blob: fc27a9926d9e52c28a63d6be8852518e5cebb4da [file] [log] [blame]
Arnd Bergmann20d29d72010-01-30 12:24:26 +00001#include <linux/etherdevice.h>
2#include <linux/if_macvlan.h>
3#include <linux/interrupt.h>
4#include <linux/nsproxy.h>
5#include <linux/compat.h>
6#include <linux/if_tun.h>
7#include <linux/module.h>
8#include <linux/skbuff.h>
9#include <linux/cache.h>
10#include <linux/sched.h>
11#include <linux/types.h>
Tejun Heo5a0e3ad2010-03-24 17:04:11 +090012#include <linux/slab.h>
Arnd Bergmann20d29d72010-01-30 12:24:26 +000013#include <linux/init.h>
14#include <linux/wait.h>
15#include <linux/cdev.h>
16#include <linux/fs.h>
17
18#include <net/net_namespace.h>
19#include <net/rtnetlink.h>
20#include <net/sock.h>
Arnd Bergmannb9fb9ee2010-02-18 05:48:17 +000021#include <linux/virtio_net.h>
Arnd Bergmann20d29d72010-01-30 12:24:26 +000022
23/*
24 * A macvtap queue is the central object of this driver, it connects
25 * an open character device to a macvlan interface. There can be
26 * multiple queues on one interface, which map back to queues
27 * implemented in hardware on the underlying device.
28 *
29 * macvtap_proto is used to allocate queues through the sock allocation
30 * mechanism.
31 *
32 * TODO: multiqueue support is currently not implemented, even though
33 * macvtap is basically prepared for that. We will need to add this
34 * here as well as in virtio-net and qemu to get line rate on 10gbit
35 * adapters from a guest.
36 */
37struct macvtap_queue {
38 struct sock sk;
39 struct socket sock;
Eric Dumazet43815482010-04-29 11:01:49 +000040 struct socket_wq wq;
Michael S. Tsirkin55afbd02010-04-29 13:50:48 +030041 int vnet_hdr_sz;
Arnd Bergmann20d29d72010-01-30 12:24:26 +000042 struct macvlan_dev *vlan;
43 struct file *file;
Arnd Bergmannb9fb9ee2010-02-18 05:48:17 +000044 unsigned int flags;
Arnd Bergmann20d29d72010-01-30 12:24:26 +000045};
46
47static struct proto macvtap_proto = {
48 .name = "macvtap",
49 .owner = THIS_MODULE,
50 .obj_size = sizeof (struct macvtap_queue),
51};
52
53/*
54 * Minor number matches netdev->ifindex, so need a potentially
55 * large value. This also makes it possible to split the
56 * tap functionality out again in the future by offering it
57 * from other drivers besides macvtap. As long as every device
58 * only has one tap, the interface numbers assure that the
59 * device nodes are unique.
60 */
David S. Miller1ebed712010-07-10 19:25:50 -070061static dev_t macvtap_major;
Arnd Bergmann20d29d72010-01-30 12:24:26 +000062#define MACVTAP_NUM_DEVS 65536
63static struct class *macvtap_class;
64static struct cdev macvtap_cdev;
65
Arnd Bergmann501c7742010-02-18 05:46:50 +000066static const struct proto_ops macvtap_socket_ops;
67
Arnd Bergmann20d29d72010-01-30 12:24:26 +000068/*
69 * RCU usage:
Arnd Bergmann02df55d2010-02-18 05:45:36 +000070 * The macvtap_queue and the macvlan_dev are loosely coupled, the
71 * pointers from one to the other can only be read while rcu_read_lock
72 * or macvtap_lock is held.
Arnd Bergmann20d29d72010-01-30 12:24:26 +000073 *
Arnd Bergmann02df55d2010-02-18 05:45:36 +000074 * Both the file and the macvlan_dev hold a reference on the macvtap_queue
75 * through sock_hold(&q->sk). When the macvlan_dev goes away first,
76 * q->vlan becomes inaccessible. When the files gets closed,
77 * macvtap_get_queue() fails.
Arnd Bergmann20d29d72010-01-30 12:24:26 +000078 *
Arnd Bergmann02df55d2010-02-18 05:45:36 +000079 * There may still be references to the struct sock inside of the
80 * queue from outbound SKBs, but these never reference back to the
81 * file or the dev. The data structure is freed through __sk_free
82 * when both our references and any pending SKBs are gone.
Arnd Bergmann20d29d72010-01-30 12:24:26 +000083 */
84static DEFINE_SPINLOCK(macvtap_lock);
85
86/*
Krishna Kumar1565c7c2010-08-04 06:15:59 +000087 * get_slot: return a [unused/occupied] slot in vlan->taps[]:
88 * - if 'q' is NULL, return the first empty slot;
89 * - otherwise, return the slot this pointer occupies.
