1 /* 2 * TUN - Universal TUN/TAP device driver. 3 * Copyright (C) 1999-2002 Maxim Krasnyansky <maxk@qualcomm.com> 4 * 5 * This program is free software; you can redistribute it and/or modify 6 * it under the terms of the GNU General Public License as published by 7 * the Free Software Foundation; either version 2 of the License, or 8 * (at your option) any later version. 9 * 10 * This program is distributed in the hope that it will be useful, 11 * but WITHOUT ANY WARRANTY; without even the implied warranty of 12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the 13 * GNU General Public License for more details. 14 * 15 * $Id: tun.c,v 1.15 2002/03/01 02:44:24 maxk Exp $ 16 */ 17 18 /* 19 * Changes: 20 * 21 * Mike Kershaw <dragorn@kismetwireless.net> 2005/08/14 22 * Add TUNSETLINK ioctl to set the link encapsulation 23 * 24 * Mark Smith <markzzzsmith@yahoo.com.au> 25 * Use random_ether_addr() for tap MAC address. 26 * 27 * Harald Roelle <harald.roelle@ifi.lmu.de> 2004/04/20 28 * Fixes in packet dropping, queue length setting and queue wakeup. 29 * Increased default tx queue length. 30 * Added ethtool API. 31 * Minor cleanups 32 * 33 * Daniel Podlejski <underley@underley.eu.org> 34 * Modifications for 2.3.99-pre5 kernel. 35 */ 36 37 #define DRV_NAME "tun" 38 #define DRV_VERSION "1.6" 39 #define DRV_DESCRIPTION "Universal TUN/TAP device driver" 40 #define DRV_COPYRIGHT "(C) 1999-2004 Max Krasnyansky <maxk@qualcomm.com>" 41 42 #include <linux/module.h> 43 #include <linux/errno.h> 44 #include <linux/kernel.h> 45 #include <linux/major.h> 46 #include <linux/slab.h> 47 #include <linux/smp_lock.h> 48 #include <linux/poll.h> 49 #include <linux/fcntl.h> 50 #include <linux/init.h> 51 #include <linux/skbuff.h> 52 #include <linux/netdevice.h> 53 #include <linux/etherdevice.h> 54 #include <linux/miscdevice.h> 55 #include <linux/ethtool.h> 56 #include <linux/rtnetlink.h> 57 #include <linux/if.h> 58 #include <linux/if_arp.h> 59 #include <linux/if_ether.h> 60 #include <linux/if_tun.h> 61 #include <linux/crc32.h> 62 #include <linux/nsproxy.h> 63 #include <linux/virtio_net.h> 64 #include <net/net_namespace.h> 65 #include <net/netns/generic.h> 66 #include <net/rtnetlink.h> 67 #include <net/sock.h> 68 69 #include <asm/system.h> 70 #include <asm/uaccess.h> 71 72 /* Uncomment to enable debugging */ 73 /* #define TUN_DEBUG 1 */ 74 75 #ifdef TUN_DEBUG 76 static int debug; 77 78 #define DBG if(tun->debug)printk 79 #define DBG1 if(debug==2)printk 80 #else 81 #define DBG( a... ) 82 #define DBG1( a... ) 83 #endif 84 85 #define FLT_EXACT_COUNT 8 86 struct tap_filter { 87 unsigned int count; /* Number of addrs. Zero means disabled */ 88 u32 mask[2]; /* Mask of the hashed addrs */ 89 unsigned char addr[FLT_EXACT_COUNT][ETH_ALEN]; 90 }; 91 92 struct tun_file { 93 atomic_t count; 94 struct tun_struct *tun; 95 struct net *net; 96 }; 97 98 struct tun_sock; 99 100 struct tun_struct { 101 struct tun_file *tfile; 102 unsigned int flags; 103 uid_t owner; 104 gid_t group; 105 106 struct sk_buff_head readq; 107 108 struct net_device *dev; 109 struct fasync_struct *fasync; 110 111 struct tap_filter txflt; 112 struct sock *sk; 113 struct socket socket; 114 115 #ifdef TUN_DEBUG 116 int debug; 117 #endif 118 }; 119 120 struct tun_sock { 121 struct sock sk; 122 struct tun_struct *tun; 123 }; 124 125 static inline struct tun_sock *tun_sk(struct sock *sk) 126 { 127 return container_of(sk, struct tun_sock, sk); 128 } 129 130 static int tun_attach(struct tun_struct *tun, struct file *file) 131 { 132 struct tun_file *tfile = file->private_data; 133 const struct cred *cred = current_cred(); 134 int err; 135 136 ASSERT_RTNL(); 137 138 /* Check permissions */ 139 if (((tun->owner != -1 && cred->euid != tun->owner) || 140 (tun->group != -1 && !in_egroup_p(tun->group))) && 141 !capable(CAP_NET_ADMIN)) 142 return -EPERM; 143 144 netif_tx_lock_bh(tun->dev); 145 146 err = -EINVAL; 147 if (tfile->tun) 148 goto out; 149 150 err = -EBUSY; 151 if (tun->tfile) 152 goto out; 153 154 err = 0; 155 tfile->tun = tun; 156 tun->tfile = tfile; 157 dev_hold(tun->dev); 158 sock_hold(tun->sk); 159 atomic_inc(&tfile->count); 160 161 out: 162 netif_tx_unlock_bh(tun->dev); 163 return err; 164 } 165 166 static void __tun_detach(struct tun_struct *tun) 167 { 168 /* Detach from net device */ 169 netif_tx_lock_bh(tun->dev); 170 tun->tfile = NULL; 171 netif_tx_unlock_bh(tun->dev); 172 173 /* Drop read queue */ 174 skb_queue_purge(&tun->readq); 175 176 /* Drop the extra count on the net device */ 177 dev_put(tun->dev); 178 } 179 180 static