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