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 #include "tun_vnet.h" 30 31 #define TAP_IFFEATURES (IFF_VNET_HDR | IFF_MULTI_QUEUE) 32 33 static struct proto tap_proto = { 34 .name = "tap", 35 .owner = THIS_MODULE, 36 .obj_size = sizeof(struct tap_queue), 37 }; 38 39 #define TAP_NUM_DEVS (1U << MINORBITS) 40 41 static LIST_HEAD(major_list); 42 43 struct major_info { 44 struct rcu_head rcu; 45 dev_t major; 46 struct idr minor_idr; 47 spinlock_t minor_lock; 48 const char *device_name; 49 struct list_head next; 50 }; 51 52 #define GOODCOPY_LEN 128 53 54 static const struct proto_ops tap_socket_ops; 55 56 #define RX_OFFLOADS (NETIF_F_GRO | NETIF_F_LRO) 57 #define TAP_FEATURES (NETIF_F_GSO | NETIF_F_SG | NETIF_F_FRAGLIST) 58 59 static struct tap_dev *tap_dev_get_rcu(const struct net_device *dev) 60 { 61 return rcu_dereference(dev->rx_handler_data); 62 } 63 64 /* 65 * RCU usage: 66 * The tap_queue and the macvlan_dev are loosely coupled, the 67 * pointers from one to the other can only be read while rcu_read_lock 68 * or rtnl is held. 69 * 70 * Both the file and the macvlan_dev hold a reference on the tap_queue 71 * through sock_hold(&q->sk). When the macvlan_dev goes away first, 72 * q->vlan becomes inaccessible. When the files gets closed, 73 * tap_get_queue() fails. 74 * 75 * There may still be references to the struct sock inside of the 76 * queue from outbound SKBs, but these never reference back to the 77 * file or the dev. The data structure is freed through __sk_free 78 * when both our references and any pending SKBs are gone. 79 */ 80 81 static int tap_enable_queue(struct tap_dev *tap, struct file *file, 82 struct tap_queue *q) 83 { 84 int err = -EINVAL; 85 86 ASSERT_RTNL(); 87 88 if (q->enabled) 89 goto out; 90 91 err = 0; 92 rcu_assign_pointer(tap->taps[tap->numvtaps], q); 93 q->queue_index = tap->numvtaps; 94 q->enabled = true; 95 96 tap->numvtaps++; 97 out: 98 return err; 99 } 100 101 /* Requires RTNL */ 102 static int tap_set_queue(struct tap_dev *tap, struct file *file, 103 struct tap_queue *q) 104 { 105 if (tap->numqueues == MAX_TAP_QUEUES) 106 return -EBUSY; 107 108 rcu_assign_pointer(q->tap, tap); 109 rcu_assign_pointer(tap->taps[tap->numvtaps], q); 110 sock_hold(&q->sk); 111 112 q->file = file; 113 q->queue_index = tap->numvtaps; 114 q->enabled = true; 115 file->private_data = q; 116 list_add_tail(&q->next, &tap->queue_list); 117 118 tap->numvtaps++; 119 tap->numqueues++; 120 121 return 0; 122 } 123 124 static int tap_disable_queue(struct tap_queue *q) 125 { 126 struct tap_dev *tap; 127 struct tap_queue *nq; 128 129 ASSERT_RTNL(); 130 if (!q->enabled) 131 return -EINVAL; 132 133 tap = rtnl_dereference(q->tap); 134 135 if (tap) { 136 int index = q->queue_index; 137 BUG_ON(index >= tap->numvtaps); 138 nq = rtnl_dereference(tap->taps[tap->numvtaps - 1]); 139 nq->queue_index = index; 140 141 rcu_assign_pointer(tap->taps[index], nq); 142 RCU_INIT_POINTER(tap->taps[tap->numvtaps - 1], NULL); 143 q->enabled = false; 144 145 tap->numvtaps--; 146 } 147 148 return 0; 149 } 150 151 /* 152 * The file owning the queue got closed, give up both 153 * the reference that the files holds as well as the 154 * one from the macvlan_dev if that still exists. 155 * 156 * Using the spinlock makes sure that we don't get 157 * to the queue again after destroying it. 158 */ 159 static void tap_put_queue(struct tap_queue *q) 160 { 161 struct tap_dev *tap; 162 163 rtnl_lock(); 164 tap = rtnl_dereference(q->tap); 165 166 if (tap) { 167 if (q->enabled) 168 BUG_ON(tap_disable_queue(q)); 169 170 tap->numqueues--; 171 RCU_INIT_POINTER(q->tap, NULL); 172 sock_put(&q->sk); 173 list_del_init(&q->next); 174 } 175 176 rtnl_unlock(); 177 178 synchronize_rcu(); 179 sock_put(&q->sk); 180 } 181 182 /* 183 * Select a queue based on the rxq of the device on which this packet 184 * arrived. If the incoming device is not mq, calculate a flow hash 185 * to select a queue. If all fails, find the first available queue. 186 * Cache vlan->numvtaps since it can become zero during the execution 187 * of this function. 