Arnd Bergmann20d29d72010-01-30 12:24:26 +000090 */
Krishna Kumar1565c7c2010-08-04 06:15:59 +000091static int get_slot(struct macvlan_dev *vlan, struct macvtap_queue *q)
92{
93 int i;
94
95 for (i = 0; i < MAX_MACVTAP_QUEUES; i++) {
96 if (rcu_dereference(vlan->taps[i]) == q)
97 return i;
98 }
99
100 /* Should never happen */
101 BUG_ON(1);
102}
103
Arnd Bergmann20d29d72010-01-30 12:24:26 +0000104static int macvtap_set_queue(struct net_device *dev, struct file *file,
105 struct macvtap_queue *q)
106{
107 struct macvlan_dev *vlan = netdev_priv(dev);
Krishna Kumar1565c7c2010-08-04 06:15:59 +0000108 int index;
Arnd Bergmann20d29d72010-01-30 12:24:26 +0000109 int err = -EBUSY;
110
111 spin_lock(&macvtap_lock);
Krishna Kumar1565c7c2010-08-04 06:15:59 +0000112 if (vlan->numvtaps == MAX_MACVTAP_QUEUES)
Arnd Bergmann20d29d72010-01-30 12:24:26 +0000113 goto out;
114
115 err = 0;
Krishna Kumar1565c7c2010-08-04 06:15:59 +0000116 index = get_slot(vlan, NULL);
Arnd Bergmann02df55d2010-02-18 05:45:36 +0000117 rcu_assign_pointer(q->vlan, vlan);
Krishna Kumar1565c7c2010-08-04 06:15:59 +0000118 rcu_assign_pointer(vlan->taps[index], q);
Arnd Bergmann02df55d2010-02-18 05:45:36 +0000119 sock_hold(&q->sk);
Arnd Bergmann20d29d72010-01-30 12:24:26 +0000120
121 q->file = file;
Arnd Bergmann02df55d2010-02-18 05:45:36 +0000122 file->private_data = q;
Arnd Bergmann20d29d72010-01-30 12:24:26 +0000123
Krishna Kumar1565c7c2010-08-04 06:15:59 +0000124 vlan->numvtaps++;
125
Arnd Bergmann20d29d72010-01-30 12:24:26 +0000126out:
127 spin_unlock(&macvtap_lock);
128 return err;
129}
130
131/*
Arnd Bergmann02df55d2010-02-18 05:45:36 +0000132 * The file owning the queue got closed, give up both
133 * the reference that the files holds as well as the
134 * one from the macvlan_dev if that still exists.
Arnd Bergmann20d29d72010-01-30 12:24:26 +0000135 *
136 * Using the spinlock makes sure that we don't get
137 * to the queue again after destroying it.
Arnd Bergmann20d29d72010-01-30 12:24:26 +0000138 */
Arnd Bergmann02df55d2010-02-18 05:45:36 +0000139static void macvtap_put_queue(struct macvtap_queue *q)
Arnd Bergmann20d29d72010-01-30 12:24:26 +0000140{
Arnd Bergmann02df55d2010-02-18 05:45:36 +0000141 struct macvlan_dev *vlan;
Arnd Bergmann20d29d72010-01-30 12:24:26 +0000142
143 spin_lock(&macvtap_lock);
Arnd Bergmann02df55d2010-02-18 05:45:36 +0000144 vlan = rcu_dereference(q->vlan);
145 if (vlan) {
Krishna Kumar1565c7c2010-08-04 06:15:59 +0000146 int index = get_slot(vlan, q);
147
148 rcu_assign_pointer(vlan->taps[index], NULL);
Arnd Bergmann02df55d2010-02-18 05:45:36 +0000149 rcu_assign_pointer(q->vlan, NULL);
150 sock_put(&q->sk);
Krishna Kumar1565c7c2010-08-04 06:15:59 +0000151 --vlan->numvtaps;
Arnd Bergmann20d29d72010-01-30 12:24:26 +0000152 }
153
Arnd Bergmann20d29d72010-01-30 12:24:26 +0000154 spin_unlock(&macvtap_lock);
155
156 synchronize_rcu();
157 sock_put(&q->sk);
158}
159
160/*
Krishna Kumar1565c7c2010-08-04 06:15:59 +0000161 * Select a queue based on the rxq of the device on which this packet
162 * arrived. If the incoming device is not mq, calculate a flow hash
163 * to select a queue. If all fails, find the first available queue.
164 * Cache vlan->numvtaps since it can become zero during the execution
165 * of this function.
Arnd Bergmann20d29d72010-01-30 12:24:26 +0000166 */
167static struct macvtap_queue *macvtap_get_queue(struct net_device *dev,
168 struct sk_buff *skb)
169{
170 struct macvlan_dev *vlan = netdev_priv(dev);
Krishna Kumar1565c7c2010-08-04 06:15:59 +0000171 struct macvtap_queue *tap = NULL;
172 int numvtaps = vlan->numvtaps;
173 __u32 rxq;
Arnd Bergmann20d29d72010-01-30 12:24:26 +0000174
Krishna Kumar1565c7c2010-08-04 06:15:59 +0000175 if (!numvtaps)
176 goto out;
177
178 if (likely(skb_rx_queue_recorded(skb))) {
179 rxq = skb_get_rx_queue(skb);
180
181 while (unlikely(rxq >= numvtaps))
182 rxq -= numvtaps;
183
184 tap = rcu_dereference(vlan->taps[rxq]);
185 if (tap)
186 goto out;
187 }
188
189 /* Check if we can use flow to select a queue */
190 rxq = skb_get_rxhash(skb);
191 if (rxq) {
192 tap = rcu_dereference(vlan->taps[rxq % numvtaps]);
193 if (tap)
194 goto out;
195 }
196
197 /* Everything failed - find first available queue */
198 for (rxq = 0; rxq < MAX_MACVTAP_QUEUES; rxq++) {
199 tap = rcu_dereference(vlan->taps[rxq]);
200 if (tap)
201 break;
202 }
203
204out:
205 return tap;
Arnd Bergmann20d29d72010-01-30 12:24:26 +0000206}
207
Arnd Bergmann02df55d2010-02-18 05:45:36 +0000208/*
209 * The net_device is going away, give up the reference
Krishna Kumar1565c7c2010-08-04 06:15:59 +0000210 * that it holds on all queues and safely set the pointer
211 * from the queues to NULL.
