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 52 static inline u16 macvtap16_to_cpu(struct macvtap_queue *q, __virtio16 val) 53 { 54 return __virtio16_to_cpu(q->flags & MACVTAP_VNET_LE, val); 55 } 56 57 static inline __virtio16 cpu_to_macvtap16(struct macvtap_queue *q, u16 val) 58 { 59 return __cpu_to_virtio16(q->flags & MACVTAP_VNET_LE, val); 60 } 61 62 static struct proto macvtap_proto = { 63 .name = "macvtap", 64 .owner = THIS_MODULE, 65 .obj_size = sizeof (struct macvtap_queue), 66 }; 67 68 /* 69 * Variables for dealing with macvtaps device numbers. 70 */ 71 static dev_t macvtap_major; 72 #define MACVTAP_NUM_DEVS (1U << MINORBITS) 73 static DEFINE_MUTEX(minor_lock); 74 static DEFINE_IDR(minor_idr); 75 76 #define GOODCOPY_LEN 128 77 static struct class *macvtap_class; 78 static struct cdev macvtap_cdev; 79 80 static const struct proto_ops macvtap_socket_ops; 81 82 #define TUN_OFFLOADS (NETIF_F_HW_CSUM | NETIF_F_TSO_ECN | NETIF_F_TSO | \ 83 NETIF_F_TSO6 | NETIF_F_UFO) 84 #define RX_OFFLOADS (NETIF_F_GRO | NETIF_F_LRO) 85 #define TAP_FEATURES (NETIF_F_GSO | NETIF_F_SG) 86 87 static struct macvlan_dev *macvtap_get_vlan_rcu(const struct net_device *dev) 88 { 89 return rcu_dereference(dev->rx_handler_data); 90 } 91 92 /* 93 * RCU usage: 94 * The macvtap_queue and the macvlan_dev are loosely coupled, the 95 * pointers from one to the other can only be read while rcu_read_lock 96 * or rtnl is held. 97 * 98 * Both the file and the macvlan_dev hold a reference on the macvtap_queue 99 * through sock_hold(&q->sk). When the macvlan_dev goes away first, 100 * q->vlan becomes inaccessible. When the files gets closed, 101 * macvtap_get_queue() fails. 102 * 103 * There may still be references to the struct sock inside of the 104 * queue from outbound SKBs, but these never reference back to the 105 * file or the dev. The data structure is freed through __sk_free 106 * when both our references and any pending SKBs are gone. 107 */ 108 109 static int macvtap_enable_queue(struct net_device *dev, struct file *file, 110 struct macvtap_queue *q) 111 { 112 struct macvlan_dev *vlan = netdev_priv(dev); 113 int err = -EINVAL; 114 115 ASSERT_RTNL(); 116 117 if (q->enabled) 118 goto out; 119 120 err = 0; 121 rcu_assign_pointer(vlan->taps[vlan->numvtaps], q); 122 q->queue_index = vlan->numvtaps; 123 q->enabled = true; 124 125 vlan->numvtaps++; 126 out: 127 return err; 128 } 129 130 /* Requires RTNL */ 131 static int macvtap_set_queue(struct net_device *dev, struct file *file, 132 struct macvtap_queue *q) 133 { 134 struct macvlan_dev *vlan = netdev_priv(dev); 135 136 if (vlan->numqueues == MAX_MACVTAP_QUEUES) 137 return -EBUSY; 138 139 rcu_assign_pointer(q->vlan, vlan); 140 rcu_assign_pointer(vlan->taps[vlan->numvtaps], q); 141 sock_hold(&q->sk); 142 143 q->file = file; 144 q->queue_index = vlan->numvtaps; 145 q->enabled = true; 146 file->private_data = q; 147 list_add_tail(&q->next, &vlan->queue_list); 148 149 vlan->numvtaps++; 150 vlan->numqueues++; 151 152 return 0; 153 } 154 155 static int macvtap_disable_queue(struct macvtap_queue *q) 156 { 157 struct macvlan_dev *vlan; 158 struct macvtap_queue *nq; 159 160 ASSERT_RTNL(); 161 if (!q->enabled) 162 return -EINVAL; 163 164 vlan = rtnl_dereference(q->vlan); 165 166 if (vlan) { 167 int index = q->queue_index; 168 BUG_ON(index >= vlan->numvtaps); 169 nq = rtnl_dereference(vlan->taps[vlan->numvtaps - 1]); 170 nq->queue_index = index; 171 172 rcu_assign_pointer(vlan->taps[index], nq); 173 RCU_INIT_POINTER(vlan->taps[vlan->numvtaps - 1], NULL); 174 q->enabled = false; 175 176 vlan->numvtaps--; 177 } 178 179 return 0; 180 } 181 182 /* 183 * The file owning the queue got closed, give up both 184 * the reference that the files holds as well as the 185 * one from the macvlan_dev if that still exists. 186 * 187 * Using the spinlock makes sure that we don't get 188 * to the queue again after destroying it. 189 */ 190 static void macvtap_put_queue(struct macvtap_queue *q) 191 { 192 struct macvlan_dev *vlan; 193 194 rtnl_lock(); 195 vlan = rtnl_dereference(q->vlan); 196 197 if (vlan) { 198 if (q->enabled) 199 BUG_ON(macvtap_disable_queue(q)); 200 201 vlan->numqueues--; 202 RCU_INIT_POINTER(q->vlan, NULL); 203 sock_put(&q->sk); 204 list_del_init(&q->next); 205 } 206 207 rtnl_unlock(); 208 209 synchronize_rcu(); 210 sock_put(&q->sk); 211 } 212 213 /* 214 * Select a queue based on the rxq of the device on which this packet 215 * arrived. If the incoming device is not mq, calculate a flow hash 216 * to select a queue. If all fails, find the first available queue. 