98ee6cc2875d0500ce5e5de0d34e3ac026a0aaa7
[sfrench/cifs-2.6.git] / drivers / net / tap.c
1 #include <linux/etherdevice.h>
2 #include <linux/if_tap.h>
3 #include <linux/if_vlan.h>
4 #include <linux/interrupt.h>
5 #include <linux/nsproxy.h>
6 #include <linux/compat.h>
7 #include <linux/if_tun.h>
8 #include <linux/module.h>
9 #include <linux/skbuff.h>
10 #include <linux/cache.h>
11 #include <linux/sched/signal.h>
12 #include <linux/types.h>
13 #include <linux/slab.h>
14 #include <linux/wait.h>
15 #include <linux/cdev.h>
16 #include <linux/idr.h>
17 #include <linux/fs.h>
18 #include <linux/uio.h>
19
20 #include <net/net_namespace.h>
21 #include <net/rtnetlink.h>
22 #include <net/sock.h>
23 #include <linux/virtio_net.h>
24 #include <linux/skb_array.h>
25
26 #define TAP_IFFEATURES (IFF_VNET_HDR | IFF_MULTI_QUEUE)
27
28 #define TAP_VNET_LE 0x80000000
29 #define TAP_VNET_BE 0x40000000
30
31 #ifdef CONFIG_TUN_VNET_CROSS_LE
32 static inline bool tap_legacy_is_little_endian(struct tap_queue *q)
33 {
34         return q->flags & TAP_VNET_BE ? false :
35                 virtio_legacy_is_little_endian();
36 }
37
38 static long tap_get_vnet_be(struct tap_queue *q, int __user *sp)
39 {
40         int s = !!(q->flags & TAP_VNET_BE);
41
42         if (put_user(s, sp))
43                 return -EFAULT;
44
45         return 0;
46 }
47
48 static long tap_set_vnet_be(struct tap_queue *q, int __user *sp)
49 {
50         int s;
51
52         if (get_user(s, sp))
53                 return -EFAULT;
54
55         if (s)
56                 q->flags |= TAP_VNET_BE;
57         else
58                 q->flags &= ~TAP_VNET_BE;
59
60         return 0;
61 }
62 #else
63 static inline bool tap_legacy_is_little_endian(struct tap_queue *q)
64 {
65         return virtio_legacy_is_little_endian();
66 }
67
68 static long tap_get_vnet_be(struct tap_queue *q, int __user *argp)
69 {
70         return -EINVAL;
71 }
72
73 static long tap_set_vnet_be(struct tap_queue *q, int __user *argp)
74 {
75         return -EINVAL;
76 }
77 #endif /* CONFIG_TUN_VNET_CROSS_LE */
78
79 static inline bool tap_is_little_endian(struct tap_queue *q)
80 {
81         return q->flags & TAP_VNET_LE ||
82                 tap_legacy_is_little_endian(q);
83 }
84
85 static inline u16 tap16_to_cpu(struct tap_queue *q, __virtio16 val)
86 {
87         return __virtio16_to_cpu(tap_is_little_endian(q), val);
88 }
89
90 static inline __virtio16 cpu_to_tap16(struct tap_queue *q, u16 val)
91 {
92         return __cpu_to_virtio16(tap_is_little_endian(q), val);
93 }
94
95 static struct proto tap_proto = {
96         .name = "tap",
97         .owner = THIS_MODULE,
98         .obj_size = sizeof(struct tap_queue),
99 };
100
101 #define TAP_NUM_DEVS (1U << MINORBITS)
102
103 static LIST_HEAD(major_list);
104
105 struct major_info {
106         struct rcu_head rcu;
107         dev_t major;
108         struct idr minor_idr;
109         spinlock_t minor_lock;
110         const char *device_name;
111         struct list_head next;
112 };
113
114 #define GOODCOPY_LEN 128
115
116 static const struct proto_ops tap_socket_ops;
117
118 #define RX_OFFLOADS (NETIF_F_GRO | NETIF_F_LRO)
119 #define TAP_FEATURES (NETIF_F_GSO | NETIF_F_SG | NETIF_F_FRAGLIST)
120
121 static struct tap_dev *tap_dev_get_rcu(const struct net_device *dev)
122 {
123         return rcu_dereference(dev->rx_handler_data);
124 }
125
126 /*
127  * RCU usage:
128  * The tap_queue and the macvlan_dev are loosely coupled, the
129  * pointers from one to the other can only be read while rcu_read_lock
130  * or rtnl is held.
131  *
132  * Both the file and the macvlan_dev hold a reference on the tap_queue
133  * through sock_hold(&q->sk). When the macvlan_dev goes away first,
134  * q->vlan becomes inaccessible. When the files gets closed,
135  * tap_get_queue() fails.
136  *
137  * There may still be references to the struct sock inside of the
138  * queue from outbound SKBs, but these never reference back to the
139  * file or the dev. The data structure is freed through __sk_free
140  * when both our references and any pending SKBs are gone.
