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
tap_legacy_is_little_endian(struct tap_queue * q)33 static 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
tap_get_vnet_be(struct tap_queue * q,int __user * sp)39 static 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
tap_set_vnet_be(struct tap_queue * q,int __user * sp)49 static 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
tap_legacy_is_little_endian(struct tap_queue * q)64 static inline bool tap_legacy_is_little_endian(struct tap_queue *q)
65 {
66 return virtio_legacy_is_little_endian();
67 }
68
tap_get_vnet_be(struct tap_queue * q,int __user * argp)69 static long tap_get_vnet_be(struct tap_queue *q, int __user *argp)
70 {
71 return -EINVAL;
72 }
73
tap_set_vnet_be(struct tap_queue * q,int __user * argp)74 static 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
tap_is_little_endian(struct tap_queue * q)80 static 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
tap16_to_cpu(struct tap_queue * q,__virtio16 val)86 static 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
cpu_to_tap16(struct tap_queue * q,u16 val)91 static 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
96 static 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
104 static LIST_HEAD(major_list);
105
106 struct 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
117 static 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
tap_dev_get_rcu(const struct net_device * dev)122 static 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
tap_enable_queue(struct tap_dev * tap,struct file * file,struct tap_queue * q)144 static 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++;
160 out:
161 return err;
162 }
163
164 /* Requires RTNL */
tap_set_queue(struct tap_dev * tap,struct file * file,struct tap_queue * q)165 static 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
tap_disable_queue(struct tap_queue * q)187 static 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 */
tap_put_queue(struct tap_queue * q)222 static 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 */
tap_get_queue(struct tap_dev * tap,struct sk_buff * skb)252 static 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
287 single:
288 queue = rcu_dereference(tap->taps[0]);
289 out:
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 */
tap_del_queues(struct tap_dev * tap)298 void 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 }
316 EXPORT_SYMBOL_GPL(tap_del_queues);
317
tap_handle_frame(struct sk_buff ** pskb)318 rx_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 enum skb_drop_reason drop_reason;
326
327 tap = tap_dev_get_rcu(dev);
328 if (!tap)
329 return RX_HANDLER_PASS;
330
331 q = tap_get_queue(tap, skb);
332 if (!q)
333 return RX_HANDLER_PASS;
334
335 skb_push(skb, ETH_HLEN);
336
337 /* Apply the forward feature mask so that we perform segmentation
338 * according to users wishes. This only works if VNET_HDR is
339 * enabled.
340 */
341 if (q->flags & IFF_VNET_HDR)
342 features |= tap->tap_features;
343 if (netif_needs_gso(skb, features)) {
344 struct sk_buff *segs = __skb_gso_segment(skb, features, false);
345 struct sk_buff *next;
346
347 if (IS_ERR(segs)) {
348 drop_reason = SKB_DROP_REASON_SKB_GSO_SEG;
349 goto drop;
350 }
351
352 if (!segs) {
353 if (ptr_ring_produce(&q->ring, skb)) {
354 drop_reason = SKB_DROP_REASON_FULL_RING;
355 goto drop;
356 }
357 goto wake_up;
358 }
359
360 consume_skb(skb);
361 skb_list_walk_safe(segs, skb, next) {
362 skb_mark_not_on_list(skb);
363 if (ptr_ring_produce(&q->ring, skb)) {
364 drop_reason = SKB_DROP_REASON_FULL_RING;
365 kfree_skb_reason(skb, drop_reason);
366 kfree_skb_list_reason(next, drop_reason);
367 break;
368 }
369 }
370 } else {
371 /* If we receive a partial checksum and the tap side
372 * doesn't support checksum offload, compute the checksum.
373 * Note: it doesn't matter which checksum feature to
374 * check, we either support them all or none.
