Loading...
1// SPDX-License-Identifier: GPL-2.0-only
2#include <linux/etherdevice.h>
3#include <linux/if_tap.h>
4#include <linux/if_vlan.h>
5#include <linux/interrupt.h>
6#include <linux/nsproxy.h>
7#include <linux/compat.h>
8#include <linux/if_tun.h>
9#include <linux/module.h>
10#include <linux/skbuff.h>
11#include <linux/cache.h>
12#include <linux/sched/signal.h>
13#include <linux/types.h>
14#include <linux/slab.h>
15#include <linux/wait.h>
16#include <linux/cdev.h>
17#include <linux/idr.h>
18#include <linux/fs.h>
19#include <linux/uio.h>
20
21#include <net/gso.h>
22#include <net/net_namespace.h>
23#include <net/rtnetlink.h>
24#include <net/sock.h>
25#include <net/xdp.h>
26#include <linux/virtio_net.h>
27#include <linux/skb_array.h>
28
29#define TAP_IFFEATURES (IFF_VNET_HDR | IFF_MULTI_QUEUE)
30
31#define TAP_VNET_LE 0x80000000
32#define TAP_VNET_BE 0x40000000
33
34#ifdef CONFIG_TUN_VNET_CROSS_LE
35static inline bool tap_legacy_is_little_endian(struct tap_queue *q)
36{
37 return q->flags & TAP_VNET_BE ? false :
38 virtio_legacy_is_little_endian();
39}
40
41static long tap_get_vnet_be(struct tap_queue *q, int __user *sp)
42{
43 int s = !!(q->flags & TAP_VNET_BE);
44
45 if (put_user(s, sp))
46 return -EFAULT;
47
48 return 0;
49}
50
51static long tap_set_vnet_be(struct tap_queue *q, int __user *sp)
52{
53 int s;
54
55 if (get_user(s, sp))
56 return -EFAULT;
57
58 if (s)
59 q->flags |= TAP_VNET_BE;
60 else
61 q->flags &= ~TAP_VNET_BE;
62
63 return 0;
64}
65#else
66static inline bool tap_legacy_is_little_endian(struct tap_queue *q)
67{
68 return virtio_legacy_is_little_endian();
69}
70
71static long tap_get_vnet_be(struct tap_queue *q, int __user *argp)
72{
73 return -EINVAL;
74}
75
76static long tap_set_vnet_be(struct tap_queue *q, int __user *argp)
77{
78 return -EINVAL;
79}
80#endif /* CONFIG_TUN_VNET_CROSS_LE */
81
82static inline bool tap_is_little_endian(struct tap_queue *q)
83{
84 return q->flags & TAP_VNET_LE ||
85 tap_legacy_is_little_endian(q);
86}
87
88static inline u16 tap16_to_cpu(struct tap_queue *q, __virtio16 val)
89{
90 return __virtio16_to_cpu(tap_is_little_endian(q), val);
91}
92
93static inline __virtio16 cpu_to_tap16(struct tap_queue *q, u16 val)
94{
95 return __cpu_to_virtio16(tap_is_little_endian(q), val);
96}
97
98static struct proto tap_proto = {
99 .name = "tap",
100 .owner = THIS_MODULE,
101 .obj_size = sizeof(struct tap_queue),
102};
103
104#define TAP_NUM_DEVS (1U << MINORBITS)
105
106static LIST_HEAD(major_list);
107
108struct major_info {
109 struct rcu_head rcu;
110 dev_t major;
111 struct idr minor_idr;
112 spinlock_t minor_lock;
113 const char *device_name;
114 struct list_head next;
115};
116
117#define GOODCOPY_LEN 128
118
119static const struct proto_ops tap_socket_ops;
120
121#define RX_OFFLOADS (NETIF_F_GRO | NETIF_F_LRO)
122#define TAP_FEATURES (NETIF_F_GSO | NETIF_F_SG | NETIF_F_FRAGLIST)
123
124static struct tap_dev *tap_dev_get_rcu(const struct net_device *dev)
125{
126 return rcu_dereference(dev->rx_handler_data);
127}
128
129/*
130 * RCU usage:
131 * The tap_queue and the macvlan_dev are loosely coupled, the
132 * pointers from one to the other can only be read while rcu_read_lock
133 * or rtnl is held.
134 *
135 * Both the file and the macvlan_dev hold a reference on the tap_queue
136 * through sock_hold(&q->sk). When the macvlan_dev goes away first,
137 * q->vlan becomes inaccessible. When the files gets closed,
138 * tap_get_queue() fails.
139 *
140 * There may still be references to the struct sock inside of the
141 * queue from outbound SKBs, but these never reference back to the
142 * file or the dev. The data structure is freed through __sk_free
143 * when both our references and any pending SKBs are gone.
144 */
145
146static int tap_enable_queue(struct tap_dev *tap, struct file *file,
147 struct tap_queue *q)
148{
149 int err = -EINVAL;
150
151 ASSERT_RTNL();
152
153 if (q->enabled)
154 goto out;
155
156 err = 0;
157 rcu_assign_pointer(tap->taps[tap->numvtaps], q);
158 q->queue_index = tap->numvtaps;
159 q->enabled = true;
160
161 tap->numvtaps++;
162out:
163 return err;
164}
165
166/* Requires RTNL */
167static int tap_set_queue(struct tap_dev *tap, struct file *file,
168 struct tap_queue *q)
169{
170 if (tap->numqueues == MAX_TAP_QUEUES)
171 return -EBUSY;
172
173 rcu_assign_pointer(q->tap, tap);
174 rcu_assign_pointer(tap->taps[tap->numvtaps], q);
175 sock_hold(&q->sk);
176
177 q->file = file;
178 q->queue_index = tap->numvtaps;
179 q->enabled = true;
180 file->private_data = q;
181 list_add_tail(&q->next, &tap->queue_list);
182
183 tap->numvtaps++;
184 tap->numqueues++;
185
186 return 0;
187}
188
189static int tap_disable_queue(struct tap_queue *q)
190{
191 struct tap_dev *tap;
192 struct tap_queue *nq;
193
194 ASSERT_RTNL();
195 if (!q->enabled)
196 return -EINVAL;
197
198 tap = rtnl_dereference(q->tap);
199
200 if (tap) {
201 int index = q->queue_index;
202 BUG_ON(index >= tap->numvtaps);
203 nq = rtnl_dereference(tap->taps[tap->numvtaps - 1]);
204 nq->queue_index = index;
205
206 rcu_assign_pointer(tap->taps[index], nq);
207 RCU_INIT_POINTER(tap->taps[tap->numvtaps - 1], NULL);
208 q->enabled = false;
209
210 tap->numvtaps--;
211 }
212
213 return 0;
214}
215
216/*
217 * The file owning the queue got closed, give up both
218 * the reference that the files holds as well as the
219 * one from the macvlan_dev if that still exists.
220 *
221 * Using the spinlock makes sure that we don't get
222 * to the queue again after destroying it.
223 */
224static void tap_put_queue(struct tap_queue *q)
225{
226 struct tap_dev *tap;
227
228 rtnl_lock();
229 tap = rtnl_dereference(q->tap);
230
231 if (tap) {
232 if (q->enabled)
233 BUG_ON(tap_disable_queue(q));
234
235 tap->numqueues--;
236 RCU_INIT_POINTER(q->tap, NULL);
237 sock_put(&q->sk);
238 list_del_init(&q->next);
239 }
240
241 rtnl_unlock();
242
243 synchronize_rcu();
244 sock_put(&q->sk);
245}
246
247/*
248 * Select a queue based on the rxq of the device on which this packet
249 * arrived. If the incoming device is not mq, calculate a flow hash
250 * to select a queue. If all fails, find the first available queue.
251 * Cache vlan->numvtaps since it can become zero during the execution
252 * of this function.
253 */
254static struct tap_queue *tap_get_queue(struct tap_dev *tap,
255 struct sk_buff *skb)
256{
257 struct tap_queue *queue = NULL;
258 /* Access to taps array is protected by rcu, but access to numvtaps
259 * isn't. Below we use it to lookup a queue, but treat it as a hint
260 * and validate that the result isn't NULL - in case we are
261 * racing against queue removal.
262 */
263 int numvtaps = READ_ONCE(tap->numvtaps);
264 __u32 rxq;
265
266 if (!numvtaps)
267 goto out;
268
269 if (numvtaps == 1)
270 goto single;
271
272 /* Check if we can use flow to select a queue */
273 rxq = skb_get_hash(skb);
274 if (rxq) {
275 queue = rcu_dereference(tap->taps[rxq % numvtaps]);
276 goto out;
277 }
278
279 if (likely(skb_rx_queue_recorded(skb))) {
280 rxq = skb_get_rx_queue(skb);
281
282 while (unlikely(rxq >= numvtaps))
283 rxq -= numvtaps;
284
285 queue = rcu_dereference(tap->taps[rxq]);
286 goto out;
287 }
288
289single:
290 queue = rcu_dereference(tap->taps[0]);
291out:
292 return queue;
293}
294
295/*
296 * The net_device is going away, give up the reference
297 * that it holds on all queues and safely set the pointer
298 * from the queues to NULL.
299 */
300void tap_del_queues(struct tap_dev *tap)
301{
302 struct tap_queue *q, *tmp;
303
304 ASSERT_RTNL();
305 list_for_each_entry_safe(q, tmp, &tap->queue_list, next) {
306 list_del_init(&q->next);
307 RCU_INIT_POINTER(q->tap, NULL);
308 if (q->enabled)
309 tap->numvtaps--;
310 tap->numqueues--;
311 sock_put(&q->sk);
312 }
313 BUG_ON(tap->numvtaps);
314 BUG_ON(tap->numqueues);
315 /* guarantee that any future tap_set_queue will fail */
316 tap->numvtaps = MAX_TAP_QUEUES;
317}
318EXPORT_SYMBOL_GPL(tap_del_queues);
319
320rx_handler_result_t tap_handle_frame(struct sk_buff **pskb)
321{
322 struct sk_buff *skb = *pskb;
323 struct net_device *dev = skb->dev;
324 struct tap_dev *tap;
325 struct tap_queue *q;
326 netdev_features_t features = TAP_FEATURES;
327 enum skb_drop_reason drop_reason;
328
329 tap = tap_dev_get_rcu(dev);
330 if (!tap)
331 return RX_HANDLER_PASS;
332
333 q = tap_get_queue(tap, skb);
334 if (!q)
335 return RX_HANDLER_PASS;
336
337 skb_push(skb, ETH_HLEN);
338
339 /* Apply the forward feature mask so that we perform segmentation
340 * according to users wishes. This only works if VNET_HDR is
341 * enabled.
342 */
343 if (q->flags & IFF_VNET_HDR)
344 features |= tap->tap_features;
345 if (netif_needs_gso(skb, features)) {
346 struct sk_buff *segs = __skb_gso_segment(skb, features, false);
347 struct sk_buff *next;
348
349 if (IS_ERR(segs)) {
350 drop_reason = SKB_DROP_REASON_SKB_GSO_SEG;
351 goto drop;
352 }
353
354 if (!segs) {
355 if (ptr_ring_produce(&q->ring, skb)) {
356 drop_reason = SKB_DROP_REASON_FULL_RING;
357 goto drop;
358 }
359 goto wake_up;
360 }
361
362 consume_skb(skb);
363 skb_list_walk_safe(segs, skb, next) {
364 skb_mark_not_on_list(skb);
365 if (ptr_ring_produce(&q->ring, skb)) {
366 drop_reason = SKB_DROP_REASON_FULL_RING;
367 kfree_skb_reason(skb, drop_reason);
368 kfree_skb_list_reason(next, drop_reason);
369 break;
370 }
371 }
372 } else {
373 /* If we receive a partial checksum and the tap side
374 * doesn't support checksum offload, compute the checksum.
