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/wait.h> 15 #include <linux/cdev.h> 16 #include <linux/idr.h> 17 #include <linux/fs.h> 18 #include <linux/uio.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 */ 35 struct macvtap_queue { 36 struct sock sk; 37 struct socket sock; 38 struct socket_wq wq; 39 int vnet_hdr_sz; 40 struct macvlan_dev __rcu *vlan; 41 struct file *file; 42 unsigned int flags; 43 u16 queue_index; 44 bool enabled; 45 struct list_head next; 46 }; 47 48 #define MACVTAP_FEATURES (IFF_VNET_HDR | IFF_MULTI_QUEUE) 49 50 #define MACVTAP_VNET_LE 0x80000000 51 #define MACVTAP_VNET_BE 0x40000000 52 53 #ifdef CONFIG_TUN_VNET_CROSS_LE 54 static inline bool macvtap_legacy_is_little_endian(struct macvtap_queue *q) 55 { 56 return q->flags & MACVTAP_VNET_BE ? false : 57 virtio_legacy_is_little_endian(); 58 } 59 60 static long macvtap_get_vnet_be(struct macvtap_queue *q, int __user *sp) 61 { 62 int s = !!(q->flags & MACVTAP_VNET_BE); 63 64 if (put_user(s, sp)) 65 return -EFAULT; 66 67 return 0; 68 } 69 70 static long macvtap_set_vnet_be(struct macvtap_queue *q, int __user *sp) 71 { 72 int s; 73 74 if (get_user(s, sp)) 75 return -EFAULT; 76 77 if (s) 78 q->flags |= MACVTAP_VNET_BE; 79 else 80 q->flags &= ~MACVTAP_VNET_BE; 81 82 return 0; 83 } 84 #else 85 static inline bool macvtap_legacy_is_little_endian(struct macvtap_queue *q) 86 { 87 return virtio_legacy_is_little_endian(); 88 } 89 90 static long macvtap_get_vnet_be(struct macvtap_queue *q, int __user *argp) 91 { 92 return -EINVAL; 93 } 94 95 static long macvtap_set_vnet_be(struct macvtap_queue *q, int __user *argp) 96 { 97 return -EINVAL; 98 } 99 #endif /* CONFIG_TUN_VNET_CROSS_LE */ 100 101 static inline bool macvtap_is_little_endian(struct macvtap_queue *q) 102 { 103 return q->flags & MACVTAP_VNET_LE || 104 macvtap_legacy_is_little_endian(q); 105 } 106 107 static inline u16 macvtap16_to_cpu(struct macvtap_queue *q, __virtio16 val) 108 { 109 return __virtio16_to_cpu(macvtap_is_little_endian(q), val); 110 } 111 112 static inline __virtio16 cpu_to_macvtap16(struct macvtap_queue *q, u16 val) 113 { 114 return __cpu_to_virtio16(macvtap_is_little_endian(q), val); 115 } 116 117 static struct proto macvtap_proto = { 118 .name = "macvtap", 119 .owner = THIS_MODULE, 120 .obj_size = sizeof (struct macvtap_queue), 121 }; 122 123 /* 124 * Variables for dealing with macvtaps device numbers. 125 */ 126 static dev_t macvtap_major; 127 #define MACVTAP_NUM_DEVS (1U << MINORBITS) 128 static DEFINE_MUTEX(minor_lock); 129 static DEFINE_IDR(minor_idr); 130 131 #define GOODCOPY_LEN 128 132 static const void *macvtap_net_namespace(struct device *d) 133 { 134 struct net_device *dev = to_net_dev(d->parent); 135 return dev_net(dev); 136 } 137 138 static struct class macvtap_class = { 139 .name = "macvtap", 140 .owner = THIS_MODULE, 141 .ns_type = &net_ns_type_operations, 142 .namespace = macvtap_net_namespace, 143 }; 144 static struct cdev macvtap_cdev; 145 146 static const struct proto_ops macvtap_socket_ops; 147 148 #define TUN_OFFLOADS (NETIF_F_HW_CSUM | NETIF_F_TSO_ECN | NETIF_F_TSO | \ 149 NETIF_F_TSO6 | NETIF_F_UFO) 150 #define RX_OFFLOADS (NETIF_F_GRO | NETIF_F_LRO) 151 #define TAP_FEATURES (NETIF_F_GSO | NETIF_F_SG | NETIF_F_FRAGLIST) 152 153 static struct macvlan_dev *macvtap_get_vlan_rcu(const struct net_device *dev) 154 { 155 return rcu_dereference(dev->rx_handler_data); 156 } 157 158 /* 159 * RCU usage: 160 * The macvtap_queue and the macvlan_dev are loosely coupled, the 161 * pointers from one to the other can only be read while rcu_read_lock 162 * or rtnl is held. 163 * 164 * Both the file and the macvlan_dev hold a reference on the macvtap_queue 165 * through sock_hold(&q->sk). When the macvlan_dev goes away first, 166 * q->vlan becomes inaccessible. When the files gets closed, 167 * macvtap_get_queue() fails. 