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 struct class *macvtap_class; 133 static struct cdev macvtap_cdev; 134 135 static const struct proto_ops macvtap_socket_ops; 136 137 #define TUN_OFFLOADS (NETIF_F_HW_CSUM | NETIF_F_TSO_ECN | NETIF_F_TSO | \ 138 NETIF_F_TSO6 | NETIF_F_UFO) 139 #define RX_OFFLOADS (NETIF_F_GRO | NETIF_F_LRO) 140 #define TAP_FEATURES (NETIF_F_GSO | NETIF_F_SG | NETIF_F_FRAGLIST) 141 142 static struct macvlan_dev *macvtap_get_vlan_rcu(const struct net_device *dev) 143 { 144 return rcu_dereference(dev->rx_handler_data); 145 } 146 147 /* 148 * RCU usage: 149 * The macvtap_queue and the macvlan_dev are loosely coupled, the 150 * pointers from one to the other can only be read while rcu_read_lock 151 * or rtnl is held. 152 * 153 * Both the file and the macvlan_dev hold a reference on the macvtap_queue 154 * through sock_hold(&q->sk). When the macvlan_dev goes away first, 155 * q->vlan becomes inaccessible. When the files gets closed, 156 * macvtap_get_queue() fails. 157 * 158 * There may still be references to the struct sock inside of the 159 * queue from outbound SKBs, but these never reference back to the 160 * file or the dev. The data structure is freed through __sk_free 161 * when both our references and any pending SKBs are gone. 162 */ 163 164 static int macvtap_enable_queue(struct net_device *dev, struct file *file, 165 struct macvtap_queue *q) 166 { 167 struct macvlan_dev *vlan = netdev_priv(dev); 168 int err = -EINVAL; 169 170 ASSERT_RTNL(); 171 172 if (q->enabled) 173 goto out; 174 175 err = 0; 176 rcu_assign_pointer(vlan->taps[vlan->numvtaps], q); 177 q->queue_index = vlan->numvtaps; 178 q->enabled = true; 179 180 vlan->numvtaps++; 181 out: 182 return err; 183 } 184 185 /* Requires RTNL */ 186 static int macvtap_set_queue(struct net_device *dev, struct file *file, 187 struct macvtap_queue *q) 188 { 189 struct macvlan_dev *vlan = netdev_priv(dev); 190 191 if (vlan->numqueues == MAX_MACVTAP_QUEUES) 192 return -EBUSY; 193 194 rcu_assign_pointer(q->vlan, vlan); 195 rcu_assign_pointer(vlan->taps[vlan->numvtaps], q); 196 sock_hold(&q->sk); 197 198 q->file = file; 199 q->queue_index = vlan->numvtaps; 200 q->enabled = true; 201 file->private_data = q; 202 list_add_tail(&q->next, &vlan->queue_list); 203 204 vlan->numvtaps++; 205 vlan->numqueues++; 206 207 return 0; 208 } 209 210 static int macvtap_disable_queue(struct macvtap_queue *q) 211 { 212 struct macvlan_dev *vlan; 213 struct macvtap_queue *nq; 214 215 ASSERT_RTNL(); 216 if (!q->enabled) 217 return -EINVAL; 218 219 vlan = rtnl_dereference(q->vlan); 220 221 if (vlan) { 222 int index = q->queue_index; 223 BUG_ON(index >= vlan->numvtaps); 224 nq = rtnl_dereference(vlan->taps[vlan->numvtaps - 1]); 225 nq->queue_index = index; 226 227 rcu_assign_pointer(vlan->taps[index], nq); 228 RCU_INIT_POINTER(vlan->taps[vlan->numvtaps - 1], NULL); 229 q->enabled = false; 230 231 vlan->numvtaps--; 232 } 233 234 return 0; 235 } 236 237 /* 238 * The file owning the queue got closed, give up both 239 * the reference that the files holds as well as the 240 * one from the macvlan_dev if that still exists. 241 * 242 * Using the spinlock makes sure that we don't get 243 * to the queue again after destroying it. 244 */ 245 static void macvtap_put_queue(struct macvtap_queue *q) 246 { 247 struct macvlan_dev *vlan; 248 249 rtnl_lock(); 250 vlan = rtnl_dereference(q->vlan); 251 252 if (vlan) { 253 if (q->enabled) 254 BUG_ON(macvtap_disable_queue(q)); 255 256 vlan->numqueues--; 257 RCU_INIT_POINTER(q->vlan, NULL); 258 sock_put(&q->sk); 259 list_del_init(&q->next); 260 } 261 262 rtnl_unlock(); 263 264 synchronize_rcu(); 265 sock_put(&q->sk); 266 } 267 268 /* 269 * Select a queue based on the rxq of the device on which this packet 270 * arrived. If the incoming device is not mq, calculate a flow hash 271 * to select a queue. If all fails, find the first available queue. 272 * Cache vlan->numvtaps since it can become zero during the execution 273 * of this function. 