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/ipv6.h> 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 static struct proto macvtap_proto = { 49 .name = "macvtap", 50 .owner = THIS_MODULE, 51 .obj_size = sizeof (struct macvtap_queue), 52 }; 53 54 /* 55 * Variables for dealing with macvtaps device numbers. 56 */ 57 static dev_t macvtap_major; 58 #define MACVTAP_NUM_DEVS (1U << MINORBITS) 59 static DEFINE_MUTEX(minor_lock); 60 static DEFINE_IDR(minor_idr); 61 62 #define GOODCOPY_LEN 128 63 static struct class *macvtap_class; 64 static struct cdev macvtap_cdev; 65 66 static const struct proto_ops macvtap_socket_ops; 67 68 #define TUN_OFFLOADS (NETIF_F_HW_CSUM | NETIF_F_TSO_ECN | NETIF_F_TSO | \ 69 NETIF_F_TSO6) 70 #define RX_OFFLOADS (NETIF_F_GRO | NETIF_F_LRO) 71 #define TAP_FEATURES (NETIF_F_GSO | NETIF_F_SG) 72 73 static struct macvlan_dev *macvtap_get_vlan_rcu(const struct net_device *dev) 74 { 75 return rcu_dereference(dev->rx_handler_data); 76 } 77 78 /* 79 * RCU usage: 80 * The macvtap_queue and the macvlan_dev are loosely coupled, the 81 * pointers from one to the other can only be read while rcu_read_lock 82 * or rtnl is held. 83 * 84 * Both the file and the macvlan_dev hold a reference on the macvtap_queue 85 * through sock_hold(&q->sk). When the macvlan_dev goes away first, 86 * q->vlan becomes inaccessible. When the files gets closed, 87 * macvtap_get_queue() fails. 88 * 89 * There may still be references to the struct sock inside of the 90 * queue from outbound SKBs, but these never reference back to the 91 * file or the dev. The data structure is freed through __sk_free 92 * when both our references and any pending SKBs are gone. 93 */ 94 95 static int macvtap_enable_queue(struct net_device *dev, struct file *file, 96 struct macvtap_queue *q) 97 { 98 struct macvlan_dev *vlan = netdev_priv(dev); 99 int err = -EINVAL; 100 101 ASSERT_RTNL(); 102 103 if (q->enabled) 104 goto out; 105 106 err = 0; 107 rcu_assign_pointer(vlan->taps[vlan->numvtaps], q); 108 q->queue_index = vlan->numvtaps; 109 q->enabled = true; 110 111 vlan->numvtaps++; 112 out: 113 return err; 114 } 115 116 /* Requires RTNL */ 117 static int macvtap_set_queue(struct net_device *dev, struct file *file, 118 struct macvtap_queue *q) 119 { 120 struct macvlan_dev *vlan = netdev_priv(dev); 121 122 if (vlan->numqueues == MAX_MACVTAP_QUEUES) 123 return -EBUSY; 124 125 rcu_assign_pointer(q->vlan, vlan); 126 rcu_assign_pointer(vlan->taps[vlan->numvtaps], q); 127 sock_hold(&q->sk); 128 129 q->file = file; 130 q->queue_index = vlan->numvtaps; 131 q->enabled = true; 132 file->private_data = q; 133 list_add_tail(&q->next, &vlan->queue_list); 134 135 vlan->numvtaps++; 136 vlan->numqueues++; 137 138 return 0; 139 } 140 141 static int macvtap_disable_queue(struct macvtap_queue *q) 142 { 143 struct macvlan_dev *vlan; 144 struct macvtap_queue *nq; 145 146 ASSERT_RTNL(); 147 if (!q->enabled) 148 return -EINVAL; 149 150 vlan = rtnl_dereference(q->vlan); 151 152 if (vlan) { 153 int index = q->queue_index; 154 BUG_ON(index >= vlan->numvtaps); 155 nq = rtnl_dereference(vlan->taps[vlan->numvtaps - 1]); 156 nq->queue_index = index; 157 158 rcu_assign_pointer(vlan->taps[index], nq); 159 RCU_INIT_POINTER(vlan->taps[vlan->numvtaps - 1], NULL); 160 q->enabled = false; 161 162 vlan->numvtaps--; 163 } 164 165 return 0; 166 } 167 168 /* 169 * The file owning the queue got closed, give up both 170 * the reference that the files holds as well as the 171 * one from the macvlan_dev if that still exists. 172 * 173 * Using the spinlock makes sure that we don't get 174 * to the queue again after destroying it. 175 */ 176 static void macvtap_put_queue(struct macvtap_queue *q) 177 { 178 struct macvlan_dev *vlan; 179 180 rtnl_lock(); 181 vlan = rtnl_dereference(q->vlan); 182 183 if (vlan) { 184 if (q->enabled) 185 BUG_ON(macvtap_disable_queue(q)); 186 187 vlan->numqueues--; 188 RCU_INIT_POINTER(q->vlan, NULL); 189 sock_put(&q->sk); 190 list_del_init(&q->next); 191 } 192 193 rtnl_unlock(); 194 195 synchronize_rcu(); 196 sock_put(&q->sk); 197 } 198 199 /* 200 * Select a queue based on the rxq of the device on which this packet 201 * arrived. If the incoming device is not mq, calculate a flow hash 202 * to select a queue. If all fails, find the first available queue. 203 * Cache vlan->numvtaps since it can become zero during the execution 204 * of this function. 205 */ 206 static struct macvtap_queue *macvtap_get_queue(struct net_device *dev, 207 struct sk_buff *skb) 208 { 209 struct macvlan_dev *vlan = netdev_priv(dev); 210 struct macvtap_queue *tap = NULL; 211 /* Access to taps array is protected by rcu, but access to numvtaps 212 * isn't. Below we use it to lookup a queue, but treat it as a hint 213 * and validate that the result isn't NULL - in case we are 214 * racing against queue removal. 215 */ 216 int numvtaps = ACCESS_ONCE(vlan->numvtaps); 217 __u32 rxq; 218 219 if (!numvtaps) 220 goto out; 221 222 /* Check if we can use flow to select a queue */ 223 rxq = skb_get_hash(skb); 224 if (rxq) { 225 tap = rcu_dereference(vlan->taps[rxq % numvtaps]); 226 goto out; 227 } 228 229 if (likely(skb_rx_queue_recorded(skb))) { 230 rxq = skb_get_rx_queue(skb); 231 232 while (unlikely(rxq >= numvtaps)) 233 rxq -= numvtaps; 234 235 tap = rcu_dereference(vlan->taps[rxq]); 236 goto out; 237 } 238 239 tap = rcu_dereference(vlan->taps[0]); 240 out: 241 return tap; 242 } 243 244 /* 245 * The net_device is going away, give up the reference 246 * that it holds on all queues and safely set the pointer 247 * from the queues to NULL. 248 */ 249 static void macvtap_del_queues(struct net_device *dev) 250 { 251 struct macvlan_dev *vlan = netdev_priv(dev); 252 struct macvtap_queue *q, *tmp, *qlist[MAX_MACVTAP_QUEUES]; 253 int i, j = 0; 254 255 ASSERT_RTNL(); 256 list_for_each_entry_safe(q, tmp, &vlan->queue_list, next) { 257 list_del_init(&q->next); 258 qlist[j++] = q; 259 RCU_INIT_POINTER(q->vlan, NULL); 260 if (q->enabled) 261 vlan->numvtaps--; 262 vlan->numqueues--; 263 } 264 for (i = 0; i < vlan->numvtaps; i++) 265 RCU_INIT_POINTER(vlan->taps[i], NULL); 