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/if_vlan.h> 64 #include <linux/crc32.h> 65 #include <linux/nsproxy.h> 66 #include <linux/virtio_net.h> 67 #include <linux/rcupdate.h> 68 #include <net/net_namespace.h> 69 #include <net/netns/generic.h> 70 #include <net/rtnetlink.h> 71 #include <net/sock.h> 72 #include <linux/seq_file.h> 73 #include <linux/uio.h> 74 75 #include <asm/uaccess.h> 76 77 /* Uncomment to enable debugging */ 78 /* #define TUN_DEBUG 1 */ 79 80 #ifdef TUN_DEBUG 81 static int debug; 82 83 #define tun_debug(level, tun, fmt, args...) \ 84 do { \ 85 if (tun->debug) \ 86 netdev_printk(level, tun->dev, fmt, ##args); \ 87 } while (0) 88 #define DBG1(level, fmt, args...) \ 89 do { \ 90 if (debug == 2) \ 91 printk(level fmt, ##args); \ 92 } while (0) 93 #else 94 #define tun_debug(level, tun, fmt, args...) \ 95 do { \ 96 if (0) \ 97 netdev_printk(level, tun->dev, fmt, ##args); \ 98 } while (0) 99 #define DBG1(level, fmt, args...) \ 100 do { \ 101 if (0) \ 102 printk(level fmt, ##args); \ 103 } while (0) 104 #endif 105 106 /* TUN device flags */ 107 108 /* IFF_ATTACH_QUEUE is never stored in device flags, 109 * overload it to mean fasync when stored there. 110 */ 111 #define TUN_FASYNC IFF_ATTACH_QUEUE 112 /* High bits in flags field are unused. */ 113 #define TUN_VNET_LE 0x80000000 114 #define TUN_VNET_BE 0x40000000 115 116 #define TUN_FEATURES (IFF_NO_PI | IFF_ONE_QUEUE | IFF_VNET_HDR | \ 117 IFF_MULTI_QUEUE) 118 #define GOODCOPY_LEN 128 119 120 #define FLT_EXACT_COUNT 8 121 struct tap_filter { 122 unsigned int count; /* Number of addrs. Zero means disabled */ 123 u32 mask[2]; /* Mask of the hashed addrs */ 124 unsigned char addr[FLT_EXACT_COUNT][ETH_ALEN]; 125 }; 126 127 /* MAX_TAP_QUEUES 256 is chosen to allow rx/tx queues to be equal 128 * to max number of VCPUs in guest. */ 129 #define MAX_TAP_QUEUES 256 130 #define MAX_TAP_FLOWS 4096 131 132 #define TUN_FLOW_EXPIRE (3 * HZ) 133 134 /* A tun_file connects an open character device to a tuntap netdevice. It 135 * also contains all socket related structures (except sock_fprog and tap_filter) 136 * to serve as one transmit queue for tuntap device. The sock_fprog and 137 * tap_filter were kept in tun_struct since they were used for filtering for the 138 * netdevice not for a specific queue (at least I didn't see the requirement for 139 * this). 140 * 141 * RCU usage: 142 * The tun_file and tun_struct are loosely coupled, the pointer from one to the 143 * other can only be read while rcu_read_lock or rtnl_lock is held. 144 */ 145 struct tun_file { 146 struct sock sk; 147 struct socket socket; 148 struct socket_wq wq; 149 struct tun_struct __rcu *tun; 150 struct fasync_struct *fasync; 151 /* only used for fasnyc */ 152 unsigned int flags; 153 union { 154 u16 queue_index; 155 unsigned int ifindex; 156 }; 157 struct list_head next; 158 struct tun_struct *detached; 159 }; 160 161 struct tun_flow_entry { 162 struct hlist_node hash_link; 163 struct rcu_head rcu; 164 struct tun_struct *tun; 165 166 u32 rxhash; 167 u32 rps_rxhash; 168 int queue_index; 169 unsigned long updated; 170 }; 171 172 #define TUN_NUM_FLOW_ENTRIES 1024 173 174 /* Since the socket were moved to tun_file, to preserve the behavior of persist 175 * device, socket filter, sndbuf and vnet header size were restore when the 176 * file were attached to a persist device. 177 */ 178 struct tun_struct { 179 struct tun_file __rcu *tfiles[MAX_TAP_QUEUES]; 180 unsigned int numqueues; 181 unsigned int flags; 182 kuid_t owner; 183 kgid_t group; 184 185 struct net_device *dev; 186 netdev_features_t set_features; 187 #define TUN_USER_FEATURES (NETIF_F_HW_CSUM|NETIF_F_TSO_ECN|NETIF_F_TSO| \ 188 NETIF_F_TSO6|NETIF_F_UFO) 189 190 int vnet_hdr_sz; 191 int sndbuf; 192 struct tap_filter txflt; 193 struct sock_fprog fprog; 194 /* protected by rtnl lock */ 195 bool filter_attached; 196 #ifdef TUN_DEBUG 197 int debug; 198 #endif 199 spinlock_t lock; 200 struct hlist_head flows[TUN_NUM_FLOW_ENTRIES]; 201 struct timer_list flow_gc_timer; 202 unsigned long ageing_time; 203 unsigned int numdisabled; 204 struct list_head disabled; 205 void *security; 206 u32 flow_count; 207 }; 208 209 #ifdef CONFIG_TUN_VNET_CROSS_LE 210 static inline bool tun_legacy_is_little_endian(struct tun_struct *tun) 211 { 212 return tun->flags & TUN_VNET_BE ? false : 213 virtio_legacy_is_little_endian(); 214 } 215 216 static long tun_get_vnet_be(struct tun_struct *tun, int __user *argp) 217 { 218 int be = !!(tun->flags & TUN_VNET_BE); 219 220 if (put_user(be, argp)) 221 return -EFAULT; 222 223 return 0; 224 } 225 226 static long tun_set_vnet_be(struct tun_struct *tun, int __user *argp) 227 { 228 int be; 229 230 if (get_user(be, argp)) 231 return -EFAULT; 232 233 if (be) 234 tun->flags |= TUN_VNET_BE; 235 else 236 tun->flags &= ~TUN_VNET_BE; 237 238 return 0; 239 } 240 #else 241 static inline bool tun_legacy_is_little_endian(struct tun_struct *tun) 242 { 243 return virtio_legacy_is_little_endian(); 244 } 245 246 static long tun_get_vnet_be(struct tun_struct *tun, int __user *argp) 247 { 248 return -EINVAL; 249 } 250 251 static long tun_set_vnet_be(struct tun_struct *tun, int __user *argp) 252 { 253 return -EINVAL; 254 } 255 #endif /* CONFIG_TUN_VNET_CROSS_LE */ 256 257 static inline bool tun_is_little_endian(struct tun_struct *tun) 258 { 259 return tun->flags & TUN_VNET_LE || 260 tun_legacy_is_little_endian(tun); 261 } 262 263 static inline u16 tun16_to_cpu(struct tun_struct *tun, __virtio16 val) 264 { 265 return __virtio16_to_cpu(tun_is_little_endian(tun), val); 266 } 267 268 static inline __virtio16 cpu_to_tun16(struct tun_struct *tun, u16 val) 269 { 270 return __cpu_to_virtio16(tun_is_little_endian(tun), val); 271 } 272 273 static inline u32 tun_hashfn(u32 rxhash) 274 { 275 return rxhash & 0x3ff; 276 } 277 278 static struct tun_flow_entry *tun_flow_find(struct hlist_head *head, u32 rxhash) 279 { 280 struct tun_flow_entry *e; 281 282 hlist_for_each_entry_rcu(e, head, hash_link) { 283 if (e->rxhash == rxhash) 284 return e; 285 } 286 return NULL; 287 } 288 289 static struct tun_flow_entry *tun_flow_create(struct tun_struct *tun, 290 struct hlist_head *head, 291 u32 rxhash, u16 queue_index) 292 { 293 struct tun_flow_entry *e = kmalloc(sizeof(*e), GFP_ATOMIC); 294 295 if (e) { 296 tun_debug(KERN_INFO, tun, "create flow: hash %u index %u\n", 297 rxhash, queue_index); 298 e->updated = jiffies; 299 e->rxhash = rxhash; 300 e->rps_rxhash = 0; 301 e->queue_index = queue_index; 302 e->tun = tun; 303 hlist_add_head_rcu(&e->hash_link, head); 304 ++tun->flow_count; 305 } 306 return e; 307 } 308 309 static void tun_flow_delete(struct tun_struct *tun, struct tun_flow_entry *e) 310 { 311 tun_debug(KERN_INFO, tun, "delete flow: hash %u index %u\n", 312 e->rxhash, e->queue_index); 313 hlist_del_rcu(&e->hash_link); 314 kfree_rcu(e, rcu); 315 --tun->flow_count; 316 } 317 318 static void tun_flow_flush(struct tun_struct *tun) 319 { 320 int i; 321 322 spin_lock_bh(&tun->lock); 323 for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++) { 324 struct tun_flow_entry *e; 325 struct hlist_node *n; 326 327 hlist_for_each_entry_safe(e, n, &tun->flows[i], hash_link) 328 tun_flow_delete(tun, e); 329 } 330 spin_unlock_bh(&tun->lock); 331 } 332 333 static void tun_flow_delete_by_queue(struct tun_struct *tun, u16 queue_index) 334 { 335 int i; 336 337 spin_lock_bh(&tun->lock); 338 for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++) { 339 struct tun_flow_entry *e; 340 struct hlist_node *n; 341 342 hlist_for_each_entry_safe(e, n, &tun->flows[i], hash_link) { 343 if (e->queue_index == queue_index) 344 tun_flow_delete(tun, e); 345 } 346 } 347 spin_unlock_bh(&tun->lock); 348 } 349 350 static void tun_flow_cleanup(unsigned long data) 351 { 352 struct tun_struct *tun = (struct tun_struct *)data; 353 unsigned long delay = tun->ageing_time; 354 unsigned long next_timer = jiffies + delay; 355 unsigned long count = 0; 356 int i; 357 358 tun_debug(KERN_INFO, tun, "tun_flow_cleanup\n"); 359 360 spin_lock_bh(&tun->lock); 361 for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++) { 362 struct tun_flow_entry *e; 363 struct hlist_node *n; 364 365 hlist_for_each_entry_safe(e, n, &tun->flows[i], hash_link) { 366 unsigned long this_timer; 367 count++; 368 this_timer = e->updated + delay; 369 if (time_before_eq(this_timer, jiffies)) 370 tun_flow_delete(tun, e); 371 else if (time_before(this_timer, next_timer)) 372 next_timer = this_timer; 373 } 374 } 375 376 if (count) 377 mod_timer(&tun->flow_gc_timer, round_jiffies_up(next_timer)); 378 spin_unlock_bh(&tun->lock); 379 } 380 381 static void tun_flow_update(struct tun_struct *tun, u32 rxhash, 382 struct tun_file *tfile) 383 { 384 struct hlist_head *head; 385 struct tun_flow_entry *e; 386 unsigned long delay = tun->ageing_time; 387 u16 queue_index = tfile->queue_index; 388 389 if (!rxhash) 390 return; 391 else 392 head = &tun->flows[tun_hashfn(rxhash)]; 393 394 rcu_read_lock(); 395 396 /* We may get a very small possibility of OOO during switching, not 397 * worth to optimize.