1 // SPDX-License-Identifier: GPL-2.0-only 2 #include <linux/etherdevice.h> 3 #include <linux/if_tap.h> 4 #include <linux/if_vlan.h> 5 #include <linux/interrupt.h> 6 #include <linux/nsproxy.h> 7 #include <linux/compat.h> 8 #include <linux/if_tun.h> 9 #include <linux/module.h> 10 #include <linux/skbuff.h> 11 #include <linux/cache.h> 12 #include <linux/sched/signal.h> 13 #include <linux/types.h> 14 #include <linux/slab.h> 15 #include <linux/wait.h> 16 #include <linux/cdev.h> 17 #include <linux/idr.h> 18 #include <linux/fs.h> 19 #include <linux/uio.h> 20 21 #include <net/gso.h> 22 #include <net/net_namespace.h> 23 #include <net/rtnetlink.h> 24 #include <net/sock.h> 25 #include <net/xdp.h> 26 #include <linux/virtio_net.h> 27 #include <linux/skb_array.h> 28 29 #define TAP_IFFEATURES (IFF_VNET_HDR | IFF_MULTI_QUEUE) 30 31 #define TAP_VNET_LE 0x80000000 32 #define TAP_VNET_BE 0x40000000 33 34 #ifdef CONFIG_TUN_VNET_CROSS_LE 35 static inline bool tap_legacy_is_little_endian(struct tap_queue *q) 36 { 37 return q->flags & TAP_VNET_BE ? false : 38 virtio_legacy_is_little_endian(); 39 } 40 41 static long tap_get_vnet_be(struct tap_queue *q, int __user *sp) 42 { 43 int s = !!(q->flags & TAP_VNET_BE); 44 45 if (put_user(s, sp)) 46 return -EFAULT; 47 48 return 0; 49 } 50 51 static long tap_set_vnet_be(struct tap_queue *q, int __user *sp) 52 { 53 int s; 54 55 if (get_user(s, sp)) 56 return -EFAULT; 57 58 if (s) 59 q->flags |= TAP_VNET_BE; 60 else 61 q->flags &= ~TAP_VNET_BE; 62 63 return 0; 64 } 65 #else 66 static inline bool tap_legacy_is_little_endian(struct tap_queue *q) 67 { 68 return virtio_legacy_is_little_endian(); 69 } 70 71 static long tap_get_vnet_be(struct tap_queue *q, int __user *argp) 72 { 73 return -EINVAL; 74 } 75 76 static long tap_set_vnet_be(struct tap_queue *q, int __user *argp) 77 { 78 return -EINVAL; 79 } 80 #endif /* CONFIG_TUN_VNET_CROSS_LE */ 81 82 static inline bool tap_is_little_endian(struct tap_queue *q) 83 { 84 return q->flags & TAP_VNET_LE || 85 tap_legacy_is_little_endian(q); 86 } 87 88 static inline u16 tap16_to_cpu(struct tap_queue *q, __virtio16 val) 89 { 90 return __virtio16_to_cpu(tap_is_little_endian(q), val); 91 } 92 93 static inline __virtio16 cpu_to_tap16(struct tap_queue *q, u16 val) 94 { 95 return __cpu_to_virtio16(tap_is_little_endian(q), val); 96 } 97 98 static struct proto tap_proto = { 99 .name = "tap", 100 .owner = THIS_MODULE, 101 .obj_size = sizeof(struct tap_queue), 102 }; 103 104 #define TAP_NUM_DEVS (1U << MINORBITS) 105 106 static LIST_HEAD(major_list); 107 108 struct major_info { 109 struct rcu_head rcu; 110 dev_t major; 111 struct idr minor_idr; 112 spinlock_t minor_lock; 113 const char *device_name; 114 struct list_head next; 115 }; 116 117 #define GOODCOPY_LEN 128 118 119 static const struct proto_ops tap_socket_ops; 120 121 #define RX_OFFLOADS (NETIF_F_GRO | NETIF_F_LRO) 122 #define TAP_FEATURES (NETIF_F_GSO | NETIF_F_SG | NETIF_F_FRAGLIST) 123 124 static struct tap_dev *tap_dev_get_rcu(const struct net_device *dev) 125 { 126 return rcu_dereference(dev->rx_handler_data); 127 } 128 129 /* 130 * RCU usage: 131 * The tap_queue and the macvlan_dev are loosely coupled, the 132 * pointers from one to the other can only be read while rcu_read_lock 133 * or rtnl is held. 134 * 135 * Both the file and the macvlan_dev hold a reference on the tap_queue 136 * through sock_hold(&q->sk). When the macvlan_dev goes away first, 137 * q->vlan becomes inaccessible. When the files gets closed, 138 * tap_get_queue() fails. 139 * 140 * There may still be references to the struct sock inside of the 141 * queue from outbound SKBs, but these never reference back to the 142 * file or the dev. The data structure is freed through __sk_free 143 * when both our references and any pending SKBs are gone. 144 */ 145 146 static int tap_enable_queue(struct tap_dev *tap, struct file *file, 147 struct tap_queue *q) 148 { 149 int err = -EINVAL; 150 151 ASSERT_RTNL(); 152 153 if (q->enabled) 154 goto out; 155 156 err = 0; 157 rcu_assign_pointer(tap->taps[tap->numvtaps], q); 158 q->queue_index = tap->numvtaps; 159 q->enabled = true; 160 161 tap->numvtaps++; 162 out: 163 return err; 164 } 165 166 /* Requires RTNL */ 167 static int tap_set_queue(struct tap_dev *tap, struct file *file, 168 struct tap_queue *q) 169 { 170 if (tap->numqueues == MAX_TAP_QUEUES) 171 return -EBUSY; 172 173 rcu_assign_pointer(q->tap, tap); 174 