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 skb_array_cleanup(&q->skb_array); 279 sock_put(&q->sk); 280 } 281 282 /* 283 * Select a queue based on the rxq of the device on which this packet 284 * arrived. If the incoming device is not mq, calculate a flow hash 285 * to select a queue. If all fails, find the first available queue. 286 * Cache vlan->numvtaps since it can become zero during the execution 287 * of this function. 288 */ 289 static struct macvtap_queue *macvtap_get_queue(struct net_device *dev, 290 struct sk_buff *skb) 291 { 292 struct macvlan_dev *vlan = netdev_priv(dev); 293 struct macvtap_queue *tap = NULL; 294 /* Access to taps array is protected by rcu, but access to numvtaps 295 * isn't. Below we use it to lookup a queue, but treat it as a hint 296 * and validate that the result isn't NULL - in case we are 297 * racing against queue removal. 298 */ 299 int numvtaps = ACCESS_ONCE(vlan->numvtaps); 300 __u32 rxq; 301 302 if (!numvtaps) 303 goto out; 304 305 if (numvtaps == 1) 306 goto single; 307 308 /* Check if we can use flow to select a queue */ 309 rxq = skb_get_hash(skb); 310 if (rxq) { 311 tap = rcu_dereference(vlan->taps[rxq % numvtaps]); 312 goto out; 313 } 314 315 if (likely(skb_rx_queue_recorded(skb))) { 316 rxq = skb_get_rx_queue(skb); 317 318 while (unlikely(rxq >= numvtaps)) 319 rxq -= numvtaps; 320 321 tap = rcu_dereference(vlan->taps[rxq]); 322 goto out; 323 } 324 325 single: 326 tap = rcu_dereference(vlan->taps[0]); 327 out: 328 return tap; 329 } 330 331 /* 332 * The net_device is going away, give up the reference 333 * that it holds on all queues and safely set the pointer 334 * from the queues to NULL. 335 */ 336 static void macvtap_del_queues(struct net_device *dev) 337 { 338 struct macvlan_dev *vlan = netdev_priv(dev); 339 struct macvtap_queue *q, *tmp; 340 341 ASSERT_RTNL(); 342 list_for_each_entry_safe(q, tmp, &vlan->queue_list, next) { 343 list_del_init(&q->next); 344 RCU_INIT_POINTER(q->vlan, NULL); 345 if (q->enabled) 346 vlan->numvtaps--; 347 vlan->numqueues--; 348 sock_put(&q->sk); 349 } 350 BUG_ON(vlan->numvtaps); 351 BUG_ON(vlan->numqueues); 352 /* guarantee that any future macvtap_set_queue will fail */ 353 vlan->numvtaps = MAX_MACVTAP_QUEUES; 354 } 355 356 static rx_handler_result_t macvtap_handle_frame(struct sk_buff **pskb) 357 { 358 struct sk_buff *skb = *pskb; 359 struct net_device *dev = skb->dev; 360 struct macvlan_dev *vlan; 361 struct macvtap_queue *q; 362 netdev_features_t features = TAP_FEATURES; 363 364 vlan = macvtap_get_vlan_rcu(dev); 365 if (!vlan) 366 return RX_HANDLER_PASS; 367 368 q = macvtap_get_queue(dev, skb); 369 if (!q) 370 return RX_HANDLER_PASS; 371 372 if (__skb_array_full(&q->skb_array)) 373 goto drop; 374 375 skb_push(skb, ETH_HLEN); 376 377 /* Apply the forward feature mask so that we perform segmentation 378 * according to users wishes. This only works if VNET_HDR is 379 * enabled. 380 */ 381 if (q->flags & IFF_VNET_HDR) 382 features |= vlan->tap_features; 383 if (netif_needs_gso(skb, features)) { 384 struct sk_buff *segs = __skb_gso_segment(skb, features, false); 385 386 if (IS_ERR(segs)) 387 goto drop; 388 389 if (!segs) { 390 if (skb_array_produce(&q->skb_array, skb)) 391 goto drop; 392 goto wake_up; 393 } 394 395 consume_skb(skb); 396 while (segs) { 397 struct sk_buff *nskb = segs->next; 398 399 segs->next = NULL; 400 if (skb_array_produce(&q->skb_array, segs)) { 401 kfree_skb(segs); 402 kfree_skb_list(nskb); 403 break; 404 } 405 segs = nskb; 406 } 407 } else { 408 /* If we receive a partial checksum and the tap side 409 * doesn't support checksum offload, compute the checksum. 