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 19 #include <net/net_namespace.h> 20 #include <net/rtnetlink.h> 21 #include <net/sock.h> 22 #include <linux/virtio_net.h> 23 24 /* 25 * A macvtap queue is the central object of this driver, it connects 26 * an open character device to a macvlan interface. There can be 27 * multiple queues on one interface, which map back to queues 28 * implemented in hardware on the underlying device. 29 * 30 * macvtap_proto is used to allocate queues through the sock allocation 31 * mechanism. 32 * 33 */ 34 struct macvtap_queue { 35 struct sock sk; 36 struct socket sock; 37 struct socket_wq wq; 38 int vnet_hdr_sz; 39 struct macvlan_dev __rcu *vlan; 40 struct file *file; 41 unsigned int flags; 42 u16 queue_index; 43 bool enabled; 44 struct list_head next; 45 }; 46 47 static struct proto macvtap_proto = { 48 .name = "macvtap", 49 .owner = THIS_MODULE, 50 .obj_size = sizeof (struct macvtap_queue), 51 }; 52 53 /* 54 * Variables for dealing with macvtaps device numbers. 55 */ 56 static dev_t macvtap_major; 57 #define MACVTAP_NUM_DEVS (1U << MINORBITS) 58 static DEFINE_MUTEX(minor_lock); 59 static DEFINE_IDR(minor_idr); 60 61 #define GOODCOPY_LEN 128 62 static struct class *macvtap_class; 63 static struct cdev macvtap_cdev; 64 65 static const struct proto_ops macvtap_socket_ops; 66 67 #define TUN_OFFLOADS (NETIF_F_HW_CSUM | NETIF_F_TSO_ECN | NETIF_F_TSO | \ 68 NETIF_F_TSO6 | NETIF_F_UFO) 69 #define RX_OFFLOADS (NETIF_F_GRO | NETIF_F_LRO) 70 #define TAP_FEATURES (NETIF_F_GSO | NETIF_F_SG) 71 72 static struct macvlan_dev *macvtap_get_vlan_rcu(const struct net_device *dev) 73 { 74 return rcu_dereference(dev->rx_handler_data); 75 } 76 77 /* 78 * RCU usage: 79 * The macvtap_queue and the macvlan_dev are loosely coupled, the 80 * pointers from one to the other can only be read while rcu_read_lock 81 * or rtnl is held. 82 * 83 * Both the file and the macvlan_dev hold a reference on the macvtap_queue 84 * through sock_hold(&q->sk). When the macvlan_dev goes away first, 85 * q->vlan becomes inaccessible. When the files gets closed, 86 * macvtap_get_queue() fails. 87 * 88 * There may still be references to the struct sock inside of the 89 * queue from outbound SKBs, but these never reference back to the 90 * file or the dev. The data structure is freed through __sk_free 91 * when both our references and any pending SKBs are gone. 92 */ 93 94 static int macvtap_enable_queue(struct net_device *dev, struct file *file, 95 struct macvtap_queue *q) 96 { 97 struct macvlan_dev *vlan = netdev_priv(dev); 98 int err = -EINVAL; 99 100 ASSERT_RTNL(); 101 102 if (q->enabled) 103 goto out; 104 105 err = 0; 106 rcu_assign_pointer(vlan->taps[vlan->numvtaps], q); 107 q->queue_index = vlan->numvtaps; 108 q->enabled = true; 109 110 vlan->numvtaps++; 111 out: 112 return err; 113 } 114 115 static int macvtap_set_queue(struct net_device *dev, struct file *file, 116 struct macvtap_queue *q) 117 { 118 struct macvlan_dev *vlan = netdev_priv(dev); 119 int err = -EBUSY; 120 121 rtnl_lock(); 122 if (vlan->numqueues == MAX_MACVTAP_QUEUES) 123 goto out; 124 125 err = 0; 126 rcu_assign_pointer(q->vlan, vlan); 127 rcu_assign_pointer(vlan->taps[vlan->numvtaps], q); 128 sock_hold(&q->sk); 129 130 q->file = file; 131 q->queue_index = vlan->numvtaps; 132 q->enabled = true; 133 file->private_data = q; 134 list_add_tail(&q->next, &vlan->queue_list); 135 136 vlan->numvtaps++; 137 vlan->numqueues++; 138 139 out: 140 rtnl_unlock(); 141 return err; 142 } 143 144 static int macvtap_disable_queue(struct macvtap_queue *q) 145 { 146 struct macvlan_dev *vlan; 147 struct macvtap_queue *nq; 148 149 ASSERT_RTNL(); 150 if (!q->enabled) 151 return -EINVAL; 152 153 vlan = rtnl_dereference(q->vlan); 154 155 if (vlan) { 156 int index = q->queue_index; 157 BUG_ON(index >= vlan->numvtaps); 158 nq = rtnl_dereference(vlan->taps[vlan->numvtaps - 1]); 159 nq->queue_index = index; 160 161 rcu_assign_pointer(vlan->taps[index], nq); 162 RCU_INIT_POINTER(vlan->taps[vlan->numvtaps - 1], NULL); 163 q->enabled = false; 164 165 vlan->numvtaps--; 166 } 167 168 return 0; 169 } 170 171 /* 172 * The file owning the queue got closed, give up both 173 * the reference that the files holds as well as the 174 * one from the macvlan_dev if that still exists. 