375 * Note: it doesn't matter which checksum feature to
376 * check, we either support them all or none.
377 */
378 if (skb->ip_summed == CHECKSUM_PARTIAL &&
379 !(features & NETIF_F_CSUM_MASK) &&
380 skb_checksum_help(skb)) {
381 drop_reason = SKB_DROP_REASON_SKB_CSUM;
382 goto drop;
383 }
384 if (ptr_ring_produce(&q->ring, skb)) {
385 drop_reason = SKB_DROP_REASON_FULL_RING;
386 goto drop;
387 }
388 }
389
390wake_up:
391 wake_up_interruptible_poll(sk_sleep(&q->sk), EPOLLIN | EPOLLRDNORM | EPOLLRDBAND);
392 return RX_HANDLER_CONSUMED;
393
394drop:
395 /* Count errors/drops only here, thus don't care about args. */
396 if (tap->count_rx_dropped)
397 tap->count_rx_dropped(tap);
398 kfree_skb_reason(skb, drop_reason);
399 return RX_HANDLER_CONSUMED;
400}
401EXPORT_SYMBOL_GPL(tap_handle_frame);
402
403static struct major_info *tap_get_major(int major)
404{
405 struct major_info *tap_major;
406
407 list_for_each_entry_rcu(tap_major, &major_list, next) {
408 if (tap_major->major == major)
409 return tap_major;
410 }
411
412 return NULL;
413}
414
415int tap_get_minor(dev_t major, struct tap_dev *tap)
416{
417 int retval = -ENOMEM;
418 struct major_info *tap_major;
419
420 rcu_read_lock();
421 tap_major = tap_get_major(MAJOR(major));
422 if (!tap_major) {
423 retval = -EINVAL;
424 goto unlock;
425 }
426
427 spin_lock(&tap_major->minor_lock);
428 retval = idr_alloc(&tap_major->minor_idr, tap, 1, TAP_NUM_DEVS, GFP_ATOMIC);
429 if (retval >= 0) {
430 tap->minor = retval;
431 } else if (retval == -ENOSPC) {
432 netdev_err(tap->dev, "Too many tap devices\n");
433 retval = -EINVAL;
434 }
435 spin_unlock(&tap_major->minor_lock);
436
437unlock:
438 rcu_read_unlock();
439 return retval < 0 ? retval : 0;
440}
441EXPORT_SYMBOL_GPL(tap_get_minor);
442
443void tap_free_minor(dev_t major, struct tap_dev *tap)
444{
445 struct major_info *tap_major;
446
447 rcu_read_lock();
448 tap_major = tap_get_major(MAJOR(major));
449 if (!tap_major) {
450 goto unlock;
451 }
452
453 spin_lock(&tap_major->minor_lock);
454 if (tap->minor) {
455 idr_remove(&tap_major->minor_idr, tap->minor);
456 tap->minor = 0;
457 }
458 spin_unlock(&tap_major->minor_lock);
459
460unlock:
461 rcu_read_unlock();
462}
463EXPORT_SYMBOL_GPL(tap_free_minor);
464
465static struct tap_dev *dev_get_by_tap_file(int major, int minor)
466{
467 struct net_device *dev = NULL;
468 struct tap_dev *tap;
469 struct major_info *tap_major;
470
471 rcu_read_lock();
472 tap_major = tap_get_major(major);
473 if (!tap_major) {
474 tap = NULL;
475 goto unlock;
476 }
477
478 spin_lock(&tap_major->minor_lock);
479 tap = idr_find(&tap_major->minor_idr, minor);
480 if (tap) {
481 dev = tap->dev;
482 dev_hold(dev);
483 }
484 spin_unlock(&tap_major->minor_lock);
485
486unlock:
487 rcu_read_unlock();
488 return tap;
489}
490
491static void tap_sock_write_space(struct sock *sk)
492{
493 wait_queue_head_t *wqueue;
494
495 if (!sock_writeable(sk) ||
496 !test_and_clear_bit(SOCKWQ_ASYNC_NOSPACE, &sk->sk_socket->flags))
497 return;
498
499 wqueue = sk_sleep(sk);
500 if (wqueue && waitqueue_active(wqueue))
501 wake_up_interruptible_poll(wqueue, EPOLLOUT | EPOLLWRNORM | EPOLLWRBAND);
502}
503
504static void tap_sock_destruct(struct sock *sk)
505{
506 struct tap_queue *q = container_of(sk, struct tap_queue, sk);
507
508 ptr_ring_cleanup(&q->ring, __skb_array_destroy_skb);
509}
510
511static int tap_open(struct inode *inode, struct file *file)
512{
513 struct net *net = current->nsproxy->net_ns;
514 struct tap_dev *tap;
515 struct tap_queue *q;
516 int err = -ENODEV;
517
518 rtnl_lock();
519 tap = dev_get_by_tap_file(imajor(inode), iminor(inode));
520 if (!tap)
521 goto err;
522
523 err = -ENOMEM;
524 q = (struct tap_queue *)sk_alloc(net, AF_UNSPEC, GFP_KERNEL,
525 &tap_proto, 0);
526 if (!q)
527 goto err;
528 if (ptr_ring_init(&q->ring, tap->dev->tx_queue_len, GFP_KERNEL)) {
529 sk_free(&q->sk);
530 goto err;
531 }
532
533 init_waitqueue_head(&q->sock.wq.wait);
534 q->sock.type = SOCK_RAW;
535 q->sock.state = SS_CONNECTED;
536 q->sock.file = file;
537 q->sock.ops = &tap_socket_ops;
538 sock_init_data_uid(&q->sock, &q->sk, current_fsuid());
539 q->sk.sk_write_space = tap_sock_write_space;
540 q->sk.sk_destruct = tap_sock_destruct;
541 q->flags = IFF_VNET_HDR | IFF_NO_PI | IFF_TAP;
542 q->vnet_hdr_sz = sizeof(struct virtio_net_hdr);
543
544 /*
545 * so far only KVM virtio_net uses tap, enable zero copy between
546 * guest kernel and host kernel when lower device supports zerocopy
547 *
548 * The macvlan supports zerocopy iff the lower device supports zero
549 * copy so we don't have to look at the lower device directly.
550 */
551 if ((tap->dev->features & NETIF_F_HIGHDMA) && (tap->dev->features & NETIF_F_SG))
552 sock_set_flag(&q->sk, SOCK_ZEROCOPY);
553
554 err = tap_set_queue(tap, file, q);
555 if (err) {
556 /* tap_sock_destruct() will take care of freeing ptr_ring */
557 goto err_put;
558 }
559
560 /* tap groks IOCB_NOWAIT just fine, mark it as such */
561 file->f_mode |= FMODE_NOWAIT;
562
563 dev_put(tap->dev);
564
565 rtnl_unlock();
566 return err;
567
568err_put:
569 sock_put(&q->sk);
570err:
571 if (tap)
572 dev_put(tap->dev);
573
574 rtnl_unlock();
575 return err;
576}
577
578static int tap_release(struct inode *inode, struct file *file)
579{
580 struct tap_queue *q = file->private_data;
581 tap_put_queue(q);
582 return 0;
583}
584
585static __poll_t tap_poll(struct file *file, poll_table *wait)
586{
587 struct tap_queue *q = file->private_data;
588 __poll_t mask = EPOLLERR;
589
590 if (!q)
591 goto out;
592
593 mask = 0;
594 poll_wait(file, &q->sock.wq.wait, wait);
595
596 if (!ptr_ring_empty(&q->ring))
597 mask |= EPOLLIN | EPOLLRDNORM;
598
599 if (sock_writeable(&q->sk) ||
600 (!test_and_set_bit(SOCKWQ_ASYNC_NOSPACE, &q->sock.flags) &&
601 sock_writeable(&q->sk)))
602 mask |= EPOLLOUT | EPOLLWRNORM;
603
604out:
605 return mask;
606}
607
608static inline struct sk_buff *tap_alloc_skb(struct sock *sk, size_t prepad,
609 size_t len, size_t linear,
610 int noblock, int *err)
611{
612 struct sk_buff *skb;
613
614 /* Under a page? Don't bother with paged skb. */
615 if (prepad + len < PAGE_SIZE || !linear)
616 linear = len;
617
618 if (len - linear > MAX_SKB_FRAGS * (PAGE_SIZE << PAGE_ALLOC_COSTLY_ORDER))
619 linear = len - MAX_SKB_FRAGS * (PAGE_SIZE << PAGE_ALLOC_COSTLY_ORDER);
620 skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock,
621 err, PAGE_ALLOC_COSTLY_ORDER);
622 if (!skb)
623 return NULL;
624
625 skb_reserve(skb, prepad);
626 skb_put(skb, linear);
627 skb->data_len = len - linear;
628 skb->len += len - linear;
629
630 return skb;
631}
632
633/* Neighbour code has some assumptions on HH_DATA_MOD alignment */
634#define TAP_RESERVE HH_DATA_OFF(ETH_HLEN)
635
636/* Get packet from user space buffer */
637static ssize_t tap_get_user(struct tap_queue *q, void *msg_control,
638 struct iov_iter *from, int noblock)
639{
640 int good_linear = SKB_MAX_HEAD(TAP_RESERVE);
641 struct sk_buff *skb;
642 struct tap_dev *tap;
643 unsigned long total_len = iov_iter_count(from);
644 unsigned long len = total_len;
645 int err;
646 struct virtio_net_hdr vnet_hdr = { 0 };
647 int vnet_hdr_len = 0;
648 int copylen = 0;
649 int depth;
650 bool zerocopy = false;
651 size_t linear;
652 enum skb_drop_reason drop_reason;
653
654 if (q->flags & IFF_VNET_HDR) {
655 vnet_hdr_len = READ_ONCE(q->vnet_hdr_sz);
656
657 err = -EINVAL;
658 if (len < vnet_hdr_len)
659 goto err;
660 len -= vnet_hdr_len;
661
662 err = -EFAULT;
663 if (!copy_from_iter_full(&vnet_hdr, sizeof(vnet_hdr), from))
664 goto err;
665 iov_iter_advance(from, vnet_hdr_len - sizeof(vnet_hdr));
666 if ((vnet_hdr.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) &&
667 tap16_to_cpu(q, vnet_hdr.csum_start) +
668 tap16_to_cpu(q, vnet_hdr.csum_offset) + 2 >
669 tap16_to_cpu(q, vnet_hdr.hdr_len))
670 vnet_hdr.hdr_len = cpu_to_tap16(q,
671 tap16_to_cpu(q, vnet_hdr.csum_start) +
672 tap16_to_cpu(q, vnet_hdr.csum_offset) + 2);
673 err = -EINVAL;
674 if (tap16_to_cpu(q, vnet_hdr.hdr_len) > len)
675 goto err;
676 }
677
678 err = -EINVAL;
679 if (unlikely(len < ETH_HLEN))
680 goto err;
681
682 if (msg_control && sock_flag(&q->sk, SOCK_ZEROCOPY)) {
683 struct iov_iter i;
684
685 copylen = vnet_hdr.hdr_len ?
