ_TX].vq.mutex); 978 n->tx_flush = true; 979 mutex_unlock(&n->vqs[VHOST_NET_VQ_TX].vq.mutex); 980 /* Wait for all lower device DMAs done. */ 981 vhost_net_ubuf_put_and_wait(n->vqs[VHOST_NET_VQ_TX].ubufs); 982 mutex_lock(&n->vqs[VHOST_NET_VQ_TX].vq.mutex); 983 n->tx_flush = false; 984 atomic_set(&n->vqs[VHOST_NET_VQ_TX].ubufs->refcount, 1); 985 mutex_unlock(&n->vqs[VHOST_NET_VQ_TX].vq.mutex); 986 } 987 } 988 989 static int vhost_net_release(struct inode *inode, struct file *f) 990 { 991 struct vhost_net *n = f->private_data; 992 struct socket *tx_sock; 993 struct socket *rx_sock; 994 995 vhost_net_stop(n, &tx_sock, &rx_sock); 996 vhost_net_flush(n); 997 vhost_dev_stop(&n->dev); 998 vhost_dev_cleanup(&n->dev, false); 999 vhost_net_vq_reset(n); 1000 if (tx_sock) 1001 sockfd_put(tx_sock); 1002 if (rx_sock) 1003 sockfd_put(rx_sock); 1004 /* Make sure no callbacks are outstanding */ 1005 synchronize_rcu_bh(); 1006 /* We do an extra flush before freeing memory, 1007 * since jobs can re-queue themselves. */ 1008 vhost_net_flush(n); 1009 kfree(n->vqs[VHOST_NET_VQ_RX].rxq.queue); 1010 kfree(n->dev.vqs); 1011 kvfree(n); 1012 return 0; 1013 } 1014 1015 static struct socket *get_raw_socket(int fd) 1016 { 1017 struct { 1018 struct sockaddr_ll sa; 1019 char buf[MAX_ADDR_LEN]; 1020 } uaddr; 1021 int uaddr_len = sizeof uaddr, r; 1022 struct socket *sock = sockfd_lookup(fd, &r); 1023 1024 if (!sock) 1025 return ERR_PTR(-ENOTSOCK); 1026 1027 /* Parameter checking */ 1028 if (sock->sk->sk_type != SOCK_RAW) { 1029 r = -ESOCKTNOSUPPORT; 1030 goto err; 1031 } 1032 1033 r = sock->ops->getname(sock, (struct sockaddr *)&uaddr.sa, 1034 &uaddr_len, 0); 1035 if (r) 1036 goto err; 1037 1038 if (uaddr.sa.sll_family != AF_PACKET) { 1039 r = -EPFNOSUPPORT; 1040 goto err; 1041 } 1042 return sock; 1043 err: 1044 sockfd_put(sock); 1045 return ERR_PTR(r); 1046 } 1047 1048 static struct skb_array *get_tap_skb_array(int fd) 1049 { 1050 struct skb_array *array; 1051 struct file *file = fget(fd); 1052 1053 if (!file) 1054 return NULL; 1055 array = tun_get_skb_array(file); 1056 if (!IS_ERR(array)) 1057 goto out; 1058 array = tap_get_skb_array(file); 1059 if (!IS_ERR(array)) 1060 goto out; 1061 array = NULL; 1062 out: 1063 fput(file); 1064 return array; 1065 } 1066 1067 static struct socket *get_tap_socket(int fd) 1068 { 1069 struct file *file = fget(fd); 1070 struct socket *sock; 1071 1072 if (!file) 1073 return ERR_PTR(-EBADF); 1074 sock = tun_get_socket(file); 1075 if (!IS_ERR(sock)) 1076 return sock; 1077 sock = tap_get_socket(file); 1078 if (IS_ERR(sock)) 1079 fput(file); 1080 return sock; 1081 } 1082 1083 static struct socket *get_socket(int fd) 1084 { 1085 struct socket *sock; 1086 1087 /* special case to disable backend */ 1088 if (fd == -1) 1089 return