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 } 510 copy -= size; 511 offset1 += size; 512 offset = 0; 513 } 514 515 if (len == offset1) 516 return 0; 517 518 while (count--) { 519 struct page *page[MAX_SKB_FRAGS]; 520 int num_pages; 521 unsigned long base; 522 523 len = from->iov_len - offset1; 524 if (!len) { 525 offset1 = 0; 526 ++from; 527 continue; 528 } 529 base = (unsigned long)from->iov_base + offset1; 530 size = ((base & ~PAGE_MASK) + len + ~PAGE_MASK) >> PAGE_SHIFT; 531 num_pages = get_user_pages_fast(base, size, 0, &page[i]); 532 if ((num_pages != size) || 533 (num_pages > MAX_SKB_FRAGS - skb_shinfo(skb)->nr_frags)) 534 /* put_page is in skb free */ 535 return -EFAULT; 536 skb->data_len += len; 537 skb->len += len; 538 skb->truesize += len; 539 atomic_add(len, &skb->sk->sk_wmem_alloc); 540 while (len) { 541 int off = base & ~PAGE_MASK; 542 int size = min_t(int, len, PAGE_SIZE - off); 543 __skb_fill_page_desc(skb, i, page[i], off, size); 544 skb_shinfo(skb)->nr_frags++; 545 /* increase sk_wmem_alloc */ 546 base += size; 547 len -= size; 548 i++; 549 } 550 offset1 = 0; 551 ++from; 552 } 553 return 0; 554 } 555 556 /* 557 * macvtap_skb_from_vnet_hdr and macvtap_skb_to_vnet_hdr should 558 * be shared with the tun/tap driver. 559 */ 560 static int macvtap_skb_from_vnet_hdr(struct sk_buff *skb, 561 struct virtio_net_hdr *vnet_hdr) 562 { 563 unsigned short gso_type = 0; 564 if (vnet_hdr->gso_type != VIRTIO_NET_HDR_GSO_NONE) { 565 switch (vnet_hdr->gso_type & ~VIRTIO_NET_HDR_GSO_ECN) { 566 case VIRTIO_NET_HDR_GSO_TCPV4: 567 gso_type = SKB_GSO_TCPV4; 568 break; 569 case VIRTIO_NET_HDR_GSO_TCPV6: 570 gso_type = SKB_GSO_TCPV6; 571 break; 572 case VIRTIO_NET_HDR_GSO_UDP: 573 gso_type = SKB_GSO_UDP; 574 break; 575 default: 576 return -EINVAL; 577 } 578 579 if (vnet_hdr->gso_type & VIRTIO_NET_HDR_GSO_ECN) 580 gso_type |= SKB_GSO_TCP_ECN; 581 582 if (vnet_hdr->gso_size == 0) 583 return -EINVAL; 584 } 585 586 if (vnet_hdr->flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) { 587 if (!skb_partial_csum_set(skb, vnet_hdr->csum_start, 588 vnet_hdr->csum_offset)) 589 return -EINVAL; 590 } 591 592 if (vnet_hdr->gso_type != VIRTIO_NET_HDR_GSO_NONE) { 593 skb_shinfo(skb)->gso_size = vnet_hdr->gso_size; 594 skb_shinfo(skb)->gso_type = gso_type; 595 596 /* Header must be checked, and gso_segs computed. */ 597 skb_shinfo(skb)->gso_type |= SKB_GSO_DODGY; 598 skb_shinfo(skb)->gso_segs = 0; 599 } 600 return 0; 601 } 602 603 static int macvtap_skb_to_vnet_hdr(const struct sk_buff *skb, 604 struct virtio_net_hdr *vnet_hdr) 605 { 606 memset(vnet_hdr, 0, sizeof(*vnet_hdr)); 607 608 if (skb_is_gso(skb)) { 609 struct skb_shared_info *sinfo = skb_shinfo(skb); 610 611 /* This is a hint as to how much should be linear. */ 612 vnet_hdr->hdr_len = skb_headlen(skb); 613 vnet_hdr->gso_size = sinfo->gso_size; 614 if (sinfo->gso_type & SKB_GSO_TCPV4) 615 vnet_hdr->gso_type = VIRTIO_NET_HDR_GSO_TCPV4; 616 else if (sinfo->gso_type & SKB_GSO_TCPV6) 617 vnet_hdr->gso_type = VIRTIO_NET_HDR_GSO_TCPV6; 618 else if (sinfo->gso_type & SKB_GSO_UDP) 619 vnet_hdr->gso_type = VIRTIO_NET_HDR_GSO_UDP; 620 else 621 BUG(); 622 if (sinfo->gso_type & SKB_GSO_TCP_ECN) 623 vnet_hdr->gso_type |= VIRTIO_NET_HDR_GSO_ECN; 624 } else 625 vnet_hdr->gso_type = VIRTIO_NET_HDR_GSO_NONE; 626 627 if (skb->ip_summed == CHECKSUM_PARTIAL) { 628 vnet_hdr->flags = VIRTIO_NET_HDR_F_NEEDS_CSUM; 629 vnet_hdr->csum_start = skb_checksum_start_offset(skb); 630 vnet_hdr->csum_offset = skb->csum_offset; 631 } else if (skb->ip_summed == CHECKSUM_UNNECESSARY) { 632 vnet_hdr->flags = VIRTIO_NET_HDR_F_DATA_VALID; 633 } /* else everything is zero */ 634 635 return 0; 636 } 637 638 639 /* Get packet from user space buffer */ 640 static ssize_t macvtap_get_user(struct macvtap_queue *q, struct msghdr *m, 641 const struct iovec *iv, unsigned long total_len, 642 size_t count, int noblock) 643 { 644 struct sk_buff *skb; 645 struct macvlan_dev *vlan; 646 unsigned long len = total_len; 647 int err; 648 struct virtio_net_hdr vnet_hdr = { 0 }; 649 int vnet_hdr_len = 0; 650 int copylen; 651 bool zerocopy = false; 652 653 if (q->flags & IFF_VNET_HDR) { 654 vnet_hdr_len = q->vnet_hdr_sz; 655 656 err = -EINVAL; 657 if (len < vnet_hdr_len) 658 goto err; 659 len -= vnet_hdr_len; 660 661 err = memcpy_fromiovecend((void *)&vnet_hdr, iv, 0, 662 sizeof(vnet_hdr)); 663 if (err < 0) 664 goto err; 665 if ((vnet_hdr.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) && 666 vnet_hdr.csum_start + vnet_hdr.csum_offset + 2 > 667 vnet_hdr.hdr_len) 668 vnet_hdr.hdr_len = vnet_hdr.csum_start + 669 vnet_hdr.csum_offset + 2; 670 err = -EINVAL; 671 if (vnet_hdr.hdr_len > len) 672 goto err; 673 } 674 675 err = -EINVAL; 676 if (unlikely(len < ETH_HLEN)) 677 goto err; 678 679 if (m && m->msg_control && sock_flag(&q->sk, SOCK_ZEROCOPY)) 680 zerocopy = true; 681 682 if (zerocopy) { 683 /* There are 256 bytes to be copied in skb, so there is enough 684 * room for skb expand head in case it is used. 685 * The rest buffer is mapped from userspace. 686 */ 687 copylen = vnet_hdr.hdr_len; 688 if (!copylen) 689 copylen = GOODCOPY_LEN; 690 } else 691 copylen = len; 692 693 skb = macvtap_alloc_skb(&q->sk, NET_IP_ALIGN, copylen, 694 vnet_hdr.hdr_len, noblock, &err); 695 if (!skb) 696 goto err; 697 698 if (zerocopy) { 699 err = zerocopy_sg_from_iovec(skb, iv, vnet_hdr_len, count); 700 skb_shinfo(skb)->tx_flags |= SKBTX_DEV_ZEROCOPY; 701 } else 702 err = skb_copy_datagram_from_iovec(skb, 0, iv, vnet_hdr_len, 703 len); 704 if (err) 705 goto err_kfree; 706 707 skb_set_network_header(skb, ETH_HLEN); 708 skb_reset_mac_header(skb); 709 