1 #include <linux/etherdevice.h> 2 #include <linux/if_macvlan.h> 3 #include <linux/if_vlan.h> 4 #include <linux/interrupt.h> 5 #include <linux/nsproxy.h> 6 #include <linux/compat.h> 7 #include <linux/if_tun.h> 8 #include <linux/module.h> 9 #include <linux/skbuff.h> 10 #include <linux/cache.h> 11 #include <linux/sched.h> 12 #include <linux/types.h> 13 #include <linux/slab.h> 14 #include <linux/init.h> 15 #include <linux/wait.h> 16 #include <linux/cdev.h> 17 #include <linux/idr.h> 18 #include <linux/fs.h> 19 20 #include <net/net_namespace.h> 21 #include <net/rtnetlink.h> 22 #include <net/sock.h> 23 #include <linux/virtio_net.h> 24 25 /* 26 * A macvtap queue is the central object of this driver, it connects 27 * an open character device to a macvlan interface. There can be 28 * multiple queues on one interface, which map back to queues 29 * implemented in hardware on the underlying device. 30 * 31 * macvtap_proto is used to allocate queues through the sock allocation 32 * mechanism. 33 * 34 * TODO: multiqueue support is currently not implemented, even though 35 * macvtap is basically prepared for that. We will need to add this 36 * here as well as in virtio-net and qemu to get line rate on 10gbit 37 * adapters from a guest. 38 */ 39 struct macvtap_queue { 40 struct sock sk; 41 struct socket sock; 42 struct socket_wq wq; 43 int vnet_hdr_sz; 44 struct macvlan_dev __rcu *vlan; 45 struct file *file; 46 unsigned int flags; 47 }; 48 49 static struct proto macvtap_proto = { 50 .name = "macvtap", 51 .owner = THIS_MODULE, 52 .obj_size = sizeof (struct macvtap_queue), 53 }; 54 55 /* 56 * Variables for dealing with macvtaps device numbers. 57 */ 58 static dev_t macvtap_major; 59 #define MACVTAP_NUM_DEVS (1U << MINORBITS) 60 static DEFINE_MUTEX(minor_lock); 61 static DEFINE_IDR(minor_idr); 62 63 #define GOODCOPY_LEN 128 64 static struct class *macvtap_class; 65 static struct cdev macvtap_cdev; 66 67 static const struct proto_ops macvtap_socket_ops; 68 69 /* 70 * RCU usage: 71 * The macvtap_queue and the macvlan_dev are loosely coupled, the 72 * pointers from one to the other can only be read while rcu_read_lock 73 * or macvtap_lock is held. 74 * 75 * Both the file and the macvlan_dev hold a reference on the macvtap_queue 76 * through sock_hold(&q->sk). When the macvlan_dev goes away first, 77 * q->vlan becomes inaccessible. When the files gets closed, 78 * macvtap_get_queue() fails. 79 * 80 * There may still be references to the struct sock inside of the 81 * queue from outbound SKBs, but these never reference back to the 82 * file or the dev. The data structure is freed through __sk_free 83 * when both our references and any pending SKBs are gone. 84 */ 85 static DEFINE_SPINLOCK(macvtap_lock); 86 87 /* 88 * get_slot: return a [unused/occupied] slot in vlan->taps[]: 89 * - if 'q' is NULL, return the first empty slot; 90 * - otherwise, return the slot this pointer occupies. 91 */ 92 static int get_slot(struct macvlan_dev *vlan, struct macvtap_queue *q) 93 { 94 int i; 95 96 for (i = 0; i < MAX_MACVTAP_QUEUES; i++) { 97 if (rcu_dereference(vlan->taps[i]) == q) 98 return i; 99 } 100 101 /* Should never happen */ 102 BUG_ON(1); 103 } 104 105 static int macvtap_set_queue(struct net_device *dev, struct file *file, 106 struct macvtap_queue *q) 107 { 108 struct macvlan_dev *vlan = netdev_priv(dev); 109 int index; 110 int err = -EBUSY; 111 112 spin_lock(&macvtap_lock); 113 if (vlan->numvtaps == MAX_MACVTAP_QUEUES) 114 goto out; 115 116 err = 0; 117 index = get_slot(vlan, NULL); 118 rcu_assign_pointer(q->vlan, vlan); 119 rcu_assign_pointer(vlan->taps[index], q); 120 sock_hold(&q->sk); 121 122 q->file = file; 123 file->private_data = q; 124 125 vlan->numvtaps++; 126 127 out: 128 spin_unlock(&macvtap_lock); 129 return err; 130 } 131 132 /* 133 * The file owning the queue got closed, give up both 134 * the reference that the files holds as well as the 135 * one from the macvlan_dev if that still exists. 