void tun_detach(struct tun_struct *tun) 181 { 182 rtnl_lock(); 183 __tun_detach(tun); 184 rtnl_unlock(); 185 } 186 187 static struct tun_struct *__tun_get(struct tun_file *tfile) 188 { 189 struct tun_struct *tun = NULL; 190 191 if (atomic_inc_not_zero(&tfile->count)) 192 tun = tfile->tun; 193 194 return tun; 195 } 196 197 static struct tun_struct *tun_get(struct file *file) 198 { 199 return __tun_get(file->private_data); 200 } 201 202 static void tun_put(struct tun_struct *tun) 203 { 204 struct tun_file *tfile = tun->tfile; 205 206 if (atomic_dec_and_test(&tfile->count)) 207 tun_detach(tfile->tun); 208 } 209 210 /* TAP filterting */ 211 static void addr_hash_set(u32 *mask, const u8 *addr) 212 { 213 int n = ether_crc(ETH_ALEN, addr) >> 26; 214 mask[n >> 5] |= (1 << (n & 31)); 215 } 216 217 static unsigned int addr_hash_test(const u32 *mask, const u8 *addr) 218 { 219 int n = ether_crc(ETH_ALEN, addr) >> 26; 220 return mask[n >> 5] & (1 << (n & 31)); 221 } 222 223 static int update_filter(struct tap_filter *filter, void __user *arg) 224 { 225 struct { u8 u[ETH_ALEN]; } *addr; 226 struct tun_filter uf; 227 int err, alen, n, nexact; 228 229 if (copy_from_user(&uf, arg, sizeof(uf))) 230 return -EFAULT; 231 232 if (!uf.count) { 233 /* Disabled */ 234 filter->count = 0; 235 return 0; 236 } 237 238 alen = ETH_ALEN * uf.count; 239 addr = kmalloc(alen, GFP_KERNEL); 240 if (!addr) 241 return -ENOMEM; 242 243 if (copy_from_user(addr, arg + sizeof(uf), alen)) { 244 err = -EFAULT; 245 goto done; 246 } 247 248 /* The filter is updated without holding any locks. Which is 249 * perfectly safe. We disable it first and in the worst 250 * case we'll accept a few undesired packets. */ 251 filter->count = 0; 252 wmb(); 253 254 /* Use first set of addresses as an exact filter */ 255 for (n = 0; n < uf.count && n < FLT_EXACT_COUNT; n++) 256 memcpy(filter->addr[n], addr[n].u, ETH_ALEN); 257 258 nexact = n; 259 260 /* Remaining multicast addresses are hashed, 261 * unicast will leave the filter disabled. */ 262 memset(filter->mask, 0, sizeof(filter->mask)); 263 for (; n < uf.count; n++) { 264 if (!is_multicast_ether_addr(addr[n].u)) { 265 err = 0; /* no filter */ 266 goto done; 267 } 268 addr_hash_set(filter->mask, addr[n].u); 269 } 270 271 /* For ALLMULTI just set the mask to all ones. 272 * This overrides the mask populated above. */ 273 if ((uf.flags & TUN_FLT_ALLMULTI)) 274 memset(filter->mask, ~0, sizeof(filter->mask)); 275 276 /* Now enable the filter */ 277 wmb(); 278 filter->count = nexact; 279 280 /* Return the number of exact filters */ 281 err = nexact; 282 283 done: 284 kfree(addr); 285 return err; 286 } 287 288 /* Returns: 0 - drop, !=0 - accept */ 289 static int run_filter(struct tap_filter *filter, const struct sk_buff *skb) 290 { 291 /* Cannot use eth_hdr(skb) here because skb_mac_hdr() is incorrect 292 * at this point. */ 293 struct ethhdr *eh = (struct ethhdr *) skb->data; 294 int i; 295 296 /* Exact match */ 297 for (i = 0; i < filter->count; i++) 298 if (!compare_ether_addr(eh->h_dest, filter->addr[i])) 299 return 1; 300 301 /* Inexact match (multicast only) */ 302 if (is_multicast_ether_addr(eh->h_dest)) 303 return addr_hash_test(filter->mask, eh->h_dest); 304 305 return 0; 306 } 307 308 /* 309 * Checks whether the packet is accepted or not. 310 * Returns: 0 - drop, !=0 - accept 311 */ 312 static int check_filter(struct tap_filter *filter, const struct sk_buff *skb) 313 { 314 if (!filter->count) 315 return 1; 316 317 return run_filter(filter, skb); 318 } 319 320 /* Network device part of the driver */ 321 322 static const struct ethtool_ops tun_ethtool_ops; 323 324 /* Net device detach from fd. */ 325 static void tun_net_uninit(struct net_device *dev) 326 { 327 struct tun_struct *tun = netdev_priv(dev); 328 struct tun_file *tfile = tun->tfile; 329 330 /* Inform the methods they need to stop using the dev. 331 */ 332 if (tfile) { 333 wake_up_all(&tun->socket.wait); 334 if (atomic_dec_and_test(&tfile->count)) 335 __tun_detach(tun); 336 } 337 } 338 339 static void tun_free_netdev(struct net_device *dev) 340 { 341 struct tun_struct *tun = netdev_priv(dev); 342 343 sock_put(tun->sk); 344 } 345 346 /* Net device open. */ 347 static int tun_net_open(struct net_device *dev) 348 { 349 netif_start_queue(dev); 350 return 0; 351 } 352 353 /* Net device close. */ 354 static int tun_net_close(struct net_device *dev) 355 { 356 netif_stop_queue(dev); 357 return 0; 358 } 359 360 /* Net device start xmit */ 361 static int tun_net_xmit(struct sk_buff *skb, struct net_device *dev) 362 { 363 struct tun_struct *tun = netdev_priv(dev); 364 365 DBG(KERN_INFO "%s: tun_net_xmit %d\n", tun->dev->name, skb->len); 366 367 /* Drop packet if interface is not attached */ 368 if (!tun->tfile) 369 goto drop; 370 371 /* Drop if the filter does not like it. 372 * This is a noop if the filter is disabled. 