Arnd Bergmann02df55d2010-02-18 05:45:36 +0000212 */
Arnd Bergmann20d29d72010-01-30 12:24:26 +0000213static void macvtap_del_queues(struct net_device *dev)
214{
215 struct macvlan_dev *vlan = netdev_priv(dev);
Krishna Kumar1565c7c2010-08-04 06:15:59 +0000216 struct macvtap_queue *q, *qlist[MAX_MACVTAP_QUEUES];
217 int i, j = 0;
Arnd Bergmann20d29d72010-01-30 12:24:26 +0000218
Krishna Kumar1565c7c2010-08-04 06:15:59 +0000219 /* macvtap_put_queue can free some slots, so go through all slots */
Arnd Bergmann02df55d2010-02-18 05:45:36 +0000220 spin_lock(&macvtap_lock);
Krishna Kumar1565c7c2010-08-04 06:15:59 +0000221 for (i = 0; i < MAX_MACVTAP_QUEUES && vlan->numvtaps; i++) {
222 q = rcu_dereference(vlan->taps[i]);
223 if (q) {
224 qlist[j++] = q;
225 rcu_assign_pointer(vlan->taps[i], NULL);
226 rcu_assign_pointer(q->vlan, NULL);
227 vlan->numvtaps--;
228 }
Arnd Bergmann02df55d2010-02-18 05:45:36 +0000229 }
Krishna Kumar1565c7c2010-08-04 06:15:59 +0000230 BUG_ON(vlan->numvtaps != 0);
Arnd Bergmann02df55d2010-02-18 05:45:36 +0000231 spin_unlock(&macvtap_lock);
232
233 synchronize_rcu();
Krishna Kumar1565c7c2010-08-04 06:15:59 +0000234
235 for (--j; j >= 0; j--)
236 sock_put(&qlist[j]->sk);
Arnd Bergmann20d29d72010-01-30 12:24:26 +0000237}
238
239/*
240 * Forward happens for data that gets sent from one macvlan
241 * endpoint to another one in bridge mode. We just take
242 * the skb and put it into the receive queue.
243 */
244static int macvtap_forward(struct net_device *dev, struct sk_buff *skb)
245{
246 struct macvtap_queue *q = macvtap_get_queue(dev, skb);
247 if (!q)
Herbert Xu8a357472010-07-21 21:44:31 +0000248 goto drop;
249
250 if (skb_queue_len(&q->sk.sk_receive_queue) >= dev->tx_queue_len)
251 goto drop;
Arnd Bergmann20d29d72010-01-30 12:24:26 +0000252
253 skb_queue_tail(&q->sk.sk_receive_queue, skb);
Eric Dumazet4a4771a2010-04-25 22:20:06 +0000254 wake_up_interruptible_poll(sk_sleep(&q->sk), POLLIN | POLLRDNORM | POLLRDBAND);
Herbert Xu8a357472010-07-21 21:44:31 +0000255 return NET_RX_SUCCESS;
256
257drop:
258 kfree_skb(skb);
259 return NET_RX_DROP;
Arnd Bergmann20d29d72010-01-30 12:24:26 +0000260}
261
262/*
263 * Receive is for data from the external interface (lowerdev),
264 * in case of macvtap, we can treat that the same way as
265 * forward, which macvlan cannot.
266 */
267static int macvtap_receive(struct sk_buff *skb)
268{
269 skb_push(skb, ETH_HLEN);
270 return macvtap_forward(skb->dev, skb);
271}
272
273static int macvtap_newlink(struct net *src_net,
274 struct net_device *dev,
275 struct nlattr *tb[],
276 struct nlattr *data[])
277{
278 struct device *classdev;
279 dev_t devt;
280 int err;
281
282 err = macvlan_common_newlink(src_net, dev, tb, data,
283 macvtap_receive, macvtap_forward);
284 if (err)
285 goto out;
286
287 devt = MKDEV(MAJOR(macvtap_major), dev->ifindex);
288
289 classdev = device_create(macvtap_class, &dev->dev, devt,
290 dev, "tap%d", dev->ifindex);
291 if (IS_ERR(classdev)) {
292 err = PTR_ERR(classdev);
293 macvtap_del_queues(dev);
294 }
295
296out:
297 return err;
298}
299
300static void macvtap_dellink(struct net_device *dev,
301 struct list_head *head)
302{
303 device_destroy(macvtap_class,
304 MKDEV(MAJOR(macvtap_major), dev->ifindex));
305
306 macvtap_del_queues(dev);
307 macvlan_dellink(dev, head);
308}
309
Herbert Xu8a357472010-07-21 21:44:31 +0000310static void macvtap_setup(struct net_device *dev)
311{
312 macvlan_common_setup(dev);
313 dev->tx_queue_len = TUN_READQ_SIZE;
314}
315
Arnd Bergmann20d29d72010-01-30 12:24:26 +0000316static struct rtnl_link_ops macvtap_link_ops __read_mostly = {
317 .kind = "macvtap",
Herbert Xu8a357472010-07-21 21:44:31 +0000318 .setup = macvtap_setup,
Arnd Bergmann20d29d72010-01-30 12:24:26 +0000319 .newlink = macvtap_newlink,