141  */
142
143 static int tap_enable_queue(struct tap_dev *tap, struct file *file,
144                             struct tap_queue *q)
145 {
146         int err = -EINVAL;
147
148         ASSERT_RTNL();
149
150         if (q->enabled)
151                 goto out;
152
153         err = 0;
154         rcu_assign_pointer(tap->taps[tap->numvtaps], q);
155         q->queue_index = tap->numvtaps;
156         q->enabled = true;
157
158         tap->numvtaps++;
159 out:
160         return err;
161 }
162
163 /* Requires RTNL */
164 static int tap_set_queue(struct tap_dev *tap, struct file *file,
165                          struct tap_queue *q)
166 {
167         if (tap->numqueues == MAX_TAP_QUEUES)
168                 return -EBUSY;
169
170         rcu_assign_pointer(q->tap, tap);
171         rcu_assign_pointer(tap->taps[tap->numvtaps], q);
172         sock_hold(&q->sk);
173
174         q->file = file;
175         q->queue_index = tap->numvtaps;
176         q->enabled = true;
177         file->private_data = q;
178         list_add_tail(&q->next, &tap->queue_list);
179
180         tap->numvtaps++;
181         tap->numqueues++;
182
183         return 0;
184 }
185
186 static int tap_disable_queue(struct tap_queue *q)
187 {
188         struct tap_dev *tap;
189         struct tap_queue *nq;
190
191         ASSERT_RTNL();
192         if (!q->enabled)
193                 return -EINVAL;
194
195         tap = rtnl_dereference(q->tap);
196
197         if (tap) {
198                 int index = q->queue_index;
199                 BUG_ON(index >= tap->numvtaps);
200                 nq = rtnl_dereference(tap->taps[tap->numvtaps - 1]);
201                 nq->queue_index = index;
202
203                 rcu_assign_pointer(tap->taps[index], nq);
204                 RCU_INIT_POINTER(tap->taps[tap->numvtaps - 1], NULL);
205                 q->enabled = false;
206
207                 tap->numvtaps--;
208         }
209
210         return 0;
211 }
212
213 /*
214  * The file owning the queue got closed, give up both
215  * the reference that the files holds as well as the
216  * one from the macvlan_dev if that still exists.
217  *
218  * Using the spinlock makes sure that we don't get
219  * to the queue again after destroying it.
220  */
221 static void tap_put_queue(struct tap_queue *q)
222 {
223         struct tap_dev *tap;
224
225         rtnl_lock();
226         tap = rtnl_dereference(q->tap);
227
228         if (tap) {
229                 if (q->enabled)
230                         BUG_ON(tap_disable_queue(q));
231
232                 tap->numqueues--;
233                 RCU_INIT_POINTER(q->tap, NULL);
234                 sock_put(&q->sk);
235                 list_del_init(&q->next);
236         }
237
238         rtnl_unlock();
239
240         synchronize_rcu();
241         sock_put(&q->sk);
242 }
243
244 /*
245  * Select a queue based on the rxq of the device on which this packet
246  * arrived. If the incoming device is not mq, calculate a flow hash
247  * to select a queue. If all fails, find the first available queue.
248  * Cache vlan->numvtaps since it can become zero during the execution
249  * of this function.
250  */
251 static struct tap_queue *tap_get_queue(struct tap_dev *tap,
252                                        struct sk_buff *skb)
253 {
254         struct tap_queue *queue = NULL;
255         /* Access to taps array is protected by rcu, but access to numvtaps
256          * isn't. Below we use it to lookup a queue, but treat it as a hint
257          * and validate that the result isn't NULL - in case we are
258          * racing against queue removal.
259          */
260         int numvtaps = ACCESS_ONCE(tap->numvtaps);
261         __u32 rxq;
262
263         if (!numvtaps)
264                 goto out;
265
266         if (numvtaps == 1)
267                 goto single;
268
269         /* Check if we can use flow to select a queue */
270         rxq = skb_get_hash(skb);
271         if (rxq) {
272                 queue = rcu_dereference(tap->taps[rxq % numvtaps]);
273                 goto out;
274         }
275
276         if (likely(skb_rx_queue_recorded(skb))) {
277                 rxq = skb_get_rx_queue(skb);
278
279                 while (unlikely(rxq >= numvtaps))
280                         rxq -= numvtaps;
281
282                 queue = rcu_dereference(tap->taps[rxq]);
283                 goto out;
284         }
285
286 single:
287         queue = rcu_dereference(tap->taps[0]);
288 out:
289         return queue;
290 }
291
292 /*
293  * The net_device is going away, give up the reference
294  * that it holds on all queues and safely set the pointer
295  * from the queues to NULL.
296  */
297 void tap_del_queues(struct tap_dev *tap)
298 {
299         struct tap_queue *q, *tmp;
300
301         ASSERT_RTNL();
302         list_for_each_entry_safe(q, tmp, &tap->queue_list, next) {
303                 list_del_init(&q->next);
304                 RCU_INIT_POINTER(q->tap, NULL);
305                 if (q->enabled)
306                         tap->numvtaps--;
307                 tap->numqueues--;
308                 sock_put(&q->sk);
309         }
310         BUG_ON(tap->numvtaps);
311         BUG_ON(tap->numqueues);
312         /* guarantee that any future tap_set_queue will fail */
313         tap->numvtaps = MAX_TAP_QUEUES;
314 }
315 EXPORT_SYMBOL_GPL(tap_del_queues);
316
317 rx_handler_result_t tap_handle_frame(struct sk_buff **pskb)
318 {
319         struct sk_buff *skb = *pskb;
320         struct net_device *dev = skb->dev;
321         struct tap_dev *tap;
322         struct tap_queue *q;
323         netdev_features_t features = TAP_FEATURES;
324
325         tap = tap_dev_get_rcu(dev);
326         if (!tap)
327                 return RX_HANDLER_PASS;
328
329         q = tap_get_queue(tap, skb);
330         if (!q)
331                 return RX_HANDLER_PASS;
332
333         if (__skb_array_full(&q->skb_array))
334                 goto drop;
335
336         skb_push(skb, ETH_HLEN);
337
338         /* Apply the forward feature mask so that we perform segmentation
339          * according to users wishes.  This only works if VNET_HDR is
340          * enabled.