188 */ 189 static struct tap_queue *tap_get_queue(struct tap_dev *tap, 190 struct sk_buff *skb) 191 { 192 struct tap_queue *queue = NULL; 193 /* Access to taps array is protected by rcu, but access to numvtaps 194 * isn't. Below we use it to lookup a queue, but treat it as a hint 195 * and validate that the result isn't NULL - in case we are 196 * racing against queue removal. 197 */ 198 int numvtaps = READ_ONCE(tap->numvtaps); 199 __u32 rxq; 200 201 if (!numvtaps) 202 goto out; 203 204 if (numvtaps == 1) 205 goto single; 206 207 /* Check if we can use flow to select a queue */ 208 rxq = skb_get_hash(skb); 209 if (rxq) { 210 queue = rcu_dereference(tap->taps[rxq % numvtaps]); 211 goto out; 212 } 213 214 if (likely(skb_rx_queue_recorded(skb))) { 215 rxq = skb_get_rx_queue(skb); 216 217 while (unlikely(rxq >= numvtaps)) 218 rxq -= numvtaps; 219 220 queue = rcu_dereference(tap->taps[rxq]); 221 goto out; 222 } 223 224 single: 225 queue = rcu_dereference(tap->taps[0]); 226 out: 227 return queue; 228 } 229 230 /* 231 * The net_device is going away, give up the reference 232 * that it holds on all queues and safely set the pointer 233 * from the queues to NULL. 234 */ 235 void tap_del_queues(struct tap_dev *tap) 236 { 237 struct tap_queue *q, *tmp; 238 239 ASSERT_RTNL(); 240 list_for_each_entry_safe(q, tmp, &tap->queue_list, next) { 241 list_del_init(&q->next); 242 RCU_INIT_POINTER(q->tap, NULL); 243 if (q->enabled) 244 tap->numvtaps--; 245 tap->numqueues--; 246 sock_put(&q->sk); 247 } 248 BUG_ON(tap->numvtaps); 249 BUG_ON(tap->numqueues); 250 /* guarantee that any future tap_set_queue will fail */ 251 tap->numvtaps = MAX_TAP_QUEUES; 252 } 253 EXPORT_SYMBOL_GPL(tap_del_queues); 254 255 rx_handler_result_t tap_handle_frame(struct sk_buff **pskb) 256 { 257 struct sk_buff *skb = *pskb; 258 struct net_device *dev = skb->dev; 259 struct tap_dev *tap; 260 struct tap_queue *q; 261 netdev_features_t features = TAP_FEATURES; 262 enum skb_drop_reason drop_reason; 263 264 tap = tap_dev_get_rcu(dev); 265 if (!tap) 266 return RX_HANDLER_PASS; 267 268 q = tap_get_queue(tap, skb); 269 if (!q) 270 return RX_HANDLER_PASS; 271 272 skb_push(skb, ETH_HLEN); 273 274 /* Apply the forward feature mask so that we perform segmentation 275 * according to users wishes. This only works if VNET_HDR is 276 * enabled. 277 */ 278 if (q->flags & IFF_VNET_HDR) 279 features |= tap->tap_features; 280 if (netif_needs_gso(skb, features)) { 281 struct sk_buff *segs; 282 struct sk_buff *next; 283 284 segs = __skb_gso_segment(skb, features, false, 0); 285 if (IS_ERR(segs)) { 286 drop_reason = SKB_DROP_REASON_SKB_GSO_SEG; 287 goto drop; 288 } 289 290 if (!segs) { 291 if (ptr_ring_produce(&q->ring, skb)) { 292 drop_reason = SKB_DROP_REASON_FULL_RING; 293 goto drop; 294 } 295 goto wake_up; 296 } 297 298 consume_skb(skb); 299 skb_list_walk_safe(segs, skb, next) { 300 skb_mark_not_on_list(skb); 301 if (ptr_ring_produce(&q->ring, skb)) { 302 drop_reason = SKB_DROP_REASON_FULL_RING; 303 kfree_skb_reason(skb, drop_reason); 304 kfree_skb_list_reason(next, drop_reason); 305 break; 306 } 307 } 308 } else { 309 /* If we receive a partial checksum and the tap side 310 * doesn't support checksum offload, compute the checksum. 311 * Note: it doesn't matter which checksum feature to 312 * check, we either support them all or none. 313 */ 314 if (skb->ip_summed == CHECKSUM_PARTIAL && 315 !(features & NETIF_F_CSUM_MASK) && 316 skb_checksum_help(skb)) { 317 drop_reason = SKB_DROP_REASON_SKB_CSUM; 318 goto drop; 319 } 320 if (ptr_ring_produce(&q->ring, skb)) { 321 drop_reason = SKB_DROP_REASON_FULL_RING; 322 goto drop; 323 } 324 } 325 326 wake_up: 327 wake_up_interruptible_poll(sk_sleep(&q->sk), EPOLLIN | EPOLLRDNORM | EPOLLRDBAND); 328 return RX_HANDLER_CONSUMED; 329 330 drop: 331 /* Count errors/drops only here, thus don't care about args. */ 332 if (tap->count_rx_dropped) 333 tap->count_rx_dropped(tap); 334 kfree_skb_reason(skb, drop_reason); 335 return RX_HANDLER_CONSUMED; 336 } 337 EXPORT_SYMBOL_GPL(tap_handle_frame); 338 339 static struct major_info *tap_get_major(int major) 