686 tap16_to_cpu(q, vnet_hdr.hdr_len) : GOODCOPY_LEN;
687 if (copylen > good_linear)
688 copylen = good_linear;
689 else if (copylen < ETH_HLEN)
690 copylen = ETH_HLEN;
691 linear = copylen;
692 i = *from;
693 iov_iter_advance(&i, copylen);
694 if (iov_iter_npages(&i, INT_MAX) <= MAX_SKB_FRAGS)
695 zerocopy = true;
696 }
697
698 if (!zerocopy) {
699 copylen = len;
700 linear = tap16_to_cpu(q, vnet_hdr.hdr_len);
701 if (linear > good_linear)
702 linear = good_linear;
703 else if (linear < ETH_HLEN)
704 linear = ETH_HLEN;
705 }
706
707 skb = tap_alloc_skb(&q->sk, TAP_RESERVE, copylen,
708 linear, noblock, &err);
709 if (!skb)
710 goto err;
711
712 if (zerocopy)
713 err = zerocopy_sg_from_iter(skb, from);
714 else
715 err = skb_copy_datagram_from_iter(skb, 0, from, len);
716
717 if (err) {
718 drop_reason = SKB_DROP_REASON_SKB_UCOPY_FAULT;
719 goto err_kfree;
720 }
721
722 skb_set_network_header(skb, ETH_HLEN);
723 skb_reset_mac_header(skb);
724 skb->protocol = eth_hdr(skb)->h_proto;
725
726 rcu_read_lock();
727 tap = rcu_dereference(q->tap);
728 if (!tap) {
729 kfree_skb(skb);
730 rcu_read_unlock();
731 return total_len;
732 }
733 skb->dev = tap->dev;
734
735 if (vnet_hdr_len) {
736 err = virtio_net_hdr_to_skb(skb, &vnet_hdr,
737 tap_is_little_endian(q));
738 if (err) {
739 rcu_read_unlock();
740 drop_reason = SKB_DROP_REASON_DEV_HDR;
741 goto err_kfree;
742 }
743 }
744
745 skb_probe_transport_header(skb);
746
747 /* Move network header to the right position for VLAN tagged packets */
748 if (eth_type_vlan(skb->protocol) &&
749 vlan_get_protocol_and_depth(skb, skb->protocol, &depth) != 0)
750 skb_set_network_header(skb, depth);
751
752 /* copy skb_ubuf_info for callback when skb has no error */
753 if (zerocopy) {
754 skb_zcopy_init(skb, msg_control);
755 } else if (msg_control) {
756 struct ubuf_info *uarg = msg_control;
757 uarg->callback(NULL, uarg, false);
758 }
759
760 dev_queue_xmit(skb);
761 rcu_read_unlock();
762 return total_len;
763
764err_kfree:
765 kfree_skb_reason(skb, drop_reason);
766
767err:
768 rcu_read_lock();
769 tap = rcu_dereference(q->tap);
770 if (tap && tap->count_tx_dropped)
771 tap->count_tx_dropped(tap);
772 rcu_read_unlock();
773
774 return err;
775}
776
777static ssize_t tap_write_iter(struct kiocb *iocb, struct iov_iter *from)
778{
779 struct file *file = iocb->ki_filp;
780 struct tap_queue *q = file->private_data;
781 int noblock = 0;
782
783 if ((file->f_flags & O_NONBLOCK) || (iocb->ki_flags & IOCB_NOWAIT))
784 noblock = 1;
785
786 return tap_get_user(q, NULL, from, noblock);
787}
788
789/* Put packet to the user space buffer */
790static ssize_t tap_put_user(struct tap_queue *q,
791 const struct sk_buff *skb,
792 struct iov_iter *iter)
793{
794 int ret;
795 int vnet_hdr_len = 0;
796 int vlan_offset = 0;
797 int total;
798
799 if (q->flags & IFF_VNET_HDR) {
800 int vlan_hlen = skb_vlan_tag_present(skb) ? VLAN_HLEN : 0;
801 struct virtio_net_hdr vnet_hdr;
802
803 vnet_hdr_len = READ_ONCE(q->vnet_hdr_sz);
804 if (iov_iter_count(iter) < vnet_hdr_len)
805 return -EINVAL;
806
807 if (virtio_net_hdr_from_skb(skb, &vnet_hdr,
808 tap_is_little_endian(q), true,
809 vlan_hlen))
810 BUG();
811
812 if (copy_to_iter(&vnet_hdr, sizeof(vnet_hdr), iter) !=
813 sizeof(vnet_hdr))
814 return -EFAULT;
815
816 iov_iter_advance(iter, vnet_hdr_len - sizeof(vnet_hdr));
817 }
818 total = vnet_hdr_len;
819 total += skb->len;
820
821 if (skb_vlan_tag_present(skb)) {
822 struct {
823 __be16 h_vlan_proto;
824 __be16 h_vlan_TCI;
825 } veth;
826 veth.h_vlan_proto = skb->vlan_proto;
827 veth.h_vlan_TCI = htons(skb_vlan_tag_get(skb));
828
829 vlan_offset = offsetof(struct vlan_ethhdr, h_vlan_proto);
830 total += VLAN_HLEN;
831
832 ret = skb_copy_datagram_iter(skb, 0, iter, vlan_offset);
833 if (ret || !iov_iter_count(iter))
834 goto done;
835
836 ret = copy_to_iter(&veth, sizeof(veth), iter);
837 if (ret != sizeof(veth) || !iov_iter_count(iter))
838 goto done;
839 }
840
841 ret = skb_copy_datagram_iter(skb, vlan_offset, iter,
842 skb->len - vlan_offset);
843
844done:
845 return ret ? ret : total;
846}
847
848static ssize_t tap_do_read(struct tap_queue *q,
849 struct iov_iter *to,
850 int noblock, struct sk_buff *skb)
851{
852 DEFINE_WAIT(wait);
853 ssize_t ret = 0;
854
855 if (!iov_iter_count(to)) {
856 kfree_skb(skb);
857 return 0;
858 }
859
860 if (skb)
861 goto put;
862
863 while (1) {
864 if (!noblock)
865 prepare_to_wait(sk_sleep(&q->sk), &wait,
866 TASK_INTERRUPTIBLE);
867
868 /* Read frames from the queue */
869 skb = ptr_ring_consume(&q->ring);
870 if (skb)
871 break;
872 if (noblock) {
873 ret = -EAGAIN;
874 break;
875 }
876 if (signal_pending(current)) {
877 ret = -ERESTARTSYS;
878 break;
879 }
880 /* Nothing to read, let's sleep */
881 schedule();
882 }
883 if (!noblock)
884 finish_wait(sk_sleep(&q->sk), &wait);
885
886put:
887 if (skb) {
888 ret = tap_put_user(q, skb, to);
889 if (unlikely(ret < 0))
890 kfree_skb(skb);
891 else
892 consume_skb(skb);
893 }
894 return ret;
895}
896
897static ssize_t tap_read_iter(struct kiocb *iocb, struct iov_iter *to)
898{
899 struct file *file = iocb->ki_filp;
900 struct tap_queue *q = file->private_data;
901 ssize_t len = iov_iter_count(to), ret;
902 int noblock = 0;
903
904 if ((file->f_flags & O_NONBLOCK) || (iocb->ki_flags & IOCB_NOWAIT))
905 noblock = 1;
906
907 ret = tap_do_read(q, to, noblock, NULL);
908 ret = min_t(ssize_t, ret, len);
909 if (ret > 0)
910 iocb->ki_pos = ret;
911 return ret;
912}
913
914static struct tap_dev *tap_get_tap_dev(struct tap_queue *q)
915{
916 struct tap_dev *tap;
917
918 ASSERT_RTNL();
919 tap = rtnl_dereference(q->tap);
920 if (tap)
921 dev_hold(tap->dev);
922
923 return tap;
924}
925
926static void tap_put_tap_dev(struct tap_dev *tap)
927{
928 dev_put(tap->dev);
929}
930
931static int tap_ioctl_set_queue(struct file *file, unsigned int flags)
932{
933 struct tap_queue *q = file->private_data;
934 struct tap_dev *tap;
935 int ret;
936
937 tap = tap_get_tap_dev(q);
938 if (!tap)
939 return -EINVAL;
940
941 if (flags & IFF_ATTACH_QUEUE)
942 ret = tap_enable_queue(tap, file, q);
943 else if (flags & IFF_DETACH_QUEUE)
944 ret = tap_disable_queue(q);
945 else
946 ret = -EINVAL;
947
948 tap_put_tap_dev(tap);
949 return ret;
950}
951
952static int set_offload(struct tap_queue *q, unsigned long arg)
953{
954 struct tap_dev *tap;
955 netdev_features_t features;
956 netdev_features_t feature_mask = 0;
957
958 tap = rtnl_dereference(q->tap);
959 if (!tap)
960 return -ENOLINK;
961
962 features = tap->dev->features;
963
964 if (arg & TUN_F_CSUM) {
965 feature_mask = NETIF_F_HW_CSUM;
966
967 if (arg & (TUN_F_TSO4 | TUN_F_TSO6)) {
968 if (arg & TUN_F_TSO_ECN)
969 feature_mask |= NETIF_F_TSO_ECN;
970 if (arg & TUN_F_TSO4)
971 feature_mask |= NETIF_F_TSO;
972 if (arg & TUN_F_TSO6)
973 feature_mask |= NETIF_F_TSO6;
974 }
975
976 /* TODO: for now USO4 and USO6 should work simultaneously */
977 if ((arg & (TUN_F_USO4 | TUN_F_USO6)) == (TUN_F_USO4 | TUN_F_USO6))
978 features |= NETIF_F_GSO_UDP_L4;
979 }
980
981 /* tun/tap driver inverts the usage for TSO offloads, where
982 * setting the TSO bit means that the userspace wants to
983 * accept TSO frames and turning it off means that user space
984 * does not support TSO.
985 * For tap, we have to invert it to mean the same thing.
986 * When user space turns off TSO, we turn off GSO/LRO so that
987 * user-space will not receive TSO frames.
988 */
989 if (feature_mask & (NETIF_F_TSO | NETIF_F_TSO6) ||
990 (feature_mask & (TUN_F_USO4 | TUN_F_USO6)) == (TUN_F_USO4 | TUN_F_USO6))
991 features |= RX_OFFLOADS;
992 else
993 features &= ~RX_OFFLOADS;
994
995 /* tap_features are the same as features on tun/tap and
996 * reflect user expectations.