*/ 398 if (tun->numqueues == 1 || tfile->detached) 399 goto unlock; 400 401 e = tun_flow_find(head, rxhash); 402 if (likely(e)) { 403 /* TODO: keep queueing to old queue until it's empty? */ 404 e->queue_index = queue_index; 405 e->updated = jiffies; 406 sock_rps_record_flow_hash(e->rps_rxhash); 407 } else { 408 spin_lock_bh(&tun->lock); 409 if (!tun_flow_find(head, rxhash) && 410 tun->flow_count < MAX_TAP_FLOWS) 411 tun_flow_create(tun, head, rxhash, queue_index); 412 413 if (!timer_pending(&tun->flow_gc_timer)) 414 mod_timer(&tun->flow_gc_timer, 415 round_jiffies_up(jiffies + delay)); 416 spin_unlock_bh(&tun->lock); 417 } 418 419 unlock: 420 rcu_read_unlock(); 421 } 422 423 /** 424 * Save the hash received in the stack receive path and update the 425 * flow_hash table accordingly. 426 */ 427 static inline void tun_flow_save_rps_rxhash(struct tun_flow_entry *e, u32 hash) 428 { 429 if (unlikely(e->rps_rxhash != hash)) 430 e->rps_rxhash = hash; 431 } 432 433 /* We try to identify a flow through its rxhash first. The reason that 434 * we do not check rxq no. is because some cards(e.g 82599), chooses 435 * the rxq based on the txq where the last packet of the flow comes. As 436 * the userspace application move between processors, we may get a 437 * different rxq no. here. If we could not get rxhash, then we would 438 * hope the rxq no. may help here. 439 */ 440 static u16 tun_select_queue(struct net_device *dev, struct sk_buff *skb, 441 void *accel_priv, select_queue_fallback_t fallback) 442 { 443 struct tun_struct *tun = netdev_priv(dev); 444 struct tun_flow_entry *e; 445 u32 txq = 0; 446 u32 numqueues = 0; 447 448 rcu_read_lock(); 449 numqueues = ACCESS_ONCE(tun->numqueues); 450 451 txq = skb_get_hash(skb); 452 if (txq) { 453 e = tun_flow_find(&tun->flows[tun_hashfn(txq)], txq); 454 if (e) { 455 tun_flow_save_rps_rxhash(e, txq); 456 txq = e->queue_index; 457 } else 458 /* use multiply and shift instead of expensive divide */ 459 txq = ((u64)txq * numqueues) >> 32; 460 } else if (likely(skb_rx_queue_recorded(skb))) { 461 txq = skb_get_rx_queue(skb); 462 while (unlikely(txq >= numqueues)) 463 txq -= numqueues; 464 } 465 466 rcu_read_unlock(); 467 return txq; 468 } 469 470 static inline bool tun_not_capable(struct tun_struct *tun) 471 { 472 const struct cred *cred = current_cred(); 473 struct net *net = dev_net(tun->dev); 474 475 return ((uid_valid(tun->owner) && !uid_eq(cred->euid, tun->owner)) || 476 (gid_valid(tun->group) && !in_egroup_p(tun->group))) && 477 !ns_capable(net->user_ns, CAP_NET_ADMIN); 478 } 479 480 static void tun_set_real_num_queues(struct tun_struct *tun) 481 { 482 netif_set_real_num_tx_queues(tun->dev, tun->numqueues); 483 netif_set_real_num_rx_queues(tun->dev, tun->numqueues); 484 } 485 486 static void tun_disable_queue(struct tun_struct *tun, struct tun_file *tfile) 487 { 488 tfile->detached = tun; 489 list_add_tail(&tfile->next, &tun->disabled); 490 ++tun->numdisabled; 491 } 492 493 static struct tun_struct *tun_enable_queue(struct tun_file *tfile) 494 { 495 struct tun_struct *tun = tfile->detached; 496 497 tfile->detached = NULL; 498 list_del_init(&tfile->next); 499 --tun->numdisabled; 500 return tun; 501 } 502 503 static void tun_queue_purge(struct tun_file *tfile) 504 { 505 skb_queue_purge(&tfile->sk.sk_receive_queue); 506 skb_queue_purge(&tfile->sk.sk_error_queue); 507 } 508 509 static void __tun_detach(struct tun_file *tfile, bool clean) 510 { 511 struct tun_file *ntfile; 512 struct tun_struct *tun; 513 514 tun = rtnl_dereference(tfile->tun); 515 516 if (tun && !tfile->detached) { 517 u16 index = tfile->queue_index; 518 BUG_ON(index >= tun->numqueues); 519 520 rcu_assign_pointer(tun->tfiles[index], 521 tun->tfiles[tun->numqueues - 1]); 522 ntfile = rtnl_dereference(tun->tfiles[index]); 523 ntfile->queue_index = index; 524 525 --tun->numqueues; 526 if (clean) { 527 RCU_INIT_POINTER(tfile->tun, NULL); 528 sock_put(&tfile->sk); 529 } else 530 tun_disable_queue(tun, tfile); 531 532 synchronize_net(); 533 tun_flow_delete_by_queue(tun, tun->numqueues + 1); 534 /* Drop read queue */ 535 tun_queue_purge(tfile); 536 tun_set_real_num_queues(tun); 537 } else if (tfile->detached && clean) { 538 tun = tun_enable_queue(tfile); 539 sock_put(&tfile->sk); 540 } 541 542 if (clean) { 543 if (tun && tun->numqueues == 0 && tun->numdisabled == 0) { 544 netif_carrier_off(tun->dev); 545 546 if (!(tun->flags & IFF_PERSIST) && 547 tun->dev->reg_state == NETREG_REGISTERED) 548 unregister_netdevice(tun->dev); 549 } 550 sock_put(&tfile->sk); 551 } 552 } 553 554 static void tun_detach(struct tun_file *tfile, bool clean) 555 { 556 rtnl_lock(); 557 __tun_detach(tfile, clean); 558 rtnl_unlock(); 559 } 560 561 static void tun_detach_all(struct net_device *dev) 562 { 563 struct tun_struct *tun = netdev_priv(dev); 564 struct tun_file *tfile, *tmp; 565 int i, n = tun->numqueues; 566 567 for (i = 0; i < n; i++) { 568 tfile = rtnl_dereference(tun->tfiles[i]); 569 BUG_ON(!tfile); 570 tfile->socket.sk->sk_data_ready(tfile->socket.sk); 571 RCU_INIT_POINTER(tfile->tun, NULL); 572 --tun->numqueues; 573 } 574 list_for_each_entry(tfile, &tun->disabled, next) { 575 tfile->socket.sk->sk_data_ready(tfile->socket.sk); 576 RCU_INIT_POINTER(tfile->tun, NULL); 577 } 578 BUG_ON(tun->numqueues != 0); 579 580 synchronize_net(); 581 for (i = 0; i < n; i++) { 582 tfile = rtnl_dereference(tun->tfiles[i]); 583 /* Drop read queue */ 584 tun_queue_purge(tfile); 585 sock_put(&tfile->sk); 586 } 587 list_for_each_entry_safe(tfile, tmp, &tun->disabled, next) { 588 tun_enable_queue(tfile); 589 tun_queue_purge(tfile); 590 sock_put(&tfile->sk); 591 } 592 BUG_ON(tun->numdisabled != 0); 593 594 if (tun->flags & IFF_PERSIST) 595 module_put(THIS_MODULE); 596 } 597 598 static int tun_attach(struct tun_struct *tun, struct file *file, bool skip_filter) 599 { 600 struct tun_file *tfile = file->private_data; 601 int err; 602 603 err = security_tun_dev_attach(tfile->socket.sk, tun->security); 604 if (err < 0) 605 goto out; 606 607 err = -EINVAL; 608 if (rtnl_dereference(tfile->tun) && !tfile->detached) 609 goto out; 610 611 err = -EBUSY; 612 if (!(tun->flags & IFF_MULTI_QUEUE) && tun->numqueues == 1) 613 goto out; 614 615 err = -E2BIG; 616 if (!tfile->detached && 617 tun->numqueues + tun->numdisabled == MAX_TAP_QUEUES) 618 goto out; 619 620 err = 0; 621 622 /* Re-attach the filter to persist device */ 623 if (!skip_filter && (tun->filter_attached == true)) { 624 err = sk_attach_filter(&tun->fprog, tfile->socket.sk); 625 if (!err) 626 goto out; 627 } 628 tfile->queue_index = tun->numqueues; 629 rcu_assign_pointer(tfile->tun, tun); 630 rcu_assign_pointer(tun->tfiles[tun->numqueues], tfile); 631 tun->numqueues++; 632 633 if (tfile->detached) 634 tun_enable_queue(tfile); 635 else 636 sock_hold(&tfile->sk); 637 638 tun_set_real_num_queues(tun); 639 640 /* device is allowed to go away first, so no need to hold extra 641 * refcnt. 