blob: cb8fd5069dbe3971a001ea459a085b4d64657fa8 [file] [log] [blame]
Arnd Bergmann20d29d72010-01-30 12:24:26 +00001#include <linux/etherdevice.h>
2#include <linux/if_macvlan.h>
Basil Gorf09e2242012-05-03 22:55:24 +00003#include <linux/if_vlan.h>
Arnd Bergmann20d29d72010-01-30 12:24:26 +00004#include <linux/interrupt.h>
5#include <linux/nsproxy.h>
6#include <linux/compat.h>
7#include <linux/if_tun.h>
8#include <linux/module.h>
9#include <linux/skbuff.h>
10#include <linux/cache.h>
11#include <linux/sched.h>
12#include <linux/types.h>
Tejun Heo5a0e3ad2010-03-24 17:04:11 +090013#include <linux/slab.h>
Arnd Bergmann20d29d72010-01-30 12:24:26 +000014#include <linux/init.h>
15#include <linux/wait.h>
16#include <linux/cdev.h>
Al Viro40401532012-02-13 03:58:52 +000017#include <linux/idr.h>
Arnd Bergmann20d29d72010-01-30 12:24:26 +000018#include <linux/fs.h>
19
20#include <net/net_namespace.h>
21#include <net/rtnetlink.h>
22#include <net/sock.h>
Arnd Bergmannb9fb9ee2010-02-18 05:48:17 +000023#include <linux/virtio_net.h>
Arnd Bergmann20d29d72010-01-30 12:24:26 +000024
25/*
26 * A macvtap queue is the central object of this driver, it connects
27 * an open character device to a macvlan interface. There can be
28 * multiple queues on one interface, which map back to queues
29 * implemented in hardware on the underlying device.
30 *
31 * macvtap_proto is used to allocate queues through the sock allocation
32 * mechanism.
33 *
34 * TODO: multiqueue support is currently not implemented, even though
35 * macvtap is basically prepared for that. We will need to add this
36 * here as well as in virtio-net and qemu to get line rate on 10gbit
37 * adapters from a guest.
38 */
39struct macvtap_queue {
40 struct sock sk;
41 struct socket sock;
Eric Dumazet43815482010-04-29 11:01:49 +000042 struct socket_wq wq;
Michael S. Tsirkin55afbd02010-04-29 13:50:48 +030043 int vnet_hdr_sz;
Eric Dumazet13707f92011-01-26 19:28:23 +000044 struct macvlan_dev __rcu *vlan;
Arnd Bergmann20d29d72010-01-30 12:24:26 +000045 struct file *file;
Arnd Bergmannb9fb9ee2010-02-18 05:48:17 +000046 unsigned int flags;
Arnd Bergmann20d29d72010-01-30 12:24:26 +000047};
48
49static struct proto macvtap_proto = {
50 .name = "macvtap",
51 .owner = THIS_MODULE,
52 .obj_size = sizeof (struct macvtap_queue),
53};
54
55/*
Eric W. Biedermane09eff72011-10-20 04:29:24 +000056 * Variables for dealing with macvtaps device numbers.
Arnd Bergmann20d29d72010-01-30 12:24:26 +000057 */
David S. Miller1ebed712010-07-10 19:25:50 -070058static dev_t macvtap_major;
Eric W. Biedermane09eff72011-10-20 04:29:24 +000059#define MACVTAP_NUM_DEVS (1U << MINORBITS)
60static DEFINE_MUTEX(minor_lock);
61static DEFINE_IDR(minor_idr);
62
Shirley Ma97bc3632011-07-06 12:26:11 +000063#define GOODCOPY_LEN 128
Arnd Bergmann20d29d72010-01-30 12:24:26 +000064static struct class *macvtap_class;
65static struct cdev macvtap_cdev;
66
Arnd Bergmann501c7742010-02-18 05:46:50 +000067static const struct proto_ops macvtap_socket_ops;
68
Arnd Bergmann20d29d72010-01-30 12:24:26 +000069/*
70 * RCU usage:
Arnd Bergmann02df55d2010-02-18 05:45:36 +000071 * The macvtap_queue and the macvlan_dev are loosely coupled, the
72 * pointers from one to the other can only be read while rcu_read_lock
73 * or macvtap_lock is held.
Arnd Bergmann20d29d72010-01-30 12:24:26 +000074 *
Arnd Bergmann02df55d2010-02-18 05:45:36 +000075 * Both the file and the macvlan_dev hold a reference on the macvtap_queue
76 * through sock_hold(&q->sk). When the macvlan_dev goes away first,
77 * q->vlan becomes inaccessible. When the files gets closed,
78 * macvtap_get_queue() fails.
Arnd Bergmann20d29d72010-01-30 12:24:26 +000079 *
Arnd Bergmann02df55d2010-02-18 05:45:36 +000080 * There may still be references to the struct sock inside of the
81 * queue from outbound SKBs, but these never reference back to the
82 * file or the dev. The data structure is freed through __sk_free
83 * when both our references and any pending SKBs are gone.
Arnd Bergmann20d29d72010-01-30 12:24:26 +000084 */
85static DEFINE_SPINLOCK(macvtap_lock);
86
87/*
Krishna Kumar1565c7c2010-08-04 06:15:59 +000088 * get_slot: return a [unused/occupied] slot in vlan->taps[]:
89 * - if 'q' is NULL, return the first empty slot;
90 * - otherwise, return the slot this pointer occupies.
Arnd Bergmann20d29d72010-01-30 12:24:26 +000091 */
Krishna Kumar1565c7c2010-08-04 06:15:59 +000092static int get_slot(struct macvlan_dev *vlan, struct macvtap_queue *q)
93{
94 int i;
95
96 for (i = 0; i < MAX_MACVTAP_QUEUES; i++) {
97 if (rcu_dereference(vlan->taps[i]) == q)
98 return i;
99 }
100
101 /* Should never happen */
102 BUG_ON(1);
103}
104
Arnd Bergmann20d29d72010-01-30 12:24:26 +0000105static int macvtap_set_queue(struct net_device *dev, struct file *file,
106 struct macvtap_queue *q)
107{
108 struct macvlan_dev *vlan = netdev_priv(dev);
Krishna Kumar1565c7c2010-08-04 06:15:59 +0000109 int index;
Arnd Bergmann20d29d72010-01-30 12:24:26 +0000110 int err = -EBUSY;
111
112 spin_lock(&macvtap_lock);
Krishna Kumar1565c7c2010-08-04 06:15:59 +0000113 if (vlan->numvtaps == MAX_MACVTAP_QUEUES)
Arnd Bergmann20d29d72010-01-30 12:24:26 +0000114 goto out;
115
116 err = 0;
Krishna Kumar1565c7c2010-08-04 06:15:59 +0000117 index = get_slot(vlan, NULL);
Arnd Bergmann02df55d2010-02-18 05:45:36 +0000118 rcu_assign_pointer(q->vlan, vlan);
Krishna Kumar1565c7c2010-08-04 06:15:59 +0000119 rcu_assign_pointer(vlan->taps[index], q);
Arnd Bergmann02df55d2010-02-18 05:45:36 +0000120 sock_hold(&q->sk);
Arnd Bergmann20d29d72010-01-30 12:24:26 +0000121
122 q->file = file;
Arnd Bergmann02df55d2010-02-18 05:45:36 +0000123 file->private_data = q;
Arnd Bergmann20d29d72010-01-30 12:24:26 +0000124
Krishna Kumar1565c7c2010-08-04 06:15:59 +0000125 vlan->numvtaps++;
126
Arnd Bergmann20d29d72010-01-30 12:24:26 +0000127out:
128 spin_unlock(&macvtap_lock);
129 return err;
130}
131
132/*
Arnd Bergmann02df55d2010-02-18 05:45:36 +0000133 * The file owning the queue got closed, give up both
134 * the reference that the files holds as well as the
135 * one from the macvlan_dev if that still exists.
