blob: 8362aef0c15e534241babfddbcd6cd6912ed1a8c [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/wait.h>
15#include <linux/cdev.h>
Al Viro40401532012-02-13 03:58:52 +000016#include <linux/idr.h>
Arnd Bergmann20d29d72010-01-30 12:24:26 +000017#include <linux/fs.h>
Herbert Xu6c36d2e2014-11-07 21:22:25 +080018#include <linux/uio.h>
Arnd Bergmann20d29d72010-01-30 12:24:26 +000019
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 *
Arnd Bergmann20d29d72010-01-30 12:24:26 +000034 */
35struct macvtap_queue {
36 struct sock sk;
37 struct socket sock;
Eric Dumazet43815482010-04-29 11:01:49 +000038 struct socket_wq wq;
Michael S. Tsirkin55afbd02010-04-29 13:50:48 +030039 int vnet_hdr_sz;
Eric Dumazet13707f92011-01-26 19:28:23 +000040 struct macvlan_dev __rcu *vlan;
Arnd Bergmann20d29d72010-01-30 12:24:26 +000041 struct file *file;
Arnd Bergmannb9fb9ee2010-02-18 05:48:17 +000042 unsigned int flags;
Jason Wang376b1aa2013-06-05 23:54:38 +000043 u16 queue_index;
Jason Wang815f2362013-06-05 23:54:39 +000044 bool enabled;
45 struct list_head next;
Arnd Bergmann20d29d72010-01-30 12:24:26 +000046};
47
Michael S. Tsirkin01b07fb2014-12-16 15:05:10 +020048#define MACVTAP_FEATURES (IFF_VNET_HDR | IFF_MULTI_QUEUE)
49
50#define MACVTAP_VNET_LE 0x80000000
Michael S. Tsirkin6ae7feb2014-11-23 17:24:15 +020051
52static inline u16 macvtap16_to_cpu(struct macvtap_queue *q, __virtio16 val)
53{
Michael S. Tsirkin01b07fb2014-12-16 15:05:10 +020054 return __virtio16_to_cpu(q->flags & MACVTAP_VNET_LE, val);
Michael S. Tsirkin6ae7feb2014-11-23 17:24:15 +020055}
56
57static inline __virtio16 cpu_to_macvtap16(struct macvtap_queue *q, u16 val)
58{
Michael S. Tsirkin01b07fb2014-12-16 15:05:10 +020059 return __cpu_to_virtio16(q->flags & MACVTAP_VNET_LE, val);
Michael S. Tsirkin6ae7feb2014-11-23 17:24:15 +020060}
61
Arnd Bergmann20d29d72010-01-30 12:24:26 +000062static struct proto macvtap_proto = {
63 .name = "macvtap",
64 .owner = THIS_MODULE,
65 .obj_size = sizeof (struct macvtap_queue),
66};
67
68/*
Eric W. Biedermane09eff72011-10-20 04:29:24 +000069 * Variables for dealing with macvtaps device numbers.
Arnd Bergmann20d29d72010-01-30 12:24:26 +000070 */
David S. Miller1ebed712010-07-10 19:25:50 -070071static dev_t macvtap_major;
Eric W. Biedermane09eff72011-10-20 04:29:24 +000072#define MACVTAP_NUM_DEVS (1U << MINORBITS)
73static DEFINE_MUTEX(minor_lock);
74static DEFINE_IDR(minor_idr);
75
Shirley Ma97bc3632011-07-06 12:26:11 +000076#define GOODCOPY_LEN 128
Arnd Bergmann20d29d72010-01-30 12:24:26 +000077static struct class *macvtap_class;
78static struct cdev macvtap_cdev;
79
Arnd Bergmann501c7742010-02-18 05:46:50 +000080static const struct proto_ops macvtap_socket_ops;
81
Vlad Yasevich2be5c762013-06-25 16:04:21 -040082#define TUN_OFFLOADS (NETIF_F_HW_CSUM | NETIF_F_TSO_ECN | NETIF_F_TSO | \
Vlad Yaseviche3e3c422015-02-03 16:36:17 -050083 NETIF_F_TSO6 | NETIF_F_UFO)
Vlad Yasevich2be5c762013-06-25 16:04:21 -040084#define RX_OFFLOADS (NETIF_F_GRO | NETIF_F_LRO)
Vlad Yasevicha567dd62013-08-16 15:25:00 -040085#define TAP_FEATURES (NETIF_F_GSO | NETIF_F_SG)
86
Vlad Yasevich6acf54f2013-12-11 13:27:10 -050087static struct macvlan_dev *macvtap_get_vlan_rcu(const struct net_device *dev)
88{
89 return rcu_dereference(dev->rx_handler_data);
90}
91
Arnd Bergmann20d29d72010-01-30 12:24:26 +000092/*
93 * RCU usage:
Arnd Bergmann02df55d2010-02-18 05:45:36 +000094 * The macvtap_queue and the macvlan_dev are loosely coupled, the
95 * pointers from one to the other can only be read while rcu_read_lock
Vlad Yasevich441ac0f2013-06-25 16:04:19 -040096 * or rtnl is held.