642 */ 643 644 out: 645 return err; 646 } 647 648 static struct tun_struct *__tun_get(struct tun_file *tfile) 649 { 650 struct tun_struct *tun; 651 652 rcu_read_lock(); 653 tun = rcu_dereference(tfile->tun); 654 if (tun) 655 dev_hold(tun->dev); 656 rcu_read_unlock(); 657 658 return tun; 659 } 660 661 static struct tun_struct *tun_get(struct file *file) 662 { 663 return __tun_get(file->private_data); 664 } 665 666 static void tun_put(struct tun_struct *tun) 667 { 668 dev_put(tun->dev); 669 } 670 671 /* TAP filtering */ 672 static void addr_hash_set(u32 *mask, const u8 *addr) 673 { 674 int n = ether_crc(ETH_ALEN, addr) >> 26; 675 mask[n >> 5] |= (1 << (n & 31)); 676 } 677 678 static unsigned int addr_hash_test(const u32 *mask, const u8 *addr) 679 { 680 int n = ether_crc(ETH_ALEN, addr) >> 26; 681 return mask[n >> 5] & (1 << (n & 31)); 682 } 683 684 static int update_filter(struct tap_filter *filter, void __user *arg) 685 { 686 struct { u8 u[ETH_ALEN]; } *addr; 687 struct tun_filter uf; 688 int err, alen, n, nexact; 689 690 if (copy_from_user(&uf, arg, sizeof(uf))) 691 return -EFAULT; 692 693 if (!uf.count) { 694 /* Disabled */ 695 filter->count = 0; 696 return 0; 697 } 698 699 alen = ETH_ALEN * uf.count; 700 addr = kmalloc(alen, GFP_KERNEL); 701 if (!addr) 702 return -ENOMEM; 703 704 if (copy_from_user(addr, arg + sizeof(uf), alen)) { 705 err = -EFAULT; 706 goto done; 707 } 708 709 /* The filter is updated without holding any locks. Which is 710 * perfectly safe. We disable it first and in the worst 711 * case we'll accept a few undesired packets. */ 712 filter->count = 0; 713 wmb(); 714 715 /* Use first set of addresses as an exact filter */ 716 for (n = 0; n < uf.count && n < FLT_EXACT_COUNT; n++) 717 memcpy(filter->addr[n], addr[n].u, ETH_ALEN); 718 719 nexact = n; 720 721 /* Remaining multicast addresses are hashed, 722 * unicast will leave the filter disabled. */ 723 memset(filter->mask, 0, sizeof(filter->mask)); 724 for (; n < uf.count; n++) { 725 if (!is_multicast_ether_addr(addr[n].u)) { 726 err = 0; /* no filter */ 727 goto done; 728 } 729 addr_hash_set(filter->mask, addr[n].u); 730 } 731 732 /* For ALLMULTI just set the mask to all ones. 733 * This overrides the mask populated above. */ 734 if ((uf.flags & TUN_FLT_ALLMULTI)) 735 memset(filter->mask, ~0, sizeof(filter->mask)); 736 737 /* Now enable the filter */ 738 wmb(); 739 filter->count = nexact; 740 741 /* Return the number of exact filters */ 742 err = nexact; 743 744 done: 745 kfree(addr); 746 return err; 747 } 748 749 /* Returns: 0 - drop, !=0 - accept */ 750 static int run_filter(struct tap_filter *filter, const struct sk_buff *skb) 751 { 752 /* Cannot use eth_hdr(skb) here because skb_mac_hdr() is incorrect 753 * at this point. */ 754 struct ethhdr *eh = (struct ethhdr *) skb->data; 755 int i; 756 757 /* Exact match */ 758 for (i = 0; i < filter->count; i++) 759 if (ether_addr_equal(eh->h_dest, filter->addr[i])) 760 return 1; 761 762 /* Inexact match (multicast only) */ 763 if (is_multicast_ether_addr(eh->h_dest)) 764 return addr_hash_test(filter->mask, eh->h_dest); 765 766 return 0; 767 } 768 769 /* 770 * Checks whether the packet is accepted or not. 771 * Returns: 0 - drop, !=0 - accept 772 */ 773 static int check_filter(struct tap_filter *filter, const struct sk_buff *skb) 774 { 775 if (!filter->count) 776 return 1; 777 778 return run_filter(filter, skb); 779 } 780 781 /* Network device part of the driver */ 782 783 static const struct ethtool_ops tun_ethtool_ops; 784 785 /* Net device detach from fd. */ 786 static void tun_net_uninit(struct net_device *dev) 787 { 788 tun_detach_all(dev); 789 } 790 791 /* Net device open. */ 792 static int tun_net_open(struct net_device *dev) 793 { 794 netif_tx_start_all_queues(dev); 795 return 0; 796 } 797 798 /* Net device close. */ 799 static int tun_net_close(struct net_device *dev) 800 { 801 netif_tx_stop_all_queues(dev); 802 return 0; 803 } 804 805 /* Net device start xmit */ 806 static netdev_tx_t tun_net_xmit(struct sk_buff *skb, struct net_device *dev) 807 { 808 struct tun_struct *tun = netdev_priv(dev); 809 int txq = skb->queue_mapping; 810 struct tun_file *tfile; 811 u32 numqueues = 0; 812 813 rcu_read_lock(); 814 tfile = rcu_dereference(tun->tfiles[txq]); 815 numqueues = ACCESS_ONCE(tun->numqueues); 816 817 /* Drop packet if interface is not attached */ 818 if (txq >= numqueues) 819 goto drop; 820 821 if (numqueues == 1) { 822 /* Select queue was not called for the skbuff, so we extract the 823 * RPS hash and save it into the flow_table here. 824 */ 825 __u32 rxhash; 826 827 rxhash = skb_get_hash(skb); 828 if (rxhash) { 829 struct tun_flow_entry *e; 830 e = tun_flow_find(&tun->flows[tun_hashfn(rxhash)], 831 rxhash); 832 if (e) 833 tun_flow_save_rps_rxhash(e, rxhash); 834 } 835 } 836 837 tun_debug(KERN_INFO, tun, "tun_net_xmit %d\n", skb->len); 838 839 BUG_ON(!tfile); 840 841 /* Drop if the filter does not like it. 842 * This is a noop if the filter is disabled. 843 * Filter can be enabled only for the TAP devices. */ 844 if (!check_filter(&tun->txflt, skb)) 845 goto drop; 846 847 if (tfile->socket.sk->sk_filter && 848 sk_filter(tfile->socket.sk, skb)) 849 goto drop; 850 851 /* Limit the number of packets queued by dividing txq length with the 852 * number of queues. 853 */ 854 if (skb_queue_len(&tfile->socket.sk->sk_receive_queue) * numqueues 855 >= dev->tx_queue_len) 856 goto drop; 857 858 if (unlikely(skb_orphan_frags(skb, GFP_ATOMIC))) 859 goto drop; 860 861 if (skb->sk && sk_fullsock(skb->sk)) { 862 sock_tx_timestamp(skb->sk, &skb_shinfo(skb)->tx_flags); 863 sw_tx_timestamp(skb); 864 } 865 866 /* Orphan the skb - required as we might hang on to it 867 * for indefinite time. 868 */ 869 skb_orphan(skb); 870 871 nf_reset(skb); 872 873 /* Enqueue packet */ 874 skb_queue_tail(&tfile->socket.sk->sk_receive_queue, skb); 875 876 /* Notify and wake up reader process */ 877 if (tfile->flags & TUN_FASYNC) 878 kill_fasync(&tfile->fasync, SIGIO, POLL_IN); 879 tfile->socket.sk->sk_data_ready(tfile->socket.sk); 880 881 rcu_read_unlock(); 882 return NETDEV_TX_OK; 883 884 drop: 885 dev->stats.tx_dropped++; 886 skb_tx_error(skb); 887 kfree_skb(skb); 888 rcu_read_unlock(); 889 return NET_XMIT_DROP; 890 } 891 892 static void tun_net_mclist(struct net_device *dev) 893 { 894 /* 895 * This callback is supposed to deal with mc filter in 896 * _rx_ path and has nothing to do with the _tx_ path. 897 * In rx path we always accept everything userspace gives us. 898 */ 899 } 900 901 #define MIN_MTU 68 902 #define MAX_MTU 65535 903 904 static int 905 tun_net_change_mtu(struct net_device *dev, int new_mtu) 906 { 907 if (new_mtu < MIN_MTU || new_mtu + dev->hard_header_len > MAX_MTU) 908 return -EINVAL; 909 dev->mtu = new_mtu; 910 return 0; 911 } 912 913 static netdev_features_t tun_net_fix_features(struct net_device *dev, 914 netdev_features_t features) 915 { 916 struct tun_struct *tun = netdev_priv(dev); 917 918 return (features & tun->set_features) | (features & ~TUN_USER_FEATURES); 919 } 920 #ifdef CONFIG_NET_POLL_CONTROLLER 921 static void tun_poll_controller(struct net_device *dev) 922 { 923 /* 924 * Tun only receives frames when: 925 * 1) the char device endpoint gets data from user space 926 * 2) the tun socket gets a sendmsg call from user space 927 * Since both of those are synchronous operations, we are guaranteed 928 * never to have pending data when we poll for it 929 * so there is nothing to do here but return. 