410 * Note: it doesn't matter which checksum feature to 411 * check, we either support them all or none. 412 */ 413 if (skb->ip_summed == CHECKSUM_PARTIAL && 414 !(features & NETIF_F_CSUM_MASK) && 415 skb_checksum_help(skb)) 416 goto drop; 417 if (skb_array_produce(&q->skb_array, skb)) 418 goto drop; 419 } 420 421 wake_up: 422 wake_up_interruptible_poll(sk_sleep(&q->sk), POLLIN | POLLRDNORM | POLLRDBAND); 423 return RX_HANDLER_CONSUMED; 424 425 drop: 426 /* Count errors/drops only here, thus don't care about args. */ 427 macvlan_count_rx(vlan, 0, 0, 0); 428 kfree_skb(skb); 429 return RX_HANDLER_CONSUMED; 430 } 431 432 static int macvtap_get_minor(struct macvlan_dev *vlan) 433 { 434 int retval = -ENOMEM; 435 436 mutex_lock(&minor_lock); 437 retval = idr_alloc(&minor_idr, vlan, 1, MACVTAP_NUM_DEVS, GFP_KERNEL); 438 if (retval >= 0) { 439 vlan->minor = retval; 440 } else if (retval == -ENOSPC) { 441 printk(KERN_ERR "too many macvtap devices\n"); 442 retval = -EINVAL; 443 } 444 mutex_unlock(&minor_lock); 445 return retval < 0 ? retval : 0; 446 } 447 448 static void macvtap_free_minor(struct macvlan_dev *vlan) 449 { 450 mutex_lock(&minor_lock); 451 if (vlan->minor) { 452 idr_remove(&minor_idr, vlan->minor); 453 vlan->minor = 0; 454 } 455 mutex_unlock(&minor_lock); 456 } 457 458 static struct net_device *dev_get_by_macvtap_minor(int minor) 459 { 460 struct net_device *dev = NULL; 461 struct macvlan_dev *vlan; 462 463 mutex_lock(&minor_lock); 464 vlan = idr_find(&minor_idr, minor); 465 if (vlan) { 466 dev = vlan->dev; 467 dev_hold(dev); 468 } 469 mutex_unlock(&minor_lock); 470 return dev; 471 } 472 473 static int macvtap_newlink(struct net *src_net, 474 struct net_device *dev, 475 struct nlattr *tb[], 476 struct nlattr *data[]) 477 { 478 struct macvlan_dev *vlan = netdev_priv(dev); 479 int err; 480 481 INIT_LIST_HEAD(&vlan->queue_list); 482 483 /* Since macvlan supports all offloads by default, make 484 * tap support all offloads also. 485 */ 486 vlan->tap_features = TUN_OFFLOADS; 487 488 err = netdev_rx_handler_register(dev, macvtap_handle_frame, vlan); 489 if (err) 490 return err; 491 492 /* Don't put anything that may fail after macvlan_common_newlink 493 * because we can't undo what it does. 494 */ 495 return macvlan_common_newlink(src_net, dev, tb, data); 496 } 497 498 static void macvtap_dellink(struct net_device *dev, 499 struct list_head *head) 500 { 501 netdev_rx_handler_unregister(dev); 502 macvtap_del_queues(dev); 503 macvlan_dellink(dev, head); 504 } 505 506 static void macvtap_setup(struct net_device *dev) 507 { 508 macvlan_common_setup(dev); 509 dev->tx_queue_len = TUN_READQ_SIZE; 510 } 511 512 static struct rtnl_link_ops macvtap_link_ops __read_mostly = { 513 .kind = "macvtap", 514 .setup = macvtap_setup, 515 .newlink = macvtap_newlink, 516 .dellink = macvtap_dellink, 517 }; 518 519 520 static void macvtap_sock_write_space(struct sock *sk) 521 { 522 wait_queue_head_t *wqueue; 523 524 if (!sock_writeable(sk) || 525 !test_and_clear_bit(SOCKWQ_ASYNC_NOSPACE, &sk->sk_socket->flags)) 526 return; 527 528 wqueue = sk_sleep(sk); 529 if (wqueue && waitqueue_active(wqueue)) 530 wake_up_interruptible_poll(wqueue, POLLOUT | POLLWRNORM | POLLWRBAND); 531 } 532 533 static void macvtap_sock_destruct(struct sock *sk) 534 { 535 struct macvtap_queue *q = container_of(sk, struct macvtap_queue, sk); 536 struct sk_buff *skb; 537 538 while ((skb = skb_array_consume(&q->skb_array)) != NULL) 539 kfree_skb(skb); 540 } 541 542 static int macvtap_open(struct inode *inode, struct file *file) 543 { 544 struct net *net = current->nsproxy->net_ns; 545 struct net_device *dev; 546 struct