375 */
376 if (skb->ip_summed == CHECKSUM_PARTIAL &&
377 !(features & NETIF_F_CSUM_MASK) &&
378 skb_checksum_help(skb)) {
379 drop_reason = SKB_DROP_REASON_SKB_CSUM;
380 goto drop;
381 }
382 if (ptr_ring_produce(&q->ring, skb)) {
383 drop_reason = SKB_DROP_REASON_FULL_RING;
384 goto drop;
385 }
386 }
387
388 wake_up:
389 wake_up_interruptible_poll(sk_sleep(&q->sk), EPOLLIN | EPOLLRDNORM | EPOLLRDBAND);
390 return RX_HANDLER_CONSUMED;
391
392 drop:
393 /* Count errors/drops only here, thus don't care about args. */
394 if (tap->count_rx_dropped)
395 tap->count_rx_dropped(tap);
396 kfree_skb_reason(skb, drop_reason);
397 return RX_HANDLER_CONSUMED;
398 }
399 EXPORT_SYMBOL_GPL(tap_handle_frame);
400
tap_get_major(int major)401 static struct major_info *tap_get_major(int major)
402 {
403 struct major_info *tap_major;
404
405 list_for_each_entry_rcu(tap_major, &major_list, next) {
406 if (tap_major->major == major)
407 return tap_major;
408 }
409
410 return NULL;
411 }
412
tap_get_minor(dev_t major,struct tap_dev * tap)413 int tap_get_minor(dev_t major, struct tap_dev *tap)
414 {
415 int retval = -ENOMEM;
416 struct major_info *tap_major;
417
418 rcu_read_lock();
419 tap_major = tap_get_major(MAJOR(major));
420 if (!tap_major) {
421 retval = -EINVAL;
422 goto unlock;
423 }
424
425 spin_lock(&tap_major->minor_lock);
426 retval = idr_alloc(&tap_major->minor_idr, tap, 1, TAP_NUM_DEVS, GFP_ATOMIC);
427 if (retval >= 0) {
428 tap->minor = retval;
429 } else if (retval == -ENOSPC) {
430 netdev_err(tap->dev, "Too many tap devices\n");
431 retval = -EINVAL;
432 }
433 spin_unlock(&tap_major->minor_lock);
434
435 unlock:
436 rcu_read_unlock();
437 return retval < 0 ? retval : 0;
438 }
439 EXPORT_SYMBOL_GPL(tap_get_minor);
440
tap_free_minor(dev_t major,struct tap_dev * tap)441 void tap_free_minor(dev_t major, struct tap_dev *tap)
442 {
443 struct major_info *tap_major;
444
445 rcu_read_lock();
446 tap_major = tap_get_major(MAJOR(major));
447 if (!tap_major) {
448 goto unlock;
449 }
450
451 spin_lock(&tap_major->minor_lock);
452 if (tap->minor) {
453 idr_remove(&tap_major->minor_idr, tap->minor);
454 tap->minor = 0;
455 }
456 spin_unlock(&tap_major->minor_lock);
457
458 unlock:
459 rcu_read_unlock();
460 }
461 EXPORT_SYMBOL_GPL(tap_free_minor);
462
dev_get_by_tap_file(int major,int minor)463 static struct tap_dev *dev_get_by_tap_file(int major, int minor)
464 {
465 struct net_device *dev = NULL;
466 struct tap_dev *tap;
467 struct major_info *tap_major;
468
469 rcu_read_lock();
470 tap_major = tap_get_major(major);
471 if (!tap_major) {
472 tap = NULL;
473 goto unlock;
474 }
475
476 spin_lock(&tap_major->minor_lock);
477 tap = idr_find(&tap_major->minor_idr, minor);
478 if (tap) {
479 dev = tap->dev;
480 dev_hold(dev);
481 }
482 spin_unlock(&tap_major->minor_lock);
483
484 unlock:
485 rcu_read_unlock();
486 return tap;
487 }
488
tap_sock_write_space(struct sock * sk)489 static void tap_sock_write_space(struct sock *sk)
490 {
491 wait_queue_head_t *wqueue;
492
493 if (!sock_writeable(sk) ||
494 !test_and_clear_bit(SOCKWQ_ASYNC_NOSPACE, &sk->sk_socket->flags))
495 return;
496
497 wqueue = sk_sleep(sk);
498 if (wqueue && waitqueue_active(wqueue))
499 wake_up_interruptible_poll(wqueue, EPOLLOUT | EPOLLWRNORM | EPOLLWRBAND);
500 }
501
tap_sock_destruct(struct sock * sk)502 static void tap_sock_destruct(struct sock *sk)
503 {
504 struct tap_queue *q = container_of(sk, struct tap_queue, sk);
505
506 ptr_ring_cleanup(&q->ring, __skb_array_destroy_skb);
507 }
508
tap_open(struct inode * inode,struct file * file)509 static int tap_open(struct inode *inode, struct file *file)
510 {
511 struct net *net = current->nsproxy->net_ns;
512 struct tap_dev *tap;
513 struct tap_queue *q;
514 int err = -ENODEV;
515
516 rtnl_lock();
517 tap = dev_get_by_tap_file(imajor(inode), iminor(inode));
518 if (!tap)
519 goto err;
520
521 err = -ENOMEM;
522 q = (struct tap_queue *)sk_alloc(net, AF_UNSPEC, GFP_KERNEL,
523 &tap_proto, 0);
524 if (!q)
525 goto err;
526 if (ptr_ring_init(&q->ring, tap->dev->tx_queue_len, GFP_KERNEL)) {
527 sk_free(&q->sk);
528 goto err;
529 }
530
531 init_waitqueue_head(&q->sock.wq.wait);
532 q->sock.type = SOCK_RAW;
533 q->sock.state = SS_CONNECTED;
534 q->sock.file = file;
535 q->sock.ops = &tap_socket_ops;
536 sock_init_data(&q->sock, &q->sk);
537 q->sk.sk_write_space = tap_sock_write_space;
538 q->sk.sk_destruct = tap_sock_destruct;
539 q->flags = IFF_VNET_HDR | IFF_NO_PI | IFF_TAP;
540 q->vnet_hdr_sz = sizeof(struct virtio_net_hdr);
541
542 /*
543 * so far only KVM virtio_net uses tap, enable zero copy between
544 * guest kernel and host kernel when lower device supports zerocopy
545 *
546 * The macvlan supports zerocopy iff the lower device supports zero
547 * copy so we don't have to look at the lower device directly.