930 * We need this though so netpoll recognizes us as an interface that 931 * supports polling, which enables bridge devices in virt setups to 932 * still use netconsole 933 */ 934 return; 935 } 936 #endif 937 static const struct net_device_ops tun_netdev_ops = { 938 .ndo_uninit = tun_net_uninit, 939 .ndo_open = tun_net_open, 940 .ndo_stop = tun_net_close, 941 .ndo_start_xmit = tun_net_xmit, 942 .ndo_change_mtu = tun_net_change_mtu, 943 .ndo_fix_features = tun_net_fix_features, 944 .ndo_select_queue = tun_select_queue, 945 #ifdef CONFIG_NET_POLL_CONTROLLER 946 .ndo_poll_controller = tun_poll_controller, 947 #endif 948 }; 949 950 static const struct net_device_ops tap_netdev_ops = { 951 .ndo_uninit = tun_net_uninit, 952 .ndo_open = tun_net_open, 953 .ndo_stop = tun_net_close, 954 .ndo_start_xmit = tun_net_xmit, 955 .ndo_change_mtu = tun_net_change_mtu, 956 .ndo_fix_features = tun_net_fix_features, 957 .ndo_set_rx_mode = tun_net_mclist, 958 .ndo_set_mac_address = eth_mac_addr, 959 .ndo_validate_addr = eth_validate_addr, 960 .ndo_select_queue = tun_select_queue, 961 #ifdef CONFIG_NET_POLL_CONTROLLER 962 .ndo_poll_controller = tun_poll_controller, 963 #endif 964 .ndo_features_check = passthru_features_check, 965 }; 966 967 static void tun_flow_init(struct tun_struct *tun) 968 { 969 int i; 970 971 for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++) 972 INIT_HLIST_HEAD(&tun->flows[i]); 973 974 tun->ageing_time = TUN_FLOW_EXPIRE; 975 setup_timer(&tun->flow_gc_timer, tun_flow_cleanup, (unsigned long)tun); 976 mod_timer(&tun->flow_gc_timer, 977 round_jiffies_up(jiffies + tun->ageing_time)); 978 } 979 980 static void tun_flow_uninit(struct tun_struct *tun) 981 { 982 del_timer_sync(&tun->flow_gc_timer); 983 tun_flow_flush(tun); 984 } 985 986 /* Initialize net device. */ 987 static void tun_net_init(struct net_device *dev) 988 { 989 struct tun_struct *tun = netdev_priv(dev); 990 991 switch (tun->flags & TUN_TYPE_MASK) { 992 case IFF_TUN: 993 dev->netdev_ops = &tun_netdev_ops; 994 995 /* Point-to-Point TUN Device */ 996 dev->hard_header_len = 0; 997 dev->addr_len = 0; 998 dev->mtu = 1500; 999 1000 /* Zero header length */ 1001 dev->type = ARPHRD_NONE; 1002 dev->flags = IFF_POINTOPOINT | IFF_NOARP | IFF_MULTICAST; 1003 dev->tx_queue_len = TUN_READQ_SIZE; /* We prefer our own queue length */ 1004 break; 1005 1006 case IFF_TAP: 1007 dev->netdev_ops = &tap_netdev_ops; 1008 /* Ethernet TAP Device */ 1009 ether_setup(dev); 1010 dev->priv_flags &= ~IFF_TX_SKB_SHARING; 1011 dev->priv_flags |= IFF_LIVE_ADDR_CHANGE; 1012 1013 eth_hw_addr_random(dev); 1014 1015 dev->tx_queue_len = TUN_READQ_SIZE; /* We prefer our own queue length */ 1016 break; 1017 } 1018 } 1019 1020 /* Character device part */ 1021 1022 /* Poll */ 1023 static unsigned int tun_chr_poll(struct file *file, poll_table *wait) 1024 { 1025 struct tun_file *tfile = file->private_data; 1026 struct tun_struct *tun = __tun_get(tfile); 1027 struct sock *sk; 1028 unsigned int mask = 0; 1029 1030 if (!tun) 1031 return POLLERR; 1032 1033 sk = tfile->socket.sk; 1034 1035 tun_debug(KERN_INFO, tun, "tun_chr_poll\n"); 1036 1037 poll_wait(file, sk_sleep(sk), wait); 1038 1039 if (!skb_queue_empty(&sk->sk_receive_queue)) 1040 mask |= POLLIN | POLLRDNORM; 1041 1042 if (sock_writeable(sk) || 1043 (!test_and_set_bit(SOCKWQ_ASYNC_NOSPACE, &sk->sk_socket->flags) && 1044 sock_writeable(sk))) 1045 mask |= POLLOUT | POLLWRNORM; 1046 1047 if (tun->dev->reg_state != NETREG_REGISTERED) 1048 mask = POLLERR; 1049 1050 tun_put(tun); 1051 return mask; 1052 } 1053 1054 /* prepad is the amount to reserve at front. len is length after that. 1055 * linear is a hint as to how much to copy (usually headers). */ 1056 static struct sk_buff *tun_alloc_skb(struct tun_file *tfile, 1057 size_t prepad, size_t len, 1058 size_t linear, int noblock) 1059 { 1060 struct sock *sk = tfile->socket.sk; 1061 struct sk_buff *skb; 1062 int err; 1063 1064 /* Under a page? Don't bother with paged skb. */ 1065 if (prepad + len < PAGE_SIZE || !linear) 1066 linear = len; 1067 1068 skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock, 1069 &err, 0); 1070 if (!skb) 1071 return ERR_PTR(err); 1072 1073 skb_reserve(skb, prepad); 1074 skb_put(skb, linear); 1075 skb->data_len = len - linear; 1076 skb->len += len - linear; 1077 1078 return skb; 1079 } 1080 1081 /* Get packet from user space buffer */ 1082 static ssize_t tun_get_user(struct tun_struct *tun, struct tun_file *tfile, 1083 void *msg_control, struct iov_iter *from, 1084 int noblock) 1085 { 1086 struct tun_pi pi = { 0, cpu_to_be16(ETH_P_IP) }; 1087 struct sk_buff *skb; 1088 size_t total_len = iov_iter_count(from); 1089 size_t len = total_len, align = NET_SKB_PAD, linear; 1090 struct virtio_net_hdr gso = { 0 }; 1091 int good_linear; 1092 int copylen; 1093 bool zerocopy = false; 1094 int err; 1095 u32 rxhash; 1096 ssize_t n; 1097 1098 if (!(tun->dev->flags & IFF_UP)) 1099 return -EIO; 1100 1101 if (!(tun->flags & IFF_NO_PI)) { 1102 if (len < sizeof(pi)) 1103 return -EINVAL; 1104 len -= sizeof(pi); 1105 1106 n = copy_from_iter(&pi, sizeof(pi), from); 1107 if (n != sizeof(pi)) 1108 return -EFAULT; 1109 } 1110 1111 if (tun->flags & IFF_VNET_HDR) { 1112 if (len < tun->vnet_hdr_sz) 1113 return -EINVAL; 1114 len -= tun->vnet_hdr_sz; 1115 1116 n = copy_from_iter(&gso, sizeof(gso), from); 1117 if (n != sizeof(gso)) 1118 return -EFAULT; 1119 1120 if ((gso.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) && 1121 tun16_to_cpu(tun, gso.csum_start) + tun16_to_cpu(tun, gso.csum_offset) + 2 > tun16_to_cpu(tun, gso.hdr_len)) 1122 gso.hdr_len = cpu_to_tun16(tun, tun16_to_cpu(tun, gso.csum_start) + tun16_to_cpu(tun, gso.csum_offset) + 2); 1123 1124 if (tun16_to_cpu(tun, gso.hdr_len) > len) 1125 return -EINVAL; 1126 iov_iter_advance(from, tun->vnet_hdr_sz - sizeof(gso)); 1127 } 1128 1129 if ((tun->flags & TUN_TYPE_MASK) == IFF_TAP) { 1130 align += NET_IP_ALIGN; 1131 if (unlikely(len < ETH_HLEN || 1132 (gso.hdr_len && tun16_to_cpu(tun, gso.hdr_len) < ETH_HLEN))) 1133 return -EINVAL; 1134 } 1135 1136 good_linear = SKB_MAX_HEAD(align); 1137 1138 if (msg_control) { 1139 struct iov_iter i = *from; 1140 1141 /* There are 256 bytes to be copied in skb, so there is 1142 * enough room for skb expand head in case it is used. 1143 * The rest of the buffer is mapped from userspace. 1144 */ 1145 copylen = gso.hdr_len ? tun16_to_cpu(tun, gso.hdr_len) : GOODCOPY_LEN; 1146 if (copylen > good_linear) 1147 copylen = good_linear; 1148 linear = copylen; 1149 iov_iter_advance(&i, copylen); 1150 if (iov_iter_npages(&i, INT_MAX) <= MAX_SKB_FRAGS) 1151 zerocopy = true; 1152 } 1153 1154 if (!zerocopy) { 1155 copylen = len; 1156 if (tun16_to_cpu(tun, gso.hdr_len) > good_linear) 1157 linear = good_linear; 1158 else 1159 linear = tun16_to_cpu(tun, gso.hdr_len); 1160 } 1161 1162 skb = tun_alloc_skb(tfile, align, copylen, linear, noblock); 1163 if (IS_ERR(skb)) { 1164 if (PTR_ERR(skb) != -EAGAIN) 1165 tun->dev->stats.rx_dropped++; 1166 return PTR_ERR(skb); 1167 } 1168 1169 if (zerocopy) 1170 err = zerocopy_sg_from_iter(skb, from); 1171 else { 1172 err = skb_copy_datagram_from_iter(skb, 0, from, len); 1173 if (!err && msg_control) { 1174 struct ubuf_info *uarg = msg_control; 1175 uarg->callback(uarg, false); 1176 } 1177 } 1178 1179 if (err) { 1180 tun->dev->stats.rx_dropped++; 1181 kfree_skb(skb); 1182 return -EFAULT; 1183 } 1184 1185 if (gso.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) { 1186 if (!skb_partial_csum_set(skb, tun16_to_cpu(tun, gso.csum_start), 1187 tun16_to_cpu(tun, gso.csum_offset))) { 1188 tun->dev->stats.rx_frame_errors++; 1189 kfree_skb(skb); 1190 return -EINVAL; 1191 } 1192 } 1193 1194 switch (tun->flags & TUN_TYPE_MASK) { 1195 case IFF_TUN: 1196 if (tun->flags & IFF_NO_PI) { 1197 switch (skb->data[0] & 0xf0) { 1198 case 0x40: 1199 pi.proto = htons(ETH_P_IP); 1200 break; 1201 case 0x60: 1202 pi.proto = htons(ETH_P_IPV6); 1203 break; 1204 default: 1205 tun->dev->stats.rx_dropped++; 1206 kfree_skb(skb); 1207 return -EINVAL; 1208 } 1209 } 1210 1211 skb_reset_mac_header(skb); 1212 skb->protocol = pi.proto; 1213 skb->dev = tun->dev; 1214 break; 1215 case IFF_TAP: 1216 skb->protocol = eth_type_trans(skb, tun->dev); 1217 break; 1218 } 1219 1220 if (gso.gso_type != VIRTIO_NET_HDR_GSO_NONE) { 1221 pr_debug("GSO!