373 * Filter can be enabled only for the TAP devices. */ 374 if (!check_filter(&tun->txflt, skb)) 375 goto drop; 376 377 if (skb_queue_len(&tun->readq) >= dev->tx_queue_len) { 378 if (!(tun->flags & TUN_ONE_QUEUE)) { 379 /* Normal queueing mode. */ 380 /* Packet scheduler handles dropping of further packets. */ 381 netif_stop_queue(dev); 382 383 /* We won't see all dropped packets individually, so overrun 384 * error is more appropriate. */ 385 dev->stats.tx_fifo_errors++; 386 } else { 387 /* Single queue mode. 388 * Driver handles dropping of all packets itself. */ 389 goto drop; 390 } 391 } 392 393 /* Enqueue packet */ 394 skb_queue_tail(&tun->readq, skb); 395 dev->trans_start = jiffies; 396 397 /* Notify and wake up reader process */ 398 if (tun->flags & TUN_FASYNC) 399 kill_fasync(&tun->fasync, SIGIO, POLL_IN); 400 wake_up_interruptible(&tun->socket.wait); 401 return 0; 402 403 drop: 404 dev->stats.tx_dropped++; 405 kfree_skb(skb); 406 return 0; 407 } 408 409 static void tun_net_mclist(struct net_device *dev) 410 { 411 /* 412 * This callback is supposed to deal with mc filter in 413 * _rx_ path and has nothing to do with the _tx_ path. 414 * In rx path we always accept everything userspace gives us. 415 */ 416 return; 417 } 418 419 #define MIN_MTU 68 420 #define MAX_MTU 65535 421 422 static int 423 tun_net_change_mtu(struct net_device *dev, int new_mtu) 424 { 425 if (new_mtu < MIN_MTU || new_mtu + dev->hard_header_len > MAX_MTU) 426 return -EINVAL; 427 dev->mtu = new_mtu; 428 return 0; 429 } 430 431 static const struct net_device_ops tun_netdev_ops = { 432 .ndo_uninit = tun_net_uninit, 433 .ndo_open = tun_net_open, 434 .ndo_stop = tun_net_close, 435 .ndo_start_xmit = tun_net_xmit, 436 .ndo_change_mtu = tun_net_change_mtu, 437 }; 438 439 static const struct net_device_ops tap_netdev_ops = { 440 .ndo_uninit = tun_net_uninit, 441 .ndo_open = tun_net_open, 442 .ndo_stop = tun_net_close, 443 .ndo_start_xmit = tun_net_xmit, 444 .ndo_change_mtu = tun_net_change_mtu, 445 .ndo_set_multicast_list = tun_net_mclist, 446 .ndo_set_mac_address = eth_mac_addr, 447 .ndo_validate_addr = eth_validate_addr, 448 }; 449 450 /* Initialize net device. */ 451 static void tun_net_init(struct net_device *dev) 452 { 453 struct tun_struct *tun = netdev_priv(dev); 454 455 switch (tun->flags & TUN_TYPE_MASK) { 456 case TUN_TUN_DEV: 457 dev->netdev_ops = &tun_netdev_ops; 458 459 /* Point-to-Point TUN Device */ 460 dev->hard_header_len = 0; 461 dev->addr_len = 0; 462 dev->mtu = 1500; 463 464 /* Zero header length */ 465 dev->type = ARPHRD_NONE; 466 dev->flags = IFF_POINTOPOINT | IFF_NOARP | IFF_MULTICAST; 467 dev->tx_queue_len = TUN_READQ_SIZE; /* We prefer our own queue length */ 468 break; 469 470 case TUN_TAP_DEV: 471 dev->netdev_ops = &tap_netdev_ops; 472 /* Ethernet TAP Device */ 473 ether_setup(dev); 474 475 random_ether_addr(dev->dev_addr); 476 477 dev->tx_queue_len = TUN_READQ_SIZE; /* We prefer our own queue length */ 478 break; 479 } 480 } 481 482 /* Character device part */ 483 484 /* Poll */ 485 static unsigned int tun_chr_poll(struct file *file, poll_table * wait) 486 { 487 struct tun_file *tfile = file->private_data; 488 struct tun_struct *tun = __tun_get(tfile); 489 struct sock *sk = tun->sk; 490 unsigned int mask = 0; 491 492 if (!tun) 493 return POLLERR; 494 495 DBG(KERN_INFO "%s: tun_chr_poll\n", tun->dev->name); 496 497 poll_wait(file, &tun->socket.wait, wait); 498 499 if (!skb_queue_empty(&tun->readq)) 500 mask |= POLLIN | POLLRDNORM; 501 502 if (sock_writeable(sk) || 503 (!test_and_set_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags) && 504 sock_writeable(sk))) 505 mask |= POLLOUT | POLLWRNORM; 506 507 if (tun->dev->reg_state != NETREG_REGISTERED) 508 mask = POLLERR; 509 510 tun_put(tun); 511 return mask; 512 } 513 514 /* prepad is the amount to reserve at front. len is length after that. 515 * linear is a hint as to how much to copy (usually headers). */ 516 static inline struct sk_buff *tun_alloc_skb(struct tun_struct *tun, 517 size_t prepad, size_t len, 518 size_t linear, int noblock) 519 { 520 struct sock *sk = tun->sk; 521 struct sk_buff *skb; 522 int err; 523 524 /* Under a page? Don't bother with paged skb. */ 525 if (prepad + len < PAGE_SIZE || !linear) 526 linear = len; 527 528 skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock, 529 &err); 530 if (!skb) 531 return ERR_PTR(err); 532 533 skb_reserve(skb, prepad); 534 skb_put(skb, linear); 535 skb->data_len = len - linear; 536 skb->len += len - linear; 537 538 return skb; 539 } 540 541 /* Get packet from user space buffer */ 542 static __inline__ ssize_t tun_get_user(struct tun_struct *tun, 543 struct iovec *iv, size_t count, 544 int noblock) 545 { 546 struct tun_pi pi = { 0, cpu_to_be16(ETH_P_IP) }; 547 struct sk_buff *skb; 548 size_t len = count, align = 0; 549 struct virtio_net_hdr gso = { 0 }; 550 551 if (!