Arnd Bergmann20d29d72010-01-30 12:24:26 +0000136 *
137 * Using the spinlock makes sure that we don't get
138 * to the queue again after destroying it.
Arnd Bergmann20d29d72010-01-30 12:24:26 +0000139 */
Arnd Bergmann02df55d2010-02-18 05:45:36 +0000140static void macvtap_put_queue(struct macvtap_queue *q)
Arnd Bergmann20d29d72010-01-30 12:24:26 +0000141{
Arnd Bergmann02df55d2010-02-18 05:45:36 +0000142 struct macvlan_dev *vlan;
Arnd Bergmann20d29d72010-01-30 12:24:26 +0000143
144 spin_lock(&macvtap_lock);
Eric Dumazet13707f92011-01-26 19:28:23 +0000145 vlan = rcu_dereference_protected(q->vlan,
146 lockdep_is_held(&macvtap_lock));
Arnd Bergmann02df55d2010-02-18 05:45:36 +0000147 if (vlan) {
Krishna Kumar1565c7c2010-08-04 06:15:59 +0000148 int index = get_slot(vlan, q);
149
Eric Dumazet2cfa5a02011-11-23 07:09:32 +0000150 RCU_INIT_POINTER(vlan->taps[index], NULL);
151 RCU_INIT_POINTER(q->vlan, NULL);
Arnd Bergmann02df55d2010-02-18 05:45:36 +0000152 sock_put(&q->sk);
Krishna Kumar1565c7c2010-08-04 06:15:59 +0000153 --vlan->numvtaps;
Arnd Bergmann20d29d72010-01-30 12:24:26 +0000154 }
155
Arnd Bergmann20d29d72010-01-30 12:24:26 +0000156 spin_unlock(&macvtap_lock);
157
158 synchronize_rcu();
159 sock_put(&q->sk);
160}
161
162/*
Krishna Kumar1565c7c2010-08-04 06:15:59 +0000163 * Select a queue based on the rxq of the device on which this packet
164 * arrived. If the incoming device is not mq, calculate a flow hash
165 * to select a queue. If all fails, find the first available queue.
166 * Cache vlan->numvtaps since it can become zero during the execution
167 * of this function.
Arnd Bergmann20d29d72010-01-30 12:24:26 +0000168 */
169static struct macvtap_queue *macvtap_get_queue(struct net_device *dev,
170 struct sk_buff *skb)
171{
172 struct macvlan_dev *vlan = netdev_priv(dev);
Krishna Kumar1565c7c2010-08-04 06:15:59 +0000173 struct macvtap_queue *tap = NULL;
174 int numvtaps = vlan->numvtaps;
175 __u32 rxq;
Arnd Bergmann20d29d72010-01-30 12:24:26 +0000176
Krishna Kumar1565c7c2010-08-04 06:15:59 +0000177 if (!numvtaps)
178 goto out;
179
Krishna Kumaref0002b2011-11-23 22:17:14 +0000180 /* Check if we can use flow to select a queue */
181 rxq = skb_get_rxhash(skb);
182 if (rxq) {
183 tap = rcu_dereference(vlan->taps[rxq % numvtaps]);
184 if (tap)
185 goto out;
186 }
187
Krishna Kumar1565c7c2010-08-04 06:15:59 +0000188 if (likely(skb_rx_queue_recorded(skb))) {
189 rxq = skb_get_rx_queue(skb);
190
191 while (unlikely(rxq >= numvtaps))
192 rxq -= numvtaps;
193
194 tap = rcu_dereference(vlan->taps[rxq]);
195 if (tap)
196 goto out;
197 }
198
Krishna Kumar1565c7c2010-08-04 06:15:59 +0000199 /* Everything failed - find first available queue */
200 for (rxq = 0; rxq < MAX_MACVTAP_QUEUES; rxq++) {
201 tap = rcu_dereference(vlan->taps[rxq]);
202 if (tap)
203 break;
204 }
205
206out:
207 return tap;
Arnd Bergmann20d29d72010-01-30 12:24:26 +0000208}
209
Arnd Bergmann02df55d2010-02-18 05:45:36 +0000210/*
211 * The net_device is going away, give up the reference
Krishna Kumar1565c7c2010-08-04 06:15:59 +0000212 * that it holds on all queues and safely set the pointer
213 * from the queues to NULL.
Arnd Bergmann02df55d2010-02-18 05:45:36 +0000214 */
Arnd Bergmann20d29d72010-01-30 12:24:26 +0000215static void macvtap_del_queues(struct net_device *dev)
216{
217 struct macvlan_dev *vlan = netdev_priv(dev);
Krishna Kumar1565c7c2010-08-04 06:15:59 +0000218 struct macvtap_queue *q, *qlist[MAX_MACVTAP_QUEUES];
219 int i, j = 0;
Arnd Bergmann20d29d72010-01-30 12:24:26 +0000220
Krishna Kumar1565c7c2010-08-04 06:15:59 +0000221 /* macvtap_put_queue can free some slots, so go through all slots */
Arnd Bergmann02df55d2010-02-18 05:45:36 +0000222 spin_lock(&macvtap_lock);
Krishna Kumar1565c7c2010-08-04 06:15:59 +0000223 for (i = 0; i < MAX_MACVTAP_QUEUES && vlan->numvtaps; i++) {
Eric Dumazet13707f92011-01-26 19:28:23 +0000224 q = rcu_dereference_protected(vlan->taps[i],
225 lockdep_is_held(&macvtap_lock));
Krishna Kumar1565c7c2010-08-04 06:15:59 +0000226 if (q) {
227 qlist[j++] = q;
Eric Dumazet2cfa5a02011-11-23 07:09:32 +0000228 RCU_INIT_POINTER(vlan->taps[i], NULL);
229 RCU_INIT_POINTER(q->vlan, NULL);
Krishna Kumar1565c7c2010-08-04 06:15:59 +0000230 vlan->numvtaps--;
231 }
Arnd Bergmann02df55d2010-02-18 05:45:36 +0000232 }
Krishna Kumar1565c7c2010-08-04 06:15:59 +0000233 BUG_ON(vlan->numvtaps != 0);
Eric W. Biederman99f34b32011-10-20 04:26:01 +0000234 /* guarantee that any future macvtap_set_queue will fail */
235 vlan->numvtaps = MAX_MACVTAP_QUEUES;
Arnd Bergmann02df55d2010-02-18 05:45:36 +0000236 spin_unlock(&macvtap_lock);
237
238 synchronize_rcu();
Krishna Kumar1565c7c2010-08-04 06:15:59 +0000239
240 for (--j; j >= 0; j--)
241 sock_put(&qlist[j]->sk);
Arnd Bergmann20d29d72010-01-30 12:24:26 +0000242}
243
244/*
245 * Forward happens for data that gets sent from one macvlan
246 * endpoint to another one in bridge mode. We just take
247 * the skb and put it into the receive queue.