\n"); 1222 switch (gso.gso_type & ~VIRTIO_NET_HDR_GSO_ECN) { 1223 case VIRTIO_NET_HDR_GSO_TCPV4: 1224 skb_shinfo(skb)->gso_type = SKB_GSO_TCPV4; 1225 break; 1226 case VIRTIO_NET_HDR_GSO_TCPV6: 1227 skb_shinfo(skb)->gso_type = SKB_GSO_TCPV6; 1228 break; 1229 case VIRTIO_NET_HDR_GSO_UDP: 1230 skb_shinfo(skb)->gso_type = SKB_GSO_UDP; 1231 break; 1232 default: 1233 tun->dev->stats.rx_frame_errors++; 1234 kfree_skb(skb); 1235 return -EINVAL; 1236 } 1237 1238 if (gso.gso_type & VIRTIO_NET_HDR_GSO_ECN) 1239 skb_shinfo(skb)->gso_type |= SKB_GSO_TCP_ECN; 1240 1241 skb_shinfo(skb)->gso_size = tun16_to_cpu(tun, gso.gso_size); 1242 if (skb_shinfo(skb)->gso_size == 0) { 1243 tun->dev->stats.rx_frame_errors++; 1244 kfree_skb(skb); 1245 return -EINVAL; 1246 } 1247 1248 /* Header must be checked, and gso_segs computed. */ 1249 skb_shinfo(skb)->gso_type |= SKB_GSO_DODGY; 1250 skb_shinfo(skb)->gso_segs = 0; 1251 } 1252 1253 /* copy skb_ubuf_info for callback when skb has no error */ 1254 if (zerocopy) { 1255 skb_shinfo(skb)->destructor_arg = msg_control; 1256 skb_shinfo(skb)->tx_flags |= SKBTX_DEV_ZEROCOPY; 1257 skb_shinfo(skb)->tx_flags |= SKBTX_SHARED_FRAG; 1258 } 1259 1260 skb_reset_network_header(skb); 1261 skb_probe_transport_header(skb, 0); 1262 1263 rxhash = skb_get_hash(skb); 1264 netif_rx_ni(skb); 1265 1266 tun->dev->stats.rx_packets++; 1267 tun->dev->stats.rx_bytes += len; 1268 1269 tun_flow_update(tun, rxhash, tfile); 1270 return total_len; 1271 } 1272 1273 static ssize_t tun_chr_write_iter(struct kiocb *iocb, struct iov_iter *from) 1274 { 1275 struct file *file = iocb->ki_filp; 1276 struct tun_struct *tun = tun_get(file); 1277 struct tun_file *tfile = file->private_data; 1278 ssize_t result; 1279 1280 if (!tun) 1281 return -EBADFD; 1282 1283 result = tun_get_user(tun, tfile, NULL, from, file->f_flags & O_NONBLOCK); 1284 1285 tun_put(tun); 1286 return result; 1287 } 1288 1289 /* Put packet to the user space buffer */ 1290 static ssize_t tun_put_user(struct tun_struct *tun, 1291 struct tun_file *tfile, 1292 struct sk_buff *skb, 1293 struct iov_iter *iter) 1294 { 1295 struct tun_pi pi = { 0, skb->protocol }; 1296 ssize_t total; 1297 int vlan_offset = 0; 1298 int vlan_hlen = 0; 1299 int vnet_hdr_sz = 0; 1300 1301 if (skb_vlan_tag_present(skb)) 1302 vlan_hlen = VLAN_HLEN; 1303 1304 if (tun->flags & IFF_VNET_HDR) 1305 vnet_hdr_sz = tun->vnet_hdr_sz; 1306 1307 total = skb->len + vlan_hlen + vnet_hdr_sz; 1308 1309 if (!(tun->flags & IFF_NO_PI)) { 1310 if (iov_iter_count(iter) < sizeof(pi)) 1311 return -EINVAL; 1312 1313 total += sizeof(pi); 1314 if (iov_iter_count(iter) < total) { 1315 /* Packet will be striped */ 1316 pi.flags |= TUN_PKT_STRIP; 1317 } 1318 1319 if (copy_to_iter(&pi, sizeof(pi), iter) != sizeof(pi)) 1320 return -EFAULT; 1321 } 1322 1323 if (vnet_hdr_sz) { 1324 struct virtio_net_hdr gso = { 0 }; /* no info leak */ 1325 if (iov_iter_count(iter) < vnet_hdr_sz) 1326 return -EINVAL; 1327 1328 if (skb_is_gso(skb)) { 1329 struct skb_shared_info *sinfo = skb_shinfo(skb); 1330 1331 /* This is a hint as to how much should be linear. */ 1332 gso.hdr_len = cpu_to_tun16(tun, skb_headlen(skb)); 1333 gso.gso_size = cpu_to_tun16(tun, sinfo->gso_size); 1334 if (sinfo->gso_type & SKB_GSO_TCPV4) 1335 gso.gso_type = VIRTIO_NET_HDR_GSO_TCPV4; 1336 else if (sinfo->gso_type & SKB_GSO_TCPV6) 1337 gso.gso_type = VIRTIO_NET_HDR_GSO_TCPV6; 1338 else if (sinfo->gso_type & SKB_GSO_UDP) 1339 gso.gso_type = VIRTIO_NET_HDR_GSO_UDP; 1340 else { 1341 pr_err("unexpected GSO type: " 1342 "0x%x, gso_size %d, hdr_len %d\n", 1343 sinfo->gso_type, tun16_to_cpu(tun, gso.gso_size), 1344 tun16_to_cpu(tun, gso.hdr_len)); 1345 print_hex_dump(KERN_ERR, "tun: ", 1346 DUMP_PREFIX_NONE, 1347 16, 1, skb->head, 1348 min((int)tun16_to_cpu(tun, gso.hdr_len), 64), true); 1349 WARN_ON_ONCE(1); 1350 return -EINVAL; 1351 } 1352 if (sinfo->gso_type & SKB_GSO_TCP_ECN) 1353 gso.gso_type |= VIRTIO_NET_HDR_GSO_ECN; 1354 } else 1355 gso.gso_type = VIRTIO_NET_HDR_GSO_NONE; 1356 1357 if (skb->ip_summed == CHECKSUM_PARTIAL) { 1358 gso.flags = VIRTIO_NET_HDR_F_NEEDS_CSUM; 1359 gso.csum_start = cpu_to_tun16(tun, skb_checksum_start_offset(skb) + 1360 vlan_hlen); 1361 gso.csum_offset = cpu_to_tun16(tun, skb->csum_offset); 1362 } else if (skb->ip_summed == CHECKSUM_UNNECESSARY) { 1363 gso.flags = VIRTIO_NET_HDR_F_DATA_VALID; 1364 } /* else everything is zero */ 1365 1366 if (copy_to_iter(&gso, sizeof(gso), iter) != sizeof(gso)) 1367 return -EFAULT; 1368 1369 iov_iter_advance(iter, vnet_hdr_sz - sizeof(gso)); 1370 } 1371 1372 if (vlan_hlen) { 1373 int ret; 1374 struct { 1375 __be16 h_vlan_proto; 1376 __be16 h_vlan_TCI; 1377 } veth; 1378 1379 veth.h_vlan_proto = skb->vlan_proto; 1380 veth.h_vlan_TCI = htons(skb_vlan_tag_get(skb)); 1381 1382 vlan_offset = offsetof(struct vlan_ethhdr, h_vlan_proto); 1383 1384 ret = skb_copy_datagram_iter(skb, 0, iter, vlan_offset); 1385 if (ret || !iov_iter_count(iter)) 1386 goto done; 1387 1388 ret = copy_to_iter(&veth, sizeof(veth), iter); 1389 if (ret != sizeof(veth) || !iov_iter_count(iter)) 1390 goto done; 1391 } 1392 1393 skb_copy_datagram_iter(skb, vlan_offset, iter, skb->len - vlan_offset); 1394 1395 done: 1396 tun->dev->stats.tx_packets++; 1397 tun->dev->stats.tx_bytes += skb->len + vlan_hlen; 1398 1399 return total; 1400 } 1401 1402 static ssize_t tun_do_read(struct tun_struct *tun, struct tun_file *tfile, 1403 struct iov_iter *to, 1404 int noblock) 1405 { 1406 struct sk_buff *skb; 1407 ssize_t ret; 1408 int peeked, err, off = 0; 1409 1410 tun_debug(KERN_INFO, tun, "tun_do_read\n"); 1411 1412 if (!iov_iter_count(to)) 1413 return 0; 1414 1415 if (tun->dev->reg_state != NETREG_REGISTERED) 1416 return -EIO; 1417 1418 /* Read frames from queue */ 1419 skb = __skb_recv_datagram(tfile->socket.sk, noblock ? MSG_DONTWAIT : 0, 1420 &peeked, &off, &err); 1421 if (!skb) 1422 return err; 1423 1424 ret = tun_put_user(tun, tfile, skb, to); 1425 if (unlikely(ret < 0)) 1426 kfree_skb(skb); 1427 else 1428 consume_skb(skb); 1429 1430 return ret; 1431 } 1432 1433 static ssize_t tun_chr_read_iter(struct kiocb *iocb, struct iov_iter *to) 1434 { 1435 struct file *file = iocb->ki_filp; 1436 struct tun_file *tfile = file->private_data; 1437 struct tun_struct *tun = __tun_get(tfile); 1438 ssize_t len = iov_iter_count(to), ret; 1439 1440 if (!tun) 1441 return -EBADFD; 1442 ret = tun_do_read(tun, tfile, to, file->f_flags & O_NONBLOCK); 1443 ret = min_t(ssize_t, ret, len); 1444 if (ret > 0) 1445 iocb->ki_pos = ret; 1446 tun_put(tun); 1447 return ret; 1448 } 1449 1450 static void tun_free_netdev(struct net_device *dev) 1451 { 1452 struct tun_struct *tun = netdev_priv(dev); 1453 1454 BUG_ON(!(list_empty(&tun->disabled))); 1455 tun_flow_uninit(tun); 1456 security_tun_dev_free_security(tun->security); 1457 free_netdev(dev); 1458 } 1459 1460 static void tun_setup(struct net_device *dev) 1461 { 1462 struct tun_struct *tun = netdev_priv(dev); 1463 1464 tun->owner = INVALID_UID; 1465 tun->group = INVALID_GID; 1466 1467 dev->ethtool_ops = &tun_ethtool_ops; 1468 dev->destructor = tun_free_netdev; 1469 } 1470 1471 /* Trivial set of netlink ops to allow deleting tun or tap 1472 * device with netlink. 