(tun->flags & TUN_NO_PI)) { 552 if ((len -= sizeof(pi)) > count) 553 return -EINVAL; 554 555 if(memcpy_fromiovec((void *)&pi, iv, sizeof(pi))) 556 return -EFAULT; 557 } 558 559 if (tun->flags & TUN_VNET_HDR) { 560 if ((len -= sizeof(gso)) > count) 561 return -EINVAL; 562 563 if (memcpy_fromiovec((void *)&gso, iv, sizeof(gso))) 564 return -EFAULT; 565 566 if (gso.hdr_len > len) 567 return -EINVAL; 568 } 569 570 if ((tun->flags & TUN_TYPE_MASK) == TUN_TAP_DEV) { 571 align = NET_IP_ALIGN; 572 if (unlikely(len < ETH_HLEN || 573 (gso.hdr_len && gso.hdr_len < ETH_HLEN))) 574 return -EINVAL; 575 } 576 577 skb = tun_alloc_skb(tun, align, len, gso.hdr_len, noblock); 578 if (IS_ERR(skb)) { 579 if (PTR_ERR(skb) != -EAGAIN) 580 tun->dev->stats.rx_dropped++; 581 return PTR_ERR(skb); 582 } 583 584 if (skb_copy_datagram_from_iovec(skb, 0, iv, len)) { 585 tun->dev->stats.rx_dropped++; 586 kfree_skb(skb); 587 return -EFAULT; 588 } 589 590 if (gso.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) { 591 if (!skb_partial_csum_set(skb, gso.csum_start, 592 gso.csum_offset)) { 593 tun->dev->stats.rx_frame_errors++; 594 kfree_skb(skb); 595 return -EINVAL; 596 } 597 } else if (tun->flags & TUN_NOCHECKSUM) 598 skb->ip_summed = CHECKSUM_UNNECESSARY; 599 600 switch (tun->flags & TUN_TYPE_MASK) { 601 case TUN_TUN_DEV: 602 if (tun->flags & TUN_NO_PI) { 603 switch (skb->data[0] & 0xf0) { 604 case 0x40: 605 pi.proto = htons(ETH_P_IP); 606 break; 607 case 0x60: 608 pi.proto = htons(ETH_P_IPV6); 609 break; 610 default: 611 tun->dev->stats.rx_dropped++; 612 kfree_skb(skb); 613 return -EINVAL; 614 } 615 } 616 617 skb_reset_mac_header(skb); 618 skb->protocol = pi.proto; 619 skb->dev = tun->dev; 620 break; 621 case TUN_TAP_DEV: 622 skb->protocol = eth_type_trans(skb, tun->dev); 623 break; 624 }; 625 626 if (gso.gso_type != VIRTIO_NET_HDR_GSO_NONE) { 627 pr_debug("GSO!\n"); 628 switch (gso.gso_type & ~VIRTIO_NET_HDR_GSO_ECN) { 629 case VIRTIO_NET_HDR_GSO_TCPV4: 630 skb_shinfo(skb)->gso_type = SKB_GSO_TCPV4; 631 break; 632 case VIRTIO_NET_HDR_GSO_TCPV6: 633 skb_shinfo(skb)->gso_type = SKB_GSO_TCPV6; 634 break; 635 default: 636 tun->dev->stats.rx_frame_errors++; 637 kfree_skb(skb); 638 return -EINVAL; 639 } 640 641 if (gso.gso_type & VIRTIO_NET_HDR_GSO_ECN) 642 skb_shinfo(skb)->gso_type |= SKB_GSO_TCP_ECN; 643 644 skb_shinfo(skb)->gso_size = gso.gso_size; 645 if (skb_shinfo(skb)->gso_size == 0) { 646 tun->dev->stats.rx_frame_errors++; 647 kfree_skb(skb); 648 return -EINVAL; 649 } 650 651 /* Header must be checked, and gso_segs computed. */ 652 skb_shinfo(skb)->gso_type |= SKB_GSO_DODGY; 653 skb_shinfo(skb)->gso_segs = 0; 654 } 655 656 netif_rx_ni(skb); 657 658 tun->dev->stats.rx_packets++; 659 tun->dev->stats.rx_bytes += len; 660 661 return count; 662 } 663 664 static ssize_t tun_chr_aio_write(struct kiocb *iocb, const struct iovec *iv, 665 unsigned long count, loff_t pos) 666 { 667 struct file *file = iocb->ki_filp; 668 struct tun_struct *tun = tun_get(file); 669 ssize_t result; 670 671 if (!tun) 672 return -EBADFD; 673 674 DBG(KERN_INFO "%s: tun_chr_write %ld\n", tun->dev->name, count); 675 676 result = tun_get_user(tun, (struct iovec *)iv, iov_length(iv, count), 677 file->f_flags & O_NONBLOCK); 678 679 tun_put(tun); 680 return result; 681 } 682 683 /* Put packet to the user space buffer */ 684 static __inline__ ssize_t tun_put_user(struct tun_struct *tun, 685 struct sk_buff *skb, 686 struct iovec *iv, int len) 687 { 688 struct tun_pi pi = { 0, skb->protocol }; 689 ssize_t total = 0; 690 691 if (!