341          */
342         if (q->flags & IFF_VNET_HDR)
343                 features |= tap->tap_features;
344         if (netif_needs_gso(skb, features)) {
345                 struct sk_buff *segs = __skb_gso_segment(skb, features, false);
346
347                 if (IS_ERR(segs))
348                         goto drop;
349
350                 if (!segs) {
351                         if (skb_array_produce(&q->skb_array, skb))
352                                 goto drop;
353                         goto wake_up;
354                 }
355
356                 consume_skb(skb);
357                 while (segs) {
358                         struct sk_buff *nskb = segs->next;
359
360                         segs->next = NULL;
361                         if (skb_array_produce(&q->skb_array, segs)) {
362                                 kfree_skb(segs);
363                                 kfree_skb_list(nskb);
364                                 break;
365                         }
366                         segs = nskb;
367                 }
368         } else {
369                 /* If we receive a partial checksum and the tap side
370                  * doesn't support checksum offload, compute the checksum.
371                  * Note: it doesn't matter which checksum feature to
372                  *        check, we either support them all or none.
373                  */
374                 if (skb->ip_summed == CHECKSUM_PARTIAL &&
375                     !(features & NETIF_F_CSUM_MASK) &&
376                     skb_checksum_help(skb))
377                         goto drop;
378                 if (skb_array_produce(&q->skb_array, skb))
379                         goto drop;
380         }
381
382 wake_up:
383         wake_up_interruptible_poll(sk_sleep(&q->sk), POLLIN | POLLRDNORM | POLLRDBAND);
384         return RX_HANDLER_CONSUMED;
385
386 drop:
387         /* Count errors/drops only here, thus don't care about args. */
388         if (tap->count_rx_dropped)
389                 tap->count_rx_dropped(tap);
390         kfree_skb(skb);
391         return RX_HANDLER_CONSUMED;
392 }
393 EXPORT_SYMBOL_GPL(tap_handle_frame);
394
395 static struct major_info *tap_get_major(int major)
396 {
397         struct major_info *tap_major;
398
399         list_for_each_entry_rcu(tap_major, &major_list, next) {
400                 if (tap_major->major == major)
401                         return tap_major;
402         }
403
404         return NULL;
405 }
406
407 int tap_get_minor(dev_t major, struct tap_dev *tap)
408 {
409         int retval = -ENOMEM;
410         struct major_info *tap_major;
411
412         rcu_read_lock();
413         tap_major = tap_get_major(MAJOR(major));
414         if (!tap_major) {
415                 retval = -EINVAL;
416                 goto unlock;
417         }
418
419         spin_lock(&tap_major->minor_lock);
420         retval = idr_alloc(&tap_major->minor_idr, tap, 1, TAP_NUM_DEVS, GFP_ATOMIC);
421         if (retval >= 0) {
422                 tap->minor = retval;
423         } else if (retval == -ENOSPC) {
424                 netdev_err(tap->dev, "Too many tap devices\n");
425                 retval = -EINVAL;
426         }
427         spin_unlock(&tap_major->minor_lock);
428
429 unlock:
430         rcu_read_unlock();
431         return retval < 0 ? retval : 0;
432 }
433 EXPORT_SYMBOL_GPL(tap_get_minor);
434
435 void tap_free_minor(dev_t major, struct tap_dev *tap)
436 {
437         struct major_info *tap_major;
438
439         rcu_read_lock();
440         tap_major = tap_get_major(MAJOR(major));
441         if (!tap_major) {
442                 goto unlock;
443         }
444
445         spin_lock(&tap_major->minor_lock);
446         if (tap->minor) {
447                 idr_remove(&tap_major->minor_idr, tap->minor);
448                 tap->minor = 0;
449         }
450         spin_unlock(&tap_major->minor_lock);
451
452 unlock:
453         rcu_read_unlock();
454 }
455 EXPORT_SYMBOL_GPL(tap_free_minor);
456
457 static struct tap_dev *dev_get_by_tap_file(int major, int minor)
458 {
459         struct net_device *dev = NULL;
460         struct tap_dev *tap;
461         struct major_info *tap_major;
462
463         rcu_read_lock();
464         tap_major = tap_get_major(major);
465         if (!tap_major) {
466                 tap = NULL;
467                 goto unlock;
468         }
469
470         spin_lock(&tap_major->minor_lock);
471         tap = idr_find(&tap_major->minor_idr, minor);
472         if (tap) {
473                 dev = tap->dev;
474                 dev_hold(dev);
475         }
476         spin_unlock(&tap_major->minor_lock);
477
478 unlock:
479         rcu_read_unlock();
480         return tap;
481 }
482
483 static void tap_sock_write_space(struct sock *sk)
484 {
485         wait_queue_head_t *wqueue;
486
487         if (!sock_writeable(sk) ||
488             !test_and_clear_bit(SOCKWQ_ASYNC_NOSPACE, &sk->sk_socket->flags))
489                 return;
490
491         wqueue = sk_sleep(sk);
492         if (wqueue && waitqueue_active(wqueue))
493                 wake_up_interruptible_poll(wqueue, POLLOUT | POLLWRNORM | POLLWRBAND);
494 }
495
496 static void tap_sock_destruct(struct sock *sk)
497 {
498         struct tap_queue *q = container_of(sk, struct tap_queue, sk);
499
500         skb_array_cleanup(&q->skb_array);
501 }
502
503 static int tap_open(struct inode *inode, struct file *file)
504 {
505         struct net *net = current->nsproxy->net_ns;
506         struct tap_dev *tap;
507         struct tap_queue *q;
508         int err = -ENODEV;
509
510         rtnl_lock();
511         tap = dev_get_by_tap_file(imajor(inode), iminor(inode));
512         if (!tap)
513                 goto err;
514
515         err = -ENOMEM;
516         q = (struct tap_queue *)sk_alloc(net, AF_UNSPEC, GFP_KERNEL,
517                                              &tap_proto, 0);
518         if (!q)
519                 goto err;
520         if (skb_array_init(&q->skb_array, tap->dev->tx_queue_len, GFP_KERNEL)) {
521                 sk_free(&q->sk);
522                 goto err;
523         }
524
525         RCU_INIT_POINTER(q->sock.wq, &q->wq);
526         init_waitqueue_head(&q->wq.wait);
527         q->sock.type = SOCK_RAW;
528         q->sock.state = SS_CONNECTED;
529         q->sock.file = file;
530         q->sock.ops = &tap_socket_ops;
531         sock_init_data(&q->sock, &q->sk);
532         q->sk.sk_write_space = tap_sock_write_space;
533         q->sk.sk_destruct = tap_sock_destruct;
534         q->flags = IFF_VNET_HDR | IFF_NO_PI | IFF_TAP;
535         q->vnet_hdr_sz = sizeof(struct virtio_net_hdr);
536
537         /*
538          * so far only KVM virtio_net uses tap, enable zero copy between
539          * guest kernel and host kernel when lower device supports zerocopy
540          *
541          * The macvlan supports zerocopy iff the lower device supports zero
542          * copy so we don't have to look at the lower device directly.