1473 */ 1474 static int tun_validate(struct nlattr *tb[], struct nlattr *data[]) 1475 { 1476 return -EINVAL; 1477 } 1478 1479 static struct rtnl_link_ops tun_link_ops __read_mostly = { 1480 .kind = DRV_NAME, 1481 .priv_size = sizeof(struct tun_struct), 1482 .setup = tun_setup, 1483 .validate = tun_validate, 1484 }; 1485 1486 static void tun_sock_write_space(struct sock *sk) 1487 { 1488 struct tun_file *tfile; 1489 wait_queue_head_t *wqueue; 1490 1491 if (!sock_writeable(sk)) 1492 return; 1493 1494 if (!test_and_clear_bit(SOCKWQ_ASYNC_NOSPACE, &sk->sk_socket->flags)) 1495 return; 1496 1497 wqueue = sk_sleep(sk); 1498 if (wqueue && waitqueue_active(wqueue)) 1499 wake_up_interruptible_sync_poll(wqueue, POLLOUT | 1500 POLLWRNORM | POLLWRBAND); 1501 1502 tfile = container_of(sk, struct tun_file, sk); 1503 kill_fasync(&tfile->fasync, SIGIO, POLL_OUT); 1504 } 1505 1506 static int tun_sendmsg(struct socket *sock, struct msghdr *m, size_t total_len) 1507 { 1508 int ret; 1509 struct tun_file *tfile = container_of(sock, struct tun_file, socket); 1510 struct tun_struct *tun = __tun_get(tfile); 1511 1512 if (!tun) 1513 return -EBADFD; 1514 1515 ret = tun_get_user(tun, tfile, m->msg_control, &m->msg_iter, 1516 m->msg_flags & MSG_DONTWAIT); 1517 tun_put(tun); 1518 return ret; 1519 } 1520 1521 static int tun_recvmsg(struct socket *sock, struct msghdr *m, size_t total_len, 1522 int flags) 1523 { 1524 struct tun_file *tfile = container_of(sock, struct tun_file, socket); 1525 struct tun_struct *tun = __tun_get(tfile); 1526 int ret; 1527 1528 if (!tun) 1529 return -EBADFD; 1530 1531 if (flags & ~(MSG_DONTWAIT|MSG_TRUNC|MSG_ERRQUEUE)) { 1532 ret = -EINVAL; 1533 goto out; 1534 } 1535 if (flags & MSG_ERRQUEUE) { 1536 ret = sock_recv_errqueue(sock->sk, m, total_len, 1537 SOL_PACKET, TUN_TX_TIMESTAMP); 1538 goto out; 1539 } 1540 ret = tun_do_read(tun, tfile, &m->msg_iter, flags & MSG_DONTWAIT); 1541 if (ret > (ssize_t)total_len) { 1542 m->msg_flags |= MSG_TRUNC; 1543 ret = flags & MSG_TRUNC ? ret : total_len; 1544 } 1545 out: 1546 tun_put(tun); 1547 return ret; 1548 } 1549 1550 /* Ops structure to mimic raw sockets with tun */ 1551 static const struct proto_ops tun_socket_ops = { 1552 .sendmsg = tun_sendmsg, 1553 .recvmsg = tun_recvmsg, 1554 }; 1555 1556 static struct proto tun_proto = { 1557 .name = "tun", 1558 .owner = THIS_MODULE, 1559 .obj_size = sizeof(struct tun_file), 1560 }; 1561 1562 static int tun_flags(struct tun_struct *tun) 1563 { 1564 return tun->flags & (TUN_FEATURES | IFF_PERSIST | IFF_TUN | IFF_TAP); 1565 } 1566 1567 static ssize_t tun_show_flags(struct device *dev, struct device_attribute *attr, 1568 char *buf) 1569 { 1570 struct tun_struct *tun = netdev_priv(to_net_dev(dev)); 1571 return sprintf(buf, "0x%x\n", tun_flags(tun)); 1572 } 1573 1574 static ssize_t tun_show_owner(struct device *dev, struct device_attribute *attr, 1575 char *buf) 1576 { 1577 struct tun_struct *tun = netdev_priv(to_net_dev(dev)); 1578 return uid_valid(tun->owner)? 1579 sprintf(buf, "%u\n", 1580 from_kuid_munged(current_user_ns(), tun->owner)): 1581 sprintf(buf, "-1\n"); 1582 } 1583 1584 static ssize_t tun_show_group(struct device *dev, struct device_attribute *attr, 1585 char *buf) 1586 { 1587 struct tun_struct *tun = netdev_priv(to_net_dev(dev)); 1588 return gid_valid(tun->group) ? 1589 sprintf(buf, "%u\n", 1590 from_kgid_munged(current_user_ns(), tun->group)): 1591 sprintf(buf, "-1\n"); 1592 } 1593 1594 static DEVICE_ATTR(tun_flags, 0444, tun_show_flags, NULL); 1595 static DEVICE_ATTR(owner, 0444, tun_show_owner, NULL); 1596 static DEVICE_ATTR(group, 0444, tun_show_group, NULL); 1597 1598 static struct attribute *tun_dev_attrs[] = { 1599 &dev_attr_tun_flags.attr, 1600 &dev_attr_owner.attr, 1601 &dev_attr_group.attr, 1602 NULL 1603 }; 1604 1605 static const struct attribute_group tun_attr_group = { 1606 .attrs = tun_dev_attrs 1607 }; 1608 1609 static int tun_set_iff(struct net *net, struct file *file, struct ifreq *ifr) 1610 { 1611 struct tun_struct *tun; 1612 struct tun_file *tfile = file->private_data; 1613 struct net_device *dev; 1614 int err; 1615 1616 if (tfile->detached) 1617 return -EINVAL; 1618 1619 dev = __dev_get_by_name(net, ifr->ifr_name); 1620 if (dev) { 1621 if (ifr->ifr_flags & IFF_TUN_EXCL) 1622 return -EBUSY; 1623 if ((ifr->ifr_flags & IFF_TUN) && dev->netdev_ops == &tun_netdev_ops) 1624 tun = netdev_priv(dev); 1625 else if ((ifr->ifr_flags & IFF_TAP) && dev->netdev_ops == &tap_netdev_ops) 1626 tun = netdev_priv(dev); 1627 else 1628 return -EINVAL; 1629 1630 if (!!(ifr->ifr_flags & IFF_MULTI_QUEUE) != 1631 !!(tun->flags & IFF_MULTI_QUEUE)) 1632 return -EINVAL; 1633 1634 if (tun_not_capable(tun)) 1635 return -EPERM; 1636 err = security_tun_dev_open(tun->security); 1637 if (err < 0) 1638 return err; 1639 1640 err = tun_attach(tun, file, ifr->ifr_flags & IFF_NOFILTER); 1641 if (err < 0) 1642 return err; 1643 1644 if (tun->flags & IFF_MULTI_QUEUE && 1645 (tun->numqueues + tun->numdisabled > 1)) { 1646 /* One or more queue has already been attached, no need 1647 * to initialize the device again. 1648 */ 1649 return 0; 1650 } 1651 } 1652 else { 1653 char *name; 1654 unsigned long flags = 0; 1655 int queues = ifr->ifr_flags & IFF_MULTI_QUEUE ? 1656 MAX_TAP_QUEUES : 1; 1657 1658 if (!ns_capable(net->user_ns, CAP_NET_ADMIN)) 1659 return -EPERM; 1660 err = security_tun_dev_create(); 1661 if (err < 0) 1662 return err; 1663 1664 /* Set dev type */ 1665 if (ifr->ifr_flags & IFF_TUN) { 1666 /* TUN device */ 1667 flags |= IFF_TUN; 1668 name = "tun%d"; 1669 } else if (ifr->ifr_flags & IFF_TAP) { 1670 /* TAP device */ 1671 flags |= IFF_TAP; 1672 name = "tap%d"; 1673 } else 1674 return -EINVAL; 1675 1676 if (*ifr->ifr_name) 1677 name = ifr->ifr_name; 1678 1679 dev = alloc_netdev_mqs(sizeof(struct tun_struct), name, 1680 NET_NAME_UNKNOWN, tun_setup, queues, 1681 queues); 1682 1683 if (!dev) 1684 return -ENOMEM; 1685 1686 dev_net_set(dev, net); 1687 dev->rtnl_link_ops = &tun_link_ops; 1688 dev->ifindex = tfile->ifindex; 1689 dev->sysfs_groups[0] = &tun_attr_group; 1690 1691 tun = netdev_priv(dev); 1692 tun->dev = dev; 1693 tun->flags = flags; 1694 tun->txflt.count = 0; 1695 tun->vnet_hdr_sz = sizeof(struct virtio_net_hdr); 1696 1697 tun->filter_attached = false; 1698 tun->sndbuf = tfile->socket.sk->sk_sndbuf; 1699 1700 spin_lock_init(&tun->lock); 1701 1702 err = security_tun_dev_alloc_security(&tun->security); 1703 if (err < 0) 1704 goto err_free_dev; 1705 1706 tun_net_init(dev); 1707 tun_flow_init(tun); 1708 1709 dev->hw_features = NETIF_F_SG | NETIF_F_FRAGLIST | 1710 TUN_USER_FEATURES | NETIF_F_HW_VLAN_CTAG_TX | 1711 NETIF_F_HW_VLAN_STAG_TX; 1712 dev->features = dev->hw_features; 1713 dev->vlan_features = dev->features & 1714 ~(NETIF_F_HW_VLAN_CTAG_TX | 1715 NETIF_F_HW_VLAN_STAG_TX); 1716 1717 INIT_LIST_HEAD(&tun->disabled); 1718 err = tun_attach(tun, file, false); 1719 if (err < 0) 1720 goto err_free_flow; 1721 1722 err = register_netdevice(tun->dev); 1723 if (err < 0) 1724 goto err_detach; 1725 } 1726 1727 netif_carrier_on(tun->dev); 1728 1729 tun_debug(KERN_INFO, tun, "tun_set_iff\n"); 1730 1731 tun->flags = (tun->flags & ~TUN_FEATURES) | 1732 (ifr->ifr_flags & TUN_FEATURES); 1733 1734 /* Make sure persistent devices do not get stuck in 1735 * xoff state. 1736 */ 1737 if (netif_running(tun->dev)) 1738 netif_tx_wake_all_queues(tun->dev); 1739 1740 strcpy(ifr->ifr_name, tun->dev->name); 1741 return 0; 1742 1743 err_detach: 1744 tun_detach_all(dev); 1745 err_free_flow: 1746 tun_flow_uninit(tun); 1747 security_tun_dev_free_security(tun->security); 1748 err_free_dev: 1749 free_netdev(dev); 1750 return err; 1751 } 1752 1753 static void tun_get_iff(struct net *net, struct tun_struct *tun, 1754 struct ifreq *ifr) 1755 { 1756 tun_debug(KERN_INFO, tun, "tun_get_iff\n"); 1757 1758 strcpy(ifr->ifr_name, tun->dev->name); 1759 1760 ifr->ifr_flags = tun_flags(tun); 1761 1762 } 1763 1764 /* This is like a cut-down ethtool ops, except done via tun fd so no 1765 * privs required. */ 1766 static int set_offload(struct tun_struct *tun, unsigned long arg) 1767 { 1768 netdev_features_t features = 0; 1769 1770 if (arg & TUN_F_CSUM) { 1771 features |= NETIF_F_HW_CSUM; 1772 arg &= ~TUN_F_CSUM; 1773 1774 if (arg & (TUN_F_TSO4|TUN_F_TSO6)) { 1775 if (arg & TUN_F_TSO_ECN) { 1776 features |= NETIF_F_TSO_ECN; 1777 arg &= ~TUN_F_TSO_ECN; 1778 } 1779 if (arg & TUN_F_TSO4) 1780 features |= NETIF_F_TSO; 1781 if (arg & TUN_F_TSO6) 1782 features |= NETIF_F_TSO6; 1783 arg &= ~(TUN_F_TSO4|TUN_F_TSO6); 1784 } 1785 1786 if (arg & TUN_F_UFO) { 1787 features |= NETIF_F_UFO; 1788 arg &= ~TUN_F_UFO; 1789 } 1790 } 1791 