548 */
549 if ((tap->dev->features & NETIF_F_HIGHDMA) && (tap->dev->features & NETIF_F_SG))
550 sock_set_flag(&q->sk, SOCK_ZEROCOPY);
551
552 err = tap_set_queue(tap, file, q);
553 if (err) {
554 /* tap_sock_destruct() will take care of freeing ptr_ring */
555 goto err_put;
556 }
557
558 dev_put(tap->dev);
559
560 rtnl_unlock();
561 return err;
562
563 err_put:
564 sock_put(&q->sk);
565 err:
566 if (tap)
567 dev_put(tap->dev);
568
569 rtnl_unlock();
570 return err;
571 }
572
tap_release(struct inode * inode,struct file * file)573 static int tap_release(struct inode *inode, struct file *file)
574 {
575 struct tap_queue *q = file->private_data;
576 tap_put_queue(q);
577 return 0;
578 }
579
tap_poll(struct file * file,poll_table * wait)580 static __poll_t tap_poll(struct file *file, poll_table *wait)
581 {
582 struct tap_queue *q = file->private_data;
583 __poll_t mask = EPOLLERR;
584
585 if (!q)
586 goto out;
587
588 mask = 0;
589 poll_wait(file, &q->sock.wq.wait, wait);
590
591 if (!ptr_ring_empty(&q->ring))
592 mask |= EPOLLIN | EPOLLRDNORM;
593
594 if (sock_writeable(&q->sk) ||
595 (!test_and_set_bit(SOCKWQ_ASYNC_NOSPACE, &q->sock.flags) &&
596 sock_writeable(&q->sk)))
597 mask |= EPOLLOUT | EPOLLWRNORM;
598
599 out:
600 return mask;
601 }
602
tap_alloc_skb(struct sock * sk,size_t prepad,size_t len,size_t linear,int noblock,int * err)603 static inline struct sk_buff *tap_alloc_skb(struct sock *sk, size_t prepad,
604 size_t len, size_t linear,
605 int noblock, int *err)
606 {
607 struct sk_buff *skb;
608
609 /* Under a page? Don't bother with paged skb. */
610 if (prepad + len < PAGE_SIZE || !linear)
611 linear = len;
612
613 skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock,
614 err, 0);
615 if (!skb)
616 return NULL;
617
618 skb_reserve(skb, prepad);
619 skb_put(skb, linear);
620 skb->data_len = len - linear;
621 skb->len += len - linear;
622
623 return skb;
624 }
625
626 /* Neighbour code has some assumptions on HH_DATA_MOD alignment */
627 #define TAP_RESERVE HH_DATA_OFF(ETH_HLEN)
628
629 /* Get packet from user space buffer */
tap_get_user(struct tap_queue * q,void * msg_control,struct iov_iter * from,int noblock)630 static ssize_t tap_get_user(struct tap_queue *q, void *msg_control,
631 struct iov_iter *from, int noblock)
632 {
633 int good_linear = SKB_MAX_HEAD(TAP_RESERVE);
634 struct sk_buff *skb;
635 struct tap_dev *tap;
636 unsigned long total_len = iov_iter_count(from);
637 unsigned long len = total_len;
638 int err;
639 struct virtio_net_hdr vnet_hdr = { 0 };
640 int vnet_hdr_len = 0;
641 int copylen = 0;
642 int depth;
643 bool zerocopy = false;
644 size_t linear;
645 enum skb_drop_reason drop_reason;
646
647 if (q->flags & IFF_VNET_HDR) {
648 vnet_hdr_len = READ_ONCE(q->vnet_hdr_sz);
649
650 err = -EINVAL;
651 if (len < vnet_hdr_len)
652 goto err;
653 len -= vnet_hdr_len;
654
655 err = -EFAULT;
656 if (!copy_from_iter_full(&vnet_hdr, sizeof(vnet_hdr), from))
657 goto err;
658 iov_iter_advance(from, vnet_hdr_len - sizeof(vnet_hdr));
659 if ((vnet_hdr.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) &&
660 tap16_to_cpu(q, vnet_hdr.csum_start) +
661 tap16_to_cpu(q, vnet_hdr.csum_offset) + 2 >
662 tap16_to_cpu(q, vnet_hdr.hdr_len))
663 vnet_hdr.hdr_len = cpu_to_tap16(q,
664 tap16_to_cpu(q, vnet_hdr.csum_start) +
665 tap16_to_cpu(q, vnet_hdr.csum_offset) + 2);
666 err = -EINVAL;
667 if (tap16_to_cpu(q, vnet_hdr.hdr_len) > len)
668 goto err;
669 }
670
671 err = -EINVAL;
672 if (unlikely(len < ETH_HLEN))
673 goto err;
674
675 if (msg_control && sock_flag(&q->sk, SOCK_ZEROCOPY)) {
676 struct iov_iter i;
677
678 copylen = vnet_hdr.hdr_len ?