266 BUG_ON(vlan->numvtaps); 267 BUG_ON(vlan->numqueues); 268 /* guarantee that any future macvtap_set_queue will fail */ 269 vlan->numvtaps = MAX_MACVTAP_QUEUES; 270 271 for (--j; j >= 0; j--) 272 sock_put(&qlist[j]->sk); 273 } 274 275 static rx_handler_result_t macvtap_handle_frame(struct sk_buff **pskb) 276 { 277 struct sk_buff *skb = *pskb; 278 struct net_device *dev = skb->dev; 279 struct macvlan_dev *vlan; 280 struct macvtap_queue *q; 281 netdev_features_t features = TAP_FEATURES; 282 283 vlan = macvtap_get_vlan_rcu(dev); 284 if (!vlan) 285 return RX_HANDLER_PASS; 286 287 q = macvtap_get_queue(dev, skb); 288 if (!q) 289 return RX_HANDLER_PASS; 290 291 if (skb_queue_len(&q->sk.sk_receive_queue) >= dev->tx_queue_len) 292 goto drop; 293 294 skb_push(skb, ETH_HLEN); 295 296 /* Apply the forward feature mask so that we perform segmentation 297 * according to users wishes. This only works if VNET_HDR is 298 * enabled. 299 */ 300 if (q->flags & IFF_VNET_HDR) 301 features |= vlan->tap_features; 302 if (netif_needs_gso(dev, skb, features)) { 303 struct sk_buff *segs = __skb_gso_segment(skb, features, false); 304 305 if (IS_ERR(segs)) 306 goto drop; 307 308 if (!segs) { 309 skb_queue_tail(&q->sk.sk_receive_queue, skb); 310 goto wake_up; 311 } 312 313 kfree_skb(skb); 314 while (segs) { 315 struct sk_buff *nskb = segs->next; 316 317 segs->next = NULL; 318 skb_queue_tail(&q->sk.sk_receive_queue, segs); 319 segs = nskb; 320 } 321 } else { 322 /* If we receive a partial checksum and the tap side 323 * doesn't support checksum offload, compute the checksum. 324 * Note: it doesn't matter which checksum feature to 325 * check, we either support them all or none. 326 */ 327 if (skb->ip_summed == CHECKSUM_PARTIAL && 328 !(features & NETIF_F_ALL_CSUM) && 329 skb_checksum_help(skb)) 330 goto drop; 331 skb_queue_tail(&q->sk.sk_receive_queue, skb); 332 } 333 334 wake_up: 335 wake_up_interruptible_poll(sk_sleep(&q->sk), POLLIN | POLLRDNORM | POLLRDBAND); 336 return RX_HANDLER_CONSUMED; 337 338 drop: 339 /* Count errors/drops only here, thus don't care about args. */ 340 macvlan_count_rx(vlan, 0, 0, 0); 341 kfree_skb(skb); 342 return RX_HANDLER_CONSUMED; 343 } 344 345 static int macvtap_get_minor(struct macvlan_dev *vlan) 346 { 347 int retval = -ENOMEM; 348 349 mutex_lock(&minor_lock); 350 retval = idr_alloc(&minor_idr, vlan, 1, MACVTAP_NUM_DEVS, GFP_KERNEL); 351 if (retval >= 0) { 352 vlan->minor = retval; 353 } else if (retval == -ENOSPC) { 354 printk(KERN_ERR "too many macvtap devices\n"); 355 retval = -EINVAL; 356 } 357 mutex_unlock(&minor_lock); 358 return retval < 0 ? retval : 0; 359 } 360 361 static void macvtap_free_minor(struct macvlan_dev *vlan) 362 { 363 mutex_lock(&minor_lock); 364 if (vlan->minor) { 365 idr_remove(&minor_idr, vlan->minor); 366 vlan->minor = 0; 367 } 368 mutex_unlock(&minor_lock); 369 } 370 371 static struct net_device *dev_get_by_macvtap_minor(int minor) 372 { 373 struct net_device *dev = NULL; 374 struct macvlan_dev *vlan; 375 376 mutex_lock(&minor_lock); 377 vlan = idr_find(&minor_idr, minor); 378 if (vlan) { 379 dev = vlan->dev; 380 dev_hold(dev); 381 } 382 mutex_unlock(&minor_lock); 383 return dev; 384 } 385 386 static int macvtap_newlink(struct net *src_net, 387 struct net_device *dev, 388 struct nlattr *tb[], 389 struct nlattr *data[]) 390 { 391 struct macvlan_dev *vlan = netdev_priv(dev); 392 int err; 393 394 INIT_LIST_HEAD(&vlan->queue_list); 395 396 /* Since macvlan supports all offloads by default, make 397 * tap support all offloads also. 398 */ 399 vlan->tap_features = TUN_OFFLOADS; 400 401 err = netdev_rx_handler_register(dev, macvtap_handle_frame, vlan); 402 if (err) 403 return err; 404 405 /* Don't put anything that may fail after macvlan_common_newlink 406 * because we can't undo what it does. 407 */ 408 return macvlan_common_newlink(src_net, dev, tb, data); 409 } 410 411 static void macvtap_dellink(struct net_device *dev, 412 struct list_head *head) 413 { 414 netdev_rx_handler_unregister(dev); 415 macvtap_del_queues(dev); 416 macvlan_dellink(dev, head); 417 } 418 419 static void macvtap_setup(struct net_device *dev) 420 { 421 macvlan_common_setup(dev); 422 dev->tx_queue_len = TUN_READQ_SIZE; 423 } 424 425 static struct rtnl_link_ops macvtap_link_ops __read_mostly = { 426 .kind = "macvtap", 427 .setup = macvtap_setup, 428 .newlink = macvtap_newlink, 429 .dellink = macvtap_dellink, 430 }; 431 432 433 static void macvtap_sock_write_space(struct sock *sk) 434 { 435 wait_queue_head_t *wqueue; 436 437 if (!sock_writeable(sk) || 438 !test_and_clear_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags)) 439 return; 440 441 wqueue = sk_sleep(sk); 442 if (wqueue && waitqueue_active(wqueue)) 443 wake_up_interruptible_poll(wqueue, POLLOUT | POLLWRNORM | POLLWRBAND); 444 } 445 446 static void macvtap_sock_destruct(struct sock *sk) 447 { 448 skb_queue_purge(&sk->sk_receive_queue); 449 } 450 451 static int macvtap_open(struct inode *inode, struct file *file) 452 { 453 struct net *net = current->nsproxy->net_ns; 454 struct net_device *dev; 455 struct macvtap_queue *q; 456 int err = -ENODEV; 457 458 rtnl_lock(); 459 dev = dev_get_by_macvtap_minor(iminor(inode)); 460 if (!dev) 461 goto out; 462 463 err = -ENOMEM; 464 q = (struct macvtap_queue *)sk_alloc(net, AF_UNSPEC, GFP_KERNEL, 465 &macvtap_proto); 466 if (!q) 467 goto out; 468 469 RCU_INIT_POINTER(q->sock.wq, &q->wq); 470 init_waitqueue_head(&q->wq.wait); 471 q->sock.type = SOCK_RAW; 472 q->sock.state = SS_CONNECTED; 473 q->sock.file = file; 474 q->sock.ops = &macvtap_socket_ops; 475 sock_init_data(&q->sock, &q->sk); 476 q->sk.sk_write_space = macvtap_sock_write_space; 477 q->sk.sk_destruct = macvtap_sock_destruct; 478 q->flags = IFF_VNET_HDR | IFF_NO_PI | IFF_TAP; 479 q->vnet_hdr_sz = sizeof(struct virtio_net_hdr); 480 481 /* 482 * so far only KVM virtio_net uses macvtap, enable zero copy between 483 * guest kernel and host kernel when lower device supports zerocopy 484 * 485 * The macvlan supports zerocopy iff the lower device supports zero 486 * copy so we don't have to look at the lower device directly. 