136 * 137 * Using the spinlock makes sure that we don't get 138 * to the queue again after destroying it. 139 */ 140 static void macvtap_put_queue(struct macvtap_queue *q) 141 { 142 struct macvlan_dev *vlan; 143 144 spin_lock(&macvtap_lock); 145 vlan = rcu_dereference_protected(q->vlan, 146 lockdep_is_held(&macvtap_lock)); 147 if (vlan) { 148 int index = get_slot(vlan, q); 149 150 RCU_INIT_POINTER(vlan->taps[index], NULL); 151 RCU_INIT_POINTER(q->vlan, NULL); 152 sock_put(&q->sk); 153 --vlan->numvtaps; 154 } 155 156 spin_unlock(&macvtap_lock); 157 158 synchronize_rcu(); 159 sock_put(&q->sk); 160 } 161 162 /* 163 * Select a queue based on the rxq of the device on which this packet 164 * arrived. If the incoming device is not mq, calculate a flow hash 165 * to select a queue. If all fails, find the first available queue. 166 * Cache vlan->numvtaps since it can become zero during the execution 167 * of this function. 168 */ 169 static struct macvtap_queue *macvtap_get_queue(struct net_device *dev, 170 struct sk_buff *skb) 171 { 172 struct macvlan_dev *vlan = netdev_priv(dev); 173 struct macvtap_queue *tap = NULL; 174 int numvtaps = vlan->numvtaps; 175 __u32 rxq; 176 177 if (!numvtaps) 178 goto out; 179 180 /* Check if we can use flow to select a queue */ 181 rxq = skb_get_rxhash(skb); 182 if (rxq) { 183 tap = rcu_dereference(vlan->taps[rxq % numvtaps]); 184 if (tap) 185 goto out; 186 } 187 188 if (likely(skb_rx_queue_recorded(skb))) { 189 rxq = skb_get_rx_queue(skb); 190 191 while (unlikely(rxq >= numvtaps)) 192 rxq -= numvtaps; 193 194 tap = rcu_dereference(vlan->taps[rxq]); 195 if (tap) 196 goto out; 197 } 198 199 /* Everything failed - find first available queue */ 200 for (rxq = 0; rxq < MAX_MACVTAP_QUEUES; rxq++) { 201 tap = rcu_dereference(vlan->taps[rxq]); 202 if (tap) 203 break; 204 } 205 206 out: 207 return tap; 208 } 209 210 /* 211 * The net_device is going away, give up the reference 212 * that it holds on all queues and safely set the pointer 213 * from the queues to NULL. 214 */ 215 static void macvtap_del_queues(struct net_device *dev) 216 { 217 struct macvlan_dev *vlan = netdev_priv(dev); 218 struct macvtap_queue *q, *qlist[MAX_MACVTAP_QUEUES]; 219 int i, j = 0; 220 221 /* macvtap_put_queue can free some slots, so go through all slots */ 222 spin_lock(&macvtap_lock); 223 for (i = 0; i < MAX_MACVTAP_QUEUES && vlan->numvtaps; i++) { 224 q = rcu_dereference_protected(vlan->taps[i], 225 lockdep_is_held(&macvtap_lock)); 226 if (q) { 227 qlist[j++] = q; 228 RCU_INIT_POINTER(vlan->taps[i], NULL); 229 RCU_INIT_POINTER(q->vlan, NULL); 230 vlan->numvtaps--; 231 } 232 } 233 BUG_ON(vlan->numvtaps != 0); 234 /* guarantee that any future macvtap_set_queue will fail */ 235 vlan->numvtaps = MAX_MACVTAP_QUEUES; 236 spin_unlock(&macvtap_lock); 237 238 synchronize_rcu(); 239 240 for (--j; j >= 0; j--) 241 sock_put(&qlist[j]->sk); 242 } 243 244 /* 245 * Forward happens for data that gets sent from one macvlan 246 * endpoint to another one in bridge mode. We just take 247 * the skb and put it into the receive queue. 248 */ 249 static int macvtap_forward(struct net_device *dev, struct sk_buff *skb) 250 { 251 struct macvtap_queue *q = macvtap_get_queue(dev, skb); 252 if (!q) 253 goto drop; 254 255 if (skb_queue_len(&q->sk.sk_receive_queue) >= dev->tx_queue_len) 256 goto drop; 257 258 skb_queue_tail(&q->sk.sk_receive_queue, skb); 259 wake_up_interruptible_poll(sk_sleep(&q->sk), POLLIN | POLLRDNORM | POLLRDBAND); 260 return NET_RX_SUCCESS; 261 262 drop: 263 kfree_skb(skb); 264 return NET_RX_DROP; 265 } 266 267 /* 268 * Receive is for data from the external interface (lowerdev), 269 * in case of macvtap, we can treat that the same way as 270 * forward, which macvlan cannot. 