248 */
249static int macvtap_forward(struct net_device *dev, struct sk_buff *skb)
250{
251 struct macvtap_queue *q = macvtap_get_queue(dev, skb);
252 if (!q)
Herbert Xu8a357472010-07-21 21:44:31 +0000253 goto drop;
254
255 if (skb_queue_len(&q->sk.sk_receive_queue) >= dev->tx_queue_len)
256 goto drop;
Arnd Bergmann20d29d72010-01-30 12:24:26 +0000257
258 skb_queue_tail(&q->sk.sk_receive_queue, skb);
Eric Dumazet4a4771a2010-04-25 22:20:06 +0000259 wake_up_interruptible_poll(sk_sleep(&q->sk), POLLIN | POLLRDNORM | POLLRDBAND);
Herbert Xu8a357472010-07-21 21:44:31 +0000260 return NET_RX_SUCCESS;
261
262drop:
263 kfree_skb(skb);
264 return NET_RX_DROP;
Arnd Bergmann20d29d72010-01-30 12:24:26 +0000265}
266
267/*
268 * Receive is for data from the external interface (lowerdev),
269 * in case of macvtap, we can treat that the same way as
270 * forward, which macvlan cannot.
271 */
272static int macvtap_receive(struct sk_buff *skb)
273{
274 skb_push(skb, ETH_HLEN);
275 return macvtap_forward(skb->dev, skb);
276}
277
Eric W. Biedermane09eff72011-10-20 04:29:24 +0000278static int macvtap_get_minor(struct macvlan_dev *vlan)
279{
280 int retval = -ENOMEM;
281 int id;
282
283 mutex_lock(&minor_lock);
284 if (idr_pre_get(&minor_idr, GFP_KERNEL) == 0)
285 goto exit;
286
287 retval = idr_get_new_above(&minor_idr, vlan, 1, &id);
288 if (retval < 0) {
289 if (retval == -EAGAIN)
290 retval = -ENOMEM;
291 goto exit;
292 }
293 if (id < MACVTAP_NUM_DEVS) {
294 vlan->minor = id;
295 } else {
296 printk(KERN_ERR "too many macvtap devices\n");
297 retval = -EINVAL;
298 idr_remove(&minor_idr, id);
299 }
300exit:
301 mutex_unlock(&minor_lock);
302 return retval;
303}
304
305static void macvtap_free_minor(struct macvlan_dev *vlan)
306{
307 mutex_lock(&minor_lock);
308 if (vlan->minor) {
309 idr_remove(&minor_idr, vlan->minor);
310 vlan->minor = 0;
311 }
312 mutex_unlock(&minor_lock);
313}
314
315static struct net_device *dev_get_by_macvtap_minor(int minor)
316{
317 struct net_device *dev = NULL;
318 struct macvlan_dev *vlan;
319
320 mutex_lock(&minor_lock);
321 vlan = idr_find(&minor_idr, minor);
322 if (vlan) {
323 dev = vlan->dev;
324 dev_hold(dev);
325 }
326 mutex_unlock(&minor_lock);
327 return dev;
328}
329
Arnd Bergmann20d29d72010-01-30 12:24:26 +0000330static int macvtap_newlink(struct net *src_net,
331 struct net_device *dev,
332 struct nlattr *tb[],
333 struct nlattr *data[])
334{
Eric W. Biederman9bf19072011-10-20 04:28:46 +0000335 /* Don't put anything that may fail after macvlan_common_newlink
336 * because we can't undo what it does.
337 */
338 return macvlan_common_newlink(src_net, dev, tb, data,
339 macvtap_receive, macvtap_forward);
Arnd Bergmann20d29d72010-01-30 12:24:26 +0000340}
341
342static void macvtap_dellink(struct net_device *dev,
343 struct list_head *head)
344{
Arnd Bergmann20d29d72010-01-30 12:24:26 +0000345 macvtap_del_queues(dev);
346 macvlan_dellink(dev, head);
347}
348
Herbert Xu8a357472010-07-21 21:44:31 +0000349static void macvtap_setup(struct net_device *dev)
350{
351 macvlan_common_setup(dev);
352 dev->tx_queue_len = TUN_READQ_SIZE;
353}
354
Arnd Bergmann20d29d72010-01-30 12:24:26 +0000355static struct rtnl_link_ops macvtap_link_ops __read_mostly = {
356 .kind = "macvtap",
Herbert Xu8a357472010-07-21 21:44:31 +0000357 .setup = macvtap_setup,
Arnd Bergmann20d29d72010-01-30 12:24:26 +0000358 .newlink = macvtap_newlink,
359 .dellink = macvtap_dellink,
360};
361
362
363static void macvtap_sock_write_space(struct sock *sk)
364{
Eric Dumazet43815482010-04-29 11:01:49 +0000365 wait_queue_head_t *wqueue;
366
Arnd Bergmann20d29d72010-01-30 12:24:26 +0000367 if (!sock_writeable(sk) ||
368 !test_and_clear_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags))
369 return;
370
Eric Dumazet43815482010-04-29 11:01:49 +0000371 wqueue = sk_sleep(sk);
372 if (wqueue && waitqueue_active(wqueue))
373 wake_up_interruptible_poll(wqueue, POLLOUT | POLLWRNORM | POLLWRBAND);
Arnd Bergmann20d29d72010-01-30 12:24:26 +0000374}
375
Eric W. Biederman2259fef2011-10-20 04:27:24 +0000376static void macvtap_sock_destruct(struct sock *sk)
377{
378 skb_queue_purge(&sk->sk_receive_queue);
379}
380
Arnd Bergmann20d29d72010-01-30 12:24:26 +0000381static int macvtap_open(struct inode *inode, struct file *file)
382{
383 struct net *net = current->nsproxy->net_ns;
Eric W. Biedermane09eff72011-10-20 04:29:24 +0000384 struct net_device *dev = dev_get_by_macvtap_minor(iminor(inode));
Arnd Bergmann20d29d72010-01-30 12:24:26 +0000385 struct macvtap_queue *q;
386 int err;
387
388 err = -ENODEV;
389 if (!dev)
390 goto out;
391
Arnd Bergmann20d29d72010-01-30 12:24:26 +0000392 err = -ENOMEM;
393 q = (struct macvtap_queue *)sk_alloc(net, AF_UNSPEC, GFP_KERNEL,
394 &macvtap_proto);
395 if (!q)
396 goto out;
397
Eric Dumazet43815482010-04-29 11:01:49 +0000398 q->sock.wq = &q->wq;
399 init_waitqueue_head(&q->wq.wait);
Arnd Bergmann20d29d72010-01-30 12:24:26 +0000400 q->sock.type = SOCK_RAW;
401 q->sock.state = SS_CONNECTED;
Arnd Bergmann501c7742010-02-18 05:46:50 +0000402 q->sock.file = file;
403 q->sock.ops = &macvtap_socket_ops;
Arnd Bergmann20d29d72010-01-30 12:24:26 +0000404 sock_init_data(&q->sock, &q->sk);
Arnd Bergmann20d29d72010-01-30 12:24:26 +0000405 q->sk.sk_write_space = macvtap_sock_write_space;
Eric W. Biederman2259fef2011-10-20 04:27:24 +0000406 q->sk.sk_destruct = macvtap_sock_destruct;
Arnd Bergmannb9fb9ee2010-02-18 05:48:17 +0000407 q->flags = IFF_VNET_HDR | IFF_NO_PI | IFF_TAP;
Michael S. Tsirkin55afbd02010-04-29 13:50:48 +0300408 q->vnet_hdr_sz = sizeof(struct virtio_net_hdr);
Arnd Bergmann20d29d72010-01-30 12:24:26 +0000409
Shirley Ma97bc3632011-07-06 12:26:11 +0000410 /*
411 * so far only KVM virtio_net uses macvtap, enable zero copy between
412 * guest kernel and host kernel when lower device supports zerocopy
Eric W. Biederman047af9cf2011-10-20 04:26:39 +0000413 *
414 * The macvlan supports zerocopy iff the lower device supports zero
415 * copy so we don't have to look at the lower device directly.