1792 /* This gives the user a way to test for new features in future by 1793 * trying to set them. */ 1794 if (arg) 1795 return -EINVAL; 1796 1797 tun->set_features = features; 1798 netdev_update_features(tun->dev); 1799 1800 return 0; 1801 } 1802 1803 static void tun_detach_filter(struct tun_struct *tun, int n) 1804 { 1805 int i; 1806 struct tun_file *tfile; 1807 1808 for (i = 0; i < n; i++) { 1809 tfile = rtnl_dereference(tun->tfiles[i]); 1810 sk_detach_filter(tfile->socket.sk); 1811 } 1812 1813 tun->filter_attached = false; 1814 } 1815 1816 static int tun_attach_filter(struct tun_struct *tun) 1817 { 1818 int i, ret = 0; 1819 struct tun_file *tfile; 1820 1821 for (i = 0; i < tun->numqueues; i++) { 1822 tfile = rtnl_dereference(tun->tfiles[i]); 1823 ret = sk_attach_filter(&tun->fprog, tfile->socket.sk); 1824 if (ret) { 1825 tun_detach_filter(tun, i); 1826 return ret; 1827 } 1828 } 1829 1830 tun->filter_attached = true; 1831 return ret; 1832 } 1833 1834 static void tun_set_sndbuf(struct tun_struct *tun) 1835 { 1836 struct tun_file *tfile; 1837 int i; 1838 1839 for (i = 0; i < tun->numqueues; i++) { 1840 tfile = rtnl_dereference(tun->tfiles[i]); 1841 tfile->socket.sk->sk_sndbuf = tun->sndbuf; 1842 } 1843 } 1844 1845 static int tun_set_queue(struct file *file, struct ifreq *ifr) 1846 { 1847 struct tun_file *tfile = file->private_data; 1848 struct tun_struct *tun; 1849 int ret = 0; 1850 1851 rtnl_lock(); 1852 1853 if (ifr->ifr_flags & IFF_ATTACH_QUEUE) { 1854 tun = tfile->detached; 1855 if (!tun) { 1856 ret = -EINVAL; 1857 goto unlock; 1858 } 1859 ret = security_tun_dev_attach_queue(tun->security); 1860 if (ret < 0) 1861 goto unlock; 1862 ret = tun_attach(tun, file, false); 1863 } else if (ifr->ifr_flags & IFF_DETACH_QUEUE) { 1864 tun = rtnl_dereference(tfile->tun); 1865 if (!tun || !(tun->flags & IFF_MULTI_QUEUE) || tfile->detached) 1866 ret = -EINVAL; 1867 else 1868 __tun_detach(tfile, false); 1869 } else 1870 ret = -EINVAL; 1871 1872 unlock: 1873 rtnl_unlock(); 1874 return ret; 1875 } 1876 1877 static long __tun_chr_ioctl(struct file *file, unsigned int cmd, 1878 unsigned long arg, int ifreq_len) 1879 { 1880 struct tun_file *tfile = file->private_data; 1881 struct tun_struct *tun; 1882 void __user* argp = (void __user*)arg; 1883 struct ifreq ifr; 1884 kuid_t owner; 1885 kgid_t group; 1886 int sndbuf; 1887 int vnet_hdr_sz; 1888 unsigned int ifindex; 1889 int le; 1890 int ret; 1891 1892 if (cmd == TUNSETIFF || cmd == TUNSETQUEUE || _IOC_TYPE(cmd) == 0x89) { 1893 if (copy_from_user(&ifr, argp, ifreq_len)) 1894 return -EFAULT; 1895 } else { 1896 memset(&ifr, 0, sizeof(ifr)); 1897 } 1898 if (cmd == TUNGETFEATURES) { 1899 /* Currently this just means: "what IFF flags are valid?". 1900 * This is needed because we never checked for invalid flags on 1901 * TUNSETIFF. 1902 */ 1903 return put_user(IFF_TUN | IFF_TAP | TUN_FEATURES, 1904 (unsigned int __user*)argp); 1905 } else if (cmd == TUNSETQUEUE) 1906 return tun_set_queue(file, &ifr); 1907 1908 ret = 0; 1909 rtnl_lock(); 1910 1911 tun = __tun_get(tfile); 1912 if (cmd == TUNSETIFF && !tun) { 1913 ifr.ifr_name[IFNAMSIZ-1] = '\0'; 1914 1915 ret = tun_set_iff(sock_net(&tfile->sk), file, &ifr); 1916 1917 if (ret) 1918 goto unlock; 1919 1920 if (copy_to_user(argp, &ifr, ifreq_len)) 1921 ret = -EFAULT; 1922 goto unlock; 1923 } 1924 if (cmd == TUNSETIFINDEX) { 1925 ret = -EPERM; 1926 if (tun) 1927 goto unlock; 1928 1929 ret = -EFAULT; 1930 if (copy_from_user(&ifindex, argp, sizeof(ifindex))) 1931 goto unlock; 1932 1933 ret = 0; 1934 tfile->ifindex = ifindex; 1935 goto unlock; 1936 } 1937 1938 ret = -EBADFD; 1939 if (!tun) 1940 goto unlock; 1941 1942 tun_debug(KERN_INFO, tun, "tun_chr_ioctl cmd %u\n", cmd); 1943 1944 ret = 0; 1945 switch (cmd) { 1946 case TUNGETIFF: 1947 tun_get_iff(current->nsproxy->net_ns, tun, &ifr); 1948 1949 if (tfile->detached) 1950 ifr.ifr_flags |= IFF_DETACH_QUEUE; 1951 if (!tfile->socket.sk->sk_filter) 1952 ifr.ifr_flags |= IFF_NOFILTER; 1953 1954 if (copy_to_user(argp, &ifr, ifreq_len)) 1955 ret = -EFAULT; 1956 break; 1957 1958 case TUNSETNOCSUM: 1959 /* Disable/Enable checksum */ 1960 1961 /* [unimplemented] */ 1962 tun_debug(KERN_INFO, tun, "ignored: set checksum %s\n", 1963 arg ? "disabled" : "enabled"); 1964 break; 1965 1966 case TUNSETPERSIST: 1967 /* Disable/Enable persist mode. Keep an extra reference to the 1968 * module to prevent the module being unprobed. 1969 */ 1970 if (arg && !(tun->flags & IFF_PERSIST)) { 1971 tun->flags |= IFF_PERSIST; 1972 __module_get(THIS_MODULE); 1973 } 1974 if (!arg && (tun->flags & IFF_PERSIST)) { 1975 tun->flags &= ~IFF_PERSIST; 1976 module_put(THIS_MODULE); 1977 } 1978 1979 tun_debug(KERN_INFO, tun, "persist %s\n", 1980 arg ? "enabled" : "disabled"); 1981 break; 1982 1983 case TUNSETOWNER: 1984 /* Set owner of the device */ 1985 owner = make_kuid(current_user_ns(), arg); 1986 if (!uid_valid(owner)) { 1987 ret = -EINVAL; 1988 break; 1989 } 1990 tun->owner = owner; 1991 tun_debug(KERN_INFO, tun, "owner set to %u\n", 1992 from_kuid(&init_user_ns, tun->owner)); 1993 break; 1994 1995 case TUNSETGROUP: 1996 /* Set group of the device */ 1997 group = make_kgid(current_user_ns(), arg); 1998 if (!gid_valid(group)) { 1999 ret = -EINVAL; 2000 break; 2001 } 2002 tun->group = group; 2003 tun_debug(KERN_INFO, tun, "group set to %u\n", 2004 from_kgid(&init_user_ns, tun->group)); 2005 break; 2006 2007 case TUNSETLINK: 2008 /* Only allow setting the type when the interface is down */ 2009 if (tun->dev->flags & IFF_UP) { 2010 tun_debug(KERN_INFO, tun, 2011 "Linktype set failed because interface is up\n"); 2012 ret = -EBUSY; 2013 } else { 2014 tun->dev->type = (int) arg; 2015 tun_debug(KERN_INFO, tun, "linktype set to %d\n", 2016 tun->dev->type); 2017 ret = 0; 2018 } 2019 break; 2020 2021 #ifdef TUN_DEBUG 2022 case TUNSETDEBUG: 2023 tun->debug = arg; 2024 break; 2025 #endif 2026 case TUNSETOFFLOAD: 2027 ret = set_offload(tun, arg); 2028 break; 2029 2030 case TUNSETTXFILTER: 2031 /* Can be set only for TAPs */ 2032 ret = -EINVAL; 2033 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP) 2034 break; 2035 ret = update_filter(&tun->txflt, (void __user *)arg); 2036 break; 2037 2038 case SIOCGIFHWADDR: 2039 /* Get hw address */ 2040 memcpy(ifr.ifr_hwaddr.sa_data, tun->dev->dev_addr, ETH_ALEN); 2041 ifr.ifr_hwaddr.sa_family = tun->dev->type; 2042 if (copy_to_user(argp, &ifr, ifreq_len)) 2043 ret = -EFAULT; 2044 break; 2045 2046 case SIOCSIFHWADDR: 2047 /* Set hw address */ 2048 tun_debug(KERN_DEBUG, tun, "set hw address: %pM\n", 2049 ifr.ifr_hwaddr.sa_data); 2050 2051 ret = dev_set_mac_address(tun->dev, &ifr.ifr_hwaddr); 2052 break; 2053 2054 case TUNGETSNDBUF: 2055 sndbuf = tfile->socket.sk->sk_sndbuf; 2056 if (copy_to_user(argp, &sndbuf, sizeof(sndbuf))) 2057 ret = -EFAULT; 2058 break; 2059 2060 case TUNSETSNDBUF: 2061 if (copy_from_user(&sndbuf, argp, sizeof(sndbuf))) { 2062 ret = -EFAULT; 2063 break; 2064 } 2065 2066 tun->sndbuf = sndbuf; 2067 tun_set_sndbuf(tun); 2068 break; 2069 2070 case TUNGETVNETHDRSZ: 2071 vnet_hdr_sz = tun->vnet_hdr_sz; 2072 if (copy_to_user(argp, &vnet_hdr_sz, sizeof(vnet_hdr_sz))) 2073 ret = -EFAULT; 2074 break; 2075 2076 case TUNSETVNETHDRSZ: 2077 if (copy_from_user(&vnet_hdr_sz, argp, sizeof(vnet_hdr_sz))) { 2078 ret = -EFAULT; 2079 break; 2080 } 2081 if (vnet_hdr_sz < (int)sizeof(struct virtio_net_hdr)) { 2082 ret = -EINVAL; 2083 break; 2084 } 2085 2086 tun->vnet_hdr_sz = vnet_hdr_sz; 2087 break; 2088 2089 case TUNGETVNETLE: 2090 le = !!