543          */
544         if ((tap->dev->features & NETIF_F_HIGHDMA) && (tap->dev->features & NETIF_F_SG))
545                 sock_set_flag(&q->sk, SOCK_ZEROCOPY);
546
547         err = tap_set_queue(tap, file, q);
548         if (err) {
549                 /* tap_sock_destruct() will take care of freeing skb_array */
550                 goto err_put;
551         }
552
553         dev_put(tap->dev);
554
555         rtnl_unlock();
556         return err;
557
558 err_put:
559         sock_put(&q->sk);
560 err:
561         if (tap)
562                 dev_put(tap->dev);
563
564         rtnl_unlock();
565         return err;
566 }
567
568 static int tap_release(struct inode *inode, struct file *file)
569 {
570         struct tap_queue *q = file->private_data;
571         tap_put_queue(q);
572         return 0;
573 }
574
575 static unsigned int tap_poll(struct file *file, poll_table *wait)
576 {
577         struct tap_queue *q = file->private_data;
578         unsigned int mask = POLLERR;
579
580         if (!q)
581                 goto out;
582
583         mask = 0;
584         poll_wait(file, &q->wq.wait, wait);
585
586         if (!skb_array_empty(&q->skb_array))
587                 mask |= POLLIN | POLLRDNORM;
588
589         if (sock_writeable(&q->sk) ||
590             (!test_and_set_bit(SOCKWQ_ASYNC_NOSPACE, &q->sock.flags) &&
591              sock_writeable(&q->sk)))
592                 mask |= POLLOUT | POLLWRNORM;
593
594 out:
595         return mask;
596 }
597
598 static inline struct sk_buff *tap_alloc_skb(struct sock *sk, size_t prepad,
599                                             size_t len, size_t linear,
600                                                 int noblock, int *err)
601 {
602         struct sk_buff *skb;
603
604         /* Under a page?  Don't bother with paged skb. */
605         if (prepad + len < PAGE_SIZE || !linear)
606                 linear = len;
607
608         skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock,
609                                    err, 0);
610         if (!skb)
611                 return NULL;
612
613         skb_reserve(skb, prepad);
614         skb_put(skb, linear);
615         skb->data_len = len - linear;
616         skb->len += len - linear;
617
618         return skb;
619 }
620
621 /* Neighbour code has some assumptions on HH_DATA_MOD alignment */
622 #define TAP_RESERVE HH_DATA_OFF(ETH_HLEN)
623
624 /* Get packet from user space buffer */
625 static ssize_t tap_get_user(struct tap_queue *q, struct msghdr *m,
626                             struct iov_iter *from, int noblock)
627 {
628         int good_linear = SKB_MAX_HEAD(TAP_RESERVE);
629         struct sk_buff *skb;
630         struct tap_dev *tap;
631         unsigned long total_len = iov_iter_count(from);
632         unsigned long len = total_len;
633         int err;
634         struct virtio_net_hdr vnet_hdr = { 0 };
635         int vnet_hdr_len = 0;
636         int copylen = 0;
637         int depth;
638         bool zerocopy = false;
639         size_t linear;
640
641         if (q->flags & IFF_VNET_HDR) {
642                 vnet_hdr_len = READ_ONCE(q->vnet_hdr_sz);
643
644                 err = -EINVAL;
645                 if (len < vnet_hdr_len)
646                         goto err;
647                 len -= vnet_hdr_len;
648
649                 err = -EFAULT;
650                 if (!copy_from_iter_full(&vnet_hdr, sizeof(vnet_hdr), from))
651                         goto err;
652                 iov_iter_advance(from, vnet_hdr_len - sizeof(vnet_hdr));
653                 if ((vnet_hdr.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) &&
654                      tap16_to_cpu(q, vnet_hdr.csum_start) +
655                      tap16_to_cpu(q, vnet_hdr.csum_offset) + 2 >
656                              tap16_to_cpu(q, vnet_hdr.hdr_len))
657                         vnet_hdr.hdr_len = cpu_to_tap16(q,
658                                  tap16_to_cpu(q, vnet_hdr.csum_start) +
659                                  tap16_to_cpu(q, vnet_hdr.csum_offset) + 2);
660                 err = -EINVAL;
661                 if (tap16_to_cpu(q, vnet_hdr.hdr_len) > len)
662                         goto err;
663         }
664
665         err = -EINVAL;
666         if (unlikely(len < ETH_HLEN))
667                 goto err;
668
669         if (m && m->msg_control && sock_flag(&q->sk, SOCK_ZEROCOPY)) {
670                 struct iov_iter i;
671
672                 copylen = vnet_hdr.hdr_len ?
