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 eth_random_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 pr_fmt(fmt) KBUILD_MODNAME ": " fmt 38 39 #define DRV_NAME "tun" 40 #define DRV_VERSION "1.6" 41 #define DRV_DESCRIPTION "Universal TUN/TAP device driver" 42 #define DRV_COPYRIGHT "(C) 1999-2004 Max Krasnyansky <maxk@qualcomm.com>" 43 44 #include <linux/module.h> 45 #include <linux/errno.h> 46 #include <linux/kernel.h> 47 #include <linux/major.h> 48 #include <linux/slab.h> 49 #include <linux/poll.h> 50 #include <linux/fcntl.h> 51 #include <linux/init.h> 52 #include <linux/skbuff.h> 53 #include <linux/netdevice.h> 54 #include <linux/etherdevice.h> 55 #include <linux/miscdevice.h> 56 #include <linux/ethtool.h> 57 #include <linux/rtnetlink.h> 58 #include <linux/compat.h> 59 #include <linux/if.h> 60 #include <linux/if_arp.h> 61 #include <linux/if_ether.h> 62 #include <linux/if_tun.h> 63 #include <linux/crc32.h> 64 #include <linux/nsproxy.h> 65 #include <linux/virtio_net.h> 66 #include <linux/rcupdate.h> 67 #include <net/net_namespace.h> 68 #include <net/netns/generic.h> 69 #include <net/rtnetlink.h> 70 #include <net/sock.h> 71 72 #include <asm/uaccess.h> 73 74 /* Uncomment to enable debugging */ 75 /* #define TUN_DEBUG 1 */ 76 77 #ifdef TUN_DEBUG 78 static int debug; 79 80 #define tun_debug(level, tun, fmt, args...) \ 81 do { \ 82 if (tun->debug) \ 83 netdev_printk(level, tun->dev, fmt, ##args); \ 84 } while (0) 85 #define DBG1(level, fmt, args...) \ 86 do { \ 87 if (debug == 2) \ 88 printk(level fmt, ##args); \ 89 } while (0) 90 #else 91 #define tun_debug(level, tun, fmt, args...) \ 92 do { \ 93 if (0) \ 94 netdev_printk(level, tun->dev, fmt, ##args); \ 95 } while (0) 96 #define DBG1(level, fmt, args...) \ 97 do { \ 98 if (0) \ 99 printk(level fmt, ##args); \ 100 } while (0) 101 #endif 102 103 #define GOODCOPY_LEN 128 104 105 #define FLT_EXACT_COUNT 8 106 struct tap_filter { 107 unsigned int count; /* Number of addrs. Zero means disabled */ 108 u32 mask[2]; /* Mask of the hashed addrs */ 109 unsigned char addr[FLT_EXACT_COUNT][ETH_ALEN]; 110 }; 111 112 /* DEFAULT_MAX_NUM_RSS_QUEUES were choosed to let the rx/tx queues allocated for 113 * the netdevice to be fit in one page. So we can make sure the success of 114 * memory allocation. TODO: increase the limit. */ 115 #define MAX_TAP_QUEUES DEFAULT_MAX_NUM_RSS_QUEUES 116 #define MAX_TAP_FLOWS 4096 117 118 #define TUN_FLOW_EXPIRE (3 * HZ) 119 120 /* A tun_file connects an open character device to a tuntap netdevice. It 121 * also contains all socket related strctures (except sock_fprog and tap_filter) 122 * to serve as one transmit queue for tuntap device. The sock_fprog and 123 * tap_filter were kept in tun_struct since they were used for filtering for the 124 * netdevice not for a specific queue (at least I didn't see the requirement for 125 * this). 126 * 127 * RCU usage: 128 * The tun_file and tun_struct are loosely coupled, the pointer from one to the 129 * other can only be read while rcu_read_lock or rtnl_lock is held. 130 */ 131 struct tun_file { 132 struct sock sk; 133 struct socket socket; 134 struct socket_wq wq; 135 struct tun_struct __rcu *tun; 136 struct net *net; 137 struct fasync_struct *fasync; 138 /* only used for fasnyc */ 139 unsigned int flags; 140 u16 queue_index; 141 struct list_head next; 142 struct tun_struct *detached; 143 }; 144 145 struct tun_flow_entry { 146 struct hlist_node hash_link; 147 struct rcu_head rcu; 148 struct tun_struct *tun; 149 150 u32 rxhash; 151 int queue_index; 152 unsigned long updated; 153 }; 154 155 #define TUN_NUM_FLOW_ENTRIES 1024 156 157 /* Since the socket were moved to tun_file, to preserve the behavior of persist 158 * device, socket filter, sndbuf and vnet header size were restore when the 159 * file were attached to a persist device. 160 */ 161 struct tun_struct { 162 struct tun_file __rcu *tfiles[MAX_TAP_QUEUES]; 163 unsigned int numqueues; 164 unsigned int flags; 165 kuid_t owner; 166 kgid_t group; 167 168 struct net_device *dev; 169 netdev_features_t set_features; 170 #define TUN_USER_FEATURES (NETIF_F_HW_CSUM|NETIF_F_TSO_ECN|NETIF_F_TSO| \ 171 NETIF_F_TSO6|NETIF_F_UFO) 172 173 int vnet_hdr_sz; 174 int sndbuf; 175 struct tap_filter txflt; 176 struct sock_fprog fprog; 177 /* protected by rtnl lock */ 178 bool filter_attached; 179 #ifdef TUN_DEBUG 180 int debug; 181 #endif 182 spinlock_t lock; 183 struct hlist_head flows[TUN_NUM_FLOW_ENTRIES]; 184 struct timer_list flow_gc_timer; 185 unsigned long ageing_time; 186 unsigned int numdisabled; 187 struct list_head disabled; 188 void *security; 189 u32 flow_count; 190 }; 191 192 static inline u32 tun_hashfn(u32 rxhash) 193 { 194 return rxhash & 0x3ff; 195 } 196 197 static struct tun_flow_entry *tun_flow_find(struct hlist_head *head, u32 rxhash) 198 { 199 struct tun_flow_entry *e; 200 201 hlist_for_each_entry_rcu(e, head, hash_link) { 202 if (e->rxhash == rxhash) 203 return e; 204 } 205 return NULL; 206 } 207 208 static struct tun_flow_entry *tun_flow_create(struct tun_struct *tun, 209 struct hlist_head *head, 210 u32 rxhash, u16 queue_index) 211 { 212 struct tun_flow_entry *e = kmalloc(sizeof(*e), GFP_ATOMIC); 213 214 if (e) { 215 tun_debug(KERN_INFO, tun, "create flow: hash %u index %u\n", 216 rxhash, queue_index); 217 e->updated = jiffies; 218 e->rxhash = rxhash; 219 e->queue_index = queue_index; 220 e->tun = tun; 221 hlist_add_head_rcu(&e->hash_link, head); 222 ++tun->flow_count; 223 } 224 return e; 225 } 226 227 static void tun_flow_delete(struct tun_struct *tun, struct tun_flow_entry *e) 228 { 229 tun_debug(KERN_INFO, tun, "delete flow: hash %u index %u\n", 230 e->rxhash, e->queue_index); 231 hlist_del_rcu(&e->hash_link); 232 kfree_rcu(e, rcu); 233 --tun->flow_count; 234 } 235 236 static void tun_flow_flush(struct tun_struct *tun) 237 { 238 int i; 239 240 spin_lock_bh(&tun->lock); 241 for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++) { 242 struct tun_flow_entry *e; 243 struct hlist_node *n; 244 245 hlist_for_each_entry_safe(e, n, &tun->flows[i], hash_link) 246 tun_flow_delete(tun, e); 247 } 248 spin_unlock_bh(&tun->lock); 249 } 250 251 static void tun_flow_delete_by_queue(struct tun_struct *tun, u16 queue_index) 252 { 253 int i; 254 255 spin_lock_bh(&tun->lock); 256 for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++) { 257 struct tun_flow_entry *e; 258 struct hlist_node *n; 259 260 hlist_for_each_entry_safe(e, n, &tun->flows[i], hash_link) { 261 if (e->queue_index == queue_index) 262 tun_flow_delete(tun, e); 263 } 264 } 265 spin_unlock_bh(&tun->lock); 266 } 267 268 static void tun_flow_cleanup(unsigned long data) 269 { 270 struct tun_struct *tun = (struct tun_struct *)data; 271 unsigned long delay = tun->ageing_time; 272 unsigned long next_timer = jiffies + delay; 273 unsigned long count = 0; 274 int i; 275 276 tun_debug(KERN_INFO, tun, "tun_flow_cleanup\n"); 277 278 spin_lock_bh(&tun->lock); 279 for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++) { 280 struct tun_flow_entry *e; 281 struct hlist_node *n; 282 283 hlist_for_each_entry_safe(e, n, &tun->flows[i], hash_link) { 284 unsigned long this_timer; 285 count++; 286 this_timer = e->updated + delay; 287 if (time_before_eq(this_timer, jiffies)) 288 tun_flow_delete(tun, e); 289 else if (time_before(this_timer, next_timer)) 290 next_timer = this_timer; 291 } 292 } 293 294 if (count) 295 mod_timer(&tun->flow_gc_timer, round_jiffies_up(next_timer)); 296 spin_unlock_bh(&tun->lock); 297 } 298 299 static void tun_flow_update(struct tun_struct *tun, u32 rxhash, 300 struct tun_file *tfile) 301 { 302 struct hlist_head *head; 303 struct tun_flow_entry *e; 304 unsigned long delay = tun->ageing_time; 305 u16 queue_index = tfile->queue_index; 306 307 if (!rxhash) 308 return; 309 else 310 head = &tun->flows[tun_hashfn(rxhash)]; 311 312 rcu_read_lock(); 313 314 /* We may get a very small possibility of OOO during switching, not 315 * worth to optimize.