175 * 176 * Using the spinlock makes sure that we don't get 177 * to the queue again after destroying it. 178 */ 179 static void macvtap_put_queue(struct macvtap_queue *q) 180 { 181 struct macvlan_dev *vlan; 182 183 rtnl_lock(); 184 vlan = rtnl_dereference(q->vlan); 185 186 if (vlan) { 187 if (q->enabled) 188 BUG_ON(macvtap_disable_queue(q)); 189 190 vlan->numqueues--; 191 RCU_INIT_POINTER(q->vlan, NULL); 192 sock_put(&q->sk); 193 list_del_init(&q->next); 194 } 195 196 rtnl_unlock(); 197 198 synchronize_rcu(); 199 sock_put(&q->sk); 200 } 201 202 /* 203 * Select a queue based on the rxq of the device on which this packet 204 * arrived. If the incoming device is not mq, calculate a flow hash 205 * to select a queue. If all fails, find the first available queue. 206 * Cache vlan->numvtaps since it can become zero during the execution 207 * of this function. 208 */ 209 static struct macvtap_queue *macvtap_get_queue(struct net_device *dev, 210 struct sk_buff *skb) 211 { 212 struct macvlan_dev *vlan = netdev_priv(dev); 213 struct macvtap_queue *tap = NULL; 214 /* Access to taps array is protected by rcu, but access to numvtaps 215 * isn't. Below we use it to lookup a queue, but treat it as a hint 216 * and validate that the result isn't NULL - in case we are 217 * racing against queue removal. 218 */ 219 int numvtaps = ACCESS_ONCE(vlan->numvtaps); 220 __u32 rxq; 221 222 if (!numvtaps) 223 goto out; 224 225 /* Check if we can use flow to select a queue */ 226 rxq = skb_get_hash(skb); 227 if (rxq) { 228 tap = rcu_dereference(vlan->taps[rxq % numvtaps]); 229 goto out; 230 } 231 232 if (likely(skb_rx_queue_recorded(skb))) { 233 rxq = skb_get_rx_queue(skb); 234 235 while (unlikely(rxq >= numvtaps)) 236 rxq -= numvtaps; 237 238 tap = rcu_dereference(vlan->taps[rxq]); 239 goto out; 240 } 241 242 tap = rcu_dereference(vlan->taps[0]); 243 out: 244 return tap; 245 } 246 247 /* 248 * The net_device is going away, give up the reference 249 * that it holds on all queues and safely set the pointer 250 * from the queues to NULL. 251 */ 252 static void macvtap_del_queues(struct net_device *dev) 253 { 254 struct macvlan_dev *vlan = netdev_priv(dev); 255 struct macvtap_queue *q, *tmp, *qlist[MAX_MACVTAP_QUEUES]; 256 int i, j = 0; 257 258 ASSERT_RTNL(); 259 list_for_each_entry_safe(q, tmp, &vlan->queue_list, next) { 260 list_del_init(&q->next); 261 qlist[j++] = q; 262 RCU_INIT_POINTER(q->vlan, NULL); 263 if (q->enabled) 264 vlan->numvtaps--; 265 vlan->numqueues--; 266 } 267 for (i = 0; i < vlan->numvtaps; i++) 268 RCU_INIT_POINTER(vlan->taps[i], NULL); 269 BUG_ON(vlan->numvtaps); 270 BUG_ON(vlan->numqueues); 271 /* guarantee that any future macvtap_set_queue will fail */ 272 vlan->numvtaps = MAX_MACVTAP_QUEUES; 273 274 for (--j; j >= 0; j--) 275 sock_put(&qlist[j]->sk); 276 } 277 278 static rx_handler_result_t macvtap_handle_frame(struct sk_buff **pskb) 279 { 280 struct sk_buff *skb = *pskb; 281 struct net_device *dev = skb->dev; 282 struct macvlan_dev *vlan; 283 struct macvtap_queue *q; 284 netdev_features_t features = TAP_FEATURES; 285 286 vlan = macvtap_get_vlan_rcu(dev); 287 if (!vlan) 288 return RX_HANDLER_PASS; 289 290 q = macvtap_get_queue(dev, skb); 291 if (!q) 292 return RX_HANDLER_PASS; 293 294 if (skb_queue_len(&q->sk.sk_receive_queue) >= dev->tx_queue_len) 295 goto drop; 296 297 skb_push(skb, ETH_HLEN); 298 299 /* Apply the forward feature mask so that we perform segmentation 300 * according to users wishes. This only works if VNET_HDR is 301 * enabled. 302 */ 303 if (q->flags & IFF_VNET_HDR) 304 features |= vlan->tap_features; 305 if (netif_needs_gso(skb, features)) { 306 struct sk_buff *segs = __skb_gso_segment(skb, features, false); 307 308 if (IS_ERR(segs)) 309 goto drop; 310 311 if (!segs) { 312 skb_queue_tail(&q->sk.sk_receive_queue, skb); 313 goto wake_up; 314 } 315 316 kfree_skb(skb); 317 while (segs) { 318 struct sk_buff *nskb = segs->next; 319 320 segs->next = NULL; 321 skb_queue_tail(&q->sk.sk_receive_queue, segs); 322 segs = nskb; 323 } 324 } else { 325 /* If we receive a partial checksum and the tap side 326 * doesn't support checksum offload, compute the checksum. 