271 */ 272 static int macvtap_receive(struct sk_buff *skb) 273 { 274 skb_push(skb, ETH_HLEN); 275 return macvtap_forward(skb->dev, skb); 276 } 277 278 static int macvtap_get_minor(struct macvlan_dev *vlan) 279 { 280 int retval = -ENOMEM; 281 int id; 282 283 mutex_lock(&minor_lock); 284 if (idr_pre_get(&minor_idr, GFP_KERNEL) == 0) 285 goto exit; 286 287 retval = idr_get_new_above(&minor_idr, vlan, 1, &id); 288 if (retval < 0) { 289 if (retval == -EAGAIN) 290 retval = -ENOMEM; 291 goto exit; 292 } 293 if (id < MACVTAP_NUM_DEVS) { 294 vlan->minor = id; 295 } else { 296 printk(KERN_ERR "too many macvtap devices\n"); 297 retval = -EINVAL; 298 idr_remove(&minor_idr, id); 299 } 300 exit: 301 mutex_unlock(&minor_lock); 302 return retval; 303 } 304 305 static void macvtap_free_minor(struct macvlan_dev *vlan) 306 { 307 mutex_lock(&minor_lock); 308 if (vlan->minor) { 309 idr_remove(&minor_idr, vlan->minor); 310 vlan->minor = 0; 311 } 312 mutex_unlock(&minor_lock); 313 } 314 315 static struct net_device *dev_get_by_macvtap_minor(int minor) 316 { 317 struct net_device *dev = NULL; 318 struct macvlan_dev *vlan; 319 320 mutex_lock(&minor_lock); 321 vlan = idr_find(&minor_idr, minor); 322 if (vlan) { 323 dev = vlan->dev; 324 dev_hold(dev); 325 } 326 mutex_unlock(&minor_lock); 327 return dev; 328 } 329 330 static int macvtap_newlink(struct net *src_net, 331 struct net_device *dev, 332 struct nlattr *tb[], 333 struct nlattr *data[]) 334 { 335 /* Don't put anything that may fail after macvlan_common_newlink 336 * because we can't undo what it does. 337 */ 338 return macvlan_common_newlink(src_net, dev, tb, data, 339 macvtap_receive, macvtap_forward); 340 } 341 342 static void macvtap_dellink(struct net_device *dev, 343 struct list_head *head) 344 { 345 macvtap_del_queues(dev); 346 macvlan_dellink(dev, head); 347 } 348 349 static void macvtap_setup(struct net_device *dev) 350 { 351 macvlan_common_setup(dev); 352 dev->tx_queue_len = TUN_READQ_SIZE; 353 } 354 355 static struct rtnl_link_ops macvtap_link_ops __read_mostly = { 356 .kind = "macvtap", 357 .setup = macvtap_setup, 358 .newlink = macvtap_newlink, 359 .dellink = macvtap_dellink, 360 }; 361 362 363 static void macvtap_sock_write_space(struct sock *sk) 364 { 365 wait_queue_head_t *wqueue; 366 367 if (!sock_writeable(sk) || 368 !test_and_clear_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags)) 369 return; 370 371 wqueue = sk_sleep(sk); 372 if (wqueue && waitqueue_active(wqueue)) 373 wake_up_interruptible_poll(wqueue, POLLOUT | POLLWRNORM | POLLWRBAND); 374 } 375 376 static void macvtap_sock_destruct(struct sock *sk) 377 { 378 skb_queue_purge(&sk->sk_receive_queue); 379 } 380 381 static int macvtap_open(struct inode *inode, struct file *file) 382 { 383 struct net *net = current->nsproxy->net_ns; 384 struct net_device *dev = dev_get_by_macvtap_minor(iminor(inode)); 385 struct macvtap_queue *q; 386 int err; 387 388 err = -ENODEV; 389 if (!dev) 390 goto out; 391 392 err = -ENOMEM; 393 q = (struct macvtap_queue *)sk_alloc(net, AF_UNSPEC, GFP_KERNEL, 394 &macvtap_proto); 395 if (!q) 396 goto out; 397 398 q->sock.wq = &q->wq; 399 init_waitqueue_head(&q->wq.wait); 400 q->sock.type = SOCK_RAW; 401 q->sock.state = SS_CONNECTED; 402 q->sock.file = file; 403 q->sock.ops = &macvtap_socket_ops; 404 sock_init_data(&q->sock, &q->sk); 405 q->sk.sk_write_space = macvtap_sock_write_space; 406 q->sk.sk_destruct = macvtap_sock_destruct; 407 q->flags = IFF_VNET_HDR | IFF_NO_PI | IFF_TAP; 408 q->vnet_hdr_sz = sizeof(struct virtio_net_hdr); 409 410 /* 411 * so far only KVM virtio_net uses macvtap, enable zero copy between 412 * guest kernel and host kernel when lower device supports zerocopy 413 * 414 * The macvlan supports zerocopy iff the lower device supports zero 415 * copy so we don't have to look at the lower device directly. 