997 */
998 tap->tap_features = feature_mask;
999 if (tap->update_features)
1000 tap->update_features(tap, features);
1001
1002 return 0;
1003}
1004
1005/*
1006 * provide compatibility with generic tun/tap interface
1007 */
1008static long tap_ioctl(struct file *file, unsigned int cmd,
1009 unsigned long arg)
1010{
1011 struct tap_queue *q = file->private_data;
1012 struct tap_dev *tap;
1013 void __user *argp = (void __user *)arg;
1014 struct ifreq __user *ifr = argp;
1015 unsigned int __user *up = argp;
1016 unsigned short u;
1017 int __user *sp = argp;
1018 struct sockaddr sa;
1019 int s;
1020 int ret;
1021
1022 switch (cmd) {
1023 case TUNSETIFF:
1024 /* ignore the name, just look at flags */
1025 if (get_user(u, &ifr->ifr_flags))
1026 return -EFAULT;
1027
1028 ret = 0;
1029 if ((u & ~TAP_IFFEATURES) != (IFF_NO_PI | IFF_TAP))
1030 ret = -EINVAL;
1031 else
1032 q->flags = (q->flags & ~TAP_IFFEATURES) | u;
1033
1034 return ret;
1035
1036 case TUNGETIFF:
1037 rtnl_lock();
1038 tap = tap_get_tap_dev(q);
1039 if (!tap) {
1040 rtnl_unlock();
1041 return -ENOLINK;
1042 }
1043
1044 ret = 0;
1045 u = q->flags;
1046 if (copy_to_user(&ifr->ifr_name, tap->dev->name, IFNAMSIZ) ||
1047 put_user(u, &ifr->ifr_flags))
1048 ret = -EFAULT;
1049 tap_put_tap_dev(tap);
1050 rtnl_unlock();
1051 return ret;
1052
1053 case TUNSETQUEUE:
1054 if (get_user(u, &ifr->ifr_flags))
1055 return -EFAULT;
1056 rtnl_lock();
1057 ret = tap_ioctl_set_queue(file, u);
1058 rtnl_unlock();
1059 return ret;
1060
1061 case TUNGETFEATURES:
1062 if (put_user(IFF_TAP | IFF_NO_PI | TAP_IFFEATURES, up))
1063 return -EFAULT;
1064 return 0;
1065
1066 case TUNSETSNDBUF:
1067 if (get_user(s, sp))
1068 return -EFAULT;
1069 if (s <= 0)
1070 return -EINVAL;
1071
1072 q->sk.sk_sndbuf = s;
1073 return 0;
1074
1075 case TUNGETVNETHDRSZ:
1076 s = q->vnet_hdr_sz;
1077 if (put_user(s, sp))
1078 return -EFAULT;
1079 return 0;
1080
1081 case TUNSETVNETHDRSZ:
1082 if (get_user(s, sp))
1083 return -EFAULT;
1084 if (s < (int)sizeof(struct virtio_net_hdr))
1085 return -EINVAL;
1086
1087 q->vnet_hdr_sz = s;
1088 return 0;
1089
1090 case TUNGETVNETLE:
1091 s = !!(q->flags & TAP_VNET_LE);
1092 if (put_user(s, sp))
1093 return -EFAULT;
1094 return 0;
1095
1096 case TUNSETVNETLE:
1097 if (get_user(s, sp))
1098 return -EFAULT;
1099 if (s)
1100 q->flags |= TAP_VNET_LE;
1101 else
1102 q->flags &= ~TAP_VNET_LE;
1103 return 0;
1104
1105 case TUNGETVNETBE:
1106 return tap_get_vnet_be(q, sp);
1107
1108 case TUNSETVNETBE:
1109 return tap_set_vnet_be(q, sp);
1110
1111 case TUNSETOFFLOAD:
1112 /* let the user check for future flags */
1113 if (arg & ~(TUN_F_CSUM | TUN_F_TSO4 | TUN_F_TSO6 |
1114 TUN_F_TSO_ECN | TUN_F_UFO |
1115 TUN_F_USO4 | TUN_F_USO6))
1116 return -EINVAL;
1117
1118 rtnl_lock();
1119 ret = set_offload(q, arg);
1120 rtnl_unlock();
1121 return ret;
1122
1123 case SIOCGIFHWADDR:
1124 rtnl_lock();
1125 tap = tap_get_tap_dev(q);
1126 if (!tap) {
1127 rtnl_unlock();
1128 return -ENOLINK;
1129 }
1130 ret = 0;
1131 dev_get_mac_address(&sa, dev_net(tap->dev), tap->dev->name);
1132 if (copy_to_user(&ifr->ifr_name, tap->dev->name, IFNAMSIZ) ||
1133 copy_to_user(&ifr->ifr_hwaddr, &sa, sizeof(sa)))
1134 ret = -EFAULT;
1135 tap_put_tap_dev(tap);
1136 rtnl_unlock();
1137 return ret;
1138
1139 case SIOCSIFHWADDR:
1140 if (copy_from_user(&sa, &ifr->ifr_hwaddr, sizeof(sa)))
1141 return -EFAULT;
1142 rtnl_lock();
1143 tap = tap_get_tap_dev(q);
1144 if (!tap) {
1145 rtnl_unlock();
1146 return -ENOLINK;
1147 }
1148 ret = dev_set_mac_address_user(tap->dev, &sa, NULL);
1149 tap_put_tap_dev(tap);
1150 rtnl_unlock();
1151 return ret;
1152
1153 default:
1154 return -EINVAL;
1155 }
1156}
1157
1158static const struct file_operations tap_fops = {
1159 .owner = THIS_MODULE,
1160 .open = tap_open,
1161 .release = tap_release,
1162 .read_iter = tap_read_iter,
1163 .write_iter = tap_write_iter,
1164 .poll = tap_poll,
1165 .llseek = no_llseek,
1166 .unlocked_ioctl = tap_ioctl,
1167 .compat_ioctl = compat_ptr_ioctl,
1168};
1169
1170static int tap_get_user_xdp(struct tap_queue *q, struct xdp_buff *xdp)
1171{
1172 struct tun_xdp_hdr *hdr = xdp->data_hard_start;
1173 struct virtio_net_hdr *gso = &hdr->gso;
1174 int buflen = hdr->buflen;
1175 int vnet_hdr_len = 0;
1176 struct tap_dev *tap;
1177 struct sk_buff *skb;
1178 int err, depth;
1179
1180 if (q->flags & IFF_VNET_HDR)
1181 vnet_hdr_len = READ_ONCE(q->vnet_hdr_sz);
1182
1183 skb = build_skb(xdp->data_hard_start, buflen);
1184 if (!skb) {
1185 err = -ENOMEM;
1186 goto err;
1187 }
1188
1189 skb_reserve(skb, xdp->data - xdp->data_hard_start);
1190 skb_put(skb, xdp->data_end - xdp->data);
1191
1192 skb_set_network_header(skb, ETH_HLEN);
1193 skb_reset_mac_header(skb);
1194 skb->protocol = eth_hdr(skb)->h_proto;
1195
1196 if (vnet_hdr_len) {
1197 err = virtio_net_hdr_to_skb(skb, gso, tap_is_little_endian(q));
1198 if (err)
1199 goto err_kfree;
1200 }
1201
1202 /* Move network header to the right position for VLAN tagged packets */
1203 if (eth_type_vlan(skb->protocol) &&
1204 vlan_get_protocol_and_depth(skb, skb->protocol, &depth) != 0)
1205 skb_set_network_header(skb, depth);
1206
1207 rcu_read_lock();
1208 tap = rcu_dereference(q->tap);
1209 if (tap) {
1210 skb->dev = tap->dev;
1211 skb_probe_transport_header(skb);
1212 dev_queue_xmit(skb);
1213 } else {
1214 kfree_skb(skb);
1215 }
1216 rcu_read_unlock();
1217
1218 return 0;
1219
1220err_kfree:
1221 kfree_skb(skb);
1222err:
1223 rcu_read_lock();
1224 tap = rcu_dereference(q->tap);
1225 if (tap && tap->count_tx_dropped)
1226 tap->count_tx_dropped(tap);
1227 rcu_read_unlock();
1228 return err;
1229}
1230
1231static int tap_sendmsg(struct socket *sock, struct msghdr *m,
1232 size_t total_len)
1233{
1234 struct tap_queue *q = container_of(sock, struct tap_queue, sock);
1235 struct tun_msg_ctl *ctl = m->msg_control;
1236 struct xdp_buff *xdp;
1237 int i;
1238
1239 if (m->msg_controllen == sizeof(struct tun_msg_ctl) &&
1240 ctl && ctl->type == TUN_MSG_PTR) {
1241 for (i = 0; i < ctl->num; i++) {
1242 xdp = &((struct xdp_buff *)ctl->ptr)[i];
1243 tap_get_user_xdp(q, xdp);
1244 }
1245 return 0;
1246 }
1247
1248 return tap_get_user(q, ctl ? ctl->ptr : NULL, &m->msg_iter,
1249 m->msg_flags & MSG_DONTWAIT);
1250}
1251
1252static int tap_recvmsg(struct socket *sock, struct msghdr *m,
1253 size_t total_len, int flags)
1254{
1255 struct tap_queue *q = container_of(sock, struct tap_queue, sock);
1256 struct sk_buff *skb = m->msg_control;
1257 int ret;
1258 if (flags & ~(MSG_DONTWAIT|MSG_TRUNC)) {
1259 kfree_skb(skb);
1260 return -EINVAL;
1261 }
1262 ret = tap_do_read(q, &m->msg_iter, flags & MSG_DONTWAIT, skb);
1263 if (ret > total_len) {
1264 m->msg_flags |= MSG_TRUNC;
1265 ret = flags & MSG_TRUNC ? ret : total_len;
1266 }
1267 return ret;
1268}
1269
1270static int tap_peek_len(struct socket *sock)
1271{
1272 struct tap_queue *q = container_of(sock, struct tap_queue,
1273 sock);
1274 return PTR_RING_PEEK_CALL(&q->ring, __skb_array_len_with_tag);
1275}
1276
1277/* Ops structure to mimic raw sockets with tun */
1278static const struct proto_ops tap_socket_ops = {
1279 .sendmsg = tap_sendmsg,
1280 .recvmsg = tap_recvmsg,
1281 .peek_len = tap_peek_len,
1282};
1283
1284/* Get an underlying socket object from tun file. Returns error unless file is
1285 * attached to a device. The returned object works like a packet socket, it
1286 * can be used for sock_sendmsg/sock_recvmsg. The caller is responsible for
1287 * holding a reference to the file for as long as the socket is in use. */
1288struct socket *tap_get_socket(struct file *file)
1289{
1290 struct tap_queue *q;
1291 if (file->f_op != &tap_fops)
1292 return ERR_PTR(-EINVAL);
1293 q = file->private_data;
1294 if (!q)
1295 return ERR_PTR(-EBADFD);
1296 return &q->sock;
1297}
1298EXPORT_SYMBOL_GPL(tap_get_socket);
1299
1300struct ptr_ring *tap_get_ptr_ring(struct file *file)
1301{
1302 struct tap_queue *q;
1303
1304 if (file->f_op != &tap_fops)
1305 return ERR_PTR(-EINVAL);
1306 q = file->private_data;
1307 if (!q)
1308 return ERR_PTR(-EBADFD);
1309 return &q->ring;
1310}
1311EXPORT_SYMBOL_GPL(tap_get_ptr_ring);
1312
1313int tap_queue_resize(struct tap_dev *tap)
1314{
1315 struct net_device *dev = tap->dev;
1316 struct tap_queue *q;
1317 struct ptr_ring **rings;
1318 int n = tap->numqueues;
1319 int ret, i = 0;
1320
1321 rings = kmalloc_array(n, sizeof(*rings), GFP_KERNEL);
1322 if (!rings)
1323 return -ENOMEM;
1324
1325 list_for_each_entry(q, &tap->queue_list, next)
1326 rings[i++] = &q->ring;
1327
1328 ret = ptr_ring_resize_multiple(rings, n,
1329 dev->tx_queue_len, GFP_KERNEL,
1330 __skb_array_destroy_skb);
1331
1332 kfree(rings);
1333 return ret;
1334}
1335EXPORT_SYMBOL_GPL(tap_queue_resize);
1336
1337static int tap_list_add(dev_t major, const char *device_name)
1338{
1339 struct major_info *tap_major;
1340
1341 tap_major = kzalloc(sizeof(*tap_major), GFP_ATOMIC);
1342 if (!tap_major)
1343 return -ENOMEM;
1344
1345 tap_major->major = MAJOR(major);
1346
1347 idr_init(&tap_major->minor_idr);
1348 spin_lock_init(&tap_major->minor_lock);
1349
1350 tap_major->device_name = device_name;
1351
1352 list_add_tail_rcu(&tap_major->next, &major_list);
1353 return 0;
1354}
1355
1356int tap_create_cdev(struct cdev *tap_cdev, dev_t *tap_major,
1357 const char *device_name, struct module *module)
1358{
1359 int err;
1360
1361 err = alloc_chrdev_region(tap_major, 0, TAP_NUM_DEVS, device_name);
1362 if (err)
1363 goto out1;
1364
1365 cdev_init(tap_cdev, &tap_fops);
1366 tap_cdev->owner = module;