Shirley Ma97bc3632011-07-06 12:26:11 +0000416 */
Eric W. Biederman047af9cf2011-10-20 04:26:39 +0000417 if ((dev->features & NETIF_F_HIGHDMA) && (dev->features & NETIF_F_SG))
418 sock_set_flag(&q->sk, SOCK_ZEROCOPY);
Shirley Ma97bc3632011-07-06 12:26:11 +0000419
Arnd Bergmann20d29d72010-01-30 12:24:26 +0000420 err = macvtap_set_queue(dev, file, q);
421 if (err)
422 sock_put(&q->sk);
423
424out:
425 if (dev)
426 dev_put(dev);
427
428 return err;
429}
430
431static int macvtap_release(struct inode *inode, struct file *file)
432{
Arnd Bergmann02df55d2010-02-18 05:45:36 +0000433 struct macvtap_queue *q = file->private_data;
434 macvtap_put_queue(q);
Arnd Bergmann20d29d72010-01-30 12:24:26 +0000435 return 0;
436}
437
438static unsigned int macvtap_poll(struct file *file, poll_table * wait)
439{
Arnd Bergmann02df55d2010-02-18 05:45:36 +0000440 struct macvtap_queue *q = file->private_data;
Arnd Bergmann20d29d72010-01-30 12:24:26 +0000441 unsigned int mask = POLLERR;
442
443 if (!q)
444 goto out;
445
446 mask = 0;
Eric Dumazet43815482010-04-29 11:01:49 +0000447 poll_wait(file, &q->wq.wait, wait);
Arnd Bergmann20d29d72010-01-30 12:24:26 +0000448
449 if (!skb_queue_empty(&q->sk.sk_receive_queue))
450 mask |= POLLIN | POLLRDNORM;
451
452 if (sock_writeable(&q->sk) ||
453 (!test_and_set_bit(SOCK_ASYNC_NOSPACE, &q->sock.flags) &&
454 sock_writeable(&q->sk)))
455 mask |= POLLOUT | POLLWRNORM;
456
457out:
Arnd Bergmann20d29d72010-01-30 12:24:26 +0000458 return mask;
459}
460
Arnd Bergmannb9fb9ee2010-02-18 05:48:17 +0000461static inline struct sk_buff *macvtap_alloc_skb(struct sock *sk, size_t prepad,
462 size_t len, size_t linear,
463 int noblock, int *err)
464{
465 struct sk_buff *skb;
466
467 /* Under a page? Don't bother with paged skb. */
468 if (prepad + len < PAGE_SIZE || !linear)
469 linear = len;
470
471 skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock,
472 err);
473 if (!skb)
474 return NULL;
475
476 skb_reserve(skb, prepad);
477 skb_put(skb, linear);
478 skb->data_len = len - linear;
479 skb->len += len - linear;
480
481 return skb;
482}
483
Shirley Ma97bc3632011-07-06 12:26:11 +0000484/* set skb frags from iovec, this can move to core network code for reuse */
485static int zerocopy_sg_from_iovec(struct sk_buff *skb, const struct iovec *from,
486 int offset, size_t count)
487{
488 int len = iov_length(from, count) - offset;
489 int copy = skb_headlen(skb);
490 int size, offset1 = 0;
491 int i = 0;
Shirley Ma97bc3632011-07-06 12:26:11 +0000492
493 /* Skip over from offset */
494 while (count && (offset >= from->iov_len)) {
495 offset -= from->iov_len;
496 ++from;
497 --count;
498 }
499
500 /* copy up to skb headlen */
501 while (count && (copy > 0)) {
502 size = min_t(unsigned int, copy, from->iov_len - offset);
503 if (copy_from_user(skb->data + offset1, from->iov_base + offset,
504 size))
505 return -EFAULT;
506 if (copy > size) {
507 ++from;
508 --count;
509 }
510 copy -= size;
511 offset1 += size;
512 offset = 0;
513 }
514
515 if (len == offset1)
516 return 0;
517
518 while (count--) {
519 struct page *page[MAX_SKB_FRAGS];
520 int num_pages;
521 unsigned long base;
522
523 len = from->iov_len - offset1;
524 if (!len) {
525 offset1 = 0;
526 ++from;
527 continue;
528 }
529 base = (unsigned long)from->iov_base + offset1;
530 size = ((base & ~PAGE_MASK) + len + ~PAGE_MASK) >> PAGE_SHIFT;
531 num_pages = get_user_pages_fast(base, size, 0, &page[i]);
532 if ((num_pages != size) ||
533 (num_pages > MAX_SKB_FRAGS - skb_shinfo(skb)->nr_frags))
534 /* put_page is in skb free */
535 return -EFAULT;
536 skb->data_len += len;
537 skb->len += len;
538 skb->truesize += len;
539 atomic_add(len, &skb->sk->sk_wmem_alloc);
540 while (len) {
Jason Wang653fc912011-09-18 23:48:31 +0000541 int off = base & ~PAGE_MASK;
542 int size = min_t(int, len, PAGE_SIZE - off);
543 __skb_fill_page_desc(skb, i, page[i], off, size);
Shirley Ma97bc3632011-07-06 12:26:11 +0000544 skb_shinfo(skb)->nr_frags++;
545 /* increase sk_wmem_alloc */
Jason Wang653fc912011-09-18 23:48:31 +0000546 base += size;
547 len -= size;
Shirley Ma97bc3632011-07-06 12:26:11 +0000548 i++;
549 }
550 offset1 = 0;
551 ++from;
552 }
553 return 0;
554}
555
Arnd Bergmannb9fb9ee2010-02-18 05:48:17 +0000556/*
557 * macvtap_skb_from_vnet_hdr and macvtap_skb_to_vnet_hdr should
558 * be shared with the tun/tap driver.
