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 #include <linux/skb_array.h> 25 26 /* 27 * A macvtap queue is the central object of this driver, it connects 28 * an open character device to a macvlan interface. There can be 29 * multiple queues on one interface, which map back to queues 30 * implemented in hardware on the underlying device. 31 * 32 * macvtap_proto is used to allocate queues through the sock allocation 33 * mechanism. 34 * 35 */ 36 struct macvtap_queue { 37 struct sock sk; 38 struct socket sock; 39 struct socket_wq wq; 40 int vnet_hdr_sz; 41 struct macvlan_dev __rcu *vlan; 42 struct file *file; 43 unsigned int flags; 44 u16 queue_index; 45 bool enabled; 46 struct list_head next; 47 struct skb_array skb_array; 48 }; 49 50 #define MACVTAP_FEATURES (IFF_VNET_HDR | IFF_MULTI_QUEUE) 51 52 #define MACVTAP_VNET_LE 0x80000000 53 #define MACVTAP_VNET_BE 0x40000000 54 55 #ifdef CONFIG_TUN_VNET_CROSS_LE 56 static inline bool macvtap_legacy_is_little_endian(struct macvtap_queue *q) 57 { 58 return q->flags & MACVTAP_VNET_BE ? false : 59 virtio_legacy_is_little_endian(); 60 } 61 62 static long macvtap_get_vnet_be(struct macvtap_queue *q, int __user *sp) 63 { 64 int s = !!(q->flags & MACVTAP_VNET_BE); 65 66 if (put_user(s, sp)) 67 return -EFAULT; 68 69 return 0; 70 } 71 72 static long macvtap_set_vnet_be(struct macvtap_queue *q, int __user *sp) 73 { 74 int s; 75 76 if (get_user(s, sp)) 77 return -EFAULT; 78 79 if (s) 80 q->flags |= MACVTAP_VNET_BE; 81 else 82 q->flags &= ~MACVTAP_VNET_BE; 83 84 return 0; 85 } 86 #else 87 static inline bool macvtap_legacy_is_little_endian(struct macvtap_queue *q) 88 { 89 return virtio_legacy_is_little_endian(); 90 } 91 92 static long macvtap_get_vnet_be(struct macvtap_queue *q, int __user *argp) 93 { 94 return -EINVAL; 95 } 96 97 static long macvtap_set_vnet_be(struct macvtap_queue *q, int __user *argp) 98 { 99 return -EINVAL; 100 } 101 #endif /* CONFIG_TUN_VNET_CROSS_LE */ 102 103 static inline bool macvtap_is_little_endian(struct macvtap_queue *q) 104 { 105 return q->flags & MACVTAP_VNET_LE || 106 macvtap_legacy_is_little_endian(q); 107 } 108 109 static inline u16 macvtap16_to_cpu(struct macvtap_queue *q, __virtio16 val) 110 { 111 return __virtio16_to_cpu(macvtap_is_little_endian(q), val); 112 } 113 114 static inline __virtio16 cpu_to_macvtap16(struct macvtap_queue *q, u16 val) 115 { 116 return __cpu_to_virtio16(macvtap_is_little_endian(q), val); 117 } 118 119 static struct proto macvtap_proto = { 120 .name = "macvtap", 121 .owner = THIS_MODULE, 122 .obj_size = sizeof (struct macvtap_queue), 123 }; 124 125 /* 126 * Variables for dealing with macvtaps device numbers. 127 */ 128 static dev_t macvtap_major; 129 #define MACVTAP_NUM_DEVS (1U << MINORBITS) 130 static DEFINE_MUTEX(minor_lock); 131 static DEFINE_IDR(minor_idr); 132 133 #define GOODCOPY_LEN 128 134 static const void *macvtap_net_namespace(struct device *d) 135 { 136 struct net_device *dev = to_net_dev(d->parent); 137 return dev_net(dev); 138 } 139 140 static struct class macvtap_class = { 141 .name = "macvtap", 142 .owner = THIS_MODULE, 143 .ns_type = &net_ns_type_operations, 144 .namespace = macvtap_net_namespace, 145 }; 146 static struct cdev macvtap_cdev; 147 148 static const struct proto_ops macvtap_socket_ops; 149 150 #define TUN_OFFLOADS (NETIF_F_HW_CSUM | NETIF_F_TSO_ECN | NETIF_F_TSO | \ 151 NETIF_F_TSO6 | NETIF_F_UFO) 152 #define RX_OFFLOADS (NETIF_F_GRO | NETIF_F_LRO) 153 #define TAP_FEATURES (NETIF_F_GSO | NETIF_F_SG | NETIF_F_FRAGLIST) 154 155 static struct macvlan_dev *macvtap_get_vlan_rcu(const struct net_device *dev) 156 { 157 return rcu_dereference(dev->rx_handler_data); 158 } 159 160 /* 161 * RCU usage: 162 * The macvtap_queue and the macvlan_dev are loosely coupled, the 163 * pointers from one to the other can only be read while rcu_read_lock 164 * or rtnl is held. 165 * 166 * Both the file and the macvlan_dev hold a reference on the macvtap_queue 167 * through sock_hold(&q->sk). When the macvlan_dev goes away first, 168 * q->vlan becomes inaccessible. When the files gets closed, 169 * macvtap_get_queue() fails. 