tun: Move read_wait into tun_file
[safe/jmp/linux-2.6] / drivers / net / tun.c
1 /*
2  *  TUN - Universal TUN/TAP device driver.
3  *  Copyright (C) 1999-2002 Maxim Krasnyansky <maxk@qualcomm.com>
4  *
5  *  This program is free software; you can redistribute it and/or modify
6  *  it under the terms of the GNU General Public License as published by
7  *  the Free Software Foundation; either version 2 of the License, or
8  *  (at your option) any later version.
9  *
10  *  This program is distributed in the hope that it will be useful,
11  *  but WITHOUT ANY WARRANTY; without even the implied warranty of
12  *  MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
13  *  GNU General Public License for more details.
14  *
15  *  $Id: tun.c,v 1.15 2002/03/01 02:44:24 maxk Exp $
16  */
17
18 /*
19  *  Changes:
20  *
21  *  Mike Kershaw <dragorn@kismetwireless.net> 2005/08/14
22  *    Add TUNSETLINK ioctl to set the link encapsulation
23  *
24  *  Mark Smith <markzzzsmith@yahoo.com.au>
25  *    Use random_ether_addr() for tap MAC address.
26  *
27  *  Harald Roelle <harald.roelle@ifi.lmu.de>  2004/04/20
28  *    Fixes in packet dropping, queue length setting and queue wakeup.
29  *    Increased default tx queue length.
30  *    Added ethtool API.
31  *    Minor cleanups
32  *
33  *  Daniel Podlejski <underley@underley.eu.org>
34  *    Modifications for 2.3.99-pre5 kernel.
35  */
36
37 #define DRV_NAME        "tun"
38 #define DRV_VERSION     "1.6"
39 #define DRV_DESCRIPTION "Universal TUN/TAP device driver"
40 #define DRV_COPYRIGHT   "(C) 1999-2004 Max Krasnyansky <maxk@qualcomm.com>"
41
42 #include <linux/module.h>
43 #include <linux/errno.h>
44 #include <linux/kernel.h>
45 #include <linux/major.h>
46 #include <linux/slab.h>
47 #include <linux/smp_lock.h>
48 #include <linux/poll.h>
49 #include <linux/fcntl.h>
50 #include <linux/init.h>
51 #include <linux/skbuff.h>
52 #include <linux/netdevice.h>
53 #include <linux/etherdevice.h>
54 #include <linux/miscdevice.h>
55 #include <linux/ethtool.h>
56 #include <linux/rtnetlink.h>
57 #include <linux/if.h>
58 #include <linux/if_arp.h>
59 #include <linux/if_ether.h>
60 #include <linux/if_tun.h>
61 #include <linux/crc32.h>
62 #include <linux/nsproxy.h>
63 #include <linux/virtio_net.h>
64 #include <net/net_namespace.h>
65 #include <net/netns/generic.h>
66
67 #include <asm/system.h>
68 #include <asm/uaccess.h>
69
70 /* Uncomment to enable debugging */
71 /* #define TUN_DEBUG 1 */
72
73 #ifdef TUN_DEBUG
74 static int debug;
75
76 #define DBG  if(tun->debug)printk
77 #define DBG1 if(debug==2)printk
78 #else
79 #define DBG( a... )
80 #define DBG1( a... )
81 #endif
82
83 #define FLT_EXACT_COUNT 8
84 struct tap_filter {
85         unsigned int    count;    /* Number of addrs. Zero means disabled */
86         u32             mask[2];  /* Mask of the hashed addrs */
87         unsigned char   addr[FLT_EXACT_COUNT][ETH_ALEN];
88 };
89
90 struct tun_file {
91         struct tun_struct *tun;
92         struct net *net;
93         wait_queue_head_t       read_wait;
94 };
95
96 struct tun_struct {
97         struct tun_file         *tfile;
98         unsigned int            flags;
99         uid_t                   owner;
100         gid_t                   group;
101
102         struct sk_buff_head     readq;
103
104         struct net_device       *dev;
105         struct fasync_struct    *fasync;
106
107         struct tap_filter       txflt;
108
109 #ifdef TUN_DEBUG
110         int debug;
111 #endif
112 };
113
114 static int tun_attach(struct tun_struct *tun, struct file *file)
115 {
116         struct tun_file *tfile = file->private_data;
117         const struct cred *cred = current_cred();
118         int err;
119
120         ASSERT_RTNL();
121
122         /* Check permissions */
123         if (((tun->owner != -1 && cred->euid != tun->owner) ||
124              (tun->group != -1 && cred->egid != tun->group)) &&
125                 !capable(CAP_NET_ADMIN))
126                 return -EPERM;
127
128         netif_tx_lock_bh(tun->dev);
129
130         err = -EINVAL;
131         if (tfile->tun)
132                 goto out;
133
134         err = -EBUSY;
135         if (tun->tfile)
136                 goto out;
137
138         err = 0;
139         tfile->tun = tun;
140         tun->tfile = tfile;
141
142 out:
143         netif_tx_unlock_bh(tun->dev);
144         return err;
145 }
146
147 static void __tun_detach(struct tun_struct *tun)
148 {
149         struct tun_file *tfile = tun->tfile;
150
151         /* Detach from net device */
152         netif_tx_lock_bh(tun->dev);
153         tfile->tun = NULL;
154         tun->tfile = NULL;
155         netif_tx_unlock_bh(tun->dev);
156
157         /* Drop read queue */
158         skb_queue_purge(&tun->readq);
159 }
160
161 static struct tun_struct *__tun_get(struct tun_file *tfile)
162 {
163         return tfile->tun;
164 }
165
166 static struct tun_struct *tun_get(struct file *file)
167 {
168         return __tun_get(file->private_data);
169 }
170
171 static void tun_put(struct tun_struct *tun)
172 {
173         /* Noop for now */
174 }
175
176 /* TAP filterting */
177 static void addr_hash_set(u32 *mask, const u8 *addr)
178 {
179         int n = ether_crc(ETH_ALEN, addr) >> 26;
180         mask[n >> 5] |= (1 << (n & 31));
181 }
182
183 static unsigned int addr_hash_test(const u32 *mask, const u8 *addr)
184 {
185         int n = ether_crc(ETH_ALEN, addr) >> 26;
186         return mask[n >> 5] & (1 << (n & 31));
187 }
188