673                         tap16_to_cpu(q, vnet_hdr.hdr_len) : GOODCOPY_LEN;
674                 if (copylen > good_linear)
675                         copylen = good_linear;
676                 else if (copylen < ETH_HLEN)
677                         copylen = ETH_HLEN;
678                 linear = copylen;
679                 i = *from;
680                 iov_iter_advance(&i, copylen);
681                 if (iov_iter_npages(&i, INT_MAX) <= MAX_SKB_FRAGS)
682                         zerocopy = true;
683         }
684
685         if (!zerocopy) {
686                 copylen = len;
687                 linear = tap16_to_cpu(q, vnet_hdr.hdr_len);
688                 if (linear > good_linear)
689                         linear = good_linear;
690                 else if (linear < ETH_HLEN)
691                         linear = ETH_HLEN;
692         }
693
694         skb = tap_alloc_skb(&q->sk, TAP_RESERVE, copylen,
695                             linear, noblock, &err);
696         if (!skb)
697                 goto err;
698
699         if (zerocopy)
700                 err = zerocopy_sg_from_iter(skb, from);
701         else
702                 err = skb_copy_datagram_from_iter(skb, 0, from, len);
703
704         if (err)
705                 goto err_kfree;
706
707         skb_set_network_header(skb, ETH_HLEN);
708         skb_reset_mac_header(skb);
709         skb->protocol = eth_hdr(skb)->h_proto;
710
711         if (vnet_hdr_len) {
712                 err = virtio_net_hdr_to_skb(skb, &vnet_hdr,
713                                             tap_is_little_endian(q));
714                 if (err)
715                         goto err_kfree;
716         }
717
718         skb_probe_transport_header(skb, ETH_HLEN);
719
720         /* Move network header to the right position for VLAN tagged packets */
721         if ((skb->protocol == htons(ETH_P_8021Q) ||
722              skb->protocol == htons(ETH_P_8021AD)) &&
723             __vlan_get_protocol(skb, skb->protocol, &depth) != 0)
724                 skb_set_network_header(skb, depth);
725
726         rcu_read_lock();
727         tap = rcu_dereference(q->tap);
728         /* copy skb_ubuf_info for callback when skb has no error */
729         if (zerocopy) {
730                 skb_shinfo(skb)->destructor_arg = m->msg_control;
731                 skb_shinfo(skb)->tx_flags |= SKBTX_DEV_ZEROCOPY;
732                 skb_shinfo(skb)->tx_flags |= SKBTX_SHARED_FRAG;
733         } else if (m && m->msg_control) {
734                 struct ubuf_info *uarg = m->msg_control;
735                 uarg->callback(uarg, false);
736         }
737
738         if (tap) {
739                 skb->dev = tap->dev;
740                 dev_queue_xmit(skb);
741         } else {
742                 kfree_skb(skb);
743         }
744         rcu_read_unlock();
745
746         return total_len;
747
748 err_kfree:
749         kfree_skb(skb);
750
751 err:
752         rcu_read_lock();
753         tap = rcu_dereference(q->tap);
754         if (tap && tap->count_tx_dropped)
755                 tap->count_tx_dropped(tap);
756         rcu_read_unlock();
757
758         return err;
759 }
760
761 static ssize_t tap_write_iter(struct kiocb *iocb, struct iov_iter *from)
762 {
763         struct file *file = iocb->ki_filp;
764         struct tap_queue *q = file->private_data;
765
766         return tap_get_user(q, NULL, from, file->f_flags & O_NONBLOCK);
767 }
768
769 /* Put packet to the user space buffer */
770 static ssize_t tap_put_user(struct tap_queue *q,
771                             const struct sk_buff *skb,
772                             struct iov_iter *iter)
773 {
774         int ret;
775         int vnet_hdr_len = 0;
776         int vlan_offset = 0;
777         int total;
778
779         if (q->flags & IFF_VNET_HDR) {
780                 struct virtio_net_hdr vnet_hdr;
781                 vnet_hdr_len = READ_ONCE(q->vnet_hdr_sz);
782                 if (iov_iter_count(iter) < vnet_hdr_len)
783                         return -EINVAL;
784
785                 if (virtio_net_hdr_from_skb(skb, &vnet_hdr,
786                                             tap_is_little_endian(q), true))
787                         BUG();
788
789                 if (copy_to_iter(&vnet_hdr, sizeof(vnet_hdr), iter) !=
790                     sizeof(vnet_hdr))
791                         return -EFAULT;
792
793                 iov_iter_advance(iter, vnet_hdr_len - sizeof(vnet_hdr));
794         }
795         total = vnet_hdr_len;
796         total += skb->len;
797
798         if (skb_vlan_tag_present(skb)) {
799                 struct {
800                         __be16 h_vlan_proto;
801                         __be16 h_vlan_TCI;
802                 } veth;
803                 veth.h_vlan_proto = skb->vlan_proto;
804                 veth.h_vlan_TCI = htons(skb_vlan_tag_get(skb));
805
806                 vlan_offset = offsetof(struct vlan_ethhdr, h_vlan_proto);
807                 total += VLAN_HLEN;
808
809                 ret = skb_copy_datagram_iter(skb, 0, iter, vlan_offset);