274 */ 275 static struct macvtap_queue *macvtap_get_queue(struct net_device *dev, 276 struct sk_buff *skb) 277 { 278 struct macvlan_dev *vlan = netdev_priv(dev); 279 struct macvtap_queue *tap = NULL; 280 /* Access to taps array is protected by rcu, but access to numvtaps 281 * isn't. Below we use it to lookup a queue, but treat it as a hint 282 * and validate that the result isn't NULL - in case we are 283 * racing against queue removal. 284 */ 285 int numvtaps = ACCESS_ONCE(vlan->numvtaps); 286 __u32 rxq; 287 288 if (!numvtaps) 289 goto out; 290 291 /* Check if we can use flow to select a queue */ 292 rxq = skb_get_hash(skb); 293 if (rxq) { 294 tap = rcu_dereference(vlan->taps[rxq % numvtaps]); 295 goto out; 296 } 297 298 if (likely(skb_rx_queue_recorded(skb))) { 299 rxq = skb_get_rx_queue(skb); 300 301 while (unlikely(rxq >= numvtaps)) 302 rxq -= numvtaps; 303 304 tap = rcu_dereference(vlan->taps[rxq]); 305 goto out; 306 } 307 308 tap = rcu_dereference(vlan->taps[0]); 309 out: 310 return tap; 311 } 312 313 /* 314 * The net_device is going away, give up the reference 315 * that it holds on all queues and safely set the pointer 316 * from the queues to NULL. 317 */ 318 static void macvtap_del_queues(struct net_device *dev) 319 { 320 struct macvlan_dev *vlan = netdev_priv(dev); 321 struct macvtap_queue *q, *tmp; 322 323 ASSERT_RTNL(); 324 list_for_each_entry_safe(q, tmp, &vlan->queue_list, next) { 325 list_del_init(&q->next); 326 RCU_INIT_POINTER(q->vlan, NULL); 327 if (q->enabled) 328 vlan->numvtaps--; 329 vlan->numqueues--; 330 sock_put(&q->sk); 331 } 332 BUG_ON(vlan->numvtaps); 333 BUG_ON(vlan->numqueues); 334 /* guarantee that any future macvtap_set_queue will fail */ 335 vlan->numvtaps = MAX_MACVTAP_QUEUES; 336 } 337 338 static rx_handler_result_t macvtap_handle_frame(struct sk_buff **pskb) 339 { 340 struct sk_buff *skb = *pskb; 341 struct net_device *dev = skb->dev; 342 struct macvlan_dev *vlan; 343 struct macvtap_queue *q; 344 netdev_features_t features = TAP_FEATURES; 345 346 vlan = macvtap_get_vlan_rcu(dev); 347 if (!vlan) 348 return RX_HANDLER_PASS; 349 350 q = macvtap_get_queue(dev, skb); 351 if (!q) 352 return RX_HANDLER_PASS; 353 354 if (skb_queue_len(&q->sk.sk_receive_queue) >= dev->tx_queue_len) 355 goto drop; 356 357 skb_push(skb, ETH_HLEN); 358 359 /* Apply the forward feature mask so that we perform segmentation 360 * according to users wishes. This only works if VNET_HDR is 361 * enabled. 362 */ 363 if (q->flags & IFF_VNET_HDR) 364 features |= vlan->tap_features; 365 if (netif_needs_gso(skb, features)) { 366 struct sk_buff *segs = __skb_gso_segment(skb, features, false); 367 368 if (IS_ERR(segs)) 369 goto drop; 370 371 if (!segs) { 372 skb_queue_tail(&q->sk.sk_receive_queue, skb); 373 goto wake_up; 374 } 375 376 kfree_skb(skb); 377 while (segs) { 378 struct sk_buff *nskb = segs->next; 379 380 segs->next = NULL; 381 skb_queue_tail(&q->sk.sk_receive_queue, segs); 382 segs = nskb; 383 } 384 } else { 385 /* If we receive a partial checksum and the tap side 386 * doesn't support checksum offload, compute the checksum. 387 * Note: it doesn't matter which checksum feature to 388 * check, we either support them all or none. 389 */ 390 if (skb->ip_summed == CHECKSUM_PARTIAL && 391 !(features & NETIF_F_ALL_CSUM) && 392 skb_checksum_help(skb)) 393 goto drop; 394 skb_queue_tail(&q->sk.sk_receive_queue, skb); 395 } 396 397 wake_up: 398 wake_up_interruptible_poll(sk_sleep(&q->sk), POLLIN | POLLRDNORM | POLLRDBAND); 399 return RX_HANDLER_CONSUMED; 400 401 drop: 402 /* Count errors/drops only here, thus don't care about args. */ 403 macvlan_count_rx(vlan, 0, 0, 0); 404 kfree_skb(skb); 405 return RX_HANDLER_CONSUMED; 406 } 407 408 static int macvtap_get_minor(struct macvlan_dev *vlan) 409 { 410 int retval = -ENOMEM; 411 412 mutex_lock(&minor_lock); 413 retval = idr_alloc(&minor_idr, vlan, 1, MACVTAP_NUM_DEVS, GFP_KERNEL); 414 if (retval >= 0) { 415 vlan->minor = retval; 416 } else if (retval == -ENOSPC) { 417 printk(KERN_ERR "too many macvtap devices\n"); 418 retval = -EINVAL; 419 } 420 mutex_unlock(&minor_lock); 421 return retval < 0 ? retval : 0; 422 } 423 424 static void macvtap_free_minor(struct macvlan_dev *vlan) 425 { 426 mutex_lock(&minor_lock); 427 if (vlan->minor) { 428 idr_remove(&minor_idr, vlan->minor); 429 vlan->minor = 0; 430 } 431 mutex_unlock(&minor_lock); 432 } 433 434 static struct net_device *dev_get_by_macvtap_minor(int