*/ 316 if (tun->numqueues == 1 || tfile->detached) 317 goto unlock; 318 319 e = tun_flow_find(head, rxhash); 320 if (likely(e)) { 321 /* TODO: keep queueing to old queue until it's empty? */ 322 e->queue_index = queue_index; 323 e->updated = jiffies; 324 } else { 325 spin_lock_bh(&tun->lock); 326 if (!tun_flow_find(head, rxhash) && 327 tun->flow_count < MAX_TAP_FLOWS) 328 tun_flow_create(tun, head, rxhash, queue_index); 329 330 if (!timer_pending(&tun->flow_gc_timer)) 331 mod_timer(&tun->flow_gc_timer, 332 round_jiffies_up(jiffies + delay)); 333 spin_unlock_bh(&tun->lock); 334 } 335 336 unlock: 337 rcu_read_unlock(); 338 } 339 340 /* We try to identify a flow through its rxhash first. The reason that 341 * we do not check rxq no. is becuase some cards(e.g 82599), chooses 342 * the rxq based on the txq where the last packet of the flow comes. As 343 * the userspace application move between processors, we may get a 344 * different rxq no. here. If we could not get rxhash, then we would 345 * hope the rxq no. may help here. 346 */ 347 static u16 tun_select_queue(struct net_device *dev, struct sk_buff *skb) 348 { 349 struct tun_struct *tun = netdev_priv(dev); 350 struct tun_flow_entry *e; 351 u32 txq = 0; 352 u32 numqueues = 0; 353 354 rcu_read_lock(); 355 numqueues = tun->numqueues; 356 357 txq = skb_get_rxhash(skb); 358 if (txq) { 359 e = tun_flow_find(&tun->flows[tun_hashfn(txq)], txq); 360 if (e) 361 txq = e->queue_index; 362 else 363 /* use multiply and shift instead of expensive divide */ 364 txq = ((u64)txq * numqueues) >> 32; 365 } else if (likely(skb_rx_queue_recorded(skb))) { 366 txq = skb_get_rx_queue(skb); 367 while (unlikely(txq >= numqueues)) 368 txq -= numqueues; 369 } 370 371 rcu_read_unlock(); 372 return txq; 373 } 374 375 static inline bool tun_not_capable(struct tun_struct *tun) 376 { 377 const struct cred *cred = current_cred(); 378 struct net *net = dev_net(tun->dev); 379 380 return ((uid_valid(tun->owner) && !uid_eq(cred->euid, tun->owner)) || 381 (gid_valid(tun->group) && !in_egroup_p(tun->group))) && 382 !ns_capable(net->user_ns, CAP_NET_ADMIN); 383 } 384 385 static void tun_set_real_num_queues(struct tun_struct *tun) 386 { 387 netif_set_real_num_tx_queues(tun->dev, tun->numqueues); 388 netif_set_real_num_rx_queues(tun->dev, tun->numqueues); 389 } 390 391 static void tun_disable_queue(struct tun_struct *tun, struct tun_file *tfile) 392 { 393 tfile->detached = tun; 394 list_add_tail(&tfile->next, &tun->disabled); 395 ++tun->numdisabled; 396 } 397 398 static struct tun_struct *tun_enable_queue(struct tun_file *tfile) 399 { 400 struct tun_struct *tun = tfile->detached; 401 402 tfile->detached = NULL; 403 list_del_init(&tfile->next); 404 --tun->numdisabled; 405 return tun; 406 } 407 408 static void __tun_detach(struct tun_file *tfile, bool clean) 409 { 410 struct tun_file *ntfile; 411 struct tun_struct *tun; 412 struct net_device *dev; 413 414 tun = rtnl_dereference(tfile->tun); 415 416 if (tun && !tfile->detached) { 417 u16 index = tfile->queue_index; 418 BUG_ON(index >= tun->numqueues); 419 dev = tun->dev; 420 421 rcu_assign_pointer(tun->tfiles[index], 422 tun->tfiles[tun->numqueues - 1]); 423 ntfile = rtnl_dereference(tun->tfiles[index]); 424 ntfile->queue_index = index; 425 426 --tun->numqueues; 427 if (clean) { 428 rcu_assign_pointer(tfile->tun, NULL); 429 sock_put(&tfile->sk); 430 } else 431 tun_disable_queue(tun, tfile); 432 433 synchronize_net(); 434 tun_flow_delete_by_queue(tun, tun->numqueues + 1); 435 /* Drop read queue */ 436 skb_queue_purge(&tfile->sk.sk_receive_queue); 437 tun_set_real_num_queues(tun); 438 } else if (tfile->detached && clean) { 439 tun = tun_enable_queue(tfile); 440 sock_put(&tfile->sk); 441 } 442 443 if (clean) { 444 if (tun && tun->numqueues == 0 && tun->numdisabled == 0) { 445 netif_carrier_off(tun->dev); 446 447 if (!(tun->flags & TUN_PERSIST) && 448 tun->dev->reg_state == NETREG_REGISTERED) 449 unregister_netdevice(tun->dev); 450 } 451 452 BUG_ON(!test_bit(SOCK_EXTERNALLY_ALLOCATED, 453 &tfile->socket.flags)); 454 sk_release_kernel(&tfile->sk); 455 } 456 } 457 458 static void tun_detach(struct tun_file *tfile, bool clean) 459 { 460 rtnl_lock(); 461 __tun_detach(tfile, clean); 462 rtnl_unlock(); 463 } 464 465 static void tun_detach_all(struct net_device *dev) 466 { 467 struct tun_struct *tun = netdev_priv(dev); 468 struct tun_file *tfile, *tmp; 469 int i, n = tun->numqueues; 470 471 for (i = 0; i < n; i++) { 472 tfile = rtnl_dereference(tun->tfiles[i]); 473 BUG_ON(!tfile); 474 wake_up_all(&tfile->wq.wait); 475 rcu_assign_pointer(tfile->tun, NULL); 476 --tun->numqueues; 477 } 478 list_for_each_entry(tfile, &tun->disabled, next) { 479 wake_up_all(&tfile->wq.wait); 480 rcu_assign_pointer(tfile->tun, NULL); 481 } 482 BUG_ON(tun->numqueues != 0); 483 484 synchronize_net(); 485 for (i = 0; i < n; i++) { 486 tfile = rtnl_dereference(tun->tfiles[i]); 487 /* Drop read queue */ 488 skb_queue_purge(&tfile->sk.sk_receive_queue); 489 sock_put(&tfile->sk); 490 } 491 list_for_each_entry_safe(tfile, tmp, &tun->disabled, next) { 492 tun_enable_queue(tfile); 493 skb_queue_purge(&tfile->sk.sk_receive_queue); 494 sock_put(&tfile->sk); 495 } 496 BUG_ON(tun->numdisabled != 0); 497 498 if (tun->flags & TUN_PERSIST) 499 module_put(THIS_MODULE); 500 } 501 502 static int tun_attach(struct tun_struct *tun, struct file *file) 503 { 504 struct tun_file *tfile = file->private_data; 505 int err; 506 507 err = security_tun_dev_attach(tfile->socket.sk, tun->security); 508 if (err < 0) 509 goto out; 510 511 err = -EINVAL; 512 if (rtnl_dereference(tfile->tun) && !tfile->detached) 513 goto out; 514 515 err = -EBUSY; 516 if (!(tun->flags & TUN_TAP_MQ) && tun->numqueues == 1) 517 goto out; 518 519 err = -E2BIG; 520 if (!tfile->detached && 521 tun->numqueues + tun->numdisabled == MAX_TAP_QUEUES) 522 goto out; 523 524 err = 0; 525 526 /* Re-attach the filter to presist device */ 527 if (tun->filter_attached == true) { 528 err = sk_attach_filter(&tun->fprog, tfile->socket.sk); 529 if (!err) 530 goto out; 531 } 532 tfile->queue_index = tun->numqueues; 533 rcu_assign_pointer(tfile->tun, tun); 534 rcu_assign_pointer(tun->tfiles[tun->numqueues], tfile); 535 tun->numqueues++; 536 537 if (tfile->detached) 538 tun_enable_queue(tfile); 539 else 540 sock_hold(&tfile->sk); 541 542 tun_set_real_num_queues(tun); 543 544 /* device is allowed to go away first, so no need to hold extra 545 * refcnt. 