170 * 171 * There may still be references to the struct sock inside of the 172 * queue from outbound SKBs, but these never reference back to the 173 * file or the dev. The data structure is freed through __sk_free 174 * when both our references and any pending SKBs are gone. 175 */ 176 177 static int macvtap_enable_queue(struct net_device *dev, struct file *file, 178 struct macvtap_queue *q) 179 { 180 struct macvlan_dev *vlan = netdev_priv(dev); 181 int err = -EINVAL; 182 183 ASSERT_RTNL(); 184 185 if (q->enabled) 186 goto out; 187 188 err = 0; 189 rcu_assign_pointer(vlan->taps[vlan->numvtaps], q); 190 q->queue_index = vlan->numvtaps; 191 q->enabled = true; 192 193 vlan->numvtaps++; 194 out: 195 return err; 196 } 197 198 /* Requires RTNL */ 199 static int macvtap_set_queue(struct net_device *dev, struct file *file, 200 struct macvtap_queue *q) 201 { 202 struct macvlan_dev *vlan = netdev_priv(dev); 203 204 if (vlan->numqueues == MAX_MACVTAP_QUEUES) 205 return -EBUSY; 206 207 rcu_assign_pointer(q->vlan, vlan); 208 rcu_assign_pointer(vlan->taps[vlan->numvtaps], q); 209 sock_hold(&q->sk); 210 211 q->file = file; 212 q->queue_index = vlan->numvtaps; 213 q->enabled = true; 214 file->private_data = q; 215 list_add_tail(&q->next, &vlan->queue_list); 216 217 vlan->numvtaps++; 218 vlan->numqueues++; 219 220 return 0; 221 } 222 223 static int macvtap_disable_queue(struct macvtap_queue *q) 224 { 225 struct macvlan_dev *vlan; 226 struct macvtap_queue *nq; 227 228 ASSERT_RTNL(); 229 if (!q->enabled) 230 return -EINVAL; 231 232 vlan = rtnl_dereference(q->vlan); 233 234 if (vlan) { 235 int index = q->queue_index; 236 BUG_ON(index >= vlan->numvtaps); 237 nq = rtnl_dereference(vlan->taps[vlan->numvtaps - 1]); 238 nq->queue_index = index; 239 240 rcu_assign_pointer(vlan->taps[index], nq); 241 RCU_INIT_POINTER(vlan->taps[vlan->numvtaps - 1], NULL); 242 q->enabled = false; 243 244 vlan->numvtaps--; 245 } 246 247 return 0; 248 } 249 250 /* 251 * The file owning the queue got closed, give up both 252 * the reference that the files holds as well as the 253 * one from the macvlan_dev if that still exists. 254 * 255 * Using the spinlock makes sure that we don't get 256 * to the queue again after destroying it. 257 */ 258 static void macvtap_put_queue(struct macvtap_queue *q) 259 { 260 struct macvlan_dev *vlan; 261 262 rtnl_lock(); 263 vlan = rtnl_dereference(q->vlan); 264 265 if (vlan) { 266 if (q->enabled) 267 BUG_ON(macvtap_disable_queue(q)); 268 269 vlan->numqueues--; 270 RCU_INIT_POINTER(q->vlan, NULL); 271 sock_put(&q->sk); 272 list_del_init(&q->next); 273 } 274 275 rtnl_unlock(); 276 277 synchronize_rcu(); 278 sock_put(&q->sk); 279 } 280 281 /* 282 * Select a queue based on the rxq of the device on which this packet 283 * arrived. If the incoming device is not mq, calculate a flow hash 284 * to select a queue. If all fails, find the first available queue. 285 * Cache vlan->numvtaps since it can become zero during the execution 286 * of this function. 287 */ 288 static struct macvtap_queue *macvtap_get_queue(struct net_device *dev, 289 struct sk_buff *skb) 290 { 291 struct macvlan_dev *vlan = netdev_priv(dev); 292 struct macvtap_queue *tap = NULL; 293 /* Access to taps array is protected by rcu, but access to numvtaps 294 * isn't. Below we use it to lookup a queue, but treat it as a hint 295 * and validate that the result isn't NULL - in case we are 296 * racing against queue removal. 297 */ 298 int numvtaps = ACCESS_ONCE(vlan->numvtaps); 299 __u32 rxq; 300 301 if (!numvtaps) 302 goto out; 303 304 if (numvtaps == 1) 305 goto single; 306 307 /* Check if we can use flow to select a queue */ 308 rxq = skb_get_hash(skb); 309 if (rxq) { 310 tap = rcu_dereference(vlan->taps[rxq % numvtaps]); 311 goto out; 312 } 313 314 if (likely(skb_rx_queue_recorded(skb))) { 315 rxq = skb_get_rx_queue(skb); 316 317 while (unlikely(rxq >= numvtaps)) 318 rxq -= numvtaps; 319 320 tap = rcu_dereference(vlan->taps[rxq]); 321 goto out; 322 } 323 324 single: 325 tap = rcu_dereference(vlan->taps[0]); 326 out: 327 return tap; 328 } 329 330 /* 331 * The net_device is going away, give up the reference 332 * that it holds on all queues and safely set the pointer 333 * from the queues to NULL. 334 */ 335 static void macvtap_del_queues(struct net_device *dev) 336 { 337 struct macvlan_dev *vlan = netdev_priv(dev); 338 struct macvtap_queue *q, *tmp; 339 340 ASSERT_RTNL(); 341 list_for_each_entry_safe(q, tmp, &vlan->queue_list, next) { 342 list_del_init(&q->next); 343 RCU_INIT_POINTER(q->vlan, NULL); 344 if (q->enabled) 345 vlan->numvtaps--; 346 vlan->numqueues--; 347 sock_put(&q->sk); 348 } 349 BUG_ON(vlan->numvtaps); 350 BUG_ON(vlan->numqueues); 351 /* guarantee that any future macvtap_set_queue will fail */ 352 vlan->numvtaps = MAX_MACVTAP_QUEUES; 353 } 354 355 static rx_handler_result_t macvtap_handle_frame(struct sk_buff **pskb) 356 { 357 struct sk_buff *skb = *pskb; 358 struct net_device *dev = skb->dev; 359 struct macvlan_dev *vlan; 360 struct macvtap_queue *q; 361 netdev_features_t features = TAP_FEATURES; 362 363 vlan = macvtap_get_vlan_rcu(dev); 364 if (!vlan) 365 return RX_HANDLER_PASS; 366 367 q = macvtap_get_queue(dev, skb); 368 if (!q) 369 return RX_HANDLER_PASS; 370 371 if (__skb_array_full(&q->skb_array)) 372 goto drop; 373 374 skb_push(skb, ETH_HLEN); 375 376 /* Apply the forward feature mask so that we perform segmentation 377 * according to users wishes. This only works if VNET_HDR is 378 * enabled. 