macvtap_queue *q; 547 int err = -ENODEV; 548 549 rtnl_lock(); 550 dev = dev_get_by_macvtap_minor(iminor(inode)); 551 if (!dev) 552 goto err; 553 554 err = -ENOMEM; 555 q = (struct macvtap_queue *)sk_alloc(net, AF_UNSPEC, GFP_KERNEL, 556 &macvtap_proto, 0); 557 if (!q) 558 goto err; 559 560 RCU_INIT_POINTER(q->sock.wq, &q->wq); 561 init_waitqueue_head(&q->wq.wait); 562 q->sock.type = SOCK_RAW; 563 q->sock.state = SS_CONNECTED; 564 q->sock.file = file; 565 q->sock.ops = &macvtap_socket_ops; 566 sock_init_data(&q->sock, &q->sk); 567 q->sk.sk_write_space = macvtap_sock_write_space; 568 q->sk.sk_destruct = macvtap_sock_destruct; 569 q->flags = IFF_VNET_HDR | IFF_NO_PI | IFF_TAP; 570 q->vnet_hdr_sz = sizeof(struct virtio_net_hdr); 571 572 /* 573 * so far only KVM virtio_net uses macvtap, enable zero copy between 574 * guest kernel and host kernel when lower device supports zerocopy 575 * 576 * The macvlan supports zerocopy iff the lower device supports zero 577 * copy so we don't have to look at the lower device directly. 578 */ 579 if ((dev->features & NETIF_F_HIGHDMA) && (dev->features & NETIF_F_SG)) 580 sock_set_flag(&q->sk, SOCK_ZEROCOPY); 581 582 err = -ENOMEM; 583 if (skb_array_init(&q->skb_array, dev->tx_queue_len, GFP_KERNEL)) 584 goto err_array; 585 586 err = macvtap_set_queue(dev, file, q); 587 if (err) 588 goto err_queue; 589 590 dev_put(dev); 591 592 rtnl_unlock(); 593 return err; 594 595 err_queue: 596 skb_array_cleanup(&q->skb_array); 597 err_array: 598 sock_put(&q->sk); 599 err: 600 if (dev) 601 dev_put(dev); 602 603 rtnl_unlock(); 604 return err; 605 } 606 607 static int macvtap_release(struct inode *inode, struct file *file) 608 { 609 struct macvtap_queue *q = file->private_data; 610 macvtap_put_queue(q); 611 return 0; 612 } 613 614 static unsigned int macvtap_poll(struct file *file, poll_table * wait) 615 { 616 struct macvtap_queue *q = file->private_data; 617 unsigned int mask = POLLERR; 618 619 if (!q) 620 goto out; 621 622 mask = 0; 623 poll_wait(file, &q->wq.wait, wait); 624 625 if (!skb_array_empty(&q->skb_array)) 626 mask |= POLLIN | POLLRDNORM; 627 628 if (sock_writeable(&q->sk) || 629 (!test_and_set_bit(SOCKWQ_ASYNC_NOSPACE, &q->sock.flags) && 630 sock_writeable(&q->sk))) 631 mask |= POLLOUT | POLLWRNORM; 632 633 out: 634 return mask; 635 } 636 637 static inline struct sk_buff *macvtap_alloc_skb(struct sock *sk, size_t prepad, 638 size_t len, size_t linear, 639 int noblock, int *err) 640 { 641 struct sk_buff *skb; 642 643 /* Under a page? Don't bother with paged skb. */ 644 if (prepad + len < PAGE_SIZE || !linear) 645 linear = len; 646 647 skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock, 648 err, 0); 649 if (!skb) 650 return NULL; 651 652 skb_reserve(skb, prepad); 653 skb_put(skb, linear); 654 skb->data_len = len - linear; 655 skb->len += len - linear; 656 657 return skb; 658 } 659 660 /* Neighbour code has some assumptions on HH_DATA_MOD alignment */ 661 #define MACVTAP_RESERVE HH_DATA_OFF(ETH_HLEN) 662 663 /* Get packet from user space buffer */ 664 static ssize_t macvtap_get_user(struct macvtap_queue *q, struct msghdr *m, 665 struct iov_iter *from, int noblock) 666 { 667 int good_linear = SKB_MAX_HEAD(MACVTAP_RESERVE); 668 struct sk_buff *skb; 669 struct macvlan_dev *vlan; 670 unsigned long total_len = iov_iter_count(from); 671 unsigned long len = total_len; 672 int err; 673 struct virtio_net_hdr vnet_hdr = { 0 }; 674 int vnet_hdr_len = 0; 675 int copylen = 0; 676 int depth; 677 bool zerocopy = false; 678 size_t linear; 679 ssize_t n; 680 681 if (q->flags & IFF_VNET_HDR) { 682 vnet_hdr_len = q->vnet_hdr_sz; 683 684 err = -EINVAL; 685 if (len < vnet_hdr_len) 686 goto err; 687 len -= vnet_hdr_len; 688 689 err = -EFAULT; 690 n = copy_from_iter(&vnet_hdr, sizeof(vnet_hdr), from); 691 if (n != sizeof(vnet_hdr)) 692 goto err; 693 iov_iter_advance(from, vnet_hdr_len - sizeof(vnet_hdr)); 694 if ((vnet_hdr.