810                 if (ret || !iov_iter_count(iter))
811                         goto done;
812
813                 ret = copy_to_iter(&veth, sizeof(veth), iter);
814                 if (ret != sizeof(veth) || !iov_iter_count(iter))
815                         goto done;
816         }
817
818         ret = skb_copy_datagram_iter(skb, vlan_offset, iter,
819                                      skb->len - vlan_offset);
820
821 done:
822         return ret ? ret : total;
823 }
824
825 static ssize_t tap_do_read(struct tap_queue *q,
826                            struct iov_iter *to,
827                            int noblock, struct sk_buff *skb)
828 {
829         DEFINE_WAIT(wait);
830         ssize_t ret = 0;
831
832         if (!iov_iter_count(to))
833                 return 0;
834
835         if (skb)
836                 goto put;
837
838         while (1) {
839                 if (!noblock)
840                         prepare_to_wait(sk_sleep(&q->sk), &wait,
841                                         TASK_INTERRUPTIBLE);
842
843                 /* Read frames from the queue */
844                 skb = skb_array_consume(&q->skb_array);
845                 if (skb)
846                         break;
847                 if (noblock) {
848                         ret = -EAGAIN;
849                         break;
850                 }
851                 if (signal_pending(current)) {
852                         ret = -ERESTARTSYS;
853                         break;
854                 }
855                 /* Nothing to read, let's sleep */
856                 schedule();
857         }
858         if (!noblock)
859                 finish_wait(sk_sleep(&q->sk), &wait);
860
861 put:
862         if (skb) {
863                 ret = tap_put_user(q, skb, to);
864                 if (unlikely(ret < 0))
865                         kfree_skb(skb);
866                 else
867                         consume_skb(skb);
868         }
869         return ret;
870 }
871
872 static ssize_t tap_read_iter(struct kiocb *iocb, struct iov_iter *to)
873 {
874         struct file *file = iocb->ki_filp;
875         struct tap_queue *q = file->private_data;
876         ssize_t len = iov_iter_count(to), ret;
877
878         ret = tap_do_read(q, to, file->f_flags & O_NONBLOCK, NULL);
879         ret = min_t(ssize_t, ret, len);
880         if (ret > 0)
881                 iocb->ki_pos = ret;
882         return ret;
883 }
884
885 static struct tap_dev *tap_get_tap_dev(struct tap_queue *q)
886 {
887         struct tap_dev *tap;
888
889         ASSERT_RTNL();
890         tap = rtnl_dereference(q->tap);
891         if (tap)
892                 dev_hold(tap->dev);
893
894         return tap;
895 }
896
897 static void tap_put_tap_dev(struct tap_dev *tap)
898 {
899         dev_put(tap->dev);
900 }
901
902 static int tap_ioctl_set_queue(struct file *file, unsigned int flags)
903 {
904         struct tap_queue *q = file->private_data;
905         struct tap_dev *tap;
906         int ret;
907
908         tap = tap_get_tap_dev(q);
909         if (!tap)
910                 return -EINVAL;
911
912         if (flags & IFF_ATTACH_QUEUE)
913                 ret = tap_enable_queue(tap, file, q);
914         else if (flags & IFF_DETACH_QUEUE)
915                 ret = tap_disable_queue(q);
916         else
917                 ret = -EINVAL;
918
919         tap_put_tap_dev(tap);
920         return ret;
921 }
922
923 static int set_offload(struct tap_queue *q, unsigned long arg)
924 {
925         struct tap_dev *tap;
926         netdev_features_t features;
927         netdev_features_t feature_mask = 0;
928
929         tap = rtnl_dereference(q->tap);
930         if (!tap)
931                 return -ENOLINK;
932
933         features = tap->dev->features;
934
935         if (arg & TUN_F_CSUM) {
936                 feature_mask = NETIF_F_HW_CSUM;
937
938                 if (arg & (TUN_F_TSO4 | TUN_F_TSO6)) {
939                         if (arg & TUN_F_TSO_ECN)
940                                 feature_mask |= NETIF_F_TSO_ECN;
941                         if (arg & TUN_F_TSO4)
942                                 feature_mask |= NETIF_F_TSO;
943                         if (arg & TUN_F_TSO6)
944                                 feature_mask |= NETIF_F_TSO6;
945                 }
946         }
947
948         /* tun/tap driver inverts the usage for TSO offloads, where
949          * setting the TSO bit means that the userspace wants to
950          * accept TSO frames and turning it off means that user space
951          * does not support TSO.
952          * For tap, we have to invert it to mean the same thing.
953          * When user space turns off TSO, we turn off GSO/LRO so that
954          * user-space will not receive TSO frames.
955          */
956         if (feature_mask & (NETIF_F_TSO | NETIF_F_TSO6))
957                 features |= RX_OFFLOADS;
958         else
959                 features &= ~RX_OFFLOADS;
960
961         /* tap_features are the same as features on tun/tap and
962          * reflect user expectations.