559 */
560static int macvtap_skb_from_vnet_hdr(struct sk_buff *skb,
561 struct virtio_net_hdr *vnet_hdr)
562{
563 unsigned short gso_type = 0;
564 if (vnet_hdr->gso_type != VIRTIO_NET_HDR_GSO_NONE) {
565 switch (vnet_hdr->gso_type & ~VIRTIO_NET_HDR_GSO_ECN) {
566 case VIRTIO_NET_HDR_GSO_TCPV4:
567 gso_type = SKB_GSO_TCPV4;
568 break;
569 case VIRTIO_NET_HDR_GSO_TCPV6:
570 gso_type = SKB_GSO_TCPV6;
571 break;
572 case VIRTIO_NET_HDR_GSO_UDP:
573 gso_type = SKB_GSO_UDP;
574 break;
575 default:
576 return -EINVAL;
577 }
578
579 if (vnet_hdr->gso_type & VIRTIO_NET_HDR_GSO_ECN)
580 gso_type |= SKB_GSO_TCP_ECN;
581
582 if (vnet_hdr->gso_size == 0)
583 return -EINVAL;
584 }
585
586 if (vnet_hdr->flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) {
587 if (!skb_partial_csum_set(skb, vnet_hdr->csum_start,
588 vnet_hdr->csum_offset))
589 return -EINVAL;
590 }
591
592 if (vnet_hdr->gso_type != VIRTIO_NET_HDR_GSO_NONE) {
593 skb_shinfo(skb)->gso_size = vnet_hdr->gso_size;
594 skb_shinfo(skb)->gso_type = gso_type;
595
596 /* Header must be checked, and gso_segs computed. */
597 skb_shinfo(skb)->gso_type |= SKB_GSO_DODGY;
598 skb_shinfo(skb)->gso_segs = 0;
599 }
600 return 0;
601}
602
603static int macvtap_skb_to_vnet_hdr(const struct sk_buff *skb,
604 struct virtio_net_hdr *vnet_hdr)
605{
606 memset(vnet_hdr, 0, sizeof(*vnet_hdr));
607
608 if (skb_is_gso(skb)) {
609 struct skb_shared_info *sinfo = skb_shinfo(skb);
610
611 /* This is a hint as to how much should be linear. */
612 vnet_hdr->hdr_len = skb_headlen(skb);
613 vnet_hdr->gso_size = sinfo->gso_size;
614 if (sinfo->gso_type & SKB_GSO_TCPV4)
615 vnet_hdr->gso_type = VIRTIO_NET_HDR_GSO_TCPV4;
616 else if (sinfo->gso_type & SKB_GSO_TCPV6)
617 vnet_hdr->gso_type = VIRTIO_NET_HDR_GSO_TCPV6;
618 else if (sinfo->gso_type & SKB_GSO_UDP)
619 vnet_hdr->gso_type = VIRTIO_NET_HDR_GSO_UDP;
620 else
621 BUG();
622 if (sinfo->gso_type & SKB_GSO_TCP_ECN)
623 vnet_hdr->gso_type |= VIRTIO_NET_HDR_GSO_ECN;
624 } else
625 vnet_hdr->gso_type = VIRTIO_NET_HDR_GSO_NONE;
626
627 if (skb->ip_summed == CHECKSUM_PARTIAL) {
628 vnet_hdr->flags = VIRTIO_NET_HDR_F_NEEDS_CSUM;
Michał Mirosław55508d62010-12-14 15:24:08 +0000629 vnet_hdr->csum_start = skb_checksum_start_offset(skb);
Arnd Bergmannb9fb9ee2010-02-18 05:48:17 +0000630 vnet_hdr->csum_offset = skb->csum_offset;
Jason Wang10a8d942011-06-10 00:56:17 +0000631 } else if (skb->ip_summed == CHECKSUM_UNNECESSARY) {
632 vnet_hdr->flags = VIRTIO_NET_HDR_F_DATA_VALID;
Arnd Bergmannb9fb9ee2010-02-18 05:48:17 +0000633 } /* else everything is zero */
634
635 return 0;
636}
637
638
Arnd Bergmann20d29d72010-01-30 12:24:26 +0000639/* Get packet from user space buffer */
Shirley Ma97bc3632011-07-06 12:26:11 +0000640static ssize_t macvtap_get_user(struct macvtap_queue *q, struct msghdr *m,
641 const struct iovec *iv, unsigned long total_len,
642 size_t count, int noblock)
Arnd Bergmann20d29d72010-01-30 12:24:26 +0000643{
644 struct sk_buff *skb;
Arnd Bergmann02df55d2010-02-18 05:45:36 +0000645 struct macvlan_dev *vlan;
Shirley Ma97bc3632011-07-06 12:26:11 +0000646 unsigned long len = total_len;
Arnd Bergmann20d29d72010-01-30 12:24:26 +0000647 int err;
Arnd Bergmannb9fb9ee2010-02-18 05:48:17 +0000648 struct virtio_net_hdr vnet_hdr = { 0 };
649 int vnet_hdr_len = 0;
Shirley Ma97bc3632011-07-06 12:26:11 +0000650 int copylen;
651 bool zerocopy = false;
Arnd Bergmann20d29d72010-01-30 12:24:26 +0000652
Arnd Bergmannb9fb9ee2010-02-18 05:48:17 +0000653 if (q->flags & IFF_VNET_HDR) {
Michael S. Tsirkin55afbd02010-04-29 13:50:48 +0300654 vnet_hdr_len = q->vnet_hdr_sz;
Arnd Bergmannb9fb9ee2010-02-18 05:48:17 +0000655
656 err = -EINVAL;
Nicolas Kaiserce3c8692011-03-04 13:49:41 +0000657 if (len < vnet_hdr_len)
Arnd Bergmannb9fb9ee2010-02-18 05:48:17 +0000658 goto err;
Nicolas Kaiserce3c8692011-03-04 13:49:41 +0000659 len -= vnet_hdr_len;
Arnd Bergmannb9fb9ee2010-02-18 05:48:17 +0000660
661 err = memcpy_fromiovecend((void *)&vnet_hdr, iv, 0,
Michael S. Tsirkin55afbd02010-04-29 13:50:48 +0300662 sizeof(vnet_hdr));
Arnd Bergmannb9fb9ee2010-02-18 05:48:17 +0000663 if (err < 0)
664 goto err;
665 if ((vnet_hdr.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) &&
666 vnet_hdr.csum_start + vnet_hdr.csum_offset + 2 >
667 vnet_hdr.hdr_len)
668 vnet_hdr.hdr_len = vnet_hdr.csum_start +
669 vnet_hdr.csum_offset + 2;
670 err = -EINVAL;
671 if (vnet_hdr.hdr_len > len)
672 goto err;
673 }
674
675 err = -EINVAL;
Arnd Bergmann20d29d72010-01-30 12:24:26 +0000676 if (unlikely(len < ETH_HLEN))
Arnd Bergmannb9fb9ee2010-02-18 05:48:17 +0000677 goto err;
Arnd Bergmann20d29d72010-01-30 12:24:26 +0000678
Shirley Ma97bc3632011-07-06 12:26:11 +0000679 if (m && m->msg_control && sock_flag(&q->sk, SOCK_ZEROCOPY))
680 zerocopy = true;
681
682 if (zerocopy) {
683 /* There are 256 bytes to be copied in skb, so there is enough
684 * room for skb expand head in case it is used.
685 * The rest buffer is mapped from userspace.