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) && 695 macvtap16_to_cpu(q, vnet_hdr.csum_start) + 696 macvtap16_to_cpu(q, vnet_hdr.csum_offset) + 2 > 697 macvtap16_to_cpu(q, vnet_hdr.hdr_len)) 698 vnet_hdr.hdr_len = cpu_to_macvtap16(q, 699 macvtap16_to_cpu(q, vnet_hdr.csum_start) + 700 macvtap16_to_cpu(q, vnet_hdr.csum_offset) + 2); 701 err = -EINVAL; 702 if (macvtap16_to_cpu(q, vnet_hdr.hdr_len) > len) 703 goto err; 704 } 705 706 err = -EINVAL; 707 if (unlikely(len < ETH_HLEN)) 708 goto err; 709 710 if (m && m->msg_control && sock_flag(&q->sk, SOCK_ZEROCOPY)) { 711 struct iov_iter i; 712 713 copylen = vnet_hdr.hdr_len ? 714 macvtap16_to_cpu(q, vnet_hdr.hdr_len) : GOODCOPY_LEN; 715 if (copylen > good_linear) 716 copylen = good_linear; 717 else if (copylen < ETH_HLEN) 718 copylen = ETH_HLEN; 719 linear = copylen; 720 i = *from; 721 iov_iter_advance(&i, copylen); 722 if (iov_iter_npages(&i, INT_MAX) <= MAX_SKB_FRAGS) 723 zerocopy = true; 724 } 725 726 if (!zerocopy) { 727 copylen = len; 728 linear = macvtap16_to_cpu(q, vnet_hdr.hdr_len); 729 if (linear > good_linear) 730 linear = good_linear; 731 else if (linear < ETH_HLEN) 732 linear = ETH_HLEN; 733 } 734 735 skb = macvtap_alloc_skb(&q->sk, MACVTAP_RESERVE, copylen, 736 linear, noblock, &err); 737 if (!skb) 738 goto err; 739 740 if (zerocopy) 741 err = zerocopy_sg_from_iter(skb, from); 742 else { 743 err = skb_copy_datagram_from_iter(skb, 0, from, len); 744 if (!err && m && m->msg_control) { 745 struct ubuf_info *uarg = m->msg_control; 746 uarg->callback(uarg, false); 747 } 748 } 749 750 if (err) 751 goto err_kfree; 752 753 skb_set_network_header(skb, ETH_HLEN); 754 skb_reset_mac_header(skb); 755 skb->protocol = eth_hdr(skb)->h_proto; 756 757 if (vnet_hdr_len) { 758 err = virtio_net_hdr_to_skb(skb, &vnet_hdr, 759 macvtap_is_little_endian(q)); 760 if (err) 761 goto err_kfree; 762 } 763 764 skb_probe_transport_header(skb, ETH_HLEN); 765 766 /* Move network header to the right position for VLAN tagged packets */ 767 if ((skb->protocol == htons(ETH_P_8021Q) || 768 skb->protocol == htons(ETH_P_8021AD)) && 769 __vlan_get_protocol(skb, skb->protocol, &depth) != 0) 770 skb_set_network_header(skb, depth); 771 772 rcu_read_lock(); 773 vlan = rcu_dereference(q->vlan); 774 /* copy skb_ubuf_info for callback when skb has no error */ 775 if (zerocopy) { 776 skb_shinfo(skb)->destructor_arg = m->msg_control; 777 skb_shinfo(skb)->tx_flags |= SKBTX_DEV_ZEROCOPY; 778 skb_shinfo(skb)->tx_flags |= SKBTX_SHARED_FRAG; 779 } 780 if (vlan) { 781 skb->dev = vlan->dev; 782 dev_queue_xmit(skb); 783 } else { 784 kfree_skb(skb); 785 } 786 rcu_read_unlock(); 787 788 return total_len; 789 790 err_kfree: 791 kfree_skb(skb); 792 793 err: 794 rcu_read_lock(); 795 vlan = rcu_dereference(q->vlan); 796 if (vlan) 797 this_cpu_inc(vlan->pcpu_stats->tx_dropped); 798 rcu_read_unlock(); 799 800 return err; 801 } 802 803 static ssize_t macvtap_write_iter(struct kiocb *iocb, struct iov_iter *from) 804 { 805 struct file *file = iocb->ki_filp; 806 struct macvtap_queue *q = file->private_data; 807 808 return macvtap_get_user(q, NULL, from, file->f_flags & O_NONBLOCK); 809 } 810 811 /* Put packet to the user space buffer */ 812 static ssize_t macvtap_put_user(struct macvtap_queue *q, 813 const struct sk_buff *skb, 814 struct iov_iter *iter) 815 { 816 int ret; 817 int vnet_hdr_len = 0; 818 int