320 .dellink = macvtap_dellink,
321};
322
323
324static void macvtap_sock_write_space(struct sock *sk)
325{
Eric Dumazet43815482010-04-29 11:01:49 +0000326 wait_queue_head_t *wqueue;
327
Arnd Bergmann20d29d72010-01-30 12:24:26 +0000328 if (!sock_writeable(sk) ||
329 !test_and_clear_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags))
330 return;
331
Eric Dumazet43815482010-04-29 11:01:49 +0000332 wqueue = sk_sleep(sk);
333 if (wqueue && waitqueue_active(wqueue))
334 wake_up_interruptible_poll(wqueue, POLLOUT | POLLWRNORM | POLLWRBAND);
Arnd Bergmann20d29d72010-01-30 12:24:26 +0000335}
336
337static int macvtap_open(struct inode *inode, struct file *file)
338{
339 struct net *net = current->nsproxy->net_ns;
340 struct net_device *dev = dev_get_by_index(net, iminor(inode));
341 struct macvtap_queue *q;
342 int err;
343
344 err = -ENODEV;
345 if (!dev)
346 goto out;
347
348 /* check if this is a macvtap device */
349 err = -EINVAL;
350 if (dev->rtnl_link_ops != &macvtap_link_ops)
351 goto out;
352
353 err = -ENOMEM;
354 q = (struct macvtap_queue *)sk_alloc(net, AF_UNSPEC, GFP_KERNEL,
355 &macvtap_proto);
356 if (!q)
357 goto out;
358
Eric Dumazet43815482010-04-29 11:01:49 +0000359 q->sock.wq = &q->wq;
360 init_waitqueue_head(&q->wq.wait);
Arnd Bergmann20d29d72010-01-30 12:24:26 +0000361 q->sock.type = SOCK_RAW;
362 q->sock.state = SS_CONNECTED;
Arnd Bergmann501c7742010-02-18 05:46:50 +0000363 q->sock.file = file;
364 q->sock.ops = &macvtap_socket_ops;
Arnd Bergmann20d29d72010-01-30 12:24:26 +0000365 sock_init_data(&q->sock, &q->sk);
Arnd Bergmann20d29d72010-01-30 12:24:26 +0000366 q->sk.sk_write_space = macvtap_sock_write_space;
Arnd Bergmannb9fb9ee2010-02-18 05:48:17 +0000367 q->flags = IFF_VNET_HDR | IFF_NO_PI | IFF_TAP;
Michael S. Tsirkin55afbd02010-04-29 13:50:48 +0300368 q->vnet_hdr_sz = sizeof(struct virtio_net_hdr);
Arnd Bergmann20d29d72010-01-30 12:24:26 +0000369
370 err = macvtap_set_queue(dev, file, q);
371 if (err)
372 sock_put(&q->sk);
373
374out:
375 if (dev)
376 dev_put(dev);
377
378 return err;
379}
380
381static int macvtap_release(struct inode *inode, struct file *file)
382{
Arnd Bergmann02df55d2010-02-18 05:45:36 +0000383 struct macvtap_queue *q = file->private_data;
384 macvtap_put_queue(q);
Arnd Bergmann20d29d72010-01-30 12:24:26 +0000385 return 0;
386}
387
388static unsigned int macvtap_poll(struct file *file, poll_table * wait)
389{
Arnd Bergmann02df55d2010-02-18 05:45:36 +0000390 struct macvtap_queue *q = file->private_data;
Arnd Bergmann20d29d72010-01-30 12:24:26 +0000391 unsigned int mask = POLLERR;
392
393 if (!q)
394 goto out;
395
396 mask = 0;
Eric Dumazet43815482010-04-29 11:01:49 +0000397 poll_wait(file, &q->wq.wait, wait);
Arnd Bergmann20d29d72010-01-30 12:24:26 +0000398
399 if (!skb_queue_empty(&q->sk.sk_receive_queue))
400 mask |= POLLIN | POLLRDNORM;
401
402 if (sock_writeable(&q->sk) ||
403 (!test_and_set_bit(SOCK_ASYNC_NOSPACE, &q->sock.flags) &&
404 sock_writeable(&q->sk)))
405 mask |= POLLOUT | POLLWRNORM;
406
407out:
Arnd Bergmann20d29d72010-01-30 12:24:26 +0000408 return mask;
409}
410
Arnd Bergmannb9fb9ee2010-02-18 05:48:17 +0000411static inline struct sk_buff *macvtap_alloc_skb(struct sock *sk, size_t prepad,
412 size_t len, size_t linear,
413 int noblock, int *err)
414{
415 struct sk_buff *skb;
416
417 /* Under a page? Don't bother with paged skb. */
418 if (prepad + len < PAGE_SIZE || !linear)
419 linear = len;
420
421 skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock,
422 err);
423 if (!skb)
424 return NULL;
425
426 skb_reserve(skb, prepad);
427 skb_put(skb, linear);
428 skb->data_len = len - linear;
429 skb->len += len - linear;
430
431 return skb;
432}
433
434/*
435 * macvtap_skb_from_vnet_hdr and macvtap_skb_to_vnet_hdr should
436 * be shared with the tun/tap driver.