416 */ 417 if ((dev->features & NETIF_F_HIGHDMA) && (dev->features & NETIF_F_SG)) 418 sock_set_flag(&q->sk, SOCK_ZEROCOPY); 419 420 err = macvtap_set_queue(dev, file, q); 421 if (err) 422 sock_put(&q->sk); 423 424 out: 425 if (dev) 426 dev_put(dev); 427 428 return err; 429 } 430 431 static int macvtap_release(struct inode *inode, struct file *file) 432 { 433 struct macvtap_queue *q = file->private_data; 434 macvtap_put_queue(q); 435 return 0; 436 } 437 438 static unsigned int macvtap_poll(struct file *file, poll_table * wait) 439 { 440 struct macvtap_queue *q = file->private_data; 441 unsigned int mask = POLLERR; 442 443 if (!q) 444 goto out; 445 446 mask = 0; 447 poll_wait(file, &q->wq.wait, wait); 448 449 if (!skb_queue_empty(&q->sk.sk_receive_queue)) 450 mask |= POLLIN | POLLRDNORM; 451 452 if (sock_writeable(&q->sk) || 453 (!test_and_set_bit(SOCK_ASYNC_NOSPACE, &q->sock.flags) && 454 sock_writeable(&q->sk))) 455 mask |= POLLOUT | POLLWRNORM; 456 457 out: 458 return mask; 459 } 460 461 static inline struct sk_buff *macvtap_alloc_skb(struct sock *sk, size_t prepad, 462 size_t len, size_t linear, 463 int noblock, int *err) 464 { 465 struct sk_buff *skb; 466 467 /* Under a page? Don't bother with paged skb. */ 468 if (prepad + len < PAGE_SIZE || !linear) 469 linear = len; 470 471 skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock, 472 err); 473 if (!skb) 474 return NULL; 475 476 skb_reserve(skb, prepad); 477 skb_put(skb, linear); 478 skb->data_len = len - linear; 479 skb->len += len - linear; 480 481 return skb; 482 } 483 484 /* set skb frags from iovec, this can move to core network code for reuse */ 485 static int zerocopy_sg_from_iovec(struct sk_buff *skb, const struct iovec *from, 486 int offset, size_t count) 487 { 488 int len = iov_length(from, count) - offset; 489 int copy = skb_headlen(skb); 490 int size, offset1 = 0; 491 int i = 0; 492 493 /* Skip over from offset */ 494 while (count && (offset >= from->iov_len)) { 495 offset -= from->iov_len; 496 ++from; 497 --count; 498 } 499 500 /* copy up to skb headlen */ 501 while (count && (copy > 0)) { 502 size = min_t(unsigned int, copy, from->iov_len - offset); 503 if (copy_from_user(skb->data + offset1, from->iov_base + offset, 504 size)) 505 return -EFAULT; 506 if (copy > size) { 507 ++from; 508 --count; 509 offset = 0; 510 } else 511 offset += size; 512 copy -= size; 513 offset1 += size; 514 } 515 516 if (len == offset1) 517 return 0; 518 519 while (count--) { 520 struct page *page[MAX_SKB_FRAGS]; 521 int num_pages; 522 unsigned long base; 523 unsigned long truesize; 524 525 len = from->iov_len - offset; 526 if (!len) { 527 offset = 0; 528 ++from; 529 continue; 530 } 531 base = (unsigned long)from->iov_base + offset; 532 size = ((base & ~PAGE_MASK) + len + ~PAGE_MASK) >> PAGE_SHIFT; 533 if (i + size > MAX_SKB_FRAGS) 534 return -EMSGSIZE; 535 num_pages = get_user_pages_fast(base, size, 0, &page[i]); 536 if (num_pages != size) { 537 for (i = 0; i < num_pages; i++) 538 put_page(page[i]); 539 return -EFAULT; 540 } 541 truesize = size * PAGE_SIZE; 542 skb->data_len += len; 543 skb->len += len; 544 skb->truesize += truesize; 545 atomic_add(truesize, &skb->sk->sk_wmem_alloc); 546 while (len) { 547 int off = base & ~PAGE_MASK; 548 int size = min_t(int, len, PAGE_SIZE - off); 549 __skb_fill_page_desc(skb, i, page[i], off, size); 550 skb_shinfo(skb)->nr_frags++; 551 /* increase sk_wmem_alloc */ 552 base += size; 553 len -= size; 554 i++; 555 } 556 offset = 0; 557 ++from; 558 } 559 return 0; 560 } 561 562 /* 563 * macvtap_skb_from_vnet_hdr and macvtap_skb_to_vnet_hdr should 564 * be shared with the tun/tap driver. 