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