379 */ 380 if (q->flags & IFF_VNET_HDR) 381 features |= vlan->tap_features; 382 if (netif_needs_gso(skb, features)) { 383 struct sk_buff *segs = __skb_gso_segment(skb, features, false); 384 385 if (IS_ERR(segs)) 386 goto drop; 387 388 if (!segs) { 389 if (skb_array_produce(&q->skb_array, skb)) 390 goto drop; 391 goto wake_up; 392 } 393 394 consume_skb(skb); 395 while (segs) { 396 struct sk_buff *nskb = segs->next; 397 398 segs->next = NULL; 399 if (skb_array_produce(&q->skb_array, segs)) { 400 kfree_skb(segs); 401 kfree_skb_list(nskb); 402 break; 403 } 404 segs = nskb; 405 } 406 } else { 407 /* If we receive a partial checksum and the tap side 408 * doesn't support checksum offload, compute the checksum. 409 * Note: it doesn't matter which checksum feature to 410 * check, we either support them all or none. 411 */ 412 if (skb->ip_summed == CHECKSUM_PARTIAL && 413 !(features & NETIF_F_CSUM_MASK) && 414 skb_checksum_help(skb)) 415 goto drop; 416 if (skb_array_produce(&q->skb_array, skb)) 417 goto drop; 418 } 419 420 wake_up: 421 wake_up_interruptible_poll(sk_sleep(&q->sk), POLLIN | POLLRDNORM | POLLRDBAND); 422 return RX_HANDLER_CONSUMED; 423 424 drop: 425 /* Count errors/drops only here, thus don't care about args. */ 426 macvlan_count_rx(vlan, 0, 0, 0); 427 kfree_skb(skb); 428 return RX_HANDLER_CONSUMED; 429 } 430 431 static int macvtap_get_minor(struct macvlan_dev *vlan) 432 { 433 int retval = -ENOMEM; 434 435 mutex_lock(&minor_lock); 436 retval = idr_alloc(&minor_idr, vlan, 1, MACVTAP_NUM_DEVS, GFP_KERNEL); 437 if (retval >= 0) { 438 vlan->minor = retval; 439 } else if (retval == -ENOSPC) { 440 printk(KERN_ERR "too many macvtap devices\n"); 441 retval = -EINVAL; 442 } 443 mutex_unlock(&minor_lock); 444 return retval < 0 ? retval : 0; 445 } 446 447 static void macvtap_free_minor(struct macvlan_dev *vlan) 448 { 449 mutex_lock(&minor_lock); 450 if (vlan->minor) { 451 idr_remove(&minor_idr, vlan->minor); 452 vlan->minor = 0; 453 } 454 mutex_unlock(&minor_lock); 455 } 456 457 static struct net_device *dev_get_by_macvtap_minor(int minor) 458 { 459 struct net_device *dev = NULL; 460 struct macvlan_dev *vlan; 461 462 mutex_lock(&minor_lock); 463 vlan = idr_find(&minor_idr, minor); 464 if (vlan) { 465 dev = vlan->dev; 466 dev_hold(dev); 467 } 468 mutex_unlock(&minor_lock); 469 return dev; 470 } 471 472 static int macvtap_newlink(struct net *src_net, 473 struct net_device *dev, 474 struct nlattr *tb[], 475 struct nlattr *data[]) 476 { 477 struct macvlan_dev *vlan = netdev_priv(dev); 478 int err; 479 480 INIT_LIST_HEAD(&vlan->queue_list); 481 482 /* Since macvlan supports all offloads by default, make 483 * tap support all offloads also. 484 */ 485 vlan->tap_features = TUN_OFFLOADS; 486 487 err = netdev_rx_handler_register(dev, macvtap_handle_frame, vlan); 488 if (err) 489 return err; 490 491 /* Don't put anything that may fail after macvlan_common_newlink 492 * because we can't undo what it does. 493 */ 494 return macvlan_common_newlink(src_net, dev, tb, data); 495 } 496 497 static void macvtap_dellink(struct net_device *dev, 498 struct list_head *head) 499 { 500 netdev_rx_handler_unregister(dev); 501 macvtap_del_queues(dev); 502 macvlan_dellink(dev, head); 503 } 504 505 static void macvtap_setup(struct net_device *dev) 506 { 507 macvlan_common_setup(dev); 508 dev->tx_queue_len = TUN_READQ_SIZE; 509 } 510 511 static struct rtnl_link_ops macvtap_link_ops __read_mostly = { 512 .kind = "macvtap", 513 .setup = macvtap_setup, 514 .newlink = macvtap_newlink, 515 .dellink = macvtap_dellink, 516 }; 517 518 519 static void macvtap_sock_write_space(struct sock *sk) 520 { 521 wait_queue_head_t *wqueue; 522 523 if (!sock_writeable(sk) || 524 !test_and_clear_bit(SOCKWQ_ASYNC_NOSPACE, &sk->sk_socket->flags)) 525 return; 526 527 wqueue = sk_sleep(sk); 