546 */ 547 548 out: 549 return err; 550 } 551 552 static struct tun_struct *__tun_get(struct tun_file *tfile) 553 { 554 struct tun_struct *tun; 555 556 rcu_read_lock(); 557 tun = rcu_dereference(tfile->tun); 558 if (tun) 559 dev_hold(tun->dev); 560 rcu_read_unlock(); 561 562 return tun; 563 } 564 565 static struct tun_struct *tun_get(struct file *file) 566 { 567 return __tun_get(file->private_data); 568 } 569 570 static void tun_put(struct tun_struct *tun) 571 { 572 dev_put(tun->dev); 573 } 574 575 /* TAP filtering */ 576 static void addr_hash_set(u32 *mask, const u8 *addr) 577 { 578 int n = ether_crc(ETH_ALEN, addr) >> 26; 579 mask[n >> 5] |= (1 << (n & 31)); 580 } 581 582 static unsigned int addr_hash_test(const u32 *mask, const u8 *addr) 583 { 584 int n = ether_crc(ETH_ALEN, addr) >> 26; 585 return mask[n >> 5] & (1 << (n & 31)); 586 } 587 588 static int update_filter(struct tap_filter *filter, void __user *arg) 589 { 590 struct { u8 u[ETH_ALEN]; } *addr; 591 struct tun_filter uf; 592 int err, alen, n, nexact; 593 594 if (copy_from_user(&uf, arg, sizeof(uf))) 595 return -EFAULT; 596 597 if (!uf.count) { 598 /* Disabled */ 599 filter->count = 0; 600 return 0; 601 } 602 603 alen = ETH_ALEN * uf.count; 604 addr = kmalloc(alen, GFP_KERNEL); 605 if (!addr) 606 return -ENOMEM; 607 608 if (copy_from_user(addr, arg + sizeof(uf), alen)) { 609 err = -EFAULT; 610 goto done; 611 } 612 613 /* The filter is updated without holding any locks. Which is 614 * perfectly safe. We disable it first and in the worst 615 * case we'll accept a few undesired packets. */ 616 filter->count = 0; 617 wmb(); 618 619 /* Use first set of addresses as an exact filter */ 620 for (n = 0; n < uf.count && n < FLT_EXACT_COUNT; n++) 621 memcpy(filter->addr[n], addr[n].u, ETH_ALEN); 622 623 nexact = n; 624 625 /* Remaining multicast addresses are hashed, 626 * unicast will leave the filter disabled. */ 627 memset(filter->mask, 0, sizeof(filter->mask)); 628 for (; n < uf.count; n++) { 629 if (!is_multicast_ether_addr(addr[n].u)) { 630 err = 0; /* no filter */ 631 goto done; 632 } 633 addr_hash_set(filter->mask, addr[n].u); 634 } 635 636 /* For ALLMULTI just set the mask to all ones. 637 * This overrides the mask populated above. */ 638 if ((uf.flags & TUN_FLT_ALLMULTI)) 639 memset(filter->mask, ~0, sizeof(filter->mask)); 640 641 /* Now enable the filter */ 642 wmb(); 643 filter->count = nexact; 644 645 /* Return the number of exact filters */ 646 err = nexact; 647 648 done: 649 kfree(addr); 650 return err; 651 } 652 653 /* Returns: 0 - drop, !=0 - accept */ 654 static int run_filter(struct tap_filter *filter, const struct sk_buff *skb) 655 { 656 /* Cannot use eth_hdr(skb) here because skb_mac_hdr() is incorrect 657 * at this point. */ 658 struct ethhdr *eh = (struct ethhdr *) skb->data; 659 int i; 660 661 /* Exact match */ 662 for (i = 0; i < filter->count; i++) 663 if (ether_addr_equal(eh->h_dest, filter->addr[i])) 664 return 1; 665 666 /* Inexact match (multicast only) */ 667 if (is_multicast_ether_addr(eh->h_dest)) 668 return addr_hash_test(filter->mask, eh->h_dest); 669 670 return 0; 671 } 672 673 /* 674 * Checks whether the packet is accepted or not. 675 * Returns: 0 - drop, !=0 - accept 676 */ 677 static int check_filter(struct tap_filter *filter, const struct sk_buff *skb) 678 { 679 if (!filter->count) 680 return 1; 681 682 return run_filter(filter, skb); 683 } 684 685 /* Network device part of the driver */ 686 687 static const struct ethtool_ops tun_ethtool_ops; 688 689 /* Net device detach from fd. */ 690 static void tun_net_uninit(struct net_device *dev) 691 { 692 tun_detach_all(dev); 693 } 694 695 /* Net device open. */ 696 static int tun_net_open(struct net_device *dev) 697 { 698 netif_tx_start_all_queues(dev); 699 return 0; 700 } 701 702 /* Net device close. */ 703 static int tun_net_close(struct net_device *dev) 704 { 705 netif_tx_stop_all_queues(dev); 706 return 0; 707 } 708 709 /* Net device start xmit */ 710 static netdev_tx_t tun_net_xmit(struct sk_buff *skb, struct net_device *dev) 711 { 712 struct tun_struct *tun = netdev_priv(dev); 713 int txq = skb->queue_mapping; 714 struct tun_file *tfile; 715 716 rcu_read_lock(); 717 tfile = rcu_dereference(tun->tfiles[txq]); 718 719 /* Drop packet if interface is not attached */ 720 if (txq >= tun->numqueues) 721 goto drop; 722 723 tun_debug(KERN_INFO, tun, "tun_net_xmit %d\n", skb->len); 724 725 BUG_ON(!tfile); 726 727 /* Drop if the filter does not like it. 728 * This is a noop if the filter is disabled. 729 * Filter can be enabled only for the TAP devices. */ 730 if (!check_filter(&tun->txflt, skb)) 731 goto drop; 732 733 if (tfile->socket.sk->sk_filter && 734 sk_filter(tfile->socket.sk, skb)) 735 goto drop; 736 737 /* Limit the number of packets queued by dividing txq length with the 738 * number of queues. 739 */ 740 if (skb_queue_len(&tfile->socket.sk->sk_receive_queue) 741 >= dev->tx_queue_len / tun->numqueues) 742 goto drop; 743 744 /* Orphan the skb - required as we might hang on to it 745 * for indefinite time. */ 746 if (unlikely(skb_orphan_frags(skb, GFP_ATOMIC))) 747 goto drop; 748 skb_orphan(skb); 749 750 nf_reset(skb); 751 752 /* Enqueue packet */ 753 skb_queue_tail(&tfile->socket.sk->sk_receive_queue, skb); 754 755 /* Notify and wake up reader process */ 756 if (tfile->flags & TUN_FASYNC) 757 kill_fasync(&tfile->fasync, SIGIO, POLL_IN); 758 wake_up_interruptible_poll(&tfile->wq.wait, POLLIN | 759 POLLRDNORM | POLLRDBAND); 760 761 rcu_read_unlock(); 762 return NETDEV_TX_OK; 763 764 drop: 765 dev->stats.tx_dropped++; 766 skb_tx_error(skb); 767 kfree_skb(skb); 768 rcu_read_unlock(); 769 return NETDEV_TX_OK; 770 } 771 772 static void tun_net_mclist(struct net_device *dev) 773 { 774 /* 775 * This callback is supposed to deal with mc filter in 776 * _rx_ path and has nothing to do with the _tx_ path. 777 * In rx path we always accept everything userspace gives us. 778 */ 779 } 780 781 #define MIN_MTU 68 782 #define MAX_MTU 65535 783 784 static int 785 tun_net_change_mtu(struct net_device *dev, int new_mtu) 786 { 787 if (new_mtu < MIN_MTU || new_mtu + dev->hard_header_len > MAX_MTU) 788 return -EINVAL; 789 dev->mtu = new_mtu; 790 return 0; 791 } 792 793 static netdev_features_t tun_net_fix_features(struct net_device *dev, 794 netdev_features_t features) 795 { 796 struct tun_struct *tun = netdev_priv(dev); 797 798 return (features & tun->set_features) | (features & ~TUN_USER_FEATURES); 799 } 800 #ifdef CONFIG_NET_POLL_CONTROLLER 801 static void tun_poll_controller(struct net_device *dev) 802 { 803 /* 804 * Tun only receives frames when: 805 * 1) the char device endpoint gets data from user space 806 * 2) the tun socket gets a sendmsg call from user space 807 * Since both of those are syncronous operations, we are guaranteed 808 * never to have pending data when we poll for it 809 * so theres nothing to do here but return. 810 * We need this though so netpoll recognizes us as an interface that 811 * supports polling, which enables bridge devices in virt setups to 812 * still use netconsole 813 */ 814 return; 815 } 816 #endif 817 static const struct net_device_ops tun_netdev_ops = { 818 .ndo_uninit = tun_net_uninit, 819 .ndo_open = tun_net_open, 820 .ndo_stop = tun_net_close, 821 .ndo_start_xmit = tun_net_xmit, 822 .ndo_change_mtu = tun_net_change_mtu, 823 .ndo_fix_features = tun_net_fix_features, 824 .ndo_select_queue = tun_select_queue, 825 #ifdef CONFIG_NET_POLL_CONTROLLER 826 .ndo_poll_controller = tun_poll_controller, 827 #endif 828 }; 829 830 static const struct net_device_ops tap_netdev_ops = { 831 .ndo_uninit = tun_net_uninit, 832 .ndo_open = tun_net_open, 833 .ndo_stop = tun_net_close, 834 .ndo_start_xmit = tun_net_xmit, 835 .ndo_change_mtu = tun_net_change_mtu, 836 .ndo_fix_features = tun_net_fix_features, 837 .ndo_set_rx_mode = tun_net_mclist, 838 .ndo_set_mac_address = eth_mac_addr, 839 .ndo_validate_addr = eth_validate_addr, 840 .ndo_select_queue = tun_select_queue, 841 #ifdef CONFIG_NET_POLL_CONTROLLER 842 .ndo_poll_controller = tun_poll_controller, 843 #endif 844 }; 845 846 static int tun_flow_init(struct tun_struct *tun) 847 { 848 int i; 849 850 for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++) 851 INIT_HLIST_HEAD(&tun->flows[i]); 852 853 tun->ageing_time = TUN_FLOW_EXPIRE; 854 setup_timer(&tun->flow_gc_timer, tun_flow_cleanup, (unsigned long)tun); 855 mod_timer(&tun->flow_gc_timer, 856 round_jiffies_up(jiffies + tun->ageing_time)); 857 858 return 0; 859 } 860 861 static void tun_flow_uninit(struct tun_struct *tun) 