327 * Note: it doesn't matter which checksum feature to 328 * check, we either support them all or none. 329 */ 330 if (skb->ip_summed == CHECKSUM_PARTIAL && 331 !(features & NETIF_F_ALL_CSUM) && 332 skb_checksum_help(skb)) 333 goto drop; 334 skb_queue_tail(&q->sk.sk_receive_queue, skb); 335 } 336 337 wake_up: 338 wake_up_interruptible_poll(sk_sleep(&q->sk), POLLIN | POLLRDNORM | POLLRDBAND); 339 return RX_HANDLER_CONSUMED; 340 341 drop: 342 /* Count errors/drops only here, thus don't care about args. */ 343 macvlan_count_rx(vlan, 0, 0, 0); 344 kfree_skb(skb); 345 return RX_HANDLER_CONSUMED; 346 } 347 348 static int macvtap_get_minor(struct macvlan_dev *vlan) 349 { 350 int retval = -ENOMEM; 351 352 mutex_lock(&minor_lock); 353 retval = idr_alloc(&minor_idr, vlan, 1, MACVTAP_NUM_DEVS, GFP_KERNEL); 354 if (retval >= 0) { 355 vlan->minor = retval; 356 } else if (retval == -ENOSPC) { 357 printk(KERN_ERR "too many macvtap devices\n"); 358 retval = -EINVAL; 359 } 360 mutex_unlock(&minor_lock); 361 return retval < 0 ? retval : 0; 362 } 363 364 static void macvtap_free_minor(struct macvlan_dev *vlan) 365 { 366 mutex_lock(&minor_lock); 367 if (vlan->minor) { 368 idr_remove(&minor_idr, vlan->minor); 369 vlan->minor = 0; 370 } 371 mutex_unlock(&minor_lock); 372 } 373 374 static struct net_device *dev_get_by_macvtap_minor(int minor) 375 { 376 struct net_device *dev = NULL; 377 struct macvlan_dev *vlan; 378 379 mutex_lock(&minor_lock); 380 vlan = idr_find(&minor_idr, minor); 381 if (vlan) { 382 dev = vlan->dev; 383 dev_hold(dev); 384 } 385 mutex_unlock(&minor_lock); 386 return dev; 387 } 388 389 static int macvtap_newlink(struct net *src_net, 390 struct net_device *dev, 391 struct nlattr *tb[], 392 struct nlattr *data[]) 393 { 394 struct macvlan_dev *vlan = netdev_priv(dev); 395 int err; 396 397 INIT_LIST_HEAD(&vlan->queue_list); 398 399 /* Since macvlan supports all offloads by default, make 400 * tap support all offloads also. 401 */ 402 vlan->tap_features = TUN_OFFLOADS; 403 404 err = netdev_rx_handler_register(dev, macvtap_handle_frame, vlan); 405 if (err) 406 return err; 407 408 /* Don't put anything that may fail after macvlan_common_newlink 409 * because we can't undo what it does. 410 */ 411 return macvlan_common_newlink(src_net, dev, tb, data); 412 } 413 414 static void macvtap_dellink(struct net_device *dev, 415 struct list_head *head) 416 { 417 netdev_rx_handler_unregister(dev); 418 macvtap_del_queues(dev); 419 macvlan_dellink(dev, head); 420 } 421 422 static void macvtap_setup(struct net_device *dev) 423 { 424 macvlan_common_setup(dev); 425 dev->tx_queue_len = TUN_READQ_SIZE; 426 } 427 428 static struct rtnl_link_ops macvtap_link_ops __read_mostly = { 429 .kind = "macvtap", 430 .setup = macvtap_setup, 431 .newlink = macvtap_newlink, 432 .dellink = macvtap_dellink, 433 }; 434 435 436 static void macvtap_sock_write_space(struct sock *sk) 437 { 438 wait_queue_head_t *wqueue; 439 440 if (!sock_writeable(sk) || 441 !test_and_clear_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags)) 442 return; 443 444 wqueue = sk_sleep(sk); 445 if (wqueue && waitqueue_active(wqueue)) 446 wake_up_interruptible_poll(wqueue, POLLOUT | POLLWRNORM | POLLWRBAND); 447 } 448 449 static void macvtap_sock_destruct(struct sock *sk) 450 { 451 skb_queue_purge(&sk->sk_receive_queue); 452 } 453 454 static int macvtap_open(struct inode *inode, struct file *file) 455 { 456 struct net *net = current->nsproxy->net_ns; 457 struct net_device *dev = dev_get_by_macvtap_minor(iminor(inode)); 458 struct macvtap_queue *q; 459 int err; 460 461 err = -ENODEV; 462 if (!dev) 463 goto out; 464 465 err = -ENOMEM; 466 q = (struct macvtap_queue *)sk_alloc(net, AF_UNSPEC, GFP_KERNEL, 467 &macvtap_proto); 468 if (!q) 469 goto out; 470 471 RCU_INIT_POINTER(q->sock.wq, &q->wq); 472 init_waitqueue_head(&q->wq.wait); 473 q->sock.type = SOCK_RAW; 474 q->sock.state = SS_CONNECTED; 475 q->sock.file = file; 476 q->sock.ops = &macvtap_socket_ops; 477 sock_init_data(&q->sock, &q->sk); 478 q->sk.sk_write_space = macvtap_sock_write_space; 479 q->sk.sk_destruct = macvtap_sock_destruct; 480 q->flags = IFF_VNET_HDR | IFF_NO_PI | IFF_TAP; 481 q->vnet_hdr_sz = sizeof(struct virtio_net_hdr); 482 483 /* 484 * so far only KVM virtio_net uses macvtap, enable zero copy between 485 * guest kernel and host kernel when lower device supports zerocopy 486 * 487 * The macvlan supports zerocopy iff the lower device supports zero 488 * copy so we don't have to look at the lower device directly. 