skb->protocol = eth_hdr(skb)->h_proto; 710 711 if (vnet_hdr_len) { 712 err = macvtap_skb_from_vnet_hdr(skb, &vnet_hdr); 713 if (err) 714 goto err_kfree; 715 } 716 717 rcu_read_lock_bh(); 718 vlan = rcu_dereference_bh(q->vlan); 719 /* copy skb_ubuf_info for callback when skb has no error */ 720 if (zerocopy) 721 skb_shinfo(skb)->destructor_arg = m->msg_control; 722 if (vlan) 723 macvlan_start_xmit(skb, vlan->dev); 724 else 725 kfree_skb(skb); 726 rcu_read_unlock_bh(); 727 728 return total_len; 729 730 err_kfree: 731 kfree_skb(skb); 732 733 err: 734 rcu_read_lock_bh(); 735 vlan = rcu_dereference_bh(q->vlan); 736 if (vlan) 737 vlan->dev->stats.tx_dropped++; 738 rcu_read_unlock_bh(); 739 740 return err; 741 } 742 743 static ssize_t macvtap_aio_write(struct kiocb *iocb, const struct iovec *iv, 744 unsigned long count, loff_t pos) 745 { 746 struct file *file = iocb->ki_filp; 747 ssize_t result = -ENOLINK; 748 struct macvtap_queue *q = file->private_data; 749 750 result = macvtap_get_user(q, NULL, iv, iov_length(iv, count), count, 751 file->f_flags & O_NONBLOCK); 752 return result; 753 } 754 755 /* Put packet to the user space buffer */ 756 static ssize_t macvtap_put_user(struct macvtap_queue *q, 757 const struct sk_buff *skb, 758 const struct iovec *iv, int len) 759 { 760 struct macvlan_dev *vlan; 761 int ret; 762 int vnet_hdr_len = 0; 763 int vlan_offset = 0; 764 int copied; 765 766 if (q->flags & IFF_VNET_HDR) { 767 struct virtio_net_hdr vnet_hdr; 768 vnet_hdr_len = q->vnet_hdr_sz; 769 if ((len -= vnet_hdr_len) < 0) 770 return -EINVAL; 771 772 ret = macvtap_skb_to_vnet_hdr(skb, &vnet_hdr); 773 if (ret) 774 return ret; 775 776 if (memcpy_toiovecend(iv, (void *)&vnet_hdr, 0, sizeof(vnet_hdr))) 777 return -EFAULT; 778 } 779 copied = vnet_hdr_len; 780 781 if (!vlan_tx_tag_present(skb)) 782 len = min_t(int, skb->len, len); 783 else { 784 int copy; 785 struct { 786 __be16 h_vlan_proto; 787 __be16 h_vlan_TCI; 788 } veth; 789 veth.h_vlan_proto = htons(ETH_P_8021Q); 790 veth.h_vlan_TCI = htons(vlan_tx_tag_get(skb)); 791 792 vlan_offset = offsetof(struct vlan_ethhdr, h_vlan_proto); 793 len = min_t(int, skb->len + VLAN_HLEN, len); 794 795 copy = min_t(int, vlan_offset, len); 796 ret = skb_copy_datagram_const_iovec(skb, 0, iv, copied, copy); 797 len -= copy; 798 copied += copy; 799 if (ret || !len) 800 goto done; 801 802 copy = min_t(int, sizeof(veth), len); 803 ret = memcpy_toiovecend(iv, (void *)&veth, copied, copy); 804 len -= copy; 805 copied += copy; 806 if (ret || !len) 807 goto done; 808 } 809 810 ret = skb_copy_datagram_const_iovec(skb, vlan_offset, iv, copied, len); 811 copied += len; 812 813 done: 814 rcu_read_lock_bh(); 815 vlan = rcu_dereference_bh(q->vlan); 816 if (vlan) 817 macvlan_count_rx(vlan, copied - vnet_hdr_len, ret == 0, 0); 818 rcu_read_unlock_bh(); 819 820 return