168 * 169 * There may still be references to the struct sock inside of the 170 * queue from outbound SKBs, but these never reference back to the 171 * file or the dev. The data structure is freed through __sk_free 172 * when both our references and any pending SKBs are gone. 173 */ 174 175 static int macvtap_enable_queue(struct net_device *dev, struct file *file, 176 struct macvtap_queue *q) 177 { 178 struct macvlan_dev *vlan = netdev_priv(dev); 179 int err = -EINVAL; 180 181 ASSERT_RTNL(); 182 183 if (q->enabled) 184 goto out; 185 186 err = 0; 187 rcu_assign_pointer(vlan->taps[vlan->numvtaps], q); 188 q->queue_index = vlan->numvtaps; 189 q->enabled = true; 190 191 vlan->numvtaps++; 192 out: 193 return err; 194 } 195 196 /* Requires RTNL */ 197 static int macvtap_set_queue(struct net_device *dev, struct file *file, 198 struct macvtap_queue *q) 199 { 200 struct macvlan_dev *vlan = netdev_priv(dev); 201 202 if (vlan->numqueues == MAX_MACVTAP_QUEUES) 203 return -EBUSY; 204 205 rcu_assign_pointer(q->vlan, vlan); 206 rcu_assign_pointer(vlan->taps[vlan->numvtaps], q); 207 sock_hold(&q->sk); 208 209 q->file = file; 210 q->queue_index = vlan->numvtaps; 211 q->enabled = true; 212 file->private_data = q; 213 list_add_tail(&q->next, &vlan->queue_list); 214 215 vlan->numvtaps++; 216 vlan->numqueues++; 217 218 return 0; 219 } 220 221 static int macvtap_disable_queue(struct macvtap_queue *q) 222 { 223 struct macvlan_dev *vlan; 224 struct macvtap_queue *nq; 225 226 ASSERT_RTNL(); 227 if (!q->enabled) 228 return -EINVAL; 229 230 vlan = rtnl_dereference(q->vlan); 231 232 if (vlan) { 233 int index = q->queue_index; 234 BUG_ON(index >= vlan->numvtaps); 235 nq = rtnl_dereference(vlan->taps[vlan->numvtaps - 1]); 236 nq->queue_index = index; 237 238 rcu_assign_pointer(vlan->taps[index], nq); 239 RCU_INIT_POINTER(vlan->taps[vlan->numvtaps - 1], NULL); 240 q->enabled = false; 241 242 vlan->numvtaps--; 243 } 244 245 return 0; 246 } 247 248 /* 249 * The file owning the queue got closed, give up both 250 * the reference that the files holds as well as the 251 * one from the macvlan_dev if that still exists. 252 * 253 * Using the spinlock makes sure that we don't get 254 * to the queue again after destroying it. 255 */ 256 static void macvtap_put_queue(struct macvtap_queue *q) 257 { 258 struct macvlan_dev *vlan; 259 260 rtnl_lock(); 261 vlan = rtnl_dereference(q->vlan); 262 263 if (vlan) { 264 if (q->enabled) 265 BUG_ON(macvtap_disable_queue(q)); 266 267 vlan->numqueues--; 268 RCU_INIT_POINTER(q->vlan, NULL); 269 sock_put(&q->sk); 270 list_del_init(&q->next); 271 } 272 273 rtnl_unlock(); 274 275 synchronize_rcu(); 276 sock_put(&q->sk); 277 } 278 279 /* 280 * Select a queue based on the rxq of the device on which this packet 281 * arrived. If the incoming device is not mq, calculate a flow hash 282 * to select a queue. If all fails, find the first available queue. 283 * Cache vlan->numvtaps since it can become zero during the execution 284 * of this function. 285 */ 286 static struct macvtap_queue *macvtap_get_queue(struct net_device *dev, 287 struct sk_buff *skb) 288 { 289 struct macvlan_dev *vlan = netdev_priv(dev); 290 struct macvtap_queue *tap = NULL; 291 /* Access to taps array is protected by rcu, but access to numvtaps 292 * isn't. Below we use it to lookup a queue, but treat it as a hint 293 * and validate that the result isn't NULL - in case we are 294 * racing against queue removal. 295 */ 296 int numvtaps = ACCESS_ONCE(vlan->numvtaps); 297 __u32 rxq; 298 299 if (!numvtaps) 300 goto out; 301 302 /* Check if we can use flow to select a queue */ 303 rxq = skb_get_hash(skb); 304 if (rxq) { 305 tap = rcu_dereference(vlan->taps[rxq % numvtaps]); 306 goto out; 307 } 308 309 if (likely(skb_rx_queue_recorded(skb))) { 310 rxq = skb_get_rx_queue(skb); 311 312 while (unlikely(rxq >= numvtaps)) 313 rxq -= numvtaps; 314 315 tap = rcu_dereference(vlan->taps[rxq]); 316 goto out; 317 } 318 319 tap = rcu_dereference(vlan->taps[0]); 320 out: 321 return tap; 322 } 323 324 /* 325 * The net_device is going away, give up the reference 326 * that it holds on all queues and safely set the pointer 327 * from the queues to NULL. 328 */ 329 static void macvtap_del_queues(struct net_device *dev) 330 { 331 struct macvlan_dev *vlan = netdev_priv(dev); 332 struct macvtap_queue *q, *tmp; 333 334 ASSERT_RTNL(); 335 list_for_each_entry_safe(q, tmp, &vlan->queue_list, next) { 336 list_del_init(&q->next); 337 RCU_INIT_POINTER(q->vlan, NULL); 338 if (q->enabled) 339 vlan->numvtaps--; 340 vlan->numqueues--; 341 sock_put(&q->sk); 342 } 343 BUG_ON(vlan->numvtaps); 344 BUG_ON(vlan->numqueues); 345 /* guarantee that any future