189 static int update_filter(struct tap_filter *filter, void __user *arg)
190 {
191         struct { u8 u[ETH_ALEN]; } *addr;
192         struct tun_filter uf;
193         int err, alen, n, nexact;
194
195         if (copy_from_user(&uf, arg, sizeof(uf)))
196                 return -EFAULT;
197
198         if (!uf.count) {
199                 /* Disabled */
200                 filter->count = 0;
201                 return 0;
202         }
203
204         alen = ETH_ALEN * uf.count;
205         addr = kmalloc(alen, GFP_KERNEL);
206         if (!addr)
207                 return -ENOMEM;
208
209         if (copy_from_user(addr, arg + sizeof(uf), alen)) {
210                 err = -EFAULT;
211                 goto done;
212         }
213
214         /* The filter is updated without holding any locks. Which is
215          * perfectly safe. We disable it first and in the worst
216          * case we'll accept a few undesired packets. */
217         filter->count = 0;
218         wmb();
219
220         /* Use first set of addresses as an exact filter */
221         for (n = 0; n < uf.count && n < FLT_EXACT_COUNT; n++)
222                 memcpy(filter->addr[n], addr[n].u, ETH_ALEN);
223
224         nexact = n;
225
226         /* The rest is hashed */
227         memset(filter->mask, 0, sizeof(filter->mask));
228         for (; n < uf.count; n++)
229                 addr_hash_set(filter->mask, addr[n].u);
230
231         /* For ALLMULTI just set the mask to all ones.
232          * This overrides the mask populated above. */
233         if ((uf.flags & TUN_FLT_ALLMULTI))
234                 memset(filter->mask, ~0, sizeof(filter->mask));
235
236         /* Now enable the filter */
237         wmb();
238         filter->count = nexact;
239
240         /* Return the number of exact filters */
241         err = nexact;
242
243 done:
244         kfree(addr);
245         return err;
246 }
247
248 /* Returns: 0 - drop, !=0 - accept */
249 static int run_filter(struct tap_filter *filter, const struct sk_buff *skb)
250 {
251         /* Cannot use eth_hdr(skb) here because skb_mac_hdr() is incorrect
252          * at this point. */
253         struct ethhdr *eh = (struct ethhdr *) skb->data;
254         int i;
255
256         /* Exact match */
257         for (i = 0; i < filter->count; i++)
258                 if (!compare_ether_addr(eh->h_dest, filter->addr[i]))
259                         return 1;
260
261         /* Inexact match (multicast only) */
262         if (is_multicast_ether_addr(eh->h_dest))
263                 return addr_hash_test(filter->mask, eh->h_dest);
264
265         return 0;
266 }
267
268 /*
269  * Checks whether the packet is accepted or not.
270  * Returns: 0 - drop, !=0 - accept
271  */
272 static int check_filter(struct tap_filter *filter, const struct sk_buff *skb)
273 {
274         if (!filter->count)
275                 return 1;
276
277         return run_filter(filter, skb);
278 }
279
280 /* Network device part of the driver */
281
282 static const struct ethtool_ops tun_ethtool_ops;
283
284 /* Net device open. */
285 static int tun_net_open(struct net_device *dev)
286 {
287         netif_start_queue(dev);
288         return 0;
289 }
290
291 /* Net device close. */
292 static int tun_net_close(struct net_device *dev)
293 {
294         netif_stop_queue(dev);
295         return 0;
296 }
297
298 /* Net device start xmit */
299 static int tun_net_xmit(struct sk_buff *skb, struct net_device *dev)
300 {
301         struct tun_struct *tun = netdev_priv(dev);
302
303         DBG(KERN_INFO "%s: tun_net_xmit %d\n", tun->dev->name, skb->len);
304
305         /* Drop packet if interface is not attached */
306         if (!tun->tfile)
307                 goto drop;
308
309         /* Drop if the filter does not like it.
310          * This is a noop if the filter is disabled.
311          * Filter can be enabled only for the TAP devices. */
312         if (!check_filter(&tun->txflt, skb))
313                 goto drop;
314
315         if (skb_queue_len(&tun->readq) >= dev->tx_queue_len) {
316                 if (!(tun->flags & TUN_ONE_QUEUE)) {
317                         /* Normal queueing mode. */
318                         /* Packet scheduler handles dropping of further packets. */
319                         netif_stop_queue(dev);
320
321                         /* We won't see all dropped packets individually, so overrun
322                          * error is more appropriate. */
323                         dev->stats.tx_fifo_errors++;
324                 } else {
325                         /* Single queue mode.
326                          * Driver handles dropping of all packets itself. */
327                         goto drop;
328                 }
329         }
330
331         /* Enqueue packet */
332         skb_queue_tail(&tun->readq, skb);
333         dev->trans_start = jiffies;
334
335         /* Notify and wake up reader process */
336         if (tun->flags & TUN_FASYNC)
337                 kill_fasync(&tun->fasync, SIGIO, POLL_IN);
338         wake_up_interruptible(&tun->tfile->read_wait);
339         return 0;
340
341 drop:
342         dev->stats.tx_dropped++;
343         kfree_skb(skb);
344         return 0;
345 }
346
347 static void tun_net_mclist(struct net_device *dev)
348 {
349         /*
350          * This callback is supposed to deal with mc filter in
351          * _rx_ path and has nothing to do with the _tx_ path.
352          * In rx path we always accept everything userspace gives us.