217 * Cache vlan->numvtaps since it can become zero during the execution 218 * of this function. 219 */ 220 static struct macvtap_queue *macvtap_get_queue(struct net_device *dev, 221 struct sk_buff *skb) 222 { 223 struct macvlan_dev *vlan = netdev_priv(dev); 224 struct macvtap_queue *tap = NULL; 225 /* Access to taps array is protected by rcu, but access to numvtaps 226 * isn't. Below we use it to lookup a queue, but treat it as a hint 227 * and validate that the result isn't NULL - in case we are 228 * racing against queue removal. 229 */ 230 int numvtaps = ACCESS_ONCE(vlan->numvtaps); 231 __u32 rxq; 232 233 if (!numvtaps) 234 goto out; 235 236 /* Check if we can use flow to select a queue */ 237 rxq = skb_get_hash(skb); 238 if (rxq) { 239 tap = rcu_dereference(vlan->taps[rxq % numvtaps]); 240 goto out; 241 } 242 243 if (likely(skb_rx_queue_recorded(skb))) { 244 rxq = skb_get_rx_queue(skb); 245 246 while (unlikely(rxq >= numvtaps)) 247 rxq -= numvtaps; 248 249 tap = rcu_dereference(vlan->taps[rxq]); 250 goto out; 251 } 252 253 tap = rcu_dereference(vlan->taps[0]); 254 out: 255 return tap; 256 } 257 258 /* 259 * The net_device is going away, give up the reference 260 * that it holds on all queues and safely set the pointer 261 * from the queues to NULL. 262 */ 263 static void macvtap_del_queues(struct net_device *dev) 264 { 265 struct macvlan_dev *vlan = netdev_priv(dev); 266 struct macvtap_queue *q, *tmp; 267 268 ASSERT_RTNL(); 269 list_for_each_entry_safe(q, tmp, &vlan->queue_list, next) { 270 list_del_init(&q->next); 271 RCU_INIT_POINTER(q->vlan, NULL); 272 if (q->enabled) 273 vlan->numvtaps--; 274 vlan->numqueues--; 275 sock_put(&q->sk); 276 } 277 BUG_ON(vlan->numvtaps); 278 BUG_ON(vlan->numqueues); 279 /* guarantee that any future macvtap_set_queue will fail */ 280 vlan->numvtaps = MAX_MACVTAP_QUEUES; 281 } 282 283 static rx_handler_result_t macvtap_handle_frame(struct sk_buff **pskb) 284 { 285 struct sk_buff *skb = *pskb; 286 struct net_device *dev = skb->dev; 287 struct macvlan_dev *vlan; 288 struct macvtap_queue *q; 289 netdev_features_t features = TAP_FEATURES; 290 291 vlan = macvtap_get_vlan_rcu(dev); 292 if (!vlan) 293 return RX_HANDLER_PASS; 294 295 q = macvtap_get_queue(dev, skb); 296 if (!q) 297 return RX_HANDLER_PASS; 298 299 if (skb_queue_len(&q->sk.sk_receive_queue) >= dev->tx_queue_len) 300 goto drop; 301 302 skb_push(skb, ETH_HLEN); 303 304 /* Apply the forward feature mask so that we perform segmentation 305 * according to users wishes. This only works if VNET_HDR is 306 * enabled. 307 */ 308 if (q->flags & IFF_VNET_HDR) 309 features |= vlan->tap_features; 310 if (netif_needs_gso(skb, features)) { 311 struct sk_buff *segs = __skb_gso_segment(skb, features, false); 312 313 if (IS_ERR(segs)) 314 goto drop; 315 316 if (!segs) { 317 skb_queue_tail(&q->sk.sk_receive_queue, skb); 318 goto wake_up; 319 } 320 321 kfree_skb(skb); 322 while (segs) { 323 struct sk_buff *nskb = segs->next; 324 325 segs->next = NULL; 326 skb_queue_tail(&q->sk.sk_receive_queue, segs); 327 segs = nskb; 328 } 329 } else { 330 /* If we receive a partial checksum and the tap side 331 * doesn't support checksum offload, compute the checksum. 332 * Note: it doesn't matter which checksum feature to 333 * check, we either support them all or none. 334 */ 335 if (skb->ip_summed == CHECKSUM_PARTIAL && 336 !