vlan_offset = 0; 819 int total; 820 821 if (q->flags & IFF_VNET_HDR) { 822 struct virtio_net_hdr vnet_hdr; 823 vnet_hdr_len = q->vnet_hdr_sz; 824 if (iov_iter_count(iter) < vnet_hdr_len) 825 return -EINVAL; 826 827 ret = virtio_net_hdr_from_skb(skb, &vnet_hdr, 828 macvtap_is_little_endian(q)); 829 if (ret) 830 BUG(); 831 832 if (copy_to_iter(&vnet_hdr, sizeof(vnet_hdr), iter) != 833 sizeof(vnet_hdr)) 834 return -EFAULT; 835 836 iov_iter_advance(iter, vnet_hdr_len - sizeof(vnet_hdr)); 837 } 838 total = vnet_hdr_len; 839 total += skb->len; 840 841 if (skb_vlan_tag_present(skb)) { 842 struct { 843 __be16 h_vlan_proto; 844 __be16 h_vlan_TCI; 845 } veth; 846 veth.h_vlan_proto = skb->vlan_proto; 847 veth.h_vlan_TCI = htons(skb_vlan_tag_get(skb)); 848 849 vlan_offset = offsetof(struct vlan_ethhdr, h_vlan_proto); 850 total += VLAN_HLEN; 851 852 ret = skb_copy_datagram_iter(skb, 0, iter, vlan_offset); 853 if (ret || !iov_iter_count(iter)) 854 goto done; 855 856 ret = copy_to_iter(&veth, sizeof(veth), iter); 857 if (ret != sizeof(veth) || !iov_iter_count(iter)) 858 goto done; 859 } 860 861 ret = skb_copy_datagram_iter(skb, vlan_offset, iter, 862 skb->len - vlan_offset); 863 864 done: 865 return ret ? ret : total; 866 } 867 868 static ssize_t macvtap_do_read(struct macvtap_queue *q, 869 struct iov_iter *to, 870 int noblock) 871 { 872 DEFINE_WAIT(wait); 873 struct sk_buff *skb; 874 ssize_t ret = 0; 875 876 if (!iov_iter_count(to)) 877 return 0; 878 879 while (1) { 880 if (!noblock) 881 prepare_to_wait(sk_sleep(&q->sk), &wait, 882 TASK_INTERRUPTIBLE); 883 884 /* Read frames from the queue */ 885 skb = skb_array_consume(&q->skb_array); 886 if (skb) 887 break; 888 if (noblock) { 889 ret = -EAGAIN; 890 break; 891 } 892 if (signal_pending(current)) { 893 ret = -ERESTARTSYS; 894 break; 895 } 896 /* Nothing to read, let's sleep */ 897 schedule(); 898 } 899 if (!noblock) 900 finish_wait(sk_sleep(&q->sk), &wait); 901 902 if (skb) { 903 ret = macvtap_put_user(q, skb, to); 904 if (unlikely(ret < 0)) 905 kfree_skb(skb); 906 else 907 consume_skb(skb); 908 } 909 return ret; 910 } 911 912 static ssize_t macvtap_read_iter(struct kiocb *iocb, struct iov_iter *to) 913 { 914 struct file *file = iocb->ki_filp; 915 struct macvtap_queue *q = file->private_data; 916 ssize_t len = iov_iter_count(to), ret; 917 918 ret = macvtap_do_read(q, to, file->f_flags & O_NONBLOCK); 919 ret = min_t(ssize_t, ret, len); 920 if (ret > 0) 921 iocb->ki_pos = ret; 922 return ret; 923 } 924 925 static struct macvlan_dev *macvtap_get_vlan(struct macvtap_queue *q) 926 { 927 struct macvlan_dev *vlan; 928 929 ASSERT_RTNL(); 930 vlan = rtnl_dereference(q->vlan); 931 if (vlan) 932 dev_hold(vlan->dev); 933 934 return vlan; 935 } 936 937 static void macvtap_put_vlan(struct macvlan_dev *vlan) 938 { 939 dev_put(vlan->dev); 940 } 941 942 static int macvtap_ioctl_set_queue(struct file *file, unsigned int flags) 943 { 944 struct macvtap_queue *q = file->private_data; 945 struct macvlan_dev *vlan; 946 int ret; 947 948 vlan = macvtap_get_vlan(q); 949 if (!vlan) 950 return -EINVAL; 951 952 if (flags & IFF_ATTACH_QUEUE) 953 ret = macvtap_enable_queue(vlan->dev, file, q); 954 else if (flags & IFF_DETACH_QUEUE) 955 ret = macvtap_disable_queue(q); 956 else 957 ret = -EINVAL; 958 959 macvtap_put_vlan(vlan); 960 return ret; 961 } 962 963 static int set_offload(struct macvtap_queue *q, unsigned long arg) 964 { 965 struct macvlan_dev *vlan; 966 netdev_features_t features; 967 netdev_features_t feature_mask = 0; 968 969 vlan = rtnl_dereference(q->vlan); 970 if (!vlan) 971 return -ENOLINK; 972 973 features = vlan->dev->features; 974 975 if (arg & TUN_F_CSUM) { 976 feature_mask = NETIF_F_HW_CSUM; 977 978 if (arg & (TUN_F_TSO4 | TUN_F_TSO6)) { 979 if (arg & TUN_F_TSO_ECN) 980 feature_mask |= NETIF_F_TSO_ECN; 981 if (arg & TUN_F_TSO4) 982 feature_mask |= NETIF_F_TSO; 983 if (arg & TUN_F_TSO6) 984 feature_mask |= NETIF_F_TSO6; 985 } 986 987 if (arg & TUN_F_UFO) 988 feature_mask |= NETIF_F_UFO; 989 } 990 991 /* tun/tap driver inverts the usage for TSO offloads, where 992 * setting the TSO bit means that the userspace wants to 993 * accept TSO frames and turning it off means that user space 994 * does not support TSO. 995 * For macvtap, we have to invert it to mean the same thing. 996 * When user space turns off TSO, we turn off GSO/LRO so that 997 * user-space will not receive TSO frames. 998 */ 999 if (feature_mask & (NETIF_F_TSO | NETIF_F_TSO6 | NETIF_F_UFO)) 1000 features |= RX_OFFLOADS; 1001 else 1002 features &= ~RX_OFFLOADS; 1003 1004 /* tap_features are the same as features on tun/tap and 1005 * reflect user expectations. 1006 */ 1007 vlan->tap_features = feature_mask; 1008 vlan->set_features = features; 1009 netdev_update_features(vlan->dev); 1010 1011 return 0; 1012 } 1013 1014 /* 1015 * provide compatibility with generic tun/tap interface 1016 */ 1017 static long macvtap_ioctl(struct file *file, unsigned int cmd, 1018 unsigned long arg) 1019 { 1020 struct macvtap_queue *q = file->private_data; 1021 struct macvlan_dev *vlan; 1022 void __user *argp = (void __user *)arg; 1023 struct ifreq __user *ifr = argp; 1024 unsigned int __user *up = argp; 1025 unsigned short u; 1026 int __user *sp = argp; 1027 struct sockaddr sa; 1028 int s; 1029 int ret; 1030 1031 switch (cmd) { 1032 case TUNSETIFF: 1033 /* ignore the name, just look at flags */ 1034 if (get_user(u, &ifr->ifr_flags)) 1035 return -EFAULT; 1036 1037 ret = 0; 1038 if ((u & ~MACVTAP_FEATURES) != (IFF_NO_PI | IFF_TAP)) 1039 ret = -EINVAL; 1040 else 1041 q->flags = (q->flags & ~MACVTAP_FEATURES) | u; 1042 1043 return ret; 1044 1045 case TUNGETIFF: 1046 rtnl_lock(); 1047 vlan = macvtap_get_vlan(q); 1048 if (!vlan) { 1049 rtnl_unlock(); 1050 return -ENOLINK; 1051 } 1052 1053 ret = 0; 1054 u = q->flags; 1055 if (copy_to_user(&ifr->ifr_name, vlan->dev->name, IFNAMSIZ) || 1056 put_user(u, &ifr->ifr_flags)) 1057 ret = -EFAULT; 1058 macvtap_put_vlan(vlan); 1059 rtnl_unlock(); 1060 return ret; 1061 1062 case TUNSETQUEUE: 1063 if (get_user(u, &ifr->ifr_flags)) 1064 return -EFAULT; 1065 rtnl_lock(); 1066 ret = macvtap_ioctl_set_queue(file, u); 1067 rtnl_unlock(); 1068 return ret; 1069 1070 case TUNGETFEATURES: 1071 if (put_user(IFF_TAP | IFF_NO_PI | MACVTAP_FEATURES, up)) 1072 return -EFAULT; 1073 return 0; 1074 1075 case TUNSETSNDBUF: 1076 if (get_user(s, sp)) 1077 return -EFAULT; 1078 1079 q->sk.sk_sndbuf = s; 1080 return 0; 1081 1082 case TUNGETVNETHDRSZ: 1083 s = q->vnet_hdr_sz; 1084 if (put_user(s, sp)) 1085 return -EFAULT; 1086 return 0; 1087 1088 case TUNSETVNETHDRSZ: 1089 if (get_user(s, sp)) 1090 return -EFAULT; 1091 if (s < (int)sizeof(struct virtio_net_hdr)) 1092 return -EINVAL; 1093 1094 q->vnet_hdr_sz = s; 1095 return 0; 1096 1097 case TUNGETVNETLE: 1098 s = !!