489 */ 490 if ((dev->features & NETIF_F_HIGHDMA) && (dev->features & NETIF_F_SG)) 491 sock_set_flag(&q->sk, SOCK_ZEROCOPY); 492 493 err = macvtap_set_queue(dev, file, q); 494 if (err) 495 sock_put(&q->sk); 496 497 out: 498 if (dev) 499 dev_put(dev); 500 501 return err; 502 } 503 504 static int macvtap_release(struct inode *inode, struct file *file) 505 { 506 struct macvtap_queue *q = file->private_data; 507 macvtap_put_queue(q); 508 return 0; 509 } 510 511 static unsigned int macvtap_poll(struct file *file, poll_table * wait) 512 { 513 struct macvtap_queue *q = file->private_data; 514 unsigned int mask = POLLERR; 515 516 if (!q) 517 goto out; 518 519 mask = 0; 520 poll_wait(file, &q->wq.wait, wait); 521 522 if (!skb_queue_empty(&q->sk.sk_receive_queue)) 523 mask |= POLLIN | POLLRDNORM; 524 525 if (sock_writeable(&q->sk) || 526 (!test_and_set_bit(SOCK_ASYNC_NOSPACE, &q->sock.flags) && 527 sock_writeable(&q->sk))) 528 mask |= POLLOUT | POLLWRNORM; 529 530 out: 531 return mask; 532 } 533 534 static inline struct sk_buff *macvtap_alloc_skb(struct sock *sk, size_t prepad, 535 size_t len, size_t linear, 536 int noblock, int *err) 537 { 538 struct sk_buff *skb; 539 540 /* Under a page? Don't bother with paged skb. */ 541 if (prepad + len < PAGE_SIZE || !linear) 542 linear = len; 543 544 skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock, 545 err, 0); 546 if (!skb) 547 return NULL; 548 549 skb_reserve(skb, prepad); 550 skb_put(skb, linear); 551 skb->data_len = len - linear; 552 skb->len += len - linear; 553 554 return skb; 555 } 556 557 /* 558 * macvtap_skb_from_vnet_hdr and macvtap_skb_to_vnet_hdr should 559 * be shared with the tun/tap driver. 560 */ 561 static int macvtap_skb_from_vnet_hdr(struct sk_buff *skb, 562 struct virtio_net_hdr *vnet_hdr) 563 { 564 unsigned short gso_type = 0; 565 if (vnet_hdr->gso_type != VIRTIO_NET_HDR_GSO_NONE) { 566 switch (vnet_hdr->gso_type & ~VIRTIO_NET_HDR_GSO_ECN) { 567 case VIRTIO_NET_HDR_GSO_TCPV4: 568 gso_type = SKB_GSO_TCPV4; 569 break; 570 case VIRTIO_NET_HDR_GSO_TCPV6: 571 gso_type = SKB_GSO_TCPV6; 572 break; 573 case VIRTIO_NET_HDR_GSO_UDP: 574 gso_type = SKB_GSO_UDP; 575 break; 576 default: 577 return -EINVAL; 578 } 579 580 if (vnet_hdr->gso_type & VIRTIO_NET_HDR_GSO_ECN) 581 gso_type |= SKB_GSO_TCP_ECN; 582 583 if (vnet_hdr->gso_size == 0) 584 return -EINVAL; 585 } 586 587 if (vnet_hdr->flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) { 588 if (!skb_partial_csum_set(skb, vnet_hdr->csum_start, 589 vnet_hdr->csum_offset)) 590 return -EINVAL; 591 } 592 593 if (vnet_hdr->gso_type != VIRTIO_NET_HDR_GSO_NONE) { 594 skb_shinfo(skb)->gso_size = vnet_hdr->gso_size; 595 skb_shinfo(skb)->gso_type = gso_type; 596 597 /* Header must be checked, and gso_segs computed. */ 598 skb_shinfo(skb)->gso_type |= SKB_GSO_DODGY; 599 skb_shinfo(skb)->gso_segs = 0; 600 } 601 return 0; 602 } 603 604 static void macvtap_skb_to_vnet_hdr(const struct sk_buff *skb, 605 struct virtio_net_hdr *vnet_hdr) 606 { 607 memset(vnet_hdr, 0, sizeof(*vnet_hdr)); 608 609 if (skb_is_gso(skb)) { 610 struct skb_shared_info *sinfo = skb_shinfo(skb); 611 612 /* This is a hint as to how much should be linear. */ 613 vnet_hdr->hdr_len = skb_headlen(skb); 614 vnet_hdr->gso_size = sinfo->gso_size; 615 if (sinfo->gso_type & SKB_GSO_TCPV4) 616 vnet_hdr->gso_type = VIRTIO_NET_HDR_GSO_TCPV4; 617 else if (sinfo->gso_type & SKB_GSO_TCPV6) 618 vnet_hdr->gso_type = VIRTIO_NET_HDR_GSO_TCPV6; 619 else if (sinfo->gso_type & SKB_GSO_UDP) 620 vnet_hdr->gso_type = VIRTIO_NET_HDR_GSO_UDP; 621 else 622 BUG(); 