437 */
438static int macvtap_skb_from_vnet_hdr(struct sk_buff *skb,
439 struct virtio_net_hdr *vnet_hdr)
440{
441 unsigned short gso_type = 0;
442 if (vnet_hdr->gso_type != VIRTIO_NET_HDR_GSO_NONE) {
443 switch (vnet_hdr->gso_type & ~VIRTIO_NET_HDR_GSO_ECN) {
444 case VIRTIO_NET_HDR_GSO_TCPV4:
445 gso_type = SKB_GSO_TCPV4;
446 break;
447 case VIRTIO_NET_HDR_GSO_TCPV6:
448 gso_type = SKB_GSO_TCPV6;
449 break;
450 case VIRTIO_NET_HDR_GSO_UDP:
451 gso_type = SKB_GSO_UDP;
452 break;
453 default:
454 return -EINVAL;
455 }
456
457 if (vnet_hdr->gso_type & VIRTIO_NET_HDR_GSO_ECN)
458 gso_type |= SKB_GSO_TCP_ECN;
459
460 if (vnet_hdr->gso_size == 0)
461 return -EINVAL;
462 }
463
464 if (vnet_hdr->flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) {
465 if (!skb_partial_csum_set(skb, vnet_hdr->csum_start,
466 vnet_hdr->csum_offset))
467 return -EINVAL;
468 }
469
470 if (vnet_hdr->gso_type != VIRTIO_NET_HDR_GSO_NONE) {
471 skb_shinfo(skb)->gso_size = vnet_hdr->gso_size;
472 skb_shinfo(skb)->gso_type = gso_type;
473
474 /* Header must be checked, and gso_segs computed. */
475 skb_shinfo(skb)->gso_type |= SKB_GSO_DODGY;
476 skb_shinfo(skb)->gso_segs = 0;
477 }
478 return 0;
479}
480
481static int macvtap_skb_to_vnet_hdr(const struct sk_buff *skb,
482 struct virtio_net_hdr *vnet_hdr)
483{
484 memset(vnet_hdr, 0, sizeof(*vnet_hdr));
485
486 if (skb_is_gso(skb)) {
487 struct skb_shared_info *sinfo = skb_shinfo(skb);
488
489 /* This is a hint as to how much should be linear. */
490 vnet_hdr->hdr_len = skb_headlen(skb);
491 vnet_hdr->gso_size = sinfo->gso_size;
492 if (sinfo->gso_type & SKB_GSO_TCPV4)
493 vnet_hdr->gso_type = VIRTIO_NET_HDR_GSO_TCPV4;
494 else if (sinfo->gso_type & SKB_GSO_TCPV6)
495 vnet_hdr->gso_type = VIRTIO_NET_HDR_GSO_TCPV6;
496 else if (sinfo->gso_type & SKB_GSO_UDP)
497 vnet_hdr->gso_type = VIRTIO_NET_HDR_GSO_UDP;
498 else
499 BUG();
500 if (sinfo->gso_type & SKB_GSO_TCP_ECN)
501 vnet_hdr->gso_type |= VIRTIO_NET_HDR_GSO_ECN;
502 } else
503 vnet_hdr->gso_type = VIRTIO_NET_HDR_GSO_NONE;
504
505 if (skb->ip_summed == CHECKSUM_PARTIAL) {
506 vnet_hdr->flags = VIRTIO_NET_HDR_F_NEEDS_CSUM;
Michał Mirosław55508d62010-12-14 15:24:08 +0000507 vnet_hdr->csum_start = skb_checksum_start_offset(skb);
Arnd Bergmannb9fb9ee2010-02-18 05:48:17 +0000508 vnet_hdr->csum_offset = skb->csum_offset;
509 } /* else everything is zero */
510
511 return 0;
512}
513
514
Arnd Bergmann20d29d72010-01-30 12:24:26 +0000515/* Get packet from user space buffer */
516static ssize_t macvtap_get_user(struct macvtap_queue *q,
517 const struct iovec *iv, size_t count,
518 int noblock)
519{
520 struct sk_buff *skb;
Arnd Bergmann02df55d2010-02-18 05:45:36 +0000521 struct macvlan_dev *vlan;
Arnd Bergmann20d29d72010-01-30 12:24:26 +0000522 size_t len = count;
523 int err;
Arnd Bergmannb9fb9ee2010-02-18 05:48:17 +0000524 struct virtio_net_hdr vnet_hdr = { 0 };
525 int vnet_hdr_len = 0;
Arnd Bergmann20d29d72010-01-30 12:24:26 +0000526
Arnd Bergmannb9fb9ee2010-02-18 05:48:17 +0000527 if (q->flags & IFF_VNET_HDR) {
Michael S. Tsirkin55afbd02010-04-29 13:50:48 +0300528 vnet_hdr_len = q->vnet_hdr_sz;
Arnd Bergmannb9fb9ee2010-02-18 05:48:17 +0000529
530 err = -EINVAL;
Nicolas Kaiserce3c8692011-03-04 13:49:41 +0000531 if (len < vnet_hdr_len)
Arnd Bergmannb9fb9ee2010-02-18 05:48:17 +0000532 goto err;
Nicolas Kaiserce3c8692011-03-04 13:49:41 +0000533 len -= vnet_hdr_len;
Arnd Bergmannb9fb9ee2010-02-18 05:48:17 +0000534
535 err = memcpy_fromiovecend((void *)&vnet_hdr, iv, 0,
Michael S. Tsirkin55afbd02010-04-29 13:50:48 +0300536 sizeof(vnet_hdr));
Arnd Bergmannb9fb9ee2010-02-18 05:48:17 +0000537 if (err < 0)
538 goto err;
539 if ((vnet_hdr.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) &&
540 vnet_hdr.csum_start + vnet_hdr.csum_offset + 2 >
541 vnet_hdr.hdr_len)
542 vnet_hdr.hdr_len = vnet_hdr.csum_start +
543 vnet_hdr.csum_offset + 2;
544 err = -EINVAL;
545 if (vnet_hdr.hdr_len > len)
546 goto err;
547 }
548
549 err = -EINVAL;
Arnd Bergmann20d29d72010-01-30 12:24:26 +0000550 if (unlikely(len < ETH_HLEN))
Arnd Bergmannb9fb9ee2010-02-18 05:48:17 +0000551 goto err;
Arnd Bergmann20d29d72010-01-30 12:24:26 +0000552