340 { 341 struct major_info *tap_major; 342 343 list_for_each_entry_rcu(tap_major, &major_list, next) { 344 if (tap_major->major == major) 345 return tap_major; 346 } 347 348 return NULL; 349 } 350 351 int tap_get_minor(dev_t major, struct tap_dev *tap) 352 { 353 int retval = -ENOMEM; 354 struct major_info *tap_major; 355 356 rcu_read_lock(); 357 tap_major = tap_get_major(MAJOR(major)); 358 if (!tap_major) { 359 retval = -EINVAL; 360 goto unlock; 361 } 362 363 spin_lock(&tap_major->minor_lock); 364 retval = idr_alloc(&tap_major->minor_idr, tap, 1, TAP_NUM_DEVS, GFP_ATOMIC); 365 if (retval >= 0) { 366 tap->minor = retval; 367 } else if (retval == -ENOSPC) { 368 netdev_err(tap->dev, "Too many tap devices\n"); 369 retval = -EINVAL; 370 } 371 spin_unlock(&tap_major->minor_lock); 372 373 unlock: 374 rcu_read_unlock(); 375 return retval < 0 ? retval : 0; 376 } 377 EXPORT_SYMBOL_GPL(tap_get_minor); 378 379 void tap_free_minor(dev_t major, struct tap_dev *tap) 380 { 381 struct major_info *tap_major; 382 383 rcu_read_lock(); 384 tap_major = tap_get_major(MAJOR(major)); 385 if (!tap_major) { 386 goto unlock; 387 } 388 389 spin_lock(&tap_major->minor_lock); 390 if (tap->minor) { 391 idr_remove(&tap_major->minor_idr, tap->minor); 392 tap->minor = 0; 393 } 394 spin_unlock(&tap_major->minor_lock); 395 396 unlock: 397 rcu_read_unlock(); 398 } 399 EXPORT_SYMBOL_GPL(tap_free_minor); 400 401 static struct tap_dev *dev_get_by_tap_file(int major, int minor) 402 { 403 struct net_device *dev = NULL; 404 struct tap_dev *tap; 405 struct major_info *tap_major; 406 407 rcu_read_lock(); 408 tap_major = tap_get_major(major); 409 if (!tap_major) { 410 tap = NULL; 411 goto unlock; 412 } 413 414 spin_lock(&tap_major->minor_lock); 415 tap = idr_find(&tap_major->minor_idr, minor); 416 if (tap) { 417 dev = tap->dev; 418 dev_hold(dev); 419 } 420 spin_unlock(&tap_major->minor_lock); 421 422 unlock: 423 rcu_read_unlock(); 424 return tap; 425 } 426 427 static void tap_sock_write_space(struct sock *sk) 428 { 429 wait_queue_head_t *wqueue; 430 431 if (!sock_writeable(sk) || 432 !test_and_clear_bit(SOCKWQ_ASYNC_NOSPACE, &sk->sk_socket->flags)) 433 return; 434 435 wqueue = sk_sleep(sk); 436 if (wqueue && waitqueue_active(wqueue)) 437 wake_up_interruptible_poll(wqueue, EPOLLOUT | EPOLLWRNORM | EPOLLWRBAND); 438 } 439 440 static void tap_sock_destruct(struct sock *sk) 441 { 442 struct tap_queue *q = container_of(sk, struct tap_queue, sk); 443 444 ptr_ring_cleanup(&q->ring, __skb_array_destroy_skb); 445 } 446 447 static int tap_open(struct inode *inode, struct file *file) 448 { 449 struct net *net = current->nsproxy->net_ns; 450 struct tap_dev *tap; 451 struct tap_queue *q; 452 int err = -ENODEV; 453 454 rtnl_lock(); 455 tap = dev_get_by_tap_file(imajor(inode), iminor(inode)); 456 if (!tap) 457 goto err; 458 459 err = -ENOMEM; 460 q = (struct tap_queue *)sk_alloc(net, AF_UNSPEC, GFP_KERNEL, 461 &tap_proto, 0); 462 if (!q) 463 goto err; 464 if (ptr_ring_init(&q->ring, tap->dev->tx_queue_len, GFP_KERNEL)) { 465 sk_free(&q->sk); 466 goto err; 467 } 468 469 init_waitqueue_head(&q->sock.wq.wait); 470 q->sock.type = SOCK_RAW; 471 q->sock.state = SS_CONNECTED; 472 q->sock.file = file; 473 q->sock.ops = &tap_socket_ops; 474 sock_init_data_uid(&q->sock, &q->sk, current_fsuid()); 475 q->sk.sk_write_space = tap_sock_write_space; 476 q->sk.sk_destruct = tap_sock_destruct; 477 q->flags = IFF_VNET_HDR | IFF_NO_PI | IFF_TAP; 478 q->vnet_hdr_sz = sizeof(struct virtio_net_hdr); 479 480 /* 481 * so far only KVM virtio_net uses tap, enable zero copy between 482 * guest kernel and host kernel when lower device supports zerocopy 483 * 484 * The macvlan supports zerocopy iff the lower device supports zero 485 * copy so we don't have to look at the lower device directly. 