623 if (sinfo->gso_type & SKB_GSO_TCP_ECN) 624 vnet_hdr->gso_type |= VIRTIO_NET_HDR_GSO_ECN; 625 } else 626 vnet_hdr->gso_type = VIRTIO_NET_HDR_GSO_NONE; 627 628 if (skb->ip_summed == CHECKSUM_PARTIAL) { 629 vnet_hdr->flags = VIRTIO_NET_HDR_F_NEEDS_CSUM; 630 vnet_hdr->csum_start = skb_checksum_start_offset(skb); 631 vnet_hdr->csum_offset = skb->csum_offset; 632 } else if (skb->ip_summed == CHECKSUM_UNNECESSARY) { 633 vnet_hdr->flags = VIRTIO_NET_HDR_F_DATA_VALID; 634 } /* else everything is zero */ 635 } 636 637 /* Get packet from user space buffer */ 638 static ssize_t macvtap_get_user(struct macvtap_queue *q, struct msghdr *m, 639 const struct iovec *iv, unsigned long total_len, 640 size_t count, int noblock) 641 { 642 int good_linear = SKB_MAX_HEAD(NET_IP_ALIGN); 643 struct sk_buff *skb; 644 struct macvlan_dev *vlan; 645 unsigned long len = total_len; 646 int err; 647 struct virtio_net_hdr vnet_hdr = { 0 }; 648 int vnet_hdr_len = 0; 649 int copylen = 0; 650 bool zerocopy = false; 651 size_t linear; 652 653 if (q->flags & IFF_VNET_HDR) { 654 vnet_hdr_len = q->vnet_hdr_sz; 655 656 err = -EINVAL; 657 if (len < vnet_hdr_len) 658 goto err; 659 len -= vnet_hdr_len; 660 661 err = memcpy_fromiovecend((void *)&vnet_hdr, iv, 0, 662 sizeof(vnet_hdr)); 663 if (err < 0) 664 goto err; 665 if ((vnet_hdr.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) && 666 vnet_hdr.csum_start + vnet_hdr.csum_offset + 2 > 667 vnet_hdr.hdr_len) 668 vnet_hdr.hdr_len = vnet_hdr.csum_start + 669 vnet_hdr.csum_offset + 2; 670 err = -EINVAL; 671 if (vnet_hdr.hdr_len > len) 672 goto err; 673 } 674 675 err = -EINVAL; 676 if (unlikely(len < ETH_HLEN)) 677 goto err; 678 679 err = -EMSGSIZE; 680 if (unlikely(count > UIO_MAXIOV)) 681 goto err; 682 683 if (m && m->msg_control && sock_flag(&q->sk, SOCK_ZEROCOPY)) { 684 copylen = vnet_hdr.hdr_len ? vnet_hdr.hdr_len : GOODCOPY_LEN; 685 if (copylen > good_linear) 686 copylen = good_linear; 687 linear = copylen; 688 if (iov_pages(iv, vnet_hdr_len + copylen, count) 689 <= MAX_SKB_FRAGS) 690 zerocopy = true; 691 } 692 693 if (!zerocopy) { 694 copylen = len; 695 if (vnet_hdr.hdr_len > good_linear) 696 linear = good_linear; 697 else 698 linear = vnet_hdr.hdr_len; 699 } 700 701 skb = macvtap_alloc_skb(&q->sk, NET_IP_ALIGN, copylen, 702 linear, noblock, &err); 703 if (!skb) 704 goto err; 705 706 if (zerocopy) 707 err = zerocopy_sg_from_iovec(skb, iv, vnet_hdr_len, count); 708 else { 709 err = skb_copy_datagram_from_iovec(skb, 0, iv, vnet_hdr_len, 710 len); 711 if (!err && m && m->msg_control) { 712 struct ubuf_info *uarg = m->msg_control; 713 uarg->callback(uarg, false); 714 } 715 } 716 717 if (err) 718 goto err_kfree; 719 720 skb_set_network_header(skb, ETH_HLEN); 721 skb_reset_mac_header(skb); 722 skb->protocol = eth_hdr(skb)->h_proto; 723 724 if (vnet_hdr_len) { 725 err = macvtap_skb_from_vnet_hdr(skb, &vnet_hdr); 726 if (err) 727 goto err_kfree; 728 } 729 730 skb_probe_transport_header(skb, ETH_HLEN); 731 732 rcu_read_lock(); 733 vlan = rcu_dereference(q->vlan); 734 /* copy skb_ubuf_info for callback when skb has no error */ 735 if (zerocopy) { 736 skb_shinfo(skb)->destructor_arg = m->msg_control; 737 skb_shinfo(skb)->tx_flags |= SKBTX_DEV_ZEROCOPY; 738 skb_shinfo(skb)->tx_flags |= SKBTX_SHARED_FRAG; 739 } 740 if (vlan) { 741 skb->dev = vlan->dev; 742 dev_queue_xmit(skb); 743 } else { 744 kfree_skb(skb); 745 } 746 rcu_read_unlock(); 747 748 return total_len; 749 750 err_kfree: 751 kfree_skb(skb); 752 753 err: 754 rcu_read_lock(); 755 vlan = rcu_dereference(q->vlan); 756 if (vlan) 757 this_cpu_inc(vlan->pcpu_stats->tx_dropped); 758 rcu_read_unlock(); 759 760 return err; 761 } 762 763 static ssize_t macvtap_aio_write(struct kiocb *iocb, const struct iovec *iv, 764 unsigned long count, loff_t pos) 765 { 766 struct file *file = iocb->ki_filp; 767 ssize_t result = -ENOLINK; 768 struct macvtap_queue *q = file->private_data; 769 770 result = macvtap_get_user(q, NULL, iv, iov_length(iv, count), count, 771 file->f_flags & O_NONBLOCK); 772 return result; 773 } 774 775 /* Put packet to the user space buffer */ 776 static ssize_t macvtap_put_user(struct