Arnd Bergmannb9fb9ee2010-02-18 05:48:17 +0000553 skb = macvtap_alloc_skb(&q->sk, NET_IP_ALIGN, len, vnet_hdr.hdr_len,
554 noblock, &err);
Arnd Bergmann02df55d2010-02-18 05:45:36 +0000555 if (!skb)
556 goto err;
Arnd Bergmann20d29d72010-01-30 12:24:26 +0000557
Arnd Bergmannb9fb9ee2010-02-18 05:48:17 +0000558 err = skb_copy_datagram_from_iovec(skb, 0, iv, vnet_hdr_len, len);
Arnd Bergmann02df55d2010-02-18 05:45:36 +0000559 if (err)
Arnd Bergmannb9fb9ee2010-02-18 05:48:17 +0000560 goto err_kfree;
Arnd Bergmann20d29d72010-01-30 12:24:26 +0000561
562 skb_set_network_header(skb, ETH_HLEN);
Arnd Bergmannb9fb9ee2010-02-18 05:48:17 +0000563 skb_reset_mac_header(skb);
564 skb->protocol = eth_hdr(skb)->h_proto;
565
566 if (vnet_hdr_len) {
567 err = macvtap_skb_from_vnet_hdr(skb, &vnet_hdr);
568 if (err)
569 goto err_kfree;
570 }
571
Arnd Bergmann02df55d2010-02-18 05:45:36 +0000572 rcu_read_lock_bh();
573 vlan = rcu_dereference(q->vlan);
574 if (vlan)
575 macvlan_start_xmit(skb, vlan->dev);
576 else
577 kfree_skb(skb);
578 rcu_read_unlock_bh();
Arnd Bergmann20d29d72010-01-30 12:24:26 +0000579
580 return count;
Arnd Bergmann02df55d2010-02-18 05:45:36 +0000581
Arnd Bergmannb9fb9ee2010-02-18 05:48:17 +0000582err_kfree:
583 kfree_skb(skb);
584
Arnd Bergmann02df55d2010-02-18 05:45:36 +0000585err:
586 rcu_read_lock_bh();
587 vlan = rcu_dereference(q->vlan);
588 if (vlan)
Eric Dumazet1ac9ad12011-01-12 12:13:14 +0000589 vlan->dev->stats.tx_dropped++;
Arnd Bergmann02df55d2010-02-18 05:45:36 +0000590 rcu_read_unlock_bh();
591
Arnd Bergmann02df55d2010-02-18 05:45:36 +0000592 return err;
Arnd Bergmann20d29d72010-01-30 12:24:26 +0000593}
594
595static ssize_t macvtap_aio_write(struct kiocb *iocb, const struct iovec *iv,
596 unsigned long count, loff_t pos)
597{
598 struct file *file = iocb->ki_filp;
599 ssize_t result = -ENOLINK;
Arnd Bergmann02df55d2010-02-18 05:45:36 +0000600 struct macvtap_queue *q = file->private_data;
Arnd Bergmann20d29d72010-01-30 12:24:26 +0000601
602 result = macvtap_get_user(q, iv, iov_length(iv, count),
603 file->f_flags & O_NONBLOCK);
Arnd Bergmann20d29d72010-01-30 12:24:26 +0000604 return result;
605}
606
607/* Put packet to the user space buffer */
608static ssize_t macvtap_put_user(struct macvtap_queue *q,
609 const struct sk_buff *skb,
610 const struct iovec *iv, int len)
611{
Arnd Bergmann02df55d2010-02-18 05:45:36 +0000612 struct macvlan_dev *vlan;
Arnd Bergmann20d29d72010-01-30 12:24:26 +0000613 int ret;
Arnd Bergmannb9fb9ee2010-02-18 05:48:17 +0000614 int vnet_hdr_len = 0;
615
616 if (q->flags & IFF_VNET_HDR) {
617 struct virtio_net_hdr vnet_hdr;
Michael S. Tsirkin55afbd02010-04-29 13:50:48 +0300618 vnet_hdr_len = q->vnet_hdr_sz;
Arnd Bergmannb9fb9ee2010-02-18 05:48:17 +0000619 if ((len -= vnet_hdr_len) < 0)
620 return -EINVAL;
621
622 ret = macvtap_skb_to_vnet_hdr(skb, &vnet_hdr);
623 if (ret)
624 return ret;
625
Michael S. Tsirkin55afbd02010-04-29 13:50:48 +0300626 if (memcpy_toiovecend(iv, (void *)&vnet_hdr, 0, sizeof(vnet_hdr)))
Arnd Bergmannb9fb9ee2010-02-18 05:48:17 +0000627 return -EFAULT;
628 }
Arnd Bergmann20d29d72010-01-30 12:24:26 +0000629
630 len = min_t(int, skb->len, len);
631
Arnd Bergmannb9fb9ee2010-02-18 05:48:17 +0000632 ret = skb_copy_datagram_const_iovec(skb, 0, iv, vnet_hdr_len, len);
Arnd Bergmann20d29d72010-01-30 12:24:26 +0000633
Arnd Bergmann02df55d2010-02-18 05:45:36 +0000634 rcu_read_lock_bh();
635 vlan = rcu_dereference(q->vlan);
Arnd Bergmann501c7742010-02-18 05:46:50 +0000636 if (vlan)
637 macvlan_count_rx(vlan, len, ret == 0, 0);
Arnd Bergmann02df55d2010-02-18 05:45:36 +0000638 rcu_read_unlock_bh();
Arnd Bergmann20d29d72010-01-30 12:24:26 +0000639
Arnd Bergmannb9fb9ee2010-02-18 05:48:17 +0000640 return ret ? ret : (len + vnet_hdr_len);
Arnd Bergmann20d29d72010-01-30 12:24:26 +0000641}
642
Arnd Bergmann501c7742010-02-18 05:46:50 +0000643static ssize_t macvtap_do_read(struct macvtap_queue *q, struct kiocb *iocb,
644 const struct iovec *iv, unsigned long len,
645 int noblock)
Arnd Bergmann20d29d72010-01-30 12:24:26 +0000646{
Arnd Bergmann20d29d72010-01-30 12:24:26 +0000647 DECLARE_WAITQUEUE(wait, current);