487 */ 488 if ((dev->features & NETIF_F_HIGHDMA) && (dev->features & NETIF_F_SG)) 489 sock_set_flag(&q->sk, SOCK_ZEROCOPY); 490 491 err = macvtap_set_queue(dev, file, q); 492 if (err) 493 sock_put(&q->sk); 494 495 out: 496 if (dev) 497 dev_put(dev); 498 499 rtnl_unlock(); 500 return err; 501 } 502 503 static int macvtap_release(struct inode *inode, struct file *file) 504 { 505 struct macvtap_queue *q = file->private_data; 506 macvtap_put_queue(q); 507 return 0; 508 } 509 510 static unsigned int macvtap_poll(struct file *file, poll_table * wait) 511 { 512 struct macvtap_queue *q = file->private_data; 513 unsigned int mask = POLLERR; 514 515 if (!q) 516 goto out; 517 518 mask = 0; 519 poll_wait(file, &q->wq.wait, wait); 520 521 if (!skb_queue_empty(&q->sk.sk_receive_queue)) 522 mask |= POLLIN | POLLRDNORM; 523 524 if (sock_writeable(&q->sk) || 525 (!test_and_set_bit(SOCK_ASYNC_NOSPACE, &q->sock.flags) && 526 sock_writeable(&q->sk))) 527 mask |= POLLOUT | POLLWRNORM; 528 529 out: 530 return mask; 531 } 532 533 static inline struct sk_buff *macvtap_alloc_skb(struct sock *sk, size_t prepad, 534 size_t len, size_t linear, 535 int noblock, int *err) 536 { 537 struct sk_buff *skb; 538 539 /* Under a page? Don't bother with paged skb. */ 540 if (prepad + len < PAGE_SIZE || !linear) 541 linear = len; 542 543 skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock, 544 err, 0); 545 if (!skb) 546 return NULL; 547 548 skb_reserve(skb, prepad); 549 skb_put(skb, linear); 550 skb->data_len = len - linear; 551 skb->len += len - linear; 552 553 return skb; 554 } 555 556 /* 557 * macvtap_skb_from_vnet_hdr and macvtap_skb_to_vnet_hdr should 558 * be shared with the tun/tap driver. 559 */ 560 static int macvtap_skb_from_vnet_hdr(struct sk_buff *skb, 561 struct virtio_net_hdr *vnet_hdr) 562 { 563 unsigned short gso_type = 0; 564 if (vnet_hdr->gso_type != VIRTIO_NET_HDR_GSO_NONE) { 565 switch (vnet_hdr->gso_type & ~VIRTIO_NET_HDR_GSO_ECN) { 566 case VIRTIO_NET_HDR_GSO_TCPV4: 567 gso_type = SKB_GSO_TCPV4; 568 break; 569 case VIRTIO_NET_HDR_GSO_TCPV6: 570 gso_type = SKB_GSO_TCPV6; 571 break; 572 case VIRTIO_NET_HDR_GSO_UDP: 573 pr_warn_once("macvtap: %s: using disabled UFO feature; please fix this program\n", 574 current->comm); 575 gso_type = SKB_GSO_UDP; 576 if (skb->protocol == htons(ETH_P_IPV6)) 577 ipv6_proxy_select_ident(skb); 578 break; 579 default: 580 return -EINVAL; 581 } 582 583 if (vnet_hdr->gso_type & VIRTIO_NET_HDR_GSO_ECN) 584 gso_type |= SKB_GSO_TCP_ECN; 585 586 if (vnet_hdr->gso_size == 0) 587 return -EINVAL; 588 } 589 590 if (vnet_hdr->flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) { 591 if (!skb_partial_csum_set(skb, vnet_hdr->csum_start, 592 vnet_hdr->csum_offset)) 593 return -EINVAL; 594 } 595 596 if (vnet_hdr->gso_type != VIRTIO_NET_HDR_GSO_NONE) { 597 skb_shinfo(skb)->gso_size = vnet_hdr->gso_size; 598 skb_shinfo(skb)->gso_type = gso_type; 599 600 /* Header must be checked, and gso_segs computed. */ 601 skb_shinfo(skb)->gso_type |= SKB_GSO_DODGY; 602 skb_shinfo(skb)->gso_segs = 0; 603 } 604 return 0; 605 } 606 607 static void macvtap_skb_to_vnet_hdr(const struct sk_buff *skb, 608 struct virtio_net_hdr *vnet_hdr) 609 { 610 memset(vnet_hdr, 0, sizeof(*vnet_hdr)); 611 612 if (skb_is_gso(skb)) { 613 struct skb_shared_info *sinfo = skb_shinfo(skb); 614 615 /* This is a hint as to how much should be linear. */ 616 vnet_hdr->hdr_len = skb_headlen(skb); 617 vnet_hdr->gso_size = sinfo->gso_size; 618 if (sinfo->gso_type & SKB_GSO_TCPV4) 619 vnet_hdr->gso_type = VIRTIO_NET_HDR_GSO_TCPV4; 620 else if (sinfo->gso_type & SKB_GSO_TCPV6) 621 vnet_hdr->gso_type = VIRTIO_NET_HDR_GSO_TCPV6; 622 else 623 BUG(); 624 if (sinfo->gso_type & SKB_GSO_TCP_ECN) 625 vnet_hdr->gso_type |= VIRTIO_NET_HDR_GSO_ECN; 626 } else 627 vnet_hdr->gso_type = VIRTIO_NET_HDR_GSO_NONE; 628 629 if (skb->ip_summed == CHECKSUM_PARTIAL) { 630 vnet_hdr->flags = VIRTIO_NET_HDR_F_NEEDS_CSUM; 631 vnet_hdr->csum_start = skb_checksum_start_offset(skb); 632 vnet_hdr->csum_offset = skb->csum_offset; 633 } else if (skb->ip_summed == CHECKSUM_UNNECESSARY) { 634 vnet_hdr->flags = VIRTIO_NET_HDR_F_DATA_VALID; 635 } /* else everything is zero */ 636 } 637 638 /* Get packet from user space buffer */ 639 static ssize_t macvtap_get_user(struct macvtap_queue *q, struct msghdr *m, 640 const struct iovec *iv, unsigned long total_len, 641 size_t count, int noblock) 642 { 643 int good_linear = SKB_MAX_HEAD(NET_IP_ALIGN); 644 struct sk_buff *skb; 645 struct macvlan_dev *vlan; 646 unsigned long len = total_len; 647 int err; 648 struct virtio_net_hdr vnet_hdr = { 0 }; 649 int vnet_hdr_len = 0; 650 int copylen = 0; 651 bool zerocopy = false; 652 size_t linear; 653 654 if (q->flags & IFF_VNET_HDR) { 655 vnet_hdr_len = q->vnet_hdr_sz; 656 657 err = -EINVAL; 658 if (len < vnet_hdr_len) 659 goto err; 660 len -= vnet_hdr_len; 661 662 err = memcpy_fromiovecend((void *)&vnet_hdr, iv, 0, 663 sizeof(vnet_hdr)); 664 if (err < 0) 665 goto err; 666 if ((vnet_hdr.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) && 667 vnet_hdr.csum_start + vnet_hdr.csum_offset + 2 > 668 vnet_hdr.hdr_len) 669 vnet_hdr.hdr_len = vnet_hdr.csum_start + 670 vnet_hdr.csum_offset + 2; 671 err = -EINVAL; 672 if (vnet_hdr.hdr_len > len) 673 goto err; 674 } 675 676 err = -EINVAL; 677 if (unlikely(len < ETH_HLEN)) 678 goto err; 679 680 err = -EMSGSIZE; 681 if (unlikely(count > UIO_MAXIOV)) 682 goto err; 683 684 if (m && m->msg_control && sock_flag(&q->sk, SOCK_ZEROCOPY)) { 685 copylen = vnet_hdr.hdr_len ? vnet_hdr.hdr_len : GOODCOPY_LEN; 686 if (copylen > good_linear) 687 copylen = good_linear; 688 linear = copylen; 689 if (iov_pages(iv, vnet_hdr_len + copylen, count) 690 <= MAX_SKB_FRAGS) 691 zerocopy = true; 692 } 693 694 if (!zerocopy) { 695 copylen = len; 696 if (vnet_hdr.hdr_len > good_linear) 697 linear = good_linear; 698 else 699 linear = vnet_hdr.hdr_len; 700 } 701 702 skb = macvtap_alloc_skb(&q->sk, NET_IP_ALIGN, copylen, 703 linear, noblock, &err); 704 if (!skb) 705 goto err; 706 707 if (zerocopy) 708 err = zerocopy_sg_from_iovec(skb, iv, vnet_hdr_len, count); 