353          */
354         return;
355 }
356
357 #define MIN_MTU 68
358 #define MAX_MTU 65535
359
360 static int
361 tun_net_change_mtu(struct net_device *dev, int new_mtu)
362 {
363         if (new_mtu < MIN_MTU || new_mtu + dev->hard_header_len > MAX_MTU)
364                 return -EINVAL;
365         dev->mtu = new_mtu;
366         return 0;
367 }
368
369 static const struct net_device_ops tun_netdev_ops = {
370         .ndo_open               = tun_net_open,
371         .ndo_stop               = tun_net_close,
372         .ndo_start_xmit         = tun_net_xmit,
373         .ndo_change_mtu         = tun_net_change_mtu,
374 };
375
376 static const struct net_device_ops tap_netdev_ops = {
377         .ndo_open               = tun_net_open,
378         .ndo_stop               = tun_net_close,
379         .ndo_start_xmit         = tun_net_xmit,
380         .ndo_change_mtu         = tun_net_change_mtu,
381         .ndo_set_multicast_list = tun_net_mclist,
382         .ndo_set_mac_address    = eth_mac_addr,
383         .ndo_validate_addr      = eth_validate_addr,
384 };
385
386 /* Initialize net device. */
387 static void tun_net_init(struct net_device *dev)
388 {
389         struct tun_struct *tun = netdev_priv(dev);
390
391         switch (tun->flags & TUN_TYPE_MASK) {
392         case TUN_TUN_DEV:
393                 dev->netdev_ops = &tun_netdev_ops;
394
395                 /* Point-to-Point TUN Device */
396                 dev->hard_header_len = 0;
397                 dev->addr_len = 0;
398                 dev->mtu = 1500;
399
400                 /* Zero header length */
401                 dev->type = ARPHRD_NONE;
402                 dev->flags = IFF_POINTOPOINT | IFF_NOARP | IFF_MULTICAST;
403                 dev->tx_queue_len = TUN_READQ_SIZE;  /* We prefer our own queue length */
404                 break;
405
406         case TUN_TAP_DEV:
407                 dev->netdev_ops = &tap_netdev_ops;
408                 /* Ethernet TAP Device */
409                 ether_setup(dev);
410
411                 random_ether_addr(dev->dev_addr);
412
413                 dev->tx_queue_len = TUN_READQ_SIZE;  /* We prefer our own queue length */
414                 break;
415         }
416 }
417
418 /* Character device part */
419
420 /* Poll */
421 static unsigned int tun_chr_poll(struct file *file, poll_table * wait)
422 {
423         struct tun_file *tfile = file->private_data;
424         struct tun_struct *tun = __tun_get(tfile);
425         unsigned int mask = POLLOUT | POLLWRNORM;
426
427         if (!tun)
428                 return POLLERR;
429
430         DBG(KERN_INFO "%s: tun_chr_poll\n", tun->dev->name);
431
432         poll_wait(file, &tfile->read_wait, wait);
433
434         if (!skb_queue_empty(&tun->readq))
435                 mask |= POLLIN | POLLRDNORM;
436
437         tun_put(tun);
438         return mask;
439 }
440
441 /* prepad is the amount to reserve at front.  len is length after that.
442  * linear is a hint as to how much to copy (usually headers). */
443 static struct sk_buff *tun_alloc_skb(size_t prepad, size_t len, size_t linear,
444                                      gfp_t gfp)
445 {
446         struct sk_buff *skb;
447         unsigned int i;
448
449         skb = alloc_skb(prepad + len, gfp|__GFP_NOWARN);
450         if (skb) {
451                 skb_reserve(skb, prepad);
452                 skb_put(skb, len);
453                 return skb;
454         }
455
456         /* Under a page?  Don't bother with paged skb. */
457         if (prepad + len < PAGE_SIZE)
458                 return NULL;
459
460         /* Start with a normal skb, and add pages. */
461         skb = alloc_skb(prepad + linear, gfp);
462         if (!skb)
463                 return NULL;
464
465         skb_reserve(skb, prepad);
466         skb_put(skb, linear);
467
468         len -= linear;
469
470         for (i = 0; i < MAX_SKB_FRAGS; i++) {
471                 skb_frag_t *f = &skb_shinfo(skb)->frags[i];
472
473                 f->page = alloc_page(gfp|__GFP_ZERO);
474                 if (!f->page)
475                         break;
476
477                 f->page_offset = 0;
478                 f->size = PAGE_SIZE;
479
480                 skb->data_len += PAGE_SIZE;
481                 skb->len += PAGE_SIZE;
482                 skb->truesize += PAGE_SIZE;
483                 skb_shinfo(skb)->nr_frags++;
484
485                 if (len < PAGE_SIZE) {
486                         len = 0;
487                         break;
488                 }
489                 len -= PAGE_SIZE;
490         }
491
492         /* Too large, or alloc fail? */
493         if (unlikely(len)) {
494                 kfree_skb(skb);
495                 skb = NULL;
496         }
497
498         return skb;
499 }
500
501 /* Get packet from user space buffer */
502 static __inline__ ssize_t tun_get_user(struct tun_struct *tun, struct iovec *iv, size_t count)
503 {
504         struct tun_pi pi = { 0, __constant_htons(ETH_P_IP) };
505         struct sk_buff *skb;
506         size_t len = count, align = 0;
507         struct virtio_net_hdr gso = { 0 };
508
509         if (!(tun->flags & TUN_NO_PI)) {
510                 if ((len -= sizeof(pi)) > count)
511                         return -EINVAL;
512
513                 if(memcpy_fromiovec((void *)&pi, iv, sizeof(pi)))
514                         return -EFAULT;
515         }
516
517         if (tun->flags & TUN_VNET_HDR) {
518                 if ((len -= sizeof(gso)) > count)
519                         return -EINVAL;
520
521                 if (memcpy_fromiovec((void *)&gso, iv, sizeof(gso)))
522                         return -EFAULT;
523
524                 if (gso.hdr_len > len)
525                         return -EINVAL;
526         }
527
528         if ((tun->flags & TUN_TYPE_MASK) == TUN_TAP_DEV) {
529                 align = NET_IP_ALIGN;
530                 if (unlikely(len < ETH_HLEN))
531                         return -EINVAL;
532         }
533
534         if (!(skb = tun_alloc_skb(align, len, gso.hdr_len, GFP_KERNEL))) {