minor) 435 { 436 struct net_device *dev = NULL; 437 struct macvlan_dev *vlan; 438 439 mutex_lock(&minor_lock); 440 vlan = idr_find(&minor_idr, minor); 441 if (vlan) { 442 dev = vlan->dev; 443 dev_hold(dev); 444 } 445 mutex_unlock(&minor_lock); 446 return dev; 447 } 448 449 static int macvtap_newlink(struct net *src_net, 450 struct net_device *dev, 451 struct nlattr *tb[], 452 struct nlattr *data[]) 453 { 454 struct macvlan_dev *vlan = netdev_priv(dev); 455 int err; 456 457 INIT_LIST_HEAD(&vlan->queue_list); 458 459 /* Since macvlan supports all offloads by default, make 460 * tap support all offloads also. 461 */ 462 vlan->tap_features = TUN_OFFLOADS; 463 464 err = netdev_rx_handler_register(dev, macvtap_handle_frame, vlan); 465 if (err) 466 return err; 467 468 /* Don't put anything that may fail after macvlan_common_newlink 469 * because we can't undo what it does. 470 */ 471 return macvlan_common_newlink(src_net, dev, tb, data); 472 } 473 474 static void macvtap_dellink(struct net_device *dev, 475 struct list_head *head) 476 { 477 netdev_rx_handler_unregister(dev); 478 macvtap_del_queues(dev); 479 macvlan_dellink(dev, head); 480 } 481 482 static void macvtap_setup(struct net_device *dev) 483 { 484 macvlan_common_setup(dev); 485 dev->tx_queue_len = TUN_READQ_SIZE; 486 } 487 488 static struct rtnl_link_ops macvtap_link_ops __read_mostly = { 489 .kind = "macvtap", 490 .setup = macvtap_setup, 491 .newlink = macvtap_newlink, 492 .dellink = macvtap_dellink, 493 }; 494 495 496 static void macvtap_sock_write_space(struct sock *sk) 497 { 498 wait_queue_head_t *wqueue; 499 500 if (!sock_writeable(sk) || 501 !test_and_clear_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags)) 502 return; 503 504 wqueue = sk_sleep(sk); 505 if (wqueue && waitqueue_active(wqueue)) 506 wake_up_interruptible_poll(wqueue, POLLOUT | POLLWRNORM | POLLWRBAND); 507 } 508 509 static void macvtap_sock_destruct(struct sock *sk) 510 { 511 skb_queue_purge(&sk->sk_receive_queue); 512 } 513 514 static int macvtap_open(struct inode *inode, struct file *file) 515 { 516 struct net *net = current->nsproxy->net_ns; 517 struct net_device *dev; 518 struct macvtap_queue *q; 519 int err = -ENODEV; 520 521 rtnl_lock(); 522 dev = dev_get_by_macvtap_minor(iminor(inode)); 523 if (!dev) 524 goto out; 525 526 err = -ENOMEM; 527 q = (struct macvtap_queue *)sk_alloc(net, AF_UNSPEC, GFP_KERNEL, 528 &macvtap_proto, 0); 529 if (!q) 530 goto out; 531 532 RCU_INIT_POINTER(q->sock.wq, &q->wq); 533 init_waitqueue_head(&q->wq.wait); 534 q->sock.type = SOCK_RAW; 535 q->sock.state = SS_CONNECTED; 536 q->sock.file = file; 537 q->sock.ops = &macvtap_socket_ops; 538 sock_init_data(&q->sock, &q->sk); 539 q->sk.sk_write_space = macvtap_sock_write_space; 540 q->sk.sk_destruct = macvtap_sock_destruct; 541 q->flags = IFF_VNET_HDR | IFF_NO_PI | IFF_TAP; 542 q->vnet_hdr_sz = sizeof(struct virtio_net_hdr); 543 544 /* 545 * so far only KVM virtio_net uses macvtap, enable zero copy between 546 * guest kernel and host kernel when lower device supports zerocopy 547 * 548 * The macvlan supports zerocopy iff the lower device supports zero 549 * copy so we don't have to look at the lower device directly. 550 */ 551 if ((dev->features & NETIF_F_HIGHDMA) && (dev->features & NETIF_F_SG)) 552 sock_set_flag(&q->sk, SOCK_ZEROCOPY); 553 554 err = macvtap_set_queue(dev, file, q); 555 if (err) 556 sock_put(&q->sk); 557 558 out: 559 if (dev) 560 dev_put(dev); 561 562 rtnl_unlock(); 563 return err; 564 } 565 566 static int macvtap_release(struct inode *inode, struct file *file) 567 { 568 struct macvtap_queue *q = file->private_data; 569 macvtap_put_queue(q); 570 return 0; 571 } 572 573 static unsigned int macvtap_poll(struct file *file, poll_table * wait) 574 { 575 struct macvtap_queue *q = file->private_data; 576 unsigned int mask = POLLERR; 577 578 if (!q) 579 goto out; 580 581 mask = 0; 582 poll_wait(file, &q->wq.wait, wait); 583 584 if (!skb_queue_empty(&q->sk.sk_receive_queue)) 585 mask |= POLLIN | POLLRDNORM; 586 587 if (sock_writeable(&q->sk) || 588 (!test_and_set_bit(SOCK_ASYNC_NOSPACE, &q->sock.flags) && 589 sock_writeable(&q->sk))) 590 mask |= POLLOUT | POLLWRNORM; 591 592 out: 593 return mask; 594 } 595 596 