679 tap16_to_cpu(q, vnet_hdr.hdr_len) : GOODCOPY_LEN;
680 if (copylen > good_linear)
681 copylen = good_linear;
682 else if (copylen < ETH_HLEN)
683 copylen = ETH_HLEN;
684 linear = copylen;
685 i = *from;
686 iov_iter_advance(&i, copylen);
687 if (iov_iter_npages(&i, INT_MAX) <= MAX_SKB_FRAGS)
688 zerocopy = true;
689 }
690
691 if (!zerocopy) {
692 copylen = len;
693 linear = tap16_to_cpu(q, vnet_hdr.hdr_len);
694 if (linear > good_linear)
695 linear = good_linear;
696 else if (linear < ETH_HLEN)
697 linear = ETH_HLEN;
698 }
699
700 skb = tap_alloc_skb(&q->sk, TAP_RESERVE, copylen,
701 linear, noblock, &err);
702 if (!skb)
703 goto err;
704
705 if (zerocopy)
706 err = zerocopy_sg_from_iter(skb, from);
707 else
708 err = skb_copy_datagram_from_iter(skb, 0, from, len);
709
710 if (err) {
711 drop_reason = SKB_DROP_REASON_SKB_UCOPY_FAULT;
712 goto err_kfree;
713 }
714
715 skb_set_network_header(skb, ETH_HLEN);
716 skb_reset_mac_header(skb);
717 skb->protocol = eth_hdr(skb)->h_proto;
718
719 rcu_read_lock();
720 tap = rcu_dereference(q->tap);
721 if (!tap) {
722 kfree_skb(skb);
723 rcu_read_unlock();
724 return total_len;
725 }
726 skb->dev = tap->dev;
727
728 if (vnet_hdr_len) {
729 err = virtio_net_hdr_to_skb(skb, &vnet_hdr,
730 tap_is_little_endian(q));
731 if (err) {
732 rcu_read_unlock();
733 drop_reason = SKB_DROP_REASON_DEV_HDR;
734 goto err_kfree;
735 }
736 }
737
738 skb_probe_transport_header(skb);
739
740 /* Move network header to the right position for VLAN tagged packets */
741 if (eth_type_vlan(skb->protocol) &&
742 __vlan_get_protocol(skb, skb->protocol, &depth) != 0)
743 skb_set_network_header(skb, depth);
744
745 /* copy skb_ubuf_info for callback when skb has no error */
746 if (zerocopy) {
747 skb_zcopy_init(skb, msg_control);
748 } else if (msg_control) {
749 struct ubuf_info *uarg = msg_control;
750 uarg->callback(NULL, uarg, false);
751 }
752
753 dev_queue_xmit(skb);
754 rcu_read_unlock();
755 return total_len;
756
757 err_kfree:
758 kfree_skb_reason(skb, drop_reason);
759
760 err:
761 rcu_read_lock();
762 tap = rcu_dereference(q->tap);
763 if (tap && tap->count_tx_dropped)
764 tap->count_tx_dropped(tap);
765 rcu_read_unlock();
766
767 return err;
768 }
769
tap_write_iter(struct kiocb * iocb,struct iov_iter * from)770 static ssize_t tap_write_iter(struct kiocb *iocb, struct iov_iter *from)
771 {
772 struct file *file = iocb->ki_filp;
773 struct tap_queue *q = file->private_data;
774
775 return tap_get_user(q, NULL, from, file->f_flags & O_NONBLOCK);
776 }
777
778 /* Put packet to the user space buffer */
tap_put_user(struct tap_queue * q,const struct sk_buff * skb,struct iov_iter * iter)779 static ssize_t tap_put_user(struct tap_queue *q,
780 const struct sk_buff *skb,
781 struct iov_iter *iter)
782 {
783 int ret;
784 int vnet_hdr_len = 0;
785 int vlan_offset = 0;
786 int total;
787
788 if (q->flags & IFF_VNET_HDR) {
789 int vlan_hlen = skb_vlan_tag_present(skb) ? VLAN_HLEN : 0;
790 struct virtio_net_hdr vnet_hdr;
791
792 vnet_hdr_len = READ_ONCE(q->vnet_hdr_sz);
793 if (iov_iter_count(iter) < vnet_hdr_len)
794 return -EINVAL;
795
796 if (virtio_net_hdr_from_skb(skb, &vnet_hdr,
797 tap_is_little_endian(q), true,
798 vlan_hlen))
799 BUG();
800
801 if (copy_to_iter(&vnet_hdr, sizeof(vnet_hdr), iter) !=
802 sizeof(vnet_hdr))
803 return -EFAULT;
804
805 iov_iter_advance(iter, vnet_hdr_len - sizeof(vnet_hdr));
806 }
807 total = vnet_hdr_len;
808 total += skb->len;
809
810 if (skb_vlan_tag_present(skb)) {
811 struct {
812 __be16 h_vlan_proto;
813 __be16 h_vlan_TCI;
814 } veth;
815 veth.h_vlan_proto = skb->vlan_proto;
816 veth.h_vlan_TCI = htons(skb_vlan_tag_get(skb));
817
818 vlan_offset = offsetof(struct vlan_ethhdr, h_vlan_proto);
819 total += VLAN_HLEN;
820
821 ret = skb_copy_datagram_iter(skb, 0, iter, vlan_offset);
822 if (ret || !iov_iter_count(iter))
823 goto done;
824