528 if (wqueue && waitqueue_active(wqueue)) 529 wake_up_interruptible_poll(wqueue, POLLOUT | POLLWRNORM | POLLWRBAND); 530 } 531 532 static void macvtap_sock_destruct(struct sock *sk) 533 { 534 struct macvtap_queue *q = container_of(sk, struct macvtap_queue, sk); 535 536 skb_array_cleanup(&q->skb_array); 537 } 538 539 static int macvtap_open(struct inode *inode, struct file *file) 540 { 541 struct net *net = current->nsproxy->net_ns; 542 struct net_device *dev; 543 struct macvtap_queue *q; 544 int err = -ENODEV; 545 546 rtnl_lock(); 547 dev = dev_get_by_macvtap_minor(iminor(inode)); 548 if (!dev) 549 goto err; 550 551 err = -ENOMEM; 552 q = (struct macvtap_queue *)sk_alloc(net, AF_UNSPEC, GFP_KERNEL, 553 &macvtap_proto, 0); 554 if (!q) 555 goto err; 556 557 RCU_INIT_POINTER(q->sock.wq, &q->wq); 558 init_waitqueue_head(&q->wq.wait); 559 q->sock.type = SOCK_RAW; 560 q->sock.state = SS_CONNECTED; 561 q->sock.file = file; 562 q->sock.ops = &macvtap_socket_ops; 563 sock_init_data(&q->sock, &q->sk); 564 q->sk.sk_write_space = macvtap_sock_write_space; 565 q->sk.sk_destruct = macvtap_sock_destruct; 566 q->flags = IFF_VNET_HDR | IFF_NO_PI | IFF_TAP; 567 q->vnet_hdr_sz = sizeof(struct virtio_net_hdr); 568 569 /* 570 * so far only KVM virtio_net uses macvtap, enable zero copy between 571 * guest kernel and host kernel when lower device supports zerocopy 572 * 573 * The macvlan supports zerocopy iff the lower device supports zero 574 * copy so we don't have to look at the lower device directly. 575 */ 576 if ((dev->features & NETIF_F_HIGHDMA) && (dev->features & NETIF_F_SG)) 577 sock_set_flag(&q->sk, SOCK_ZEROCOPY); 578 579 err = -ENOMEM; 580 if (skb_array_init(&q->skb_array, dev->tx_queue_len, GFP_KERNEL)) 581 goto err_array; 582 583 err = macvtap_set_queue(dev, file, q); 584 if (err) 585 goto err_queue; 586 587 dev_put(dev); 588 589 rtnl_unlock(); 590 return err; 591 592 err_queue: 593 skb_array_cleanup(&q->skb_array); 594 err_array: 595 sock_put(&q->sk); 596 err: 597 if (dev) 598 dev_put(dev); 599 600 rtnl_unlock(); 601 return err; 602 } 603 604 static int macvtap_release(struct inode *inode, struct file *file) 605 { 606 struct macvtap_queue *q = file->private_data; 607 macvtap_put_queue(q); 608 return 0; 609 } 610 611 static unsigned int macvtap_poll(struct file *file, poll_table * wait) 612 { 613 struct macvtap_queue *q = file->private_data; 614 unsigned int mask = POLLERR; 615 616 if (!q) 617 goto out; 618 619 mask = 0; 620 poll_wait(file, &q->wq.wait, wait); 621 622 if (!skb_array_empty(&q->skb_array)) 623 mask |= POLLIN | POLLRDNORM; 624 625 if (sock_writeable(&q->sk) || 626 (!test_and_set_bit(SOCKWQ_ASYNC_NOSPACE, &q->sock.flags) && 627 sock_writeable(&q->sk))) 628 mask |= POLLOUT | POLLWRNORM; 629 630 out: 631 return mask; 632 } 633 634 static inline struct sk_buff *macvtap_alloc_skb(struct sock *sk, size_t prepad, 635 size_t len, size_t linear, 636 int noblock, int *err) 637 { 638 struct sk_buff *skb; 639 640 /* Under a page? Don't bother with paged skb. */ 641 if (prepad + len < PAGE_SIZE || !linear) 642 linear = len; 643 644 skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock, 645 err, 0); 646 if (!skb) 647 return NULL; 648 649 skb_reserve(skb, prepad); 650 skb_put(skb, linear); 651 skb->data_len = len - linear; 652 skb->len += len - linear; 653 654 return skb; 655 } 656 657 /* Neighbour code has some assumptions on HH_DATA_MOD alignment */ 658 #define MACVTAP_RESERVE HH_DATA_OFF(ETH_HLEN) 659 660 /* Get packet from user space buffer */ 661 static ssize_t macvtap_get_user(struct macvtap_queue *q, struct msghdr *m, 662 struct iov_iter *from, int noblock) 663 { 664 int good_linear = SKB_MAX_HEAD(MACVTAP_RESERVE); 665 struct sk_buff *skb; 666 struct macvlan_dev *vlan; 667 unsigned long total_len = iov_iter_count(from); 668 unsigned long len = total_len; 669 int err; 670 struct virtio_net_hdr vnet_hdr = { 0 }; 671 int vnet_hdr_len = 0; 672 int copylen = 0; 673 int depth; 674 bool zerocopy = false; 675 size_t linear; 676 ssize_t n; 677 678 if (q->flags & IFF_VNET_HDR) { 679 vnet_hdr_len = q->vnet_hdr_sz; 680 681 err = -EINVAL; 682 if (len < vnet_hdr_len) 683 goto err; 684 len -= vnet_hdr_len; 685 686 err = -EFAULT; 687 n = copy_from_iter(&vnet_hdr, sizeof(vnet_hdr), from); 688 if (n != sizeof(vnet_hdr)) 689 goto err; 690 iov_iter_advance(from, vnet_hdr_len - sizeof(vnet_hdr)); 691 if ((vnet_hdr.