648 struct sk_buff *skb;
Arnd Bergmann501c7742010-02-18 05:46:50 +0000649 ssize_t ret = 0;
Arnd Bergmann20d29d72010-01-30 12:24:26 +0000650
Eric Dumazet4a4771a2010-04-25 22:20:06 +0000651 add_wait_queue(sk_sleep(&q->sk), &wait);
Arnd Bergmann20d29d72010-01-30 12:24:26 +0000652 while (len) {
653 current->state = TASK_INTERRUPTIBLE;
654
655 /* Read frames from the queue */
656 skb = skb_dequeue(&q->sk.sk_receive_queue);
657 if (!skb) {
Arnd Bergmann501c7742010-02-18 05:46:50 +0000658 if (noblock) {
Arnd Bergmann20d29d72010-01-30 12:24:26 +0000659 ret = -EAGAIN;
660 break;
661 }
662 if (signal_pending(current)) {
663 ret = -ERESTARTSYS;
664 break;
665 }
666 /* Nothing to read, let's sleep */
667 schedule();
668 continue;
669 }
670 ret = macvtap_put_user(q, skb, iv, len);
671 kfree_skb(skb);
672 break;
673 }
674
675 current->state = TASK_RUNNING;
Eric Dumazet4a4771a2010-04-25 22:20:06 +0000676 remove_wait_queue(sk_sleep(&q->sk), &wait);
Arnd Bergmann501c7742010-02-18 05:46:50 +0000677 return ret;
678}
Arnd Bergmann20d29d72010-01-30 12:24:26 +0000679
Arnd Bergmann501c7742010-02-18 05:46:50 +0000680static ssize_t macvtap_aio_read(struct kiocb *iocb, const struct iovec *iv,
681 unsigned long count, loff_t pos)
682{
683 struct file *file = iocb->ki_filp;
684 struct macvtap_queue *q = file->private_data;
685 ssize_t len, ret = 0;
686
687 len = iov_length(iv, count);
688 if (len < 0) {
689 ret = -EINVAL;
690 goto out;
691 }
692
693 ret = macvtap_do_read(q, iocb, iv, len, file->f_flags & O_NONBLOCK);
694 ret = min_t(ssize_t, ret, len); /* XXX copied from tun.c. Why? */
Arnd Bergmann20d29d72010-01-30 12:24:26 +0000695out:
Arnd Bergmann20d29d72010-01-30 12:24:26 +0000696 return ret;
697}
698
699/*
700 * provide compatibility with generic tun/tap interface
701 */
702static long macvtap_ioctl(struct file *file, unsigned int cmd,
703 unsigned long arg)
704{
Arnd Bergmann02df55d2010-02-18 05:45:36 +0000705 struct macvtap_queue *q = file->private_data;
706 struct macvlan_dev *vlan;
Arnd Bergmann20d29d72010-01-30 12:24:26 +0000707 void __user *argp = (void __user *)arg;
708 struct ifreq __user *ifr = argp;
709 unsigned int __user *up = argp;
710 unsigned int u;
Michael S. Tsirkin55afbd02010-04-29 13:50:48 +0300711 int __user *sp = argp;
712 int s;
Arnd Bergmann02df55d2010-02-18 05:45:36 +0000713 int ret;
Arnd Bergmann20d29d72010-01-30 12:24:26 +0000714
715 switch (cmd) {
716 case TUNSETIFF:
717 /* ignore the name, just look at flags */
718 if (get_user(u, &ifr->ifr_flags))
719 return -EFAULT;
Arnd Bergmannb9fb9ee2010-02-18 05:48:17 +0000720
721 ret = 0;
722 if ((u & ~IFF_VNET_HDR) != (IFF_NO_PI | IFF_TAP))
723 ret = -EINVAL;
724 else
725 q->flags = u;
726
727 return ret;
Arnd Bergmann20d29d72010-01-30 12:24:26 +0000728
729 case TUNGETIFF:
Arnd Bergmann02df55d2010-02-18 05:45:36 +0000730 rcu_read_lock_bh();
731 vlan = rcu_dereference(q->vlan);
732 if (vlan)
733 dev_hold(vlan->dev);
734 rcu_read_unlock_bh();
Arnd Bergmann20d29d72010-01-30 12:24:26 +0000735
Arnd Bergmann02df55d2010-02-18 05:45:36 +0000736 if (!vlan)
737 return -ENOLINK;
738
739 ret = 0;
Arnd Bergmann20d29d72010-01-30 12:24:26 +0000740 if (copy_to_user(&ifr->ifr_name, q->vlan->dev->name, IFNAMSIZ) ||
Arnd Bergmannb9fb9ee2010-02-18 05:48:17 +0000741 put_user(q->flags, &ifr->ifr_flags))
Arnd Bergmann02df55d2010-02-18 05:45:36 +0000742 ret = -EFAULT;
743 dev_put(vlan->dev);
744 return ret;
Arnd Bergmann20d29d72010-01-30 12:24:26 +0000745
746 case TUNGETFEATURES:
Arnd Bergmannb9fb9ee2010-02-18 05:48:17 +0000747 if (put_user(IFF_TAP | IFF_NO_PI | IFF_VNET_HDR, up))
Arnd Bergmann20d29d72010-01-30 12:24:26 +0000748 return -EFAULT;
749 return 0;
750
751 case TUNSETSNDBUF:
752 if (get_user(u, up))
753 return -EFAULT;
754
Arnd Bergmann20d29d72010-01-30 12:24:26 +0000755 q->sk.sk_sndbuf = u;
Arnd Bergmann20d29d72010-01-30 12:24:26 +0000756 return 0;
757
Michael S. Tsirkin55afbd02010-04-29 13:50:48 +0300758 case TUNGETVNETHDRSZ:
759 s = q->vnet_hdr_sz;
760 if (put_user(s, sp))
761 return -EFAULT;
762 return 0;
763
764 case TUNSETVNETHDRSZ:
765 if (get_user(s, sp))
766 return -EFAULT;
767 if (s < (int)sizeof(struct virtio_net_hdr))