rcu_assign_pointer(tap->taps[tap->numvtaps], q); 175 sock_hold(&q->sk); 176 177 q->file = file; 178 q->queue_index = tap->numvtaps; 179 q->enabled = true; 180 file->private_data = q; 181 list_add_tail(&q->next, &tap->queue_list); 182 183 tap->numvtaps++; 184 tap->numqueues++; 185 186 return 0; 187 } 188 189 static int tap_disable_queue(struct tap_queue *q) 190 { 191 struct tap_dev *tap; 192 struct tap_queue *nq; 193 194 ASSERT_RTNL(); 195 if (!q->enabled) 196 return -EINVAL; 197 198 tap = rtnl_dereference(q->tap); 199 200 if (tap) { 201 int index = q->queue_index; 202 BUG_ON(index >= tap->numvtaps); 203 nq = rtnl_dereference(tap->taps[tap->numvtaps - 1]); 204 nq->queue_index = index; 205 206 rcu_assign_pointer(tap->taps[index], nq); 207 RCU_INIT_POINTER(tap->taps[tap->numvtaps - 1], NULL); 208 q->enabled = false; 209 210 tap->numvtaps--; 211 } 212 213 return 0; 214 } 215 216 /* 217 * The file owning the queue got closed, give up both 218 * the reference that the files holds as well as the 219 * one from the macvlan_dev if that still exists. 220 * 221 * Using the spinlock makes sure that we don't get 222 * to the queue again after destroying it. 223 */ 224 static void tap_put_queue(struct tap_queue *q) 225 { 226 struct tap_dev *tap; 227 228 rtnl_lock(); 229 tap = rtnl_dereference(q->tap); 230 231 if (tap) { 232 if (q->enabled) 233 BUG_ON(tap_disable_queue(q)); 234 235 tap->numqueues--; 236 RCU_INIT_POINTER(q->tap, NULL); 237 sock_put(&q->sk); 238 list_del_init(&q->next); 239 } 240 241 rtnl_unlock(); 242 243 synchronize_rcu(); 244 sock_put(&q->sk); 245 } 246 247 /* 248 * Select a queue based on the rxq of the device on which this packet 249 * arrived. If the incoming device is not mq, calculate a flow hash 250 * to select a queue. If all fails, find the first available queue. 251 * Cache vlan->numvtaps since it can become zero during the execution 252 * of this function. 253 */ 254 static struct tap_queue *tap_get_queue(struct tap_dev *tap, 255 struct sk_buff *skb) 256 { 257 struct tap_queue *queue = NULL; 258 /* Access to taps array is protected by rcu, but access to numvtaps 259 * isn't. Below we use it to lookup a queue, but treat it as a hint 260 * and validate that the result isn't NULL - in case we are 261 * racing against queue removal. 262 */ 263 int numvtaps = READ_ONCE(tap->numvtaps); 264 __u32 rxq; 265 266 if (!numvtaps) 267 goto out; 268 269 if (numvtaps == 1) 270 goto single; 271 272 /* Check if we can use flow to select a queue */ 273 rxq = skb_get_hash(skb); 274 if (rxq) { 275 queue = rcu_dereference(tap->taps[rxq % numvtaps]); 276 goto out; 277 } 278 279 if (likely(skb_rx_queue_recorded(skb))) { 280 rxq = skb_get_rx_queue(skb); 281 282 while (unlikely(rxq >= numvtaps)) 283 rxq -= numvtaps; 284 285 queue = rcu_dereference(tap->taps[rxq]); 286 goto out; 287 } 288 289 single: 290 queue = rcu_dereference(tap->taps[0]); 291 out: 292 return queue; 293 } 294 295 /* 296 * The net_device is going away, give up the reference 297 * that it holds on all queues and safely set the pointer 298 * from the queues to NULL. 299 */ 300 void tap_del_queues(struct tap_dev *tap) 301 { 302 struct tap_queue *q, *tmp; 303 304 ASSERT_RTNL(); 305 list_for_each_entry_safe(q, tmp, &tap->queue_list, next) { 306 list_del_init(&q->next); 307 RCU_INIT_POINTER(q->tap, NULL); 308 if (q->enabled) 309 tap->numvtaps--; 310 tap->numqueues--; 311 sock_put(&q->sk); 312 } 313 BUG_ON(tap->numvtaps); 314 BUG_ON(tap->numqueues); 315 /* guarantee that any future tap_set_queue will fail */ 316 tap->numvtaps = MAX_TAP_QUEUES; 317 } 318 EXPORT_SYMBOL_GPL(tap_del_queues); 319 320 rx_handler_result_t tap_handle_frame(struct sk_buff **pskb) 321 { 322 struct sk_buff *skb = *pskb; 323 struct net_device *dev = skb->dev; 324 struct tap_dev *tap; 325 struct tap_queue *q; 326 netdev_features_t features = TAP_FEATURES; 327 enum skb_drop_reason drop_reason; 328 329 tap = tap_dev_get_rcu(dev); 330 if (!tap) 331 return RX_HANDLER_PASS; 332 333 q = tap_get_queue(tap, skb); 334 if (!q) 335 return RX_HANDLER_PASS; 336 337 skb_push(skb, ETH_HLEN); 338 339 /* Apply the forward feature mask so that we perform segmentation 340 * according to users wishes. This only works if VNET_HDR is 341 * enabled. 