static inline struct sk_buff *macvtap_alloc_skb(struct sock *sk, size_t prepad, 597 size_t len, size_t linear, 598 int noblock, int *err) 599 { 600 struct sk_buff *skb; 601 602 /* Under a page? Don't bother with paged skb. */ 603 if (prepad + len < PAGE_SIZE || !linear) 604 linear = len; 605 606 skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock, 607 err, 0); 608 if (!skb) 609 return NULL; 610 611 skb_reserve(skb, prepad); 612 skb_put(skb, linear); 613 skb->data_len = len - linear; 614 skb->len += len - linear; 615 616 return skb; 617 } 618 619 /* 620 * macvtap_skb_from_vnet_hdr and macvtap_skb_to_vnet_hdr should 621 * be shared with the tun/tap driver. 622 */ 623 static int macvtap_skb_from_vnet_hdr(struct macvtap_queue *q, 624 struct sk_buff *skb, 625 struct virtio_net_hdr *vnet_hdr) 626 { 627 unsigned short gso_type = 0; 628 if (vnet_hdr->gso_type != VIRTIO_NET_HDR_GSO_NONE) { 629 switch (vnet_hdr->gso_type & ~VIRTIO_NET_HDR_GSO_ECN) { 630 case VIRTIO_NET_HDR_GSO_TCPV4: 631 gso_type = SKB_GSO_TCPV4; 632 break; 633 case VIRTIO_NET_HDR_GSO_TCPV6: 634 gso_type = SKB_GSO_TCPV6; 635 break; 636 case VIRTIO_NET_HDR_GSO_UDP: 637 gso_type = SKB_GSO_UDP; 638 break; 639 default: 640 return -EINVAL; 641 } 642 643 if (vnet_hdr->gso_type & VIRTIO_NET_HDR_GSO_ECN) 644 gso_type |= SKB_GSO_TCP_ECN; 645 646 if (vnet_hdr->gso_size == 0) 647 return -EINVAL; 648 } 649 650 if (vnet_hdr->flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) { 651 if (!skb_partial_csum_set(skb, macvtap16_to_cpu(q, vnet_hdr->csum_start), 652 macvtap16_to_cpu(q, vnet_hdr->csum_offset))) 653 return -EINVAL; 654 } 655 656 if (vnet_hdr->gso_type != VIRTIO_NET_HDR_GSO_NONE) { 657 skb_shinfo(skb)->gso_size = macvtap16_to_cpu(q, vnet_hdr->gso_size); 658 skb_shinfo(skb)->gso_type = gso_type; 659 660 /* Header must be checked, and gso_segs computed. */ 661 skb_shinfo(skb)->gso_type |= SKB_GSO_DODGY; 662 skb_shinfo(skb)->gso_segs = 0; 663 } 664 return 0; 665 } 666 667 static void macvtap_skb_to_vnet_hdr(struct macvtap_queue *q, 668 const struct sk_buff *skb, 669 struct virtio_net_hdr *vnet_hdr) 670 { 671 memset(vnet_hdr, 0, sizeof(*vnet_hdr)); 672 673 if (skb_is_gso(skb)) { 674 struct skb_shared_info *sinfo = skb_shinfo(skb); 675 676 /* This is a hint as to how much should be linear. */ 677 vnet_hdr->hdr_len = cpu_to_macvtap16(q, skb_headlen(skb)); 678 vnet_hdr->gso_size = cpu_to_macvtap16(q, sinfo->gso_size); 679 if (sinfo->gso_type & SKB_GSO_TCPV4) 680 vnet_hdr->gso_type = VIRTIO_NET_HDR_GSO_TCPV4; 681 else if (sinfo->gso_type & SKB_GSO_TCPV6) 682 vnet_hdr->gso_type = VIRTIO_NET_HDR_GSO_TCPV6; 683 else if (sinfo->gso_type & SKB_GSO_UDP) 684 vnet_hdr->gso_type = VIRTIO_NET_HDR_GSO_UDP; 685 else 686 BUG(); 687 if (sinfo->gso_type & SKB_GSO_TCP_ECN) 688 vnet_hdr->gso_type |= VIRTIO_NET_HDR_GSO_ECN; 689 } else 690 vnet_hdr->gso_type = VIRTIO_NET_HDR_GSO_NONE; 691 692 if (skb->ip_summed == CHECKSUM_PARTIAL) { 693 vnet_hdr->flags = VIRTIO_NET_HDR_F_NEEDS_CSUM; 694 if (skb_vlan_tag_present(skb)) 695 vnet_hdr->csum_start = cpu_to_macvtap16(q, 696 skb_checksum_start_offset(skb) + VLAN_HLEN); 697 else 698 vnet_hdr->csum_start = cpu_to_macvtap16(q, 699 skb_checksum_start_offset(skb)); 700 vnet_hdr->csum_offset = cpu_to_macvtap16(q, skb->csum_offset); 701 } else if (skb->ip_summed == CHECKSUM_UNNECESSARY) { 702 vnet_hdr->flags = VIRTIO_NET_HDR_F_DATA_VALID; 703 } /* else everything is zero */ 704 } 705 706 /* Neighbour code has some assumptions on HH_DATA_MOD alignment */ 707 #define MACVTAP_RESERVE HH_DATA_OFF(ETH_HLEN) 708 709 /* Get packet from user space buffer */ 710 static ssize_t macvtap_get_user(struct macvtap_queue *q, struct msghdr *m, 711 struct iov_iter *from, int noblock) 712 { 713 int good_linear = SKB_MAX_HEAD(MACVTAP_RESERVE); 714 struct sk_buff *skb; 715 struct macvlan_dev *vlan; 716 unsigned long total_len = iov_iter_count(from); 717 unsigned long len = total_len; 718 int err; 719 struct virtio_net_hdr vnet_hdr = { 0 }; 720 int vnet_hdr_len = 0; 721 int copylen = 0; 722 int depth; 723 bool zerocopy = false; 724 size_t linear; 725 ssize_t n; 726 727 if (q->flags & IFF_VNET_HDR) { 728 vnet_hdr_len = q->vnet_hdr_sz; 729 730 err = -EINVAL; 731 if (len < vnet_hdr_len) 732 goto err; 733 len -= vnet_hdr_len; 734 735 err = -EFAULT; 736 n = copy_from_iter(&vnet_hdr, sizeof(vnet_hdr), from); 737 if (n != sizeof(vnet_hdr)) 738 goto err; 739 iov_iter_advance(from, vnet_hdr_len - sizeof(vnet_hdr)); 740 if ((vnet_hdr.