486 */ 487 if ((tap->dev->features & NETIF_F_HIGHDMA) && (tap->dev->features & NETIF_F_SG)) 488 sock_set_flag(&q->sk, SOCK_ZEROCOPY); 489 490 err = tap_set_queue(tap, file, q); 491 if (err) { 492 /* tap_sock_destruct() will take care of freeing ptr_ring */ 493 goto err_put; 494 } 495 496 /* tap groks IOCB_NOWAIT just fine, mark it as such */ 497 file->f_mode |= FMODE_NOWAIT; 498 499 dev_put(tap->dev); 500 501 rtnl_unlock(); 502 return err; 503 504 err_put: 505 sock_put(&q->sk); 506 err: 507 if (tap) 508 dev_put(tap->dev); 509 510 rtnl_unlock(); 511 return err; 512 } 513 514 static int tap_release(struct inode *inode, struct file *file) 515 { 516 struct tap_queue *q = file->private_data; 517 tap_put_queue(q); 518 return 0; 519 } 520 521 static __poll_t tap_poll(struct file *file, poll_table *wait) 522 { 523 struct tap_queue *q = file->private_data; 524 __poll_t mask = EPOLLERR; 525 526 if (!q) 527 goto out; 528 529 mask = 0; 530 poll_wait(file, &q->sock.wq.wait, wait); 531 532 if (!ptr_ring_empty(&q->ring)) 533 mask |= EPOLLIN | EPOLLRDNORM; 534 535 if (sock_writeable(&q->sk) || 536 (!test_and_set_bit(SOCKWQ_ASYNC_NOSPACE, &q->sock.flags) && 537 sock_writeable(&q->sk))) 538 mask |= EPOLLOUT | EPOLLWRNORM; 539 540 out: 541 return mask; 542 } 543 544 static inline struct sk_buff *tap_alloc_skb(struct sock *sk, size_t prepad, 545 size_t len, size_t linear, 546 int noblock, int *err) 547 { 548 struct sk_buff *skb; 549 550 /* Under a page? Don't bother with paged skb. */ 551 if (prepad + len < PAGE_SIZE || !linear) 552 linear = len; 553 554 if (len - linear > MAX_SKB_FRAGS * (PAGE_SIZE << PAGE_ALLOC_COSTLY_ORDER)) 555 linear = len - MAX_SKB_FRAGS * (PAGE_SIZE << PAGE_ALLOC_COSTLY_ORDER); 556 skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock, 557 err, PAGE_ALLOC_COSTLY_ORDER); 558 if (!skb) 559 return NULL; 560 561 skb_reserve(skb, prepad); 562 skb_put(skb, linear); 563 skb->data_len = len - linear; 564 skb->len += len - linear; 565 566 return skb; 567 } 568 569 /* Neighbour code has some assumptions on HH_DATA_MOD alignment */ 570 #define TAP_RESERVE HH_DATA_OFF(ETH_HLEN) 571 572 /* Get packet from user space buffer */ 573 static ssize_t tap_get_user(struct tap_queue *q, void *msg_control, 574 struct iov_iter *from, int noblock) 575 { 576 int good_linear = SKB_MAX_HEAD(TAP_RESERVE); 577 struct sk_buff *skb; 578 struct tap_dev *tap; 579 unsigned long total_len = iov_iter_count(from); 580 unsigned long len = total_len; 581 int err; 582 struct virtio_net_hdr vnet_hdr = { 0 }; 583 int vnet_hdr_len = 0; 584 int hdr_len = 0; 585 int copylen = 0; 586 int depth; 587 bool zerocopy = false; 588 size_t linear; 589 enum skb_drop_reason drop_reason; 590 591 if (q->flags & IFF_VNET_HDR) { 592 vnet_hdr_len = READ_ONCE(q->vnet_hdr_sz); 593 594 hdr_len = tun_vnet_hdr_get(vnet_hdr_len, q->flags, from, &vnet_hdr); 595 if (hdr_len < 0) { 596 err = hdr_len; 597 goto err; 598 } 599 600 len -= vnet_hdr_len; 601 } 602 603 err = -EINVAL; 604 if (unlikely(len < ETH_HLEN)) 605 goto err; 606 607 if (msg_control && sock_flag(&q->sk, SOCK_ZEROCOPY)) { 608 struct iov_iter i; 609 610 copylen = clamp(hdr_len ?: GOODCOPY_LEN, ETH_HLEN, good_linear); 611 linear = copylen; 612 i = *from; 613 iov_iter_advance(&i, copylen); 614 if (iov_iter_npages(&i, INT_MAX) <= MAX_SKB_FRAGS) 615 zerocopy = true; 616 } 617 618 if (!zerocopy) { 619 copylen = len; 620 linear = clamp(hdr_len, ETH_HLEN, good_linear); 621 } 622 623 skb = tap_alloc_skb(&q->sk, TAP_RESERVE, copylen, 624 linear, noblock, &err); 625 if (!skb) 626 goto err; 627 628 if (zerocopy) 629 err = zerocopy_sg_from_iter(skb, from); 630 else 631 err = skb_copy_datagram_from_iter(skb, 0, from, len); 632 633 if (err) { 634 drop_reason = SKB_DROP_REASON_SKB_UCOPY_FAULT; 635 goto err_kfree; 636 } 637 638 skb_set_network_header(skb, ETH_HLEN); 639 skb_reset_mac_header(skb); 640 skb->protocol = eth_hdr(skb)->h_proto; 641 642 rcu_read_lock(); 643 tap = rcu_dereference(q->tap); 644 if (!tap) { 645 kfree_skb(skb); 646 rcu_read_unlock(); 647 return total_len; 648 } 649 skb->dev = tap->dev; 650 651 if (vnet_hdr_len) { 652 err = tun_vnet_hdr_to_skb(q->flags, skb, &vnet_hdr); 653 if (err) { 654 rcu_read_unlock(); 655 drop_reason = SKB_DROP_REASON_DEV_HDR; 656 goto err_kfree; 657 } 658 } 659 660 skb_probe_transport_header(skb); 661 662 /* Move network header to the right position for VLAN tagged