342 */ 343 if (q->flags & IFF_VNET_HDR) 344 features |= tap->tap_features; 345 if (netif_needs_gso(skb, features)) { 346 struct sk_buff *segs = __skb_gso_segment(skb, features, false); 347 struct sk_buff *next; 348 349 if (IS_ERR(segs)) { 350 drop_reason = SKB_DROP_REASON_SKB_GSO_SEG; 351 goto drop; 352 } 353 354 if (!segs) { 355 if (ptr_ring_produce(&q->ring, skb)) { 356 drop_reason = SKB_DROP_REASON_FULL_RING; 357 goto drop; 358 } 359 goto wake_up; 360 } 361 362 consume_skb(skb); 363 skb_list_walk_safe(segs, skb, next) { 364 skb_mark_not_on_list(skb); 365 if (ptr_ring_produce(&q->ring, skb)) { 366 drop_reason = SKB_DROP_REASON_FULL_RING; 367 kfree_skb_reason(skb, drop_reason); 368 kfree_skb_list_reason(next, drop_reason); 369 break; 370 } 371 } 372 } else { 373 /* If we receive a partial checksum and the tap side 374 * doesn't support checksum offload, compute the checksum. 375 * Note: it doesn't matter which checksum feature to 376 * check, we either support them all or none. 377 */ 378 if (skb->ip_summed == CHECKSUM_PARTIAL && 379 !(features & NETIF_F_CSUM_MASK) && 380 skb_checksum_help(skb)) { 381 drop_reason = SKB_DROP_REASON_SKB_CSUM; 382 goto drop; 383 } 384 if (ptr_ring_produce(&q->ring, skb)) { 385 drop_reason = SKB_DROP_REASON_FULL_RING; 386 goto drop; 387 } 388 } 389 390 wake_up: 391 wake_up_interruptible_poll(sk_sleep(&q->sk), EPOLLIN | EPOLLRDNORM | EPOLLRDBAND); 392 return RX_HANDLER_CONSUMED; 393 394 drop: 395 /* Count errors/drops only here, thus don't care about args. */ 396 if (tap->count_rx_dropped) 397 tap->count_rx_dropped(tap); 398 kfree_skb_reason(skb, drop_reason); 399 return RX_HANDLER_CONSUMED; 400 } 401 EXPORT_SYMBOL_GPL(tap_handle_frame); 402 403 static struct major_info *tap_get_major(int major) 404 { 405 struct major_info *tap_major; 406 407 list_for_each_entry_rcu(tap_major, &major_list, next) { 408 if (tap_major->major == major) 409 return tap_major; 410 } 411 412 return NULL; 413 } 414 415 int tap_get_minor(dev_t major, struct tap_dev *tap) 416 { 417 int retval = -ENOMEM; 418 struct major_info *tap_major; 419 420 rcu_read_lock(); 421 tap_major = tap_get_major(MAJOR(major)); 422 if (!tap_major) { 423 retval = -EINVAL; 424 goto unlock; 425 } 426 427 spin_lock(&tap_major->minor_lock); 428 retval = idr_alloc(&tap_major->minor_idr, tap, 1, TAP_NUM_DEVS, GFP_ATOMIC); 429 if (retval >= 0) { 430 tap->minor = retval; 431 } else if (retval == -ENOSPC) { 432 netdev_err(tap->dev, "Too many tap devices\n"); 433 retval = -EINVAL; 434 } 435 spin_unlock(&tap_major->minor_lock); 436 437 unlock: 438 rcu_read_unlock(); 439 return retval < 0 ? retval : 0; 440 } 441 EXPORT_SYMBOL_GPL(tap_get_minor); 442 443 void tap_free_minor(dev_t major, struct tap_dev *tap) 444 { 445 struct major_info *tap_major; 446 447 rcu_read_lock(); 448 tap_major = tap_get_major(MAJOR(major)); 449 if (!tap_major) { 450 goto unlock; 451 } 452 453 spin_lock(&tap_major->minor_lock); 454 if (tap->minor) { 455 idr_remove(&tap_major->minor_idr, tap->minor); 456 tap->minor = 0; 457 } 458 spin_unlock(&tap_major->minor_lock); 459 460 unlock: 461 rcu_read_unlock(); 462 } 463 EXPORT_SYMBOL_GPL(tap_free_minor); 464 465 static struct tap_dev *dev_get_by_tap_file(int major, int minor) 466 { 467 struct net_device *dev = NULL; 468 struct tap_dev *tap; 469 struct major_info *tap_major; 470 471 rcu_read_lock(); 472 tap_major = tap_get_major(major); 473 if (!tap_major) { 474 tap = NULL; 475 goto unlock; 476 } 477 478 spin_lock(&tap_major->minor_lock); 479 tap = idr_find(&tap_major->minor_idr, minor); 480 if (tap) { 481 dev = tap->dev; 482 dev_hold(dev); 483 } 484 spin_unlock(&tap_major->minor_lock); 485 486 unlock: 487 rcu_read_unlock(); 488 return tap; 489 } 490 491 static void tap_sock_write_space(struct sock *sk) 492 { 493 wait_queue_head_t *wqueue; 494 495 if (!sock_writeable(sk) || 496 !test_and_clear_bit(SOCKWQ_ASYNC_NOSPACE, &sk->sk_socket->flags)) 497 return; 498 499 wqueue = sk_sleep(sk); 500 if (wqueue && waitqueue_active(wqueue)) 501 wake_up_interruptible_poll(wqueue, EPOLLOUT | EPOLLWRNORM | EPOLLWRBAND); 502 } 503 504 static void tap_sock_destruct(struct sock *sk) 505 { 506 struct tap_queue *q = container_of(sk, struct tap_queue, sk); 507 508 ptr_ring_cleanup(&q->ring, __skb_array_destroy_skb); 509 } 510 511 static int tap_open(struct inode *inode, struct file *file) 512 { 513 struct net *net = current->nsproxy->net_ns; 514 struct tap_dev *tap; 515 struct tap_queue *q; 516 int err = -ENODEV; 517 518 rtnl_lock(); 519 tap = dev_get_by_tap_file(imajor(inode), iminor(inode)); 520 if (!tap) 521 goto err; 522 523 err = -ENOMEM; 524 q = (struct tap_queue *)sk_alloc(net, AF_UNSPEC, GFP_KERNEL, 525 &tap_proto, 0); 526 if (!q) 527 goto err; 528 if (ptr_ring_init(&q->ring, tap->dev->tx_queue_len, GFP_KERNEL)) { 529 sk_free(&q->sk); 530 goto err; 531 } 532 533 init_waitqueue_head(&q->sock.wq.wait); 534 q->sock.type = SOCK_RAW; 535 q->sock.state = SS_CONNECTED; 536 q->sock.file = file; 537 q->sock.ops = &tap_socket_ops; 538 sock_init_data_uid(&q->sock, &q->sk, current_fsuid()); 539 q->sk.sk_write_space = tap_sock_write_space; 540 q->sk.sk_destruct = tap_sock_destruct; 541 q->flags = IFF_VNET_HDR | IFF_NO_PI | IFF_TAP; 542 q->vnet_hdr_sz = sizeof(struct virtio_net_hdr); 543 544 /* 545 * so far only KVM virtio_net uses tap, enable zero copy between 546 * guest kernel and host kernel when lower device supports zerocopy 547 * 548 * The macvlan supports zerocopy iff the lower device supports zero 549 * copy so we don't have to look at the lower device directly. 550 */ 551 if ((tap->dev->features & NETIF_F_HIGHDMA) && (tap->dev->features & NETIF_F_SG)) 552 sock_set_flag(&q->sk, SOCK_ZEROCOPY); 553 554 err = tap_set_queue(tap, file, q); 555 if (err) { 556 /* tap_sock_destruct() will take care of freeing ptr_ring */ 557 goto err_put; 558 } 559 560 /* tap groks IOCB_NOWAIT just fine, mark it as such */ 561 file->f_mode |= FMODE_NOWAIT; 562 563 dev_put(tap->dev); 564 565 rtnl_unlock(); 566 return err; 567 568 err_put: 569 sock_put(&q->sk); 570 err: 571 if (tap) 572 dev_put(tap->dev); 573 574 rtnl_unlock(); 575 return err; 576 } 577 578 static int tap_release(struct inode *inode, struct file *file) 579 { 580 struct tap_queue *q = file->private_data; 581 tap_put_queue(q); 582 return 0; 583 } 584 585 static __poll_t tap_poll(struct file *file, poll_table *wait) 586 { 587 struct tap_queue *q = file->private_data; 588 __poll_t mask = EPOLLERR; 589 590 if (!q) 591 goto out; 592 593 mask = 0; 594 poll_wait(file, &q->sock.wq.wait, wait); 595 596 if (!ptr_ring_empty(&q->ring)) 597 mask |= EPOLLIN | EPOLLRDNORM; 598 599 if (sock_writeable(&q->sk) || 600 (!test_and_set_bit(SOCKWQ_ASYNC_NOSPACE, &q->sock.flags) && 601 sock_writeable(&q->sk))) 602 mask |= EPOLLOUT | EPOLLWRNORM; 603 604 out: 605 return mask; 606 } 607 608 static inline struct sk_buff *tap_alloc_skb(struct sock *sk, size_t prepad, 609 size_t len, size_t linear, 610 int noblock, int *err) 611 { 612 struct sk_buff *skb; 613 614 /* Under a page? Don't bother with paged skb. */ 615 if (prepad + len < PAGE_SIZE || !linear) 616 linear = len; 617 618 if (len - linear > MAX_SKB_FRAGS * (PAGE_SIZE << PAGE_ALLOC_COSTLY_ORDER)) 619 linear = len - MAX_SKB_FRAGS * (PAGE_SIZE << PAGE_ALLOC_COSTLY_ORDER); 620 skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock, 621 err, PAGE_ALLOC_COSTLY_ORDER); 622 if (!skb) 623 return NULL; 624 625 skb_reserve(skb, prepad); 626 skb_put(skb, linear); 627 skb->data_len = len - linear; 628 skb->len += len - linear; 629 630 return skb; 631 } 632 633 /* Neighbour code has some assumptions on HH_DATA_MOD alignment */ 634 #define TAP_RESERVE HH_DATA_OFF(ETH_HLEN) 635 636 /* Get packet from user space buffer */ 637 static ssize_t tap_get_user(struct tap_queue *q, void *msg_control, 638 struct iov_iter *from, int noblock) 639 { 640 int good_linear = SKB_MAX_HEAD(TAP_RESERVE); 641 struct sk_buff *skb; 642 struct tap_dev *tap; 643 unsigned long total_len = iov_iter_count(from); 644 unsigned long len = total_len; 645 int err; 646 struct virtio_net_hdr vnet_hdr = { 0 }; 647 int vnet_hdr_len = 0; 648 int copylen = 0; 649 int depth; 650 bool zerocopy = false; 651 size_t linear; 652 enum skb_drop_reason drop_reason; 653 654 if (q->flags & IFF_VNET_HDR) { 655 vnet_hdr_len = READ_ONCE(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 = -EFAULT; 663 if (!copy_from_iter_full(&vnet_hdr, sizeof(vnet_hdr), from)) 664 goto err; 665 iov_iter_advance(from, vnet_hdr_len - sizeof(vnet_hdr)); 666 if ((vnet_hdr.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) && 667 tap16_to_cpu(q, vnet_hdr.csum_start) + 668 tap16_to_cpu(q, vnet_hdr.csum_offset) + 2 > 669 tap16_to_cpu(q, vnet_hdr.hdr_len)) 670 vnet_hdr.hdr_len = cpu_to_tap16(q, 671 tap16_to_cpu(q, vnet_hdr.csum_start) + 672 tap16_to_cpu(q, vnet_hdr.csum_offset) + 2); 673 err = -EINVAL; 674 if (tap16_to_cpu(q, vnet_hdr.hdr_len) > len) 675 goto err; 676 } 677 678 err = -EINVAL; 679 if (unlikely(len < ETH_HLEN)) 680 goto err; 681 682 if (msg_control && sock_flag(&q->sk, SOCK_ZEROCOPY)) { 683 struct iov_iter i; 684 685 copylen = vnet_hdr.hdr_len ? 