(features & NETIF_F_ALL_CSUM) && 337 skb_checksum_help(skb)) 338 goto drop; 339 skb_queue_tail(&q->sk.sk_receive_queue, skb); 340 } 341 342 wake_up: 343 wake_up_interruptible_poll(sk_sleep(&q->sk), POLLIN | POLLRDNORM | POLLRDBAND); 344 return RX_HANDLER_CONSUMED; 345 346 drop: 347 /* Count errors/drops only here, thus don't care about args. */ 348 macvlan_count_rx(vlan, 0, 0, 0); 349 kfree_skb(skb); 350 return RX_HANDLER_CONSUMED; 351 } 352 353 static int macvtap_get_minor(struct macvlan_dev *vlan) 354 { 355 int retval = -ENOMEM; 356 357 mutex_lock(&minor_lock); 358 retval = idr_alloc(&minor_idr, vlan, 1, MACVTAP_NUM_DEVS, GFP_KERNEL); 359 if (retval >= 0) { 360 vlan->minor = retval; 361 } else if (retval == -ENOSPC) { 362 printk(KERN_ERR "too many macvtap devices\n"); 363 retval = -EINVAL; 364 } 365 mutex_unlock(&minor_lock); 366 return retval < 0 ? retval : 0; 367 } 368 369 static void macvtap_free_minor(struct macvlan_dev *vlan) 370 { 371 mutex_lock(&minor_lock); 372 if (vlan->minor) { 373 idr_remove(&minor_idr, vlan->minor); 374 vlan->minor = 0; 375 } 376 mutex_unlock(&minor_lock); 377 } 378 379 static struct net_device *dev_get_by_macvtap_minor(int minor) 380 { 381 struct net_device *dev = NULL; 382 struct macvlan_dev *vlan; 383 384 mutex_lock(&minor_lock); 385 vlan = idr_find(&minor_idr, minor); 386 if (vlan) { 387 dev = vlan->dev; 388 dev_hold(dev); 389 } 390 mutex_unlock(&minor_lock); 391 return dev; 392 } 393 394 static int macvtap_newlink(struct net *src_net, 395 struct net_device *dev, 396 struct nlattr *tb[], 397 struct nlattr *data[]) 398 { 399 struct macvlan_dev *vlan = netdev_priv(dev); 400 int err; 401 402 INIT_LIST_HEAD(&vlan->queue_list); 403 404 /* Since macvlan supports all offloads by default, make 405 * tap support all offloads also. 406 */ 407 vlan->tap_features = TUN_OFFLOADS; 408 409 err = netdev_rx_handler_register(dev, macvtap_handle_frame, vlan); 410 if (err) 411 return err; 412 413 /* Don't put anything that may fail after macvlan_common_newlink 414 * because we can't undo what it does. 415 */ 416 return macvlan_common_newlink(src_net, dev, tb, data); 417 } 418 419 static void macvtap_dellink(struct net_device *dev, 420 struct list_head *head) 421 { 422 netdev_rx_handler_unregister(dev); 423 macvtap_del_queues(dev); 424 macvlan_dellink(dev, head); 425 } 426 427 static void macvtap_setup(struct net_device *dev) 428 { 429 macvlan_common_setup(dev); 430 dev->tx_queue_len = TUN_READQ_SIZE; 431 } 432 433 static struct rtnl_link_ops macvtap_link_ops __read_mostly = { 434 .kind = "macvtap", 435 .setup = macvtap_setup, 436 .newlink = macvtap_newlink, 437 .dellink = macvtap_dellink, 438 }; 439 440 441 static void macvtap_sock_write_space(struct sock *sk) 442 { 443 wait_queue_head_t *wqueue; 444 445 if (!sock_writeable(sk) || 446 !test_and_clear_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags)) 447 return; 448 449 wqueue = sk_sleep(sk); 450 if (wqueue && waitqueue_active(wqueue)) 451 wake_up_interruptible_poll(wqueue, POLLOUT | POLLWRNORM | POLLWRBAND); 452 } 453 454 static void macvtap_sock_destruct(struct sock *sk) 455 { 456 skb_queue_purge(&sk->sk_receive_queue); 457 } 458 459 static int macvtap_open(struct inode *inode, struct file *file) 460 { 461 struct net *net = current->nsproxy->net_ns; 462 struct net_device *dev; 463 struct macvtap_queue *q; 464 int err = -ENODEV; 465 466 rtnl_lock(); 467 dev = dev_get_by_macvtap_minor(iminor(inode)); 468 if (!dev) 469 goto out; 470 471 err = -ENOMEM; 472 q = (struct macvtap_queue *)sk_alloc(net, AF_UNSPEC, GFP_KERNEL, 473 &macvtap_proto, 0); 474 if (!q) 475 goto out; 476 477 RCU_INIT_POINTER(q->sock.wq, &q->wq); 478 init_waitqueue_head(&q->wq.wait); 479 q->sock.type = SOCK_RAW; 480 q->sock.state = SS_CONNECTED; 481 q->sock.file = file; 482 q->sock.ops = &macvtap_socket_ops; 483 sock_init_data(&q->sock, &q->sk); 484 q->sk.sk_write_space = macvtap_sock_write_space; 485 q->sk.sk_destruct = macvtap_sock_destruct; 486 q->flags = IFF_VNET_HDR | IFF_NO_PI | IFF_TAP; 487 q->vnet_hdr_sz = sizeof(struct virtio_net_hdr); 488 489 /* 490 * so far only KVM virtio_net uses macvtap, enable zero copy between 491 * guest kernel and host kernel when lower device supports zerocopy 492 * 493 * The macvlan supports zerocopy iff the lower device supports zero 494 * copy so we don't have to look at the lower device directly. 