(tun->flags & TUN_NO_PI)) { 692 if ((len -= sizeof(pi)) < 0) 693 return -EINVAL; 694 695 if (len < skb->len) { 696 /* Packet will be striped */ 697 pi.flags |= TUN_PKT_STRIP; 698 } 699 700 if (memcpy_toiovec(iv, (void *) &pi, sizeof(pi))) 701 return -EFAULT; 702 total += sizeof(pi); 703 } 704 705 if (tun->flags & TUN_VNET_HDR) { 706 struct virtio_net_hdr gso = { 0 }; /* no info leak */ 707 if ((len -= sizeof(gso)) < 0) 708 return -EINVAL; 709 710 if (skb_is_gso(skb)) { 711 struct skb_shared_info *sinfo = skb_shinfo(skb); 712 713 /* This is a hint as to how much should be linear. */ 714 gso.hdr_len = skb_headlen(skb); 715 gso.gso_size = sinfo->gso_size; 716 if (sinfo->gso_type & SKB_GSO_TCPV4) 717 gso.gso_type = VIRTIO_NET_HDR_GSO_TCPV4; 718 else if (sinfo->gso_type & SKB_GSO_TCPV6) 719 gso.gso_type = VIRTIO_NET_HDR_GSO_TCPV6; 720 else 721 BUG(); 722 if (sinfo->gso_type & SKB_GSO_TCP_ECN) 723 gso.gso_type |= VIRTIO_NET_HDR_GSO_ECN; 724 } else 725 gso.gso_type = VIRTIO_NET_HDR_GSO_NONE; 726 727 if (skb->ip_summed == CHECKSUM_PARTIAL) { 728 gso.flags = VIRTIO_NET_HDR_F_NEEDS_CSUM; 729 gso.csum_start = skb->csum_start - skb_headroom(skb); 730 gso.csum_offset = skb->csum_offset; 731 } /* else everything is zero */ 732 733 if (unlikely(memcpy_toiovec(iv, (void *)&gso, sizeof(gso)))) 734 return -EFAULT; 735 total += sizeof(gso); 736 } 737 738 len = min_t(int, skb->len, len); 739 740 skb_copy_datagram_iovec(skb, 0, iv, len); 741 total += len; 742 743 tun->dev->stats.tx_packets++; 744 tun->dev->stats.tx_bytes += len; 745 746 return total; 747 } 748 749 static ssize_t tun_chr_aio_read(struct kiocb *iocb, const struct iovec *iv, 750 unsigned long count, loff_t pos) 751 { 752 struct file *file = iocb->ki_filp; 753 struct tun_file *tfile = file->private_data; 754 struct tun_struct *tun = __tun_get(tfile); 755 DECLARE_WAITQUEUE(wait, current); 756 struct sk_buff *skb; 757 ssize_t len, ret = 0; 758 759 if (!tun) 760 return -EBADFD; 761 762 DBG(KERN_INFO "%s: tun_chr_read\n", tun->dev->name); 763 764 len = iov_length(iv, count); 765 if (len < 0) { 766 ret = -EINVAL; 767 goto out; 768 } 769 770 add_wait_queue(&tun->socket.wait, &wait); 771 while (len) { 772 current->state = TASK_INTERRUPTIBLE; 773 774 /* Read frames from the queue */ 775 if (!(skb=skb_dequeue(&tun->readq))) { 776 if (file->f_flags & O_NONBLOCK) { 777 ret = -EAGAIN; 778 break; 779 } 780 if (signal_pending(current)) { 781 ret = -ERESTARTSYS; 782 break; 783 } 784 if (tun->dev->reg_state != NETREG_REGISTERED) { 785 ret = -EIO; 786 break; 787 } 788 789 /* Nothing to read, let's sleep */ 790 schedule(); 791 continue; 792 } 793 netif_wake_queue(tun->dev); 794 795 ret = tun_put_user(tun, skb, (struct iovec *) iv, len); 796 kfree_skb(skb); 797 break; 798 } 799 800 current->state = TASK_RUNNING; 801 remove_wait_queue(&tun->socket.wait, &wait); 802 803 out: 804 tun_put(tun); 805 return ret; 806 } 807 808 static void tun_setup(struct net_device *dev) 809 { 810 struct tun_struct *tun = netdev_priv(dev); 811 812 skb_queue_head_init(&tun->readq); 813 814 tun->owner = -1; 815 tun->group = -1; 816 817 dev->ethtool_ops = &tun_ethtool_ops; 818 dev->destructor = tun_free_netdev; 819 } 820 821 /* Trivial set of netlink ops to allow deleting tun or tap 822 * device with netlink. 823 */ 824 static int tun_validate(struct nlattr *tb[], struct nlattr *data[]) 825 { 826 return -EINVAL; 827 } 828 829 static struct rtnl_link_ops tun_link_ops __read_mostly = { 830 .kind = DRV_NAME, 831 .priv_size = sizeof(struct tun_struct), 832 .setup = tun_setup, 833 .validate = tun_validate, 834 }; 835 836 static void tun_sock_write_space(struct sock *sk) 837 { 838 struct tun_struct *tun; 839 840 if (!sock_writeable(sk)) 841 return; 842 843 if (sk->sk_sleep && waitqueue_active(sk->sk_sleep)) 844 wake_up_interruptible_sync(sk->sk_sleep); 845 846 if (!test_and_clear_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags)) 847 return; 848 849 tun = container_of(sk, struct tun_sock, sk)->tun; 850 kill_fasync(&tun->fasync, SIGIO, POLL_OUT); 851 } 852 853 static void tun_sock_destruct(struct sock *sk) 854 { 855 free_netdev(container_of(sk, struct tun_sock, sk)->tun->dev); 856 } 857 858 static struct proto tun_proto = { 859 .name = "tun", 860 .owner = THIS_MODULE, 861 .obj_size = sizeof(struct tun_sock), 862 }; 863 864 static int tun_set_iff(struct net *net, struct file *file, struct ifreq *ifr) 865 { 866 struct sock *sk; 867 struct tun_struct *tun; 868 struct net_device *dev; 869 int err; 870 871 dev = __dev_get_by_name(net, ifr->ifr_name); 872 if (dev) { 873 if ((ifr->ifr_flags & IFF_TUN) && dev->netdev_ops == &tun_netdev_ops) 874 tun = netdev_priv(dev); 875 else if ((ifr->ifr_flags & IFF_TAP) && dev->netdev_ops == &tap_netdev_ops) 876 tun = netdev_priv(dev); 877 else 878 return -EINVAL; 879 880 err = tun_attach(tun, file); 881 if (err < 0) 882 return err; 883 } 884 else { 885 char *name; 886 unsigned long flags = 0; 887 