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) && 741 macvtap16_to_cpu(q, vnet_hdr.csum_start) + 742 macvtap16_to_cpu(q, vnet_hdr.csum_offset) + 2 > 743 macvtap16_to_cpu(q, vnet_hdr.hdr_len)) 744 vnet_hdr.hdr_len = cpu_to_macvtap16(q, 745 macvtap16_to_cpu(q, vnet_hdr.csum_start) + 746 macvtap16_to_cpu(q, vnet_hdr.csum_offset) + 2); 747 err = -EINVAL; 748 if (macvtap16_to_cpu(q, vnet_hdr.hdr_len) > len) 749 goto err; 750 } 751 752 err = -EINVAL; 753 if (unlikely(len < ETH_HLEN)) 754 goto err; 755 756 if (m && m->msg_control && sock_flag(&q->sk, SOCK_ZEROCOPY)) { 757 struct iov_iter i; 758 759 copylen = vnet_hdr.hdr_len ? 760 macvtap16_to_cpu(q, vnet_hdr.hdr_len) : GOODCOPY_LEN; 761 if (copylen > good_linear) 762 copylen = good_linear; 763 linear = copylen; 764 i = *from; 765 iov_iter_advance(&i, copylen); 766 if (iov_iter_npages(&i, INT_MAX) <= MAX_SKB_FRAGS) 767 zerocopy = true; 768 } 769 770 if (!zerocopy) { 771 copylen = len; 772 if (macvtap16_to_cpu(q, vnet_hdr.hdr_len) > good_linear) 773 linear = good_linear; 774 else 775 linear = macvtap16_to_cpu(q, vnet_hdr.hdr_len); 776 } 777 778 skb = macvtap_alloc_skb(&q->sk, MACVTAP_RESERVE, copylen, 779 linear, noblock, &err); 780 if (!skb) 781 goto err; 782 783 if (zerocopy) 784 err = zerocopy_sg_from_iter(skb, from); 785 else { 786 err = skb_copy_datagram_from_iter(skb, 0, from, len); 787 if (!err && m && m->msg_control) { 788 struct ubuf_info *uarg = m->msg_control; 789 uarg->callback(uarg, false); 790 } 791 } 792 793 if (err) 794 goto err_kfree; 795 796 skb_set_network_header(skb, ETH_HLEN); 797 skb_reset_mac_header(skb); 798 skb->protocol = eth_hdr(skb)->h_proto; 799 800 if (vnet_hdr_len) { 801 err = macvtap_skb_from_vnet_hdr(q, skb, &vnet_hdr); 802 if (err) 803 goto err_kfree; 804 } 805 806 skb_probe_transport_header(skb, ETH_HLEN); 807 808 /* Move network header to the right position for VLAN tagged packets */ 809 if ((skb->protocol == htons(ETH_P_8021Q) || 810 skb->protocol == htons(ETH_P_8021AD)) && 811 __vlan_get_protocol(skb, skb->protocol, &depth) != 0) 812 skb_set_network_header(skb, depth); 813 814 rcu_read_lock(); 815 vlan = rcu_dereference(q->vlan); 816 /* copy skb_ubuf_info for callback when skb has no error */ 817 if (zerocopy) { 818 skb_shinfo(skb)->destructor_arg = m->msg_control; 819 skb_shinfo(skb)->tx_flags |= SKBTX_DEV_ZEROCOPY; 820 skb_shinfo(skb)->tx_flags |= SKBTX_SHARED_FRAG; 821 } 822 if (vlan) { 823 skb->dev = vlan->dev; 824 dev_queue_xmit(skb); 825 } else { 826 kfree_skb(skb); 827 } 828 rcu_read_unlock(); 829 830 return total_len; 831 832 err_kfree: 833 kfree_skb(skb); 834 835 err: 836 rcu_read_lock(); 837 vlan = rcu_dereference(q->vlan); 838 if (vlan) 839 this_cpu_inc(vlan->pcpu_stats->tx_dropped); 840 rcu_read_unlock(); 841 842 return err; 843 } 844 845 static ssize_t macvtap_write_iter(struct kiocb *iocb, struct iov_iter *from) 846 { 847 struct file *file = iocb->ki_filp; 848 struct macvtap_queue *q = file->private_data; 849 850 return macvtap_get_user(q, NULL, from, file->f_flags & O_NONBLOCK); 851 } 852 853 /* Put packet to the user space buffer */ 854 static ssize_t macvtap_put_user(struct macvtap_queue *q, 855 const struct sk_buff *skb, 856 struct iov_iter *iter) 857 { 858 int ret; 859 int vnet_hdr_len = 0; 860 int vlan_offset = 0; 861 int total; 862 863 if (q->flags & IFF_VNET_HDR) { 864 struct virtio_net_hdr vnet_hdr; 865 vnet_hdr_len = q->vnet_hdr_sz; 866 if (iov_iter_count(iter) < vnet_hdr_len) 867 return -EINVAL; 868 869 macvtap_skb_to_vnet_hdr(q, skb, &vnet_hdr); 870 871 if (copy_to_iter(&vnet_hdr, sizeof(vnet_hdr), iter) != 872 sizeof(vnet_hdr)) 873 return -EFAULT; 874 875 iov_iter_advance(iter, vnet_hdr_len - sizeof(vnet_hdr)); 876 } 877 total = vnet_hdr_len; 878 total += skb->len; 879 880 if (skb_vlan_tag_present(skb)) { 881 struct { 882 __be16 h_vlan_proto; 883 __be16 h_vlan_TCI; 884 } veth; 885 veth.h_vlan_proto = skb->vlan_proto; 886 veth.h_vlan_TCI = htons(skb_vlan_tag_get(skb)); 