565 */ 566 static int macvtap_skb_from_vnet_hdr(struct sk_buff *skb, 567 struct virtio_net_hdr *vnet_hdr) 568 { 569 unsigned short gso_type = 0; 570 if (vnet_hdr->gso_type != VIRTIO_NET_HDR_GSO_NONE) { 571 switch (vnet_hdr->gso_type & ~VIRTIO_NET_HDR_GSO_ECN) { 572 case VIRTIO_NET_HDR_GSO_TCPV4: 573 gso_type = SKB_GSO_TCPV4; 574 break; 575 case VIRTIO_NET_HDR_GSO_TCPV6: 576 gso_type = SKB_GSO_TCPV6; 577 break; 578 case VIRTIO_NET_HDR_GSO_UDP: 579 gso_type = SKB_GSO_UDP; 580 break; 581 default: 582 return -EINVAL; 583 } 584 585 if (vnet_hdr->gso_type & VIRTIO_NET_HDR_GSO_ECN) 586 gso_type |= SKB_GSO_TCP_ECN; 587 588 if (vnet_hdr->gso_size == 0) 589 return -EINVAL; 590 } 591 592 if (vnet_hdr->flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) { 593 if (!skb_partial_csum_set(skb, vnet_hdr->csum_start, 594 vnet_hdr->csum_offset)) 595 return -EINVAL; 596 } 597 598 if (vnet_hdr->gso_type != VIRTIO_NET_HDR_GSO_NONE) { 599 skb_shinfo(skb)->gso_size = vnet_hdr->gso_size; 600 skb_shinfo(skb)->gso_type = gso_type; 601 602 /* Header must be checked, and gso_segs computed. */ 603 skb_shinfo(skb)->gso_type |= SKB_GSO_DODGY; 604 skb_shinfo(skb)->gso_segs = 0; 605 } 606 return 0; 607 } 608 609 static int macvtap_skb_to_vnet_hdr(const struct sk_buff *skb, 610 struct virtio_net_hdr *vnet_hdr) 611 { 612 memset(vnet_hdr, 0, sizeof(*vnet_hdr)); 613 614 if (skb_is_gso(skb)) { 615 struct skb_shared_info *sinfo = skb_shinfo(skb); 616 617 /* This is a hint as to how much should be linear. */ 618 vnet_hdr->hdr_len = skb_headlen(skb); 619 vnet_hdr->gso_size = sinfo->gso_size; 620 if (sinfo->gso_type & SKB_GSO_TCPV4) 621 vnet_hdr->gso_type = VIRTIO_NET_HDR_GSO_TCPV4; 622 else if (sinfo->gso_type & SKB_GSO_TCPV6) 623 vnet_hdr->gso_type = VIRTIO_NET_HDR_GSO_TCPV6; 624 else if (sinfo->gso_type & SKB_GSO_UDP) 625 vnet_hdr->gso_type = VIRTIO_NET_HDR_GSO_UDP; 626 else 627 BUG(); 628 if (sinfo->gso_type & SKB_GSO_TCP_ECN) 629 vnet_hdr->gso_type |= VIRTIO_NET_HDR_GSO_ECN; 630 } else 631 vnet_hdr->gso_type = VIRTIO_NET_HDR_GSO_NONE; 632 633 if (skb->ip_summed == CHECKSUM_PARTIAL) { 634 vnet_hdr->flags = VIRTIO_NET_HDR_F_NEEDS_CSUM; 635 vnet_hdr->csum_start = skb_checksum_start_offset(skb); 636 vnet_hdr->csum_offset = skb->csum_offset; 637 } else if (skb->ip_summed == CHECKSUM_UNNECESSARY) { 638 vnet_hdr->flags = VIRTIO_NET_HDR_F_DATA_VALID; 639 } /* else everything is zero */ 640 641 return 0; 642 } 643 644 645 /* Get packet from user space buffer */ 646 static ssize_t macvtap_get_user(struct macvtap_queue *q, struct msghdr *m, 647 const struct iovec *iv, unsigned long total_len, 648 size_t count, int noblock) 649 { 650 struct sk_buff *skb; 651 struct macvlan_dev *vlan; 652 unsigned long len = total_len; 653 int err; 654 struct virtio_net_hdr vnet_hdr = { 0 }; 655 int vnet_hdr_len = 0; 656 int copylen = 0; 657 bool zerocopy = false; 658 659 if (q->flags & IFF_VNET_HDR) { 660 vnet_hdr_len = q->vnet_hdr_sz; 661 662 err = -EINVAL; 663 if (len < vnet_hdr_len) 664 goto err; 665 len -= vnet_hdr_len; 666 667 err = memcpy_fromiovecend((void *)&vnet_hdr, iv, 0, 668 sizeof(vnet_hdr)); 669 if (err < 0) 670 goto err; 671 if ((vnet_hdr.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) && 672 vnet_hdr.csum_start + vnet_hdr.csum_offset + 2 > 673 vnet_hdr.hdr_len) 674 vnet_hdr.hdr_len = vnet_hdr.csum_start + 675 vnet_hdr.csum_offset + 2; 676 err = -EINVAL; 677 if (vnet_hdr.hdr_len > len) 678 goto err; 679 } 680 681 err = -EINVAL; 682 if (unlikely(len < ETH_HLEN)) 683 goto err; 684 685 err = -EMSGSIZE; 686 if (unlikely(count > UIO_MAXIOV)) 687 goto err; 688 689 if (m && m->msg_control && sock_flag(&q->sk, SOCK_ZEROCOPY)) 690 zerocopy = true; 691 692 if (zerocopy) { 693 /* Userspace may produce vectors with count greater than 694 * MAX_SKB_FRAGS, so we need to linearize parts of the skb 695 * to let the rest of data to be fit in the frags. 696 */ 697 if (count > MAX_SKB_FRAGS) { 698 copylen = iov_length(iv, count - MAX_SKB_FRAGS); 699 if (copylen < vnet_hdr_len) 700 copylen = 0; 701 else 702 copylen -= vnet_hdr_len; 703 } 704 /* There are 256 bytes to be copied in skb, so there is enough 705 * room for skb expand head in case it is used. 706 * The rest buffer is mapped from userspace. 