495 */ 496 if ((dev->features & NETIF_F_HIGHDMA) && (dev->features & NETIF_F_SG)) 497 sock_set_flag(&q->sk, SOCK_ZEROCOPY); 498 499 err = macvtap_set_queue(dev, file, q); 500 if (err) 501 sock_put(&q->sk); 502 503 out: 504 if (dev) 505 dev_put(dev); 506 507 rtnl_unlock(); 508 return err; 509 } 510 511 static int macvtap_release(struct inode *inode, struct file *file) 512 { 513 struct macvtap_queue *q = file->private_data; 514 macvtap_put_queue(q); 515 return 0; 516 } 517 518 static unsigned int macvtap_poll(struct file *file, poll_table * wait) 519 { 520 struct macvtap_queue *q = file->private_data; 521 unsigned int mask = POLLERR; 522 523 if (!q) 524 goto out; 525 526 mask = 0; 527 poll_wait(file, &q->wq.wait, wait); 528 529 if (!skb_queue_empty(&q->sk.sk_receive_queue)) 530 mask |= POLLIN | POLLRDNORM; 531 532 if (sock_writeable(&q->sk) || 533 (!test_and_set_bit(SOCK_ASYNC_NOSPACE, &q->sock.flags) && 534 sock_writeable(&q->sk))) 535 mask |= POLLOUT | POLLWRNORM; 536 537 out: 538 return mask; 539 } 540 541 static inline struct sk_buff *macvtap_alloc_skb(struct sock *sk, size_t prepad, 542 size_t len, size_t linear, 543 int noblock, int *err) 544 { 545 struct sk_buff *skb; 546 547 /* Under a page? Don't bother with paged skb. */ 548 if (prepad + len < PAGE_SIZE || !linear) 549 linear = len; 550 551 skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock, 552 err, 0); 553 if (!skb) 554 return NULL; 555 556 skb_reserve(skb, prepad); 557 skb_put(skb, linear); 558 skb->data_len = len - linear; 559 skb->len += len - linear; 560 561 return skb; 562 } 563 564 /* 565 * macvtap_skb_from_vnet_hdr and macvtap_skb_to_vnet_hdr should 566 * be shared with the tun/tap driver. 567 */ 568 static int macvtap_skb_from_vnet_hdr(struct macvtap_queue *q, 569 struct sk_buff *skb, 570 struct virtio_net_hdr *vnet_hdr) 571 { 572 unsigned short gso_type = 0; 573 if (vnet_hdr->gso_type != VIRTIO_NET_HDR_GSO_NONE) { 574 switch (vnet_hdr->gso_type & ~VIRTIO_NET_HDR_GSO_ECN) { 575 case VIRTIO_NET_HDR_GSO_TCPV4: 576 gso_type = SKB_GSO_TCPV4; 577 break; 578 case VIRTIO_NET_HDR_GSO_TCPV6: 579 gso_type = SKB_GSO_TCPV6; 580 break; 581 case VIRTIO_NET_HDR_GSO_UDP: 582 gso_type = SKB_GSO_UDP; 583 break; 584 default: 585 return -EINVAL; 586 } 587 588 if (vnet_hdr->gso_type & VIRTIO_NET_HDR_GSO_ECN) 589 gso_type |= SKB_GSO_TCP_ECN; 590 591 if (vnet_hdr->gso_size == 0) 592 return -EINVAL; 593 } 594 595 if (vnet_hdr->flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) { 596 if (!skb_partial_csum_set(skb, macvtap16_to_cpu(q, vnet_hdr->csum_start), 597 macvtap16_to_cpu(q, vnet_hdr->csum_offset))) 598 return -EINVAL; 599 } 600 601 if (vnet_hdr->gso_type != VIRTIO_NET_HDR_GSO_NONE) { 602 skb_shinfo(skb)->gso_size = macvtap16_to_cpu(q, vnet_hdr->gso_size); 603 skb_shinfo(skb)->gso_type = gso_type; 604 605 /* Header must be checked, and gso_segs computed. */ 606 skb_shinfo(skb)->gso_type |= SKB_GSO_DODGY; 607 skb_shinfo(skb)->gso_segs = 0; 608 } 609 return 0; 610 } 611 612 static void macvtap_skb_to_vnet_hdr(struct macvtap_queue *q, 613 const struct sk_buff *skb, 614 struct virtio_net_hdr *vnet_hdr) 615 { 616 memset(vnet_hdr, 0, sizeof(*vnet_hdr)); 617 618 if (skb_is_gso(skb)) { 619 struct skb_shared_info *sinfo = skb_shinfo(skb); 620 621 /* This is a hint as to how much should be linear. */ 622 vnet_hdr->hdr_len = cpu_to_macvtap16(q, skb_headlen(skb)); 623 vnet_hdr->gso_size = cpu_to_macvtap16(q, sinfo->gso_size); 624 if (sinfo->gso_type & SKB_GSO_TCPV4) 625 vnet_hdr->gso_type = VIRTIO_NET_HDR_GSO_TCPV4; 626 else if (sinfo->gso_type & SKB_GSO_TCPV6) 627 vnet_hdr->gso_type = VIRTIO_NET_HDR_GSO_TCPV6; 628 else if (sinfo->gso_type & SKB_GSO_UDP) 629 vnet_hdr->gso_type = VIRTIO_NET_HDR_GSO_UDP; 630 else 631 BUG(); 632 if (sinfo->gso_type & SKB_GSO_TCP_ECN) 633 vnet_hdr->gso_type |= VIRTIO_NET_HDR_GSO_ECN; 634 } else 635 vnet_hdr->gso_type = VIRTIO_NET_HDR_GSO_NONE; 636 637 if (skb->ip_summed == CHECKSUM_PARTIAL) { 638 vnet_hdr->flags = VIRTIO_NET_HDR_F_NEEDS_CSUM; 639 if (skb_vlan_tag_present(skb)) 640 vnet_hdr->csum_start = cpu_to_macvtap16(q, 641 skb_checksum_start_offset(skb) + VLAN_HLEN); 642 else 643 vnet_hdr->csum_start = cpu_to_macvtap16(q, 644 skb_checksum_start_offset(skb)); 645 vnet_hdr->csum_offset = cpu_to_macvtap16(q, skb->csum_offset); 646 } else if (skb->ip_summed == CHECKSUM_UNNECESSARY) { 647 vnet_hdr->flags = VIRTIO_NET_HDR_F_DATA_VALID; 648 } /* else everything is zero */ 649 } 650 651 /* Neighbour code has some assumptions on HH_DATA_MOD alignment */ 652 #define MACVTAP_RESERVE HH_DATA_OFF(ETH_HLEN) 653 654 /* Get packet from user space buffer */ 655 static ssize_t