887 888 vlan_offset = offsetof(struct vlan_ethhdr, h_vlan_proto); 889 total += VLAN_HLEN; 890 891 ret = skb_copy_datagram_iter(skb, 0, iter, vlan_offset); 892 if (ret || !iov_iter_count(iter)) 893 goto done; 894 895 ret = copy_to_iter(&veth, sizeof(veth), iter); 896 if (ret != sizeof(veth) || !iov_iter_count(iter)) 897 goto done; 898 } 899 900 ret = skb_copy_datagram_iter(skb, vlan_offset, iter, 901 skb->len - vlan_offset); 902 903 done: 904 return ret ? ret : total; 905 } 906 907 static ssize_t macvtap_do_read(struct macvtap_queue *q, 908 struct iov_iter *to, 909 int noblock) 910 { 911 DEFINE_WAIT(wait); 912 struct sk_buff *skb; 913 ssize_t ret = 0; 914 915 if (!iov_iter_count(to)) 916 return 0; 917 918 while (1) { 919 if (!noblock) 920 prepare_to_wait(sk_sleep(&q->sk), &wait, 921 TASK_INTERRUPTIBLE); 922 923 /* Read frames from the queue */ 924 skb = skb_dequeue(&q->sk.sk_receive_queue); 925 if (skb) 926 break; 927 if (noblock) { 928 ret = -EAGAIN; 929 break; 930 } 931 if (signal_pending(current)) { 932 ret = -ERESTARTSYS; 933 break; 934 } 935 /* Nothing to read, let's sleep */ 936 schedule(); 937 } 938 if (skb) { 939 ret = macvtap_put_user(q, skb, to); 940 if (unlikely(ret < 0)) 941 kfree_skb(skb); 942 else 943 consume_skb(skb); 944 } 945 if (!noblock) 946 finish_wait(sk_sleep(&q->sk), &wait); 947 return ret; 948 } 949 950 static ssize_t macvtap_read_iter(struct kiocb *iocb, struct iov_iter *to) 951 { 952 struct file *file = iocb->ki_filp; 953 struct macvtap_queue *q = file->private_data; 954 ssize_t len = iov_iter_count(to), ret; 955 956 ret = macvtap_do_read(q, to, file->f_flags & O_NONBLOCK); 957 ret = min_t(ssize_t, ret, len); 958 if (ret > 0) 959 iocb->ki_pos = ret; 960 return ret; 961 } 962 963 static struct macvlan_dev *macvtap_get_vlan(struct macvtap_queue *q) 964 { 965 struct macvlan_dev *vlan; 966 967 ASSERT_RTNL(); 968 vlan = rtnl_dereference(q->vlan); 969 if (vlan) 970 dev_hold(vlan->dev); 971 972 return vlan; 973 } 974 975 static void macvtap_put_vlan(struct macvlan_dev *vlan) 976 { 977 dev_put(vlan->dev); 978 } 979 980 static int macvtap_ioctl_set_queue(struct file *file, unsigned int flags) 981 { 982 struct macvtap_queue *q = file->private_data; 983 struct macvlan_dev *vlan; 984 int ret; 985 986 vlan = macvtap_get_vlan(q); 987 if (!vlan) 988 return -EINVAL; 989 990 if (flags & IFF_ATTACH_QUEUE) 991 ret = macvtap_enable_queue(vlan->dev, file, q); 992 else if (flags & IFF_DETACH_QUEUE) 993 ret = macvtap_disable_queue(q); 994 else 995 ret = -EINVAL; 996 997 macvtap_put_vlan(vlan); 998 return ret; 999 } 1000 1001 static int set_offload(struct macvtap_queue *q, unsigned long arg) 1002 { 1003 struct macvlan_dev *vlan; 1004 netdev_features_t features; 1005 netdev_features_t feature_mask = 0; 1006 1007 vlan = rtnl_dereference(q->vlan); 1008 if (!vlan) 1009 return -ENOLINK; 1010 1011 features = vlan->dev->features; 1012 1013 if (arg & TUN_F_CSUM) { 1014 feature_mask = NETIF_F_HW_CSUM; 1015 1016 if (arg & (TUN_F_TSO4 | TUN_F_TSO6)) { 1017 if (arg & TUN_F_TSO_ECN) 1018 feature_mask |= NETIF_F_TSO_ECN; 1019 if (arg & TUN_F_TSO4) 1020 feature_mask |= NETIF_F_TSO; 1021 if (arg & TUN_F_TSO6) 1022 feature_mask |= NETIF_F_TSO6; 1023 } 1024 1025 if (arg & TUN_F_UFO) 1026 feature_mask |= NETIF_F_UFO; 1027 } 1028 1029 /* tun/tap driver inverts the usage for TSO offloads, where 1030 * setting the TSO bit means that the userspace wants to 1031 * accept TSO frames and turning it off means that user space 1032 * does not support TSO. 1033 * For macvtap, we have to invert it to mean the same thing. 1034 * When user space turns off TSO, we turn off GSO/LRO so that 1035 * user-space will not receive TSO frames. 1036 */ 1037 if (feature_mask & (NETIF_F_TSO | NETIF_F_TSO6 | NETIF_F_UFO)) 1038 features |= RX_OFFLOADS; 1039 else 1040 features &= ~RX_OFFLOADS; 1041 1042 /* tap_features are the same as features on tun/tap and 1043 * reflect user expectations. 