macvtap_set_queue will fail */ 346 vlan->numvtaps = MAX_MACVTAP_QUEUES; 347 } 348 349 static rx_handler_result_t macvtap_handle_frame(struct sk_buff **pskb) 350 { 351 struct sk_buff *skb = *pskb; 352 struct net_device *dev = skb->dev; 353 struct macvlan_dev *vlan; 354 struct macvtap_queue *q; 355 netdev_features_t features = TAP_FEATURES; 356 357 vlan = macvtap_get_vlan_rcu(dev); 358 if (!vlan) 359 return RX_HANDLER_PASS; 360 361 q = macvtap_get_queue(dev, skb); 362 if (!q) 363 return RX_HANDLER_PASS; 364 365 if (skb_queue_len(&q->sk.sk_receive_queue) >= dev->tx_queue_len) 366 goto drop; 367 368 skb_push(skb, ETH_HLEN); 369 370 /* Apply the forward feature mask so that we perform segmentation 371 * according to users wishes. This only works if VNET_HDR is 372 * enabled. 373 */ 374 if (q->flags & IFF_VNET_HDR) 375 features |= vlan->tap_features; 376 if (netif_needs_gso(skb, features)) { 377 struct sk_buff *segs = __skb_gso_segment(skb, features, false); 378 379 if (IS_ERR(segs)) 380 goto drop; 381 382 if (!segs) { 383 skb_queue_tail(&q->sk.sk_receive_queue, skb); 384 goto wake_up; 385 } 386 387 consume_skb(skb); 388 while (segs) { 389 struct sk_buff *nskb = segs->next; 390 391 segs->next = NULL; 392 skb_queue_tail(&q->sk.sk_receive_queue, segs); 393 segs = nskb; 394 } 395 } else { 396 /* If we receive a partial checksum and the tap side 397 * doesn't support checksum offload, compute the checksum. 398 * Note: it doesn't matter which checksum feature to 399 * check, we either support them all or none. 400 */ 401 if (skb->ip_summed == CHECKSUM_PARTIAL && 402 !(features & NETIF_F_CSUM_MASK) && 403 skb_checksum_help(skb)) 404 goto drop; 405 skb_queue_tail(&q->sk.sk_receive_queue, skb); 406 } 407 408 wake_up: 409 wake_up_interruptible_poll(sk_sleep(&q->sk), POLLIN | POLLRDNORM | POLLRDBAND); 410 return RX_HANDLER_CONSUMED; 411 412 drop: 413 /* Count errors/drops only here, thus don't care about args. */ 414 macvlan_count_rx(vlan, 0, 0, 0); 415 kfree_skb(skb); 416 return RX_HANDLER_CONSUMED; 417 } 418 419 static int macvtap_get_minor(struct macvlan_dev *vlan) 420 { 421 int retval = -ENOMEM; 422 423 mutex_lock(&minor_lock); 424 retval = idr_alloc(&minor_idr, vlan, 1, MACVTAP_NUM_DEVS, GFP_KERNEL); 425 if (retval >= 0) { 426 vlan->minor = retval; 427 } else if (retval == -ENOSPC) { 428 printk(KERN_ERR "too many macvtap devices\n"); 429 retval = -EINVAL; 430 } 431 mutex_unlock(&minor_lock); 432 return retval < 0 ? retval : 0; 433 } 434 435 static void macvtap_free_minor(struct macvlan_dev *vlan) 436 { 437 mutex_lock(&minor_lock); 438 if (vlan->minor) { 439 idr_remove(&minor_idr, vlan->minor); 440 vlan->minor = 0; 441 } 442 mutex_unlock(&minor_lock); 443 } 444 445 static struct net_device *dev_get_by_macvtap_minor(int minor) 446 { 447 struct net_device *dev = NULL; 448 struct macvlan_dev *vlan; 449 450 mutex_lock(&minor_lock); 451 vlan = idr_find(&minor_idr, minor); 452 if (vlan) { 453 dev = vlan->dev; 454 dev_hold(dev); 455 } 456 mutex_unlock(&minor_lock); 457 return dev; 458 } 459 460 static int macvtap_newlink(struct net *src_net, 461 struct net_device *dev, 462 struct nlattr *tb[], 463 struct nlattr *data[]) 464 { 465 struct macvlan_dev *vlan = netdev_priv(dev); 466 int err; 467 468 INIT_LIST_HEAD(&vlan->queue_list); 469 470 /* Since macvlan supports all offloads by default, make 471 * tap support all offloads also. 472 */ 473 vlan->tap_features = TUN_OFFLOADS; 474 475 err = netdev_rx_handler_register(dev, macvtap_handle_frame, vlan); 476 if (err) 477 return err; 478 479 /* Don't put anything that may fail after macvlan_common_newlink 480 * because we can't undo what it does. 481 */ 482 return macvlan_common_newlink(src_net, dev, tb, data); 483 } 484 485 static void macvtap_dellink(struct net_device *dev, 486 struct list_head *head) 487 { 488 netdev_rx_handler_unregister(dev); 489 macvtap_del_queues(dev); 490 macvlan_dellink(dev, head); 491 } 492 493 static void macvtap_setup(struct net_device *dev) 494 { 495 macvlan_common_setup(dev); 496 dev->tx_queue_len = TUN_READQ_SIZE; 497 } 498 499 static struct rtnl_link_ops macvtap_link_ops __read_mostly = { 500 .kind = "macvtap", 501 .setup = macvtap_setup, 502 .newlink = macvtap_newlink, 503 .dellink = macvtap_dellink, 504 }; 505 506 507 