825 ret = copy_to_iter(&veth, sizeof(veth), iter);
826 if (ret != sizeof(veth) || !iov_iter_count(iter))
827 goto done;
828 }
829
830 ret = skb_copy_datagram_iter(skb, vlan_offset, iter,
831 skb->len - vlan_offset);
832
833 done:
834 return ret ? ret : total;
835 }
836
tap_do_read(struct tap_queue * q,struct iov_iter * to,int noblock,struct sk_buff * skb)837 static ssize_t tap_do_read(struct tap_queue *q,
838 struct iov_iter *to,
839 int noblock, struct sk_buff *skb)
840 {
841 DEFINE_WAIT(wait);
842 ssize_t ret = 0;
843
844 if (!iov_iter_count(to)) {
845 kfree_skb(skb);
846 return 0;
847 }
848
849 if (skb)
850 goto put;
851
852 while (1) {
853 if (!noblock)
854 prepare_to_wait(sk_sleep(&q->sk), &wait,
855 TASK_INTERRUPTIBLE);
856
857 /* Read frames from the queue */
858 skb = ptr_ring_consume(&q->ring);
859 if (skb)
860 break;
861 if (noblock) {
862 ret = -EAGAIN;
863 break;
864 }
865 if (signal_pending(current)) {
866 ret = -ERESTARTSYS;
867 break;
868 }
869 /* Nothing to read, let's sleep */
870 schedule();
871 }
872 if (!noblock)
873 finish_wait(sk_sleep(&q->sk), &wait);
874
875 put:
876 if (skb) {
877 ret = tap_put_user(q, skb, to);
878 if (unlikely(ret < 0))
879 kfree_skb(skb);
880 else
881 consume_skb(skb);
882 }
883 return ret;
884 }
885
tap_read_iter(struct kiocb * iocb,struct iov_iter * to)886 static ssize_t tap_read_iter(struct kiocb *iocb, struct iov_iter *to)
887 {
888 struct file *file = iocb->ki_filp;
889 struct tap_queue *q = file->private_data;
890 ssize_t len = iov_iter_count(to), ret;
891
892 ret = tap_do_read(q, to, file->f_flags & O_NONBLOCK, NULL);
893 ret = min_t(ssize_t, ret, len);
894 if (ret > 0)
895 iocb->ki_pos = ret;
896 return ret;
897 }
898
tap_get_tap_dev(struct tap_queue * q)899 static struct tap_dev *tap_get_tap_dev(struct tap_queue *q)
900 {
901 struct tap_dev *tap;
902
903 ASSERT_RTNL();
904 tap = rtnl_dereference(q->tap);
905 if (tap)
906 dev_hold(tap->dev);
907
908 return tap;
909 }
910
tap_put_tap_dev(struct tap_dev * tap)911 static void tap_put_tap_dev(struct tap_dev *tap)
912 {
913 dev_put(tap->dev);
914 }
915
tap_ioctl_set_queue(struct file * file,unsigned int flags)916 static int tap_ioctl_set_queue(struct file *file, unsigned int flags)
917 {
918 struct tap_queue *q = file->private_data;
919 struct tap_dev *tap;
920 int ret;
921
922 tap = tap_get_tap_dev(q);
923 if (!tap)
924 return -EINVAL;
925
926 if (flags & IFF_ATTACH_QUEUE)
927 ret = tap_enable_queue(tap, file, q);
928 else if (flags & IFF_DETACH_QUEUE)
929 ret = tap_disable_queue(q);
930 else
931 ret = -EINVAL;
932
933 tap_put_tap_dev(tap);
934 return ret;
935 }
936
set_offload(struct tap_queue * q,unsigned long arg)937 static int set_offload(struct tap_queue *q, unsigned long arg)
938 {
939 struct tap_dev *tap;
940 netdev_features_t features;
941 netdev_features_t feature_mask = 0;
942
943 tap = rtnl_dereference(q->tap);
944 if (!tap)
945 return -ENOLINK;
946
947 features = tap->dev->features;
948
949 if (arg & TUN_F_CSUM) {
950 feature_mask = NETIF_F_HW_CSUM;
951
952 if (arg & (TUN_F_TSO4 | TUN_F_TSO6)) {
953 if (arg & TUN_F_TSO_ECN)
954 feature_mask |= NETIF_F_TSO_ECN;
955 if (arg & TUN_F_TSO4)
956 feature_mask |= NETIF_F_TSO;
957 if (arg & TUN_F_TSO6)
958 feature_mask |= NETIF_F_TSO6;
959 }
960 }
961
962 /* tun/tap driver inverts the usage for TSO offloads, where
963 * setting the TSO bit means that the userspace wants to
964 * accept TSO frames and turning it off means that user space
965 * does not support TSO.
966 * For tap, we have to invert it to mean the same thing.
967 * When user space turns off TSO, we turn off GSO/LRO so that
968 * user-space will not receive TSO frames.
969 */
970 if (feature_mask & (NETIF_F_TSO | NETIF_F_TSO6))
971 features |= RX_OFFLOADS;
972 else
973 features &= ~RX_OFFLOADS;
974
975 /* tap_features are the same as features on tun/tap and
976 * reflect user expectations.