macvtap_get_user(struct macvtap_queue *q, struct msghdr *m, 656 struct iov_iter *from, int noblock) 657 { 658 int good_linear = SKB_MAX_HEAD(MACVTAP_RESERVE); 659 struct sk_buff *skb; 660 struct macvlan_dev *vlan; 661 unsigned long total_len = iov_iter_count(from); 662 unsigned long len = total_len; 663 int err; 664 struct virtio_net_hdr vnet_hdr = { 0 }; 665 int vnet_hdr_len = 0; 666 int copylen = 0; 667 bool zerocopy = false; 668 size_t linear; 669 ssize_t n; 670 671 if (q->flags & IFF_VNET_HDR) { 672 vnet_hdr_len = q->vnet_hdr_sz; 673 674 err = -EINVAL; 675 if (len < vnet_hdr_len) 676 goto err; 677 len -= vnet_hdr_len; 678 679 err = -EFAULT; 680 n = copy_from_iter(&vnet_hdr, sizeof(vnet_hdr), from); 681 if (n != sizeof(vnet_hdr)) 682 goto err; 683 iov_iter_advance(from, vnet_hdr_len - sizeof(vnet_hdr)); 684 if ((vnet_hdr.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) && 685 macvtap16_to_cpu(q, vnet_hdr.csum_start) + 686 macvtap16_to_cpu(q, vnet_hdr.csum_offset) + 2 > 687 macvtap16_to_cpu(q, vnet_hdr.hdr_len)) 688 vnet_hdr.hdr_len = cpu_to_macvtap16(q, 689 macvtap16_to_cpu(q, vnet_hdr.csum_start) + 690 macvtap16_to_cpu(q, vnet_hdr.csum_offset) + 2); 691 err = -EINVAL; 692 if (macvtap16_to_cpu(q, vnet_hdr.hdr_len) > len) 693 goto err; 694 } 695 696 err = -EINVAL; 697 if (unlikely(len < ETH_HLEN)) 698 goto err; 699 700 if (m && m->msg_control && sock_flag(&q->sk, SOCK_ZEROCOPY)) { 701 struct iov_iter i; 702 703 copylen = vnet_hdr.hdr_len ? 704 macvtap16_to_cpu(q, vnet_hdr.hdr_len) : GOODCOPY_LEN; 705 if (copylen > good_linear) 706 copylen = good_linear; 707 linear = copylen; 708 i = *from; 709 iov_iter_advance(&i, copylen); 710 if (iov_iter_npages(&i, INT_MAX) <= MAX_SKB_FRAGS) 711 zerocopy = true; 712 } 713 714 if (!zerocopy) { 715 copylen = len; 716 if (macvtap16_to_cpu(q, vnet_hdr.hdr_len) > good_linear) 717 linear = good_linear; 718 else 719 linear = macvtap16_to_cpu(q, vnet_hdr.hdr_len); 720 } 721 722 skb = macvtap_alloc_skb(&q->sk, MACVTAP_RESERVE, copylen, 723 linear, noblock, &err); 724 if (!skb) 725 goto err; 726 727 if (zerocopy) 728 err = zerocopy_sg_from_iter(skb, from); 729 else { 730 err = skb_copy_datagram_from_iter(skb, 0, from, len); 731 if (!err && m && m->msg_control) { 732 struct ubuf_info *uarg = m->msg_control; 733 uarg->callback(uarg, false); 734 } 735 } 736 737 if (err) 738 goto err_kfree; 739 740 skb_set_network_header(skb, ETH_HLEN); 741 skb_reset_mac_header(skb); 742 skb->protocol = eth_hdr(skb)->h_proto; 743 744 if (vnet_hdr_len) { 745 err = macvtap_skb_from_vnet_hdr(q, skb, &vnet_hdr); 746 if (err) 747 goto err_kfree; 748 } 749 750 skb_probe_transport_header(skb, ETH_HLEN); 751 752 rcu_read_lock(); 753 vlan = rcu_dereference(q->vlan); 754 /* copy skb_ubuf_info for callback when skb has no error */ 755 if (zerocopy) { 756 skb_shinfo(skb)->destructor_arg = m->msg_control; 757 skb_shinfo(skb)->tx_flags |= SKBTX_DEV_ZEROCOPY; 758 skb_shinfo(skb)->tx_flags |= SKBTX_SHARED_FRAG; 759 } 760 if (vlan) { 761 skb->dev = vlan->dev; 762 dev_queue_xmit(skb); 763 } else { 764 kfree_skb(skb); 765 } 766 rcu_read_unlock(); 767 768 return total_len; 769 770 err_kfree: 771 kfree_skb(skb); 772 773 err: 774 rcu_read_lock(); 775 vlan = rcu_dereference(q->vlan); 776 if (vlan) 777 this_cpu_inc(vlan->pcpu_stats->tx_dropped); 778 rcu_read_unlock(); 779 780 return err; 781 } 782 783 static ssize_t macvtap_write_iter(struct kiocb *iocb, struct iov_iter *from) 784 { 785 struct file *file = iocb->ki_filp; 786 struct macvtap_queue *q = file->private_data; 787 788 return macvtap_get_user(q, NULL, from, file->f_flags & O_NONBLOCK); 789 } 790 791 /* Put packet to the user space buffer */ 792 static ssize_t macvtap_put_user(struct macvtap_queue *q, 793 const struct sk_buff *skb, 794 struct iov_iter *iter) 795 { 796 int ret; 797 int vnet_hdr_len = 0; 798 int vlan_offset = 0; 799 int total; 800 801 if (q->flags & IFF_VNET_HDR) { 802 struct virtio_net_hdr vnet_hdr; 803 vnet_hdr_len = q->vnet_hdr_sz; 804 if (iov_iter_count(iter) < vnet_hdr_len) 805 return -EINVAL; 806 807 macvtap_skb_to_vnet_hdr(q, skb, &vnet_hdr); 808 809 if (copy_to_iter(&vnet_hdr, sizeof(vnet_hdr), iter) != 810 sizeof(vnet_hdr)) 811 return -EFAULT; 812 813 iov_iter_advance(iter, vnet_hdr_len - sizeof(vnet_hdr)); 814 } 815 total = vnet_hdr_len; 816 total += skb->len; 817 818 if (skb_vlan_tag_present(skb)) { 819 struct { 820 __be16 h_vlan_proto; 821 __be16 h_vlan_TCI; 822 } veth; 823 veth.h_vlan_proto = skb->vlan_proto; 824 veth.h_vlan_TCI = htons(skb_vlan_tag_get(skb)); 825 826 vlan_offset = offsetof(struct vlan_ethhdr, h_vlan_proto); 827 total += VLAN_HLEN; 828 829 ret = skb_copy_datagram_iter(skb, 0, iter, vlan_offset); 830 if (ret || !iov_iter_count(iter)) 831 goto done; 832 833 ret = copy_to_iter(&veth, sizeof(veth), iter); 834 if (ret != sizeof(veth) || !iov_iter_count(iter)) 835 goto done; 836 } 837 838 ret = skb_copy_datagram_iter(skb, vlan_offset, iter, 839 skb->len - vlan_offset); 840 841 done: 842 return ret ? ret : total; 843 } 844 845 static ssize_t macvtap_do_read(struct macvtap_queue *q, 846 struct iov_iter *to, 847 int noblock) 848 { 849 DEFINE_WAIT(wait); 850 struct sk_buff *skb; 851 ssize_t ret = 0; 852 853 if (!iov_iter_count(to)) 854 return 0; 855 856 while (1) { 857 if (!noblock) 858 prepare_to_wait(sk_sleep(&q->sk), &wait, 859 TASK_INTERRUPTIBLE); 860 861 /* Read frames from the queue */ 862 skb = skb_dequeue(&q->sk.sk_receive_queue); 863 if (skb) 864 break; 865 if (noblock) { 866 ret = -EAGAIN; 867 break; 868 } 869 if (signal_pending(current)) { 870 ret = -ERESTARTSYS; 871 break; 872 } 873 /* Nothing to read, let's sleep */ 874 schedule(); 875 } 876 if (skb) { 877 ret = macvtap_put_user(q, skb, to); 878 if (unlikely(ret < 0)) 879 kfree_skb(skb); 880 else 881 consume_skb(skb); 882 } 883 if (!noblock) 884 finish_wait(sk_sleep(&q->sk), &wait); 885 return ret; 886 } 887 888 static ssize_t macvtap_read_iter(struct kiocb *iocb, struct iov_iter *to) 889 { 890 struct file *file = iocb->ki_filp; 891 struct macvtap_queue *q = file->private_data; 892 ssize_t len = iov_iter_count(to), ret; 893 894 ret = macvtap_do_read(q, to, file->f_flags & O_NONBLOCK); 895 ret = min_t(ssize_t, ret, len); 896 if (ret > 0) 897 iocb->ki_pos = ret; 898 return ret; 899 } 900 901 static struct macvlan_dev *macvtap_get_vlan(struct macvtap_queue *q) 902 { 903 struct macvlan_dev *vlan; 904 905 ASSERT_RTNL(); 906 vlan = rtnl_dereference(q->vlan); 907 if (vlan) 908 dev_hold(vlan->dev); 909 910 return vlan; 911 } 912 913 static void macvtap_put_vlan(struct macvlan_dev *vlan) 914 { 915 dev_put(vlan->dev); 916 } 917 918 static int macvtap_ioctl_set_queue(struct file *file, unsigned int flags) 919 { 920 struct macvtap_queue *q = file->private_data; 921 struct macvlan_dev *vlan; 922 int ret; 923 924 vlan = macvtap_get_vlan(q); 925 if (!vlan) 926 return -EINVAL; 927 928 if (flags & IFF_ATTACH_QUEUE) 929 ret = macvtap_enable_queue(vlan->dev, file, q); 930 else if (flags & IFF_DETACH_QUEUE) 931 ret = macvtap_disable_queue(q); 932 else 933 ret = -EINVAL; 934 935 macvtap_put_vlan(vlan); 936 return ret; 937 } 938 939 static int set_offload(struct macvtap_queue *q, unsigned long arg) 940 { 941 struct macvlan_dev *vlan; 942 netdev_features_t features; 943 netdev_features_t feature_mask = 0; 944 945 vlan = rtnl_dereference(q->vlan); 946 if (!vlan) 947 return -ENOLINK; 948 949 features = vlan->dev->features; 950 951 if (arg & TUN_F_CSUM) { 952 feature_mask = NETIF_F_HW_CSUM; 953 954 if (arg & (TUN_F_TSO4 | TUN_F_TSO6)) { 955 if (arg & TUN_F_TSO_ECN) 956 feature_mask |= NETIF_F_TSO_ECN; 957 if (arg & TUN_F_TSO4) 958 feature_mask |= NETIF_F_TSO; 959 if (arg & TUN_F_TSO6) 960 feature_mask |= NETIF_F_TSO6; 961 } 962 963 if (arg & TUN_F_UFO) 964 feature_mask |= NETIF_F_UFO; 965 } 966 967 /* tun/tap driver inverts the usage for TSO offloads, where 968 * setting the TSO bit means that the userspace wants to 969 * accept TSO frames and turning it off means that user space 970 * does not support TSO. 971 * For macvtap, we have to invert it to mean the same thing. 972 * When user space turns off TSO, we turn off GSO/LRO so that 973 * user-space will not receive TSO frames. 974 */ 975 if (feature_mask & (NETIF_F_TSO | NETIF_F_TSO6 | NETIF_F_UFO)) 976 features |= RX_OFFLOADS; 977 else 978 features &= ~RX_OFFLOADS; 979 980 /* tap_features are the same as features on tun/tap and 981 * reflect user expectations. 