862 { 863 del_timer_sync(&tun->flow_gc_timer); 864 tun_flow_flush(tun); 865 } 866 867 /* Initialize net device. */ 868 static void tun_net_init(struct net_device *dev) 869 { 870 struct tun_struct *tun = netdev_priv(dev); 871 872 switch (tun->flags & TUN_TYPE_MASK) { 873 case TUN_TUN_DEV: 874 dev->netdev_ops = &tun_netdev_ops; 875 876 /* Point-to-Point TUN Device */ 877 dev->hard_header_len = 0; 878 dev->addr_len = 0; 879 dev->mtu = 1500; 880 881 /* Zero header length */ 882 dev->type = ARPHRD_NONE; 883 dev->flags = IFF_POINTOPOINT | IFF_NOARP | IFF_MULTICAST; 884 dev->tx_queue_len = TUN_READQ_SIZE; /* We prefer our own queue length */ 885 break; 886 887 case TUN_TAP_DEV: 888 dev->netdev_ops = &tap_netdev_ops; 889 /* Ethernet TAP Device */ 890 ether_setup(dev); 891 dev->priv_flags &= ~IFF_TX_SKB_SHARING; 892 dev->priv_flags |= IFF_LIVE_ADDR_CHANGE; 893 894 eth_hw_addr_random(dev); 895 896 dev->tx_queue_len = TUN_READQ_SIZE; /* We prefer our own queue length */ 897 break; 898 } 899 } 900 901 /* Character device part */ 902 903 /* Poll */ 904 static unsigned int tun_chr_poll(struct file *file, poll_table *wait) 905 { 906 struct tun_file *tfile = file->private_data; 907 struct tun_struct *tun = __tun_get(tfile); 908 struct sock *sk; 909 unsigned int mask = 0; 910 911 if (!tun) 912 return POLLERR; 913 914 sk = tfile->socket.sk; 915 916 tun_debug(KERN_INFO, tun, "tun_chr_poll\n"); 917 918 poll_wait(file, &tfile->wq.wait, wait); 919 920 if (!skb_queue_empty(&sk->sk_receive_queue)) 921 mask |= POLLIN | POLLRDNORM; 922 923 if (sock_writeable(sk) || 924 (!test_and_set_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags) && 925 sock_writeable(sk))) 926 mask |= POLLOUT | POLLWRNORM; 927 928 if (tun->dev->reg_state != NETREG_REGISTERED) 929 mask = POLLERR; 930 931 tun_put(tun); 932 return mask; 933 } 934 935 /* prepad is the amount to reserve at front. len is length after that. 936 * linear is a hint as to how much to copy (usually headers). */ 937 static struct sk_buff *tun_alloc_skb(struct tun_file *tfile, 938 size_t prepad, size_t len, 939 size_t linear, int noblock) 940 { 941 struct sock *sk = tfile->socket.sk; 942 struct sk_buff *skb; 943 int err; 944 945 /* Under a page? Don't bother with paged skb. */ 946 if (prepad + len < PAGE_SIZE || !linear) 947 linear = len; 948 949 skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock, 950 &err); 951 if (!skb) 952 return ERR_PTR(err); 953 954 skb_reserve(skb, prepad); 955 skb_put(skb, linear); 956 skb->data_len = len - linear; 957 skb->len += len - linear; 958 959 return skb; 960 } 961 962 /* set skb frags from iovec, this can move to core network code for reuse */ 963 static int zerocopy_sg_from_iovec(struct sk_buff *skb, const struct iovec *from, 964 int offset, size_t count) 965 { 966 int len = iov_length(from, count) - offset; 967 int copy = skb_headlen(skb); 968 int size, offset1 = 0; 969 int i = 0; 970 971 /* Skip over from offset */ 972 while (count && (offset >= from->iov_len)) { 973 offset -= from->iov_len; 974 ++from; 975 --count; 976 } 977 978 /* copy up to skb headlen */ 979 while (count && (copy > 0)) { 980 size = min_t(unsigned int, copy, from->iov_len - offset); 981 if (copy_from_user(skb->data + offset1, from->iov_base + offset, 982 size)) 983 return -EFAULT; 984 if (copy > size) { 985 ++from; 986 --count; 987 offset = 0; 988 } else 989 offset += size; 990 copy -= size; 991 offset1 += size; 992 } 993 994 if (len == offset1) 995 return 0; 996 997 while (count--) { 998 struct page *page[MAX_SKB_FRAGS]; 999 int num_pages; 1000 unsigned long base; 1001 unsigned long truesize; 1002 1003 len = from->iov_len - offset; 1004 if (!len) { 1005 offset = 0; 1006 ++from; 1007 continue; 1008 } 1009 base = (unsigned long)from->iov_base + offset; 1010 size = ((base & ~PAGE_MASK) + len + ~PAGE_MASK) >> PAGE_SHIFT; 1011 if (i + size > MAX_SKB_FRAGS) 1012 return -EMSGSIZE; 1013 num_pages = get_user_pages_fast(base, size, 0, &page[i]); 1014 if (num_pages != size) { 1015 for (i = 0; i < num_pages; i++) 1016 put_page(page[i]); 1017 return -EFAULT; 1018 } 1019 truesize = size * PAGE_SIZE; 1020 skb->data_len += len; 1021 skb->len += len; 1022 skb->truesize += truesize; 1023 atomic_add(truesize, &skb->sk->sk_wmem_alloc); 1024 while (len) { 1025 int off = base & ~PAGE_MASK; 1026 int size = min_t(int, len, PAGE_SIZE - off); 1027 __skb_fill_page_desc(skb, i, page[i], off, size); 1028 skb_shinfo(skb)->nr_frags++; 1029 /* increase sk_wmem_alloc */ 1030 base += size; 1031 len -= size; 1032 i++; 1033 } 1034 offset = 0; 1035 ++from; 1036 } 1037 return 0; 1038 } 1039 1040 /* Get packet from user space buffer */ 1041 static ssize_t tun_get_user(struct tun_struct *tun, struct tun_file *tfile, 1042 void *msg_control, const struct iovec *iv, 1043 size_t total_len, size_t count, int noblock) 1044 { 1045 struct tun_pi pi = { 0, cpu_to_be16(ETH_P_IP) }; 1046 struct sk_buff *skb; 1047 size_t len = total_len, align = NET_SKB_PAD; 1048 struct virtio_net_hdr gso = { 0 }; 1049 int offset = 0; 1050 int copylen; 1051 bool zerocopy = false; 1052 int err; 1053 u32 rxhash; 1054 1055 if (!(tun->flags & TUN_NO_PI)) { 1056 if ((len -= sizeof(pi)) > total_len) 1057 return -EINVAL; 1058 1059 if (memcpy_fromiovecend((void *)&pi, iv, 0, sizeof(pi))) 1060 return -EFAULT; 1061 offset += sizeof(pi); 1062 } 1063 1064 if (tun->flags & TUN_VNET_HDR) { 1065 if ((len -= tun->vnet_hdr_sz) > total_len) 1066 return -EINVAL; 1067 1068 if (memcpy_fromiovecend((void *)&gso, iv, offset, sizeof(gso))) 1069 return -EFAULT; 1070 1071 if ((gso.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) && 1072 gso.csum_start + gso.csum_offset + 2 > gso.hdr_len) 1073 gso.hdr_len = gso.csum_start + gso.csum_offset + 2; 1074 1075 if (gso.hdr_len > len) 1076 return -EINVAL; 1077 offset += tun->vnet_hdr_sz; 1078 } 1079 1080 if ((tun->flags & TUN_TYPE_MASK) == TUN_TAP_DEV) { 1081 align += NET_IP_ALIGN; 1082 if (unlikely(len < ETH_HLEN || 1083 (gso.hdr_len && gso.hdr_len < ETH_HLEN))) 1084 return -EINVAL; 1085 } 1086 1087 if (msg_control) 1088 zerocopy = true; 1089 1090 if (zerocopy) { 1091 /* Userspace may produce vectors with count greater than 1092 * MAX_SKB_FRAGS, so we need to linearize parts of the skb 1093 * to let the rest of data to be fit in the frags. 1094 */ 1095 if (count > MAX_SKB_FRAGS) { 1096 copylen = iov_length(iv, count - MAX_SKB_FRAGS); 1097 if (copylen < offset) 1098 copylen = 0; 1099 else 1100 copylen -= offset; 1101 } else 1102 copylen = 0; 1103 /* There are 256 bytes to be copied in skb, so there is enough 1104 * room for skb expand head in case it is used. 1105 * The rest of the buffer is mapped from userspace. 1106 */ 1107 if (copylen < gso.hdr_len) 1108 copylen = gso.hdr_len; 1109 if (!copylen) 1110 copylen = GOODCOPY_LEN; 1111 } else 1112 copylen = len; 1113 1114 skb = tun_alloc_skb(tfile, align, copylen, gso.hdr_len, noblock); 1115 if (IS_ERR(skb)) { 1116 if (PTR_ERR(skb) != -EAGAIN) 1117 tun->dev->stats.rx_dropped++; 1118 return PTR_ERR(skb); 1119 } 1120 1121 if (zerocopy) 1122 err = zerocopy_sg_from_iovec(skb, iv, offset, count); 1123 else 1124 err = skb_copy_datagram_from_iovec(skb, 0, iv, offset, len); 1125 1126 if (err) { 1127 tun->dev->stats.rx_dropped++; 1128 kfree_skb(skb); 1129 return -EFAULT; 1130 } 1131 1132 if (gso.