macvtap_queue *q, 777 const struct sk_buff *skb, 778 const struct iovec *iv, int len) 779 { 780 int ret; 781 int vnet_hdr_len = 0; 782 int vlan_offset = 0; 783 int copied, total; 784 785 if (q->flags & IFF_VNET_HDR) { 786 struct virtio_net_hdr vnet_hdr; 787 vnet_hdr_len = q->vnet_hdr_sz; 788 if ((len -= vnet_hdr_len) < 0) 789 return -EINVAL; 790 791 macvtap_skb_to_vnet_hdr(skb, &vnet_hdr); 792 793 if (memcpy_toiovecend(iv, (void *)&vnet_hdr, 0, sizeof(vnet_hdr))) 794 return -EFAULT; 795 } 796 total = copied = vnet_hdr_len; 797 total += skb->len; 798 799 if (!vlan_tx_tag_present(skb)) 800 len = min_t(int, skb->len, len); 801 else { 802 int copy; 803 struct { 804 __be16 h_vlan_proto; 805 __be16 h_vlan_TCI; 806 } veth; 807 veth.h_vlan_proto = skb->vlan_proto; 808 veth.h_vlan_TCI = htons(vlan_tx_tag_get(skb)); 809 810 vlan_offset = offsetof(struct vlan_ethhdr, h_vlan_proto); 811 len = min_t(int, skb->len + VLAN_HLEN, len); 812 total += VLAN_HLEN; 813 814 copy = min_t(int, vlan_offset, len); 815 ret = skb_copy_datagram_const_iovec(skb, 0, iv, copied, copy); 816 len -= copy; 817 copied += copy; 818 if (ret || !len) 819 goto done; 820 821 copy = min_t(int, sizeof(veth), len); 822 ret = memcpy_toiovecend(iv, (void *)&veth, copied, copy); 823 len -= copy; 824 copied += copy; 825 if (ret || !len) 826 goto done; 827 } 828 829 ret = skb_copy_datagram_const_iovec(skb, vlan_offset, iv, copied, len); 830 831 done: 832 return ret ? ret : total; 833 } 834 835 static ssize_t macvtap_do_read(struct macvtap_queue *q, 836 const struct iovec *iv, unsigned long len, 837 int noblock) 838 { 839 DEFINE_WAIT(wait); 840 struct sk_buff *skb; 841 ssize_t ret = 0; 842 843 while (len) { 844 if (!noblock) 845 prepare_to_wait(sk_sleep(&q->sk), &wait, 846 TASK_INTERRUPTIBLE); 847 848 /* Read frames from the queue */ 849 skb = skb_dequeue(&q->sk.sk_receive_queue); 850 if (!skb) { 851 if (noblock) { 852 ret = -EAGAIN; 853 break; 854 } 855 if (signal_pending(current)) { 856 ret = -ERESTARTSYS; 857 break; 858 } 859 /* Nothing to read, let's sleep */ 860 schedule(); 861 continue; 862 } 863 ret = macvtap_put_user(q, skb, iv, len); 864 kfree_skb(skb); 865 break; 866 } 867 868 if (!noblock) 869 finish_wait(sk_sleep(&q->sk), &wait); 870 return ret; 871 } 872 873 static ssize_t macvtap_aio_read(struct kiocb *iocb, const struct iovec *iv, 874 unsigned long count, loff_t pos) 875 { 876 struct file *file = iocb->ki_filp; 877 struct macvtap_queue *q = file->private_data; 878 ssize_t len, ret = 0; 879 880 len = iov_length(iv, count); 881 if (len < 0) { 882 ret = -EINVAL; 883 goto out; 884 } 885 886 ret = macvtap_do_read(q, iv, len, file->f_flags & O_NONBLOCK); 887 ret = min_t(ssize_t, ret, len); 888 if (ret > 0) 889 iocb->ki_pos = ret; 890 out: 891 return ret; 892 } 893 894 static struct macvlan_dev *macvtap_get_vlan(struct macvtap_queue *q) 895 { 896 struct macvlan_dev *vlan; 897 898 ASSERT_RTNL(); 899 vlan = rtnl_dereference(q->vlan); 900 if (vlan) 901 dev_hold(vlan->dev); 902 903 return vlan; 904 } 905 906 static void macvtap_put_vlan(struct macvlan_dev *vlan) 907 { 908 dev_put(vlan->dev); 909 } 910 911 static int macvtap_ioctl_set_queue(struct file *file, unsigned int flags) 912 { 913 struct macvtap_queue *q = file->private_data; 914 struct macvlan_dev *vlan; 915 int ret; 916 917 vlan = macvtap_get_vlan(q); 918 if (!vlan) 919 return -EINVAL; 920 921 if (flags & IFF_ATTACH_QUEUE) 922 ret = macvtap_enable_queue(vlan->dev, file, q); 923 else if (flags & IFF_DETACH_QUEUE) 924 ret = macvtap_disable_queue(q); 925 else 926 ret = -EINVAL; 927 928 macvtap_put_vlan(vlan); 929 return ret; 930 } 931 932 static int set_offload(struct macvtap_queue *q, unsigned long arg) 933 { 934 struct macvlan_dev *vlan; 935 netdev_features_t features; 936 netdev_features_t feature_mask = 0; 937 938 vlan = rtnl_dereference(q->vlan); 939 if (!vlan) 940 return -ENOLINK; 941 942 features = vlan->dev->features; 943 944 if (arg & TUN_F_CSUM) { 945 feature_mask = NETIF_F_HW_CSUM; 946 947 if (arg & (TUN_F_TSO4 | TUN_F_TSO6)) { 948 if (arg & TUN_F_TSO_ECN) 949 feature_mask |= NETIF_F_TSO_ECN; 950 if (arg & TUN_F_TSO4) 951 feature_mask |= NETIF_F_TSO; 952 if (arg & TUN_F_TSO6) 953 feature_mask |= NETIF_F_TSO6; 954 } 955 956 if (arg & TUN_F_UFO) 957 feature_mask |= NETIF_F_UFO; 958 } 959 960 /* tun/tap driver inverts the usage for TSO offloads, where 961 * setting the TSO bit means that the userspace wants to 962 * accept TSO frames and turning it off means that user space 963 * does not support TSO. 964 * For macvtap, we have to invert it to mean the same thing. 