686 tap16_to_cpu(q, vnet_hdr.hdr_len) : GOODCOPY_LEN; 687 if (copylen > good_linear) 688 copylen = good_linear; 689 else if (copylen < ETH_HLEN) 690 copylen = ETH_HLEN; 691 linear = copylen; 692 i = *from; 693 iov_iter_advance(&i, copylen); 694 if (iov_iter_npages(&i, INT_MAX) <= MAX_SKB_FRAGS) 695 zerocopy = true; 696 } 697 698 if (!zerocopy) { 699 copylen = len; 700 linear = tap16_to_cpu(q, vnet_hdr.hdr_len); 701 if (linear > good_linear) 702 linear = good_linear; 703 else if (linear < ETH_HLEN) 704 linear = ETH_HLEN; 705 } 706 707 skb = tap_alloc_skb(&q->sk, TAP_RESERVE, copylen, 708 linear, noblock, &err); 709 if (!skb) 710 goto err; 711 712 if (zerocopy) 713 err = zerocopy_sg_from_iter(skb, from); 714 else 715 err = skb_copy_datagram_from_iter(skb, 0, from, len); 716 717 if (err) { 718 drop_reason = SKB_DROP_REASON_SKB_UCOPY_FAULT; 719 goto err_kfree; 720 } 721 722 skb_set_network_header(skb, ETH_HLEN); 723 skb_reset_mac_header(skb); 724 skb->protocol = eth_hdr(skb)->h_proto; 725 726 rcu_read_lock(); 727 tap = rcu_dereference(q->tap); 728 if (!tap) { 729 kfree_skb(skb); 730 rcu_read_unlock(); 731 return total_len; 732 } 733 skb->dev = tap->dev; 734 735 if (vnet_hdr_len) { 736 err = virtio_net_hdr_to_skb(skb, &vnet_hdr, 737 tap_is_little_endian(q)); 738 if (err) { 739 rcu_read_unlock(); 740 drop_reason = SKB_DROP_REASON_DEV_HDR; 741 goto err_kfree; 742 } 743 } 744 745 skb_probe_transport_header(skb); 746 747 /* Move network header to the right position for VLAN tagged packets */ 748 if (eth_type_vlan(skb->protocol) && 749 vlan_get_protocol_and_depth(skb, skb->protocol, &depth) != 0) 750 skb_set_network_header(skb, depth); 751 752 /* copy skb_ubuf_info for callback when skb has no error */ 753 if (zerocopy) { 754 skb_zcopy_init(skb, msg_control); 755 } else if (msg_control) { 756 struct ubuf_info *uarg = msg_control; 757 uarg->callback(NULL, uarg, false); 758 } 759 760 dev_queue_xmit(skb); 761 rcu_read_unlock(); 762 return total_len; 763 764 err_kfree: 765 kfree_skb_reason(skb, drop_reason); 766 767 err: 768 rcu_read_lock(); 769 tap = rcu_dereference(q->tap); 770 if (tap && tap->count_tx_dropped) 771 tap->count_tx_dropped(tap); 772 rcu_read_unlock(); 773 774 return err; 775 } 776 777 static ssize_t tap_write_iter(struct kiocb *iocb, struct iov_iter *from) 778 { 779 struct file *file = iocb->ki_filp; 780 struct tap_queue *q = file->private_data; 781 int noblock = 0; 782 783 if ((file->f_flags & O_NONBLOCK) || (iocb->ki_flags & IOCB_NOWAIT)) 784 noblock = 1; 785 786 return tap_get_user(q, NULL, from, noblock); 787 } 788 789 /* Put packet to the user space buffer */ 790 static ssize_t tap_put_user(struct tap_queue *q, 791 const struct sk_buff *skb, 792 struct iov_iter *iter) 793 { 794 int ret; 795 int vnet_hdr_len = 0; 796 int vlan_offset = 0; 797 int total; 798 799 if (q->flags & IFF_VNET_HDR) { 800 int vlan_hlen = skb_vlan_tag_present(skb) ? VLAN_HLEN : 0; 801 struct virtio_net_hdr vnet_hdr; 802 803 vnet_hdr_len = READ_ONCE(q->vnet_hdr_sz); 804 if (iov_iter_count(iter) < vnet_hdr_len) 805 return -EINVAL; 806 807 if (virtio_net_hdr_from_skb(skb, &vnet_hdr, 808 tap_is_little_endian(q), true, 809 vlan_hlen)) 810 BUG(); 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 tap_do_read(struct tap_queue *q, 849 struct iov_iter *to, 850 int noblock, struct sk_buff *skb) 851 { 852 DEFINE_WAIT(wait); 853 ssize_t ret = 0; 854 855 if (!iov_iter_count(to)) { 856 kfree_skb(skb); 857 return 0; 858 } 859 860 if (skb) 861 goto put; 862 863 while (1) { 864 if (!noblock) 865 prepare_to_wait(sk_sleep(&q->sk), &wait, 866 TASK_INTERRUPTIBLE); 867 868 /* Read frames from the queue */ 869 skb = ptr_ring_consume(&q->ring); 870 if (skb) 871 break; 872 if (noblock) { 873 ret = -EAGAIN; 874 break; 875 } 876 if (signal_pending(current)) { 877 ret = -ERESTARTSYS; 878 break; 879 } 880 /* Nothing to read, let's sleep */ 881 schedule(); 882 } 883 if (!noblock) 884 finish_wait(sk_sleep(&q->sk), &wait); 885 886 put: 887 if (skb) { 888 ret = tap_put_user(q, skb, to); 889 if (unlikely(ret < 0)) 890 kfree_skb(skb); 891 else 892 consume_skb(skb); 893 } 894 return ret; 895 } 896 897 static ssize_t tap_read_iter(struct kiocb *iocb, struct iov_iter *to) 898 { 899 struct file *file = iocb->ki_filp; 900 struct tap_queue *q = file->private_data; 901 ssize_t