535                 tun->dev->stats.rx_dropped++;
536                 return -ENOMEM;
537         }
538
539         if (skb_copy_datagram_from_iovec(skb, 0, iv, len)) {
540                 tun->dev->stats.rx_dropped++;
541                 kfree_skb(skb);
542                 return -EFAULT;
543         }
544
545         if (gso.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) {
546                 if (!skb_partial_csum_set(skb, gso.csum_start,
547                                           gso.csum_offset)) {
548                         tun->dev->stats.rx_frame_errors++;
549                         kfree_skb(skb);
550                         return -EINVAL;
551                 }
552         } else if (tun->flags & TUN_NOCHECKSUM)
553                 skb->ip_summed = CHECKSUM_UNNECESSARY;
554
555         switch (tun->flags & TUN_TYPE_MASK) {
556         case TUN_TUN_DEV:
557                 if (tun->flags & TUN_NO_PI) {
558                         switch (skb->data[0] & 0xf0) {
559                         case 0x40:
560                                 pi.proto = htons(ETH_P_IP);
561                                 break;
562                         case 0x60:
563                                 pi.proto = htons(ETH_P_IPV6);
564                                 break;
565                         default:
566                                 tun->dev->stats.rx_dropped++;
567                                 kfree_skb(skb);
568                                 return -EINVAL;
569                         }
570                 }
571
572                 skb_reset_mac_header(skb);
573                 skb->protocol = pi.proto;
574                 skb->dev = tun->dev;
575                 break;
576         case TUN_TAP_DEV:
577                 skb->protocol = eth_type_trans(skb, tun->dev);
578                 break;
579         };
580
581         if (gso.gso_type != VIRTIO_NET_HDR_GSO_NONE) {
582                 pr_debug("GSO!\n");
583                 switch (gso.gso_type & ~VIRTIO_NET_HDR_GSO_ECN) {
584                 case VIRTIO_NET_HDR_GSO_TCPV4:
585                         skb_shinfo(skb)->gso_type = SKB_GSO_TCPV4;
586                         break;
587                 case VIRTIO_NET_HDR_GSO_TCPV6:
588                         skb_shinfo(skb)->gso_type = SKB_GSO_TCPV6;
589                         break;
590                 default:
591                         tun->dev->stats.rx_frame_errors++;
592                         kfree_skb(skb);
593                         return -EINVAL;
594                 }
595
596                 if (gso.gso_type & VIRTIO_NET_HDR_GSO_ECN)
597                         skb_shinfo(skb)->gso_type |= SKB_GSO_TCP_ECN;
598
599                 skb_shinfo(skb)->gso_size = gso.gso_size;
600                 if (skb_shinfo(skb)->gso_size == 0) {
601                         tun->dev->stats.rx_frame_errors++;
602                         kfree_skb(skb);
603                         return -EINVAL;
604                 }
605
606                 /* Header must be checked, and gso_segs computed. */
607                 skb_shinfo(skb)->gso_type |= SKB_GSO_DODGY;
608                 skb_shinfo(skb)->gso_segs = 0;
609         }
610
611         netif_rx_ni(skb);
612
613         tun->dev->stats.rx_packets++;
614         tun->dev->stats.rx_bytes += len;
615
616         return count;
617 }
618
619 static ssize_t tun_chr_aio_write(struct kiocb *iocb, const struct iovec *iv,
620                               unsigned long count, loff_t pos)
621 {
622         struct tun_struct *tun = tun_get(iocb->ki_filp);
623         ssize_t result;
624
625         if (!tun)
626                 return -EBADFD;
627
628         DBG(KERN_INFO "%s: tun_chr_write %ld\n", tun->dev->name, count);
629
630         result = tun_get_user(tun, (struct iovec *) iv, iov_length(iv, count));
631
632         tun_put(tun);
633         return result;
634 }
635
636 /* Put packet to the user space buffer */
637 static __inline__ ssize_t tun_put_user(struct tun_struct *tun,
638                                        struct sk_buff *skb,
639                                        struct iovec *iv, int len)
640 {
641         struct tun_pi pi = { 0, skb->protocol };
642         ssize_t total = 0;
643
644         if (!(tun->flags & TUN_NO_PI)) {
645                 if ((len -= sizeof(pi)) < 0)
646                         return -EINVAL;
647
648                 if (len < skb->len) {
649                         /* Packet will be striped */
650                         pi.flags |= TUN_PKT_STRIP;
651                 }
652
653                 if (memcpy_toiovec(iv, (void *) &pi, sizeof(pi)))
654                         return -EFAULT;
655                 total += sizeof(pi);
656         }
657
658         if (tun->flags & TUN_VNET_HDR) {
659                 struct virtio_net_hdr gso = { 0 }; /* no info leak */
660                 if ((len -= sizeof(gso)) < 0)
661                         return -EINVAL;
662
663                 if (skb_is_gso(skb)) {
664                         struct skb_shared_info *sinfo = skb_shinfo(skb);
665
666                         /* This is a hint as to how much should be linear. */
667                         gso.hdr_len = skb_headlen(skb);
668                         gso.gso_size = sinfo->gso_size;
669                         if (sinfo->gso_type & SKB_GSO_TCPV4)
670                                 gso.gso_type = VIRTIO_NET_HDR_GSO_TCPV4;
671                         else if (sinfo->gso_type & SKB_GSO_TCPV6)
672                                 gso.gso_type = VIRTIO_NET_HDR_GSO_TCPV6;
673                         else
674                                 BUG();
675                         if (sinfo->gso_type & SKB_GSO_TCP_ECN)
676                                 gso.gso_type |= VIRTIO_NET_HDR_GSO_ECN;
677                 } else
678                         gso.gso_type = VIRTIO_NET_HDR_GSO_NONE;
679
680                 if (skb->ip_summed == CHECKSUM_PARTIAL) {
681                         gso.flags = VIRTIO_NET_HDR_F_NEEDS_CSUM;
682                         gso.csum_start = skb->csum_start - skb_headroom(skb);