(tun->flags & TUN_VNET_LE); 2091 if (put_user(le, (int __user *)argp)) 2092 ret = -EFAULT; 2093 break; 2094 2095 case TUNSETVNETLE: 2096 if (get_user(le, (int __user *)argp)) { 2097 ret = -EFAULT; 2098 break; 2099 } 2100 if (le) 2101 tun->flags |= TUN_VNET_LE; 2102 else 2103 tun->flags &= ~TUN_VNET_LE; 2104 break; 2105 2106 case TUNGETVNETBE: 2107 ret = tun_get_vnet_be(tun, argp); 2108 break; 2109 2110 case TUNSETVNETBE: 2111 ret = tun_set_vnet_be(tun, argp); 2112 break; 2113 2114 case TUNATTACHFILTER: 2115 /* Can be set only for TAPs */ 2116 ret = -EINVAL; 2117 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP) 2118 break; 2119 ret = -EFAULT; 2120 if (copy_from_user(&tun->fprog, argp, sizeof(tun->fprog))) 2121 break; 2122 2123 ret = tun_attach_filter(tun); 2124 break; 2125 2126 case TUNDETACHFILTER: 2127 /* Can be set only for TAPs */ 2128 ret = -EINVAL; 2129 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP) 2130 break; 2131 ret = 0; 2132 tun_detach_filter(tun, tun->numqueues); 2133 break; 2134 2135 case TUNGETFILTER: 2136 ret = -EINVAL; 2137 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP) 2138 break; 2139 ret = -EFAULT; 2140 if (copy_to_user(argp, &tun->fprog, sizeof(tun->fprog))) 2141 break; 2142 ret = 0; 2143 break; 2144 2145 default: 2146 ret = -EINVAL; 2147 break; 2148 } 2149 2150 unlock: 2151 rtnl_unlock(); 2152 if (tun) 2153 tun_put(tun); 2154 return ret; 2155 } 2156 2157 static long tun_chr_ioctl(struct file *file, 2158 unsigned int cmd, unsigned long arg) 2159 { 2160 return __tun_chr_ioctl(file, cmd, arg, sizeof (struct ifreq)); 2161 } 2162 2163 #ifdef CONFIG_COMPAT 2164 static long tun_chr_compat_ioctl(struct file *file, 2165 unsigned int cmd, unsigned long arg) 2166 { 2167 switch (cmd) { 2168 case TUNSETIFF: 2169 case TUNGETIFF: 2170 case TUNSETTXFILTER: 2171 case TUNGETSNDBUF: 2172 case TUNSETSNDBUF: 2173 case SIOCGIFHWADDR: 2174 case SIOCSIFHWADDR: 2175 arg = (unsigned long)compat_ptr(arg); 2176 break; 2177 default: 2178 arg = (compat_ulong_t)arg; 2179 break; 2180 } 2181 2182 /* 2183 * compat_ifreq is shorter than ifreq, so we must not access beyond 2184 * the end of that structure. All fields that are used in this 2185 * driver are compatible though, we don't need to convert the 2186 * contents. 2187 */ 2188 return __tun_chr_ioctl(file, cmd, arg, sizeof(struct compat_ifreq)); 2189 } 2190 #endif /* CONFIG_COMPAT */ 2191 2192 static int tun_chr_fasync(int fd, struct file *file, int on) 2193 { 2194 struct tun_file *tfile = file->private_data; 2195 int ret; 2196 2197 if ((ret = fasync_helper(fd, file, on, &tfile->fasync)) < 0) 2198 goto out; 2199 2200 if (on) { 2201 __f_setown(file, task_pid(current), PIDTYPE_PID, 0); 2202 tfile->flags |= TUN_FASYNC; 2203 } else 2204 tfile->flags &= ~TUN_FASYNC; 2205 ret = 0; 2206 out: 2207 return ret; 2208 } 2209 2210 static int tun_chr_open(struct inode *inode, struct file * file) 2211 { 2212 struct net *net = current->nsproxy->net_ns; 2213 struct tun_file *tfile; 2214 2215 DBG1(KERN_INFO, "tunX: tun_chr_open\n"); 2216 2217 tfile = (struct tun_file *)sk_alloc(net, AF_UNSPEC, GFP_KERNEL, 2218 &tun_proto, 0); 2219 if (!tfile) 2220 return -ENOMEM; 2221 RCU_INIT_POINTER(tfile->tun, NULL); 2222 tfile->flags = 0; 2223 tfile->ifindex = 0; 2224 2225 init_waitqueue_head(&tfile->wq.wait); 2226 RCU_INIT_POINTER(tfile->socket.wq, &tfile->wq); 2227 2228 tfile->socket.file = file; 2229 tfile->socket.ops = &tun_socket_ops; 2230 2231 sock_init_data(&tfile->socket, &tfile->sk); 2232 2233 tfile->sk.sk_write_space = tun_sock_write_space; 2234 tfile->sk.sk_sndbuf = INT_MAX; 2235 2236 file->private_data = tfile; 2237 INIT_LIST_HEAD(&tfile->next); 2238 2239 sock_set_flag(&tfile->sk, SOCK_ZEROCOPY); 2240 2241 return 0; 2242 } 2243 2244 static int tun_chr_close(struct inode *inode, struct file *file) 2245 { 2246 struct tun_file *tfile = file->private_data; 2247 2248 tun_detach(tfile, true); 2249 2250 return 0; 2251 } 2252 2253 #ifdef CONFIG_PROC_FS 2254 static void tun_chr_show_fdinfo(struct seq_file *m, struct file *f) 2255 { 2256 struct tun_struct *tun; 2257 struct ifreq ifr; 2258 2259 memset(&ifr, 0, sizeof(ifr)); 2260 2261 rtnl_lock(); 2262 tun = tun_get(f); 2263 if (tun) 2264 tun_get_iff(current->nsproxy->net_ns, tun, &ifr); 2265 rtnl_unlock(); 2266 2267 if (tun) 2268 tun_put(tun); 2269 2270 seq_printf(m, "iff:\t%s\n", ifr.ifr_name); 2271 } 2272 #endif 2273 2274 static const struct file_operations tun_fops = { 2275 .owner = THIS_MODULE, 2276 .llseek = no_llseek, 2277 .read_iter = tun_chr_read_iter, 2278 .write_iter = tun_chr_write_iter, 2279 .poll = tun_chr_poll, 2280 .unlocked_ioctl = tun_chr_ioctl, 2281 #ifdef CONFIG_COMPAT 2282 .compat_ioctl = tun_chr_compat_ioctl, 2283 #endif 2284 .open = tun_chr_open, 2285 .release = tun_chr_close, 2286 .fasync = tun_chr_fasync, 2287 #ifdef CONFIG_PROC_FS 2288 .show_fdinfo = tun_chr_show_fdinfo, 2289 #endif 2290 }; 2291 2292 static struct miscdevice tun_miscdev = { 2293 .minor = TUN_MINOR, 2294 .name = "tun", 2295 .nodename = "net/tun", 2296 .fops = &tun_fops, 2297 }; 2298 2299 /* ethtool interface */ 2300 2301 static int tun_get_settings(struct net_device *dev, struct ethtool_cmd *cmd) 2302 { 2303 cmd->supported = 0; 2304 cmd->advertising = 0; 2305 ethtool_cmd_speed_set(cmd, SPEED_10); 2306 cmd->duplex = DUPLEX_FULL; 2307 cmd->port = PORT_TP; 2308 cmd->phy_address = 0; 2309 cmd->transceiver = XCVR_INTERNAL; 2310 cmd->autoneg = AUTONEG_DISABLE; 2311 cmd->maxtxpkt = 0; 2312 cmd->maxrxpkt = 0; 2313 return 0; 2314 } 2315 2316 static void tun_get_drvinfo(struct net_device *dev, struct ethtool_drvinfo *info) 2317 { 2318 struct tun_struct *tun = netdev_priv(dev); 2319 2320 strlcpy(info->driver, DRV_NAME, sizeof(info->driver)); 2321 strlcpy(info->version, DRV_VERSION, sizeof(info->version)); 2322 2323 switch (tun->flags & TUN_TYPE_MASK) { 2324 case IFF_TUN: 2325 strlcpy(info->bus_info, "tun", sizeof(info->bus_info)); 2326 break; 2327 case IFF_TAP: 2328 strlcpy(info->bus_info, "tap", sizeof(info->bus_info)); 2329 break; 2330 } 2331 } 2332 2333 static u32 tun_get_msglevel(struct net_device *dev) 2334 { 2335 #ifdef TUN_DEBUG 2336 struct tun_struct *tun = netdev_priv(dev); 2337 return tun->debug; 2338 #else 2339 return -EOPNOTSUPP; 2340 #endif 2341 } 2342 2343 static void tun_set_msglevel(struct net_device *dev, u32 value) 2344 { 2345 #ifdef TUN_DEBUG 2346 struct tun_struct *tun = netdev_priv(dev); 2347 tun->debug = value; 2348 #endif 2349 } 2350 2351 static const struct ethtool_ops tun_ethtool_ops = { 2352 .get_settings = tun_get_settings, 2353 .get_drvinfo = tun_get_drvinfo, 2354 .get_msglevel = tun_get_msglevel, 2355 .set_msglevel = tun_set_msglevel, 2356 .get_link = ethtool_op_get_link, 2357 .get_ts_info = ethtool_op_get_ts_info, 2358 }; 2359 2360 2361 static int __init tun_init(void) 2362 { 2363 int ret = 0; 2364 2365 pr_info("%s, %s\n", DRV_DESCRIPTION, DRV_VERSION); 2366 pr_info("%s\n", DRV_COPYRIGHT); 2367 2368 ret = rtnl_link_register(&tun_link_ops); 2369 if (ret) { 2370 pr_err("Can't register link_ops\n"); 2371 goto err_linkops; 2372 } 2373 2374 ret = misc_register(&tun_miscdev); 2375 if (ret) { 2376 pr_err("Can't register misc device %d\n", TUN_MINOR); 2377 goto err_misc; 2378 } 2379 return 0; 2380 err_misc: 2381 rtnl_link_unregister(&tun_link_ops); 2382 err_linkops: 2383 return ret; 2384 } 2385 2386 static void tun_cleanup(void) 2387 { 2388 misc_deregister(&tun_miscdev); 2389 rtnl_link_unregister(&tun_link_ops); 2390 } 2391 2392 /* Get an underlying socket object from tun file. Returns error unless file is 2393 * attached to a device. The returned object works like a packet socket, it 2394 * can be used for sock_sendmsg/sock_recvmsg. The caller is responsible for 2395 * holding a reference to the file for as long as the socket is in use. */ 2396 struct socket *tun_get_socket(struct file *file) 2397 { 2398 struct tun_file *tfile; 2399 if (file->f_op != &tun_fops) 2400 return ERR_PTR(-EINVAL); 2401 tfile = file->private_data; 2402 if (!tfile) 2403 return ERR_PTR(-EBADFD); 2404 return &tfile->socket; 2405 } 2406 EXPORT_SYMBOL_GPL(tun_get_socket); 2407 2408 module_init(tun_init); 2409 module_exit(tun_cleanup); 2410 MODULE_DESCRIPTION(DRV_DESCRIPTION); 2411 MODULE_AUTHOR(DRV_COPYRIGHT); 2412 MODULE_LICENSE("GPL"); 2413 MODULE_ALIAS_MISCDEV(TUN_MINOR); 2414 MODULE_ALIAS("devname:net/tun"); 2415