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) && 692 macvtap16_to_cpu(q, vnet_hdr.csum_start) + 693 macvtap16_to_cpu(q, vnet_hdr.csum_offset) + 2 > 694 macvtap16_to_cpu(q, vnet_hdr.hdr_len)) 695 vnet_hdr.hdr_len = cpu_to_macvtap16(q, 696 macvtap16_to_cpu(q, vnet_hdr.csum_start) + 697 macvtap16_to_cpu(q, vnet_hdr.csum_offset) + 2); 698 err = -EINVAL; 699 if (macvtap16_to_cpu(q, vnet_hdr.hdr_len) > len) 700 goto err; 701 } 702 703 err = -EINVAL; 704 if (unlikely(len < ETH_HLEN)) 705 goto err; 706 707 if (m && m->msg_control && sock_flag(&q->sk, SOCK_ZEROCOPY)) { 708 struct iov_iter i; 709 710 copylen = vnet_hdr.hdr_len ? 711 macvtap16_to_cpu(q, vnet_hdr.hdr_len) : GOODCOPY_LEN; 712 if (copylen > good_linear) 713 copylen = good_linear; 714 else if (copylen < ETH_HLEN) 715 copylen = ETH_HLEN; 716 linear = copylen; 717 i = *from; 718 iov_iter_advance(&i, copylen); 719 if (iov_iter_npages(&i, INT_MAX) <= MAX_SKB_FRAGS) 720 zerocopy = true; 721 } 722 723 if (!zerocopy) { 724 copylen = len; 725 linear = macvtap16_to_cpu(q, vnet_hdr.hdr_len); 726 if (linear > good_linear) 727 linear = good_linear; 728 else if (linear < ETH_HLEN) 729 linear = ETH_HLEN; 730 } 731 732 skb = macvtap_alloc_skb(&q->sk, MACVTAP_RESERVE, copylen, 733 linear, noblock, &err); 734 if (!skb) 735 goto err; 736 737 if (zerocopy) 738 err = zerocopy_sg_from_iter(skb, from); 739 else { 740 err = skb_copy_datagram_from_iter(skb, 0, from, len); 741 if (!err && m && m->msg_control) { 742 struct ubuf_info *uarg = m->msg_control; 743 uarg->callback(uarg, false); 744 } 745 } 746 747 if (err) 748 goto err_kfree; 749 750 skb_set_network_header(skb, ETH_HLEN); 751 skb_reset_mac_header(skb); 752 skb->protocol = eth_hdr(skb)->h_proto; 753 754 if (vnet_hdr_len) { 755 err = virtio_net_hdr_to_skb(skb, &vnet_hdr, 756 macvtap_is_little_endian(q)); 757 if (err) 758 goto err_kfree; 759 } 760 761 skb_probe_transport_header(skb, ETH_HLEN); 762 763 /* Move network header to the right position for VLAN tagged packets */ 764 if ((skb->protocol == htons(ETH_P_8021Q) || 765 skb->protocol == htons(ETH_P_8021AD)) && 766 __vlan_get_protocol(skb, skb->protocol, &depth) != 0) 767 skb_set_network_header(skb, depth); 768 769 rcu_read_lock(); 770 vlan = rcu_dereference(q->vlan); 771 /* copy skb_ubuf_info for callback when skb has no error */ 772 if (zerocopy) { 773 skb_shinfo(skb)->destructor_arg = m->msg_control; 774 skb_shinfo(skb)->tx_flags |= SKBTX_DEV_ZEROCOPY; 775 skb_shinfo(skb)->tx_flags |= SKBTX_SHARED_FRAG; 776 } 777 if (vlan) { 778 skb->dev = vlan->dev; 779 dev_queue_xmit(skb); 780 } else { 781 kfree_skb(skb); 782 } 783 rcu_read_unlock(); 784 785 return total_len; 786 787 err_kfree: 788 kfree_skb(skb); 789 790 err: 791 rcu_read_lock(); 792 vlan = rcu_dereference(q->vlan); 793 if (vlan) 794 this_cpu_inc(vlan->pcpu_stats->tx_dropped); 795 rcu_read_unlock(); 796 797 return err; 798 } 799 800 static ssize_t macvtap_write_iter(struct kiocb *iocb, struct iov_iter *from) 801 { 802 struct file *file = iocb->ki_filp; 803 struct macvtap_queue *q = file->private_data; 804 805 return macvtap_get_user(q, NULL, from, file->f_flags & O_NONBLOCK); 806 } 807 808 /* Put packet to the user space buffer */ 809 static ssize_t macvtap_put_user(struct macvtap_queue *q, 810 const struct sk_buff *skb, 811 struct iov_iter *iter) 812 { 813 int ret; 814 int vnet_hdr_len = 0; 815 int vlan_offset = 0; 816 int total; 817 818 if (q->flags & IFF_VNET_HDR) { 819 struct virtio_net_hdr vnet_hdr; 820 vnet_hdr_len = q->vnet_hdr_sz; 821 if (iov_iter_count(iter) < vnet_hdr_len) 822 return -EINVAL; 823 824 ret = virtio_net_hdr_from_skb(skb, &vnet_hdr, 825 macvtap_is_little_endian(q)); 826 if (ret) 827 BUG(); 828 829 if (copy_to_iter(&vnet_hdr, sizeof(vnet_hdr), iter) != 830 sizeof(vnet_hdr)) 831 return -EFAULT; 832 833 iov_iter_advance(iter, vnet_hdr_len - sizeof(vnet_hdr)); 834 } 835 total = vnet_hdr_len; 836 total += skb->len; 837 838 if (skb_vlan_tag_present(skb)) { 839 struct { 840 __be16 h_vlan_proto; 841 __be16 h_vlan_TCI; 842 } veth; 843 veth.h_vlan_proto = skb->vlan_proto; 844 veth.h_vlan_TCI = htons(skb_vlan_tag_get(skb)); 845 846 vlan_offset = offsetof(struct vlan_ethhdr, h_vlan_proto); 847 total += VLAN_HLEN; 848 849 ret = skb_copy_datagram_iter(skb, 0, iter, vlan_offset); 850 if (ret || !iov_iter_count(iter)) 851 goto done; 852 853 ret = copy_to_iter(&veth, sizeof(veth), iter); 854 if (ret != sizeof(veth) || !iov_iter_count(iter)) 855 goto done; 856 } 857 858 ret = skb_copy_datagram_iter(skb, vlan_offset, iter, 859 skb->len - vlan_offset); 860 861 done: 862 return ret ? ret : total; 863 } 864 865 static ssize_t macvtap_do_read(struct macvtap_queue *q, 866 struct iov_iter *to, 867 int noblock) 868 { 869 DEFINE_WAIT(wait); 870 struct sk_buff *skb; 871 ssize_t ret = 0; 872 873 if (!iov_iter_count(to)) 874 return 0; 875 876 while (1) { 877 if (!noblock) 878 prepare_to_wait(sk_sleep(&q->sk), &wait, 879 TASK_INTERRUPTIBLE); 880 881 /* Read frames from the queue */ 882 skb = skb_array_consume(&q->skb_array); 883 if (skb) 884 break; 885 if (noblock) { 886 ret = -EAGAIN; 887 break; 888 } 889 if (signal_pending(current)) { 890 ret = -ERESTARTSYS; 891 break; 892 } 893 /* Nothing to read, let's sleep */ 894 schedule(); 895 } 896 if (!noblock) 897 finish_wait(sk_sleep(&q->sk), &wait); 898 899 if (skb) { 900 ret = macvtap_put_user(q, skb, to); 901 if (unlikely(ret < 0)) 902 kfree_skb(skb); 903 else 904 consume_skb(skb); 905 } 906 return ret; 907 } 908 909 static ssize_t macvtap_read_iter(struct kiocb *iocb, struct iov_iter *to) 910 { 911 struct file *file = iocb->ki_filp; 912 struct macvtap_queue *q = file->private_data; 913 ssize_t len = iov_iter_count(to), ret; 914 915 ret = macvtap_do_read(q, to, file->f_flags & O_NONBLOCK); 916 ret = min_t(ssize_t, ret, len); 917 if (ret > 0) 918 iocb->ki_pos = ret; 919 return ret; 920 } 921 922 static struct macvlan_dev *macvtap_get_vlan(struct macvtap_queue *q) 923 { 924 struct macvlan_dev *vlan; 925 926 ASSERT_RTNL(); 927 vlan = rtnl_dereference(q->vlan); 928 if (vlan) 929 dev_hold(vlan->dev); 930 931 return vlan; 932 } 933 934 static void macvtap_put_vlan(struct macvlan_dev *vlan) 935 { 936 dev_put(vlan->dev); 937 } 938 939 static int macvtap_ioctl_set_queue(struct file *file, unsigned int flags) 940 { 941 struct macvtap_queue *q = file->private_data; 942 struct macvlan_dev *vlan; 943 int ret; 944 945 vlan = macvtap_get_vlan(q); 946 if (!vlan) 947 return -EINVAL; 948 949 if (flags & IFF_ATTACH_QUEUE) 950 ret = macvtap_enable_queue(vlan->dev, file, q); 951 else if (flags & IFF_DETACH_QUEUE) 952 ret = macvtap_disable_queue(q); 953 else 954 ret = -EINVAL; 955 956 macvtap_put_vlan(vlan); 957 return ret; 958 } 959 960 static int set_offload(struct macvtap_queue *q, unsigned long arg) 961 { 962 struct macvlan_dev *vlan; 963 netdev_features_t features; 964 netdev_features_t feature_mask = 0; 965 966 vlan = rtnl_dereference(q->vlan); 967 if (!vlan) 968 return -ENOLINK; 969 970 features = vlan->dev->features; 971 972 if (arg & TUN_F_CSUM) { 973 feature_mask = NETIF_F_HW_CSUM; 974 975 if (arg & (TUN_F_TSO4 | TUN_F_TSO6)) { 976 if (arg & TUN_F_TSO_ECN) 977 feature_mask |= NETIF_F_TSO_ECN; 978 if (arg & TUN_F_TSO4) 979 feature_mask |= NETIF_F_TSO; 980 if (arg & TUN_F_TSO6) 981 feature_mask |= NETIF_F_TSO6; 982 } 983 984 if (arg & TUN_F_UFO) 985 feature_mask |= NETIF_F_UFO; 986 } 987 988 /* tun/tap driver inverts the usage for TSO offloads, where 989 * setting the TSO bit means that the userspace wants to 990 * accept TSO frames and turning it off means that user space 991 * does not support TSO. 992 * For macvtap, we have to invert it to mean the same thing. 993 * When user space turns off TSO, we turn off GSO/LRO so that 994 * user-space will not receive TSO frames. 995 */ 996 if (feature_mask & (NETIF_F_TSO | NETIF_F_TSO6 | NETIF_F_UFO)) 997 features |= RX_OFFLOADS; 998 else 999 features &= ~RX_OFFLOADS; 1000 1001 /* tap_features are the same as features on tun/tap and 1002 * reflect user expectations. 1003 */ 1004 vlan->tap_features = feature_mask; 1005 vlan->set_features = features; 1006 netdev_update_features(vlan->dev); 1007 1008 return 0; 1009 } 1010 1011 /* 1012 * provide compatibility with generic tun/tap interface 1013 */ 1014 static long macvtap_ioctl(struct file *file, unsigned int cmd, 1015 unsigned long arg) 1016 { 1017 struct macvtap_queue *q = file->private_data; 1018 struct macvlan_dev *vlan; 1019 void __user *argp = (void __user *)arg; 1020 struct ifreq __user *ifr = argp; 1021 unsigned int __user *up = argp; 1022 unsigned short u; 1023 int __user *sp = argp; 1024 struct sockaddr sa; 1025 int s; 1026 int ret; 1027 1028 switch (cmd) { 1029 case TUNSETIFF: 1030 /* ignore the name, just look at flags */ 1031 if (get_user(u, &ifr->ifr_flags)) 1032 return -EFAULT; 1033 1034 ret = 0; 1035 if ((u & ~MACVTAP_FEATURES) != (IFF_NO_PI | IFF_TAP)) 1036 ret = -EINVAL; 1037 else 1038 q->flags = (q->flags & ~MACVTAP_FEATURES) | u; 1039 1040 return ret; 1041 1042 case