683                         gso.csum_offset = skb->csum_offset;
684                 } /* else everything is zero */
685
686                 if (unlikely(memcpy_toiovec(iv, (void *)&gso, sizeof(gso))))
687                         return -EFAULT;
688                 total += sizeof(gso);
689         }
690
691         len = min_t(int, skb->len, len);
692
693         skb_copy_datagram_iovec(skb, 0, iv, len);
694         total += len;
695
696         tun->dev->stats.tx_packets++;
697         tun->dev->stats.tx_bytes += len;
698
699         return total;
700 }
701
702 static ssize_t tun_chr_aio_read(struct kiocb *iocb, const struct iovec *iv,
703                             unsigned long count, loff_t pos)
704 {
705         struct file *file = iocb->ki_filp;
706         struct tun_file *tfile = file->private_data;
707         struct tun_struct *tun = __tun_get(tfile);
708         DECLARE_WAITQUEUE(wait, current);
709         struct sk_buff *skb;
710         ssize_t len, ret = 0;
711
712         if (!tun)
713                 return -EBADFD;
714
715         DBG(KERN_INFO "%s: tun_chr_read\n", tun->dev->name);
716
717         len = iov_length(iv, count);
718         if (len < 0) {
719                 ret = -EINVAL;
720                 goto out;
721         }
722
723         add_wait_queue(&tfile->read_wait, &wait);
724         while (len) {
725                 current->state = TASK_INTERRUPTIBLE;
726
727                 /* Read frames from the queue */
728                 if (!(skb=skb_dequeue(&tun->readq))) {
729                         if (file->f_flags & O_NONBLOCK) {
730                                 ret = -EAGAIN;
731                                 break;
732                         }
733                         if (signal_pending(current)) {
734                                 ret = -ERESTARTSYS;
735                                 break;
736                         }
737
738                         /* Nothing to read, let's sleep */
739                         schedule();
740                         continue;
741                 }
742                 netif_wake_queue(tun->dev);
743
744                 ret = tun_put_user(tun, skb, (struct iovec *) iv, len);
745                 kfree_skb(skb);
746                 break;
747         }
748
749         current->state = TASK_RUNNING;
750         remove_wait_queue(&tfile->read_wait, &wait);
751
752 out:
753         tun_put(tun);
754         return ret;
755 }
756
757 static void tun_setup(struct net_device *dev)
758 {
759         struct tun_struct *tun = netdev_priv(dev);
760
761         skb_queue_head_init(&tun->readq);
762
763         tun->owner = -1;
764         tun->group = -1;
765
766         dev->ethtool_ops = &tun_ethtool_ops;
767         dev->destructor = free_netdev;
768         dev->features |= NETIF_F_NETNS_LOCAL;
769 }
770
771 static int tun_set_iff(struct net *net, struct file *file, struct ifreq *ifr)
772 {
773         struct tun_struct *tun;
774         struct net_device *dev;
775         int err;
776
777         dev = __dev_get_by_name(net, ifr->ifr_name);
778         if (dev) {
779                 if ((ifr->ifr_flags & IFF_TUN) && dev->netdev_ops == &tun_netdev_ops)
780                         tun = netdev_priv(dev);
781                 else if ((ifr->ifr_flags & IFF_TAP) && dev->netdev_ops == &tap_netdev_ops)
782                         tun = netdev_priv(dev);
783                 else
784                         return -EINVAL;
785
786                 err = tun_attach(tun, file);
787                 if (err < 0)
788                         return err;
789         }
790         else {
791                 char *name;
792                 unsigned long flags = 0;
793
794                 err = -EINVAL;
795
796                 if (!capable(CAP_NET_ADMIN))
797                         return -EPERM;
798
799                 /* Set dev type */
800                 if (ifr->ifr_flags & IFF_TUN) {
801                         /* TUN device */
802                         flags |= TUN_TUN_DEV;
803                         name = "tun%d";
804                 } else if (ifr->ifr_flags & IFF_TAP) {
805                         /* TAP device */
806                         flags |= TUN_TAP_DEV;
807                         name = "tap%d";
808                 } else
809                         goto failed;
810
811                 if (*ifr->ifr_name)
812                         name = ifr->ifr_name;
813
814                 dev = alloc_netdev(sizeof(struct tun_struct), name,
815                                    tun_setup);
816                 if (!dev)
817                         return -ENOMEM;
818
819                 dev_net_set(dev, net);
820
821                 tun = netdev_priv(dev);
822                 tun->dev = dev;
823                 tun->flags = flags;
824                 tun->txflt.count = 0;
825
826                 tun_net_init(dev);
827
828                 if (strchr(dev->name, '%')) {
829                         err = dev_alloc_name(dev, dev->name);
830                         if (err < 0)
831                                 goto err_free_dev;
832                 }
833
834                 err = register_netdevice(tun->dev);
835                 if (err < 0)
836                         goto err_free_dev;
837
838                 err = tun_attach(tun, file);
839                 if (err < 0)
840                         goto err_free_dev;
841         }
842
843         DBG(KERN_INFO "%s: tun_set_iff\n", tun->dev->name);
844
845         if (ifr->ifr_flags & IFF_NO_PI)
846                 tun->flags |= TUN_NO_PI;
847         else
848                 tun->flags &= ~TUN_NO_PI;
849
850         if (ifr->ifr_flags & IFF_ONE_QUEUE)
851                 tun->flags |= TUN_ONE_QUEUE;
852         else
853                 tun->flags &= ~TUN_ONE_QUEUE;
854
855         if (ifr->ifr_flags & IFF_VNET_HDR)
856                 tun->flags |= TUN_VNET_HDR;
857         else
858                 tun->flags &= ~TUN_VNET_HDR;
859
860         /* Make sure persistent devices do not get stuck in
861          * xoff state.