965 * When user space turns off TSO, we turn off GSO/LRO so that 966 * user-space will not receive TSO frames. 967 */ 968 if (feature_mask & (NETIF_F_TSO | NETIF_F_TSO6 | NETIF_F_UFO)) 969 features |= RX_OFFLOADS; 970 else 971 features &= ~RX_OFFLOADS; 972 973 /* tap_features are the same as features on tun/tap and 974 * reflect user expectations. 975 */ 976 vlan->tap_features = feature_mask; 977 vlan->set_features = features; 978 netdev_update_features(vlan->dev); 979 980 return 0; 981 } 982 983 /* 984 * provide compatibility with generic tun/tap interface 985 */ 986 static long macvtap_ioctl(struct file *file, unsigned int cmd, 987 unsigned long arg) 988 { 989 struct macvtap_queue *q = file->private_data; 990 struct macvlan_dev *vlan; 991 void __user *argp = (void __user *)arg; 992 struct ifreq __user *ifr = argp; 993 unsigned int __user *up = argp; 994 unsigned int u; 995 int __user *sp = argp; 996 int s; 997 int ret; 998 999 switch (cmd) { 1000 case TUNSETIFF: 1001 /* ignore the name, just look at flags */ 1002 if (get_user(u, &ifr->ifr_flags)) 1003 return -EFAULT; 1004 1005 ret = 0; 1006 if ((u & ~(IFF_VNET_HDR | IFF_MULTI_QUEUE)) != 1007 (IFF_NO_PI | IFF_TAP)) 1008 ret = -EINVAL; 1009 else 1010 q->flags = u; 1011 1012 return ret; 1013 1014 case TUNGETIFF: 1015 rtnl_lock(); 1016 vlan = macvtap_get_vlan(q); 1017 if (!vlan) { 1018 rtnl_unlock(); 1019 return -ENOLINK; 1020 } 1021 1022 ret = 0; 1023 if (copy_to_user(&ifr->ifr_name, vlan->dev->name, IFNAMSIZ) || 1024 put_user(q->flags, &ifr->ifr_flags)) 1025 ret = -EFAULT; 1026 macvtap_put_vlan(vlan); 1027 rtnl_unlock(); 1028 return ret; 1029 1030 case TUNSETQUEUE: 1031 if (get_user(u, &ifr->ifr_flags)) 1032 return -EFAULT; 1033 rtnl_lock(); 1034 ret = macvtap_ioctl_set_queue(file, u); 1035 rtnl_unlock(); 1036 return ret; 1037 1038 case TUNGETFEATURES: 1039 if (put_user(IFF_TAP | IFF_NO_PI | IFF_VNET_HDR | 1040 IFF_MULTI_QUEUE, up)) 1041 return -EFAULT; 1042 return 0; 1043 1044 case TUNSETSNDBUF: 1045 if (get_user(u, up)) 1046 return -EFAULT; 1047 1048 q->sk.sk_sndbuf = u; 1049 return 0; 1050 1051 case TUNGETVNETHDRSZ: 1052 s = q->vnet_hdr_sz; 1053 if (put_user(s, sp)) 1054 return -EFAULT; 1055 return 0; 1056 1057 case TUNSETVNETHDRSZ: 1058 if (get_user(s, sp)) 1059 return -EFAULT; 1060 if (s < (int)sizeof(struct virtio_net_hdr)) 1061 return -EINVAL; 1062 1063 q->vnet_hdr_sz = s; 1064 return 0; 1065 1066 case TUNSETOFFLOAD: 1067 /* let the user check for future flags */ 1068 if (arg & ~(TUN_F_CSUM | TUN_F_TSO4 | TUN_F_TSO6 | 1069 TUN_F_TSO_ECN | TUN_F_UFO)) 1070 return -EINVAL; 1071 1072 rtnl_lock(); 1073 ret = set_offload(q, arg); 1074 rtnl_unlock(); 1075 return ret; 1076 1077 default: 1078 return -EINVAL; 1079 } 1080 } 1081 1082 #ifdef CONFIG_COMPAT 1083 static long macvtap_compat_ioctl(struct file *file, unsigned int cmd, 1084 unsigned long arg) 1085 { 1086 return macvtap_ioctl(file, cmd, (unsigned long)compat_ptr(arg)); 1087 } 1088 #endif 1089 1090 static const struct file_operations macvtap_fops = { 1091 .owner = THIS_MODULE, 1092 .open = macvtap_open, 1093 .release = macvtap_release, 1094 .aio_read = macvtap_aio_read, 1095 .aio_write = macvtap_aio_write, 1096 .poll = macvtap_poll, 1097 .llseek = no_llseek, 1098 .unlocked_ioctl = macvtap_ioctl, 1099 #ifdef CONFIG_COMPAT 1100 .compat_ioctl = macvtap_compat_ioctl, 1101 #endif 1102 }; 1103 1104 static int macvtap_sendmsg(struct kiocb *iocb, struct socket *sock, 1105 struct msghdr *m, size_t total_len) 1106 { 1107 struct macvtap_queue *q = container_of(sock, struct macvtap_queue, sock); 1108 return macvtap_get_user(q, m, m->msg_iov, total_len, m->msg_iovlen, 1109 m->msg_flags & MSG_DONTWAIT); 1110 } 1111 