len = iov_iter_count(to), ret; 902 int noblock = 0; 903 904 if ((file->f_flags & O_NONBLOCK) || (iocb->ki_flags & IOCB_NOWAIT)) 905 noblock = 1; 906 907 ret = tap_do_read(q, to, noblock, NULL); 908 ret = min_t(ssize_t, ret, len); 909 if (ret > 0) 910 iocb->ki_pos = ret; 911 return ret; 912 } 913 914 static struct tap_dev *tap_get_tap_dev(struct tap_queue *q) 915 { 916 struct tap_dev *tap; 917 918 ASSERT_RTNL(); 919 tap = rtnl_dereference(q->tap); 920 if (tap) 921 dev_hold(tap->dev); 922 923 return tap; 924 } 925 926 static void tap_put_tap_dev(struct tap_dev *tap) 927 { 928 dev_put(tap->dev); 929 } 930 931 static int tap_ioctl_set_queue(struct file *file, unsigned int flags) 932 { 933 struct tap_queue *q = file->private_data; 934 struct tap_dev *tap; 935 int ret; 936 937 tap = tap_get_tap_dev(q); 938 if (!tap) 939 return -EINVAL; 940 941 if (flags & IFF_ATTACH_QUEUE) 942 ret = tap_enable_queue(tap, file, q); 943 else if (flags & IFF_DETACH_QUEUE) 944 ret = tap_disable_queue(q); 945 else 946 ret = -EINVAL; 947 948 tap_put_tap_dev(tap); 949 return ret; 950 } 951 952 static int set_offload(struct tap_queue *q, unsigned long arg) 953 { 954 struct tap_dev *tap; 955 netdev_features_t features; 956 netdev_features_t feature_mask = 0; 957 958 tap = rtnl_dereference(q->tap); 959 if (!tap) 960 return -ENOLINK; 961 962 features = tap->dev->features; 963 964 if (arg & TUN_F_CSUM) { 965 feature_mask = NETIF_F_HW_CSUM; 966 967 if (arg & (TUN_F_TSO4 | TUN_F_TSO6)) { 968 if (arg & TUN_F_TSO_ECN) 969 feature_mask |= NETIF_F_TSO_ECN; 970 if (arg & TUN_F_TSO4) 971 feature_mask |= NETIF_F_TSO; 972 if (arg & TUN_F_TSO6) 973 feature_mask |= NETIF_F_TSO6; 974 } 975 976 /* TODO: for now USO4 and USO6 should work simultaneously */ 977 if ((arg & (TUN_F_USO4 | TUN_F_USO6)) == (TUN_F_USO4 | TUN_F_USO6)) 978 features |= NETIF_F_GSO_UDP_L4; 979 } 980 981 /* tun/tap driver inverts the usage for TSO offloads, where 982 * setting the TSO bit means that the userspace wants to 983 * accept TSO frames and turning it off means that user space 984 * does not support TSO. 985 * For tap, we have to invert it to mean the same thing. 986 * When user space turns off TSO, we turn off GSO/LRO so that 987 * user-space will not receive TSO frames. 988 */ 989 if (feature_mask & (NETIF_F_TSO | NETIF_F_TSO6) || 990 (feature_mask & (TUN_F_USO4 | TUN_F_USO6)) == (TUN_F_USO4 | TUN_F_USO6)) 991 features |= RX_OFFLOADS; 992 else 993 features &= ~RX_OFFLOADS; 994 995 /* tap_features are the same as features on tun/tap and 996 * reflect user expectations. 997 */ 998 tap->tap_features = feature_mask; 999 if (tap->update_features) 1000 tap->update_features(tap, features); 1001 1002 return 0; 1003 } 1004 1005 /* 1006 * provide compatibility with generic tun/tap interface 1007 */ 1008 static long tap_ioctl(struct file *file, unsigned int cmd, 1009 unsigned long arg) 1010 { 1011 struct tap_queue *q = file->private_data; 1012 struct tap_dev *tap; 1013 void __user *argp = (void __user *)arg; 1014 struct ifreq __user *ifr = argp; 1015 unsigned int __user *up = argp; 1016 unsigned short u; 1017 int __user *sp = argp; 1018 struct sockaddr sa; 1019 int s; 1020 int ret; 1021 1022 switch (cmd) { 1023 case TUNSETIFF: 1024 /* ignore the name, just look at flags */ 1025 if (get_user(u, &ifr->ifr_flags)) 1026 return -EFAULT; 1027 1028 ret = 0; 1029 if ((u & ~TAP_IFFEATURES) != (IFF_NO_PI | IFF_TAP)) 1030 ret = -EINVAL; 1031 else 1032 q->flags = (q->flags & ~TAP_IFFEATURES) | u; 1033 1034 return ret; 1035 1036 case TUNGETIFF: 1037 rtnl_lock(); 1038 tap = tap_get_tap_dev(q); 1039 if (!tap) { 1040 rtnl_unlock(); 1041 return -ENOLINK; 1042 } 1043 1044 ret = 0; 1045 u = q->flags; 1046 if (copy_to_user(&ifr->ifr_name, tap->dev->name, IFNAMSIZ) || 1047 put_user(u, &ifr->ifr_flags)) 1048 ret = -EFAULT; 1049 tap_put_tap_dev(tap); 1050 rtnl_unlock(); 1051 return ret; 1052 1053 case TUNSETQUEUE: 1054 if (get_user(u, &ifr->ifr_flags)) 1055 return -EFAULT; 1056 rtnl_lock(); 1057 ret = tap_ioctl_set_queue(file, u); 1058 rtnl_unlock(); 1059 return ret; 1060 1061 case TUNGETFEATURES: 1062 if (put_user(IFF_TAP | IFF_NO_PI | TAP_IFFEATURES, up)) 1063 return -EFAULT; 1064 return 0; 1065 1066 case TUNSETSNDBUF: 1067 if (get_user(s, sp)) 1068 return -EFAULT; 1069 if (s <= 0) 1070 return -EINVAL; 1071 1072 q->sk.sk_sndbuf = s; 1073 return 0; 1074 1075 case TUNGETVNETHDRSZ: 1076 s = q->vnet_hdr_sz; 1077 if (put_user(s, sp)) 1078 return -EFAULT; 1079 return 0; 1080 1081 case TUNSETVNETHDRSZ: 1082 if (get_user(s, sp)) 1083 return -EFAULT; 1084 if (s < (int)sizeof(struct virtio_net_hdr)) 1085 return -EINVAL; 1086 1087 q->vnet_hdr_sz = s; 1088 return 0; 1089 1090 case TUNGETVNETLE: 1091 s = !!