888 err = -EINVAL; 889 890 if (!capable(CAP_NET_ADMIN)) 891 return -EPERM; 892 893 /* Set dev type */ 894 if (ifr->ifr_flags & IFF_TUN) { 895 /* TUN device */ 896 flags |= TUN_TUN_DEV; 897 name = "tun%d"; 898 } else if (ifr->ifr_flags & IFF_TAP) { 899 /* TAP device */ 900 flags |= TUN_TAP_DEV; 901 name = "tap%d"; 902 } else 903 goto failed; 904 905 if (*ifr->ifr_name) 906 name = ifr->ifr_name; 907 908 dev = alloc_netdev(sizeof(struct tun_struct), name, 909 tun_setup); 910 if (!dev) 911 return -ENOMEM; 912 913 dev_net_set(dev, net); 914 dev->rtnl_link_ops = &tun_link_ops; 915 916 tun = netdev_priv(dev); 917 tun->dev = dev; 918 tun->flags = flags; 919 tun->txflt.count = 0; 920 921 err = -ENOMEM; 922 sk = sk_alloc(net, AF_UNSPEC, GFP_KERNEL, &tun_proto); 923 if (!sk) 924 goto err_free_dev; 925 926 init_waitqueue_head(&tun->socket.wait); 927 sock_init_data(&tun->socket, sk); 928 sk->sk_write_space = tun_sock_write_space; 929 sk->sk_sndbuf = INT_MAX; 930 931 tun->sk = sk; 932 container_of(sk, struct tun_sock, sk)->tun = tun; 933 934 tun_net_init(dev); 935 936 if (strchr(dev->name, '%')) { 937 err = dev_alloc_name(dev, dev->name); 938 if (err < 0) 939 goto err_free_sk; 940 } 941 942 err = -EINVAL; 943 err = register_netdevice(tun->dev); 944 if (err < 0) 945 goto err_free_sk; 946 947 sk->sk_destruct = tun_sock_destruct; 948 949 err = tun_attach(tun, file); 950 if (err < 0) 951 goto failed; 952 } 953 954 DBG(KERN_INFO "%s: tun_set_iff\n", tun->dev->name); 955 956 if (ifr->ifr_flags & IFF_NO_PI) 957 tun->flags |= TUN_NO_PI; 958 else 959 tun->flags &= ~TUN_NO_PI; 960 961 if (ifr->ifr_flags & IFF_ONE_QUEUE) 962 tun->flags |= TUN_ONE_QUEUE; 963 else 964 tun->flags &= ~TUN_ONE_QUEUE; 965 966 if (ifr->ifr_flags & IFF_VNET_HDR) 967 tun->flags |= TUN_VNET_HDR; 968 else 969 tun->flags &= ~TUN_VNET_HDR; 970 971 /* Make sure persistent devices do not get stuck in 972 * xoff state. 973 */ 974 if (netif_running(tun->dev)) 975 netif_wake_queue(tun->dev); 976 977 strcpy(ifr->ifr_name, tun->dev->name); 978 return 0; 979 980 err_free_sk: 981 sock_put(sk); 982 err_free_dev: 983 free_netdev(dev); 984 failed: 985 return err; 986 } 987 988 static int tun_get_iff(struct net *net, struct file *file, struct ifreq *ifr) 989 { 990 struct tun_struct *tun = tun_get(file); 991 992 if (!tun) 993 return -EBADFD; 994 995 DBG(KERN_INFO "%s: tun_get_iff\n", tun->dev->name); 996 997 strcpy(ifr->ifr_name, tun->dev->name); 998 999 ifr->ifr_flags = 0; 1000 1001 if (ifr->ifr_flags & TUN_TUN_DEV) 1002 ifr->ifr_flags |= IFF_TUN; 1003 else 1004 ifr->ifr_flags |= IFF_TAP; 1005 1006 if (tun->flags & TUN_NO_PI) 1007 ifr->ifr_flags |= IFF_NO_PI; 1008 1009 if (tun->flags & TUN_ONE_QUEUE) 1010 ifr->ifr_flags |= IFF_ONE_QUEUE; 1011 1012 if (tun->flags & TUN_VNET_HDR) 1013 ifr->ifr_flags |= IFF_VNET_HDR; 1014 1015 tun_put(tun); 1016 return 0; 1017 } 1018 1019 /* This is like a cut-down ethtool ops, except done via tun fd so no 1020 * privs required. */ 1021 static int set_offload(struct net_device *dev, unsigned long arg) 1022 { 1023 unsigned int old_features, features; 1024 1025 old_features = dev->features; 1026 /* Unset features, set them as we chew on the arg. */ 1027 features = (old_features & ~(NETIF_F_HW_CSUM|NETIF_F_SG|NETIF_F_FRAGLIST 1028 |NETIF_F_TSO_ECN|NETIF_F_TSO|NETIF_F_TSO6)); 1029 1030 if (arg & TUN_F_CSUM) { 1031 features |= NETIF_F_HW_CSUM|NETIF_F_SG|NETIF_F_FRAGLIST; 1032 arg &= ~TUN_F_CSUM; 1033 1034 if (arg & (TUN_F_TSO4|TUN_F_TSO6)) { 1035 if (arg & TUN_F_TSO_ECN) { 1036 features |= NETIF_F_TSO_ECN; 1037 arg &= ~TUN_F_TSO_ECN; 1038 } 1039 if (arg & TUN_F_TSO4) 1040 features |= NETIF_F_TSO; 1041 if (arg & TUN_F_TSO6) 1042 features |= NETIF_F_TSO6; 1043 arg &= ~(TUN_F_TSO4|TUN_F_TSO6); 1044 } 1045 } 1046 1047 /* This gives the user a way to test for new features in future by 1048 * trying to set them. */ 1049 if (arg) 1050 return -EINVAL; 1051 1052 dev->features = features; 1053 if (old_features != dev->features) 1054 netdev_features_change(dev); 1055 1056 return 0; 1057 } 1058 1059 static int tun_chr_ioctl(struct inode *inode, struct file *file, 1060 unsigned int cmd, unsigned long arg) 1061 { 1062 struct tun_file *tfile = file->private_data; 1063 struct tun_struct *tun; 1064 void __user* argp = (void __user*)arg; 1065 struct ifreq ifr; 1066 int sndbuf; 1067 int ret; 1068 1069 if (cmd == TUNSETIFF || _IOC_TYPE(cmd) == 0x89) 1070 if (copy_from_user(&ifr, argp, sizeof ifr)) 1071 return -EFAULT; 1072 1073 if (cmd == TUNGETFEATURES) { 1074 /* Currently this just means: "what IFF flags are valid?". 