1367 err = cdev_add(tap_cdev, *tap_major, TAP_NUM_DEVS);
1368 if (err)
1369 goto out2;
1370
1371 err = tap_list_add(*tap_major, device_name);
1372 if (err)
1373 goto out3;
1374
1375 return 0;
1376
1377out3:
1378 cdev_del(tap_cdev);
1379out2:
1380 unregister_chrdev_region(*tap_major, TAP_NUM_DEVS);
1381out1:
1382 return err;
1383}
1384EXPORT_SYMBOL_GPL(tap_create_cdev);
1385
1386void tap_destroy_cdev(dev_t major, struct cdev *tap_cdev)
1387{
1388 struct major_info *tap_major, *tmp;
1389
1390 cdev_del(tap_cdev);
1391 unregister_chrdev_region(major, TAP_NUM_DEVS);
1392 list_for_each_entry_safe(tap_major, tmp, &major_list, next) {
1393 if (tap_major->major == MAJOR(major)) {
1394 idr_destroy(&tap_major->minor_idr);
1395 list_del_rcu(&tap_major->next);
1396 kfree_rcu(tap_major, rcu);
1397 }
1398 }
1399}
1400EXPORT_SYMBOL_GPL(tap_destroy_cdev);
1401
1402MODULE_DESCRIPTION("Common library for drivers implementing the TAP interface");
1403MODULE_AUTHOR("Arnd Bergmann <arnd@arndb.de>");
1404MODULE_AUTHOR("Sainath Grandhi <sainath.grandhi@intel.com>");
1405MODULE_LICENSE("GPL");
1// SPDX-License-Identifier: GPL-2.0-only
2#include <linux/etherdevice.h>
3#include <linux/if_tap.h>
4#include <linux/if_vlan.h>
5#include <linux/interrupt.h>
6#include <linux/nsproxy.h>
7#include <linux/compat.h>
8#include <linux/if_tun.h>
9#include <linux/module.h>
10#include <linux/skbuff.h>
11#include <linux/cache.h>
12#include <linux/sched/signal.h>
13#include <linux/types.h>
14#include <linux/slab.h>
15#include <linux/wait.h>
16#include <linux/cdev.h>
17#include <linux/idr.h>
18#include <linux/fs.h>
19#include <linux/uio.h>
20
21#include <net/net_namespace.h>
22#include <net/rtnetlink.h>
23#include <net/sock.h>
24#include <linux/virtio_net.h>
25#include <linux/skb_array.h>
26
27#define TAP_IFFEATURES (IFF_VNET_HDR | IFF_MULTI_QUEUE)
28
29#define TAP_VNET_LE 0x80000000
30#define TAP_VNET_BE 0x40000000
31
32#ifdef CONFIG_TUN_VNET_CROSS_LE
33static inline bool tap_legacy_is_little_endian(struct tap_queue *q)
34{
35 return q->flags & TAP_VNET_BE ? false :
36 virtio_legacy_is_little_endian();
37}
38
39static long tap_get_vnet_be(struct tap_queue *q, int __user *sp)
40{
41 int s = !!(q->flags & TAP_VNET_BE);
42
43 if (put_user(s, sp))
44 return -EFAULT;
45
46 return 0;
47}
48
49static long tap_set_vnet_be(struct tap_queue *q, int __user *sp)
50{
51 int s;
52
53 if (get_user(s, sp))
54 return -EFAULT;
55
56 if (s)
57 q->flags |= TAP_VNET_BE;
58 else
59 q->flags &= ~TAP_VNET_BE;
60
61 return 0;
62}
63#else
64static inline bool tap_legacy_is_little_endian(struct tap_queue *q)
65{
66 return virtio_legacy_is_little_endian();
67}
68
69static long tap_get_vnet_be(struct tap_queue *q, int __user *argp)
70{
71 return -EINVAL;
72}
73
74static long tap_set_vnet_be(struct tap_queue *q, int __user *argp)
75{
76 return -EINVAL;
77}
78#endif /* CONFIG_TUN_VNET_CROSS_LE */
79
80static inline bool tap_is_little_endian(struct tap_queue *q)
81{
82 return q->flags & TAP_VNET_LE ||
83 tap_legacy_is_little_endian(q);
84}
85
86static inline u16 tap16_to_cpu(struct tap_queue *q, __virtio16 val)
87{
88 return __virtio16_to_cpu(tap_is_little_endian(q), val);
89}
90
91static inline __virtio16 cpu_to_tap16(struct tap_queue *q, u16 val)
92{
93 return __cpu_to_virtio16(tap_is_little_endian(q), val);
94}
95
96static struct proto tap_proto = {
97 .name = "tap",
98 .owner = THIS_MODULE,
99 .obj_size = sizeof(struct tap_queue),
100};
101
102#define TAP_NUM_DEVS (1U << MINORBITS)
103
104static LIST_HEAD(major_list);
105
106struct major_info {
107 struct rcu_head rcu;
108 dev_t major;
109 struct idr minor_idr;
110 spinlock_t minor_lock;
111 const char *device_name;
112 struct list_head next;
113};
114
115#define GOODCOPY_LEN 128
116
117static const struct proto_ops tap_socket_ops;
118
119#define RX_OFFLOADS (NETIF_F_GRO | NETIF_F_LRO)
120#define TAP_FEATURES (NETIF_F_GSO | NETIF_F_SG | NETIF_F_FRAGLIST)
121
122static struct tap_dev *tap_dev_get_rcu(const struct net_device *dev)
123{
124 return rcu_dereference(dev->rx_handler_data);
125}
126
127/*
128 * RCU usage:
129 * The tap_queue and the macvlan_dev are loosely coupled, the
130 * pointers from one to the other can only be read while rcu_read_lock
131 * or rtnl is held.
132 *
133 * Both the file and the macvlan_dev hold a reference on the tap_queue
134 * through sock_hold(&q->sk). When the macvlan_dev goes away first,
135 * q->vlan becomes inaccessible. When the files gets closed,
136 * tap_get_queue() fails.
137 *
138 * There may still be references to the struct sock inside of the
139 * queue from outbound SKBs, but these never reference back to the
140 * file or the dev. The data structure is freed through __sk_free
141 * when both our references and any pending SKBs are gone.
142 */
143
144static int tap_enable_queue(struct tap_dev *tap, struct file *file,
145 struct tap_queue *q)
146{
147 int err = -EINVAL;
148
149 ASSERT_RTNL();
150
151 if (q->enabled)
152 goto out;
153
154 err = 0;
155 rcu_assign_pointer(tap->taps[tap->numvtaps], q);
156 q->queue_index = tap->numvtaps;
157 q->enabled = true;
158
159 tap->numvtaps++;
160out:
161 return err;
162}
163
164/* Requires RTNL */
165static int tap_set_queue(struct tap_dev *tap, struct file *file,
166 struct tap_queue *q)
167{
168 if (tap->numqueues == MAX_TAP_QUEUES)
169 return -EBUSY;
170
171 rcu_assign_pointer(q->tap, tap);
172 rcu_assign_pointer(tap->taps[tap->numvtaps], q);
173 sock_hold(&q->sk);
174
175 q->file = file;
176 q->queue_index = tap->numvtaps;
177 q->enabled = true;
178 file->private_data = q;
179 list_add_tail(&q->next, &tap->queue_list);
180
181 tap->numvtaps++;
182 tap->numqueues++;
183
184 return 0;
185}
186
187static int tap_disable_queue(struct tap_queue *q)
188{
189 struct tap_dev *tap;
190 struct tap_queue *nq;
191
192 ASSERT_RTNL();
193 if (!q->enabled)
194 return -EINVAL;
195
196 tap = rtnl_dereference(q->tap);
197
198 if (tap) {
199 int index = q->queue_index;
200 BUG_ON(index >= tap->numvtaps);
201 nq = rtnl_dereference(tap->taps[tap->numvtaps - 1]);
202 nq->queue_index = index;
203
204 rcu_assign_pointer(tap->taps[index], nq);
205 RCU_INIT_POINTER(tap->taps[tap->numvtaps - 1], NULL);
206 q->enabled = false;
207
208 tap->numvtaps--;
209 }
210
211 return 0;
212}
213
214/*
215 * The file owning the queue got closed, give up both
216 * the reference that the files holds as well as the
217 * one from the macvlan_dev if that still exists.
218 *
219 * Using the spinlock makes sure that we don't get
220 * to the queue again after destroying it.
221 */
222static void tap_put_queue(struct tap_queue *q)
223{
224 struct tap_dev *tap;
225
226 rtnl_lock();
227 tap = rtnl_dereference(q->tap);
228
229 if (tap) {
230 if (q->enabled)
231 BUG_ON(tap_disable_queue(q));
232
233 tap->numqueues--;
234 RCU_INIT_POINTER(q->tap, NULL);
235 sock_put(&q->sk);
236 list_del_init(&q->next);
237 }
238
239 rtnl_unlock();
240
241 synchronize_rcu();
242 sock_put(&q->sk);
243}
244
245/*
246 * Select a queue based on the rxq of the device on which this packet
247 * arrived. If the incoming device is not mq, calculate a flow hash
248 * to select a queue. If all fails, find the first available queue.
249 * Cache vlan->numvtaps since it can become zero during the execution
250 * of this function.
251 */
252static struct tap_queue *tap_get_queue(struct tap_dev *tap,
253 struct sk_buff *skb)
254{
255 struct tap_queue *queue = NULL;
256 /* Access to taps array is protected by rcu, but access to numvtaps
257 * isn't. Below we use it to lookup a queue, but treat it as a hint
258 * and validate that the result isn't NULL - in case we are
259 * racing against queue removal.
260 */
261 int numvtaps = READ_ONCE(tap->numvtaps);
262 __u32 rxq;
263
264 if (!numvtaps)
265 goto out;
266
267 if (numvtaps == 1)
268 goto single;
269
270 /* Check if we can use flow to select a queue */
271 rxq = skb_get_hash(skb);
272 if (rxq) {
273 queue = rcu_dereference(tap->taps[rxq % numvtaps]);
274 goto out;
275 }
276
277 if (likely(skb_rx_queue_recorded(skb))) {
278 rxq = skb_get_rx_queue(skb);
279
280 while (unlikely(rxq >= numvtaps))
281 rxq -= numvtaps;
282
283 queue = rcu_dereference(tap->taps[rxq]);
284 goto out;
285 }
286
287single:
288 queue = rcu_dereference(tap->taps[0]);
289out:
290 return queue;
291}
292
293/*
294 * The net_device is going away, give up the reference
295 * that it holds on all queues and safely set the pointer
296 * from the queues to NULL.
297 */
298void tap_del_queues(struct tap_dev *tap)
299{
300 struct tap_queue *q, *tmp;
301
302 ASSERT_RTNL();
303 list_for_each_entry_safe(q, tmp, &tap->queue_list, next) {
304 list_del_init(&q->next);
305 RCU_INIT_POINTER(q->tap, NULL);
306 if (q->enabled)
307 tap->numvtaps--;
308 tap->numqueues--;
309 sock_put(&q->sk);
310 }
311 BUG_ON(tap->numvtaps);
312 BUG_ON(tap->numqueues);
313 /* guarantee that any future tap_set_queue will fail */
314 tap->numvtaps = MAX_TAP_QUEUES;
315}
316EXPORT_SYMBOL_GPL(tap_del_queues);
317
318rx_handler_result_t tap_handle_frame(struct sk_buff **pskb)
319{
320 struct sk_buff *skb = *pskb;
321 struct net_device *dev = skb->dev;
322 struct tap_dev *tap;
323 struct tap_queue *q;
324 netdev_features_t features = TAP_FEATURES;
325
326 tap = tap_dev_get_rcu(dev);
327 if (!tap)
328 return RX_HANDLER_PASS;
329
330 q = tap_get_queue(tap, skb);
331 if (!q)
332 return RX_HANDLER_PASS;
333
334 skb_push(skb, ETH_HLEN);
335
336 /* Apply the forward feature mask so that we perform segmentation
337 * according to users wishes. This only works if VNET_HDR is
338 * enabled.