static void macvtap_sock_write_space(struct sock *sk) 508 { 509 wait_queue_head_t *wqueue; 510 511 if (!sock_writeable(sk) || 512 !test_and_clear_bit(SOCKWQ_ASYNC_NOSPACE, &sk->sk_socket->flags)) 513 return; 514 515 wqueue = sk_sleep(sk); 516 if (wqueue && waitqueue_active(wqueue)) 517 wake_up_interruptible_poll(wqueue, POLLOUT | POLLWRNORM | POLLWRBAND); 518 } 519 520 static void macvtap_sock_destruct(struct sock *sk) 521 { 522 skb_queue_purge(&sk->sk_receive_queue); 523 } 524 525 static int macvtap_open(struct inode *inode, struct file *file) 526 { 527 struct net *net = current->nsproxy->net_ns; 528 struct net_device *dev; 529 struct macvtap_queue *q; 530 int err = -ENODEV; 531 532 rtnl_lock(); 533 dev = dev_get_by_macvtap_minor(iminor(inode)); 534 if (!dev) 535 goto out; 536 537 err = -ENOMEM; 538 q = (struct macvtap_queue *)sk_alloc(net, AF_UNSPEC, GFP_KERNEL, 539 &macvtap_proto, 0); 540 if (!q) 541 goto out; 542 543 RCU_INIT_POINTER(q->sock.wq, &q->wq); 544 init_waitqueue_head(&q->wq.wait); 545 q->sock.type = SOCK_RAW; 546 q->sock.state = SS_CONNECTED; 547 q->sock.file = file; 548 q->sock.ops = &macvtap_socket_ops; 549 sock_init_data(&q->sock, &q->sk); 550 q->sk.sk_write_space = macvtap_sock_write_space; 551 q->sk.sk_destruct = macvtap_sock_destruct; 552 q->flags = IFF_VNET_HDR | IFF_NO_PI | IFF_TAP; 553 q->vnet_hdr_sz = sizeof(struct virtio_net_hdr); 554 555 /* 556 * so far only KVM virtio_net uses macvtap, enable zero copy between 557 * guest kernel and host kernel when lower device supports zerocopy 558 * 559 * The macvlan supports zerocopy iff the lower device supports zero 560 * copy so we don't have to look at the lower device directly. 561 */ 562 if ((dev->features & NETIF_F_HIGHDMA) && (dev->features & NETIF_F_SG)) 563 sock_set_flag(&q->sk, SOCK_ZEROCOPY); 564 565 err = macvtap_set_queue(dev, file, q); 566 if (err) 567 sock_put(&q->sk); 568 569 out: 570 if (dev) 571 dev_put(dev); 572 573 rtnl_unlock(); 574 return err; 575 } 576 577 static int macvtap_release(struct inode *inode, struct file *file) 578 { 579 struct macvtap_queue *q = file->private_data; 580 macvtap_put_queue(q); 581 return 0; 582 } 583 584 static unsigned int macvtap_poll(struct file *file, poll_table * wait) 585 { 586 struct macvtap_queue *q = file->private_data; 587 unsigned int mask = POLLERR; 588 589 if (!q) 590 goto out; 591 592 mask = 0; 593 poll_wait(file, &q->wq.wait, wait); 594 595 if (!skb_queue_empty(&q->sk.sk_receive_queue)) 596 mask |= POLLIN | POLLRDNORM; 597 598 if (sock_writeable(&q->sk) || 599 (!test_and_set_bit(SOCKWQ_ASYNC_NOSPACE, &q->sock.flags) && 600 sock_writeable(&q->sk))) 601 mask |= POLLOUT | POLLWRNORM; 602 603 out: 604 return mask; 605 } 606 607 static inline struct sk_buff *macvtap_alloc_skb(struct sock *sk, size_t prepad, 608 size_t len, size_t linear, 609 int noblock, int *err) 610 { 611 struct sk_buff *skb; 612 613 /* Under a page? Don't bother with paged skb. */ 614 if (prepad + len < PAGE_SIZE || !linear) 615 linear = len; 616 617 skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock, 618 err, 0); 619 if (!skb) 620 return NULL; 621 622 skb_reserve(skb, prepad); 623 skb_put(skb, linear); 624 skb->data_len = len - linear; 625 skb->len += len - linear; 626 627 return skb; 628 } 629 630 /* Neighbour code has some assumptions on HH_DATA_MOD alignment */ 631 #define MACVTAP_RESERVE HH_DATA_OFF(ETH_HLEN) 632 633 /* Get packet from user space buffer */ 634 static ssize_t macvtap_get_user(struct macvtap_queue *q, struct msghdr *m, 635 struct iov_iter *from, int noblock) 636 { 637 int good_linear = SKB_MAX_HEAD(MACVTAP_RESERVE); 638 struct sk_buff *skb; 639 struct macvlan_dev *vlan; 640 unsigned long total_len = iov_iter_count(from); 641 unsigned long len = total_len; 642 int err; 643 struct virtio_net_hdr vnet_hdr = { 0 }; 644 int vnet_hdr_len = 0; 645 int copylen = 0; 646 int depth; 647 bool zerocopy = false; 648 size_t linear; 649 ssize_t n; 650 651 if (q->flags & IFF_VNET_HDR) { 652 vnet_hdr_len = q->vnet_hdr_sz; 653 654 err = -EINVAL; 655 if (len < vnet_hdr_len) 656 goto err; 657 len -= vnet_hdr_len; 658 659 err = -EFAULT; 660 n = copy_from_iter(&vnet_hdr, sizeof(vnet_hdr), from); 661 if (n != sizeof(vnet_hdr)) 662 goto err; 663 iov_iter_advance(from, vnet_hdr_len - sizeof(vnet_hdr)); 664 if ((vnet_hdr.