862          */
863         if (netif_running(tun->dev))
864                 netif_wake_queue(tun->dev);
865
866         strcpy(ifr->ifr_name, tun->dev->name);
867         return 0;
868
869  err_free_dev:
870         free_netdev(dev);
871  failed:
872         return err;
873 }
874
875 static int tun_get_iff(struct net *net, struct file *file, struct ifreq *ifr)
876 {
877         struct tun_struct *tun = tun_get(file);
878
879         if (!tun)
880                 return -EBADFD;
881
882         DBG(KERN_INFO "%s: tun_get_iff\n", tun->dev->name);
883
884         strcpy(ifr->ifr_name, tun->dev->name);
885
886         ifr->ifr_flags = 0;
887
888         if (ifr->ifr_flags & TUN_TUN_DEV)
889                 ifr->ifr_flags |= IFF_TUN;
890         else
891                 ifr->ifr_flags |= IFF_TAP;
892
893         if (tun->flags & TUN_NO_PI)
894                 ifr->ifr_flags |= IFF_NO_PI;
895
896         if (tun->flags & TUN_ONE_QUEUE)
897                 ifr->ifr_flags |= IFF_ONE_QUEUE;
898
899         if (tun->flags & TUN_VNET_HDR)
900                 ifr->ifr_flags |= IFF_VNET_HDR;
901
902         tun_put(tun);
903         return 0;
904 }
905
906 /* This is like a cut-down ethtool ops, except done via tun fd so no
907  * privs required. */
908 static int set_offload(struct net_device *dev, unsigned long arg)
909 {
910         unsigned int old_features, features;
911
912         old_features = dev->features;
913         /* Unset features, set them as we chew on the arg. */
914         features = (old_features & ~(NETIF_F_HW_CSUM|NETIF_F_SG|NETIF_F_FRAGLIST
915                                     |NETIF_F_TSO_ECN|NETIF_F_TSO|NETIF_F_TSO6));
916
917         if (arg & TUN_F_CSUM) {
918                 features |= NETIF_F_HW_CSUM|NETIF_F_SG|NETIF_F_FRAGLIST;
919                 arg &= ~TUN_F_CSUM;
920
921                 if (arg & (TUN_F_TSO4|TUN_F_TSO6)) {
922                         if (arg & TUN_F_TSO_ECN) {
923                                 features |= NETIF_F_TSO_ECN;
924                                 arg &= ~TUN_F_TSO_ECN;
925                         }
926                         if (arg & TUN_F_TSO4)
927                                 features |= NETIF_F_TSO;
928                         if (arg & TUN_F_TSO6)
929                                 features |= NETIF_F_TSO6;
930                         arg &= ~(TUN_F_TSO4|TUN_F_TSO6);
931                 }
932         }
933
934         /* This gives the user a way to test for new features in future by
935          * trying to set them. */
936         if (arg)
937                 return -EINVAL;
938
939         dev->features = features;
940         if (old_features != dev->features)
941                 netdev_features_change(dev);
942
943         return 0;
944 }
945
946 static int tun_chr_ioctl(struct inode *inode, struct file *file,
947                          unsigned int cmd, unsigned long arg)
948 {
949         struct tun_file *tfile = file->private_data;
950         struct tun_struct *tun;
951         void __user* argp = (void __user*)arg;
952         struct ifreq ifr;
953         int ret;
954
955         if (cmd == TUNSETIFF || _IOC_TYPE(cmd) == 0x89)
956                 if (copy_from_user(&ifr, argp, sizeof ifr))
957                         return -EFAULT;
958
959         if (cmd == TUNGETFEATURES) {
960                 /* Currently this just means: "what IFF flags are valid?".
961                  * This is needed because we never checked for invalid flags on
962                  * TUNSETIFF. */
963                 return put_user(IFF_TUN | IFF_TAP | IFF_NO_PI | IFF_ONE_QUEUE |
964                                 IFF_VNET_HDR,
965                                 (unsigned int __user*)argp);
966         }
967
968         tun = __tun_get(tfile);
969         if (cmd == TUNSETIFF && !tun) {
970                 int err;
971
972                 ifr.ifr_name[IFNAMSIZ-1] = '\0';
973
974                 rtnl_lock();
975                 err = tun_set_iff(tfile->net, file, &ifr);
976                 rtnl_unlock();
977
978                 if (err)
979                         return err;
980
981                 if (copy_to_user(argp, &ifr, sizeof(ifr)))
982                         return -EFAULT;
983                 return 0;
984         }
985
986
987         if (!tun)
988                 return -EBADFD;
989
990         DBG(KERN_INFO "%s: tun_chr_ioctl cmd %d\n", tun->dev->name, cmd);
991
992         ret = 0;
993         switch (cmd) {
994         case TUNGETIFF:
995                 ret = tun_get_iff(current->nsproxy->net_ns, file, &ifr);
996                 if (ret)
997                         break;
998
999                 if (copy_to_user(argp, &ifr, sizeof(ifr)))
1000                         ret = -EFAULT;
1001                 break;
1002
1003         case TUNSETNOCSUM:
1004                 /* Disable/Enable checksum */
1005                 if (arg)
1006                         tun->flags |= TUN_NOCHECKSUM;
1007                 else
1008                         tun->flags &= ~TUN_NOCHECKSUM;
1009
1010                 DBG(KERN_INFO "%s: checksum %s\n",
1011                     tun->dev->name, arg ? "disabled" : "enabled");
1012                 break;
1013
1014         case TUNSETPERSIST:
1015                 /* Disable/Enable persist mode */