768 return -EINVAL;
769
770 q->vnet_hdr_sz = s;
771 return 0;
772
Arnd Bergmann20d29d72010-01-30 12:24:26 +0000773 case TUNSETOFFLOAD:
774 /* let the user check for future flags */
775 if (arg & ~(TUN_F_CSUM | TUN_F_TSO4 | TUN_F_TSO6 |
Arnd Bergmannb9fb9ee2010-02-18 05:48:17 +0000776 TUN_F_TSO_ECN | TUN_F_UFO))
Arnd Bergmann20d29d72010-01-30 12:24:26 +0000777 return -EINVAL;
778
Arnd Bergmannb9fb9ee2010-02-18 05:48:17 +0000779 /* TODO: only accept frames with the features that
780 got enabled for forwarded frames */
781 if (!(q->flags & IFF_VNET_HDR))
782 return -EINVAL;
Arnd Bergmann20d29d72010-01-30 12:24:26 +0000783 return 0;
784
785 default:
786 return -EINVAL;
787 }
788}
789
790#ifdef CONFIG_COMPAT
791static long macvtap_compat_ioctl(struct file *file, unsigned int cmd,
792 unsigned long arg)
793{
794 return macvtap_ioctl(file, cmd, (unsigned long)compat_ptr(arg));
795}
796#endif
797
798static const struct file_operations macvtap_fops = {
799 .owner = THIS_MODULE,
800 .open = macvtap_open,
801 .release = macvtap_release,
802 .aio_read = macvtap_aio_read,
803 .aio_write = macvtap_aio_write,
804 .poll = macvtap_poll,
805 .llseek = no_llseek,
806 .unlocked_ioctl = macvtap_ioctl,
807#ifdef CONFIG_COMPAT
808 .compat_ioctl = macvtap_compat_ioctl,
809#endif
810};
811
Arnd Bergmann501c7742010-02-18 05:46:50 +0000812static int macvtap_sendmsg(struct kiocb *iocb, struct socket *sock,
813 struct msghdr *m, size_t total_len)
814{
815 struct macvtap_queue *q = container_of(sock, struct macvtap_queue, sock);
816 return macvtap_get_user(q, m->msg_iov, total_len,
817 m->msg_flags & MSG_DONTWAIT);
818}
819
820static int macvtap_recvmsg(struct kiocb *iocb, struct socket *sock,
821 struct msghdr *m, size_t total_len,
822 int flags)
823{
824 struct macvtap_queue *q = container_of(sock, struct macvtap_queue, sock);
825 int ret;
826 if (flags & ~(MSG_DONTWAIT|MSG_TRUNC))
827 return -EINVAL;
828 ret = macvtap_do_read(q, iocb, m->msg_iov, total_len,
829 flags & MSG_DONTWAIT);
830 if (ret > total_len) {
831 m->msg_flags |= MSG_TRUNC;
832 ret = flags & MSG_TRUNC ? ret : total_len;
833 }
834 return ret;
835}
836
837/* Ops structure to mimic raw sockets with tun */
838static const struct proto_ops macvtap_socket_ops = {
839 .sendmsg = macvtap_sendmsg,
840 .recvmsg = macvtap_recvmsg,
841};
842
843/* Get an underlying socket object from tun file. Returns error unless file is
844 * attached to a device. The returned object works like a packet socket, it
845 * can be used for sock_sendmsg/sock_recvmsg. The caller is responsible for
846 * holding a reference to the file for as long as the socket is in use. */
847struct socket *macvtap_get_socket(struct file *file)
848{
849 struct macvtap_queue *q;
850 if (file->f_op != &macvtap_fops)
851 return ERR_PTR(-EINVAL);
852 q = file->private_data;
853 if (!q)
854 return ERR_PTR(-EBADFD);
855 return &q->sock;
856}
857EXPORT_SYMBOL_GPL(macvtap_get_socket);
858
Arnd Bergmann20d29d72010-01-30 12:24:26 +0000859static int macvtap_init(void)
860{
861 int err;
862
863 err = alloc_chrdev_region(&macvtap_major, 0,
864 MACVTAP_NUM_DEVS, "macvtap");
865 if (err)
866 goto out1;
867
868 cdev_init(&macvtap_cdev, &macvtap_fops);
869 err = cdev_add(&macvtap_cdev, macvtap_major, MACVTAP_NUM_DEVS);
870 if (err)
871 goto out2;
872
873 macvtap_class = class_create(THIS_MODULE, "macvtap");
874 if (IS_ERR(macvtap_class)) {
875 err = PTR_ERR(macvtap_class);
876 goto out3;
877 }
878
879 err = macvlan_link_register(&macvtap_link_ops);
880 if (err)
881 goto out4;
882
883 return 0;
884
885out4:
886 class_unregister(macvtap_class);
887out3:
888 cdev_del(&macvtap_cdev);
889out2:
890 unregister_chrdev_region(macvtap_major, MACVTAP_NUM_DEVS);
891out1:
892 return err;
893}
894module_init(macvtap_init);
895
896static void macvtap_exit(void)
897{
898 rtnl_link_unregister(&macvtap_link_ops);
899 class_unregister(macvtap_class);
900 cdev_del(&macvtap_cdev);
901 unregister_chrdev_region(macvtap_major, MACVTAP_NUM_DEVS);
902}
903module_exit(macvtap_exit);
904
905MODULE_ALIAS_RTNL_LINK("macvtap");
906MODULE_AUTHOR("Arnd Bergmann <arnd@arndb.de>");
907MODULE_LICENSE("GPL");