(q->flags & TAP_VNET_LE); 1092 if (put_user(s, sp)) 1093 return -EFAULT; 1094 return 0; 1095 1096 case TUNSETVNETLE: 1097 if (get_user(s, sp)) 1098 return -EFAULT; 1099 if (s) 1100 q->flags |= TAP_VNET_LE; 1101 else 1102 q->flags &= ~TAP_VNET_LE; 1103 return 0; 1104 1105 case TUNGETVNETBE: 1106 return tap_get_vnet_be(q, sp); 1107 1108 case TUNSETVNETBE: 1109 return tap_set_vnet_be(q, sp); 1110 1111 case TUNSETOFFLOAD: 1112 /* let the user check for future flags */ 1113 if (arg & ~(TUN_F_CSUM | TUN_F_TSO4 | TUN_F_TSO6 | 1114 TUN_F_TSO_ECN | TUN_F_UFO | 1115 TUN_F_USO4 | TUN_F_USO6)) 1116 return -EINVAL; 1117 1118 rtnl_lock(); 1119 ret = set_offload(q, arg); 1120 rtnl_unlock(); 1121 return ret; 1122 1123 case SIOCGIFHWADDR: 1124 rtnl_lock(); 1125 tap = tap_get_tap_dev(q); 1126 if (!tap) { 1127 rtnl_unlock(); 1128 return -ENOLINK; 1129 } 1130 ret = 0; 1131 dev_get_mac_address(&sa, dev_net(tap->dev), tap->dev->name); 1132 if (copy_to_user(&ifr->ifr_name, tap->dev->name, IFNAMSIZ) || 1133 copy_to_user(&ifr->ifr_hwaddr, &sa, sizeof(sa))) 1134 ret = -EFAULT; 1135 tap_put_tap_dev(tap); 1136 rtnl_unlock(); 1137 return ret; 1138 1139 case SIOCSIFHWADDR: 1140 if (copy_from_user(&sa, &ifr->ifr_hwaddr, sizeof(sa))) 1141 return -EFAULT; 1142 rtnl_lock(); 1143 tap = tap_get_tap_dev(q); 1144 if (!tap) { 1145 rtnl_unlock(); 1146 return -ENOLINK; 1147 } 1148 ret = dev_set_mac_address_user(tap->dev, &sa, NULL); 1149 tap_put_tap_dev(tap); 1150 rtnl_unlock(); 1151 return ret; 1152 1153 default: 1154 return -EINVAL; 1155 } 1156 } 1157 1158 static const struct file_operations tap_fops = { 1159 .owner = THIS_MODULE, 1160 .open = tap_open, 1161 .release = tap_release, 1162 .read_iter = tap_read_iter, 1163 .write_iter = tap_write_iter, 1164 .poll = tap_poll, 1165 .llseek = no_llseek, 1166 .unlocked_ioctl = tap_ioctl, 1167 .compat_ioctl = compat_ptr_ioctl, 1168 }; 1169 1170 static int tap_get_user_xdp(struct tap_queue *q, struct xdp_buff *xdp) 1171 { 1172 struct tun_xdp_hdr *hdr = xdp->data_hard_start; 1173 struct virtio_net_hdr *gso = &hdr->gso; 1174 int buflen = hdr->buflen; 1175 int vnet_hdr_len = 0; 1176 struct tap_dev *tap; 1177 struct sk_buff *skb; 1178 int err, depth; 1179 1180 if (unlikely(xdp->data_end - xdp->data < ETH_HLEN)) { 1181 err = -EINVAL; 1182 goto err; 1183 } 1184 1185 if (q->flags & IFF_VNET_HDR) 1186 vnet_hdr_len = READ_ONCE(q->vnet_hdr_sz); 1187 1188 skb = build_skb(xdp->data_hard_start, buflen); 1189 if (!skb) { 1190 err = -ENOMEM; 1191 goto err; 1192 } 1193 1194 skb_reserve(skb, xdp->data - xdp->data_hard_start); 1195 skb_put(skb, xdp->data_end - xdp->data); 1196 1197 skb_set_network_header(skb, ETH_HLEN); 1198 skb_reset_mac_header(skb); 1199 skb->protocol = eth_hdr(skb)->h_proto; 1200 1201 if (vnet_hdr_len) { 1202 err = virtio_net_hdr_to_skb(skb, gso, tap_is_little_endian(q)); 1203 if (err) 1204 goto err_kfree; 1205 } 1206 1207 /* Move network header to the right position for VLAN tagged packets */ 1208 if (eth_type_vlan(skb->protocol) && 1209 vlan_get_protocol_and_depth(skb, skb->protocol, &depth) != 0) 1210 skb_set_network_header(skb, depth); 1211 1212 rcu_read_lock(); 1213 tap = rcu_dereference(q->tap); 1214 if (tap) { 1215 skb->dev = tap->dev; 1216 skb_probe_transport_header(skb); 1217 dev_queue_xmit(skb); 1218 } else { 1219 kfree_skb(skb); 1220 } 1221 rcu_read_unlock(); 1222 1223 return 0; 1224 1225 err_kfree: 1226 kfree_skb(skb); 1227 err: 1228 rcu_read_lock(); 1229 tap = rcu_dereference(q->tap); 1230 if (tap && tap->count_tx_dropped) 1231 tap->count_tx_dropped(tap); 1232 rcu_read_unlock(); 1233 return err; 1234 } 1235 1236 static int tap_sendmsg(struct socket *sock, struct msghdr *m, 1237 size_t total_len) 1238 { 1239 struct tap_queue *q = container_of(sock, struct tap_queue, sock); 1240 struct tun_msg_ctl *ctl = m->msg_control; 1241 struct xdp_buff *xdp; 1242 int i; 1243 1244 if (m->msg_controllen == sizeof(struct tun_msg_ctl) && 1245 ctl && ctl->type == TUN_MSG_PTR) { 1246 for (i = 0; i < ctl->num; i++) { 1247 xdp = &((struct xdp_buff *)ctl->ptr)[i]; 1248 tap_get_user_xdp(q, xdp); 1249 } 1250 return 0; 1251 } 1252 1253 return tap_get_user(q, ctl ? ctl->ptr : NULL, &m->msg_iter, 1254 m->msg_flags & MSG_DONTWAIT); 1255 } 1256 1257 static int tap_recvmsg(struct socket *sock, struct msghdr *m, 1258 size_t total_len, int flags) 1259 { 1260 struct tap_queue *q = container_of(sock, struct tap_queue, sock); 1261 struct sk_buff *skb = m->msg_control; 1262 int ret; 1263 if (flags & ~(MSG_DONTWAIT|MSG_TRUNC)) { 1264 kfree_skb(skb); 1265 return -EINVAL; 1266 } 1267 ret = tap_do_read(q, &m->msg_iter, flags & MSG_DONTWAIT, skb); 1268 if (ret > total_len) { 1269 m->msg_flags |= MSG_TRUNC; 1270 ret = flags & MSG_TRUNC ? ret : total_len; 1271 } 1272 return ret; 1273 } 1274 1275 static int tap_peek_len(struct socket *sock) 1276 { 1277 struct tap_queue *q = container_of(sock, struct tap_queue, 1278 sock); 1279 return PTR_RING_PEEK_CALL(&q->ring, __skb_array_len_with_tag); 1280 } 1281 1282 /* Ops structure to mimic raw sockets with tun */ 1283 static const struct proto_ops tap_socket_ops = { 1284 .sendmsg = tap_sendmsg, 1285 .recvmsg = tap_recvmsg, 1286 .peek_len = tap_peek_len, 1287 }; 1288 1289 /* Get an underlying socket object from tun file. Returns error unless file is 1290 * attached to a device. The returned object works like a packet socket, it 1291 * can be used for sock_sendmsg/sock_recvmsg. The caller is responsible for 1292 * holding a reference to the file for as long as the socket is in use. */ 1293 struct socket *tap_get_socket(struct file *file) 1294 { 1295 struct tap_queue *q; 1296 if (file->f_op != &tap_fops) 1297 return ERR_PTR(-EINVAL); 1298 q = file->private_data; 1299 if (!q) 1300 return ERR_PTR(-EBADFD); 1301 return &q->sock; 1302 } 1303 EXPORT_SYMBOL_GPL(tap_get_socket); 1304 1305 struct ptr_ring *tap_get_ptr_ring(struct file *file) 1306 { 1307 struct tap_queue *q; 1308 1309 if (file->f_op != &tap_fops) 1310 return ERR_PTR(-EINVAL); 1311 q = file->private_data; 1312 if (!q) 1313 return ERR_PTR(-EBADFD); 1314 return &q->ring; 1315 } 1316 EXPORT_SYMBOL_GPL(tap_get_ptr_ring); 1317 1318 int tap_queue_resize(struct tap_dev *tap) 1319 { 1320 struct net_device *dev = tap->dev; 1321 struct tap_queue *q; 1322 struct ptr_ring **rings; 1323 int n = tap->numqueues; 1324 int ret, i = 0; 1325 1326 rings = kmalloc_array(n, sizeof(*rings), GFP_KERNEL); 1327 if (!rings) 1328 return -ENOMEM; 1329 1330 list_for_each_entry(q, &tap->queue_list, next) 1331 rings[i++] = &q->ring; 1332 1333 ret = ptr_ring_resize_multiple(rings, n, 1334 dev->tx_queue_len, GFP_KERNEL, 1335 __skb_array_destroy_skb); 1336 1337 kfree(rings); 1338 return ret; 1339 } 1340 EXPORT_SYMBOL_GPL(tap_queue_resize); 1341 1342 static int tap_list_add(dev_t major, const char *device_name) 1343 { 1344 struct major_info *tap_major; 1345 1346 tap_major = kzalloc(sizeof(*tap_major), GFP_ATOMIC); 1347 if (!tap_major) 1348 return -ENOMEM; 1349 1350 tap_major->major = MAJOR(major); 1351 1352 idr_init(&tap_major->minor_idr); 1353 spin_lock_init(&tap_major->minor_lock); 1354 1355 tap_major->device_name = device_name; 1356 1357 list_add_tail_rcu(&tap_major->next, &major_list); 1358 return 0; 1359 } 1360 1361 int tap_create_cdev(struct cdev *tap_cdev, dev_t *tap_major, 1362 const char *device_name, struct module *module) 1363 { 1364 int err; 1365 1366 err = alloc_chrdev_region(tap_major, 0, TAP_NUM_DEVS, device_name); 1367 if (err) 1368 goto out1; 1369 1370 cdev_init(tap_cdev, &tap_fops); 1371 tap_cdev->owner = module; 1372 err = cdev_add(tap_cdev, *tap_major, TAP_NUM_DEVS); 1373 if (err) 1374 goto out2; 1375 1376 err = tap_list_add(*tap_major, device_name); 1377 if (err) 1378 goto out3; 1379 1380 return 0; 1381 1382 out3: 1383 cdev_del(tap_cdev); 1384 out2: 1385 unregister_chrdev_region(*tap_major, TAP_NUM_DEVS); 1386 out1: 1387 return err; 1388 } 1389 EXPORT_SYMBOL_GPL(tap_create_cdev); 1390 1391 void tap_destroy_cdev(dev_t major, struct cdev *tap_cdev) 1392 { 1393 struct major_info *tap_major, *tmp; 1394 1395 cdev_del(tap_cdev); 1396 unregister_chrdev_region(major, TAP_NUM_DEVS); 1397 list_for_each_entry_safe(tap_major, tmp, &major_list, next) { 1398 if (tap_major->major == MAJOR(major)) { 1399 idr_destroy(&tap_major->minor_idr); 1400 list_del_rcu(&tap_major->next); 1401 kfree_rcu(tap_major, rcu); 1402 } 1403 } 1404 } 1405 EXPORT_SYMBOL_GPL(tap_destroy_cdev); 1406 1407 MODULE_AUTHOR("Arnd Bergmann <arnd@arndb.de>"); 1408 MODULE_AUTHOR("Sainath Grandhi <sainath.grandhi@intel.com>"); 1409 MODULE_LICENSE("GPL"); 1410