1075 * This is needed because we never checked for invalid flags on 1076 * TUNSETIFF. */ 1077 return put_user(IFF_TUN | IFF_TAP | IFF_NO_PI | IFF_ONE_QUEUE | 1078 IFF_VNET_HDR, 1079 (unsigned int __user*)argp); 1080 } 1081 1082 tun = __tun_get(tfile); 1083 if (cmd == TUNSETIFF && !tun) { 1084 int err; 1085 1086 ifr.ifr_name[IFNAMSIZ-1] = '\0'; 1087 1088 rtnl_lock(); 1089 err = tun_set_iff(tfile->net, file, &ifr); 1090 rtnl_unlock(); 1091 1092 if (err) 1093 return err; 1094 1095 if (copy_to_user(argp, &ifr, sizeof(ifr))) 1096 return -EFAULT; 1097 return 0; 1098 } 1099 1100 1101 if (!tun) 1102 return -EBADFD; 1103 1104 DBG(KERN_INFO "%s: tun_chr_ioctl cmd %d\n", tun->dev->name, cmd); 1105 1106 ret = 0; 1107 switch (cmd) { 1108 case TUNGETIFF: 1109 ret = tun_get_iff(current->nsproxy->net_ns, file, &ifr); 1110 if (ret) 1111 break; 1112 1113 if (copy_to_user(argp, &ifr, sizeof(ifr))) 1114 ret = -EFAULT; 1115 break; 1116 1117 case TUNSETNOCSUM: 1118 /* Disable/Enable checksum */ 1119 if (arg) 1120 tun->flags |= TUN_NOCHECKSUM; 1121 else 1122 tun->flags &= ~TUN_NOCHECKSUM; 1123 1124 DBG(KERN_INFO "%s: checksum %s\n", 1125 tun->dev->name, arg ? "disabled" : "enabled"); 1126 break; 1127 1128 case TUNSETPERSIST: 1129 /* Disable/Enable persist mode */ 1130 if (arg) 1131 tun->flags |= TUN_PERSIST; 1132 else 1133 tun->flags &= ~TUN_PERSIST; 1134 1135 DBG(KERN_INFO "%s: persist %s\n", 1136 tun->dev->name, arg ? "enabled" : "disabled"); 1137 break; 1138 1139 case TUNSETOWNER: 1140 /* Set owner of the device */ 1141 tun->owner = (uid_t) arg; 1142 1143 DBG(KERN_INFO "%s: owner set to %d\n", tun->dev->name, tun->owner); 1144 break; 1145 1146 case TUNSETGROUP: 1147 /* Set group of the device */ 1148 tun->group= (gid_t) arg; 1149 1150 DBG(KERN_INFO "%s: group set to %d\n", tun->dev->name, tun->group); 1151 break; 1152 1153 case TUNSETLINK: 1154 /* Only allow setting the type when the interface is down */ 1155 rtnl_lock(); 1156 if (tun->dev->flags & IFF_UP) { 1157 DBG(KERN_INFO "%s: Linktype set failed because interface is up\n", 1158 tun->dev->name); 1159 ret = -EBUSY; 1160 } else { 1161 tun->dev->type = (int) arg; 1162 DBG(KERN_INFO "%s: linktype set to %d\n", tun->dev->name, tun->dev->type); 1163 ret = 0; 1164 } 1165 rtnl_unlock(); 1166 break; 1167 1168 #ifdef TUN_DEBUG 1169 case TUNSETDEBUG: 1170 tun->debug = arg; 1171 break; 1172 #endif 1173 case TUNSETOFFLOAD: 1174 rtnl_lock(); 1175 ret = set_offload(tun->dev, arg); 1176 rtnl_unlock(); 1177 break; 1178 1179 case TUNSETTXFILTER: 1180 /* Can be set only for TAPs */ 1181 ret = -EINVAL; 1182 if ((tun->flags & TUN_TYPE_MASK) != TUN_TAP_DEV) 1183 break; 1184 rtnl_lock(); 1185 ret = update_filter(&tun->txflt, (void __user *)arg); 1186 rtnl_unlock(); 1187 break; 1188 1189 case SIOCGIFHWADDR: 1190 /* Get hw addres */ 1191 memcpy(ifr.ifr_hwaddr.sa_data, tun->dev->dev_addr, ETH_ALEN); 1192 ifr.ifr_hwaddr.sa_family = tun->dev->type; 1193 if (copy_to_user(argp, &ifr, sizeof ifr)) 1194 ret = -EFAULT; 1195 break; 1196 1197 case SIOCSIFHWADDR: 1198 /* Set hw address */ 1199 DBG(KERN_DEBUG "%s: set hw address: %pM\n", 1200 tun->dev->name, ifr.ifr_hwaddr.sa_data); 1201 1202 rtnl_lock(); 1203 ret = dev_set_mac_address(tun->dev, &ifr.ifr_hwaddr); 1204 rtnl_unlock(); 1205 break; 1206 1207 case TUNGETSNDBUF: 1208 sndbuf = tun->sk->sk_sndbuf; 1209 if (copy_to_user(argp, &sndbuf, sizeof(sndbuf))) 1210 ret = -EFAULT; 1211 break; 1212 1213 case TUNSETSNDBUF: 1214 if (copy_from_user(&sndbuf, argp, sizeof(sndbuf))) { 1215 ret = -EFAULT; 1216 break; 1217 } 1218 1219 tun->sk->sk_sndbuf = sndbuf; 1220 break; 1221 1222 default: 1223 ret = -EINVAL; 1224 break; 1225 }; 1226 1227 tun_put(tun); 1228 return ret; 1229 } 1230 1231 static int tun_chr_fasync(int fd, struct file *file, int on) 1232 { 1233 struct tun_struct *tun = tun_get(file); 1234 int ret; 1235 1236 if (!tun) 1237 return -EBADFD; 1238 1239 DBG(KERN_INFO "%s: tun_chr_fasync %d\n", tun->dev->name, on); 1240 1241 lock_kernel(); 1242 if ((ret = fasync_helper(fd, file, on, &tun->fasync)) < 0) 1243 goto out; 1244 1245 if (on) { 1246 ret = __f_setown(file, task_pid(current), PIDTYPE_PID, 0); 1247 if (ret) 1248 goto out; 1249 tun->flags |= TUN_FASYNC; 1250 } else 1251 tun->flags &= ~TUN_FASYNC; 1252 ret = 0; 1253 out: 1254 unlock_kernel(); 1255 tun_put(tun); 1256 return ret; 1257 } 1258 1259 static int tun_chr_open(struct inode *inode, struct file * file) 1260 { 1261 struct tun_file *tfile; 1262 cycle_kernel_lock(); 1263 DBG1(KERN_INFO "tunX: tun_chr_open\n"); 1264 1265 tfile = kmalloc(sizeof(*tfile), GFP_KERNEL); 1266 if (!tfile) 1267 return -ENOMEM; 1268 atomic_set(&tfile->count, 0); 1269 tfile->tun = NULL; 1270 tfile->net = get_net(current->nsproxy->net_ns); 1271 file->private_data = tfile; 1272 return 0; 1273 } 1274 1275 static int tun_chr_close(struct inode *inode, struct file *file) 1276 { 1277 struct tun_file *tfile = file->private_data; 1278 struct tun_struct *tun = __tun_get(tfile); 1279 1280 1281 if (tun) { 1282 DBG(KERN_INFO "%s: tun_chr_close\n", tun->dev->name); 1283 1284 rtnl_lock(); 1285 __tun_detach(tun); 1286 1287 /* If desireable, unregister the netdevice. */ 1288 if (!