686 */
687 copylen = vnet_hdr.hdr_len;
688 if (!copylen)
689 copylen = GOODCOPY_LEN;
690 } else
691 copylen = len;
692
693 skb = macvtap_alloc_skb(&q->sk, NET_IP_ALIGN, copylen,
694 vnet_hdr.hdr_len, noblock, &err);
Arnd Bergmann02df55d2010-02-18 05:45:36 +0000695 if (!skb)
696 goto err;
Arnd Bergmann20d29d72010-01-30 12:24:26 +0000697
Shirley Ma97bc3632011-07-06 12:26:11 +0000698 if (zerocopy) {
699 err = zerocopy_sg_from_iovec(skb, iv, vnet_hdr_len, count);
700 skb_shinfo(skb)->tx_flags |= SKBTX_DEV_ZEROCOPY;
701 } else
702 err = skb_copy_datagram_from_iovec(skb, 0, iv, vnet_hdr_len,
703 len);
Arnd Bergmann02df55d2010-02-18 05:45:36 +0000704 if (err)
Arnd Bergmannb9fb9ee2010-02-18 05:48:17 +0000705 goto err_kfree;
Arnd Bergmann20d29d72010-01-30 12:24:26 +0000706
707 skb_set_network_header(skb, ETH_HLEN);
Arnd Bergmannb9fb9ee2010-02-18 05:48:17 +0000708 skb_reset_mac_header(skb);
709 skb->protocol = eth_hdr(skb)->h_proto;
710
711 if (vnet_hdr_len) {
712 err = macvtap_skb_from_vnet_hdr(skb, &vnet_hdr);
713 if (err)
714 goto err_kfree;
715 }
716
Arnd Bergmann02df55d2010-02-18 05:45:36 +0000717 rcu_read_lock_bh();
Eric Dumazet13707f92011-01-26 19:28:23 +0000718 vlan = rcu_dereference_bh(q->vlan);
Shirley Ma97bc3632011-07-06 12:26:11 +0000719 /* copy skb_ubuf_info for callback when skb has no error */
720 if (zerocopy)
721 skb_shinfo(skb)->destructor_arg = m->msg_control;
Arnd Bergmann02df55d2010-02-18 05:45:36 +0000722 if (vlan)
723 macvlan_start_xmit(skb, vlan->dev);
724 else
725 kfree_skb(skb);
726 rcu_read_unlock_bh();
Arnd Bergmann20d29d72010-01-30 12:24:26 +0000727
Shirley Ma97bc3632011-07-06 12:26:11 +0000728 return total_len;
Arnd Bergmann02df55d2010-02-18 05:45:36 +0000729
Arnd Bergmannb9fb9ee2010-02-18 05:48:17 +0000730err_kfree:
731 kfree_skb(skb);
732
Arnd Bergmann02df55d2010-02-18 05:45:36 +0000733err:
734 rcu_read_lock_bh();
Eric Dumazet13707f92011-01-26 19:28:23 +0000735 vlan = rcu_dereference_bh(q->vlan);
Arnd Bergmann02df55d2010-02-18 05:45:36 +0000736 if (vlan)
Eric Dumazet1ac9ad12011-01-12 12:13:14 +0000737 vlan->dev->stats.tx_dropped++;
Arnd Bergmann02df55d2010-02-18 05:45:36 +0000738 rcu_read_unlock_bh();
739
Arnd Bergmann02df55d2010-02-18 05:45:36 +0000740 return err;
Arnd Bergmann20d29d72010-01-30 12:24:26 +0000741}
742
743static ssize_t macvtap_aio_write(struct kiocb *iocb, const struct iovec *iv,
744 unsigned long count, loff_t pos)
745{
746 struct file *file = iocb->ki_filp;
747 ssize_t result = -ENOLINK;
Arnd Bergmann02df55d2010-02-18 05:45:36 +0000748 struct macvtap_queue *q = file->private_data;
Arnd Bergmann20d29d72010-01-30 12:24:26 +0000749
Shirley Ma97bc3632011-07-06 12:26:11 +0000750 result = macvtap_get_user(q, NULL, iv, iov_length(iv, count), count,
751 file->f_flags & O_NONBLOCK);
Arnd Bergmann20d29d72010-01-30 12:24:26 +0000752 return result;
753}
754
755/* Put packet to the user space buffer */
756static ssize_t macvtap_put_user(struct macvtap_queue *q,
757 const struct sk_buff *skb,
758 const struct iovec *iv, int len)
759{
Arnd Bergmann02df55d2010-02-18 05:45:36 +0000760 struct macvlan_dev *vlan;
Arnd Bergmann20d29d72010-01-30 12:24:26 +0000761 int ret;
Arnd Bergmannb9fb9ee2010-02-18 05:48:17 +0000762 int vnet_hdr_len = 0;
Basil Gorf09e2242012-05-03 22:55:24 +0000763 int vlan_offset = 0;
764 int copied;
Arnd Bergmannb9fb9ee2010-02-18 05:48:17 +0000765
766 if (q->flags & IFF_VNET_HDR) {
767 struct virtio_net_hdr vnet_hdr;
Michael S. Tsirkin55afbd02010-04-29 13:50:48 +0300768 vnet_hdr_len = q->vnet_hdr_sz;
Arnd Bergmannb9fb9ee2010-02-18 05:48:17 +0000769 if ((len -= vnet_hdr_len) < 0)
770 return -EINVAL;
771
772 ret = macvtap_skb_to_vnet_hdr(skb, &vnet_hdr);
773 if (ret)
774 return ret;
775
Michael S. Tsirkin55afbd02010-04-29 13:50:48 +0300776 if (memcpy_toiovecend(iv, (void *)&vnet_hdr, 0, sizeof(vnet_hdr)))
Arnd Bergmannb9fb9ee2010-02-18 05:48:17 +0000777 return -EFAULT;
778 }
Basil Gorf09e2242012-05-03 22:55:24 +0000779 copied = vnet_hdr_len;
Arnd Bergmann20d29d72010-01-30 12:24:26 +0000780
Basil Gorf09e2242012-05-03 22:55:24 +0000781 if (!vlan_tx_tag_present(skb))
782 len = min_t(int, skb->len, len);
783 else {
784 int copy;
785 struct {
786 __be16 h_vlan_proto;
787 __be16 h_vlan_TCI;
788 } veth;
789 veth.h_vlan_proto = htons(ETH_P_8021Q);
790 veth.h_vlan_TCI = htons(vlan_tx_tag_get(skb));
Arnd Bergmann20d29d72010-01-30 12:24:26 +0000791
Basil Gorf09e2242012-05-03 22:55:24 +0000792 vlan_offset = offsetof(struct vlan_ethhdr, h_vlan_proto);
793 len = min_t(int, skb->len + VLAN_HLEN, len);
Arnd Bergmann20d29d72010-01-30 12:24:26 +0000794
Basil Gorf09e2242012-05-03 22:55:24 +0000795 copy = min_t(int, vlan_offset, len);
796 ret = skb_copy_datagram_const_iovec(skb, 0, iv, copied, copy);
797 len -= copy;
798 copied += copy;
799 if (ret || !len)
800 goto done;
801
802 copy = min_t(int, sizeof(veth), len);
803 ret = memcpy_toiovecend(iv, (void *)&veth, copied, copy);
804 len -= copy;
805 copied += copy;
806 if (ret || !len)
807 goto done;
808 }
809
810 ret = skb_copy_datagram_const_iovec(skb, vlan_offset, iv, copied, len);
811 copied += len;
812
813done:
Arnd Bergmann02df55d2010-02-18 05:45:36 +0000814 rcu_read_lock_bh();
Eric Dumazet13707f92011-01-26 19:28:23 +0000815 vlan = rcu_dereference_bh(q->vlan);
Arnd Bergmann501c7742010-02-18 05:46:50 +0000816 if (vlan)
Basil Gorf09e2242012-05-03 22:55:24 +0000817 macvlan_count_rx(vlan, copied - vnet_hdr_len, ret == 0, 0);
Arnd Bergmann02df55d2010-02-18 05:45:36 +0000818 rcu_read_unlock_bh();
Arnd Bergmann20d29d72010-01-30 12:24:26 +0000819
Basil Gorf09e2242012-05-03 22:55:24 +0000820 return ret ? ret : copied;
Arnd Bergmann20d29d72010-01-30 12:24:26 +0000821}
822
Arnd Bergmann501c7742010-02-18 05:46:50 +0000823static ssize_t macvtap_do_read(struct macvtap_queue *q, struct kiocb *iocb,