707 */ 708 if (copylen < vnet_hdr.hdr_len) 709 copylen = vnet_hdr.hdr_len; 710 if (!copylen) 711 copylen = GOODCOPY_LEN; 712 } else 713 copylen = len; 714 715 skb = macvtap_alloc_skb(&q->sk, NET_IP_ALIGN, copylen, 716 vnet_hdr.hdr_len, noblock, &err); 717 if (!skb) 718 goto err; 719 720 if (zerocopy) 721 err = zerocopy_sg_from_iovec(skb, iv, vnet_hdr_len, count); 722 else 723 err = skb_copy_datagram_from_iovec(skb, 0, iv, vnet_hdr_len, 724 len); 725 if (err) 726 goto err_kfree; 727 728 skb_set_network_header(skb, ETH_HLEN); 729 skb_reset_mac_header(skb); 730 skb->protocol = eth_hdr(skb)->h_proto; 731 732 if (vnet_hdr_len) { 733 err = macvtap_skb_from_vnet_hdr(skb, &vnet_hdr); 734 if (err) 735 goto err_kfree; 736 } 737 738 rcu_read_lock_bh(); 739 vlan = rcu_dereference_bh(q->vlan); 740 /* copy skb_ubuf_info for callback when skb has no error */ 741 if (zerocopy) { 742 skb_shinfo(skb)->destructor_arg = m->msg_control; 743 skb_shinfo(skb)->tx_flags |= SKBTX_DEV_ZEROCOPY; 744 } 745 if (vlan) 746 macvlan_start_xmit(skb, vlan->dev); 747 else 748 kfree_skb(skb); 749 rcu_read_unlock_bh(); 750 751 return total_len; 752 753 err_kfree: 754 kfree_skb(skb); 755 756 err: 757 rcu_read_lock_bh(); 758 vlan = rcu_dereference_bh(q->vlan); 759 if (vlan) 760 vlan->dev->stats.tx_dropped++; 761 rcu_read_unlock_bh(); 762 763 return err; 764 } 765 766 static ssize_t macvtap_aio_write(struct kiocb *iocb, const struct iovec *iv, 767 unsigned long count, loff_t pos) 768 { 769 struct file *file = iocb->ki_filp; 770 ssize_t result = -ENOLINK; 771 struct macvtap_queue *q = file->private_data; 772 773 result = macvtap_get_user(q, NULL, iv, iov_length(iv, count), count, 774 file->f_flags & O_NONBLOCK); 775 return result; 776 } 777 778 /* Put packet to the user space buffer */ 779 static ssize_t macvtap_put_user(struct macvtap_queue *q, 780 const struct sk_buff *skb, 781 const struct iovec *iv, int len) 782 { 783 struct macvlan_dev *vlan; 784 int ret; 785 int vnet_hdr_len = 0; 786 int vlan_offset = 0; 787 int copied; 788 789 if (q->flags & IFF_VNET_HDR) { 790 struct virtio_net_hdr vnet_hdr; 791 vnet_hdr_len = q->vnet_hdr_sz; 792 if ((len -= vnet_hdr_len) < 0) 793 return -EINVAL; 794 795 ret = macvtap_skb_to_vnet_hdr(skb, &vnet_hdr); 796 if (ret) 797 return ret; 798 799 if (memcpy_toiovecend(iv, (void *)&vnet_hdr, 0, sizeof(vnet_hdr))) 800 return -EFAULT; 801 } 802 copied = vnet_hdr_len; 803 804 if (!vlan_tx_tag_present(skb)) 805 len = min_t(int, skb->len, len); 806 else { 807 int copy; 808 struct { 809 __be16 h_vlan_proto; 810 __be16 h_vlan_TCI; 811 } veth; 812 veth.h_vlan_proto = htons(ETH_P_8021Q); 813 veth.h_vlan_TCI = htons(vlan_tx_tag_get(skb)); 814 815 vlan_offset = offsetof(struct vlan_ethhdr, h_vlan_proto); 816 len = min_t(int, skb->len + VLAN_HLEN, len); 817 818 copy = min_t(int, vlan_offset, len); 819 ret = skb_copy_datagram_const_iovec(skb, 0, iv, copied, copy); 820 len -= copy; 821 copied += copy; 822 if (ret || !len) 823 goto done; 824 825 copy = min_t(int, sizeof(veth), len); 826 ret = memcpy_toiovecend(iv, (void *)&veth, copied, copy); 827 len -= copy; 828 copied += copy; 829 if (ret || !len) 830 goto done; 831 } 832 833 ret = skb_copy_datagram_const_iovec(skb, vlan_offset, iv, copied, len); 834 copied += len; 835 836 done: 837 rcu_read_lock_bh(); 838 vlan = rcu_dereference_bh(q->vlan); 839 if (vlan) 840 macvlan_count_rx(vlan, copied - vnet_hdr_len, ret == 0, 0); 841 rcu_read_unlock_bh(); 842 843 return ret ? ret : copied; 844 } 845 846 static ssize_t macvtap_do_read(struct macvtap_queue *q, struct kiocb *iocb, 847 const struct iovec *iv, unsigned long len, 848 int noblock) 849 { 850 DECLARE_WAITQUEUE(wait, current); 851 struct sk_buff *skb; 852 ssize_t ret = 0; 853 854 add_wait_queue(sk_sleep(&q->sk), &wait); 855 while (len) { 856 current->state = TASK_INTERRUPTIBLE; 857 858 /* Read frames from the queue */ 859 skb = skb_dequeue(&q->sk.sk_receive_queue); 860 if (!skb) { 861 if (noblock) { 862 ret = -EAGAIN; 863 break; 864 } 865 if (signal_pending(current)) { 866 ret = -ERESTARTSYS; 867 break; 868 } 869 /* Nothing to read, let's sleep */ 