NULL; 1090 sock = get_raw_socket(fd); 1091 if (!IS_ERR(sock)) 1092 return sock; 1093 sock = get_tap_socket(fd); 1094 if (!IS_ERR(sock)) 1095 return sock; 1096 return ERR_PTR(-ENOTSOCK); 1097 } 1098 1099 static long vhost_net_set_backend(struct vhost_net *n, unsigned index, int fd) 1100 { 1101 struct socket *sock, *oldsock; 1102 struct vhost_virtqueue *vq; 1103 struct vhost_net_virtqueue *nvq; 1104 struct vhost_net_ubuf_ref *ubufs, *oldubufs = NULL; 1105 int r; 1106 1107 mutex_lock(&n->dev.mutex); 1108 r = vhost_dev_check_owner(&n->dev); 1109 if (r) 1110 goto err; 1111 1112 if (index >= VHOST_NET_VQ_MAX) { 1113 r = -ENOBUFS; 1114 goto err; 1115 } 1116 vq = &n->vqs[index].vq; 1117 nvq = &n->vqs[index]; 1118 mutex_lock(&vq->mutex); 1119 1120 /* Verify that ring has been setup correctly. */ 1121 if (!vhost_vq_access_ok(vq)) { 1122 r = -EFAULT; 1123 goto err_vq; 1124 } 1125 sock = get_socket(fd); 1126 if (IS_ERR(sock)) { 1127 r = PTR_ERR(sock); 1128 goto err_vq; 1129 } 1130 1131 /* start polling new socket */ 1132 oldsock = vq->private_data; 1133 if (sock != oldsock) { 1134 ubufs = vhost_net_ubuf_alloc(vq, 1135 sock && vhost_sock_zcopy(sock)); 1136 if (IS_ERR(ubufs)) { 1137 r = PTR_ERR(ubufs); 1138 goto err_ubufs; 1139 } 1140 1141 vhost_net_disable_vq(n, vq); 1142 vq->private_data = sock; 1143 vhost_net_buf_unproduce(nvq); 1144 if (index == VHOST_NET_VQ_RX) 1145 nvq->rx_array = get_tap_skb_array(fd); 1146 r = vhost_vq_init_access(vq); 1147 if (r) 1148 goto err_used; 1149 r = vhost_net_enable_vq(n, vq); 1150 if (r) 1151 goto err_used; 1152 1153 oldubufs = nvq->ubufs; 1154 nvq->ubufs = ubufs; 1155 1156 n->tx_packets = 0; 1157 n->tx_zcopy_err = 0; 1158 n->tx_flush = false; 1159 } 1160 1161 mutex_unlock(&vq->mutex); 1162 1163 if (oldubufs) { 1164 vhost_net_ubuf_put_wait_and_free(oldubufs); 1165 mutex_lock(&vq->mutex); 1166 vhost_zerocopy_signal_used(n, vq); 1167 mutex_unlock(&vq->mutex); 1168 } 1169 1170 if (oldsock) { 1171 vhost_net_flush_vq(n, index); 1172 sockfd_put(oldsock); 1173 } 1174 1175 mutex_unlock(&n->dev.mutex); 1176 return 0; 1177 1178 err_used: 1179 vq->private_data = oldsock; 1180 vhost_net_enable_vq(n, vq); 1181 if (ubufs) 1182 vhost_net_ubuf_put_wait_and_free(ubufs); 1183 err_ubufs: 1184 sockfd_put(sock); 1185 err_vq: 1186 mutex_unlock(&vq->mutex); 1187 err: 1188 mutex_unlock(&n->dev.mutex); 1189 return r; 1190 } 1191 1192 static long vhost_net_reset_owner(struct vhost_net *n) 1193 { 1194 struct socket *tx_sock = NULL; 1195 struct socket *rx_sock = NULL; 1196 long err; 1197 struct vhost_umem *umem; 1198 1199 mutex_lock(&n->dev.mutex); 1200 err = vhost_dev_check_owner(&n->dev); 1201 if (err) 1202 goto done; 1203 umem = vhost_dev_reset_owner_prepare(); 1204 if (!umem) { 1205 err = -ENOMEM; 1206 goto done; 1207 } 1208 