1016                 if (arg)
1017                         tun->flags |= TUN_PERSIST;
1018                 else
1019                         tun->flags &= ~TUN_PERSIST;
1020
1021                 DBG(KERN_INFO "%s: persist %s\n",
1022                     tun->dev->name, arg ? "enabled" : "disabled");
1023                 break;
1024
1025         case TUNSETOWNER:
1026                 /* Set owner of the device */
1027                 tun->owner = (uid_t) arg;
1028
1029                 DBG(KERN_INFO "%s: owner set to %d\n", tun->dev->name, tun->owner);
1030                 break;
1031
1032         case TUNSETGROUP:
1033                 /* Set group of the device */
1034                 tun->group= (gid_t) arg;
1035
1036                 DBG(KERN_INFO "%s: group set to %d\n", tun->dev->name, tun->group);
1037                 break;
1038
1039         case TUNSETLINK:
1040                 /* Only allow setting the type when the interface is down */
1041                 rtnl_lock();
1042                 if (tun->dev->flags & IFF_UP) {
1043                         DBG(KERN_INFO "%s: Linktype set failed because interface is up\n",
1044                                 tun->dev->name);
1045                         ret = -EBUSY;
1046                 } else {
1047                         tun->dev->type = (int) arg;
1048                         DBG(KERN_INFO "%s: linktype set to %d\n", tun->dev->name, tun->dev->type);
1049                         ret = 0;
1050                 }
1051                 rtnl_unlock();
1052                 break;
1053
1054 #ifdef TUN_DEBUG
1055         case TUNSETDEBUG:
1056                 tun->debug = arg;
1057                 break;
1058 #endif
1059         case TUNSETOFFLOAD:
1060                 rtnl_lock();
1061                 ret = set_offload(tun->dev, arg);
1062                 rtnl_unlock();
1063                 break;
1064
1065         case TUNSETTXFILTER:
1066                 /* Can be set only for TAPs */
1067                 ret = -EINVAL;
1068                 if ((tun->flags & TUN_TYPE_MASK) != TUN_TAP_DEV)
1069                         break;
1070                 rtnl_lock();
1071                 ret = update_filter(&tun->txflt, (void __user *)arg);
1072                 rtnl_unlock();
1073                 break;
1074
1075         case SIOCGIFHWADDR:
1076                 /* Get hw addres */
1077                 memcpy(ifr.ifr_hwaddr.sa_data, tun->dev->dev_addr, ETH_ALEN);
1078                 ifr.ifr_hwaddr.sa_family = tun->dev->type;
1079                 if (copy_to_user(argp, &ifr, sizeof ifr))
1080                         ret = -EFAULT;
1081                 break;
1082
1083         case SIOCSIFHWADDR:
1084                 /* Set hw address */
1085                 DBG(KERN_DEBUG "%s: set hw address: %pM\n",
1086                         tun->dev->name, ifr.ifr_hwaddr.sa_data);
1087
1088                 rtnl_lock();
1089                 ret = dev_set_mac_address(tun->dev, &ifr.ifr_hwaddr);
1090                 rtnl_unlock();
1091                 break;
1092         default:
1093                 ret = -EINVAL;
1094                 break;
1095         };
1096
1097         tun_put(tun);
1098         return ret;
1099 }
1100
1101 static int tun_chr_fasync(int fd, struct file *file, int on)
1102 {
1103         struct tun_struct *tun = tun_get(file);
1104         int ret;
1105
1106         if (!tun)
1107                 return -EBADFD;
1108
1109         DBG(KERN_INFO "%s: tun_chr_fasync %d\n", tun->dev->name, on);
1110
1111         lock_kernel();
1112         if ((ret = fasync_helper(fd, file, on, &tun->fasync)) < 0)
1113                 goto out;
1114
1115         if (on) {
1116                 ret = __f_setown(file, task_pid(current), PIDTYPE_PID, 0);
1117                 if (ret)
1118                         goto out;
1119                 tun->flags |= TUN_FASYNC;
1120         } else
1121                 tun->flags &= ~TUN_FASYNC;
1122         ret = 0;
1123 out:
1124         unlock_kernel();
1125         tun_put(tun);
1126         return ret;
1127 }
1128
1129 static int tun_chr_open(struct inode *inode, struct file * file)
1130 {
1131         struct tun_file *tfile;
1132         cycle_kernel_lock();
1133         DBG1(KERN_INFO "tunX: tun_chr_open\n");
1134
1135         tfile = kmalloc(sizeof(*tfile), GFP_KERNEL);
1136         if (!tfile)
1137                 return -ENOMEM;
1138         tfile->tun = NULL;
1139         tfile->net = get_net(current->nsproxy->net_ns);
1140         init_waitqueue_head(&tfile->read_wait);
1141         file->private_data = tfile;
1142         return 0;
1143 }
1144
1145 static int tun_chr_close(struct inode *inode, struct file *file)
1146 {
1147         struct tun_file *tfile = file->private_data;
1148         struct tun_struct *tun = __tun_get(tfile);
1149
1150
1151         if (tun) {
1152                 DBG(KERN_INFO "%s: tun_chr_close\n", tun->dev->name);
1153
1154                 rtnl_lock();
1155                 __tun_detach(tun);
1156
1157                 /* If desireable, unregister the netdevice. */
1158                 if (!(tun->flags & TUN_PERSIST))
1159                         unregister_netdevice(tun->dev);
1160
1161                 rtnl_unlock();
1162         }
1163
1164         put_net(tfile->net);