982 */ 983 vlan->tap_features = feature_mask; 984 vlan->set_features = features; 985 netdev_update_features(vlan->dev); 986 987 return 0; 988 } 989 990 /* 991 * provide compatibility with generic tun/tap interface 992 */ 993 static long macvtap_ioctl(struct file *file, unsigned int cmd, 994 unsigned long arg) 995 { 996 struct macvtap_queue *q = file->private_data; 997 struct macvlan_dev *vlan; 998 void __user *argp = (void __user *)arg; 999 struct ifreq __user *ifr = argp; 1000 unsigned int __user *up = argp; 1001 unsigned short u; 1002 int __user *sp = argp; 1003 struct sockaddr sa; 1004 int s; 1005 int ret; 1006 1007 switch (cmd) { 1008 case TUNSETIFF: 1009 /* ignore the name, just look at flags */ 1010 if (get_user(u, &ifr->ifr_flags)) 1011 return -EFAULT; 1012 1013 ret = 0; 1014 if ((u & ~MACVTAP_FEATURES) != (IFF_NO_PI | IFF_TAP)) 1015 ret = -EINVAL; 1016 else 1017 q->flags = (q->flags & ~MACVTAP_FEATURES) | u; 1018 1019 return ret; 1020 1021 case TUNGETIFF: 1022 rtnl_lock(); 1023 vlan = macvtap_get_vlan(q); 1024 if (!vlan) { 1025 rtnl_unlock(); 1026 return -ENOLINK; 1027 } 1028 1029 ret = 0; 1030 u = q->flags; 1031 if (copy_to_user(&ifr->ifr_name, vlan->dev->name, IFNAMSIZ) || 1032 put_user(u, &ifr->ifr_flags)) 1033 ret = -EFAULT; 1034 macvtap_put_vlan(vlan); 1035 rtnl_unlock(); 1036 return ret; 1037 1038 case TUNSETQUEUE: 1039 if (get_user(u, &ifr->ifr_flags)) 1040 return -EFAULT; 1041 rtnl_lock(); 1042 ret = macvtap_ioctl_set_queue(file, u); 1043 rtnl_unlock(); 1044 return ret; 1045 1046 case TUNGETFEATURES: 1047 if (put_user(IFF_TAP | IFF_NO_PI | MACVTAP_FEATURES, up)) 1048 return -EFAULT; 1049 return 0; 1050 1051 case TUNSETSNDBUF: 1052 if (get_user(u, up)) 1053 return -EFAULT; 1054 1055 q->sk.sk_sndbuf = u; 1056 return 0; 1057 1058 case TUNGETVNETHDRSZ: 1059 s = q->vnet_hdr_sz; 1060 if (put_user(s, sp)) 1061 return -EFAULT; 1062 return 0; 1063 1064 case TUNSETVNETHDRSZ: 1065 if (get_user(s, sp)) 1066 return -EFAULT; 1067 if (s < (int)sizeof(struct virtio_net_hdr)) 1068 return -EINVAL; 1069 1070 q->vnet_hdr_sz = s; 1071 return 0; 1072 1073 case TUNGETVNETLE: 1074 s = !!(q->flags & MACVTAP_VNET_LE); 1075 if (put_user(s, sp)) 1076 return -EFAULT; 1077 return 0; 1078 1079 case TUNSETVNETLE: 1080 if (get_user(s, sp)) 1081 return -EFAULT; 1082 if (s) 1083 q->flags |= MACVTAP_VNET_LE; 1084 else 1085 q->flags &= ~MACVTAP_VNET_LE; 1086 return 0; 1087 1088 case TUNSETOFFLOAD: 1089 /* let the user check for future flags */ 1090 if (arg & ~(TUN_F_CSUM | TUN_F_TSO4 | TUN_F_TSO6 | 1091 TUN_F_TSO_ECN | TUN_F_UFO)) 1092 return -EINVAL; 1093 1094 rtnl_lock(); 1095 ret = set_offload(q, arg); 1096 rtnl_unlock(); 1097 return ret; 1098 1099 case SIOCGIFHWADDR: 1100 rtnl_lock(); 1101 vlan = macvtap_get_vlan(q); 1102 if (!vlan) { 1103 rtnl_unlock(); 1104 return -ENOLINK; 1105 } 1106 ret = 0; 1107 u = vlan->dev->type; 1108 if (copy_to_user(&ifr->ifr_name, vlan->dev->name, IFNAMSIZ) || 1109 copy_to_user(&ifr->ifr_hwaddr.sa_data, vlan->dev->dev_addr, ETH_ALEN) || 1110 put_user(u, &ifr->ifr_hwaddr.sa_family)) 1111 ret = -EFAULT; 1112 macvtap_put_vlan(vlan); 1113 rtnl_unlock(); 1114 return ret; 1115 1116 case SIOCSIFHWADDR: 1117 if (copy_from_user(&sa, &ifr->ifr_hwaddr, sizeof(sa))) 1118 return -EFAULT; 1119 rtnl_lock(); 1120 vlan = macvtap_get_vlan(q); 1121 if (!vlan) { 1122 rtnl_unlock(); 1123 return -ENOLINK; 1124 } 1125 ret = dev_set_mac_address(vlan->dev, &sa); 1126 macvtap_put_vlan(vlan); 1127 rtnl_unlock(); 1128 return ret; 1129 1130 default: 1131 return -EINVAL; 1132 } 1133 } 1134 1135 #ifdef CONFIG_COMPAT 1136 static long macvtap_compat_ioctl(struct file *file, unsigned int cmd, 1137 unsigned long arg) 1138 { 1139 return macvtap_ioctl(file, cmd, (unsigned long)compat_ptr(arg)); 1140 } 1141 #endif 1142 1143 static const struct file_operations macvtap_fops = { 1144 .owner = THIS_MODULE, 1145 .open = macvtap_open, 1146 .release = macvtap_release, 1147 .read_iter = macvtap_read_iter, 1148 .write_iter = macvtap_write_iter, 1149 .poll = macvtap_poll, 1150 .llseek = no_llseek, 1151 .unlocked_ioctl = macvtap_ioctl, 1152 #ifdef CONFIG_COMPAT 1153 .compat_ioctl = macvtap_compat_ioctl, 1154 #endif 1155 }; 1156 1157 static int macvtap_sendmsg(struct socket *sock, struct msghdr *m, 1158 size_t total_len) 1159 { 1160 struct macvtap_queue *q = container_of(sock, struct macvtap_queue, sock); 1161 return macvtap_get_user(q, m, &m->msg_iter, m->msg_flags & MSG_DONTWAIT); 1162 } 1163 1164 static int macvtap_recvmsg(struct socket *sock, struct msghdr *m, 1165 size_t total_len, int flags) 1166 { 1167 struct macvtap_queue *q = container_of(sock, struct macvtap_queue, sock); 1168 int ret; 1169 if (flags & ~(MSG_DONTWAIT|MSG_TRUNC)) 1170 return -EINVAL; 1171 ret = macvtap_do_read(q, &m->msg_iter, flags & MSG_DONTWAIT); 1172 if (ret > total_len) { 1173 m->msg_flags |= MSG_TRUNC; 1174 ret = flags & MSG_TRUNC ? ret : total_len; 1175 } 1176 return ret; 1177 } 1178 1179 /* Ops structure to mimic raw sockets with tun */ 1180 static const struct proto_ops macvtap_socket_ops = { 1181 .sendmsg = macvtap_sendmsg, 1182 .recvmsg = macvtap_recvmsg, 1183 }; 1184 1185 /* Get an underlying socket object from tun file. Returns error unless file is 1186 * attached to a device. The returned object works like a packet socket, it 1187 * can be used for sock_sendmsg/sock_recvmsg. The caller is responsible for 1188 * holding a reference to the file for as long as the socket is in use. */ 1189 struct socket *macvtap_get_socket(struct file *file) 1190 { 1191 struct macvtap_queue *q; 1192 if (file->f_op != &macvtap_fops) 1193 return ERR_PTR(-EINVAL); 1194 q = file->private_data; 1195 if (!q) 1196 return ERR_PTR(-EBADFD); 1197 return &q->sock; 1198 } 1199 EXPORT_SYMBOL_GPL(macvtap_get_socket); 1200 1201 static int macvtap_device_event(struct notifier_block *unused, 1202 unsigned long event, void *ptr) 1203 { 1204 struct net_device *dev = netdev_notifier_info_to_dev(ptr); 1205 struct macvlan_dev *vlan; 1206 struct device *classdev; 1207 dev_t devt; 1208 int err; 1209 1210 if (dev->rtnl_link_ops != &macvtap_link_ops) 1211 return NOTIFY_DONE; 1212 1213 vlan = netdev_priv(dev); 1214 1215 switch (event) { 1216 case NETDEV_REGISTER: 1217 /* Create the device node here after the network device has 1218 * been registered but before register_netdevice has 1219 * finished running. 1220 */ 1221 err = macvtap_get_minor(vlan); 1222 if (err) 1223 return notifier_from_errno(err); 1224 1225 devt = MKDEV(MAJOR(macvtap_major), vlan->minor); 1226 classdev = device_create(macvtap_class, &dev->dev, devt, 1227 dev, "tap%d", dev->ifindex); 1228 if (IS_ERR(classdev)) { 1229 macvtap_free_minor(vlan); 1230 return notifier_from_errno(PTR_ERR(classdev)); 1231 } 1232 break; 1233 case NETDEV_UNREGISTER: 1234 devt = MKDEV(MAJOR(macvtap_major), vlan->minor); 1235 device_destroy(macvtap_class, devt); 1236 macvtap_free_minor(vlan); 1237 break; 1238 } 1239 1240 return NOTIFY_DONE; 1241 } 1242 1243 static struct notifier_block macvtap_notifier_block __read_mostly = { 1244 .notifier_call = macvtap_device_event, 1245 }; 1246 1247 static int macvtap_init(void) 1248 { 1249 int err; 1250 1251 err = alloc_chrdev_region(&macvtap_major, 0, 1252 MACVTAP_NUM_DEVS, "macvtap"); 1253 if (err) 1254 goto out1; 1255 1256 cdev_init(&macvtap_cdev, &macvtap_fops); 1257 err = cdev_add(&macvtap_cdev, macvtap_major, MACVTAP_NUM_DEVS); 1258 if (err) 1259 goto out2; 1260 1261 macvtap_class = class_create(THIS_MODULE, "macvtap"); 1262 if (IS_ERR(macvtap_class)) { 1263 err = PTR_ERR(macvtap_class); 1264 goto out3; 1265 } 1266 1267 err = register_netdevice_notifier(&macvtap_notifier_block); 1268 if (err) 1269 goto out4; 1270 1271 err = macvlan_link_register(&macvtap_link_ops); 1272 if (err) 1273 goto out5; 1274 1275 return 0; 1276 1277 out5: 1278 unregister_netdevice_notifier(&macvtap_notifier_block); 1279 out4: 1280 class_unregister(macvtap_class); 1281 out3: 1282 cdev_del(&macvtap_cdev); 1283 out2: 1284 unregister_chrdev_region(macvtap_major, MACVTAP_NUM_DEVS); 1285 out1: 1286 return err; 1287 } 1288 module_init(macvtap_init); 1289 1290 static void macvtap_exit(void) 1291 { 1292 rtnl_link_unregister(&macvtap_link_ops); 1293 unregister_netdevice_notifier(&macvtap_notifier_block); 1294 class_unregister(macvtap_class); 1295 cdev_del(&macvtap_cdev); 1296 unregister_chrdev_region(macvtap_major, MACVTAP_NUM_DEVS); 1297 } 1298 module_exit(macvtap_exit); 1299 1300 MODULE_ALIAS_RTNL_LINK("macvtap"); 1301 MODULE_AUTHOR("Arnd Bergmann <arnd@arndb.de>"); 1302 MODULE_LICENSE("GPL"); 1303