(q->flags & MACVTAP_VNET_LE); 1099 if (put_user(s, sp)) 1100 return -EFAULT; 1101 return 0; 1102 1103 case TUNSETVNETLE: 1104 if (get_user(s, sp)) 1105 return -EFAULT; 1106 if (s) 1107 q->flags |= MACVTAP_VNET_LE; 1108 else 1109 q->flags &= ~MACVTAP_VNET_LE; 1110 return 0; 1111 1112 case TUNGETVNETBE: 1113 return macvtap_get_vnet_be(q, sp); 1114 1115 case TUNSETVNETBE: 1116 return macvtap_set_vnet_be(q, sp); 1117 1118 case TUNSETOFFLOAD: 1119 /* let the user check for future flags */ 1120 if (arg & ~(TUN_F_CSUM | TUN_F_TSO4 | TUN_F_TSO6 | 1121 TUN_F_TSO_ECN | TUN_F_UFO)) 1122 return -EINVAL; 1123 1124 rtnl_lock(); 1125 ret = set_offload(q, arg); 1126 rtnl_unlock(); 1127 return ret; 1128 1129 case SIOCGIFHWADDR: 1130 rtnl_lock(); 1131 vlan = macvtap_get_vlan(q); 1132 if (!vlan) { 1133 rtnl_unlock(); 1134 return -ENOLINK; 1135 } 1136 ret = 0; 1137 u = vlan->dev->type; 1138 if (copy_to_user(&ifr->ifr_name, vlan->dev->name, IFNAMSIZ) || 1139 copy_to_user(&ifr->ifr_hwaddr.sa_data, vlan->dev->dev_addr, ETH_ALEN) || 1140 put_user(u, &ifr->ifr_hwaddr.sa_family)) 1141 ret = -EFAULT; 1142 macvtap_put_vlan(vlan); 1143 rtnl_unlock(); 1144 return ret; 1145 1146 case SIOCSIFHWADDR: 1147 if (copy_from_user(&sa, &ifr->ifr_hwaddr, sizeof(sa))) 1148 return -EFAULT; 1149 rtnl_lock(); 1150 vlan = macvtap_get_vlan(q); 1151 if (!vlan) { 1152 rtnl_unlock(); 1153 return -ENOLINK; 1154 } 1155 ret = dev_set_mac_address(vlan->dev, &sa); 1156 macvtap_put_vlan(vlan); 1157 rtnl_unlock(); 1158 return ret; 1159 1160 default: 1161 return -EINVAL; 1162 } 1163 } 1164 1165 #ifdef CONFIG_COMPAT 1166 static long macvtap_compat_ioctl(struct file *file, unsigned int cmd, 1167 unsigned long arg) 1168 { 1169 return macvtap_ioctl(file, cmd, (unsigned long)compat_ptr(arg)); 1170 } 1171 #endif 1172 1173 static const struct file_operations macvtap_fops = { 1174 .owner = THIS_MODULE, 1175 .open = macvtap_open, 1176 .release = macvtap_release, 1177 .read_iter = macvtap_read_iter, 1178 .write_iter = macvtap_write_iter, 1179 .poll = macvtap_poll, 1180 .llseek = no_llseek, 1181 .unlocked_ioctl = macvtap_ioctl, 1182 #ifdef CONFIG_COMPAT 1183 .compat_ioctl = macvtap_compat_ioctl, 1184 #endif 1185 }; 1186 1187 static int macvtap_sendmsg(struct socket *sock, struct msghdr *m, 1188 size_t total_len) 1189 { 1190 struct macvtap_queue *q = container_of(sock, struct macvtap_queue, sock); 1191 return macvtap_get_user(q, m, &m->msg_iter, m->msg_flags & MSG_DONTWAIT); 1192 } 1193 1194 static int macvtap_recvmsg(struct socket *sock, struct msghdr *m, 1195 size_t total_len, int flags) 1196 { 1197 struct macvtap_queue *q = container_of(sock, struct macvtap_queue, sock); 1198 int ret; 1199 if (flags & ~(MSG_DONTWAIT|MSG_TRUNC)) 1200 return -EINVAL; 1201 ret = macvtap_do_read(q, &m->msg_iter, flags & MSG_DONTWAIT); 1202 if (ret > total_len) { 1203 m->msg_flags |= MSG_TRUNC; 1204 ret = flags & MSG_TRUNC ? ret : total_len; 1205 } 1206 return ret; 1207 } 1208 1209 static int macvtap_peek_len(struct socket *sock) 1210 { 1211 struct macvtap_queue *q = container_of(sock, struct macvtap_queue, 1212 sock); 1213 return skb_array_peek_len(&q->skb_array); 1214 } 1215 1216 /* Ops structure to mimic raw sockets with tun */ 1217 static const struct proto_ops macvtap_socket_ops = { 1218 .sendmsg = macvtap_sendmsg, 1219 .recvmsg = macvtap_recvmsg, 1220 .peek_len = macvtap_peek_len, 1221 }; 1222 1223 /* Get an underlying socket object from tun file. Returns error unless file is 1224 * attached to a device. The returned object works like a packet socket, it 1225 * can be used for sock_sendmsg/sock_recvmsg. The caller is responsible for 1226 * holding a reference to the file for as long as the socket is in use. */ 1227 struct socket *macvtap_get_socket(struct file *file) 1228 { 1229 struct macvtap_queue *q; 1230 if (file->f_op != &macvtap_fops) 1231 return ERR_PTR(-EINVAL); 1232 q = file->private_data; 1233 if (!q) 1234 return ERR_PTR(-EBADFD); 1235 return &q->sock; 1236 } 1237 EXPORT_SYMBOL_GPL(macvtap_get_socket); 1238 1239 static int macvtap_queue_resize(struct macvlan_dev *vlan) 1240 { 1241 struct net_device *dev = vlan->dev; 1242 