709 else { 710 err = skb_copy_datagram_from_iovec(skb, 0, iv, vnet_hdr_len, 711 len); 712 if (!err && m && m->msg_control) { 713 struct ubuf_info *uarg = m->msg_control; 714 uarg->callback(uarg, false); 715 } 716 } 717 718 if (err) 719 goto err_kfree; 720 721 skb_set_network_header(skb, ETH_HLEN); 722 skb_reset_mac_header(skb); 723 skb->protocol = eth_hdr(skb)->h_proto; 724 725 if (vnet_hdr_len) { 726 err = macvtap_skb_from_vnet_hdr(skb, &vnet_hdr); 727 if (err) 728 goto err_kfree; 729 } 730 731 skb_probe_transport_header(skb, ETH_HLEN); 732 733 rcu_read_lock(); 734 vlan = rcu_dereference(q->vlan); 735 /* copy skb_ubuf_info for callback when skb has no error */ 736 if (zerocopy) { 737 skb_shinfo(skb)->destructor_arg = m->msg_control; 738 skb_shinfo(skb)->tx_flags |= SKBTX_DEV_ZEROCOPY; 739 skb_shinfo(skb)->tx_flags |= SKBTX_SHARED_FRAG; 740 } 741 if (vlan) { 742 skb->dev = vlan->dev; 743 dev_queue_xmit(skb); 744 } else { 745 kfree_skb(skb); 746 } 747 rcu_read_unlock(); 748 749 return total_len; 750 751 err_kfree: 752 kfree_skb(skb); 753 754 err: 755 rcu_read_lock(); 756 vlan = rcu_dereference(q->vlan); 757 if (vlan) 758 this_cpu_inc(vlan->pcpu_stats->tx_dropped); 759 rcu_read_unlock(); 760 761 return err; 762 } 763 764 static ssize_t macvtap_aio_write(struct kiocb *iocb, const struct iovec *iv, 765 unsigned long count, loff_t pos) 766 { 767 struct file *file = iocb->ki_filp; 768 ssize_t result = -ENOLINK; 769 struct macvtap_queue *q = file->private_data; 770 771 result = macvtap_get_user(q, NULL, iv, iov_length(iv, count), count, 772 file->f_flags & O_NONBLOCK); 773 return result; 774 } 775 776 /* Put packet to the user space buffer */ 777 static ssize_t macvtap_put_user(struct macvtap_queue *q, 778 const struct sk_buff *skb, 779 const struct iovec *iv, int len) 780 { 781 int ret; 782 int vnet_hdr_len = 0; 783 int vlan_offset = 0; 784 int copied, total; 785 786 if (q->flags & IFF_VNET_HDR) { 787 struct virtio_net_hdr vnet_hdr; 788 vnet_hdr_len = q->vnet_hdr_sz; 789 if ((len -= vnet_hdr_len) < 0) 790 return -EINVAL; 791 792 macvtap_skb_to_vnet_hdr(skb, &vnet_hdr); 793 794 if (memcpy_toiovecend(iv, (void *)&vnet_hdr, 0, sizeof(vnet_hdr))) 795 return -EFAULT; 796 } 797 total = copied = vnet_hdr_len; 798 total += skb->len; 799 800 if (!vlan_tx_tag_present(skb)) 801 len = min_t(int, skb->len, len); 802 else { 803 int copy; 804 struct { 805 __be16 h_vlan_proto; 806 __be16 h_vlan_TCI; 807 } veth; 808 veth.h_vlan_proto = skb->vlan_proto; 809 veth.h_vlan_TCI = htons(vlan_tx_tag_get(skb)); 810 811 vlan_offset = offsetof(struct vlan_ethhdr, h_vlan_proto); 812 len = min_t(int, skb->len + VLAN_HLEN, len); 813 total += VLAN_HLEN; 814 815 copy = min_t(int, vlan_offset, len); 816 ret = skb_copy_datagram_const_iovec(skb, 0, iv, copied, copy); 817 len -= copy; 818 copied += copy; 819 if (ret || !len) 820 goto done; 821 822 copy = min_t(int, sizeof(veth), len); 823 ret = memcpy_toiovecend(iv, (void *)&veth, copied, copy); 824 len -= copy; 825 copied += copy; 826 if (ret || !len) 827 goto done; 828 } 829 