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) { 1133 if (!skb_partial_csum_set(skb, gso.csum_start, 1134 gso.csum_offset)) { 1135 tun->dev->stats.rx_frame_errors++; 1136 kfree_skb(skb); 1137 return -EINVAL; 1138 } 1139 } 1140 1141 switch (tun->flags & TUN_TYPE_MASK) { 1142 case TUN_TUN_DEV: 1143 if (tun->flags & TUN_NO_PI) { 1144 switch (skb->data[0] & 0xf0) { 1145 case 0x40: 1146 pi.proto = htons(ETH_P_IP); 1147 break; 1148 case 0x60: 1149 pi.proto = htons(ETH_P_IPV6); 1150 break; 1151 default: 1152 tun->dev->stats.rx_dropped++; 1153 kfree_skb(skb); 1154 return -EINVAL; 1155 } 1156 } 1157 1158 skb_reset_mac_header(skb); 1159 skb->protocol = pi.proto; 1160 skb->dev = tun->dev; 1161 break; 1162 case TUN_TAP_DEV: 1163 skb->protocol = eth_type_trans(skb, tun->dev); 1164 break; 1165 } 1166 1167 if (gso.gso_type != VIRTIO_NET_HDR_GSO_NONE) { 1168 pr_debug("GSO!\n"); 1169 switch (gso.gso_type & ~VIRTIO_NET_HDR_GSO_ECN) { 1170 case VIRTIO_NET_HDR_GSO_TCPV4: 1171 skb_shinfo(skb)->gso_type = SKB_GSO_TCPV4; 1172 break; 1173 case VIRTIO_NET_HDR_GSO_TCPV6: 1174 skb_shinfo(skb)->gso_type = SKB_GSO_TCPV6; 1175 break; 1176 case VIRTIO_NET_HDR_GSO_UDP: 1177 skb_shinfo(skb)->gso_type = SKB_GSO_UDP; 1178 break; 1179 default: 1180 tun->dev->stats.rx_frame_errors++; 1181 kfree_skb(skb); 1182 return -EINVAL; 1183 } 1184 1185 if (gso.gso_type & VIRTIO_NET_HDR_GSO_ECN) 1186 skb_shinfo(skb)->gso_type |= SKB_GSO_TCP_ECN; 1187 1188 skb_shinfo(skb)->gso_size = gso.gso_size; 1189 if (skb_shinfo(skb)->gso_size == 0) { 1190 tun->dev->stats.rx_frame_errors++; 1191 kfree_skb(skb); 1192 return -EINVAL; 1193 } 1194 1195 /* Header must be checked, and gso_segs computed. */ 1196 skb_shinfo(skb)->gso_type |= SKB_GSO_DODGY; 1197 skb_shinfo(skb)->gso_segs = 0; 1198 } 1199 1200 /* copy skb_ubuf_info for callback when skb has no error */ 1201 if (zerocopy) { 1202 skb_shinfo(skb)->destructor_arg = msg_control; 1203 skb_shinfo(skb)->tx_flags |= SKBTX_DEV_ZEROCOPY; 1204 skb_shinfo(skb)->tx_flags |= SKBTX_SHARED_FRAG; 1205 } 1206 1207 skb_reset_network_header(skb); 1208 rxhash = skb_get_rxhash(skb); 1209 netif_rx_ni(skb); 1210 1211 tun->dev->stats.rx_packets++; 1212 tun->dev->stats.rx_bytes += len; 1213 1214 tun_flow_update(tun, rxhash, tfile); 1215 return total_len; 1216 } 1217 1218 static ssize_t tun_chr_aio_write(struct kiocb *iocb, const struct iovec *iv, 1219 unsigned long count, loff_t pos) 1220 { 1221 struct file *file = iocb->ki_filp; 1222 struct tun_struct *tun = tun_get(file); 1223 struct tun_file *tfile = file->private_data; 1224 ssize_t result; 1225 1226 if (!tun) 1227 return -EBADFD; 1228 1229 tun_debug(KERN_INFO, tun, "tun_chr_write %ld\n", count); 1230 1231 result = tun_get_user(tun, tfile, NULL, iv, iov_length(iv, count), 1232 count, file->f_flags & O_NONBLOCK); 1233 1234 tun_put(tun); 1235 return result; 1236 } 1237 1238 /* Put packet to the user space buffer */ 1239 static ssize_t tun_put_user(struct tun_struct *tun, 1240 struct tun_file *tfile, 1241 struct sk_buff *skb, 1242 const struct iovec *iv, int len) 1243 { 1244 struct tun_pi pi = { 0, skb->protocol }; 1245 ssize_t total = 0; 1246 1247 if (!(tun->flags & TUN_NO_PI)) { 1248 if ((len -= sizeof(pi)) < 0) 1249 return -EINVAL; 1250 1251 if (len < skb->len) { 1252 /* Packet will be striped */ 1253 pi.flags |= TUN_PKT_STRIP; 1254 } 1255 1256 if (memcpy_toiovecend(iv, (void *) &pi, 0, sizeof(pi))) 1257 return -EFAULT; 1258 total += sizeof(pi); 1259 } 1260 1261 if (tun->flags & TUN_VNET_HDR) { 1262 struct virtio_net_hdr gso = { 0 }; /* no info leak */ 1263 if ((len -= tun->vnet_hdr_sz) < 0) 1264 return -EINVAL; 1265 1266 if (skb_is_gso(skb)) { 1267 struct skb_shared_info *sinfo = skb_shinfo(skb); 1268 1269 /* This is a hint as to how much should be linear. */ 1270 gso.hdr_len = skb_headlen(skb); 1271 gso.gso_size = sinfo->gso_size; 1272 if (sinfo->gso_type & SKB_GSO_TCPV4) 1273 gso.gso_type = VIRTIO_NET_HDR_GSO_TCPV4; 1274 else if (sinfo->gso_type & SKB_GSO_TCPV6) 1275 gso.gso_type = VIRTIO_NET_HDR_GSO_TCPV6; 1276 else if (sinfo->gso_type & SKB_GSO_UDP) 1277 gso.gso_type = VIRTIO_NET_HDR_GSO_UDP; 1278 else { 1279 pr_err("unexpected GSO type: " 1280 "0x%x, gso_size %d, hdr_len %d\n", 1281 sinfo->gso_type, gso.gso_size, 1282 gso.hdr_len); 1283 print_hex_dump(KERN_ERR, "tun: ", 1284 DUMP_PREFIX_NONE, 1285 16, 1, skb->head, 1286 min((int)gso.hdr_len, 64), true); 1287 WARN_ON_ONCE(1); 1288 return -EINVAL; 1289 } 1290 if (sinfo->gso_type & SKB_GSO_TCP_ECN) 1291 gso.gso_type |= VIRTIO_NET_HDR_GSO_ECN; 1292 } else 1293 gso.gso_type = VIRTIO_NET_HDR_GSO_NONE; 1294 1295 if (skb->ip_summed == CHECKSUM_PARTIAL) { 1296 gso.flags = VIRTIO_NET_HDR_F_NEEDS_CSUM; 1297 gso.csum_start = skb_checksum_start_offset(skb); 1298 gso.csum_offset = skb->csum_offset; 1299 } else if (skb->ip_summed == CHECKSUM_UNNECESSARY) { 1300 gso.flags = VIRTIO_NET_HDR_F_DATA_VALID; 1301 } /* else everything is zero */ 1302 1303 if (unlikely(memcpy_toiovecend(iv, (void *)&gso, total, 1304 sizeof(gso)))) 1305 return -EFAULT; 1306 total += tun->vnet_hdr_sz; 1307 } 1308 1309 len = min_t(int, skb->len, len); 1310 1311 skb_copy_datagram_const_iovec(skb, 0, iv, total, len); 1312 total += skb->len; 1313 1314 tun->dev->stats.tx_packets++; 1315 tun->dev->stats.tx_bytes += len; 1316 1317 return total; 1318 } 1319 1320 static ssize_t tun_do_read(struct tun_struct *tun, struct tun_file *tfile, 1321 struct kiocb *iocb, const struct iovec *iv, 1322 ssize_t len, int noblock) 1323 { 1324 DECLARE_WAITQUEUE(wait, current); 1325 struct sk_buff *skb; 1326 ssize_t ret = 0; 1327 1328 tun_debug(KERN_INFO, tun, "tun_do_read\n"); 1329 1330 if (unlikely(!noblock)) 1331 add_wait_queue(&tfile->wq.wait, &wait); 1332 while (len) { 1333 current->state = TASK_INTERRUPTIBLE; 1334 1335 /* Read frames from the queue */ 1336 if (!(skb = skb_dequeue(&tfile->socket.sk->sk_receive_queue))) { 1337 if (noblock) { 1338 ret = -EAGAIN; 1339 break; 1340 } 1341 if (signal_pending(current)) { 1342 ret = -ERESTARTSYS; 1343 break; 1344 } 1345 if (tun->dev->reg_state != NETREG_REGISTERED) { 1346 ret = -EIO; 1347 break; 1348 } 1349 1350 /* Nothing to read, let's sleep */ 1351 schedule(); 1352 continue; 1353 } 1354 1355 ret = tun_put_user(tun, tfile, skb, iv, len); 1356 kfree_skb(skb); 1357 break; 1358 } 1359 1360 current->state = TASK_RUNNING; 1361 if (unlikely(!noblock)) 1362 remove_wait_queue(&tfile->wq.wait, &wait); 1363 1364 return ret; 1365 } 1366 1367 static ssize_t tun_chr_aio_read(struct kiocb *iocb, const struct iovec *iv, 1368 unsigned long count, loff_t pos) 1369 { 1370 struct file *file = iocb->ki_filp; 1371 struct tun_file *tfile = file->private_data; 1372 struct tun_struct *tun = __tun_get(tfile); 1373 ssize_t len, ret; 1374 1375 if (!tun) 1376 return -EBADFD; 1377 len = iov_length(iv, count); 1378 if (len < 0) { 1379 ret = -EINVAL; 1380 goto out; 1381 } 1382 1383 ret = tun_do_read(tun, tfile, iocb, iv, len, 1384 file->f_flags & O_NONBLOCK); 1385 ret = min_t(ssize_t, ret, len); 1386 out: 1387 tun_put(tun); 1388 return ret; 1389 } 1390 1391 static void tun_free_netdev(struct net_device *dev) 1392 { 1393 struct tun_struct *tun = netdev_priv(dev); 1394 1395 BUG_ON(!(list_empty(&tun->disabled))); 1396 tun_flow_uninit(tun); 1397 security_tun_dev_free_security(tun->security); 1398 free_netdev(dev); 1399 } 1400 1401 static void tun_setup(struct net_device *dev) 1402 { 1403 struct tun_struct *tun = netdev_priv(dev); 1404 1405 tun->owner = INVALID_UID; 1406 tun->group = INVALID_GID; 1407 1408 dev->ethtool_ops = &tun_ethtool_ops; 1409 dev->destructor = tun_free_netdev; 1410 } 1411 1412 /* Trivial set of netlink ops to allow deleting tun or tap 1413 * device with netlink. 