struct macvtap_queue *q; 1243 struct skb_array **arrays; 1244 int n = vlan->numqueues; 1245 int ret, i = 0; 1246 1247 arrays = kmalloc(sizeof *arrays * n, GFP_KERNEL); 1248 if (!arrays) 1249 return -ENOMEM; 1250 1251 list_for_each_entry(q, &vlan->queue_list, next) 1252 arrays[i++] = &q->skb_array; 1253 1254 ret = skb_array_resize_multiple(arrays, n, 1255 dev->tx_queue_len, GFP_KERNEL); 1256 1257 kfree(arrays); 1258 return ret; 1259 } 1260 1261 static int macvtap_device_event(struct notifier_block *unused, 1262 unsigned long event, void *ptr) 1263 { 1264 struct net_device *dev = netdev_notifier_info_to_dev(ptr); 1265 struct macvlan_dev *vlan; 1266 struct device *classdev; 1267 dev_t devt; 1268 int err; 1269 char tap_name[IFNAMSIZ]; 1270 1271 if (dev->rtnl_link_ops != &macvtap_link_ops) 1272 return NOTIFY_DONE; 1273 1274 snprintf(tap_name, IFNAMSIZ, "tap%d", dev->ifindex); 1275 vlan = netdev_priv(dev); 1276 1277 switch (event) { 1278 case NETDEV_REGISTER: 1279 /* Create the device node here after the network device has 1280 * been registered but before register_netdevice has 1281 * finished running. 1282 */ 1283 err = macvtap_get_minor(vlan); 1284 if (err) 1285 return notifier_from_errno(err); 1286 1287 devt = MKDEV(MAJOR(macvtap_major), vlan->minor); 1288 classdev = device_create(&macvtap_class, &dev->dev, devt, 1289 dev, tap_name); 1290 if (IS_ERR(classdev)) { 1291 macvtap_free_minor(vlan); 1292 return notifier_from_errno(PTR_ERR(classdev)); 1293 } 1294 err = sysfs_create_link(&dev->dev.kobj, &classdev->kobj, 1295 tap_name); 1296 if (err) 1297 return notifier_from_errno(err); 1298 break; 1299 case NETDEV_UNREGISTER: 1300 /* vlan->minor == 0 if NETDEV_REGISTER above failed */ 1301 if (vlan->minor == 0) 1302 break; 1303 sysfs_remove_link(&dev->dev.kobj, tap_name); 1304 devt = MKDEV(MAJOR(macvtap_major), vlan->minor); 1305 device_destroy(&macvtap_class, devt); 1306 macvtap_free_minor(vlan); 1307 break; 1308 case NETDEV_CHANGE_TX_QUEUE_LEN: 1309 if (macvtap_queue_resize(vlan)) 1310 return NOTIFY_BAD; 1311 break; 1312 } 1313 1314 return NOTIFY_DONE; 1315 } 1316 1317 static struct notifier_block macvtap_notifier_block __read_mostly = { 1318 .notifier_call = macvtap_device_event, 1319 }; 1320 1321 static int macvtap_init(void) 1322 { 1323 int err; 1324 1325 err = alloc_chrdev_region(&macvtap_major, 0, 1326 MACVTAP_NUM_DEVS, "macvtap"); 1327 if (err) 1328 goto out1; 1329 1330 cdev_init(&macvtap_cdev, &macvtap_fops); 1331 err = cdev_add(&macvtap_cdev, macvtap_major, MACVTAP_NUM_DEVS); 1332 if (err) 1333 goto out2; 1334 1335 err = class_register(&macvtap_class); 1336 if (err) 1337 goto out3; 1338 1339 err = register_netdevice_notifier(&macvtap_notifier_block); 1340 if (err) 1341 goto out4; 1342 1343 err = macvlan_link_register(&macvtap_link_ops); 1344 if (err) 1345 goto out5; 1346 1347 return 0; 1348 1349 out5: 1350 unregister_netdevice_notifier(&macvtap_notifier_block); 1351 out4: 1352 class_unregister(&macvtap_class); 1353 out3: 1354 cdev_del(&macvtap_cdev); 1355 out2: 1356 unregister_chrdev_region(macvtap_major, MACVTAP_NUM_DEVS); 1357 out1: 1358 return err; 1359 } 1360 module_init(macvtap_init); 1361 1362 static void macvtap_exit(void) 1363 { 1364 rtnl_link_unregister(&macvtap_link_ops); 1365 unregister_netdevice_notifier(&macvtap_notifier_block); 1366 class_unregister(&macvtap_class); 1367 cdev_del(&macvtap_cdev); 1368 unregister_chrdev_region(macvtap_major, MACVTAP_NUM_DEVS); 1369 idr_destroy(&minor_idr); 1370 } 1371 module_exit(macvtap_exit); 1372 1373 MODULE_ALIAS_RTNL_LINK("macvtap"); 1374 MODULE_AUTHOR("Arnd Bergmann <arnd@arndb.de>"); 1375 MODULE_LICENSE("GPL"); 1376