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) && 665 macvtap16_to_cpu(q, vnet_hdr.csum_start) + 666 macvtap16_to_cpu(q, vnet_hdr.csum_offset) + 2 > 667 macvtap16_to_cpu(q, vnet_hdr.hdr_len)) 668 vnet_hdr.hdr_len = cpu_to_macvtap16(q, 669 macvtap16_to_cpu(q, vnet_hdr.csum_start) + 670 macvtap16_to_cpu(q, vnet_hdr.csum_offset) + 2); 671 err = -EINVAL; 672 if (macvtap16_to_cpu(q, vnet_hdr.hdr_len) > len) 673 goto err; 674 } 675 676 err = -EINVAL; 677 if (unlikely(len < ETH_HLEN)) 678 goto err; 679 680 if (m && m->msg_control && sock_flag(&q->sk, SOCK_ZEROCOPY)) { 681 struct iov_iter i; 682 683 copylen = vnet_hdr.hdr_len ? 684 macvtap16_to_cpu(q, vnet_hdr.hdr_len) : GOODCOPY_LEN; 685 if (copylen > good_linear) 686 copylen = good_linear; 687 else if (copylen < ETH_HLEN) 688 copylen = ETH_HLEN; 689 linear = copylen; 690 i = *from; 691 iov_iter_advance(&i, copylen); 692 if (iov_iter_npages(&i, INT_MAX) <= MAX_SKB_FRAGS) 693 zerocopy = true; 694 } 695 696 if (!zerocopy) { 697 copylen = len; 698 linear = macvtap16_to_cpu(q, vnet_hdr.hdr_len); 699 if (linear > good_linear) 700 linear = good_linear; 701 else if (linear < ETH_HLEN) 702 linear = ETH_HLEN; 703 } 704 705 skb = macvtap_alloc_skb(&q->sk, MACVTAP_RESERVE, copylen, 706 linear, noblock, &err); 707 if (!skb) 708 goto err; 709 710 if (zerocopy) 711 err = zerocopy_sg_from_iter(skb, from); 712 else { 713 err = skb_copy_datagram_from_iter(skb, 0, from, len); 714 if (!err && m && m->msg_control) { 715 struct ubuf_info *uarg = m->msg_control; 716 uarg->callback(uarg, false); 717 } 718 } 719 720 if (err) 721 goto err_kfree; 722 723 skb_set_network_header(skb, ETH_HLEN); 724 skb_reset_mac_header(skb); 725 skb->protocol = eth_hdr(skb)->h_proto; 726 727 if (vnet_hdr_len) { 728 err = virtio_net_hdr_to_skb(skb, &vnet_hdr, 729 macvtap_is_little_endian(q)); 730 if (err) 731 goto err_kfree; 732 } 733 734 skb_probe_transport_header(skb, ETH_HLEN); 735 736 /* Move network header to the right position for VLAN tagged packets */ 737 if ((skb->protocol == htons(ETH_P_8021Q) || 738 skb->protocol == htons(ETH_P_8021AD)) && 739 __vlan_get_protocol(skb, skb->protocol, &depth) != 0) 740 skb_set_network_header(skb, depth); 741 742 rcu_read_lock(); 743 vlan = rcu_dereference(q->vlan); 744 /* copy skb_ubuf_info for callback when skb has no error */ 745 if (zerocopy) { 746 skb_shinfo(skb)->destructor_arg = m->msg_control; 747 skb_shinfo(skb)->tx_flags |= SKBTX_DEV_ZEROCOPY; 748 skb_shinfo(skb)->tx_flags |= SKBTX_SHARED_FRAG; 749 } 750 if (vlan) { 751 skb->dev = vlan->dev; 752 dev_queue_xmit(skb); 753 } else { 754 kfree_skb(skb); 755 } 756 rcu_read_unlock(); 757 758 return total_len; 759 760 err_kfree: 761 kfree_skb(skb); 762 763 err: 764 rcu_read_lock(); 765 vlan = rcu_dereference(q->vlan); 766 if (vlan) 767 this_cpu_inc(vlan->pcpu_stats->tx_dropped); 768 rcu_read_unlock(); 769 770 return err; 771 } 772 773 static ssize_t macvtap_write_iter(struct kiocb *iocb, struct iov_iter *from) 774 { 775 struct file *file = iocb->ki_filp; 776 struct macvtap_queue *q = file->private_data; 777 778 return macvtap_get_user(q, NULL, from, file->f_flags & O_NONBLOCK); 779 } 780 781 /* Put packet to the user space buffer */ 782 static ssize_t macvtap_put_user(struct macvtap_queue *q, 783 const struct sk_buff *skb, 784 struct iov_iter *iter) 785 { 786 int ret; 787 int vnet_hdr_len = 0; 788 int vlan_offset = 0; 789 int total; 790 791 if (q->flags & IFF_VNET_HDR) { 792 struct virtio_net_hdr vnet_hdr; 793 vnet_hdr_len = q->vnet_hdr_sz; 794 if (iov_iter_count(iter) < vnet_hdr_len) 795 return -EINVAL; 796 797 ret = virtio_net_hdr_from_skb(skb, &vnet_hdr, 798 macvtap_is_little_endian(q)); 799 if (ret) 800 BUG(); 801 802 if (copy_to_iter(&vnet_hdr, sizeof(vnet_hdr), iter) != 803 sizeof(vnet_hdr)) 804 return -EFAULT; 805 806 iov_iter_advance(iter, vnet_hdr_len - sizeof(vnet_hdr)); 807 } 808 total = vnet_hdr_len; 809 total += skb->len; 810 811 if (skb_vlan_tag_present(skb)) { 812 struct { 813 __be16 h_vlan_proto; 814 __be16 h_vlan_TCI; 815 } veth; 816 veth.h_vlan_proto = skb->vlan_proto; 817 veth.h_vlan_TCI = htons(skb_vlan_tag_get(skb)); 818 819 vlan_offset = offsetof(struct vlan_ethhdr, h_vlan_proto); 820 total += VLAN_HLEN; 