824 const struct iovec *iv, unsigned long len,
825 int noblock)
Arnd Bergmann20d29d72010-01-30 12:24:26 +0000826{
Arnd Bergmann20d29d72010-01-30 12:24:26 +0000827 DECLARE_WAITQUEUE(wait, current);
828 struct sk_buff *skb;
Arnd Bergmann501c7742010-02-18 05:46:50 +0000829 ssize_t ret = 0;
Arnd Bergmann20d29d72010-01-30 12:24:26 +0000830
Eric Dumazet4a4771a2010-04-25 22:20:06 +0000831 add_wait_queue(sk_sleep(&q->sk), &wait);
Arnd Bergmann20d29d72010-01-30 12:24:26 +0000832 while (len) {
833 current->state = TASK_INTERRUPTIBLE;
834
835 /* Read frames from the queue */
836 skb = skb_dequeue(&q->sk.sk_receive_queue);
837 if (!skb) {
Arnd Bergmann501c7742010-02-18 05:46:50 +0000838 if (noblock) {
Arnd Bergmann20d29d72010-01-30 12:24:26 +0000839 ret = -EAGAIN;
840 break;
841 }
842 if (signal_pending(current)) {
843 ret = -ERESTARTSYS;
844 break;
845 }
846 /* Nothing to read, let's sleep */
847 schedule();
848 continue;
849 }
850 ret = macvtap_put_user(q, skb, iv, len);
851 kfree_skb(skb);
852 break;
853 }
854
855 current->state = TASK_RUNNING;
Eric Dumazet4a4771a2010-04-25 22:20:06 +0000856 remove_wait_queue(sk_sleep(&q->sk), &wait);
Arnd Bergmann501c7742010-02-18 05:46:50 +0000857 return ret;
858}
Arnd Bergmann20d29d72010-01-30 12:24:26 +0000859
Arnd Bergmann501c7742010-02-18 05:46:50 +0000860static ssize_t macvtap_aio_read(struct kiocb *iocb, const struct iovec *iv,
861 unsigned long count, loff_t pos)
862{
863 struct file *file = iocb->ki_filp;
864 struct macvtap_queue *q = file->private_data;
865 ssize_t len, ret = 0;
866
867 len = iov_length(iv, count);
868 if (len < 0) {
869 ret = -EINVAL;
870 goto out;
871 }
872
873 ret = macvtap_do_read(q, iocb, iv, len, file->f_flags & O_NONBLOCK);
874 ret = min_t(ssize_t, ret, len); /* XXX copied from tun.c. Why? */
Arnd Bergmann20d29d72010-01-30 12:24:26 +0000875out:
Arnd Bergmann20d29d72010-01-30 12:24:26 +0000876 return ret;
877}
878
879/*
880 * provide compatibility with generic tun/tap interface
881 */
882static long macvtap_ioctl(struct file *file, unsigned int cmd,
883 unsigned long arg)
884{
Arnd Bergmann02df55d2010-02-18 05:45:36 +0000885 struct macvtap_queue *q = file->private_data;
886 struct macvlan_dev *vlan;
Arnd Bergmann20d29d72010-01-30 12:24:26 +0000887 void __user *argp = (void __user *)arg;
888 struct ifreq __user *ifr = argp;
889 unsigned int __user *up = argp;
890 unsigned int u;
Michael S. Tsirkin55afbd02010-04-29 13:50:48 +0300891 int __user *sp = argp;
892 int s;
Arnd Bergmann02df55d2010-02-18 05:45:36 +0000893 int ret;
Arnd Bergmann20d29d72010-01-30 12:24:26 +0000894
895 switch (cmd) {
896 case TUNSETIFF:
897 /* ignore the name, just look at flags */
898 if (get_user(u, &ifr->ifr_flags))
899 return -EFAULT;
Arnd Bergmannb9fb9ee2010-02-18 05:48:17 +0000900
901 ret = 0;
902 if ((u & ~IFF_VNET_HDR) != (IFF_NO_PI | IFF_TAP))
903 ret = -EINVAL;
904 else
905 q->flags = u;
906
907 return ret;
Arnd Bergmann20d29d72010-01-30 12:24:26 +0000908
909 case TUNGETIFF:
Arnd Bergmann02df55d2010-02-18 05:45:36 +0000910 rcu_read_lock_bh();
Eric Dumazet13707f92011-01-26 19:28:23 +0000911 vlan = rcu_dereference_bh(q->vlan);
Arnd Bergmann02df55d2010-02-18 05:45:36 +0000912 if (vlan)
913 dev_hold(vlan->dev);
914 rcu_read_unlock_bh();
Arnd Bergmann20d29d72010-01-30 12:24:26 +0000915
Arnd Bergmann02df55d2010-02-18 05:45:36 +0000916 if (!vlan)
917 return -ENOLINK;
918
919 ret = 0;
Eric Dumazet13707f92011-01-26 19:28:23 +0000920 if (copy_to_user(&ifr->ifr_name, vlan->dev->name, IFNAMSIZ) ||
Arnd Bergmannb9fb9ee2010-02-18 05:48:17 +0000921 put_user(q->flags, &ifr->ifr_flags))
Arnd Bergmann02df55d2010-02-18 05:45:36 +0000922 ret = -EFAULT;
923 dev_put(vlan->dev);
924 return ret;
Arnd Bergmann20d29d72010-01-30 12:24:26 +0000925
926 case TUNGETFEATURES:
Arnd Bergmannb9fb9ee2010-02-18 05:48:17 +0000927 if (put_user(IFF_TAP | IFF_NO_PI | IFF_VNET_HDR, up))
Arnd Bergmann20d29d72010-01-30 12:24:26 +0000928 return -EFAULT;
929 return 0;
930
931 case TUNSETSNDBUF:
932 if (get_user(u, up))
933 return -EFAULT;
934
Arnd Bergmann20d29d72010-01-30 12:24:26 +0000935 q->sk.sk_sndbuf = u;
Arnd Bergmann20d29d72010-01-30 12:24:26 +0000936 return 0;
937
Michael S. Tsirkin55afbd02010-04-29 13:50:48 +0300938 case TUNGETVNETHDRSZ:
939 s = q->vnet_hdr_sz;
940 if (put_user(s, sp))
941 return -EFAULT;
942 return 0;
943
944 case TUNSETVNETHDRSZ:
945 if (get_user(s, sp))
946 return -EFAULT;
947 if (s < (int)sizeof(struct virtio_net_hdr))
948 return -EINVAL;
949
950 q->vnet_hdr_sz = s;
951 return 0;
952
Arnd Bergmann20d29d72010-01-30 12:24:26 +0000953 case TUNSETOFFLOAD:
954 /* let the user check for future flags */
955 if (arg & ~(TUN_F_CSUM | TUN_F_TSO4 | TUN_F_TSO6 |
Arnd Bergmannb9fb9ee2010-02-18 05:48:17 +0000956 TUN_F_TSO_ECN | TUN_F_UFO))
Arnd Bergmann20d29d72010-01-30 12:24:26 +0000957 return -EINVAL;
958
Arnd Bergmannb9fb9ee2010-02-18 05:48:17 +0000959 /* TODO: only accept frames with the features that
960 got enabled for forwarded frames */
961 if (!(q->flags & IFF_VNET_HDR))
962 return -EINVAL;
Arnd Bergmann20d29d72010-01-30 12:24:26 +0000963 return 0;
964
965 default:
966 return -EINVAL;
967 }
968}
969
970#ifdef CONFIG_COMPAT
971static long macvtap_compat_ioctl(struct file *file, unsigned int cmd,
972 unsigned long arg)
973{
974 return macvtap_ioctl(file, cmd, (unsigned long)compat_ptr(arg));
975}
976#endif
977
978static const struct file_operations macvtap_fops = {
979 .owner = THIS_MODULE,
980 .open = macvtap_open,
981 .release = macvtap_release,
982 .aio_read = macvtap_aio_read,
983 .aio_write = macvtap_aio_write,
984 .poll = macvtap_poll,
985 .llseek = no_llseek,
986 .unlocked_ioctl = macvtap_ioctl,
987#ifdef CONFIG_COMPAT
988 .compat_ioctl = macvtap_compat_ioctl,
989#endif
990};
991