1414 */ 1415 static int tun_validate(struct nlattr *tb[], struct nlattr *data[]) 1416 { 1417 return -EINVAL; 1418 } 1419 1420 static struct rtnl_link_ops tun_link_ops __read_mostly = { 1421 .kind = DRV_NAME, 1422 .priv_size = sizeof(struct tun_struct), 1423 .setup = tun_setup, 1424 .validate = tun_validate, 1425 }; 1426 1427 static void tun_sock_write_space(struct sock *sk) 1428 { 1429 struct tun_file *tfile; 1430 wait_queue_head_t *wqueue; 1431 1432 if (!sock_writeable(sk)) 1433 return; 1434 1435 if (!test_and_clear_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags)) 1436 return; 1437 1438 wqueue = sk_sleep(sk); 1439 if (wqueue && waitqueue_active(wqueue)) 1440 wake_up_interruptible_sync_poll(wqueue, POLLOUT | 1441 POLLWRNORM | POLLWRBAND); 1442 1443 tfile = container_of(sk, struct tun_file, sk); 1444 kill_fasync(&tfile->fasync, SIGIO, POLL_OUT); 1445 } 1446 1447 static int tun_sendmsg(struct kiocb *iocb, struct socket *sock, 1448 struct msghdr *m, size_t total_len) 1449 { 1450 int ret; 1451 struct tun_file *tfile = container_of(sock, struct tun_file, socket); 1452 struct tun_struct *tun = __tun_get(tfile); 1453 1454 if (!tun) 1455 return -EBADFD; 1456 ret = tun_get_user(tun, tfile, m->msg_control, m->msg_iov, total_len, 1457 m->msg_iovlen, m->msg_flags & MSG_DONTWAIT); 1458 tun_put(tun); 1459 return ret; 1460 } 1461 1462 1463 static int tun_recvmsg(struct kiocb *iocb, struct socket *sock, 1464 struct msghdr *m, size_t total_len, 1465 int flags) 1466 { 1467 struct tun_file *tfile = container_of(sock, struct tun_file, socket); 1468 struct tun_struct *tun = __tun_get(tfile); 1469 int ret; 1470 1471 if (!tun) 1472 return -EBADFD; 1473 1474 if (flags & ~(MSG_DONTWAIT|MSG_TRUNC)) 1475 return -EINVAL; 1476 ret = tun_do_read(tun, tfile, iocb, m->msg_iov, total_len, 1477 flags & MSG_DONTWAIT); 1478 if (ret > total_len) { 1479 m->msg_flags |= MSG_TRUNC; 1480 ret = flags & MSG_TRUNC ? ret : total_len; 1481 } 1482 tun_put(tun); 1483 return ret; 1484 } 1485 1486 static int tun_release(struct socket *sock) 1487 { 1488 if (sock->sk) 1489 sock_put(sock->sk); 1490 return 0; 1491 } 1492 1493 /* Ops structure to mimic raw sockets with tun */ 1494 static const struct proto_ops tun_socket_ops = { 1495 .sendmsg = tun_sendmsg, 1496 .recvmsg = tun_recvmsg, 1497 .release = tun_release, 1498 }; 1499 1500 static struct proto tun_proto = { 1501 .name = "tun", 1502 .owner = THIS_MODULE, 1503 .obj_size = sizeof(struct tun_file), 1504 }; 1505 1506 static int tun_flags(struct tun_struct *tun) 1507 { 1508 int flags = 0; 1509 1510 if (tun->flags & TUN_TUN_DEV) 1511 flags |= IFF_TUN; 1512 else 1513 flags |= IFF_TAP; 1514 1515 if (tun->flags & TUN_NO_PI) 1516 flags |= IFF_NO_PI; 1517 1518 /* This flag has no real effect. We track the value for backwards 1519 * compatibility. 1520 */ 1521 if (tun->flags & TUN_ONE_QUEUE) 1522 flags |= IFF_ONE_QUEUE; 1523 1524 if (tun->flags & TUN_VNET_HDR) 1525 flags |= IFF_VNET_HDR; 1526 1527 if (tun->flags & TUN_TAP_MQ) 1528 flags |= IFF_MULTI_QUEUE; 1529 1530 return flags; 1531 } 1532 1533 static ssize_t tun_show_flags(struct device *dev, struct device_attribute *attr, 1534 char *buf) 1535 { 1536 struct tun_struct *tun = netdev_priv(to_net_dev(dev)); 1537 return sprintf(buf, "0x%x\n", tun_flags(tun)); 1538 } 1539 1540 static ssize_t tun_show_owner(struct device *dev, struct device_attribute *attr, 1541 char *buf) 1542 { 1543 struct tun_struct *tun = netdev_priv(to_net_dev(dev)); 1544 return uid_valid(tun->owner)? 1545 sprintf(buf, "%u\n", 1546 from_kuid_munged(current_user_ns(), tun->owner)): 1547 sprintf(buf, "-1\n"); 1548 } 1549 1550 static ssize_t tun_show_group(struct device *dev, struct device_attribute *attr, 1551 char *buf) 1552 { 1553 struct tun_struct *tun = netdev_priv(to_net_dev(dev)); 1554 return gid_valid(tun->group) ? 1555 sprintf(buf, "%u\n", 1556 from_kgid_munged(current_user_ns(), tun->group)): 1557 sprintf(buf, "-1\n"); 1558 } 1559 1560 static DEVICE_ATTR(tun_flags, 0444, tun_show_flags, NULL); 1561 static DEVICE_ATTR(owner, 0444, tun_show_owner, NULL); 1562 static DEVICE_ATTR(group, 0444, tun_show_group, NULL); 1563 1564 static int tun_set_iff(struct net *net, struct file *file, struct ifreq *ifr) 1565 { 1566 struct tun_struct *tun; 1567 struct tun_file *tfile = file->private_data; 1568 struct net_device *dev; 1569 int err; 1570 1571 if (tfile->detached) 1572 return -EINVAL; 1573 1574 dev = __dev_get_by_name(net, ifr->ifr_name); 1575 if (dev) { 1576 if (ifr->ifr_flags & IFF_TUN_EXCL) 1577 return -EBUSY; 1578 if ((ifr->ifr_flags & IFF_TUN) && dev->netdev_ops == &tun_netdev_ops) 1579 tun = netdev_priv(dev); 1580 else if ((ifr->ifr_flags & IFF_TAP) && dev->netdev_ops == &tap_netdev_ops) 1581 tun = netdev_priv(dev); 1582 else 1583 return -EINVAL; 1584 1585 if (tun_not_capable(tun)) 1586 return -EPERM; 1587 err = security_tun_dev_open(tun->security); 1588 if (err < 0) 1589 return err; 1590 1591 err = tun_attach(tun, file); 1592 if (err < 0) 1593 return err; 1594 1595 if (tun->flags & TUN_TAP_MQ && 1596 (tun->numqueues + tun->numdisabled > 1)) 1597 return err; 1598 } 1599 else { 1600 char *name; 1601 unsigned long flags = 0; 1602 int queues = ifr->ifr_flags & IFF_MULTI_QUEUE ? 1603 MAX_TAP_QUEUES : 1; 1604 1605 if (!ns_capable(net->user_ns, CAP_NET_ADMIN)) 1606 return -EPERM; 1607 err = security_tun_dev_create(); 1608 if (err < 0) 1609 return err; 1610 1611 /* Set dev type */ 1612 if (ifr->ifr_flags & IFF_TUN) { 1613 /* TUN device */ 1614 flags |= TUN_TUN_DEV; 1615 name = "tun%d"; 1616 } else if (ifr->ifr_flags & IFF_TAP) { 1617 /* TAP device */ 1618 flags |= TUN_TAP_DEV; 1619 name = "tap%d"; 1620 } else 1621 return -EINVAL; 1622 1623 if (*ifr->ifr_name) 1624 name = ifr->ifr_name; 1625 1626 dev = alloc_netdev_mqs(sizeof(struct tun_struct), name, 1627 tun_setup, queues, queues); 1628 1629 if (!dev) 1630 return -ENOMEM; 1631 1632 dev_net_set(dev, net); 1633 dev->rtnl_link_ops = &tun_link_ops; 1634 1635 tun = netdev_priv(dev); 1636 tun->dev = dev; 1637 tun->flags = flags; 1638 tun->txflt.count = 0; 1639 tun->vnet_hdr_sz = sizeof(struct virtio_net_hdr); 1640 1641 tun->filter_attached = false; 1642 tun->sndbuf = tfile->socket.sk->sk_sndbuf; 1643 1644 spin_lock_init(&tun->lock); 1645 1646 err = security_tun_dev_alloc_security(&tun->security); 1647 if (err < 0) 1648 goto err_free_dev; 1649 1650 tun_net_init(dev); 1651 1652 err = tun_flow_init(tun); 1653 if (err < 0) 1654 goto err_free_dev; 1655 1656 dev->hw_features = NETIF_F_SG | NETIF_F_FRAGLIST | 1657 TUN_USER_FEATURES; 1658 dev->features = dev->hw_features; 1659 1660 INIT_LIST_HEAD(&tun->disabled); 1661 err = tun_attach(tun, file); 1662 if (err < 0) 1663 goto err_free_dev; 1664 1665 err = register_netdevice(tun->dev); 1666 if (err < 0) 1667 goto err_free_dev; 1668 1669 if (device_create_file(&tun->dev->dev, &dev_attr_tun_flags) || 1670 device_create_file(&tun->dev->dev, &dev_attr_owner) || 1671 device_create_file(&tun->dev->dev, &dev_attr_group)) 1672 pr_err("Failed to create tun sysfs files\n"); 1673 } 1674 1675 netif_carrier_on(tun->dev); 1676 1677 tun_debug(KERN_INFO, tun, "tun_set_iff\n"); 1678 1679 if (ifr->ifr_flags & IFF_NO_PI) 1680 tun->flags |= TUN_NO_PI; 1681 else 1682 tun->flags &= ~TUN_NO_PI; 1683 1684 /* This flag has no real effect. We track the value for backwards 1685 * compatibility. 1686 */ 1687 if (ifr->ifr_flags & IFF_ONE_QUEUE) 1688 tun->flags |= TUN_ONE_QUEUE; 1689 else 1690 tun->flags &= ~TUN_ONE_QUEUE; 1691 1692 if (ifr->ifr_flags & IFF_VNET_HDR) 1693 tun->flags |= TUN_VNET_HDR; 1694 else 1695 tun->flags &= ~TUN_VNET_HDR; 1696 1697 if (ifr->ifr_flags & IFF_MULTI_QUEUE) 1698 tun->flags |= TUN_TAP_MQ; 1699 else 1700 tun->flags &= ~TUN_TAP_MQ; 1701 1702 /* Make sure persistent devices do not get stuck in 1703 * xoff state. 