TUNGETIFF: 1043 rtnl_lock(); 1044 vlan = macvtap_get_vlan(q); 1045 if (!vlan) { 1046 rtnl_unlock(); 1047 return -ENOLINK; 1048 } 1049 1050 ret = 0; 1051 u = q->flags; 1052 if (copy_to_user(&ifr->ifr_name, vlan->dev->name, IFNAMSIZ) || 1053 put_user(u, &ifr->ifr_flags)) 1054 ret = -EFAULT; 1055 macvtap_put_vlan(vlan); 1056 rtnl_unlock(); 1057 return ret; 1058 1059 case TUNSETQUEUE: 1060 if (get_user(u, &ifr->ifr_flags)) 1061 return -EFAULT; 1062 rtnl_lock(); 1063 ret = macvtap_ioctl_set_queue(file, u); 1064 rtnl_unlock(); 1065 return ret; 1066 1067 case TUNGETFEATURES: 1068 if (put_user(IFF_TAP | IFF_NO_PI | MACVTAP_FEATURES, up)) 1069 return -EFAULT; 1070 return 0; 1071 1072 case TUNSETSNDBUF: 1073 if (get_user(s, sp)) 1074 return -EFAULT; 1075 1076 q->sk.sk_sndbuf = s; 1077 return 0; 1078 1079 case TUNGETVNETHDRSZ: 1080 s = q->vnet_hdr_sz; 1081 if (put_user(s, sp)) 1082 return -EFAULT; 1083 return 0; 1084 1085 case TUNSETVNETHDRSZ: 1086 if (get_user(s, sp)) 1087 return -EFAULT; 1088 if (s < (int)sizeof(struct virtio_net_hdr)) 1089 return -EINVAL; 1090 1091 q->vnet_hdr_sz = s; 1092 return 0; 1093 1094 case TUNGETVNETLE: 1095 s = !!(q->flags & MACVTAP_VNET_LE); 1096 if (put_user(s, sp)) 1097 return -EFAULT; 1098 return 0; 1099 1100 case TUNSETVNETLE: 1101 if (get_user(s, sp)) 1102 return -EFAULT; 1103 if (s) 1104 q->flags |= MACVTAP_VNET_LE; 1105 else 1106 q->flags &= ~MACVTAP_VNET_LE; 1107 return 0; 1108 1109 case TUNGETVNETBE: 1110 return macvtap_get_vnet_be(q, sp); 1111 1112 case TUNSETVNETBE: 1113 return macvtap_set_vnet_be(q, sp); 1114 1115 case TUNSETOFFLOAD: 1116 /* let the user check for future flags */ 1117 if (arg & ~(TUN_F_CSUM | TUN_F_TSO4 | TUN_F_TSO6 | 1118 TUN_F_TSO_ECN | TUN_F_UFO)) 1119 return -EINVAL; 1120 1121 rtnl_lock(); 1122 ret = set_offload(q, arg); 1123 rtnl_unlock(); 1124 return ret; 1125 1126 case SIOCGIFHWADDR: 1127 rtnl_lock(); 1128 vlan = macvtap_get_vlan(q); 1129 if (!vlan) { 1130 rtnl_unlock(); 1131 return -ENOLINK; 1132 } 1133 ret = 0; 1134 u = vlan->dev->type; 1135 if (copy_to_user(&ifr->ifr_name, vlan->dev->name, IFNAMSIZ) || 1136 copy_to_user(&ifr->ifr_hwaddr.sa_data, vlan->dev->dev_addr, ETH_ALEN) || 1137 put_user(u, &ifr->ifr_hwaddr.sa_family)) 1138 ret = -EFAULT; 1139 macvtap_put_vlan(vlan); 1140 rtnl_unlock(); 1141 return ret; 1142 1143 case SIOCSIFHWADDR: 1144 if (copy_from_user(&sa, &ifr->ifr_hwaddr, sizeof(sa))) 1145 return -EFAULT; 1146 rtnl_lock(); 1147 vlan = macvtap_get_vlan(q); 1148 if (!vlan) { 1149 rtnl_unlock(); 1150 return -ENOLINK; 1151 } 1152 ret = dev_set_mac_address(vlan->dev, &sa); 1153 macvtap_put_vlan(vlan); 1154 rtnl_unlock(); 1155 return ret; 1156 1157 default: 1158 return -EINVAL; 1159 } 1160 } 1161 1162 #ifdef CONFIG_COMPAT 1163 static long macvtap_compat_ioctl(struct file *file, unsigned int cmd, 1164 unsigned long arg) 1165 { 1166 return macvtap_ioctl(file, cmd, (unsigned long)compat_ptr(arg)); 1167 } 1168 #endif 1169 1170 static const struct file_operations macvtap_fops = { 1171 .owner = THIS_MODULE, 1172 .open = macvtap_open, 1173 .release = macvtap_release, 1174 .read_iter = macvtap_read_iter, 1175 .write_iter = macvtap_write_iter, 1176 .poll = macvtap_poll, 1177 .llseek = no_llseek, 1178 .unlocked_ioctl = macvtap_ioctl, 1179 #ifdef CONFIG_COMPAT 1180 .compat_ioctl = macvtap_compat_ioctl, 1181 #endif 1182 }; 1183 1184 static int macvtap_sendmsg(struct socket *sock, struct msghdr *m, 1185 size_t total_len) 1186 { 1187 struct macvtap_queue *q = container_of(sock, struct macvtap_queue, sock); 1188 return macvtap_get_user(q, m, &m->msg_iter, m->msg_flags & MSG_DONTWAIT); 1189 } 1190 1191 static int macvtap_recvmsg(struct socket *sock, struct msghdr *m, 1192 size_t total_len, int flags) 1193 { 1194 struct macvtap_queue *q = container_of(sock, struct macvtap_queue, sock); 1195 int ret; 1196 if (flags & ~(MSG_DONTWAIT|MSG_TRUNC)) 1197 return -EINVAL; 1198 ret = macvtap_do_read(q, &m->msg_iter, flags & MSG_DONTWAIT); 1199 if (ret > total_len) { 1200 m->msg_flags |= MSG_TRUNC; 1201 ret = flags & MSG_TRUNC ? ret : total_len; 1202 } 1203 return ret; 1204 } 1205 1206 static int macvtap_peek_len(struct socket *sock) 1207 { 1208 struct macvtap_queue *q = container_of(sock, struct macvtap_queue, 1209 sock); 1210 return skb_array_peek_len(&q->skb_array); 1211 } 1212 1213 /* Ops structure to mimic raw sockets with tun */ 1214 static const struct proto_ops macvtap_socket_ops = { 1215 .sendmsg = macvtap_sendmsg, 1216 .recvmsg = macvtap_recvmsg, 1217 .peek_len = macvtap_peek_len, 1218 }; 1219 1220 /* Get an underlying socket object from tun file. Returns error unless file is 1221 * attached to a device. The returned object works like a packet socket, it 1222 * can be used for sock_sendmsg/sock_recvmsg. The caller is responsible for 1223 * holding a reference to the file for as long as the socket is in use. */ 1224 struct socket *macvtap_get_socket(struct file *file) 1225 { 1226 struct macvtap_queue *q; 1227 if (file->f_op != &macvtap_fops) 1228 return ERR_PTR(-EINVAL); 1229 q = file->private_data; 1230 if (!q) 1231 return ERR_PTR(-EBADFD); 1232 return &q->sock; 1233 } 1234 EXPORT_SYMBOL_GPL(macvtap_get_socket); 1235 1236 static int macvtap_queue_resize(struct macvlan_dev *vlan) 1237 { 1238 struct net_device *dev = vlan->dev; 1239 struct macvtap_queue *q; 1240 struct skb_array **arrays; 1241 int n = vlan->numqueues; 1242 int ret, i = 0; 1243 1244 arrays = kmalloc(sizeof *arrays * n, GFP_KERNEL); 1245 if (!arrays) 1246 return -ENOMEM; 1247 1248 list_for_each_entry(q, &vlan->queue_list, next) 1249 arrays[i++] = &q->skb_array; 1250 1251 ret = skb_array_resize_multiple(arrays, n, 1252 dev->tx_queue_len, GFP_KERNEL); 1253 1254 kfree(arrays); 1255 return ret; 1256 } 1257 1258 static int macvtap_device_event(struct notifier_block *unused, 1259 unsigned long event, void *ptr) 1260 { 1261 struct net_device *dev = netdev_notifier_info_to_dev(ptr); 1262 struct macvlan_dev *vlan; 1263 struct device *classdev; 1264 dev_t devt; 1265 int err; 1266 char tap_name[IFNAMSIZ]; 1267 1268 if (dev->rtnl_link_ops != &macvtap_link_ops) 1269 return NOTIFY_DONE; 1270 1271 snprintf(tap_name, IFNAMSIZ, "tap%d", dev->ifindex); 1272 vlan = netdev_priv(dev); 1273 1274 switch (event) { 1275 case NETDEV_REGISTER: 1276 /* Create the device node here after the network device has 1277 * been registered but before register_netdevice has 1278 * finished running. 1279 */ 1280 err = macvtap_get_minor(vlan); 1281 if (err) 1282 return notifier_from_errno(err); 1283 1284 devt = MKDEV(MAJOR(macvtap_major), vlan->minor); 1285 classdev = device_create(&macvtap_class, &dev->dev, devt, 1286 dev, tap_name); 1287 if (IS_ERR(classdev)) { 1288 macvtap_free_minor(vlan); 1289 return notifier_from_errno(PTR_ERR(classdev)); 1290 } 1291 err = sysfs_create_link(&dev->dev.kobj, &classdev->kobj, 1292 tap_name); 1293 if (err) 1294 return notifier_from_errno(err); 1295 break; 1296 case NETDEV_UNREGISTER: 1297 /* vlan->minor == 0 if NETDEV_REGISTER above failed */ 1298 if (vlan->minor == 0) 1299 break; 1300 sysfs_remove_link(&dev->dev.kobj, tap_name); 1301 devt = MKDEV(MAJOR(macvtap_major), vlan->minor); 1302 device_destroy(&macvtap_class, devt); 1303 macvtap_free_minor(vlan); 1304 break; 1305 case NETDEV_CHANGE_TX_QUEUE_LEN: 1306 if (macvtap_queue_resize(vlan)) 1307 return NOTIFY_BAD; 1308 break; 1309 } 1310 1311 return NOTIFY_DONE; 1312 } 1313 1314 static struct notifier_block macvtap_notifier_block __read_mostly = { 1315 .notifier_call = macvtap_device_event, 1316 }; 1317 1318 static int macvtap_init(void) 1319 { 1320 int err; 1321 1322 err = alloc_chrdev_region(&macvtap_major, 0, 1323 MACVTAP_NUM_DEVS, "macvtap"); 1324 if (err) 1325 goto out1; 1326 1327 cdev_init(&macvtap_cdev, &macvtap_fops); 1328 err = cdev_add(&macvtap_cdev, macvtap_major, MACVTAP_NUM_DEVS); 1329 if (err) 1330 goto out2; 1331 1332 err = class_register(&macvtap_class); 1333 if (err) 1334 goto out3; 1335 1336 err = register_netdevice_notifier(&macvtap_notifier_block); 1337 if (err) 1338 goto out4; 1339 1340 err = macvlan_link_register(&macvtap_link_ops); 1341 if (err) 1342 goto out5; 1343 1344 return 0; 1345 1346 out5: 1347 unregister_netdevice_notifier(&macvtap_notifier_block); 1348 out4: 1349 class_unregister(&macvtap_class); 1350 out3: 1351 cdev_del(&macvtap_cdev); 1352 out2: 1353 unregister_chrdev_region(macvtap_major, MACVTAP_NUM_DEVS); 1354 out1: 1355 return err; 1356 } 1357 module_init(macvtap_init); 1358 1359 static void macvtap_exit(void) 1360 { 1361 rtnl_link_unregister(&macvtap_link_ops); 1362 unregister_netdevice_notifier(&macvtap_notifier_block); 1363 class_unregister(&macvtap_class); 1364 cdev_del(&macvtap_cdev); 1365 unregister_chrdev_region(macvtap_major, MACVTAP_NUM_DEVS); 1366 idr_destroy(&minor_idr); 1367 } 1368 module_exit(macvtap_exit); 1369 1370 MODULE_ALIAS_RTNL_LINK("macvtap"); 1371 MODULE_AUTHOR("Arnd Bergmann <arnd@arndb.de>"); 1372 MODULE_LICENSE("GPL"); 1373