(tun->flags & TUN_PERSIST)) 1289 unregister_netdevice(tun->dev); 1290 1291 rtnl_unlock(); 1292 } 1293 1294 tun = tfile->tun; 1295 if (tun) 1296 sock_put(tun->sk); 1297 1298 put_net(tfile->net); 1299 kfree(tfile); 1300 1301 return 0; 1302 } 1303 1304 static const struct file_operations tun_fops = { 1305 .owner = THIS_MODULE, 1306 .llseek = no_llseek, 1307 .read = do_sync_read, 1308 .aio_read = tun_chr_aio_read, 1309 .write = do_sync_write, 1310 .aio_write = tun_chr_aio_write, 1311 .poll = tun_chr_poll, 1312 .ioctl = tun_chr_ioctl, 1313 .open = tun_chr_open, 1314 .release = tun_chr_close, 1315 .fasync = tun_chr_fasync 1316 }; 1317 1318 static struct miscdevice tun_miscdev = { 1319 .minor = TUN_MINOR, 1320 .name = "tun", 1321 .fops = &tun_fops, 1322 }; 1323 1324 /* ethtool interface */ 1325 1326 static int tun_get_settings(struct net_device *dev, struct ethtool_cmd *cmd) 1327 { 1328 cmd->supported = 0; 1329 cmd->advertising = 0; 1330 cmd->speed = SPEED_10; 1331 cmd->duplex = DUPLEX_FULL; 1332 cmd->port = PORT_TP; 1333 cmd->phy_address = 0; 1334 cmd->transceiver = XCVR_INTERNAL; 1335 cmd->autoneg = AUTONEG_DISABLE; 1336 cmd->maxtxpkt = 0; 1337 cmd->maxrxpkt = 0; 1338 return 0; 1339 } 1340 1341 static void tun_get_drvinfo(struct net_device *dev, struct ethtool_drvinfo *info) 1342 { 1343 struct tun_struct *tun = netdev_priv(dev); 1344 1345 strcpy(info->driver, DRV_NAME); 1346 strcpy(info->version, DRV_VERSION); 1347 strcpy(info->fw_version, "N/A"); 1348 1349 switch (tun->flags & TUN_TYPE_MASK) { 1350 case TUN_TUN_DEV: 1351 strcpy(info->bus_info, "tun"); 1352 break; 1353 case TUN_TAP_DEV: 1354 strcpy(info->bus_info, "tap"); 1355 break; 1356 } 1357 } 1358 1359 static u32 tun_get_msglevel(struct net_device *dev) 1360 { 1361 #ifdef TUN_DEBUG 1362 struct tun_struct *tun = netdev_priv(dev); 1363 return tun->debug; 1364 #else 1365 return -EOPNOTSUPP; 1366 #endif 1367 } 1368 1369 static void tun_set_msglevel(struct net_device *dev, u32 value) 1370 { 1371 #ifdef TUN_DEBUG 1372 struct tun_struct *tun = netdev_priv(dev); 1373 tun->debug = value; 1374 #endif 1375 } 1376 1377 static u32 tun_get_link(struct net_device *dev) 1378 { 1379 struct tun_struct *tun = netdev_priv(dev); 1380 return !!tun->tfile; 1381 } 1382 1383 static u32 tun_get_rx_csum(struct net_device *dev) 1384 { 1385 struct tun_struct *tun = netdev_priv(dev); 1386 return (tun->flags & TUN_NOCHECKSUM) == 0; 1387 } 1388 1389 static int tun_set_rx_csum(struct net_device *dev, u32 data) 1390 { 1391 struct tun_struct *tun = netdev_priv(dev); 1392 if (data) 1393 tun->flags &= ~TUN_NOCHECKSUM; 1394 else 1395 tun->flags |= TUN_NOCHECKSUM; 1396 return 0; 1397 } 1398 1399 static const struct ethtool_ops tun_ethtool_ops = { 1400 .get_settings = tun_get_settings, 1401 .get_drvinfo = tun_get_drvinfo, 1402 .get_msglevel = tun_get_msglevel, 1403 .set_msglevel = tun_set_msglevel, 1404 .get_link = tun_get_link, 1405 .get_rx_csum = tun_get_rx_csum, 1406 .set_rx_csum = tun_set_rx_csum 1407 }; 1408 1409 1410 static int __init tun_init(void) 1411 { 1412 int ret = 0; 1413 1414 printk(KERN_INFO "tun: %s, %s\n", DRV_DESCRIPTION, DRV_VERSION); 1415 printk(KERN_INFO "tun: %s\n", DRV_COPYRIGHT); 1416 1417 ret = rtnl_link_register(&tun_link_ops); 1418 if (ret) { 1419 printk(KERN_ERR "tun: Can't register link_ops\n"); 1420 goto err_linkops; 1421 } 1422 1423 ret = misc_register(&tun_miscdev); 1424 if (ret) { 1425 printk(KERN_ERR "tun: Can't register misc device %d\n", TUN_MINOR); 1426 goto err_misc; 1427 } 1428 return 0; 1429 err_misc: 1430 rtnl_link_unregister(&tun_link_ops); 1431 err_linkops: 1432 return ret; 1433 } 1434 1435 static void tun_cleanup(void) 1436 { 1437 misc_deregister(&tun_miscdev); 1438 rtnl_link_unregister(&tun_link_ops); 1439 } 1440 1441 module_init(tun_init); 1442 module_exit(tun_cleanup); 1443 MODULE_DESCRIPTION(DRV_DESCRIPTION); 1444 MODULE_AUTHOR(DRV_COPYRIGHT); 1445 MODULE_LICENSE("GPL"); 1446 MODULE_ALIAS_MISCDEV(TUN_MINOR); 1447