339 */
340 if (q->flags & IFF_VNET_HDR)
341 features |= tap->tap_features;
342 if (netif_needs_gso(skb, features)) {
343 struct sk_buff *segs = __skb_gso_segment(skb, features, false);
344 struct sk_buff *next;
345
346 if (IS_ERR(segs))
347 goto drop;
348
349 if (!segs) {
350 if (ptr_ring_produce(&q->ring, skb))
351 goto drop;
352 goto wake_up;
353 }
354
355 consume_skb(skb);
356 skb_list_walk_safe(segs, skb, next) {
357 skb_mark_not_on_list(skb);
358 if (ptr_ring_produce(&q->ring, skb)) {
359 kfree_skb(skb);
360 kfree_skb_list(next);
361 break;
362 }
363 }
364 } else {
365 /* If we receive a partial checksum and the tap side
366 * doesn't support checksum offload, compute the checksum.
367 * Note: it doesn't matter which checksum feature to
368 * check, we either support them all or none.
369 */
370 if (skb->ip_summed == CHECKSUM_PARTIAL &&
371 !(features & NETIF_F_CSUM_MASK) &&
372 skb_checksum_help(skb))
373 goto drop;
374 if (ptr_ring_produce(&q->ring, skb))
375 goto drop;
376 }
377
378wake_up:
379 wake_up_interruptible_poll(sk_sleep(&q->sk), EPOLLIN | EPOLLRDNORM | EPOLLRDBAND);
380 return RX_HANDLER_CONSUMED;
381
382drop:
383 /* Count errors/drops only here, thus don't care about args. */
384 if (tap->count_rx_dropped)
385 tap->count_rx_dropped(tap);
386 kfree_skb(skb);
387 return RX_HANDLER_CONSUMED;
388}
389EXPORT_SYMBOL_GPL(tap_handle_frame);
390
391static struct major_info *tap_get_major(int major)
392{
393 struct major_info *tap_major;
394
395 list_for_each_entry_rcu(tap_major, &major_list, next) {
396 if (tap_major->major == major)
397 return tap_major;
398 }
399
400 return NULL;
401}
402
403int tap_get_minor(dev_t major, struct tap_dev *tap)
404{
405 int retval = -ENOMEM;
406 struct major_info *tap_major;
407
408 rcu_read_lock();
409 tap_major = tap_get_major(MAJOR(major));
410 if (!tap_major) {
411 retval = -EINVAL;
412 goto unlock;
413 }
414
415 spin_lock(&tap_major->minor_lock);
416 retval = idr_alloc(&tap_major->minor_idr, tap, 1, TAP_NUM_DEVS, GFP_ATOMIC);
417 if (retval >= 0) {
418 tap->minor = retval;
419 } else if (retval == -ENOSPC) {
420 netdev_err(tap->dev, "Too many tap devices\n");
421 retval = -EINVAL;
422 }
423 spin_unlock(&tap_major->minor_lock);
424
425unlock:
426 rcu_read_unlock();
427 return retval < 0 ? retval : 0;
428}
429EXPORT_SYMBOL_GPL(tap_get_minor);
430
431void tap_free_minor(dev_t major, struct tap_dev *tap)
432{
433 struct major_info *tap_major;
434
435 rcu_read_lock();
436 tap_major = tap_get_major(MAJOR(major));
437 if (!tap_major) {
438 goto unlock;
439 }
440
441 spin_lock(&tap_major->minor_lock);
442 if (tap->minor) {
443 idr_remove(&tap_major->minor_idr, tap->minor);
444 tap->minor = 0;
445 }
446 spin_unlock(&tap_major->minor_lock);
447
448unlock:
449 rcu_read_unlock();
450}
451EXPORT_SYMBOL_GPL(tap_free_minor);
452
453static struct tap_dev *dev_get_by_tap_file(int major, int minor)
454{
455 struct net_device *dev = NULL;
456 struct tap_dev *tap;
457 struct major_info *tap_major;
458
459 rcu_read_lock();
460 tap_major = tap_get_major(major);
461 if (!tap_major) {
462 tap = NULL;
463 goto unlock;
464 }
465
466 spin_lock(&tap_major->minor_lock);
467 tap = idr_find(&tap_major->minor_idr, minor);
468 if (tap) {
469 dev = tap->dev;
470 dev_hold(dev);
471 }
472 spin_unlock(&tap_major->minor_lock);
473
474unlock:
475 rcu_read_unlock();
476 return tap;
477}
478
479static void tap_sock_write_space(struct sock *sk)
480{
481 wait_queue_head_t *wqueue;
482
483 if (!sock_writeable(sk) ||
484 !test_and_clear_bit(SOCKWQ_ASYNC_NOSPACE, &sk->sk_socket->flags))
485 return;
486
487 wqueue = sk_sleep(sk);
488 if (wqueue && waitqueue_active(wqueue))
489 wake_up_interruptible_poll(wqueue, EPOLLOUT | EPOLLWRNORM | EPOLLWRBAND);
490}
491
492static void tap_sock_destruct(struct sock *sk)
493{
494 struct tap_queue *q = container_of(sk, struct tap_queue, sk);
495
496 ptr_ring_cleanup(&q->ring, __skb_array_destroy_skb);
497}
498
499static int tap_open(struct inode *inode, struct file *file)
500{
501 struct net *net = current->nsproxy->net_ns;
502 struct tap_dev *tap;
503 struct tap_queue *q;
504 int err = -ENODEV;
505
506 rtnl_lock();
507 tap = dev_get_by_tap_file(imajor(inode), iminor(inode));
508 if (!tap)
509 goto err;
510
511 err = -ENOMEM;
512 q = (struct tap_queue *)sk_alloc(net, AF_UNSPEC, GFP_KERNEL,
513 &tap_proto, 0);
514 if (!q)
515 goto err;
516 if (ptr_ring_init(&q->ring, tap->dev->tx_queue_len, GFP_KERNEL)) {
517 sk_free(&q->sk);
518 goto err;
519 }
520
521 init_waitqueue_head(&q->sock.wq.wait);
522 q->sock.type = SOCK_RAW;
523 q->sock.state = SS_CONNECTED;
524 q->sock.file = file;
525 q->sock.ops = &tap_socket_ops;
526 sock_init_data(&q->sock, &q->sk);
527 q->sk.sk_write_space = tap_sock_write_space;
528 q->sk.sk_destruct = tap_sock_destruct;
529 q->flags = IFF_VNET_HDR | IFF_NO_PI | IFF_TAP;
530 q->vnet_hdr_sz = sizeof(struct virtio_net_hdr);
531
532 /*
533 * so far only KVM virtio_net uses tap, enable zero copy between
534 * guest kernel and host kernel when lower device supports zerocopy
535 *
536 * The macvlan supports zerocopy iff the lower device supports zero
537 * copy so we don't have to look at the lower device directly.
538 */
539 if ((tap->dev->features & NETIF_F_HIGHDMA) && (tap->dev->features & NETIF_F_SG))
540 sock_set_flag(&q->sk, SOCK_ZEROCOPY);
541
542 err = tap_set_queue(tap, file, q);
543 if (err) {
544 /* tap_sock_destruct() will take care of freeing ptr_ring */
545 goto err_put;
546 }
547
548 dev_put(tap->dev);
549
550 rtnl_unlock();
551 return err;
552
553err_put:
554 sock_put(&q->sk);
555err:
556 if (tap)
557 dev_put(tap->dev);
558
559 rtnl_unlock();
560 return err;
561}
562
563static int tap_release(struct inode *inode, struct file *file)
564{
565 struct tap_queue *q = file->private_data;
566 tap_put_queue(q);
567 return 0;
568}
569
570static __poll_t tap_poll(struct file *file, poll_table *wait)
571{
572 struct tap_queue *q = file->private_data;
573 __poll_t mask = EPOLLERR;
574
575 if (!q)
576 goto out;
577
578 mask = 0;
579 poll_wait(file, &q->sock.wq.wait, wait);
580
581 if (!ptr_ring_empty(&q->ring))
582 mask |= EPOLLIN | EPOLLRDNORM;
583
584 if (sock_writeable(&q->sk) ||
585 (!test_and_set_bit(SOCKWQ_ASYNC_NOSPACE, &q->sock.flags) &&
586 sock_writeable(&q->sk)))
587 mask |= EPOLLOUT | EPOLLWRNORM;
588
589out:
590 return mask;
591}
592
593static inline struct sk_buff *tap_alloc_skb(struct sock *sk, size_t prepad,
594 size_t len, size_t linear,
595 int noblock, int *err)
596{
597 struct sk_buff *skb;
598
599 /* Under a page? Don't bother with paged skb. */
600 if (prepad + len < PAGE_SIZE || !linear)
601 linear = len;
602
603 skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock,
604 err, 0);
605 if (!skb)
606 return NULL;
607
608 skb_reserve(skb, prepad);
609 skb_put(skb, linear);
610 skb->data_len = len - linear;
611 skb->len += len - linear;
612
613 return skb;
614}
615
616/* Neighbour code has some assumptions on HH_DATA_MOD alignment */
617#define TAP_RESERVE HH_DATA_OFF(ETH_HLEN)
618
619/* Get packet from user space buffer */
620static ssize_t tap_get_user(struct tap_queue *q, void *msg_control,
621 struct iov_iter *from, int noblock)
622{
623 int good_linear = SKB_MAX_HEAD(TAP_RESERVE);
624 struct sk_buff *skb;
625 struct tap_dev *tap;
626 unsigned long total_len = iov_iter_count(from);
627 unsigned long len = total_len;
628 int err;
629 struct virtio_net_hdr vnet_hdr = { 0 };
630 int vnet_hdr_len = 0;
631 int copylen = 0;
632 int depth;
633 bool zerocopy = false;
634 size_t linear;
635
636 if (q->flags & IFF_VNET_HDR) {
637 vnet_hdr_len = READ_ONCE(q->vnet_hdr_sz);
638
639 err = -EINVAL;
640 if (len < vnet_hdr_len)
641 goto err;
642 len -= vnet_hdr_len;
643
644 err = -EFAULT;
645 if (!copy_from_iter_full(&vnet_hdr, sizeof(vnet_hdr), from))
646 goto err;
647 iov_iter_advance(from, vnet_hdr_len - sizeof(vnet_hdr));
648 if ((vnet_hdr.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) &&
649 tap16_to_cpu(q, vnet_hdr.csum_start) +
650 tap16_to_cpu(q, vnet_hdr.csum_offset) + 2 >
651 tap16_to_cpu(q, vnet_hdr.hdr_len))
652 vnet_hdr.hdr_len = cpu_to_tap16(q,
653 tap16_to_cpu(q, vnet_hdr.csum_start) +
654 tap16_to_cpu(q, vnet_hdr.csum_offset) + 2);
655 err = -EINVAL;
656 if (tap16_to_cpu(q, vnet_hdr.hdr_len) > len)
657 goto err;
658 }
659
660 err = -EINVAL;
661 if (unlikely(len < ETH_HLEN))
662 goto err;
663
664 if (msg_control && sock_flag(&q->sk, SOCK_ZEROCOPY)) {
665 struct iov_iter i;
666
667 copylen = vnet_hdr.hdr_len ?