ret ? ret : copied; 821 } 822 823 static ssize_t macvtap_do_read(struct macvtap_queue *q, struct kiocb *iocb, 824 const struct iovec *iv, unsigned long len, 825 int noblock) 826 { 827 DECLARE_WAITQUEUE(wait, current); 828 struct sk_buff *skb; 829 ssize_t ret = 0; 830 831 add_wait_queue(sk_sleep(&q->sk), &wait); 832 while (len) { 833 current->state = TASK_INTERRUPTIBLE; 834 835 /* Read frames from the queue */ 836 skb = skb_dequeue(&q->sk.sk_receive_queue); 837 if (!skb) { 838 if (noblock) { 839 ret = -EAGAIN; 840 break; 841 } 842 if (signal_pending(current)) { 843 ret = -ERESTARTSYS; 844 break; 845 } 846 /* Nothing to read, let's sleep */ 847 schedule(); 848 continue; 849 } 850 ret = macvtap_put_user(q, skb, iv, len); 851 kfree_skb(skb); 852 break; 853 } 854 855 current->state = TASK_RUNNING; 856 remove_wait_queue(sk_sleep(&q->sk), &wait); 857 return ret; 858 } 859 860 static ssize_t macvtap_aio_read(struct kiocb *iocb, const struct iovec *iv, 861 unsigned long count, loff_t pos) 862 { 863 struct file *file = iocb->ki_filp; 864 struct macvtap_queue *q = file->private_data; 865 ssize_t len, ret = 0; 866 867 len = iov_length(iv, count); 868 if (len < 0) { 869 ret = -EINVAL; 870 goto out; 871 } 872 873 ret = macvtap_do_read(q, iocb, iv, len, file->f_flags & O_NONBLOCK); 874 ret = min_t(ssize_t, ret, len); /* XXX copied from tun.c. Why? */ 875 out: 876 return ret; 877 } 878 879 /* 880 * provide compatibility with generic tun/tap interface 881 */ 882 static long macvtap_ioctl(struct file *file, unsigned int cmd, 883 unsigned long arg) 884 { 885 struct macvtap_queue *q = file->private_data; 886 struct macvlan_dev *vlan; 887 void __user *argp = (void __user *)arg; 888 struct ifreq __user *ifr = argp; 889 unsigned int __user *up = argp; 890 unsigned int u; 891 int __user *sp = argp; 892 int s; 893 int ret; 894 895 switch (cmd) { 896 case TUNSETIFF: 897 /* ignore the name, just look at flags */ 898 if (get_user(u, &ifr->ifr_flags)) 899 return -EFAULT; 900 901 ret = 0; 902 if ((u & ~IFF_VNET_HDR) != (IFF_NO_PI | IFF_TAP)) 903 ret = -EINVAL; 904 else 905 q->flags = u; 906 907 return ret; 908 909 case TUNGETIFF: 910 rcu_read_lock_bh(); 911 vlan = rcu_dereference_bh(q->vlan); 912 if (vlan) 913 dev_hold(vlan->dev); 914 rcu_read_unlock_bh(); 915 916 if (!vlan) 917 return -ENOLINK; 918 919 ret = 0; 920 if (copy_to_user(&ifr->ifr_name, vlan->dev->name, IFNAMSIZ) || 921 put_user(q->flags, &ifr->ifr_flags)) 922 ret = -EFAULT; 923 dev_put(vlan->dev); 924 return ret; 925 926 case TUNGETFEATURES: 927 if (put_user(IFF_TAP | IFF_NO_PI | IFF_VNET_HDR, up)) 928 return -EFAULT; 929 return 0; 930 931 case TUNSETSNDBUF: 932 if (get_user(u, up)) 933 return -EFAULT; 934 935 q->sk.sk_sndbuf = u; 936 return 0; 937 938 case TUNGETVNETHDRSZ: 939 s = q->vnet_hdr_sz; 940 if (put_user(s, sp)) 941 