Arnd Bergmann501c7742010-02-18 05:46:50 +0000992static int macvtap_sendmsg(struct kiocb *iocb, struct socket *sock,
993 struct msghdr *m, size_t total_len)
994{
995 struct macvtap_queue *q = container_of(sock, struct macvtap_queue, sock);
Shirley Ma97bc3632011-07-06 12:26:11 +0000996 return macvtap_get_user(q, m, m->msg_iov, total_len, m->msg_iovlen,
Arnd Bergmann501c7742010-02-18 05:46:50 +0000997 m->msg_flags & MSG_DONTWAIT);
998}
999
1000static int macvtap_recvmsg(struct kiocb *iocb, struct socket *sock,
1001 struct msghdr *m, size_t total_len,
1002 int flags)
1003{
1004 struct macvtap_queue *q = container_of(sock, struct macvtap_queue, sock);
1005 int ret;
1006 if (flags & ~(MSG_DONTWAIT|MSG_TRUNC))
1007 return -EINVAL;
1008 ret = macvtap_do_read(q, iocb, m->msg_iov, total_len,
1009 flags & MSG_DONTWAIT);
1010 if (ret > total_len) {
1011 m->msg_flags |= MSG_TRUNC;
1012 ret = flags & MSG_TRUNC ? ret : total_len;
1013 }
1014 return ret;
1015}
1016
1017/* Ops structure to mimic raw sockets with tun */
1018static const struct proto_ops macvtap_socket_ops = {
1019 .sendmsg = macvtap_sendmsg,
1020 .recvmsg = macvtap_recvmsg,
1021};
1022
1023/* Get an underlying socket object from tun file. Returns error unless file is
1024 * attached to a device. The returned object works like a packet socket, it
1025 * can be used for sock_sendmsg/sock_recvmsg. The caller is responsible for
1026 * holding a reference to the file for as long as the socket is in use. */
1027struct socket *macvtap_get_socket(struct file *file)
1028{
1029 struct macvtap_queue *q;
1030 if (file->f_op != &macvtap_fops)
1031 return ERR_PTR(-EINVAL);
1032 q = file->private_data;
1033 if (!q)
1034 return ERR_PTR(-EBADFD);
1035 return &q->sock;
1036}
1037EXPORT_SYMBOL_GPL(macvtap_get_socket);
1038
Eric W. Biederman9bf19072011-10-20 04:28:46 +00001039static int macvtap_device_event(struct notifier_block *unused,
1040 unsigned long event, void *ptr)
1041{
1042 struct net_device *dev = ptr;
Eric W. Biedermane09eff72011-10-20 04:29:24 +00001043 struct macvlan_dev *vlan;
Eric W. Biederman9bf19072011-10-20 04:28:46 +00001044 struct device *classdev;
1045 dev_t devt;
Eric W. Biedermane09eff72011-10-20 04:29:24 +00001046 int err;
Eric W. Biederman9bf19072011-10-20 04:28:46 +00001047
1048 if (dev->rtnl_link_ops != &macvtap_link_ops)
1049 return NOTIFY_DONE;
1050
Eric W. Biedermane09eff72011-10-20 04:29:24 +00001051 vlan = netdev_priv(dev);
Eric W. Biederman9bf19072011-10-20 04:28:46 +00001052
1053 switch (event) {
1054 case NETDEV_REGISTER:
1055 /* Create the device node here after the network device has
1056 * been registered but before register_netdevice has
1057 * finished running.
1058 */
Eric W. Biedermane09eff72011-10-20 04:29:24 +00001059 err = macvtap_get_minor(vlan);
1060 if (err)
1061 return notifier_from_errno(err);
1062
1063 devt = MKDEV(MAJOR(macvtap_major), vlan->minor);
Eric W. Biederman9bf19072011-10-20 04:28:46 +00001064 classdev = device_create(macvtap_class, &dev->dev, devt,
1065 dev, "tap%d", dev->ifindex);
Eric W. Biedermane09eff72011-10-20 04:29:24 +00001066 if (IS_ERR(classdev)) {
1067 macvtap_free_minor(vlan);
Eric W. Biederman9bf19072011-10-20 04:28:46 +00001068 return notifier_from_errno(PTR_ERR(classdev));
Eric W. Biedermane09eff72011-10-20 04:29:24 +00001069 }
Eric W. Biederman9bf19072011-10-20 04:28:46 +00001070 break;
1071 case NETDEV_UNREGISTER:
Eric W. Biedermane09eff72011-10-20 04:29:24 +00001072 devt = MKDEV(MAJOR(macvtap_major), vlan->minor);
Eric W. Biederman9bf19072011-10-20 04:28:46 +00001073 device_destroy(macvtap_class, devt);
Eric W. Biedermane09eff72011-10-20 04:29:24 +00001074 macvtap_free_minor(vlan);
Eric W. Biederman9bf19072011-10-20 04:28:46 +00001075 break;
1076 }
1077
1078 return NOTIFY_DONE;
1079}
1080
1081static struct notifier_block macvtap_notifier_block __read_mostly = {
1082 .notifier_call = macvtap_device_event,
1083};
1084
Arnd Bergmann20d29d72010-01-30 12:24:26 +00001085static int macvtap_init(void)
1086{
1087 int err;
1088
1089 err = alloc_chrdev_region(&macvtap_major, 0,
1090 MACVTAP_NUM_DEVS, "macvtap");
1091 if (err)
1092 goto out1;
1093
1094 cdev_init(&macvtap_cdev, &macvtap_fops);
1095 err = cdev_add(&macvtap_cdev, macvtap_major, MACVTAP_NUM_DEVS);
1096 if (err)
1097 goto out2;
1098
1099 macvtap_class = class_create(THIS_MODULE, "macvtap");
1100 if (IS_ERR(macvtap_class)) {
1101 err = PTR_ERR(macvtap_class);
1102 goto out3;
1103 }
1104
Eric W. Biederman9bf19072011-10-20 04:28:46 +00001105 err = register_netdevice_notifier(&macvtap_notifier_block);
Arnd Bergmann20d29d72010-01-30 12:24:26 +00001106 if (err)
1107 goto out4;
1108
Eric W. Biederman9bf19072011-10-20 04:28:46 +00001109 err = macvlan_link_register(&macvtap_link_ops);
1110 if (err)
1111 goto out5;
1112
Arnd Bergmann20d29d72010-01-30 12:24:26 +00001113 return 0;
1114
Eric W. Biederman9bf19072011-10-20 04:28:46 +00001115out5:
1116 unregister_netdevice_notifier(&macvtap_notifier_block);
Arnd Bergmann20d29d72010-01-30 12:24:26 +00001117out4:
1118 class_unregister(macvtap_class);
1119out3:
1120 cdev_del(&macvtap_cdev);
1121out2:
1122 unregister_chrdev_region(macvtap_major, MACVTAP_NUM_DEVS);
1123out1:
1124 return err;
1125}
1126module_init(macvtap_init);
1127
1128static void macvtap_exit(void)
1129{
1130 rtnl_link_unregister(&macvtap_link_ops);
Eric W. Biederman9bf19072011-10-20 04:28:46 +00001131 unregister_netdevice_notifier(&macvtap_notifier_block);
Arnd Bergmann20d29d72010-01-30 12:24:26 +00001132 class_unregister(macvtap_class);
1133 cdev_del(&macvtap_cdev);
1134 unregister_chrdev_region(macvtap_major, MACVTAP_NUM_DEVS);
1135}
1136module_exit(macvtap_exit);
1137
1138MODULE_ALIAS_RTNL_LINK("macvtap");
1139MODULE_AUTHOR("Arnd Bergmann <arnd@arndb.de>");
1140MODULE_LICENSE("GPL");