821 822 ret = skb_copy_datagram_iter(skb, 0, iter, vlan_offset); 823 if (ret || !iov_iter_count(iter)) 824 goto done; 825 826 ret = copy_to_iter(&veth, sizeof(veth), iter); 827 if (ret != sizeof(veth) || !iov_iter_count(iter)) 828 goto done; 829 } 830 831 ret = skb_copy_datagram_iter(skb, vlan_offset, iter, 832 skb->len - vlan_offset); 833 834 done: 835 return ret ? ret : total; 836 } 837 838 static ssize_t macvtap_do_read(struct macvtap_queue *q, 839 struct iov_iter *to, 840 int noblock) 841 { 842 DEFINE_WAIT(wait); 843 struct sk_buff *skb; 844 ssize_t ret = 0; 845 846 if (!iov_iter_count(to)) 847 return 0; 848 849 while (1) { 850 if (!noblock) 851 prepare_to_wait(sk_sleep(&q->sk), &wait, 852 TASK_INTERRUPTIBLE); 853 854 /* Read frames from the queue */ 855 skb = skb_dequeue(&q->sk.sk_receive_queue); 856 if (skb) 857 break; 858 if (noblock) { 859 ret = -EAGAIN; 860 break; 861 } 862 if (signal_pending(current)) { 863 ret = -ERESTARTSYS; 864 break; 865 } 866 /* Nothing to read, let's sleep */ 867 schedule(); 868 } 869 if (!noblock) 870 finish_wait(sk_sleep(&q->sk), &wait); 871 872 if (skb) { 873 ret = macvtap_put_user(q, skb, to); 874 if (unlikely(ret < 0)) 875 kfree_skb(skb); 876 else 877 consume_skb(skb); 878 } 879 return ret; 880 } 881 882 static ssize_t macvtap_read_iter(struct kiocb *iocb, struct iov_iter *to) 883 { 884 struct file *file = iocb->ki_filp; 885 struct macvtap_queue *q = file->private_data; 886 ssize_t len = iov_iter_count(to), ret; 887 888 ret = macvtap_do_read(q, to, file->f_flags & O_NONBLOCK); 889 ret = min_t(ssize_t, ret, len); 890 if (ret > 0) 891 iocb->ki_pos = ret; 892 return ret; 893 } 894 895 static struct macvlan_dev *macvtap_get_vlan(struct macvtap_queue *q) 896 { 897 struct macvlan_dev *vlan; 898 899 ASSERT_RTNL(); 900 vlan = rtnl_dereference(q->vlan); 901 if (vlan) 902 dev_hold(vlan->dev); 903 904 return vlan; 905 } 906 907 static void macvtap_put_vlan(struct macvlan_dev *vlan) 908 { 909 dev_put(vlan->dev); 910 } 911 912 static int macvtap_ioctl_set_queue(struct file *file, unsigned int flags) 913 { 914 struct macvtap_queue *q = file->private_data; 915 struct macvlan_dev *vlan; 916 int ret; 917 918 vlan = macvtap_get_vlan(q); 919 if (!vlan) 920 return -EINVAL; 921 922 if (flags & IFF_ATTACH_QUEUE) 923 ret = macvtap_enable_queue(vlan->dev, file, q); 924 else if (flags & IFF_DETACH_QUEUE) 925 ret = macvtap_disable_queue(q); 926 else 927 ret = -EINVAL; 928 929 macvtap_put_vlan(vlan); 930 return ret; 931 } 932 933 static int set_offload(struct macvtap_queue *q, unsigned long arg) 934 { 935 struct macvlan_dev *vlan; 936 netdev_features_t features; 937 netdev_features_t feature_mask = 0; 938 939 vlan = rtnl_dereference(q->vlan); 940 if (!vlan) 941 return -ENOLINK; 942 943 features = vlan->dev->features; 944 945 if (arg & TUN_F_CSUM) { 946 feature_mask = NETIF_F_HW_CSUM; 947 948 if (arg & (TUN_F_TSO4 | TUN_F_TSO6)) { 949 if (arg & TUN_F_TSO_ECN) 950 feature_mask |= NETIF_F_TSO_ECN; 951 if (arg & TUN_F_TSO4) 952 feature_mask |= NETIF_F_TSO; 953 if (arg & TUN_F_TSO6) 954 feature_mask |= NETIF_F_TSO6; 955 } 956 957 if (arg & TUN_F_UFO) 958 feature_mask |= NETIF_F_UFO; 959 } 960 961 /* tun/tap driver inverts the usage for TSO offloads, where 962 * setting the TSO bit means that the userspace wants to 963 * accept TSO frames and turning it off means that user space 964 * does not support TSO. 965 * For macvtap, we have to invert it to mean the same thing. 966 * When user space turns off TSO, we turn off GSO/LRO so that 967 * user-space will not receive TSO frames. 968 */ 969 if (feature_mask & (NETIF_F_TSO | NETIF_F_TSO6 | NETIF_F_UFO)) 970 features |= RX_OFFLOADS; 971 else 972 features &= ~RX_OFFLOADS; 973 974 /* tap_features are the same as features on tun/tap and 975 * reflect user expectations. 