1165         kfree(tfile);
1166
1167         return 0;
1168 }
1169
1170 static const struct file_operations tun_fops = {
1171         .owner  = THIS_MODULE,
1172         .llseek = no_llseek,
1173         .read  = do_sync_read,
1174         .aio_read  = tun_chr_aio_read,
1175         .write = do_sync_write,
1176         .aio_write = tun_chr_aio_write,
1177         .poll   = tun_chr_poll,
1178         .ioctl  = tun_chr_ioctl,
1179         .open   = tun_chr_open,
1180         .release = tun_chr_close,
1181         .fasync = tun_chr_fasync
1182 };
1183
1184 static struct miscdevice tun_miscdev = {
1185         .minor = TUN_MINOR,
1186         .name = "tun",
1187         .fops = &tun_fops,
1188 };
1189
1190 /* ethtool interface */
1191
1192 static int tun_get_settings(struct net_device *dev, struct ethtool_cmd *cmd)
1193 {
1194         cmd->supported          = 0;
1195         cmd->advertising        = 0;
1196         cmd->speed              = SPEED_10;
1197         cmd->duplex             = DUPLEX_FULL;
1198         cmd->port               = PORT_TP;
1199         cmd->phy_address        = 0;
1200         cmd->transceiver        = XCVR_INTERNAL;
1201         cmd->autoneg            = AUTONEG_DISABLE;
1202         cmd->maxtxpkt           = 0;
1203         cmd->maxrxpkt           = 0;
1204         return 0;
1205 }
1206
1207 static void tun_get_drvinfo(struct net_device *dev, struct ethtool_drvinfo *info)
1208 {
1209         struct tun_struct *tun = netdev_priv(dev);
1210
1211         strcpy(info->driver, DRV_NAME);
1212         strcpy(info->version, DRV_VERSION);
1213         strcpy(info->fw_version, "N/A");
1214
1215         switch (tun->flags & TUN_TYPE_MASK) {
1216         case TUN_TUN_DEV:
1217                 strcpy(info->bus_info, "tun");
1218                 break;
1219         case TUN_TAP_DEV:
1220                 strcpy(info->bus_info, "tap");
1221                 break;
1222         }
1223 }
1224
1225 static u32 tun_get_msglevel(struct net_device *dev)
1226 {
1227 #ifdef TUN_DEBUG
1228         struct tun_struct *tun = netdev_priv(dev);
1229         return tun->debug;
1230 #else
1231         return -EOPNOTSUPP;
1232 #endif
1233 }
1234
1235 static void tun_set_msglevel(struct net_device *dev, u32 value)
1236 {
1237 #ifdef TUN_DEBUG
1238         struct tun_struct *tun = netdev_priv(dev);
1239         tun->debug = value;
1240 #endif
1241 }
1242
1243 static u32 tun_get_link(struct net_device *dev)
1244 {
1245         struct tun_struct *tun = netdev_priv(dev);
1246         return !!tun->tfile;
1247 }
1248
1249 static u32 tun_get_rx_csum(struct net_device *dev)
1250 {
1251         struct tun_struct *tun = netdev_priv(dev);
1252         return (tun->flags & TUN_NOCHECKSUM) == 0;
1253 }
1254
1255 static int tun_set_rx_csum(struct net_device *dev, u32 data)
1256 {
1257         struct tun_struct *tun = netdev_priv(dev);
1258         if (data)
1259                 tun->flags &= ~TUN_NOCHECKSUM;
1260         else
1261                 tun->flags |= TUN_NOCHECKSUM;
1262         return 0;
1263 }
1264
1265 static const struct ethtool_ops tun_ethtool_ops = {
1266         .get_settings   = tun_get_settings,
1267         .get_drvinfo    = tun_get_drvinfo,
1268         .get_msglevel   = tun_get_msglevel,
1269         .set_msglevel   = tun_set_msglevel,
1270         .get_link       = tun_get_link,
1271         .get_rx_csum    = tun_get_rx_csum,
1272         .set_rx_csum    = tun_set_rx_csum
1273 };
1274
1275 static int tun_init_net(struct net *net)
1276 {
1277         return 0;
1278 }
1279
1280 static void tun_exit_net(struct net *net)
1281 {
1282         struct net_device *dev, *next;
1283
1284         rtnl_lock();
1285         for_each_netdev_safe(net, dev, next) {
1286                 if (dev->ethtool_ops != &tun_ethtool_ops)
1287                         continue;
1288                 DBG(KERN_INFO "%s cleaned up\n", dev->name);
1289                 unregister_netdevice(dev);
1290         }
1291         rtnl_unlock();
1292 }
1293
1294 static struct pernet_operations tun_net_ops = {
1295         .init = tun_init_net,
1296         .exit = tun_exit_net,
1297 };
1298
1299 static int __init tun_init(void)
1300 {
1301         int ret = 0;
1302
1303         printk(KERN_INFO "tun: %s, %s\n", DRV_DESCRIPTION, DRV_VERSION);
1304         printk(KERN_INFO "tun: %s\n", DRV_COPYRIGHT);
1305
1306         ret = register_pernet_device(&tun_net_ops);
1307         if (ret) {
1308                 printk(KERN_ERR "tun: Can't register pernet ops\n");
1309                 goto err_pernet;
1310         }
1311
1312         ret = misc_register(&tun_miscdev);
1313         if (ret) {
1314                 printk(KERN_ERR "tun: Can't register misc device %d\n", TUN_MINOR);
1315                 goto err_misc;
1316         }
1317         return 0;
1318
1319 err_misc:
1320         unregister_pernet_device(&tun_net_ops);
1321 err_pernet:
1322         return ret;
1323 }
1324
1325 static void tun_cleanup(void)
1326 {
1327         misc_deregister(&tun_miscdev);
1328         unregister_pernet_device(&tun_net_ops);
1329 }
1330
1331 module_init(tun_init);
1332 module_exit(tun_cleanup);
1333 MODULE_DESCRIPTION(DRV_DESCRIPTION);
1334 MODULE_AUTHOR(DRV_COPYRIGHT);
1335 MODULE_LICENSE("GPL");
1336 MODULE_ALIAS_MISCDEV(TUN_MINOR);