668 tap16_to_cpu(q, vnet_hdr.hdr_len) : GOODCOPY_LEN;
669 if (copylen > good_linear)
670 copylen = good_linear;
671 else if (copylen < ETH_HLEN)
672 copylen = ETH_HLEN;
673 linear = copylen;
674 i = *from;
675 iov_iter_advance(&i, copylen);
676 if (iov_iter_npages(&i, INT_MAX) <= MAX_SKB_FRAGS)
677 zerocopy = true;
678 }
679
680 if (!zerocopy) {
681 copylen = len;
682 linear = tap16_to_cpu(q, vnet_hdr.hdr_len);
683 if (linear > good_linear)
684 linear = good_linear;
685 else if (linear < ETH_HLEN)
686 linear = ETH_HLEN;
687 }
688
689 skb = tap_alloc_skb(&q->sk, TAP_RESERVE, copylen,
690 linear, noblock, &err);
691 if (!skb)
692 goto err;
693
694 if (zerocopy)
695 err = zerocopy_sg_from_iter(skb, from);
696 else
697 err = skb_copy_datagram_from_iter(skb, 0, from, len);
698
699 if (err)
700 goto err_kfree;
701
702 skb_set_network_header(skb, ETH_HLEN);
703 skb_reset_mac_header(skb);
704 skb->protocol = eth_hdr(skb)->h_proto;
705
706 if (vnet_hdr_len) {
707 err = virtio_net_hdr_to_skb(skb, &vnet_hdr,
708 tap_is_little_endian(q));
709 if (err)
710 goto err_kfree;
711 }
712
713 skb_probe_transport_header(skb);
714
715 /* Move network header to the right position for VLAN tagged packets */
716 if (eth_type_vlan(skb->protocol) &&
717 __vlan_get_protocol(skb, skb->protocol, &depth) != 0)
718 skb_set_network_header(skb, depth);
719
720 rcu_read_lock();
721 tap = rcu_dereference(q->tap);
722 /* copy skb_ubuf_info for callback when skb has no error */
723 if (zerocopy) {
724 skb_zcopy_init(skb, msg_control);
725 } else if (msg_control) {
726 struct ubuf_info *uarg = msg_control;
727 uarg->callback(NULL, uarg, false);
728 }
729
730 if (tap) {
731 skb->dev = tap->dev;
732 dev_queue_xmit(skb);
733 } else {
734 kfree_skb(skb);
735 }
736 rcu_read_unlock();
737
738 return total_len;
739
740err_kfree:
741 kfree_skb(skb);
742
743err:
744 rcu_read_lock();
745 tap = rcu_dereference(q->tap);
746 if (tap && tap->count_tx_dropped)
747 tap->count_tx_dropped(tap);
748 rcu_read_unlock();
749
750 return err;
751}
752
753static ssize_t tap_write_iter(struct kiocb *iocb, struct iov_iter *from)
754{
755 struct file *file = iocb->ki_filp;
756 struct tap_queue *q = file->private_data;
757
758 return tap_get_user(q, NULL, from, file->f_flags & O_NONBLOCK);
759}
760
761/* Put packet to the user space buffer */
762static ssize_t tap_put_user(struct tap_queue *q,
763 const struct sk_buff *skb,
764 struct iov_iter *iter)
765{
766 int ret;
767 int vnet_hdr_len = 0;
768 int vlan_offset = 0;
769 int total;
770
771 if (q->flags & IFF_VNET_HDR) {
772 int vlan_hlen = skb_vlan_tag_present(skb) ? VLAN_HLEN : 0;
773 struct virtio_net_hdr vnet_hdr;
774
775 vnet_hdr_len = READ_ONCE(q->vnet_hdr_sz);
776 if (iov_iter_count(iter) < vnet_hdr_len)
777 return -EINVAL;
778
779 if (virtio_net_hdr_from_skb(skb, &vnet_hdr,
780 tap_is_little_endian(q), true,
781 vlan_hlen))
782 BUG();
783
784 if (copy_to_iter(&vnet_hdr, sizeof(vnet_hdr), iter) !=
785 sizeof(vnet_hdr))
786 return -EFAULT;
787
788 iov_iter_advance(iter, vnet_hdr_len - sizeof(vnet_hdr));
789 }
790 total = vnet_hdr_len;
791 total += skb->len;
792
793 if (skb_vlan_tag_present(skb)) {
794 struct {
795 __be16 h_vlan_proto;
796 __be16 h_vlan_TCI;
797 } veth;
798 veth.h_vlan_proto = skb->vlan_proto;
799 veth.h_vlan_TCI = htons(skb_vlan_tag_get(skb));
800
801 vlan_offset = offsetof(struct vlan_ethhdr, h_vlan_proto);
802 total += VLAN_HLEN;
803
804 ret = skb_copy_datagram_iter(skb, 0, iter, vlan_offset);
805 if (ret || !iov_iter_count(iter))
806 goto done;
807
808 ret = copy_to_iter(&veth, sizeof(veth), iter);
809 if (ret != sizeof(veth) || !iov_iter_count(iter))
810 goto done;
811 }
812
813 ret = skb_copy_datagram_iter(skb, vlan_offset, iter,
814 skb->len - vlan_offset);
815
816done:
817 return ret ? ret : total;
818}
819
820static ssize_t tap_do_read(struct tap_queue *q,
821 struct iov_iter *to,
822 int noblock, struct sk_buff *skb)
823{
824 DEFINE_WAIT(wait);
825 ssize_t ret = 0;
826
827 if (!iov_iter_count(to)) {
828 kfree_skb(skb);
829 return 0;
830 }
831
832 if (skb)
833 goto put;
834
835 while (1) {
836 if (!noblock)
837 prepare_to_wait(sk_sleep(&q->sk), &wait,
838 TASK_INTERRUPTIBLE);
839
840 /* Read frames from the queue */
841 skb = ptr_ring_consume(&q->ring);
842 if (skb)
843 break;
844 if (noblock) {
845 ret = -EAGAIN;
846 break;
847 }
848 if (signal_pending(current)) {
849 ret = -ERESTARTSYS;
850 break;
851 }
852 /* Nothing to read, let's sleep */
853 schedule();
854 }
855 if (!noblock)
856 finish_wait(sk_sleep(&q->sk), &wait);
857
858put:
859 if (skb) {
860 ret = tap_put_user(q, skb, to);
861 if (unlikely(ret < 0))
862 kfree_skb(skb);
863 else
864 consume_skb(skb);
865 }
866 return ret;
867}
868
869static ssize_t tap_read_iter(struct kiocb *iocb, struct iov_iter *to)
870{
871 struct file *file = iocb->ki_filp;
872 struct tap_queue *q = file->private_data;
873 ssize_t len = iov_iter_count(to), ret;
874
875 ret = tap_do_read(q, to, file->f_flags & O_NONBLOCK, NULL);
876 ret = min_t(ssize_t, ret, len);
877 if (ret > 0)
878 iocb->ki_pos = ret;
879 return ret;
880}
881
882static struct tap_dev *tap_get_tap_dev(struct tap_queue *q)
883{
884 struct tap_dev *tap;
885
886 ASSERT_RTNL();
887 tap = rtnl_dereference(q->tap);
888 if (tap)
889 dev_hold(tap->dev);
890
891 return tap;
892}
893
894static void tap_put_tap_dev(struct tap_dev *tap)
895{
896 dev_put(tap->dev);
897}
898
899static int tap_ioctl_set_queue(struct file *file, unsigned int flags)
900{
901 struct tap_queue *q = file->private_data;
902 struct tap_dev *tap;
903 int ret;
904
905 tap = tap_get_tap_dev(q);
906 if (!tap)
907 return -EINVAL;
908
909 if (flags & IFF_ATTACH_QUEUE)
910 ret = tap_enable_queue(tap, file, q);
911 else if (flags & IFF_DETACH_QUEUE)
912 ret = tap_disable_queue(q);
913 else
914 ret = -EINVAL;
915
916 tap_put_tap_dev(tap);
917 return ret;
918}
919
920static int set_offload(struct tap_queue *q, unsigned long arg)
921{
922 struct tap_dev *tap;
923 netdev_features_t features;
924 netdev_features_t feature_mask = 0;
925
926 tap = rtnl_dereference(q->tap);
927 if (!tap)
928 return -ENOLINK;
929
930 features = tap->dev->features;
931
932 if (arg & TUN_F_CSUM) {
933 feature_mask = NETIF_F_HW_CSUM;
934
935 if (arg & (TUN_F_TSO4 | TUN_F_TSO6)) {
936 if (arg & TUN_F_TSO_ECN)
937 feature_mask |= NETIF_F_TSO_ECN;
938 if (arg & TUN_F_TSO4)
939 feature_mask |= NETIF_F_TSO;
940 if (arg & TUN_F_TSO6)
941 feature_mask |= NETIF_F_TSO6;
942 }
943 }
944
945 /* tun/tap driver inverts the usage for TSO offloads, where
946 * setting the TSO bit means that the userspace wants to
947 * accept TSO frames and turning it off means that user space
948 * does not support TSO.
949 * For tap, we have to invert it to mean the same thing.
950 * When user space turns off TSO, we turn off GSO/LRO so that
951 * user-space will not receive TSO frames.
952 */
953 if (feature_mask & (NETIF_F_TSO | NETIF_F_TSO6))
954 features |= RX_OFFLOADS;
955 else
956 features &= ~RX_OFFLOADS;
957
958 /* tap_features are the same as features on tun/tap and
959 * reflect user expectations.