976 */ 977 vlan->tap_features = feature_mask; 978 vlan->set_features = features; 979 netdev_update_features(vlan->dev); 980 981 return 0; 982 } 983 984 /* 985 * provide compatibility with generic tun/tap interface 986 */ 987 static long macvtap_ioctl(struct file *file, unsigned int cmd, 988 unsigned long arg) 989 { 990 struct macvtap_queue *q = file->private_data; 991 struct macvlan_dev *vlan; 992 void __user *argp = (void __user *)arg; 993 struct ifreq __user *ifr = argp; 994 unsigned int __user *up = argp; 995 unsigned short u; 996 int __user *sp = argp; 997 struct sockaddr sa; 998 int s; 999 int ret; 1000 1001 switch (cmd) { 1002 case TUNSETIFF: 1003 /* ignore the name, just look at flags */ 1004 if (get_user(u, &ifr->ifr_flags)) 1005 return -EFAULT; 1006 1007 ret = 0; 1008 if ((u & ~MACVTAP_FEATURES) != (IFF_NO_PI | IFF_TAP)) 1009 ret = -EINVAL; 1010 else 1011 q->flags = (q->flags & ~MACVTAP_FEATURES) | u; 1012 1013 return ret; 1014 1015 case TUNGETIFF: 1016 rtnl_lock(); 1017 vlan = macvtap_get_vlan(q); 1018 if (!vlan) { 1019 rtnl_unlock(); 1020 return -ENOLINK; 1021 } 1022 1023 ret = 0; 1024 u = q->flags; 1025 if (copy_to_user(&ifr->ifr_name, vlan->dev->name, IFNAMSIZ) || 1026 put_user(u, &ifr->ifr_flags)) 1027 ret = -EFAULT; 1028 macvtap_put_vlan(vlan); 1029 rtnl_unlock(); 1030 return ret; 1031 1032 case TUNSETQUEUE: 1033 if (get_user(u, &ifr->ifr_flags)) 1034 return -EFAULT; 1035 rtnl_lock(); 1036 ret = macvtap_ioctl_set_queue(file, u); 1037 rtnl_unlock(); 1038 return ret; 1039 1040 case TUNGETFEATURES: 1041 if (put_user(IFF_TAP | IFF_NO_PI | MACVTAP_FEATURES, up)) 1042 return -EFAULT; 1043 return 0; 1044 1045 case TUNSETSNDBUF: 1046 if (get_user(s, sp)) 1047 return -EFAULT; 1048 1049 q->sk.sk_sndbuf = s; 1050 return 0; 1051 1052 case TUNGETVNETHDRSZ: 1053 s = q->vnet_hdr_sz; 1054 if (put_user(s, sp)) 1055 return -EFAULT; 1056 return 0; 1057 1058 case TUNSETVNETHDRSZ: 1059 if (get_user(s, sp)) 1060 return -EFAULT; 1061 if (s < (int)sizeof(struct virtio_net_hdr)) 1062 return -EINVAL; 1063 1064 q->vnet_hdr_sz = s; 1065 return 0; 1066 1067 case TUNGETVNETLE: 1068 s = !!(q->flags & MACVTAP_VNET_LE); 1069 if (put_user(s, sp)) 1070 return -EFAULT; 1071 return 0; 1072 1073 case TUNSETVNETLE: 1074 if (get_user(s, sp)) 1075 return -EFAULT; 1076 if (s) 1077 q->flags |= MACVTAP_VNET_LE; 1078 else 1079 q->flags &= ~MACVTAP_VNET_LE; 1080 return 0; 1081 1082 case TUNGETVNETBE: 1083 return macvtap_get_vnet_be(q, sp); 1084 1085 case TUNSETVNETBE: 1086 return macvtap_set_vnet_be(q, sp); 1087 1088 case TUNSETOFFLOAD: 1089 /* let the user check for future flags */ 1090 if (arg & ~(TUN_F_CSUM | TUN_F_TSO4 | TUN_F_TSO6 | 1091 TUN_F_TSO_ECN | TUN_F_UFO)) 1092 return -EINVAL; 1093 1094 rtnl_lock(); 1095 ret = set_offload(q, arg); 1096 rtnl_unlock(); 1097 return ret; 1098 1099 case SIOCGIFHWADDR: 1100 rtnl_lock(); 1101 vlan = macvtap_get_vlan(q); 1102 if (!vlan) { 1103 rtnl_unlock(); 1104 return -ENOLINK; 1105 } 1106 ret = 0; 1107 u = vlan->dev->type; 1108 if (copy_to_user(&ifr->ifr_name, vlan->dev->name, IFNAMSIZ) || 1109 copy_to_user(&ifr->ifr_hwaddr.sa_data, vlan->dev->dev_addr, ETH_ALEN) || 1110 put_user(u, &ifr->ifr_hwaddr.sa_family)) 1111 ret = -EFAULT; 1112 macvtap_put_vlan(vlan); 1113 rtnl_unlock(); 1114 return ret; 1115 1116 case SIOCSIFHWADDR: 1117 if (copy_from_user(&sa, &ifr->ifr_hwaddr, sizeof(sa))) 1118 return -EFAULT; 1119 rtnl_lock(); 1120 vlan = macvtap_get_vlan(q); 1121 if (!vlan) { 1122 rtnl_unlock(); 1123 return -ENOLINK; 1124 } 1125 ret = dev_set_mac_address(vlan->dev, &sa); 1126 macvtap_put_vlan(vlan); 1127 rtnl_unlock(); 1128 return ret; 1129 1130 default: 1131 return -EINVAL; 1132 } 1133 } 1134 1135 #ifdef CONFIG_COMPAT 1136 static long macvtap_compat_ioctl(struct file *file, unsigned int cmd, 1137 unsigned long arg) 1138 { 1139 return macvtap_ioctl(file, cmd, (unsigned long)compat_ptr(arg)); 1140 } 1141 #endif 1142 1143 static const struct file_operations macvtap_fops = { 1144 .owner = THIS_MODULE, 1145 .open = macvtap_open, 1146 .release = macvtap_release, 1147 .read_iter = macvtap_read_iter, 1148 .write_iter = macvtap_write_iter, 1149 .poll = macvtap_poll, 1150 .llseek = no_llseek, 1151 .unlocked_ioctl = macvtap_ioctl, 1152 #ifdef CONFIG_COMPAT 1153 .compat_ioctl = macvtap_compat_ioctl, 1154 #endif 1155 }; 1156 1157 static int macvtap_sendmsg(struct socket *sock, struct msghdr *m, 1158 size_t total_len) 1159 { 1160 struct macvtap_queue *q = container_of(sock, struct macvtap_queue, sock); 1161 return macvtap_get_user(q, m, &m->msg_iter, m->msg_flags & MSG_DONTWAIT); 1162 } 1163 1164 static int macvtap_recvmsg(struct socket *sock, struct msghdr *m, 1165 size_t total_len, int flags) 1166 { 1167 struct macvtap_queue *q = container_of(sock, struct macvtap_queue, sock); 1168 int ret; 1169 if (flags & ~(MSG_DONTWAIT|MSG_TRUNC)) 1170 return -EINVAL; 1171 ret = macvtap_do_read(q, &m->msg_iter, flags & MSG_DONTWAIT); 1172 if (ret > total_len) { 1173 m->msg_flags |= MSG_TRUNC; 1174 ret = flags & MSG_TRUNC ? ret : total_len; 1175 } 1176 return ret; 1177 } 1178 1179 /* Ops structure to mimic raw sockets with tun */ 1180 static const struct proto_ops macvtap_socket_ops = { 1181 .sendmsg = macvtap_sendmsg, 1182 .recvmsg = macvtap_recvmsg, 1183 }; 1184 1185 /* Get an underlying socket object from tun file. Returns error unless file is 1186 * attached to a device. The returned object works like a packet socket, it 1187 * can be used for sock_sendmsg/sock_recvmsg. The caller is responsible for 1188 * holding a reference to the file for as long as the socket is in use. */ 1189 struct socket *macvtap_get_socket(struct file *file) 1190 { 1191 struct macvtap_queue *q; 1192 if (file->f_op != &macvtap_fops) 1193 return ERR_PTR(-EINVAL); 1194 q = file->private_data; 1195 if (!q) 1196 return ERR_PTR(-EBADFD); 1197 return &q->sock; 1198 } 1199 EXPORT_SYMBOL_GPL(macvtap_get_socket); 1200 1201 static int macvtap_device_event(struct notifier_block *unused, 1202 unsigned long event, void *ptr) 1203 { 1204 struct net_device *dev = netdev_notifier_info_to_dev(ptr); 1205 struct macvlan_dev *vlan; 1206 struct device *classdev; 1207 dev_t devt; 1208 int err; 1209 char tap_name[IFNAMSIZ]; 1210 1211 if (dev->rtnl_link_ops != &macvtap_link_ops) 1212 return NOTIFY_DONE; 1213 1214 snprintf(tap_name, IFNAMSIZ, "tap%d", dev->ifindex); 1215 vlan = netdev_priv(dev); 1216 1217 switch (event) { 1218 case NETDEV_REGISTER: 1219 /* Create the device node here after the network device has 1220 * been registered but before register_netdevice has 1221 * finished running. 1222 */ 1223 err = macvtap_get_minor(vlan); 1224 if (err) 1225 return notifier_from_errno(err); 1226 1227 devt = MKDEV(MAJOR(macvtap_major), vlan->minor); 1228 classdev = device_create(&macvtap_class, &dev->dev, devt, 1229 dev, tap_name); 1230 if (IS_ERR(classdev)) { 1231 macvtap_free_minor(vlan); 1232 return notifier_from_errno(PTR_ERR(classdev)); 1233 } 1234 err = sysfs_create_link(&dev->dev.kobj, &classdev->kobj, 1235 tap_name); 1236 if (err) 1237 return notifier_from_errno(err); 1238 break; 1239 case NETDEV_UNREGISTER: 1240 /* vlan->minor == 0 if NETDEV_REGISTER above failed */ 1241 if (vlan->minor == 0) 1242 break; 1243 sysfs_remove_link(&dev->dev.kobj, tap_name); 1244 devt = MKDEV(MAJOR(macvtap_major), vlan->minor); 1245 device_destroy(&macvtap_class, devt); 1246 macvtap_free_minor(vlan); 1247 break; 1248 } 1249 1250 return NOTIFY_DONE; 1251 } 1252 1253 static struct notifier_block macvtap_notifier_block __read_mostly = { 1254 .notifier_call = macvtap_device_event, 1255 }; 1256 1257 static int macvtap_init(void) 1258 { 1259 int err; 1260 1261 err = alloc_chrdev_region(&macvtap_major, 0, 1262 MACVTAP_NUM_DEVS, "macvtap"); 1263 if (err) 1264 goto out1; 1265 1266 cdev_init(&macvtap_cdev, &macvtap_fops); 1267 err = cdev_add(&macvtap_cdev, macvtap_major, MACVTAP_NUM_DEVS); 1268 if (err) 1269 goto out2; 1270 1271 err = class_register(&macvtap_class); 1272 if (err) 1273 goto out3; 1274 1275 err = register_netdevice_notifier(&macvtap_notifier_block); 1276 if (err) 1277 goto out4; 1278 1279 err = macvlan_link_register(&macvtap_link_ops); 1280 if (err) 1281 goto out5; 1282 1283 return 0; 1284 1285 out5: 1286 unregister_netdevice_notifier(&macvtap_notifier_block); 1287 out4: 1288 class_unregister(&macvtap_class); 1289 out3: 1290 cdev_del(&macvtap_cdev); 1291 out2: 1292 unregister_chrdev_region(macvtap_major, MACVTAP_NUM_DEVS); 1293 out1: 1294 return err; 1295 } 1296 module_init(macvtap_init); 1297 1298 static void macvtap_exit(void) 1299 { 1300 rtnl_link_unregister(&macvtap_link_ops); 1301 unregister_netdevice_notifier(&macvtap_notifier_block); 1302 class_unregister(&macvtap_class); 1303 cdev_del(&macvtap_cdev); 1304 unregister_chrdev_region(macvtap_major, MACVTAP_NUM_DEVS); 1305 idr_destroy(&minor_idr); 1306 } 1307 module_exit(macvtap_exit); 1308 1309 MODULE_ALIAS_RTNL_LINK("macvtap"); 1310 MODULE_AUTHOR("Arnd Bergmann <arnd@arndb.de>"); 1311 MODULE_LICENSE("GPL"); 1312