977 */
978 tap->tap_features = feature_mask;
979 if (tap->update_features)
980 tap->update_features(tap, features);
981
982 return 0;
983 }
984
985 /*
986 * provide compatibility with generic tun/tap interface
987 */
tap_ioctl(struct file * file,unsigned int cmd,unsigned long arg)988 static long tap_ioctl(struct file *file, unsigned int cmd,
989 unsigned long arg)
990 {
991 struct tap_queue *q = file->private_data;
992 struct tap_dev *tap;
993 void __user *argp = (void __user *)arg;
994 struct ifreq __user *ifr = argp;
995 unsigned int __user *up = argp;
996 unsigned short u;
997 int __user *sp = argp;
998 struct sockaddr sa;
999 int s;
1000 int ret;
1001
1002 switch (cmd) {
1003 case TUNSETIFF:
1004 /* ignore the name, just look at flags */
1005 if (get_user(u, &ifr->ifr_flags))
1006 return -EFAULT;
1007
1008 ret = 0;
1009 if ((u & ~TAP_IFFEATURES) != (IFF_NO_PI | IFF_TAP))
1010 ret = -EINVAL;
1011 else
1012 q->flags = (q->flags & ~TAP_IFFEATURES) | u;
1013
1014 return ret;
1015
1016 case TUNGETIFF:
1017 rtnl_lock();
1018 tap = tap_get_tap_dev(q);
1019 if (!tap) {
1020 rtnl_unlock();
1021 return -ENOLINK;
1022 }
1023
1024 ret = 0;
1025 u = q->flags;
1026 if (copy_to_user(&ifr->ifr_name, tap->dev->name, IFNAMSIZ) ||
1027 put_user(u, &ifr->ifr_flags))
1028 ret = -EFAULT;
1029 tap_put_tap_dev(tap);
1030 rtnl_unlock();
1031 return ret;
1032
1033 case TUNSETQUEUE:
1034 if (get_user(u, &ifr->ifr_flags))
1035 return -EFAULT;
1036 rtnl_lock();
1037 ret = tap_ioctl_set_queue(file, u);
1038 rtnl_unlock();
1039 return ret;
1040
1041 case TUNGETFEATURES:
1042 if (put_user(IFF_TAP | IFF_NO_PI | TAP_IFFEATURES, up))
1043 return -EFAULT;
1044 return 0;
1045
1046 case TUNSETSNDBUF:
1047 if (get_user(s, sp))
1048 return -EFAULT;
1049 if (s <= 0)
1050 return -EINVAL;
1051
1052 q->sk.sk_sndbuf = s;
1053 return 0;
1054
1055 case TUNGETVNETHDRSZ:
1056 s = q->vnet_hdr_sz;
1057 if (put_user(s, sp))
1058 return -EFAULT;
1059 return 0;
1060
1061 case TUNSETVNETHDRSZ:
1062 if (get_user(s, sp))
1063 return -EFAULT;
1064 if (s < (int)sizeof(struct virtio_net_hdr))
1065 return -EINVAL;
1066
1067 q->vnet_hdr_sz = s;
1068 return 0;
1069
1070 case TUNGETVNETLE:
1071 s = !!(q->flags & TAP_VNET_LE);
1072 if (put_user(s, sp))
1073 return -EFAULT;
1074 return 0;
1075
1076 case TUNSETVNETLE:
1077 if (get_user(s, sp))
1078 return -EFAULT;
1079 if (s)
1080 q->flags |= TAP_VNET_LE;
1081 else
1082 q->flags &= ~TAP_VNET_LE;
1083 return 0;
1084
1085 case TUNGETVNETBE:
1086 return tap_get_vnet_be(q, sp);
1087
1088 case TUNSETVNETBE:
1089 return tap_set_vnet_be(q, sp);
1090
1091 case TUNSETOFFLOAD:
1092 /* let the user check for future flags */
1093 if (arg & ~(TUN_F_CSUM | TUN_F_TSO4 | TUN_F_TSO6 |
1094 TUN_F_TSO_ECN | TUN_F_UFO))
1095 return -EINVAL;
1096
1097 rtnl_lock();
1098 ret = set_offload(q, arg);
1099 rtnl_unlock();
1100 return ret;
1101
1102 case SIOCGIFHWADDR:
1103 rtnl_lock();
1104 tap = tap_get_tap_dev(q);
1105 if (!tap) {
1106 rtnl_unlock();
1107 return -ENOLINK;
1108 }
1109 ret = 0;
1110 dev_get_mac_address(&sa, dev_net(tap->dev), tap->dev->name);
1111 if (copy_to_user(&ifr->ifr_name, tap->dev->name, IFNAMSIZ) ||
1112 copy_to_user(&ifr->ifr_hwaddr, &sa, sizeof(sa)))
1113 ret = -EFAULT;
1114 tap_put_tap_dev(tap);
1115 rtnl_unlock();
1116 return ret;
1117
1118 case SIOCSIFHWADDR:
1119 if (copy_from_user(&sa, &ifr->ifr_hwaddr, sizeof(sa)))
1120 return -EFAULT;
1121 rtnl_lock();
1122 tap = tap_get_tap_dev(q);
1123 if (!tap) {
1124 rtnl_unlock();
1125 return -ENOLINK;
1126 }
1127 ret = dev_set_mac_address_user(tap->dev, &sa, NULL);
1128 tap_put_tap_dev(tap);
1129 rtnl_unlock();
1130 return ret;
1131
1132 default:
1133 return -EINVAL;
1134 }
1135 }
1136