return -EFAULT; 942 return 0; 943 944 case TUNSETVNETHDRSZ: 945 if (get_user(s, sp)) 946 return -EFAULT; 947 if (s < (int)sizeof(struct virtio_net_hdr)) 948 return -EINVAL; 949 950 q->vnet_hdr_sz = s; 951 return 0; 952 953 case TUNSETOFFLOAD: 954 /* let the user check for future flags */ 955 if (arg & ~(TUN_F_CSUM | TUN_F_TSO4 | TUN_F_TSO6 | 956 TUN_F_TSO_ECN | TUN_F_UFO)) 957 return -EINVAL; 958 959 /* TODO: only accept frames with the features that 960 got enabled for forwarded frames */ 961 if (!(q->flags & IFF_VNET_HDR)) 962 return -EINVAL; 963 return 0; 964 965 default: 966 return -EINVAL; 967 } 968 } 969 970 #ifdef CONFIG_COMPAT 971 static long macvtap_compat_ioctl(struct file *file, unsigned int cmd, 972 unsigned long arg) 973 { 974 return macvtap_ioctl(file, cmd, (unsigned long)compat_ptr(arg)); 975 } 976 #endif 977 978 static const struct file_operations macvtap_fops = { 979 .owner = THIS_MODULE, 980 .open = macvtap_open, 981 .release = macvtap_release, 982 .aio_read = macvtap_aio_read, 983 .aio_write = macvtap_aio_write, 984 .poll = macvtap_poll, 985 .llseek = no_llseek, 986 .unlocked_ioctl = macvtap_ioctl, 987 #ifdef CONFIG_COMPAT 988 .compat_ioctl = macvtap_compat_ioctl, 989 #endif 990 }; 991 992 static int macvtap_sendmsg(struct kiocb *iocb, struct socket *sock, 993 struct msghdr *m, size_t total_len) 994 { 995 struct macvtap_queue *q = container_of(sock, struct macvtap_queue, sock); 996 return macvtap_get_user(q, m, m->msg_iov, total_len, m->msg_iovlen, 997 m->msg_flags & MSG_DONTWAIT); 998 } 999 1000 static int macvtap_recvmsg(struct kiocb *iocb, struct socket *sock, 1001 struct msghdr *m, size_t total_len, 1002 int flags) 1003 { 1004 struct macvtap_queue *q = container_of(sock, struct macvtap_queue, sock); 1005 int ret; 1006 if (flags & ~(MSG_DONTWAIT|MSG_TRUNC)) 1007 return -EINVAL; 1008 ret = macvtap_do_read(q, iocb, m->msg_iov, total_len, 1009 flags & MSG_DONTWAIT); 1010 if (ret > total_len) { 1011 m->msg_flags |= MSG_TRUNC; 1012 ret = flags & MSG_TRUNC ? ret : total_len; 1013 } 1014 return ret; 1015 } 1016 1017 /* Ops structure to mimic raw sockets with tun */ 1018 static const struct proto_ops macvtap_socket_ops = { 1019 .sendmsg = macvtap_sendmsg, 1020 .recvmsg = macvtap_recvmsg, 1021 }; 1022 1023 /* Get an underlying socket object from tun file. Returns error unless file is 1024 * attached to a device. The returned object works like a packet socket, it 1025 * can be used for sock_sendmsg/sock_recvmsg. The caller is responsible for 1026 * holding a reference to the file for as long as the socket is in use. */ 1027 struct socket *macvtap_get_socket(struct file *file) 1028 { 1029 struct macvtap_queue *q; 1030 if (file->f_op != &macvtap_fops) 1031 return ERR_PTR(-EINVAL); 1032 q = file->private_data; 1033 if (!q) 1034 return ERR_PTR(-EBADFD); 1035 