960 */
961 tap->tap_features = feature_mask;
962 if (tap->update_features)
963 tap->update_features(tap, features);
964
965 return 0;
966}
967
968/*
969 * provide compatibility with generic tun/tap interface
970 */
971static long tap_ioctl(struct file *file, unsigned int cmd,
972 unsigned long arg)
973{
974 struct tap_queue *q = file->private_data;
975 struct tap_dev *tap;
976 void __user *argp = (void __user *)arg;
977 struct ifreq __user *ifr = argp;
978 unsigned int __user *up = argp;
979 unsigned short u;
980 int __user *sp = argp;
981 struct sockaddr sa;
982 int s;
983 int ret;
984
985 switch (cmd) {
986 case TUNSETIFF:
987 /* ignore the name, just look at flags */
988 if (get_user(u, &ifr->ifr_flags))
989 return -EFAULT;
990
991 ret = 0;
992 if ((u & ~TAP_IFFEATURES) != (IFF_NO_PI | IFF_TAP))
993 ret = -EINVAL;
994 else
995 q->flags = (q->flags & ~TAP_IFFEATURES) | u;
996
997 return ret;
998
999 case TUNGETIFF:
1000 rtnl_lock();
1001 tap = tap_get_tap_dev(q);
1002 if (!tap) {
1003 rtnl_unlock();
1004 return -ENOLINK;
1005 }
1006
1007 ret = 0;
1008 u = q->flags;
1009 if (copy_to_user(&ifr->ifr_name, tap->dev->name, IFNAMSIZ) ||
1010 put_user(u, &ifr->ifr_flags))
1011 ret = -EFAULT;
1012 tap_put_tap_dev(tap);
1013 rtnl_unlock();
1014 return ret;
1015
1016 case TUNSETQUEUE:
1017 if (get_user(u, &ifr->ifr_flags))
1018 return -EFAULT;
1019 rtnl_lock();
1020 ret = tap_ioctl_set_queue(file, u);
1021 rtnl_unlock();
1022 return ret;
1023
1024 case TUNGETFEATURES:
1025 if (put_user(IFF_TAP | IFF_NO_PI | TAP_IFFEATURES, up))
1026 return -EFAULT;
1027 return 0;
1028
1029 case TUNSETSNDBUF:
1030 if (get_user(s, sp))
1031 return -EFAULT;
1032 if (s <= 0)
1033 return -EINVAL;
1034
1035 q->sk.sk_sndbuf = s;
1036 return 0;
1037
1038 case TUNGETVNETHDRSZ:
1039 s = q->vnet_hdr_sz;
1040 if (put_user(s, sp))
1041 return -EFAULT;
1042 return 0;
1043
1044 case TUNSETVNETHDRSZ:
1045 if (get_user(s, sp))
1046 return -EFAULT;
1047 if (s < (int)sizeof(struct virtio_net_hdr))
1048 return -EINVAL;
1049
1050 q->vnet_hdr_sz = s;
1051 return 0;
1052
1053 case TUNGETVNETLE:
1054 s = !!(q->flags & TAP_VNET_LE);
1055 if (put_user(s, sp))
1056 return -EFAULT;
1057 return 0;
1058
1059 case TUNSETVNETLE:
1060 if (get_user(s, sp))
1061 return -EFAULT;
1062 if (s)
1063 q->flags |= TAP_VNET_LE;
1064 else
1065 q->flags &= ~TAP_VNET_LE;
1066 return 0;
1067
1068 case TUNGETVNETBE:
1069 return tap_get_vnet_be(q, sp);
1070
1071 case TUNSETVNETBE:
1072 return tap_set_vnet_be(q, sp);
1073
1074 case TUNSETOFFLOAD:
1075 /* let the user check for future flags */
1076 if (arg & ~(TUN_F_CSUM | TUN_F_TSO4 | TUN_F_TSO6 |
1077 TUN_F_TSO_ECN | TUN_F_UFO))
1078 return -EINVAL;
1079
1080 rtnl_lock();
1081 ret = set_offload(q, arg);
1082 rtnl_unlock();
1083 return ret;
1084
1085 case SIOCGIFHWADDR:
1086 rtnl_lock();
1087 tap = tap_get_tap_dev(q);
1088 if (!tap) {
1089 rtnl_unlock();
1090 return -ENOLINK;
1091 }
1092 ret = 0;
1093 dev_get_mac_address(&sa, dev_net(tap->dev), tap->dev->name);
1094 if (copy_to_user(&ifr->ifr_name, tap->dev->name, IFNAMSIZ) ||
1095 copy_to_user(&ifr->ifr_hwaddr, &sa, sizeof(sa)))
1096 ret = -EFAULT;
1097 tap_put_tap_dev(tap);
1098 rtnl_unlock();
1099 return ret;
1100
1101 case SIOCSIFHWADDR:
1102 if (copy_from_user(&sa, &ifr->ifr_hwaddr, sizeof(sa)))
1103 return -EFAULT;
1104 rtnl_lock();
1105 tap = tap_get_tap_dev(q);
1106 if (!tap) {
1107 rtnl_unlock();
1108 return -ENOLINK;
1109 }
1110 ret = dev_set_mac_address_user(tap->dev, &sa, NULL);
1111 tap_put_tap_dev(tap);
1112 rtnl_unlock();
1113 return ret;
1114
1115 default:
1116 return -EINVAL;
1117 }
1118}
1119
1120static const struct file_operations tap_fops = {
1121 .owner = THIS_MODULE,
1122 .open = tap_open,
1123 .release = tap_release,
1124 .read_iter = tap_read_iter,
1125 .write_iter = tap_write_iter,
1126 .poll = tap_poll,
1127 .llseek = no_llseek,
1128 .unlocked_ioctl = tap_ioctl,
1129 .compat_ioctl = compat_ptr_ioctl,
1130};
1131
1132static int tap_get_user_xdp(struct tap_queue *q, struct xdp_buff *xdp)
1133{
1134 struct tun_xdp_hdr *hdr = xdp->data_hard_start;
1135 struct virtio_net_hdr *gso = &hdr->gso;
1136 int buflen = hdr->buflen;
1137 int vnet_hdr_len = 0;
1138 struct tap_dev *tap;
1139 struct sk_buff *skb;
1140 int err, depth;
1141
1142 if (q->flags & IFF_VNET_HDR)
1143 vnet_hdr_len = READ_ONCE(q->vnet_hdr_sz);
1144
1145 skb = build_skb(xdp->data_hard_start, buflen);
1146 if (!skb) {
1147 err = -ENOMEM;
1148 goto err;
1149 }
1150
1151 skb_reserve(skb, xdp->data - xdp->data_hard_start);
1152 skb_put(skb, xdp->data_end - xdp->data);
1153
1154 skb_set_network_header(skb, ETH_HLEN);
1155 skb_reset_mac_header(skb);
1156 skb->protocol = eth_hdr(skb)->h_proto;
1157
1158 if (vnet_hdr_len) {
1159 err = virtio_net_hdr_to_skb(skb, gso, tap_is_little_endian(q));
1160 if (err)
1161 goto err_kfree;
1162 }
1163
1164 /* Move network header to the right position for VLAN tagged packets */
1165 if (eth_type_vlan(skb->protocol) &&
1166 __vlan_get_protocol(skb, skb->protocol, &depth) != 0)
1167 skb_set_network_header(skb, depth);
1168
1169 rcu_read_lock();
1170 tap = rcu_dereference(q->tap);
1171 if (tap) {
1172 skb->dev = tap->dev;
1173 skb_probe_transport_header(skb);
1174 dev_queue_xmit(skb);
1175 } else {
1176 kfree_skb(skb);
1177 }
1178 rcu_read_unlock();
1179
1180 return 0;
1181
1182err_kfree:
1183 kfree_skb(skb);
1184err:
1185 rcu_read_lock();
1186 tap = rcu_dereference(q->tap);
1187 if (tap && tap->count_tx_dropped)
1188 tap->count_tx_dropped(tap);
1189 rcu_read_unlock();
1190 return err;
1191}
1192
1193static int tap_sendmsg(struct socket *sock, struct msghdr *m,
1194 size_t total_len)
1195{
1196 struct tap_queue *q = container_of(sock, struct tap_queue, sock);
1197 struct tun_msg_ctl *ctl = m->msg_control;
1198 struct xdp_buff *xdp;
1199 int i;
1200
1201 if (ctl && (ctl->type == TUN_MSG_PTR)) {
1202 for (i = 0; i < ctl->num; i++) {
1203 xdp = &((struct xdp_buff *)ctl->ptr)[i];
1204 tap_get_user_xdp(q, xdp);
1205 }
1206 return 0;
1207 }
1208
1209 return tap_get_user(q, ctl ? ctl->ptr : NULL, &m->msg_iter,
1210 m->msg_flags & MSG_DONTWAIT);
1211}
1212
1213static int tap_recvmsg(struct socket *sock, struct msghdr *m,
1214 size_t total_len, int flags)
1215{
1216 struct tap_queue *q = container_of(sock, struct tap_queue, sock);
1217 struct sk_buff *skb = m->msg_control;
1218 int ret;
1219 if (flags & ~(MSG_DONTWAIT|MSG_TRUNC)) {
1220 kfree_skb(skb);
1221 return -EINVAL;
1222 }
1223 ret = tap_do_read(q, &m->msg_iter, flags & MSG_DONTWAIT, skb);
1224 if (ret > total_len) {
1225 m->msg_flags |= MSG_TRUNC;
1226 ret = flags & MSG_TRUNC ? ret : total_len;
1227 }
1228 return ret;
1229}
1230
1231static int tap_peek_len(struct socket *sock)
1232{
1233 struct tap_queue *q = container_of(sock, struct tap_queue,
1234 sock);
1235 return PTR_RING_PEEK_CALL(&q->ring, __skb_array_len_with_tag);
1236}
1237
1238/* Ops structure to mimic raw sockets with tun */
1239static const struct proto_ops tap_socket_ops = {
1240 .sendmsg = tap_sendmsg,
1241 .recvmsg = tap_recvmsg,
1242 .peek_len = tap_peek_len,
1243};
1244
1245/* Get an underlying socket object from tun file. Returns error unless file is
1246 * attached to a device. The returned object works like a packet socket, it
1247 * can be used for sock_sendmsg/sock_recvmsg. The caller is responsible for
1248 * holding a reference to the file for as long as the socket is in use. */
1249struct socket *tap_get_socket(struct file *file)
1250{
1251 struct tap_queue *q;
1252 if (file->f_op != &tap_fops)
1253 return ERR_PTR(-EINVAL);
1254 q = file->private_data;
1255 if (!q)
1256 return ERR_PTR(-EBADFD);
1257 return &q->sock;
1258}
1259EXPORT_SYMBOL_GPL(tap_get_socket);
1260
1261struct ptr_ring *tap_get_ptr_ring(struct file *file)
1262{
1263 struct tap_queue *q;
1264
1265 if (file->f_op != &tap_fops)
1266 return ERR_PTR(-EINVAL);
1267 q = file->private_data;
1268 if (!q)
1269 return ERR_PTR(-EBADFD);
1270 return &q->ring;
1271}
1272EXPORT_SYMBOL_GPL(tap_get_ptr_ring);
1273
1274int tap_queue_resize(struct tap_dev *tap)
1275{
1276 struct net_device *dev = tap->dev;
1277 struct tap_queue *q;
1278 struct ptr_ring **rings;
1279 int n = tap->numqueues;
1280 int ret, i = 0;
1281
1282 rings = kmalloc_array(n, sizeof(*rings), GFP_KERNEL);
1283 if (!rings)
1284 return -ENOMEM;
1285
1286 list_for_each_entry(q, &tap->queue_list, next)
1287 rings[i++] = &q->ring;
1288
1289 ret = ptr_ring_resize_multiple(rings, n,
1290 dev->tx_queue_len, GFP_KERNEL,
1291 __skb_array_destroy_skb);
1292
1293 kfree(rings);
1294 return ret;
1295}
1296EXPORT_SYMBOL_GPL(tap_queue_resize);
1297
1298static int tap_list_add(dev_t major, const char *device_name)
1299{
1300 struct major_info *tap_major;
1301
1302 tap_major = kzalloc(sizeof(*tap_major), GFP_ATOMIC);
1303 if (!tap_major)
1304 return -ENOMEM;
1305
1306 tap_major->major = MAJOR(major);
1307
1308 idr_init(&tap_major->minor_idr);
1309 spin_lock_init(&tap_major->minor_lock);
1310
1311 tap_major->device_name = device_name;
1312
1313 list_add_tail_rcu(&tap_major->next, &major_list);
1314 return 0;
1315}
1316
1317int tap_create_cdev(struct cdev *tap_cdev, dev_t *tap_major,
1318 const char *device_name, struct module *module)
1319{
1320 int err;
1321
1322 err = alloc_chrdev_region(tap_major, 0, TAP_NUM_DEVS, device_name);
1323 if (err)
1324 goto out1;
1325
1326 cdev_init(tap_cdev, &tap_fops);
1327 tap_cdev->owner = module;
1328 err = cdev_add(tap_cdev, *tap_major, TAP_NUM_DEVS);
1329 if (err)
1330 goto out2;
1331
1332 err = tap_list_add(*tap_major, device_name);
1333 if (err)
1334 goto out3;
1335
1336 return 0;
1337
1338out3:
1339 cdev_del(tap_cdev);
1340out2:
1341 unregister_chrdev_region(*tap_major, TAP_NUM_DEVS);
1342out1:
1343 return err;
1344}
1345EXPORT_SYMBOL_GPL(tap_create_cdev);
1346
1347void tap_destroy_cdev(dev_t major, struct cdev *tap_cdev)
1348{
1349 struct major_info *tap_major, *tmp;
1350
1351 cdev_del(tap_cdev);
1352 unregister_chrdev_region(major, TAP_NUM_DEVS);
1353 list_for_each_entry_safe(tap_major, tmp, &major_list, next) {
1354 if (tap_major->major == MAJOR(major)) {
1355 idr_destroy(&tap_major->minor_idr);
1356 list_del_rcu(&tap_major->next);
1357 kfree_rcu(tap_major, rcu);
1358 }
1359 }
1360}
1361EXPORT_SYMBOL_GPL(tap_destroy_cdev);
1362
1363MODULE_AUTHOR("Arnd Bergmann <arnd@arndb.de>");
1364MODULE_AUTHOR("Sainath Grandhi <sainath.grandhi@intel.com>");
1365MODULE_LICENSE("GPL");