1137 static const struct file_operations tap_fops = {
1138 .owner = THIS_MODULE,
1139 .open = tap_open,
1140 .release = tap_release,
1141 .read_iter = tap_read_iter,
1142 .write_iter = tap_write_iter,
1143 .poll = tap_poll,
1144 .llseek = no_llseek,
1145 .unlocked_ioctl = tap_ioctl,
1146 .compat_ioctl = compat_ptr_ioctl,
1147 };
1148
tap_get_user_xdp(struct tap_queue * q,struct xdp_buff * xdp)1149 static int tap_get_user_xdp(struct tap_queue *q, struct xdp_buff *xdp)
1150 {
1151 struct tun_xdp_hdr *hdr = xdp->data_hard_start;
1152 struct virtio_net_hdr *gso = &hdr->gso;
1153 int buflen = hdr->buflen;
1154 int vnet_hdr_len = 0;
1155 struct tap_dev *tap;
1156 struct sk_buff *skb;
1157 int err, depth;
1158
1159 if (q->flags & IFF_VNET_HDR)
1160 vnet_hdr_len = READ_ONCE(q->vnet_hdr_sz);
1161
1162 skb = build_skb(xdp->data_hard_start, buflen);
1163 if (!skb) {
1164 err = -ENOMEM;
1165 goto err;
1166 }
1167
1168 skb_reserve(skb, xdp->data - xdp->data_hard_start);
1169 skb_put(skb, xdp->data_end - xdp->data);
1170
1171 skb_set_network_header(skb, ETH_HLEN);
1172 skb_reset_mac_header(skb);
1173 skb->protocol = eth_hdr(skb)->h_proto;
1174
1175 if (vnet_hdr_len) {
1176 err = virtio_net_hdr_to_skb(skb, gso, tap_is_little_endian(q));
1177 if (err)
1178 goto err_kfree;
1179 }
1180
1181 /* Move network header to the right position for VLAN tagged packets */
1182 if (eth_type_vlan(skb->protocol) &&
1183 __vlan_get_protocol(skb, skb->protocol, &depth) != 0)
1184 skb_set_network_header(skb, depth);
1185
1186 rcu_read_lock();
1187 tap = rcu_dereference(q->tap);
1188 if (tap) {
1189 skb->dev = tap->dev;
1190 skb_probe_transport_header(skb);
1191 dev_queue_xmit(skb);
1192 } else {
1193 kfree_skb(skb);
1194 }
1195 rcu_read_unlock();
1196
1197 return 0;
1198
1199 err_kfree:
1200 kfree_skb(skb);
1201 err:
1202 rcu_read_lock();
1203 tap = rcu_dereference(q->tap);
1204 if (tap && tap->count_tx_dropped)
1205 tap->count_tx_dropped(tap);
1206 rcu_read_unlock();
1207 return err;
1208 }
1209
tap_sendmsg(struct socket * sock,struct msghdr * m,size_t total_len)1210 static int tap_sendmsg(struct socket *sock, struct msghdr *m,
1211 size_t total_len)
1212 {
1213 struct tap_queue *q = container_of(sock, struct tap_queue, sock);
1214 struct tun_msg_ctl *ctl = m->msg_control;
1215 struct xdp_buff *xdp;
1216 int i;
1217
1218 if (m->msg_controllen == sizeof(struct tun_msg_ctl) &&
1219 ctl && ctl->type == TUN_MSG_PTR) {
1220 for (i = 0; i < ctl->num; i++) {
1221 xdp = &((struct xdp_buff *)ctl->ptr)[i];
1222 tap_get_user_xdp(q, xdp);
1223 }
1224 return 0;
1225 }
1226
1227 return tap_get_user(q, ctl ? ctl->ptr : NULL, &m->msg_iter,
1228 m->msg_flags & MSG_DONTWAIT);
1229 }
1230
tap_recvmsg(struct socket * sock,struct msghdr * m,size_t total_len,int flags)1231 static int tap_recvmsg(struct socket *sock, struct msghdr *m,
1232 size_t total_len, int flags)
1233 {
1234 struct tap_queue *q = container_of(sock, struct tap_queue, sock);
1235 struct sk_buff *skb = m->msg_control;
1236 int ret;
1237 if (flags & ~(MSG_DONTWAIT|MSG_TRUNC)) {
1238 kfree_skb(skb);
1239 return -EINVAL;
1240 }
1241 ret = tap_do_read(q, &m->msg_iter, flags & MSG_DONTWAIT, skb);
1242 if (ret > total_len) {
1243 m->msg_flags |= MSG_TRUNC;
1244 ret = flags & MSG_TRUNC ? ret : total_len;
1245 }
1246 return ret;
1247 }
1248
tap_peek_len(struct socket * sock)1249 static int tap_peek_len(struct socket *sock)
1250 {
1251 struct tap_queue *q = container_of(sock, struct tap_queue,
1252 sock);
1253 return PTR_RING_PEEK_CALL(&q->ring, __skb_array_len_with_tag);
1254 }
1255
1256 /* Ops structure to mimic raw sockets with tun */
1257 static const struct proto_ops tap_socket_ops = {
1258 .sendmsg = tap_sendmsg,
1259 .recvmsg = tap_recvmsg,