870 schedule(); 871 continue; 872 } 873 ret = macvtap_put_user(q, skb, iv, len); 874 kfree_skb(skb); 875 break; 876 } 877 878 current->state = TASK_RUNNING; 879 remove_wait_queue(sk_sleep(&q->sk), &wait); 880 return ret; 881 } 882 883 static ssize_t macvtap_aio_read(struct kiocb *iocb, const struct iovec *iv, 884 unsigned long count, loff_t pos) 885 { 886 struct file *file = iocb->ki_filp; 887 struct macvtap_queue *q = file->private_data; 888 ssize_t len, ret = 0; 889 890 len = iov_length(iv, count); 891 if (len < 0) { 892 ret = -EINVAL; 893 goto out; 894 } 895 896 ret = macvtap_do_read(q, iocb, iv, len, file->f_flags & O_NONBLOCK); 897 ret = min_t(ssize_t, ret, len); /* XXX copied from tun.c. Why? */ 898 out: 899 return ret; 900 } 901 902 /* 903 * provide compatibility with generic tun/tap interface 904 */ 905 static long macvtap_ioctl(struct file *file, unsigned int cmd, 906 unsigned long arg) 907 { 908 struct macvtap_queue *q = file->private_data; 909 struct macvlan_dev *vlan; 910 void __user *argp = (void __user *)arg; 911 struct ifreq __user *ifr = argp; 912 unsigned int __user *up = argp; 913 unsigned int u; 914 int __user *sp = argp; 915 int s; 916 int ret; 917 918 switch (cmd) { 919 case TUNSETIFF: 920 /* ignore the name, just look at flags */ 921 if (get_user(u, &ifr->ifr_flags)) 922 return -EFAULT; 923 924 ret = 0; 925 if ((u & ~IFF_VNET_HDR) != (IFF_NO_PI | IFF_TAP)) 926 ret = -EINVAL; 927 else 928 q->flags = u; 929 930 return ret; 931 932 case TUNGETIFF: 933 rcu_read_lock_bh(); 934 vlan = rcu_dereference_bh(q->vlan); 935 if (vlan) 936 dev_hold(vlan->dev); 937 rcu_read_unlock_bh(); 938 939 if (!vlan) 940 return -ENOLINK; 941 942 ret = 0; 943 if (copy_to_user(&ifr->ifr_name, vlan->dev->name, IFNAMSIZ) || 944 put_user(q->flags, &ifr->ifr_flags)) 945 ret = -EFAULT; 946 dev_put(vlan->dev); 947 return ret; 948 949 case TUNGETFEATURES: 950 if (put_user(IFF_TAP | IFF_NO_PI | IFF_VNET_HDR, up)) 951 return -EFAULT; 952 return 0; 953 954 case TUNSETSNDBUF: 955 if (get_user(u, up)) 956 return -EFAULT; 957 958 q->sk.sk_sndbuf = u; 959 return 0; 960 961 case TUNGETVNETHDRSZ: 962 s = q->vnet_hdr_sz; 963 if (put_user(s, sp)) 964 return -EFAULT; 965 return 0; 966 967 case TUNSETVNETHDRSZ: 968 if (get_user(s, sp)) 969 return -EFAULT; 970 if (s < (int)sizeof(struct virtio_net_hdr)) 971 return -EINVAL; 972 973 q->vnet_hdr_sz = s; 974 return 0; 975 976 case TUNSETOFFLOAD: 977 /* let the user check for future flags */ 978 if (arg & ~(TUN_F_CSUM | TUN_F_TSO4 | TUN_F_TSO6 | 979 TUN_F_TSO_ECN | TUN_F_UFO)) 980 return -EINVAL; 981 982 /* TODO: only accept frames with the features that 983 got enabled for forwarded frames */ 984 if (!(q->flags & IFF_VNET_HDR)) 985 return -EINVAL; 986 return 0; 987 988 default: 989 return -EINVAL; 990 } 991 } 992 993 #ifdef CONFIG_COMPAT 994 static long macvtap_compat_ioctl(struct file *file, unsigned int cmd, 995 unsigned long arg) 996 { 997 return macvtap_ioctl(file, cmd, (unsigned long)compat_ptr(arg)); 998 } 999 #endif 1000 1001 static const struct file_operations macvtap_fops = { 1002 .owner = THIS_MODULE, 1003 .open = macvtap_open, 1004 .release = macvtap_release, 1005 .aio_read = macvtap_aio_read, 1006 .aio_write = macvtap_aio_write, 1007 .poll = macvtap_poll, 1008 .llseek = no_llseek, 1009 .unlocked_ioctl = macvtap_ioctl, 1010 #ifdef CONFIG_COMPAT 1011 .compat_ioctl = macvtap_compat_ioctl, 1012 #endif 1013 }; 1014 1015 static int macvtap_sendmsg(struct kiocb *iocb, struct socket *sock, 1016 struct msghdr *m, size_t total_len) 1017 { 1018 struct macvtap_queue *q = container_of(sock, struct macvtap_queue, sock); 1019 return macvtap_get_user(q, m, m->msg_iov, total_len, m->msg_iovlen, 1020 m->msg_flags & MSG_DONTWAIT); 1021 } 1022 1023 static int macvtap_recvmsg(struct kiocb *iocb, struct socket *sock, 1024 struct msghdr *m, size_t total_len, 1025 int flags) 1026 { 1027 struct macvtap_queue *q = container_of(sock, struct