vhost_net_stop(n, &tx_sock, &rx_sock); 1209 vhost_net_flush(n); 1210 vhost_dev_reset_owner(&n->dev, umem); 1211 vhost_net_vq_reset(n); 1212 done: 1213 mutex_unlock(&n->dev.mutex); 1214 if (tx_sock) 1215 sockfd_put(tx_sock); 1216 if (rx_sock) 1217 sockfd_put(rx_sock); 1218 return err; 1219 } 1220 1221 static int vhost_net_set_features(struct vhost_net *n, u64 features) 1222 { 1223 size_t vhost_hlen, sock_hlen, hdr_len; 1224 int i; 1225 1226 hdr_len = (features & ((1ULL << VIRTIO_NET_F_MRG_RXBUF) | 1227 (1ULL << VIRTIO_F_VERSION_1))) ? 1228 sizeof(struct virtio_net_hdr_mrg_rxbuf) : 1229 sizeof(struct virtio_net_hdr); 1230 if (features & (1 << VHOST_NET_F_VIRTIO_NET_HDR)) { 1231 /* vhost provides vnet_hdr */ 1232 vhost_hlen = hdr_len; 1233 sock_hlen = 0; 1234 } else { 1235 /* socket provides vnet_hdr */ 1236 vhost_hlen = 0; 1237 sock_hlen = hdr_len; 1238 } 1239 mutex_lock(&n->dev.mutex); 1240 if ((features & (1 << VHOST_F_LOG_ALL)) && 1241 !vhost_log_access_ok(&n->dev)) 1242 goto out_unlock; 1243 1244 if ((features & (1ULL << VIRTIO_F_IOMMU_PLATFORM))) { 1245 if (vhost_init_device_iotlb(&n->dev, true)) 1246 goto out_unlock; 1247 } 1248 1249 for (i = 0; i < VHOST_NET_VQ_MAX; ++i) { 1250 mutex_lock(&n->vqs[i].vq.mutex); 1251 n->vqs[i].vq.acked_features = features; 1252 n->vqs[i].vhost_hlen = vhost_hlen; 1253 n->vqs[i].sock_hlen = sock_hlen; 1254 mutex_unlock(&n->vqs[i].vq.mutex); 1255 } 1256 mutex_unlock(&n->dev.mutex); 1257 return 0; 1258 1259 out_unlock: 1260 mutex_unlock(&n->dev.mutex); 1261 return -EFAULT; 1262 } 1263 1264 static long vhost_net_set_owner(struct vhost_net *n) 1265 { 1266 int r; 1267 1268 mutex_lock(&n->dev.mutex); 1269 if (vhost_dev_has_owner(&n->dev)) { 1270 r = -EBUSY; 1271 goto out; 1272 } 1273 r = vhost_net_set_ubuf_info(n); 1274 if (r) 1275 goto out; 1276 r = vhost_dev_set_owner(&n->dev); 1277 if (r) 1278 vhost_net_clear_ubuf_info(n); 1279 vhost_net_flush(n); 1280 out: 1281 mutex_unlock(&n->dev.mutex); 1282 return r; 1283 } 1284 1285 static long vhost_net_ioctl(struct file *f, unsigned int ioctl, 1286 unsigned long arg) 1287 { 1288 struct vhost_net *n = f->private_data; 1289 void __user *argp = (void __user *)arg; 1290 u64 __user *featurep = argp; 1291 struct vhost_vring_file backend; 1292 u64 features; 1293 int r; 1294 1295 switch (ioctl) { 1296 case VHOST_NET_SET_BACKEND: 1297 if (copy_from_user(&backend, argp, sizeof backend)) 1298 return -EFAULT; 1299 return vhost_net_set_backend(n, backend.index, backend.fd); 1300 case VHOST_GET_FEATURES: 1301 features = VHOST_NET_FEATURES; 1302 if (copy_to_user(featurep, &features, sizeof features)) 1303 return -EFAULT; 1304 return 0; 1305 case VHOST_SET_FEATURES: 1306 if (copy_from_user(&features, featurep, sizeof features)) 1307 return -EFAULT; 1308 if (features & ~VHOST_NET_FEATURES) 1309 return -EOPNOTSUPP; 1310 return vhost_net_set_features(n, features); 1311 case VHOST_RESET_OWNER: 1312 return vhost_net_reset_owner(n); 1313 case VHOST_SET_OWNER: 1314 return vhost_net_set_owner(n); 1315 default: 1316 mutex_lock(&n->dev.mutex); 1317 r = vhost_dev_ioctl(&n->dev, ioctl, argp); 1318 if (r == -ENOIOCTLCMD) 1319 r = vhost_vring_ioctl(&n->dev, ioctl, argp); 1320 else 1321 vhost_net_flush(n); 1322 mutex_unlock(&n->dev.mutex); 1323 return r; 1324 } 1325 } 1326 1327 #ifdef CONFIG_COMPAT 1328 static long vhost_net_compat_ioctl(struct file *f, unsigned int ioctl, 1329 unsigned long arg) 1330 { 1331 return vhost_net_ioctl(f, ioctl, (unsigned long)compat_ptr(arg)); 1332 } 1333 #endif 1334 1335 static ssize_t vhost_net_chr_read_iter(struct kiocb *iocb, struct iov_iter *to) 1336 { 1337 struct file *file = iocb->ki_filp; 1338 struct vhost_net *n = file->private_data; 1339 struct vhost_dev *dev = &n->dev; 1340 int noblock = file->f_flags & O_NONBLOCK; 1341 1342 return vhost_chr_read_iter(dev, to, noblock); 1343 } 1344 1345 static ssize_t vhost_net_chr_write_iter(struct kiocb *iocb, 1346 struct iov_iter *from) 1347 { 1348 struct file *file = iocb->ki_filp; 1349 struct vhost_net *n = file->private_data; 1350 struct vhost_dev *dev = &n->dev; 1351 1352 return vhost_chr_write_iter(dev, from); 1353 } 1354 1355 static unsigned int vhost_net_chr_poll(struct file *file, poll_table *wait) 1356 { 1357 struct vhost_net *n = file->private_data; 1358 struct vhost_dev *dev = &n->dev; 1359 1360 return vhost_chr_poll(file, dev, wait); 1361 } 1362 1363 static const struct file_operations vhost_net_fops = { 1364 .owner = THIS_MODULE, 1365 .release = vhost_net_release, 1366 .read_iter = vhost_net_chr_read_iter, 1367 .write_iter = vhost_net_chr_write_iter, 1368 .poll = vhost_net_chr_poll, 1369 .unlocked_ioctl = vhost_net_ioctl, 1370 #ifdef CONFIG_COMPAT 1371 .compat_ioctl = vhost_net_compat_ioctl, 1372 #endif 1373 .open = vhost_net_open, 1374 .llseek = noop_llseek, 1375 }; 1376 1377 static struct miscdevice vhost_net_misc = { 1378 .minor = VHOST_NET_MINOR, 1379 .name = "vhost-net", 1380 .fops = &vhost_net_fops, 1381 }; 1382 1383 static int vhost_net_init(void) 1384 { 1385 if (experimental_zcopytx) 1386 vhost_net_enable_zcopy(VHOST_NET_VQ_TX); 1387 return misc_register(&vhost_net_misc); 1388 } 1389 module_init(vhost_net_init); 1390 1391 static void vhost_net_exit(void) 1392 { 1393 misc_deregister(&vhost_net_misc); 1394 } 1395 module_exit(vhost_net_exit); 1396 1397 MODULE_VERSION("0.0.1"); 1398 MODULE_LICENSE("GPL v2"); 1399 MODULE_AUTHOR("Michael S. Tsirkin"); 1400 MODULE_DESCRIPTION("Host kernel accelerator for virtio net"); 1401 MODULE_ALIAS_MISCDEV(VHOST_NET_MINOR); 1402 MODULE_ALIAS("devname:vhost-net"); 1403