963          */
964         tap->tap_features = feature_mask;
965         if (tap->update_features)
966                 tap->update_features(tap, features);
967
968         return 0;
969 }
970
971 /*
972  * provide compatibility with generic tun/tap interface
973  */
974 static long tap_ioctl(struct file *file, unsigned int cmd,
975                       unsigned long arg)
976 {
977         struct tap_queue *q = file->private_data;
978         struct tap_dev *tap;
979         void __user *argp = (void __user *)arg;
980         struct ifreq __user *ifr = argp;
981         unsigned int __user *up = argp;
982         unsigned short u;
983         int __user *sp = argp;
984         struct sockaddr sa;
985         int s;
986         int ret;
987
988         switch (cmd) {
989         case TUNSETIFF:
990                 /* ignore the name, just look at flags */
991                 if (get_user(u, &ifr->ifr_flags))
992                         return -EFAULT;
993
994                 ret = 0;
995                 if ((u & ~TAP_IFFEATURES) != (IFF_NO_PI | IFF_TAP))
996                         ret = -EINVAL;
997                 else
998                         q->flags = (q->flags & ~TAP_IFFEATURES) | u;
999
1000                 return ret;
1001
1002         case TUNGETIFF:
1003                 rtnl_lock();
1004                 tap = tap_get_tap_dev(q);
1005                 if (!tap) {
1006                         rtnl_unlock();
1007                         return -ENOLINK;
1008                 }
1009
1010                 ret = 0;
1011                 u = q->flags;
1012                 if (copy_to_user(&ifr->ifr_name, tap->dev->name, IFNAMSIZ) ||
1013                     put_user(u, &ifr->ifr_flags))
1014                         ret = -EFAULT;
1015                 tap_put_tap_dev(tap);
1016                 rtnl_unlock();
1017                 return ret;
1018
1019         case TUNSETQUEUE:
1020                 if (get_user(u, &ifr->ifr_flags))
1021                         return -EFAULT;
1022                 rtnl_lock();
1023                 ret = tap_ioctl_set_queue(file, u);
1024                 rtnl_unlock();
1025                 return ret;
1026
1027         case TUNGETFEATURES:
1028                 if (put_user(IFF_TAP | IFF_NO_PI | TAP_IFFEATURES, up))
1029                         return -EFAULT;
1030                 return 0;
1031
1032         case TUNSETSNDBUF:
1033                 if (get_user(s, sp))
1034                         return -EFAULT;
1035
1036                 q->sk.sk_sndbuf = s;
1037                 return 0;
1038
1039         case TUNGETVNETHDRSZ:
1040                 s = q->vnet_hdr_sz;
1041                 if (put_user(s, sp))
1042                         return -EFAULT;
1043                 return 0;
1044
1045         case TUNSETVNETHDRSZ:
1046                 if (get_user(s, sp))
1047                         return -EFAULT;
1048                 if (s < (int)sizeof(struct virtio_net_hdr))
1049                         return -EINVAL;
1050
1051                 q->vnet_hdr_sz = s;
1052                 return 0;
1053
1054         case TUNGETVNETLE:
1055                 s = !!(q->flags & TAP_VNET_LE);
1056                 if (put_user(s, sp))
1057                         return -EFAULT;
1058                 return 0;
1059
1060         case TUNSETVNETLE:
1061                 if (get_user(s, sp))
1062                         return -EFAULT;
1063                 if (s)
1064                         q->flags |= TAP_VNET_LE;
1065                 else
1066                         q->flags &= ~TAP_VNET_LE;
1067                 return 0;
1068
1069         case TUNGETVNETBE:
1070                 return tap_get_vnet_be(q, sp);
1071
1072         case TUNSETVNETBE:
1073                 return tap_set_vnet_be(q, sp);
1074
1075         case TUNSETOFFLOAD:
1076                 /* let the user check for future flags */
1077                 if (arg & ~(TUN_F_CSUM | TUN_F_TSO4 | TUN_F_TSO6 |
1078                             TUN_F_TSO_ECN))
1079                         return -EINVAL;
1080
1081                 rtnl_lock();
1082                 ret = set_offload(q, arg);
1083                 rtnl_unlock();
1084                 return ret;
1085
1086         case SIOCGIFHWADDR:
1087                 rtnl_lock();
1088                 tap = tap_get_tap_dev(q);
1089                 if (!tap) {
1090                         rtnl_unlock();
1091                         return -ENOLINK;
1092                 }
1093                 ret = 0;
1094                 u = tap->dev->type;
1095                 if (copy_to_user(&ifr->ifr_name, tap->dev->name, IFNAMSIZ) ||
1096                     copy_to_user(&ifr->ifr_hwaddr.sa_data, tap->dev->dev_addr, ETH_ALEN) ||
1097                     put_user(u, &ifr->ifr_hwaddr.sa_family))
1098                         ret = -EFAULT;
1099                 tap_put_tap_dev(tap);
1100                 rtnl_unlock();
1101                 return ret;
1102
1103         case SIOCSIFHWADDR:
1104                 if (copy_from_user(&sa, &ifr->ifr_hwaddr, sizeof(sa)))
1105                         return -EFAULT;
1106                 rtnl_lock();
1107                 tap = tap_get_tap_dev(q);
1108                 if (!tap) {
1109                         rtnl_unlock();
1110                         return -ENOLINK;
1111                 }
1112                 ret = dev_set_mac_address(tap->dev, &sa);
1113                 tap_put_tap_dev(tap);
1114                 rtnl_unlock();
1115                 return ret;
1116
1117         default:
1118                 return -EINVAL;
1119         }
1120 }
1121
1122 #ifdef CONFIG_COMPAT
1123 static long tap_compat_ioctl(struct file *file, unsigned int cmd,
1124                              unsigned long arg)
1125 {
1126         return tap_ioctl(file, cmd, (unsigned long)compat_ptr(arg));
1127 }
1128 #endif
1129
1130 static const struct file_operations tap_fops = {