packets */ 663 if (eth_type_vlan(skb->protocol) && 664 vlan_get_protocol_and_depth(skb, skb->protocol, &depth) != 0) 665 skb_set_network_header(skb, depth); 666 667 /* copy skb_ubuf_info for callback when skb has no error */ 668 if (zerocopy) { 669 skb_zcopy_init(skb, msg_control); 670 } else if (msg_control) { 671 struct ubuf_info *uarg = msg_control; 672 uarg->ops->complete(NULL, uarg, false); 673 } 674 675 dev_queue_xmit(skb); 676 rcu_read_unlock(); 677 return total_len; 678 679 err_kfree: 680 kfree_skb_reason(skb, drop_reason); 681 682 err: 683 rcu_read_lock(); 684 tap = rcu_dereference(q->tap); 685 if (tap && tap->count_tx_dropped) 686 tap->count_tx_dropped(tap); 687 rcu_read_unlock(); 688 689 return err; 690 } 691 692 static ssize_t tap_write_iter(struct kiocb *iocb, struct iov_iter *from) 693 { 694 struct file *file = iocb->ki_filp; 695 struct tap_queue *q = file->private_data; 696 int noblock = 0; 697 698 if ((file->f_flags & O_NONBLOCK) || (iocb->ki_flags & IOCB_NOWAIT)) 699 noblock = 1; 700 701 return tap_get_user(q, NULL, from, noblock); 702 } 703 704 /* Put packet to the user space buffer */ 705 static ssize_t tap_put_user(struct tap_queue *q, 706 const struct sk_buff *skb, 707 struct iov_iter *iter) 708 { 709 int ret; 710 int vnet_hdr_len = 0; 711 int vlan_offset = 0; 712 int total; 713 714 if (q->flags & IFF_VNET_HDR) { 715 struct virtio_net_hdr vnet_hdr; 716 717 vnet_hdr_len = READ_ONCE(q->vnet_hdr_sz); 718 719 ret = tun_vnet_hdr_from_skb(q->flags, NULL, skb, &vnet_hdr); 720 if (ret) 721 return ret; 722 723 ret = tun_vnet_hdr_put(vnet_hdr_len, iter, &vnet_hdr); 724 if (ret) 725 return ret; 726 } 727 total = vnet_hdr_len; 728 total += skb->len; 729 730 if (skb_vlan_tag_present(skb)) { 731 struct { 732 __be16 h_vlan_proto; 733 __be16 h_vlan_TCI; 734 } veth; 735 veth.h_vlan_proto = skb->vlan_proto; 736 veth.h_vlan_TCI = htons(skb_vlan_tag_get(skb)); 737 738 vlan_offset = offsetof(struct vlan_ethhdr, h_vlan_proto); 739 total += VLAN_HLEN; 740 741 ret = skb_copy_datagram_iter(skb, 0, iter, vlan_offset); 742 if (ret || !iov_iter_count(iter)) 743 goto done; 744 745 ret = copy_to_iter(&veth, sizeof(veth), iter); 746 if (ret != sizeof(veth) || !iov_iter_count(iter)) 747 goto done; 748 } 749 750 ret = skb_copy_datagram_iter(skb, vlan_offset, iter, 751 skb->len - vlan_offset); 752 753 done: 754 return ret ? ret : total; 755 } 756 757 static ssize_t tap_do_read(struct tap_queue *q, 758 struct iov_iter *to, 759 int noblock, struct sk_buff *skb) 760 { 761 DEFINE_WAIT(wait); 762 ssize_t ret = 0; 763 764 if (!iov_iter_count(to)) { 765 kfree_skb(skb); 766 return 0; 767 } 768 769 if (skb) 770 goto put; 771 772 while (1) { 773 if (!noblock) 774 prepare_to_wait(sk_sleep(&q->sk), &wait, 775 TASK_INTERRUPTIBLE); 776 777 /* Read frames from the queue */ 778 skb = ptr_ring_consume(&q->ring); 779 if (skb) 780 break; 781 if (noblock) { 782 ret = -EAGAIN; 783 break; 784 } 785 if (signal_pending(current)) { 786 ret = -ERESTARTSYS; 787 break; 788 } 789 /* Nothing to read, let's sleep */ 790 schedule(); 791 } 792 if (!noblock) 793 finish_wait(sk_sleep(&q->sk), &wait); 794 795 put: 796 if (skb) { 797 ret = tap_put_user(q, skb, to); 798 if (unlikely(ret < 0)) 799 kfree_skb(skb); 800 else 801 consume_skb(skb); 802 } 803 return ret; 804 } 805 806 static ssize_t tap_read_iter(struct kiocb *iocb, struct iov_iter *to) 807 { 808 struct file *file = iocb->ki_filp; 809 struct tap_queue *q = file->private_data; 810 ssize_t len = iov_iter_count(to), ret; 811 int noblock = 0; 812 813 if ((file->f_flags & O_NONBLOCK) || (iocb->ki_flags & IOCB_NOWAIT)) 814 noblock = 1; 815 816 ret = tap_do_read(q, to, noblock, NULL); 817 ret = min_t(ssize_t, ret, len); 818 if (ret > 0) 819 iocb->ki_pos = ret; 820 return ret; 821 } 822 823 static struct tap_dev *tap_get_tap_dev(struct tap_queue *q) 824 { 825 struct tap_dev *tap; 826 827 ASSERT_RTNL(); 828 tap = rtnl_dereference(q->tap); 829 if (tap) 830 dev_hold(tap->dev); 831 832 return tap; 833 } 834 835 static void tap_put_tap_dev(struct tap_dev *tap) 836 { 837 dev_put(tap->dev); 838 } 839 840 