1112 static int macvtap_recvmsg(struct kiocb *iocb, struct socket *sock, 1113 struct msghdr *m, size_t total_len, 1114 int flags) 1115 { 1116 struct macvtap_queue *q = container_of(sock, struct macvtap_queue, sock); 1117 int ret; 1118 if (flags & ~(MSG_DONTWAIT|MSG_TRUNC)) 1119 return -EINVAL; 1120 ret = macvtap_do_read(q, m->msg_iov, total_len, 1121 flags & MSG_DONTWAIT); 1122 if (ret > total_len) { 1123 m->msg_flags |= MSG_TRUNC; 1124 ret = flags & MSG_TRUNC ? ret : total_len; 1125 } 1126 return ret; 1127 } 1128 1129 /* Ops structure to mimic raw sockets with tun */ 1130 static const struct proto_ops macvtap_socket_ops = { 1131 .sendmsg = macvtap_sendmsg, 1132 .recvmsg = macvtap_recvmsg, 1133 }; 1134 1135 /* Get an underlying socket object from tun file. Returns error unless file is 1136 * attached to a device. The returned object works like a packet socket, it 1137 * can be used for sock_sendmsg/sock_recvmsg. The caller is responsible for 1138 * holding a reference to the file for as long as the socket is in use. */ 1139 struct socket *macvtap_get_socket(struct file *file) 1140 { 1141 struct macvtap_queue *q; 1142 if (file->f_op != &macvtap_fops) 1143 return ERR_PTR(-EINVAL); 1144 q = file->private_data; 1145 if (!q) 1146 return ERR_PTR(-EBADFD); 1147 return &q->sock; 1148 } 1149 EXPORT_SYMBOL_GPL(macvtap_get_socket); 1150 1151 static int macvtap_device_event(struct notifier_block *unused, 1152 unsigned long event, void *ptr) 1153 { 1154 struct net_device *dev = netdev_notifier_info_to_dev(ptr); 1155 struct macvlan_dev *vlan; 1156 struct device *classdev; 1157 dev_t devt; 1158 int err; 1159 1160 if (dev->rtnl_link_ops != &macvtap_link_ops) 1161 return NOTIFY_DONE; 1162 1163 vlan = netdev_priv(dev); 1164 1165 switch (event) { 1166 case NETDEV_REGISTER: 1167 /* Create the device node here after the network device has 1168 * been registered but before register_netdevice has 1169 * finished running. 1170 */ 1171 err = macvtap_get_minor(vlan); 1172 if (err) 1173 return notifier_from_errno(err); 1174 1175 devt = MKDEV(MAJOR(macvtap_major), vlan->minor); 1176 classdev = device_create(macvtap_class, &dev->dev, devt, 1177 dev, "tap%d", dev->ifindex); 1178 if (IS_ERR(classdev)) { 1179 macvtap_free_minor(vlan); 1180 return notifier_from_errno(PTR_ERR(classdev)); 1181 } 1182 break; 1183 case NETDEV_UNREGISTER: 1184 devt = MKDEV(MAJOR(macvtap_major), vlan->minor); 1185 device_destroy(macvtap_class, devt); 1186 macvtap_free_minor(vlan); 1187 break; 1188 } 1189 1190 return NOTIFY_DONE; 1191 } 1192 1193 static struct notifier_block macvtap_notifier_block __read_mostly = { 1194 .notifier_call = macvtap_device_event, 1195 }; 1196 1197 static int macvtap_init(void) 1198 { 1199 int err; 1200 1201 err = alloc_chrdev_region(&macvtap_major, 0, 1202 MACVTAP_NUM_DEVS, "macvtap"); 1203 if (err) 1204 goto out1; 1205 1206 cdev_init(&macvtap_cdev, &macvtap_fops); 1207 err = cdev_add(&macvtap_cdev, macvtap_major, MACVTAP_NUM_DEVS); 1208 if (err) 1209 goto out2; 1210 1211 macvtap_class = class_create(THIS_MODULE, "macvtap"); 1212 if (IS_ERR(macvtap_class)) { 1213 err = PTR_ERR(macvtap_class); 1214 goto out3; 1215 } 1216 1217 err = register_netdevice_notifier(&macvtap_notifier_block); 1218 if (err) 1219 goto out4; 1220 1221 err = macvlan_link_register(&macvtap_link_ops); 1222 if (err) 1223 goto out5; 1224 1225 return 0; 1226 1227 out5: 1228 unregister_netdevice_notifier(&macvtap_notifier_block); 1229 out4: 1230 class_unregister(macvtap_class); 1231 out3: 1232 cdev_del(&macvtap_cdev); 1233 out2: 1234 unregister_chrdev_region(macvtap_major, MACVTAP_NUM_DEVS); 1235 out1: 1236 return err; 1237 } 1238 module_init(macvtap_init); 1239 1240 static void macvtap_exit(void) 1241 { 1242 rtnl_link_unregister(&macvtap_link_ops); 1243 unregister_netdevice_notifier(&macvtap_notifier_block); 1244 class_unregister(macvtap_class); 1245 cdev_del(&macvtap_cdev); 1246 unregister_chrdev_region(macvtap_major, MACVTAP_NUM_DEVS); 1247 } 1248 module_exit(macvtap_exit); 1249 1250 MODULE_ALIAS_RTNL_LINK("macvtap"); 1251 MODULE_AUTHOR("Arnd Bergmann <arnd@arndb.de>"); 1252 MODULE_LICENSE("GPL"); 1253