1131         .owner          = THIS_MODULE,
1132         .open           = tap_open,
1133         .release        = tap_release,
1134         .read_iter      = tap_read_iter,
1135         .write_iter     = tap_write_iter,
1136         .poll           = tap_poll,
1137         .llseek         = no_llseek,
1138         .unlocked_ioctl = tap_ioctl,
1139 #ifdef CONFIG_COMPAT
1140         .compat_ioctl   = tap_compat_ioctl,
1141 #endif
1142 };
1143
1144 static int tap_sendmsg(struct socket *sock, struct msghdr *m,
1145                        size_t total_len)
1146 {
1147         struct tap_queue *q = container_of(sock, struct tap_queue, sock);
1148         return tap_get_user(q, m, &m->msg_iter, m->msg_flags & MSG_DONTWAIT);
1149 }
1150
1151 static int tap_recvmsg(struct socket *sock, struct msghdr *m,
1152                        size_t total_len, int flags)
1153 {
1154         struct tap_queue *q = container_of(sock, struct tap_queue, sock);
1155         int ret;
1156         if (flags & ~(MSG_DONTWAIT|MSG_TRUNC))
1157                 return -EINVAL;
1158         ret = tap_do_read(q, &m->msg_iter, flags & MSG_DONTWAIT,
1159                           m->msg_control);
1160         if (ret > total_len) {
1161                 m->msg_flags |= MSG_TRUNC;
1162                 ret = flags & MSG_TRUNC ? ret : total_len;
1163         }
1164         return ret;
1165 }
1166
1167 static int tap_peek_len(struct socket *sock)
1168 {
1169         struct tap_queue *q = container_of(sock, struct tap_queue,
1170                                                sock);
1171         return skb_array_peek_len(&q->skb_array);
1172 }
1173
1174 /* Ops structure to mimic raw sockets with tun */
1175 static const struct proto_ops tap_socket_ops = {
1176         .sendmsg = tap_sendmsg,
1177         .recvmsg = tap_recvmsg,
1178         .peek_len = tap_peek_len,
1179 };
1180
1181 /* Get an underlying socket object from tun file.  Returns error unless file is
1182  * attached to a device.  The returned object works like a packet socket, it
1183  * can be used for sock_sendmsg/sock_recvmsg.  The caller is responsible for
1184  * holding a reference to the file for as long as the socket is in use. */
1185 struct socket *tap_get_socket(struct file *file)
1186 {
1187         struct tap_queue *q;
1188         if (file->f_op != &tap_fops)
1189                 return ERR_PTR(-EINVAL);
1190         q = file->private_data;
1191         if (!q)
1192                 return ERR_PTR(-EBADFD);
1193         return &q->sock;
1194 }
1195 EXPORT_SYMBOL_GPL(tap_get_socket);
1196
1197 struct skb_array *tap_get_skb_array(struct file *file)
1198 {
1199         struct tap_queue *q;
1200
1201         if (file->f_op != &tap_fops)
1202                 return ERR_PTR(-EINVAL);
1203         q = file->private_data;
1204         if (!q)
1205                 return ERR_PTR(-EBADFD);
1206         return &q->skb_array;
1207 }
1208 EXPORT_SYMBOL_GPL(tap_get_skb_array);
1209
1210 int tap_queue_resize(struct tap_dev *tap)
1211 {
1212         struct net_device *dev = tap->dev;
1213         struct tap_queue *q;
1214         struct skb_array **arrays;
1215         int n = tap->numqueues;
1216         int ret, i = 0;
1217
1218         arrays = kmalloc_array(n, sizeof(*arrays), GFP_KERNEL);
1219         if (!arrays)
1220                 return -ENOMEM;
1221
1222         list_for_each_entry(q, &tap->queue_list, next)
1223                 arrays[i++] = &q->skb_array;
1224
1225         ret = skb_array_resize_multiple(arrays, n,
1226                                         dev->tx_queue_len, GFP_KERNEL);
1227
1228         kfree(arrays);
1229         return ret;
1230 }
1231 EXPORT_SYMBOL_GPL(tap_queue_resize);
1232
1233 static int tap_list_add(dev_t major, const char *device_name)
1234 {
1235         struct major_info *tap_major;
1236
1237         tap_major = kzalloc(sizeof(*tap_major), GFP_ATOMIC);
1238         if (!tap_major)
1239                 return -ENOMEM;
1240
1241         tap_major->major = MAJOR(major);
1242
1243         idr_init(&tap_major->minor_idr);
1244         spin_lock_init(&tap_major->minor_lock);
1245
1246         tap_major->device_name = device_name;
1247
1248         list_add_tail_rcu(&tap_major->next, &major_list);
1249         return 0;
1250 }
1251
1252 int tap_create_cdev(struct cdev *tap_cdev,
1253                     dev_t *tap_major, const char *device_name)
1254 {
1255         int err;
1256
1257         err = alloc_chrdev_region(tap_major, 0, TAP_NUM_DEVS, device_name);
1258         if (err)
1259                 goto out1;
1260
1261         cdev_init(tap_cdev, &tap_fops);
1262         err = cdev_add(tap_cdev, *tap_major, TAP_NUM_DEVS);
1263         if (err)
1264                 goto out2;
1265
1266         err =  tap_list_add(*tap_major, device_name);
1267         if (err)
1268                 goto out3;
1269
1270         return 0;
1271
1272 out3:
1273         cdev_del(tap_cdev);
1274 out2:
1275         unregister_chrdev_region(*tap_major, TAP_NUM_DEVS);
1276 out1:
1277         return err;
1278 }
1279 EXPORT_SYMBOL_GPL(tap_create_cdev);
1280
1281 void tap_destroy_cdev(dev_t major, struct cdev *tap_cdev)
1282 {
1283         struct major_info *tap_major, *tmp;
1284
1285         cdev_del(tap_cdev);
1286         unregister_chrdev_region(major, TAP_NUM_DEVS);
1287         list_for_each_entry_safe(tap_major, tmp, &major_list, next) {
1288                 if (tap_major->major == MAJOR(major)) {
1289                         idr_destroy(&tap_major->minor_idr);
1290                         list_del_rcu(&tap_major->next);
1291                         kfree_rcu(tap_major, rcu);
1292                 }
1293         }
1294 }
1295 EXPORT_SYMBOL_GPL(tap_destroy_cdev);
1296
1297 MODULE_AUTHOR("Arnd Bergmann <arnd@arndb.de>");
1298 MODULE_AUTHOR("Sainath Grandhi <sainath.grandhi@intel.com>");
1299 MODULE_LICENSE("GPL");