static int tap_ioctl_set_queue(struct file *file, unsigned int flags) 841 { 842 struct tap_queue *q = file->private_data; 843 struct tap_dev *tap; 844 int ret; 845 846 tap = tap_get_tap_dev(q); 847 if (!tap) 848 return -EINVAL; 849 850 if (flags & IFF_ATTACH_QUEUE) 851 ret = tap_enable_queue(tap, file, q); 852 else if (flags & IFF_DETACH_QUEUE) 853 ret = tap_disable_queue(q); 854 else 855 ret = -EINVAL; 856 857 tap_put_tap_dev(tap); 858 return ret; 859 } 860 861 static int set_offload(struct tap_queue *q, unsigned long arg) 862 { 863 struct tap_dev *tap; 864 netdev_features_t features; 865 netdev_features_t feature_mask = 0; 866 867 tap = rtnl_dereference(q->tap); 868 if (!tap) 869 return -ENOLINK; 870 871 features = tap->dev->features; 872 873 if (arg & TUN_F_CSUM) { 874 feature_mask = NETIF_F_HW_CSUM; 875 876 if (arg & (TUN_F_TSO4 | TUN_F_TSO6)) { 877 if (arg & TUN_F_TSO_ECN) 878 feature_mask |= NETIF_F_TSO_ECN; 879 if (arg & TUN_F_TSO4) 880 feature_mask |= NETIF_F_TSO; 881 if (arg & TUN_F_TSO6) 882 feature_mask |= NETIF_F_TSO6; 883 } 884 885 /* TODO: for now USO4 and USO6 should work simultaneously */ 886 if ((arg & (TUN_F_USO4 | TUN_F_USO6)) == (TUN_F_USO4 | TUN_F_USO6)) 887 feature_mask |= NETIF_F_GSO_UDP_L4; 888 } 889 890 /* tun/tap driver inverts the usage for TSO offloads, where 891 * setting the TSO bit means that the userspace wants to 892 * accept TSO frames and turning it off means that user space 893 * does not support TSO. 894 * For tap, we have to invert it to mean the same thing. 895 * When user space turns off TSO, we turn off GSO/LRO so that 896 * user-space will not receive TSO frames. 897 */ 898 if (feature_mask & (NETIF_F_TSO | NETIF_F_TSO6 | NETIF_F_GSO_UDP_L4)) 899 features |= RX_OFFLOADS; 900 else 901 features &= ~RX_OFFLOADS; 902 903 /* tap_features are the same as features on tun/tap and 904 * reflect user expectations. 905 */ 906 tap->tap_features = feature_mask; 907 if (tap->update_features) 908 tap->update_features(tap, features); 909 910 return 0; 911 } 912 913 /* 914 * provide compatibility with generic tun/tap interface 915 */ 916 static long tap_ioctl(struct file *file, unsigned int cmd, 917 unsigned long arg) 918 { 919 struct tap_queue *q = file->private_data; 920 struct tap_dev *tap; 921 void __user *argp = (void __user *)arg; 922 struct sockaddr_storage ss = {}; 923 struct ifreq __user *ifr = argp; 924 unsigned int __user *up = argp; 925 unsigned short u; 926 int __user *sp = argp; 927 int s; 928 int ret; 929 930 switch (cmd) { 931 case TUNSETIFF: 932 /* ignore the name, just look at flags */ 933 if (get_user(u, &ifr->ifr_flags)) 934 return -EFAULT; 935 936 /* TUNGETIFF may report IFF_DETACH_QUEUE, ignore it here */ 937 u &= ~IFF_DETACH_QUEUE; 938 939 ret = 0; 940 if ((u & ~TAP_IFFEATURES) != (IFF_NO_PI | IFF_TAP)) 941 ret = -EINVAL; 942 else 943 q->flags = (q->flags & ~TAP_IFFEATURES) | u; 944 945 return ret; 946 947 case TUNGETIFF: 948 rtnl_lock(); 949 tap = tap_get_tap_dev(q); 950 if (!tap) { 951 rtnl_unlock(); 952 return -ENOLINK; 953 } 954 955 ret = 0; 956 u = q->flags; 957 if (!q->enabled) 958 u |= IFF_DETACH_QUEUE; 959 if (copy_to_user(&ifr->ifr_name, tap->dev->name, IFNAMSIZ) || 960 put_user(u, &ifr->ifr_flags)) 961 ret = -EFAULT; 962 tap_put_tap_dev(tap); 963 rtnl_unlock(); 964 return ret; 965 966 case TUNSETQUEUE: 967 if (get_user(u, &ifr->ifr_flags)) 968 return -EFAULT; 969 rtnl_lock(); 970 ret = tap_ioctl_set_queue(file, u); 971 rtnl_unlock(); 972 return ret; 973 974 case TUNGETFEATURES: 975 if (put_user(IFF_TAP | IFF_NO_PI | TAP_IFFEATURES, up)) 976 return -EFAULT; 977 return 0; 978 979 case TUNSETSNDBUF: 980 if (get_user(s, sp)) 981 return -EFAULT; 982 if (s <= 0) 983 return -EINVAL; 984 985 q->sk.sk_sndbuf = s; 986 return 0; 987 988 case TUNSETOFFLOAD: 989 /* let the user check for future flags */ 990 if (arg & ~(TUN_F_CSUM | TUN_F_TSO4 | TUN_F_TSO6 | 991 TUN_F_TSO_ECN | TUN_F_UFO | 992 TUN_F_USO4 | TUN_F_USO6)) 993 return -EINVAL; 994 995 rtnl_lock(); 996 ret = set_offload(q, arg); 997 rtnl_unlock(); 998 return ret; 999 1000 case SIOCGIFHWADDR: 1001 