1260 .peek_len = tap_peek_len,
1261 };
1262
1263 /* Get an underlying socket object from tun file. Returns error unless file is
1264 * attached to a device. The returned object works like a packet socket, it
1265 * can be used for sock_sendmsg/sock_recvmsg. The caller is responsible for
1266 * holding a reference to the file for as long as the socket is in use. */
tap_get_socket(struct file * file)1267 struct socket *tap_get_socket(struct file *file)
1268 {
1269 struct tap_queue *q;
1270 if (file->f_op != &tap_fops)
1271 return ERR_PTR(-EINVAL);
1272 q = file->private_data;
1273 if (!q)
1274 return ERR_PTR(-EBADFD);
1275 return &q->sock;
1276 }
1277 EXPORT_SYMBOL_GPL(tap_get_socket);
1278
tap_get_ptr_ring(struct file * file)1279 struct ptr_ring *tap_get_ptr_ring(struct file *file)
1280 {
1281 struct tap_queue *q;
1282
1283 if (file->f_op != &tap_fops)
1284 return ERR_PTR(-EINVAL);
1285 q = file->private_data;
1286 if (!q)
1287 return ERR_PTR(-EBADFD);
1288 return &q->ring;
1289 }
1290 EXPORT_SYMBOL_GPL(tap_get_ptr_ring);
1291
tap_queue_resize(struct tap_dev * tap)1292 int tap_queue_resize(struct tap_dev *tap)
1293 {
1294 struct net_device *dev = tap->dev;
1295 struct tap_queue *q;
1296 struct ptr_ring **rings;
1297 int n = tap->numqueues;
1298 int ret, i = 0;
1299
1300 rings = kmalloc_array(n, sizeof(*rings), GFP_KERNEL);
1301 if (!rings)
1302 return -ENOMEM;
1303
1304 list_for_each_entry(q, &tap->queue_list, next)
1305 rings[i++] = &q->ring;
1306
1307 ret = ptr_ring_resize_multiple(rings, n,
1308 dev->tx_queue_len, GFP_KERNEL,
1309 __skb_array_destroy_skb);
1310
1311 kfree(rings);
1312 return ret;
1313 }
1314 EXPORT_SYMBOL_GPL(tap_queue_resize);
1315
tap_list_add(dev_t major,const char * device_name)1316 static int tap_list_add(dev_t major, const char *device_name)
1317 {
1318 struct major_info *tap_major;
1319
1320 tap_major = kzalloc(sizeof(*tap_major), GFP_ATOMIC);
1321 if (!tap_major)
1322 return -ENOMEM;
1323
1324 tap_major->major = MAJOR(major);
1325
1326 idr_init(&tap_major->minor_idr);
1327 spin_lock_init(&tap_major->minor_lock);
1328
1329 tap_major->device_name = device_name;
1330
1331 list_add_tail_rcu(&tap_major->next, &major_list);
1332 return 0;
1333 }
1334
tap_create_cdev(struct cdev * tap_cdev,dev_t * tap_major,const char * device_name,struct module * module)1335 int tap_create_cdev(struct cdev *tap_cdev, dev_t *tap_major,
1336 const char *device_name, struct module *module)
1337 {
1338 int err;
1339
1340 err = alloc_chrdev_region(tap_major, 0, TAP_NUM_DEVS, device_name);
1341 if (err)
1342 goto out1;
1343
1344 cdev_init(tap_cdev, &tap_fops);
1345 tap_cdev->owner = module;
1346 err = cdev_add(tap_cdev, *tap_major, TAP_NUM_DEVS);
1347 if (err)
1348 goto out2;
1349
1350 err = tap_list_add(*tap_major, device_name);
1351 if (err)
1352 goto out3;
1353
1354 return 0;
1355
1356 out3:
1357 cdev_del(tap_cdev);
1358 out2:
1359 unregister_chrdev_region(*tap_major, TAP_NUM_DEVS);
1360 out1:
1361 return err;
1362 }
1363 EXPORT_SYMBOL_GPL(tap_create_cdev);
1364
tap_destroy_cdev(dev_t major,struct cdev * tap_cdev)1365 void tap_destroy_cdev(dev_t major, struct cdev *tap_cdev)
1366 {
1367 struct major_info *tap_major, *tmp;
1368
1369 cdev_del(tap_cdev);
1370 unregister_chrdev_region(major, TAP_NUM_DEVS);
1371 list_for_each_entry_safe(tap_major, tmp, &major_list, next) {
1372 if (tap_major->major == MAJOR(major)) {
1373 idr_destroy(&tap_major->minor_idr);
1374 list_del_rcu(&tap_major->next);
1375 kfree_rcu(tap_major, rcu);
1376 }
1377 }
1378 }
1379 EXPORT_SYMBOL_GPL(tap_destroy_cdev);
1380
1381 MODULE_AUTHOR("Arnd Bergmann <arnd@arndb.de>");
1382 MODULE_AUTHOR("Sainath Grandhi <sainath.grandhi@intel.com>");
1383 MODULE_LICENSE("GPL");
1384