return &q->sock; 1036 } 1037 EXPORT_SYMBOL_GPL(macvtap_get_socket); 1038 1039 static int macvtap_device_event(struct notifier_block *unused, 1040 unsigned long event, void *ptr) 1041 { 1042 struct net_device *dev = ptr; 1043 struct macvlan_dev *vlan; 1044 struct device *classdev; 1045 dev_t devt; 1046 int err; 1047 1048 if (dev->rtnl_link_ops != &macvtap_link_ops) 1049 return NOTIFY_DONE; 1050 1051 vlan = netdev_priv(dev); 1052 1053 switch (event) { 1054 case NETDEV_REGISTER: 1055 /* Create the device node here after the network device has 1056 * been registered but before register_netdevice has 1057 * finished running. 1058 */ 1059 err = macvtap_get_minor(vlan); 1060 if (err) 1061 return notifier_from_errno(err); 1062 1063 devt = MKDEV(MAJOR(macvtap_major), vlan->minor); 1064 classdev = device_create(macvtap_class, &dev->dev, devt, 1065 dev, "tap%d", dev->ifindex); 1066 if (IS_ERR(classdev)) { 1067 macvtap_free_minor(vlan); 1068 return notifier_from_errno(PTR_ERR(classdev)); 1069 } 1070 break; 1071 case NETDEV_UNREGISTER: 1072 devt = MKDEV(MAJOR(macvtap_major), vlan->minor); 1073 device_destroy(macvtap_class, devt); 1074 macvtap_free_minor(vlan); 1075 break; 1076 } 1077 1078 return NOTIFY_DONE; 1079 } 1080 1081 static struct notifier_block macvtap_notifier_block __read_mostly = { 1082 .notifier_call = macvtap_device_event, 1083 }; 1084 1085 static int macvtap_init(void) 1086 { 1087 int err; 1088 1089 err = alloc_chrdev_region(&macvtap_major, 0, 1090 MACVTAP_NUM_DEVS, "macvtap"); 1091 if (err) 1092 goto out1; 1093 1094 cdev_init(&macvtap_cdev, &macvtap_fops); 1095 err = cdev_add(&macvtap_cdev, macvtap_major, MACVTAP_NUM_DEVS); 1096 if (err) 1097 goto out2; 1098 1099 macvtap_class = class_create(THIS_MODULE, "macvtap"); 1100 if (IS_ERR(macvtap_class)) { 1101 err = PTR_ERR(macvtap_class); 1102 goto out3; 1103 } 1104 1105 err = register_netdevice_notifier(&macvtap_notifier_block); 1106 if (err) 1107 goto out4; 1108 1109 err = macvlan_link_register(&macvtap_link_ops); 1110 if (err) 1111 goto out5; 1112 1113 return 0; 1114 1115 out5: 1116 unregister_netdevice_notifier(&macvtap_notifier_block); 1117 out4: 1118 class_unregister(macvtap_class); 1119 out3: 1120 cdev_del(&macvtap_cdev); 1121 out2: 1122 unregister_chrdev_region(macvtap_major, MACVTAP_NUM_DEVS); 1123 out1: 1124 return err; 1125 } 1126 module_init(macvtap_init); 1127 1128 static void macvtap_exit(void) 1129 { 1130 rtnl_link_unregister(&macvtap_link_ops); 1131 unregister_netdevice_notifier(&macvtap_notifier_block); 1132 class_unregister(macvtap_class); 1133 cdev_del(&macvtap_cdev); 1134 unregister_chrdev_region(macvtap_major, MACVTAP_NUM_DEVS); 1135 } 1136 module_exit(macvtap_exit); 1137 1138 MODULE_ALIAS_RTNL_LINK("macvtap"); 1139 MODULE_AUTHOR("Arnd Bergmann <arnd@arndb.de>"); 1140 MODULE_LICENSE("GPL"); 1141