rtnl_lock(); 1002 tap = tap_get_tap_dev(q); 1003 if (!tap) { 1004 rtnl_unlock(); 1005 return -ENOLINK; 1006 } 1007 ret = 0; 1008 netif_get_mac_address((struct sockaddr *)&ss, dev_net(tap->dev), 1009 tap->dev->name); 1010 if (copy_to_user(&ifr->ifr_name, tap->dev->name, IFNAMSIZ) || 1011 copy_to_user(&ifr->ifr_hwaddr, &ss, sizeof(ifr->ifr_hwaddr))) 1012 ret = -EFAULT; 1013 tap_put_tap_dev(tap); 1014 rtnl_unlock(); 1015 return ret; 1016 1017 case SIOCSIFHWADDR: 1018 if (copy_from_user(&ss, &ifr->ifr_hwaddr, sizeof(ifr->ifr_hwaddr))) 1019 return -EFAULT; 1020 rtnl_lock(); 1021 tap = tap_get_tap_dev(q); 1022 if (!tap) { 1023 rtnl_unlock(); 1024 return -ENOLINK; 1025 } 1026 if (tap->dev->addr_len > sizeof(ifr->ifr_hwaddr)) 1027 ret = -EINVAL; 1028 else 1029 ret = dev_set_mac_address_user(tap->dev, &ss, NULL); 1030 tap_put_tap_dev(tap); 1031 rtnl_unlock(); 1032 return ret; 1033 1034 default: 1035 return tun_vnet_ioctl(&q->vnet_hdr_sz, &q->flags, cmd, sp); 1036 } 1037 } 1038 1039 static const struct file_operations tap_fops = { 1040 .owner = THIS_MODULE, 1041 .open = tap_open, 1042 .release = tap_release, 1043 .read_iter = tap_read_iter, 1044 .write_iter = tap_write_iter, 1045 .poll = tap_poll, 1046 .unlocked_ioctl = tap_ioctl, 1047 .compat_ioctl = compat_ptr_ioctl, 1048 }; 1049 1050 static int tap_get_user_xdp(struct tap_queue *q, struct xdp_buff *xdp) 1051 { 1052 struct virtio_net_hdr *gso = xdp->data_hard_start; 1053 int buflen = xdp->frame_sz; 1054 int vnet_hdr_len = 0; 1055 struct tap_dev *tap; 1056 struct sk_buff *skb; 1057 int err, depth; 1058 1059 if (unlikely(xdp->data_end - xdp->data < ETH_HLEN)) { 1060 put_page(virt_to_head_page(xdp->data)); 1061 err = -EINVAL; 1062 goto err; 1063 } 1064 1065 if (q->flags & IFF_VNET_HDR) 1066 vnet_hdr_len = READ_ONCE(q->vnet_hdr_sz); 1067 1068 skb = build_skb(xdp->data_hard_start, buflen); 1069 if (!skb) { 1070 put_page(virt_to_head_page(xdp->data)); 1071 err = -ENOMEM; 1072 goto err; 1073 } 1074 1075 skb_reserve(skb, xdp->data - xdp->data_hard_start); 1076 skb_put(skb, xdp->data_end - xdp->data); 1077 1078 skb_set_network_header(skb, ETH_HLEN); 1079 skb_reset_mac_header(skb); 1080 skb->protocol = eth_hdr(skb)->h_proto; 1081 1082 rcu_read_lock(); 1083 tap = rcu_dereference(q->tap); 1084 if (!tap) { 1085 kfree_skb(skb); 1086 rcu_read_unlock(); 1087 return 0; 1088 } 1089 skb->dev = tap->dev; 1090 1091 if (vnet_hdr_len) { 1092 err = tun_vnet_hdr_to_skb(q->flags, skb, gso); 1093 if (err) { 1094 rcu_read_unlock(); 1095 goto err_kfree; 1096 } 1097 } 1098 1099 skb_probe_transport_header(skb); 1100 1101 /* Move network header to the right position for VLAN tagged packets */ 1102 if (eth_type_vlan(skb->protocol) && 1103 vlan_get_protocol_and_depth(skb, skb->protocol, &depth) != 0) 1104 skb_set_network_header(skb, depth); 1105 1106 dev_queue_xmit(skb); 1107 rcu_read_unlock(); 1108 1109 return 0; 1110 1111 err_kfree: 1112 kfree_skb(skb); 1113 err: 1114 rcu_read_lock(); 1115 tap = rcu_dereference(q->tap); 1116 if (tap && tap->count_tx_dropped) 1117 tap->count_tx_dropped(tap); 1118 rcu_read_unlock(); 1119 return err; 1120 } 1121 1122 static int tap_sendmsg(struct socket *sock, struct msghdr *m, 1123 size_t total_len) 1124 { 1125 struct tap_queue *q = container_of(sock, struct tap_queue, sock); 1126 struct tun_msg_ctl *ctl = m->msg_control; 1127 struct xdp_buff *xdp; 1128 int i; 1129 1130 if (m->msg_controllen == sizeof(struct tun_msg_ctl) && 1131 ctl && ctl->type == TUN_MSG_PTR) { 1132 for (i = 0; i < ctl->num; i++) { 1133 xdp = &((struct xdp_buff *)ctl->ptr)[i]; 1134 tap_get_user_xdp(q, xdp); 1135 } 1136 return 0; 1137 } 1138 1139 return tap_get_user(q, ctl ? ctl->ptr : NULL, &m->msg_iter, 1140 m->msg_flags & MSG_DONTWAIT); 1141 } 1142 1143 static int tap_recvmsg(struct socket *sock, struct msghdr *m, 1144 size_t total_len, int flags) 1145 { 1146 struct tap_queue *q = container_of(sock, struct tap_queue, sock); 1147 struct sk_buff *skb = m->msg_control; 1148 int ret; 1149 if (flags & ~(MSG_DONTWAIT|MSG_TRUNC)) { 1150 kfree_skb(skb); 1151 return -EINVAL; 1152 } 1153 ret = tap_do_read(q, &m->msg_iter, flags & MSG_DONTWAIT, skb); 1154 if (ret > total_len) { 1155 m->msg_flags |= MSG_TRUNC; 1156 ret = flags & MSG_TRUNC ? ret : total_len; 1157 } 1158 return ret; 1159 } 1160 1161 static int tap_peek_len(struct socket *sock) 1162 { 1163 struct tap_queue *q = container_of(sock, struct tap_queue, 1164 sock); 1165 return PTR_RING_PEEK_CALL(&q->ring, __skb_array_len_with_tag); 1166 } 1167 1168 /* Ops structure to mimic raw sockets with tun */ 1169 static const struct proto_ops tap_socket_ops = { 1170 .sendmsg = tap_sendmsg, 1171 .recvmsg = tap_recvmsg, 1172 .peek_len = tap_peek_len, 1173 }; 1174 1175 /* Get an underlying socket object from tun file. Returns error unless file is 1176 * attached to a device. The returned object works like a packet socket, it 1177 * can be used for sock_sendmsg/sock_recvmsg. The caller is responsible for 1178 * holding a reference to the file for as long as the socket is in use. */ 1179 struct socket *tap_get_socket(struct file *file) 1180 { 1181 struct tap_queue *q; 1182 if (file->f_op != &tap_fops) 1183 return ERR_PTR(-EINVAL); 1184 q = file->private_data; 1185 if (!q) 1186 return ERR_PTR(-EBADFD); 1187 return &q->sock; 1188 } 1189 EXPORT_SYMBOL_GPL(tap_get_socket); 1190 1191 struct ptr_ring *tap_get_ptr_ring(struct file *file) 1192 { 1193 struct tap_queue *q; 1194 1195 if (file->f_op != &tap_fops) 1196 return ERR_PTR(-EINVAL); 1197 q = file->private_data; 1198 if (!q) 1199 return ERR_PTR(-EBADFD); 1200 return &q->ring; 1201 } 1202 EXPORT_SYMBOL_GPL(tap_get_ptr_ring); 1203 1204 int tap_queue_resize(struct tap_dev *tap) 1205 { 1206 struct net_device *dev = tap->dev; 1207 struct tap_queue *q; 1208 struct ptr_ring **rings; 1209 int n = tap->numqueues; 1210 int ret, i = 0; 1211 1212 rings = kmalloc_objs(*rings, n); 1213 if (!rings) 1214 return -ENOMEM; 1215 1216 list_for_each_entry(q, &tap->queue_list, next) 1217 rings[i++] = &q->ring; 1218 1219 ret = ptr_ring_resize_multiple_bh(rings, n, 1220 dev->tx_queue_len, GFP_KERNEL, 1221 __skb_array_destroy_skb); 1222 1223 kfree(rings); 1224 return ret; 1225 } 1226 EXPORT_SYMBOL_GPL(tap_queue_resize); 1227 1228 static int tap_list_add(dev_t major, const char *device_name) 1229 { 1230 struct major_info *tap_major; 1231 1232 tap_major = kzalloc_obj(*tap_major, GFP_ATOMIC); 1233 if (!tap_major) 1234 return -ENOMEM; 1235 1236 tap_major->major = MAJOR(major); 1237 1238 idr_init(&tap_major->minor_idr); 1239 spin_lock_init(&tap_major->minor_lock); 1240 1241 tap_major->device_name = device_name; 1242 1243 list_add_tail_rcu(&tap_major->next, &major_list); 1244 return 0; 1245 } 1246 1247 int tap_create_cdev(struct cdev *tap_cdev, dev_t *tap_major, 1248 const char *device_name, struct module *module) 1249 { 1250 int err; 1251 1252 err = alloc_chrdev_region(tap_major, 0, TAP_NUM_DEVS, device_name); 1253 if (err) 1254 goto out1; 1255 1256 cdev_init(tap_cdev, &tap_fops); 1257 tap_cdev->owner = module; 1258 err = cdev_add(tap_cdev, *tap_major, TAP_NUM_DEVS); 1259 if (err) 1260 goto out2; 1261 1262 err = tap_list_add(*tap_major, device_name); 1263 if (err) 1264 goto out3; 1265 1266 return 0; 1267 1268 out3: 1269 cdev_del(tap_cdev); 1270 out2: 1271 unregister_chrdev_region(*tap_major, TAP_NUM_DEVS); 1272 out1: 1273 return err; 1274 } 1275 EXPORT_SYMBOL_GPL(tap_create_cdev); 1276 1277 void tap_destroy_cdev(dev_t major, struct cdev *tap_cdev) 1278 { 1279 struct major_info *tap_major, *tmp; 1280 1281 cdev_del(tap_cdev); 1282 unregister_chrdev_region(major, TAP_NUM_DEVS); 1283 list_for_each_entry_safe(tap_major, tmp, &major_list, next) { 1284 if (tap_major->major == MAJOR(major)) { 1285 idr_destroy(&tap_major->minor_idr); 1286 list_del_rcu(&tap_major->next); 1287 kfree_rcu(tap_major, rcu); 1288 } 1289 } 1290 } 1291 EXPORT_SYMBOL_GPL(tap_destroy_cdev); 1292 1293 MODULE_DESCRIPTION("Common library for drivers implementing the TAP interface"); 1294 MODULE_AUTHOR("Arnd Bergmann <arnd@arndb.de>"); 1295 MODULE_AUTHOR("Sainath Grandhi <sainath.grandhi@intel.com>"); 1296 MODULE_LICENSE("GPL"); 1297 MODULE_IMPORT_NS("NETDEV_INTERNAL"); 1298