macvtap_queue, sock); 1028 int ret; 1029 if (flags & ~(MSG_DONTWAIT|MSG_TRUNC)) 1030 return -EINVAL; 1031 ret = macvtap_do_read(q, iocb, m->msg_iov, total_len, 1032 flags & MSG_DONTWAIT); 1033 if (ret > total_len) { 1034 m->msg_flags |= MSG_TRUNC; 1035 ret = flags & MSG_TRUNC ? ret : total_len; 1036 } 1037 return ret; 1038 } 1039 1040 /* Ops structure to mimic raw sockets with tun */ 1041 static const struct proto_ops macvtap_socket_ops = { 1042 .sendmsg = macvtap_sendmsg, 1043 .recvmsg = macvtap_recvmsg, 1044 }; 1045 1046 /* Get an underlying socket object from tun file. Returns error unless file is 1047 * attached to a device. The returned object works like a packet socket, it 1048 * can be used for sock_sendmsg/sock_recvmsg. The caller is responsible for 1049 * holding a reference to the file for as long as the socket is in use. */ 1050 struct socket *macvtap_get_socket(struct file *file) 1051 { 1052 struct macvtap_queue *q; 1053 if (file->f_op != &macvtap_fops) 1054 return ERR_PTR(-EINVAL); 1055 q = file->private_data; 1056 if (!q) 1057 return ERR_PTR(-EBADFD); 1058 return &q->sock; 1059 } 1060 EXPORT_SYMBOL_GPL(macvtap_get_socket); 1061 1062 static int macvtap_device_event(struct notifier_block *unused, 1063 unsigned long event, void *ptr) 1064 { 1065 struct net_device *dev = ptr; 1066 struct macvlan_dev *vlan; 1067 struct device *classdev; 1068 dev_t devt; 1069 int err; 1070 1071 if (dev->rtnl_link_ops != &macvtap_link_ops) 1072 return NOTIFY_DONE; 1073 1074 vlan = netdev_priv(dev); 1075 1076 switch (event) { 1077 case NETDEV_REGISTER: 1078 /* Create the device node here after the network device has 1079 * been registered but before register_netdevice has 1080 * finished running. 1081 */ 1082 err = macvtap_get_minor(vlan); 1083 if (err) 1084 return notifier_from_errno(err); 1085 1086 devt = MKDEV(MAJOR(macvtap_major), vlan->minor); 1087 classdev = device_create(macvtap_class, &dev->dev, devt, 1088 dev, "tap%d", dev->ifindex); 1089 if (IS_ERR(classdev)) { 1090 macvtap_free_minor(vlan); 1091 return notifier_from_errno(PTR_ERR(classdev)); 1092 } 1093 break; 1094 case NETDEV_UNREGISTER: 1095 devt = MKDEV(MAJOR(macvtap_major), vlan->minor); 1096 device_destroy(macvtap_class, devt); 1097 macvtap_free_minor(vlan); 1098 break; 1099 } 1100 1101 return NOTIFY_DONE; 1102 } 1103 1104 static struct notifier_block macvtap_notifier_block __read_mostly = { 1105 .notifier_call = macvtap_device_event, 1106 }; 1107 1108 static int macvtap_init(void) 1109 { 1110 int err; 1111 1112 err = alloc_chrdev_region(&macvtap_major, 0, 1113 MACVTAP_NUM_DEVS, "macvtap"); 1114 if (err) 1115 goto out1; 1116 1117 cdev_init(&macvtap_cdev, &macvtap_fops); 1118 err = cdev_add(&macvtap_cdev, macvtap_major, MACVTAP_NUM_DEVS); 1119 if (err) 1120 goto out2; 1121 1122 macvtap_class = class_create(THIS_MODULE, "macvtap"); 1123 if (IS_ERR(macvtap_class)) { 1124 err = PTR_ERR(macvtap_class); 1125 goto out3; 1126 } 1127 1128 err = register_netdevice_notifier(&macvtap_notifier_block); 1129 if (err) 1130 goto out4; 1131 1132 err = macvlan_link_register(&macvtap_link_ops); 1133 if (err) 1134 goto out5; 1135 1136 return 0; 1137 1138 out5: 1139 unregister_netdevice_notifier(&macvtap_notifier_block); 1140 out4: 1141 class_unregister(macvtap_class); 1142 out3: 1143 cdev_del(&macvtap_cdev); 1144 out2: 1145 unregister_chrdev_region(macvtap_major, MACVTAP_NUM_DEVS); 1146 out1: 1147 return err; 1148 } 1149 module_init(macvtap_init); 1150 1151 static void macvtap_exit(void) 1152 { 1153 rtnl_link_unregister(&macvtap_link_ops); 1154 unregister_netdevice_notifier(&macvtap_notifier_block); 1155 class_unregister(macvtap_class); 1156 cdev_del(&macvtap_cdev); 1157 unregister_chrdev_region(macvtap_major, MACVTAP_NUM_DEVS); 1158 } 1159 module_exit(macvtap_exit); 1160 1161 MODULE_ALIAS_RTNL_LINK("macvtap"); 1162 MODULE_AUTHOR("Arnd Bergmann <arnd@arndb.de>"); 1163 MODULE_LICENSE("GPL"); 1164