1044 */ 1045 vlan->tap_features = feature_mask; 1046 vlan->set_features = features; 1047 netdev_update_features(vlan->dev); 1048 1049 return 0; 1050 } 1051 1052 /* 1053 * provide compatibility with generic tun/tap interface 1054 */ 1055 static long macvtap_ioctl(struct file *file, unsigned int cmd, 1056 unsigned long arg) 1057 { 1058 struct macvtap_queue *q = file->private_data; 1059 struct macvlan_dev *vlan; 1060 void __user *argp = (void __user *)arg; 1061 struct ifreq __user *ifr = argp; 1062 unsigned int __user *up = argp; 1063 unsigned short u; 1064 int __user *sp = argp; 1065 struct sockaddr sa; 1066 int s; 1067 int ret; 1068 1069 switch (cmd) { 1070 case TUNSETIFF: 1071 /* ignore the name, just look at flags */ 1072 if (get_user(u, &ifr->ifr_flags)) 1073 return -EFAULT; 1074 1075 ret = 0; 1076 if ((u & ~MACVTAP_FEATURES) != (IFF_NO_PI | IFF_TAP)) 1077 ret = -EINVAL; 1078 else 1079 q->flags = (q->flags & ~MACVTAP_FEATURES) | u; 1080 1081 return ret; 1082 1083 case TUNGETIFF: 1084 rtnl_lock(); 1085 vlan = macvtap_get_vlan(q); 1086 if (!vlan) { 1087 rtnl_unlock(); 1088 return -ENOLINK; 1089 } 1090 1091 ret = 0; 1092 u = q->flags; 1093 if (copy_to_user(&ifr->ifr_name, vlan->dev->name, IFNAMSIZ) || 1094 put_user(u, &ifr->ifr_flags)) 1095 ret = -EFAULT; 1096 macvtap_put_vlan(vlan); 1097 rtnl_unlock(); 1098 return ret; 1099 1100 case TUNSETQUEUE: 1101 if (get_user(u, &ifr->ifr_flags)) 1102 return -EFAULT; 1103 rtnl_lock(); 1104 ret = macvtap_ioctl_set_queue(file, u); 1105 rtnl_unlock(); 1106 return ret; 1107 1108 case TUNGETFEATURES: 1109 if (put_user(IFF_TAP | IFF_NO_PI | MACVTAP_FEATURES, up)) 1110 return -EFAULT; 1111 return 0; 1112 1113 case TUNSETSNDBUF: 1114 if (get_user(s, sp)) 1115 return -EFAULT; 1116 1117 q->sk.sk_sndbuf = s; 1118 return 0; 1119 1120 case TUNGETVNETHDRSZ: 1121 s = q->vnet_hdr_sz; 1122 if (put_user(s, sp)) 1123 return -EFAULT; 1124 return 0; 1125 1126 case TUNSETVNETHDRSZ: 1127 if (get_user(s, sp)) 1128 return -EFAULT; 1129 if (s < (int)sizeof(struct virtio_net_hdr)) 1130 return -EINVAL; 1131 1132 q->vnet_hdr_sz = s; 1133 return 0; 1134 1135 case TUNGETVNETLE: 1136 s = !!(q->flags & MACVTAP_VNET_LE); 1137 if (put_user(s, sp)) 1138 return -EFAULT; 1139 return 0; 1140 1141 case TUNSETVNETLE: 1142 if (get_user(s, sp)) 1143 return -EFAULT; 1144 if (s) 1145 q->flags |= MACVTAP_VNET_LE; 1146 else 1147 q->flags &= ~MACVTAP_VNET_LE; 1148 return 0; 1149 1150 case TUNGETVNETBE: 1151 return macvtap_get_vnet_be(q, sp); 1152 1153 case TUNSETVNETBE: 1154 return macvtap_set_vnet_be(q, sp); 1155 1156 case TUNSETOFFLOAD: 1157 /* let the user check for future flags */ 1158 if (arg & ~(TUN_F_CSUM | TUN_F_TSO4 | TUN_F_TSO6 | 1159 TUN_F_TSO_ECN | TUN_F_UFO)) 1160 return -EINVAL; 1161 1162 rtnl_lock(); 1163 ret = set_offload(q, arg); 1164 rtnl_unlock(); 1165 return ret; 1166 1167 case SIOCGIFHWADDR: 1168 rtnl_lock(); 1169 vlan = macvtap_get_vlan(q); 1170 if (!vlan) { 1171 rtnl_unlock(); 1172 return -ENOLINK; 1173 } 1174 ret = 0; 1175 u = vlan->dev->type; 1176 if (copy_to_user(&ifr->ifr_name, vlan->dev->name, IFNAMSIZ) || 1177 copy_to_user(&ifr->ifr_hwaddr.sa_data, vlan->dev->dev_addr, ETH_ALEN) || 1178 put_user(u, &ifr->ifr_hwaddr.sa_family)) 1179 ret = -EFAULT; 1180 macvtap_put_vlan(vlan); 1181 rtnl_unlock(); 1182 return ret; 1183 1184 case SIOCSIFHWADDR: 1185 if (copy_from_user(&sa, &ifr->ifr_hwaddr, sizeof(sa))) 1186 return -EFAULT; 1187 rtnl_lock(); 1188 vlan = macvtap_get_vlan(q); 1189 if (!vlan) { 1190 rtnl_unlock(); 1191 return -ENOLINK; 1192 } 1193 ret = dev_set_mac_address(vlan->dev, &sa); 1194 macvtap_put_vlan(vlan); 1195 rtnl_unlock(); 1196 return ret; 1197 1198 default: 1199 return -EINVAL; 1200 } 1201 } 1202 1203 #ifdef CONFIG_COMPAT 1204 static long macvtap_compat_ioctl(struct file *file, unsigned int cmd, 1205 unsigned long arg) 1206 { 1207 return macvtap_ioctl(file, cmd, (unsigned long)compat_ptr(arg)); 1208 } 1209 #endif 1210 1211 static const struct file_operations macvtap_fops = { 1212 .owner = THIS_MODULE, 1213 .open = macvtap_open, 1214 .release = macvtap_release, 1215 .read_iter = macvtap_read_iter, 1216 .write_iter = macvtap_write_iter, 1217 .poll = macvtap_poll, 1218 .llseek = no_llseek, 1219 .unlocked_ioctl = macvtap_ioctl, 1220 #ifdef CONFIG_COMPAT 1221 .compat_ioctl = macvtap_compat_ioctl, 1222 #endif 1223 }; 1224 1225 static int macvtap_sendmsg(struct socket *sock, struct msghdr *m, 1226 size_t total_len) 1227 { 1228 struct macvtap_queue *q = container_of(sock, struct macvtap_queue, sock); 