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