830 ret = skb_copy_datagram_const_iovec(skb, vlan_offset, iv, copied, len); 831 832 done: 833 return ret ? ret : total; 834 } 835 836 static ssize_t macvtap_do_read(struct macvtap_queue *q, 837 const struct iovec *iv, unsigned long len, 838 int noblock) 839 { 840 DEFINE_WAIT(wait); 841 struct sk_buff *skb; 842 ssize_t ret = 0; 843 844 while (len) { 845 if (!noblock) 846 prepare_to_wait(sk_sleep(&q->sk), &wait, 847 TASK_INTERRUPTIBLE); 848 849 /* Read frames from the queue */ 850 skb = skb_dequeue(&q->sk.sk_receive_queue); 851 if (!skb) { 852 if (noblock) { 853 ret = -EAGAIN; 854 break; 855 } 856 if (signal_pending(current)) { 857 ret = -ERESTARTSYS; 858 break; 859 } 860 /* Nothing to read, let's sleep */ 861 schedule(); 862 continue; 863 } 864 ret = macvtap_put_user(q, skb, iv, len); 865 kfree_skb(skb); 866 break; 867 } 868 869 if (!noblock) 870 finish_wait(sk_sleep(&q->sk), &wait); 871 return ret; 872 } 873 874 static ssize_t macvtap_aio_read(struct kiocb *iocb, const struct iovec *iv, 875 unsigned long count, loff_t pos) 876 { 877 struct file *file = iocb->ki_filp; 878 struct macvtap_queue *q = file->private_data; 879 ssize_t len, ret = 0; 880 881 len = iov_length(iv, count); 882 if (len < 0) { 883 ret = -EINVAL; 884 goto out; 885 } 886 887 ret = macvtap_do_read(q, iv, len, file->f_flags & O_NONBLOCK); 888 ret = min_t(ssize_t, ret, len); 889 if (ret > 0) 890 iocb->ki_pos = ret; 891 out: 892 return ret; 893 } 894 895 static struct macvlan_dev *macvtap_get_vlan(struct macvtap_queue *q) 896 { 897 struct macvlan_dev *vlan; 898 899 ASSERT_RTNL(); 900 vlan = rtnl_dereference(q->vlan); 901 if (vlan) 902 dev_hold(vlan->dev); 903 904 return vlan; 905 } 906 907 static void macvtap_put_vlan(struct macvlan_dev *vlan) 908 { 909 dev_put(vlan->dev); 910 } 911 912 static int macvtap_ioctl_set_queue(struct file *file, unsigned int flags) 913 { 914 struct macvtap_queue *q = file->private_data; 915 struct macvlan_dev *vlan; 916 int ret; 917 918 vlan = macvtap_get_vlan(q); 919 if (!vlan) 920 return -EINVAL; 921 922 if (flags & IFF_ATTACH_QUEUE) 923 ret = macvtap_enable_queue(vlan->dev, file, q); 924 else if (flags & IFF_DETACH_QUEUE) 925 ret = macvtap_disable_queue(q); 926 else 927 ret = -EINVAL; 928 929 macvtap_put_vlan(vlan); 930 return ret; 931 } 932 933 static int set_offload(struct macvtap_queue *q, unsigned long arg) 934 { 935 struct macvlan_dev *vlan; 936 netdev_features_t features; 937 netdev_features_t feature_mask = 0; 938 939 vlan = rtnl_dereference(q->vlan); 940 if (!vlan) 941 return -ENOLINK; 942 943 features = vlan->dev->features; 944 945 if (arg & TUN_F_CSUM) { 946 feature_mask = NETIF_F_HW_CSUM; 947 948 if (arg & (TUN_F_TSO4 | TUN_F_TSO6)) { 949 if (arg & TUN_F_TSO_ECN) 950 feature_mask |= NETIF_F_TSO_ECN; 951 if (arg & TUN_F_TSO4) 952 feature_mask |= NETIF_F_TSO; 953 if (arg & TUN_F_TSO6) 954 feature_mask |= NETIF_F_TSO6; 955 } 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)) 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)) 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