1229 return macvtap_get_user(q, m, &m->msg_iter, m->msg_flags & MSG_DONTWAIT); 1230 } 1231 1232 static int macvtap_recvmsg(struct socket *sock, struct msghdr *m, 1233 size_t total_len, int flags) 1234 { 1235 struct macvtap_queue *q = container_of(sock, struct macvtap_queue, sock); 1236 int ret; 1237 if (flags & ~(MSG_DONTWAIT|MSG_TRUNC)) 1238 return -EINVAL; 1239 ret = macvtap_do_read(q, &m->msg_iter, flags & MSG_DONTWAIT); 1240 if (ret > total_len) { 1241 m->msg_flags |= MSG_TRUNC; 1242 ret = flags & MSG_TRUNC ? ret : total_len; 1243 } 1244 return ret; 1245 } 1246 1247 /* Ops structure to mimic raw sockets with tun */ 1248 static const struct proto_ops macvtap_socket_ops = { 1249 .sendmsg = macvtap_sendmsg, 1250 .recvmsg = macvtap_recvmsg, 1251 }; 1252 1253 /* Get an underlying socket object from tun file. Returns error unless file is 1254 * attached to a device. The returned object works like a packet socket, it 1255 * can be used for sock_sendmsg/sock_recvmsg. The caller is responsible for 1256 * holding a reference to the file for as long as the socket is in use. */ 1257 struct socket *macvtap_get_socket(struct file *file) 1258 { 1259 struct macvtap_queue *q; 1260 if (file->f_op != &macvtap_fops) 1261 return ERR_PTR(-EINVAL); 1262 q = file->private_data; 1263 if (!q) 1264 return ERR_PTR(-EBADFD); 1265 return &q->sock; 1266 } 1267 EXPORT_SYMBOL_GPL(macvtap_get_socket); 1268 1269 static int macvtap_device_event(struct notifier_block *unused, 1270 unsigned long event, void *ptr) 1271 { 1272 struct net_device *dev = netdev_notifier_info_to_dev(ptr); 1273 struct macvlan_dev *vlan; 1274 struct device *classdev; 1275 dev_t devt; 1276 int err; 1277 1278 if (dev->rtnl_link_ops != &macvtap_link_ops) 1279 return NOTIFY_DONE; 1280 1281 vlan = netdev_priv(dev); 1282 1283 switch (event) { 1284 case NETDEV_REGISTER: 1285 /* Create the device node here after the network device has 1286 * been registered but before register_netdevice has 1287 * finished running. 1288 */ 1289 err = macvtap_get_minor(vlan); 1290 if (err) 1291 return notifier_from_errno(err); 1292 1293 devt = MKDEV(MAJOR(macvtap_major), vlan->minor); 1294 classdev = device_create(macvtap_class, &dev->dev, devt, 1295 dev, "tap%d", dev->ifindex); 1296 if (IS_ERR(classdev)) { 1297 macvtap_free_minor(vlan); 1298 return notifier_from_errno(PTR_ERR(classdev)); 1299 } 1300 break; 1301 case NETDEV_UNREGISTER: 1302 devt = MKDEV(MAJOR(macvtap_major), vlan->minor); 1303 device_destroy(macvtap_class, devt); 1304 macvtap_free_minor(vlan); 1305 break; 1306 } 1307 1308 return NOTIFY_DONE; 1309 } 1310 1311 static struct notifier_block macvtap_notifier_block __read_mostly = { 1312 .notifier_call = macvtap_device_event, 1313 }; 1314 1315 static int macvtap_init(void) 1316 { 1317 int err; 1318 1319 err = alloc_chrdev_region(&macvtap_major, 0, 1320 MACVTAP_NUM_DEVS, "macvtap"); 1321 if (err) 1322 goto out1; 1323 1324 cdev_init(&macvtap_cdev, &macvtap_fops); 1325 err = cdev_add(&macvtap_cdev, macvtap_major, MACVTAP_NUM_DEVS); 1326 if (err) 1327 goto out2; 1328 1329 macvtap_class = class_create(THIS_MODULE, "macvtap"); 1330 if (IS_ERR(macvtap_class)) { 1331 err = PTR_ERR(macvtap_class); 1332 goto out3; 1333 } 1334 1335 err = register_netdevice_notifier(&macvtap_notifier_block); 1336 if (err) 1337 goto out4; 1338 1339 err = macvlan_link_register(&macvtap_link_ops); 1340 if (err) 1341 goto out5; 1342 1343 return 0; 1344 1345 out5: 1346 unregister_netdevice_notifier(&macvtap_notifier_block); 1347 out4: 1348 class_unregister(macvtap_class); 1349 out3: 1350 cdev_del(&macvtap_cdev); 1351 out2: 1352 unregister_chrdev_region(macvtap_major, MACVTAP_NUM_DEVS); 1353 out1: 1354 return err; 1355 } 1356 module_init(macvtap_init); 1357 1358 static void macvtap_exit(void) 1359 { 1360 rtnl_link_unregister(&macvtap_link_ops); 1361 unregister_netdevice_notifier(&macvtap_notifier_block); 1362 class_unregister(macvtap_class); 1363 cdev_del(&macvtap_cdev); 1364 unregister_chrdev_region(macvtap_major, MACVTAP_NUM_DEVS); 1365 idr_destroy(&minor_idr); 1366 } 1367 module_exit(macvtap_exit); 1368 1369 MODULE_ALIAS_RTNL_LINK("macvtap"); 1370 MODULE_AUTHOR("Arnd Bergmann <arnd@arndb.de>"); 1371 MODULE_LICENSE("GPL"); 1372