1704 */ 1705 if (netif_running(tun->dev)) 1706 netif_tx_wake_all_queues(tun->dev); 1707 1708 strcpy(ifr->ifr_name, tun->dev->name); 1709 return 0; 1710 1711 err_free_dev: 1712 free_netdev(dev); 1713 return err; 1714 } 1715 1716 static void tun_get_iff(struct net *net, struct tun_struct *tun, 1717 struct ifreq *ifr) 1718 { 1719 tun_debug(KERN_INFO, tun, "tun_get_iff\n"); 1720 1721 strcpy(ifr->ifr_name, tun->dev->name); 1722 1723 ifr->ifr_flags = tun_flags(tun); 1724 1725 } 1726 1727 /* This is like a cut-down ethtool ops, except done via tun fd so no 1728 * privs required. */ 1729 static int set_offload(struct tun_struct *tun, unsigned long arg) 1730 { 1731 netdev_features_t features = 0; 1732 1733 if (arg & TUN_F_CSUM) { 1734 features |= NETIF_F_HW_CSUM; 1735 arg &= ~TUN_F_CSUM; 1736 1737 if (arg & (TUN_F_TSO4|TUN_F_TSO6)) { 1738 if (arg & TUN_F_TSO_ECN) { 1739 features |= NETIF_F_TSO_ECN; 1740 arg &= ~TUN_F_TSO_ECN; 1741 } 1742 if (arg & TUN_F_TSO4) 1743 features |= NETIF_F_TSO; 1744 if (arg & TUN_F_TSO6) 1745 features |= NETIF_F_TSO6; 1746 arg &= ~(TUN_F_TSO4|TUN_F_TSO6); 1747 } 1748 1749 if (arg & TUN_F_UFO) { 1750 features |= NETIF_F_UFO; 1751 arg &= ~TUN_F_UFO; 1752 } 1753 } 1754 1755 /* This gives the user a way to test for new features in future by 1756 * trying to set them. */ 1757 if (arg) 1758 return -EINVAL; 1759 1760 tun->set_features = features; 1761 netdev_update_features(tun->dev); 1762 1763 return 0; 1764 } 1765 1766 static void tun_detach_filter(struct tun_struct *tun, int n) 1767 { 1768 int i; 1769 struct tun_file *tfile; 1770 1771 for (i = 0; i < n; i++) { 1772 tfile = rtnl_dereference(tun->tfiles[i]); 1773 sk_detach_filter(tfile->socket.sk); 1774 } 1775 1776 tun->filter_attached = false; 1777 } 1778 1779 static int tun_attach_filter(struct tun_struct *tun) 1780 { 1781 int i, ret = 0; 1782 struct tun_file *tfile; 1783 1784 for (i = 0; i < tun->numqueues; i++) { 1785 tfile = rtnl_dereference(tun->tfiles[i]); 1786 ret = sk_attach_filter(&tun->fprog, tfile->socket.sk); 1787 if (ret) { 1788 tun_detach_filter(tun, i); 1789 return ret; 1790 } 1791 } 1792 1793 tun->filter_attached = true; 1794 return ret; 1795 } 1796 1797 static void tun_set_sndbuf(struct tun_struct *tun) 1798 { 1799 struct tun_file *tfile; 1800 int i; 1801 1802 for (i = 0; i < tun->numqueues; i++) { 1803 tfile = rtnl_dereference(tun->tfiles[i]); 1804 tfile->socket.sk->sk_sndbuf = tun->sndbuf; 1805 } 1806 } 1807 1808 static int tun_set_queue(struct file *file, struct ifreq *ifr) 1809 { 1810 struct tun_file *tfile = file->private_data; 1811 struct tun_struct *tun; 1812 int ret = 0; 1813 1814 rtnl_lock(); 1815 1816 if (ifr->ifr_flags & IFF_ATTACH_QUEUE) { 1817 tun = tfile->detached; 1818 if (!tun) { 1819 ret = -EINVAL; 1820 goto unlock; 1821 } 1822 ret = security_tun_dev_attach_queue(tun->security); 1823 if (ret < 0) 1824 goto unlock; 1825 ret = tun_attach(tun, file); 1826 } else if (ifr->ifr_flags & IFF_DETACH_QUEUE) { 1827 tun = rtnl_dereference(tfile->tun); 1828 if (!tun || !(tun->flags & TUN_TAP_MQ) || tfile->detached) 1829 ret = -EINVAL; 1830 else 1831 __tun_detach(tfile, false); 1832 } else 1833 ret = -EINVAL; 1834 1835 unlock: 1836 rtnl_unlock(); 1837 return ret; 1838 } 1839 1840 static long __tun_chr_ioctl(struct file *file, unsigned int cmd, 1841 unsigned long arg, int ifreq_len) 1842 { 1843 struct tun_file *tfile = file->private_data; 1844 struct tun_struct *tun; 1845 void __user* argp = (void __user*)arg; 1846 struct ifreq ifr; 1847 kuid_t owner; 1848 kgid_t group; 1849 int sndbuf; 1850 int vnet_hdr_sz; 1851 int ret; 1852 1853 if (cmd == TUNSETIFF || cmd == TUNSETQUEUE || _IOC_TYPE(cmd) == 0x89) { 1854 if (copy_from_user(&ifr, argp, ifreq_len)) 1855 return -EFAULT; 1856 } else { 1857 memset(&ifr, 0, sizeof(ifr)); 1858 } 1859 if (cmd == TUNGETFEATURES) { 1860 /* Currently this just means: "what IFF flags are valid?". 1861 * This is needed because we never checked for invalid flags on 1862 * TUNSETIFF. */ 1863 return put_user(IFF_TUN | IFF_TAP | IFF_NO_PI | IFF_ONE_QUEUE | 1864 IFF_VNET_HDR | IFF_MULTI_QUEUE, 1865 (unsigned int __user*)argp); 1866 } else if (cmd == TUNSETQUEUE) 1867 return tun_set_queue(file, &ifr); 1868 1869 ret = 0; 1870 rtnl_lock(); 1871 1872 tun = __tun_get(tfile); 1873 if (cmd == TUNSETIFF && !tun) { 1874 ifr.ifr_name[IFNAMSIZ-1] = '\0'; 1875 1876 ret = tun_set_iff(tfile->net, file, &ifr); 1877 1878 if (ret) 1879 goto unlock; 1880 1881 if (copy_to_user(argp, &ifr, ifreq_len)) 1882 ret = -EFAULT; 1883 goto unlock; 1884 } 1885 1886 ret = -EBADFD; 1887 if (!tun) 1888 goto unlock; 1889 1890 tun_debug(KERN_INFO, tun, "tun_chr_ioctl cmd %u\n", cmd); 1891 1892 ret = 0; 1893 switch (cmd) { 1894 case TUNGETIFF: 1895 tun_get_iff(current->nsproxy->net_ns, tun, &ifr); 1896 1897 if (copy_to_user(argp, &ifr, ifreq_len)) 1898 ret = -EFAULT; 1899 break; 1900 1901 case TUNSETNOCSUM: 1902 /* Disable/Enable checksum */ 1903 1904 /* [unimplemented] */ 1905 tun_debug(KERN_INFO, tun, "ignored: set checksum %s\n", 1906 arg ? "disabled" : "enabled"); 1907 break; 1908 1909 case TUNSETPERSIST: 1910 /* Disable/Enable persist mode. Keep an extra reference to the 1911 * module to prevent the module being unprobed. 1912 */ 1913 if (arg && !(tun->flags & TUN_PERSIST)) { 1914 tun->flags |= TUN_PERSIST; 1915 __module_get(THIS_MODULE); 1916 } 1917 if (!arg && (tun->flags & TUN_PERSIST)) { 1918 tun->flags &= ~TUN_PERSIST; 1919 module_put(THIS_MODULE); 1920 } 1921 1922 tun_debug(KERN_INFO, tun, "persist %s\n", 1923 arg ? "enabled" : "disabled"); 1924 break; 1925 1926 case TUNSETOWNER: 1927 /* Set owner of the device */ 1928 owner = make_kuid(current_user_ns(), arg); 1929 if (!uid_valid(owner)) { 1930 ret = -EINVAL; 1931 break; 1932 } 1933 tun->owner = owner; 1934 tun_debug(KERN_INFO, tun, "owner set to %u\n", 1935 from_kuid(&init_user_ns, tun->owner)); 1936 break; 1937 1938 case TUNSETGROUP: 1939 /* Set group of the device */ 1940 group = make_kgid(current_user_ns(), arg); 1941 if (!gid_valid(group)) { 1942 ret = -EINVAL; 1943 break; 1944 } 1945 tun->group = group; 1946 tun_debug(KERN_INFO, tun, "group set to %u\n", 1947 from_kgid(&init_user_ns, tun->group)); 1948 break; 1949 1950 case TUNSETLINK: 1951 /* Only allow setting the type when the interface is down */ 1952 if (tun->dev->flags & IFF_UP) { 1953 tun_debug(KERN_INFO, tun, 1954 "Linktype set failed because interface is up\n"); 1955 ret = -EBUSY; 1956 } else { 1957 tun->dev->type = (int) arg; 1958 tun_debug(KERN_INFO, tun, "linktype set to %d\n", 1959 tun->dev->type); 1960 ret = 0; 1961 } 1962 break; 1963 1964 #ifdef TUN_DEBUG 1965 case TUNSETDEBUG: 1966 tun->debug = arg; 1967 break; 1968 #endif 1969 case TUNSETOFFLOAD: 1970 ret = set_offload(tun, arg); 1971 break; 1972 1973 case TUNSETTXFILTER: 1974 /* Can be set only for TAPs */ 1975 ret = -EINVAL; 1976 if ((tun->flags & TUN_TYPE_MASK) != TUN_TAP_DEV) 1977 break; 1978 ret = update_filter(&tun->txflt, (void __user *)arg); 1979 break; 1980 1981 case SIOCGIFHWADDR: 1982 /* Get hw address */ 1983 memcpy(ifr.ifr_hwaddr.sa_data, tun->dev->dev_addr, ETH_ALEN); 1984 ifr.ifr_hwaddr.sa_family = tun->dev->type; 1985 if (copy_to_user(argp, &ifr, ifreq_len)) 1986 ret = -EFAULT; 1987 break; 1988 1989 case SIOCSIFHWADDR: 1990 /* Set hw address */ 1991 tun_debug(KERN_DEBUG, tun, "set hw address: %pM\n", 1992 ifr.ifr_hwaddr.sa_data); 1993 1994 ret = dev_set_mac_address(tun->dev, &ifr.ifr_hwaddr); 1995 break; 1996 1997 case TUNGETSNDBUF: 1998 sndbuf = tfile->socket.sk->sk_sndbuf; 1999 if (copy_to_user(argp, &sndbuf, sizeof(sndbuf))) 2000 ret = -EFAULT; 2001 break; 2002 2003 case TUNSETSNDBUF: 2004 if (copy_from_user(&sndbuf, argp, sizeof(sndbuf))) { 2005 ret = -EFAULT; 2006 break; 2007 } 2008 2009 tun->sndbuf = sndbuf; 2010 tun_set_sndbuf(tun); 2011 break; 2012 2013 case TUNGETVNETHDRSZ: 2014 vnet_hdr_sz = tun->vnet_hdr_sz; 2015 if (copy_to_user(argp, &vnet_hdr_sz, sizeof(vnet_hdr_sz))) 2016 ret = -EFAULT; 2017 break; 2018 2019 case TUNSETVNETHDRSZ: 2020 if (copy_from_user(&vnet_hdr_sz, argp, sizeof(vnet_hdr_sz))) { 2021 ret = -EFAULT; 2022 break; 2023 } 2024 if (vnet_hdr_sz < (int)sizeof(struct virtio_net_hdr)) { 2025 ret = -EINVAL; 2026 break; 2027 } 2028 2029 tun->vnet_hdr_sz = vnet_hdr_sz; 2030 break; 2031 2032 case TUNATTACHFILTER: 2033 /* Can be set only for TAPs */ 2034 ret = -EINVAL; 2035 if ((tun->flags & TUN_TYPE_MASK) != TUN_TAP_DEV) 2036 break; 2037 ret = -EFAULT; 2038 if (copy_from_user(&tun->fprog, argp, sizeof(tun->fprog))) 2039 break; 2040 2041 ret = tun_attach_filter(tun); 2042 break; 2043 2044 case TUNDETACHFILTER: 2045 /* Can be set only for TAPs */ 2046 ret = -EINVAL; 2047 if ((tun->flags & TUN_TYPE_MASK) != TUN_TAP_DEV) 2048 break; 2049 ret = 0; 2050 tun_detach_filter(tun, tun->numqueues); 2051 break; 2052 2053 default: 2054 ret = -EINVAL; 2055 break; 2056 } 2057 2058 unlock: 2059 rtnl_unlock(); 2060 if (tun) 2061 tun_put(tun); 2062 return ret; 2063 } 2064 2065 static long tun_chr_ioctl(struct file *file, 2066 unsigned int cmd, unsigned long arg) 2067 { 2068 return __tun_chr_ioctl(file, cmd, arg, sizeof (struct ifreq)); 2069 } 2070 2071 #ifdef CONFIG_COMPAT 2072 static long tun_chr_compat_ioctl(struct file *file, 2073 unsigned int cmd, unsigned long arg) 2074 { 2075 switch (cmd) { 2076 case TUNSETIFF: 2077 case TUNGETIFF: 2078 case TUNSETTXFILTER: 2079 case TUNGETSNDBUF: 2080 case TUNSETSNDBUF: 2081 case SIOCGIFHWADDR: 2082 case SIOCSIFHWADDR: 2083 arg = (unsigned long)compat_ptr(arg); 2084 break; 2085 default: 2086 arg = (compat_ulong_t)arg; 2087 break; 2088 } 2089 2090 /* 2091 * compat_ifreq is shorter than ifreq, so we must not access beyond 2092 * the end of that structure. All fields that are used in this 2093 * driver are compatible though, we don't need to convert the 2094 * contents. 2095 */ 2096 return __tun_chr_ioctl(file, cmd, arg, sizeof(struct compat_ifreq)); 2097 } 2098 #endif /* CONFIG_COMPAT */ 2099 2100 static int tun_chr_fasync(int fd, struct file *file, int on) 2101 { 2102 struct tun_file *tfile = file->private_data; 2103 int ret; 2104 2105 if ((ret = fasync_helper(fd, file, on, &tfile->fasync)) < 0) 2106 goto out; 2107 2108 if (on) { 2109 ret = __f_setown(file, task_pid(current), PIDTYPE_PID, 0); 2110 if (ret) 2111 goto out; 2112 tfile->flags |= TUN_FASYNC; 2113 } else 2114 tfile->flags &= ~TUN_FASYNC; 2115 ret = 0; 2116 out: 2117 return ret; 2118 } 2119 2120 static int tun_chr_open(struct inode *inode, struct file * file) 2121 { 2122 struct tun_file *tfile; 2123 2124 DBG1(KERN_INFO, "tunX: tun_chr_open\n"); 2125 2126 tfile = (struct tun_file *)sk_alloc(&init_net, AF_UNSPEC, GFP_KERNEL, 2127 &tun_proto); 2128 if (!tfile) 2129 return -ENOMEM; 2130 rcu_assign_pointer(tfile->tun, NULL); 2131 tfile->net = get_net(current->nsproxy->net_ns); 2132 tfile->flags = 0; 2133 2134 rcu_assign_pointer(tfile->socket.wq, &tfile->wq); 2135 init_waitqueue_head(&tfile->wq.wait); 2136 2137 tfile->socket.file = file; 2138 tfile->socket.ops = &tun_socket_ops; 2139 2140 sock_init_data(&tfile->socket, &tfile->sk); 2141 sk_change_net(&tfile->sk, tfile->net); 2142 2143 tfile->sk.sk_write_space = tun_sock_write_space; 2144 tfile->sk.sk_sndbuf = INT_MAX; 2145 2146 file->private_data = tfile; 2147 set_bit(SOCK_EXTERNALLY_ALLOCATED, &tfile->socket.flags); 2148 INIT_LIST_HEAD(&tfile->next); 2149 2150 return 0; 2151 } 2152 2153 static int tun_chr_close(struct inode *inode, struct file *file) 2154 { 2155 struct tun_file *tfile = file->private_data; 2156 struct net *net = tfile->net; 2157 2158 tun_detach(tfile, true); 2159 put_net(net); 2160 2161 return 0; 2162 } 2163 2164 static const struct file_operations tun_fops = { 2165 .owner = THIS_MODULE, 2166 .llseek = no_llseek, 2167 .read = do_sync_read, 2168 .aio_read = tun_chr_aio_read, 2169 .write = do_sync_write, 2170 .aio_write = tun_chr_aio_write, 2171 .poll = tun_chr_poll, 2172 .unlocked_ioctl = tun_chr_ioctl, 2173 #ifdef CONFIG_COMPAT 2174 .compat_ioctl = tun_chr_compat_ioctl, 2175 #endif 2176 .open = tun_chr_open, 2177 .release = tun_chr_close, 2178 .fasync = tun_chr_fasync 2179 }; 2180 2181 static struct miscdevice tun_miscdev = { 2182 .minor = TUN_MINOR, 2183 .name = "tun", 2184 .nodename = "net/tun", 2185 .fops = &tun_fops, 2186 }; 2187 2188 /* ethtool interface */ 2189 2190 static int tun_get_settings(struct net_device *dev, struct ethtool_cmd *cmd) 2191 { 2192 cmd->supported = 0; 2193 cmd->advertising = 0; 2194 ethtool_cmd_speed_set(cmd, SPEED_10); 2195 cmd->duplex = DUPLEX_FULL; 2196 cmd->port = PORT_TP; 2197 cmd->phy_address = 0; 2198 cmd->transceiver = XCVR_INTERNAL; 2199 cmd->autoneg = AUTONEG_DISABLE; 2200 cmd->maxtxpkt = 0; 2201 cmd->maxrxpkt = 0; 2202 return 0; 2203 } 2204 2205 static void tun_get_drvinfo(struct net_device *dev, struct ethtool_drvinfo *info) 2206 { 2207 struct tun_struct *tun = netdev_priv(dev); 2208 2209 strlcpy(info->driver, DRV_NAME, sizeof(info->driver)); 2210 strlcpy(info->version, DRV_VERSION, sizeof(info->version)); 2211 2212 switch (tun->flags & TUN_TYPE_MASK) { 2213 case TUN_TUN_DEV: 2214 strlcpy(info->bus_info, "tun", sizeof(info->bus_info)); 2215 break; 2216 case TUN_TAP_DEV: 2217 strlcpy(info->bus_info, "tap", sizeof(info->bus_info)); 2218 break; 2219 } 2220 } 2221 2222 static u32 tun_get_msglevel(struct net_device *dev) 2223 { 2224 #ifdef TUN_DEBUG 2225 struct tun_struct *tun = netdev_priv(dev); 2226 return tun->debug; 2227 #else 2228 return -EOPNOTSUPP; 2229 #endif 2230 } 2231 2232 static void tun_set_msglevel(struct net_device *dev, u32 value) 2233 { 2234 #ifdef TUN_DEBUG 2235 struct tun_struct *tun = netdev_priv(dev); 2236 tun->debug = value; 2237 #endif 2238 } 2239 2240 static const struct ethtool_ops tun_ethtool_ops = { 2241 .get_settings = tun_get_settings, 2242 .get_drvinfo = tun_get_drvinfo, 2243 .get_msglevel = tun_get_msglevel, 2244 .set_msglevel = tun_set_msglevel, 2245 .get_link = ethtool_op_get_link, 2246 }; 2247 2248 2249 static int __init tun_init(void) 2250 { 2251 int ret = 0; 2252 2253 pr_info("%s, %s\n", DRV_DESCRIPTION, DRV_VERSION); 2254 pr_info("%s\n", DRV_COPYRIGHT); 2255 2256 ret = rtnl_link_register(&tun_link_ops); 2257 if (ret) { 2258 pr_err("Can't register link_ops\n"); 2259 goto err_linkops; 2260 } 2261 2262 ret = misc_register(&tun_miscdev); 2263 if (ret) { 2264 pr_err("Can't register misc device %d\n", TUN_MINOR); 2265 goto err_misc; 2266 } 2267 return 0; 2268 err_misc: 2269 rtnl_link_unregister(&tun_link_ops); 2270 err_linkops: 2271 return ret; 2272 } 2273 2274 static void tun_cleanup(void) 2275 { 2276 misc_deregister(&tun_miscdev); 2277 rtnl_link_unregister(&tun_link_ops); 2278 } 2279 2280 /* Get an underlying socket object from tun file. Returns error unless file is 2281 * attached to a device. The returned object works like a packet socket, it 2282 * can be used for sock_sendmsg/sock_recvmsg. The caller is responsible for 2283 * holding a reference to the file for as long as the socket is in use. */ 2284 struct socket *tun_get_socket(struct file *file) 2285 { 2286 struct tun_file *tfile; 2287 if (file->f_op != &tun_fops) 2288 return ERR_PTR(-EINVAL); 2289 tfile = file->private_data; 2290 if (!tfile) 2291 return ERR_PTR(-EBADFD); 2292 return &tfile->socket; 2293 } 2294 EXPORT_SYMBOL_GPL(tun_get_socket); 2295 2296 module_init(tun_init); 2297 module_exit(tun_cleanup); 2298 MODULE_DESCRIPTION(DRV_DESCRIPTION); 2299 MODULE_AUTHOR(DRV_COPYRIGHT); 2300 MODULE_LICENSE("GPL"); 2301 MODULE_ALIAS_MISCDEV(TUN_MINOR); 2302 MODULE_ALIAS("devname:net/tun"); 2303