Arnd Bergmann20d29d72010-01-30 12:24:26 +000097 *
Arnd Bergmann02df55d2010-02-18 05:45:36 +000098 * Both the file and the macvlan_dev hold a reference on the macvtap_queue
99 * through sock_hold(&q->sk). When the macvlan_dev goes away first,
100 * q->vlan becomes inaccessible. When the files gets closed,
101 * macvtap_get_queue() fails.
Arnd Bergmann20d29d72010-01-30 12:24:26 +0000102 *
Arnd Bergmann02df55d2010-02-18 05:45:36 +0000103 * There may still be references to the struct sock inside of the
104 * queue from outbound SKBs, but these never reference back to the
105 * file or the dev. The data structure is freed through __sk_free
106 * when both our references and any pending SKBs are gone.
Arnd Bergmann20d29d72010-01-30 12:24:26 +0000107 */
Arnd Bergmann20d29d72010-01-30 12:24:26 +0000108
Jason Wang815f2362013-06-05 23:54:39 +0000109static int macvtap_enable_queue(struct net_device *dev, struct file *file,
Arnd Bergmann20d29d72010-01-30 12:24:26 +0000110 struct macvtap_queue *q)
111{
112 struct macvlan_dev *vlan = netdev_priv(dev);
Jason Wang815f2362013-06-05 23:54:39 +0000113 int err = -EINVAL;
114
Vlad Yasevich441ac0f2013-06-25 16:04:19 -0400115 ASSERT_RTNL();
Jason Wang815f2362013-06-05 23:54:39 +0000116
117 if (q->enabled)
118 goto out;
119
120 err = 0;
121 rcu_assign_pointer(vlan->taps[vlan->numvtaps], q);
122 q->queue_index = vlan->numvtaps;
123 q->enabled = true;
124
125 vlan->numvtaps++;
126out:
Jason Wang815f2362013-06-05 23:54:39 +0000127 return err;
128}
129
Vlad Yasevich40b8fe42014-09-22 16:34:17 -0400130/* Requires RTNL */
Jason Wang815f2362013-06-05 23:54:39 +0000131static int macvtap_set_queue(struct net_device *dev, struct file *file,
132 struct macvtap_queue *q)
133{
134 struct macvlan_dev *vlan = netdev_priv(dev);
Arnd Bergmann20d29d72010-01-30 12:24:26 +0000135
Jason Wang815f2362013-06-05 23:54:39 +0000136 if (vlan->numqueues == MAX_MACVTAP_QUEUES)
Vlad Yasevich40b8fe42014-09-22 16:34:17 -0400137 return -EBUSY;
Arnd Bergmann20d29d72010-01-30 12:24:26 +0000138
Arnd Bergmann02df55d2010-02-18 05:45:36 +0000139 rcu_assign_pointer(q->vlan, vlan);
Jason Wang376b1aa2013-06-05 23:54:38 +0000140 rcu_assign_pointer(vlan->taps[vlan->numvtaps], q);
Arnd Bergmann02df55d2010-02-18 05:45:36 +0000141 sock_hold(&q->sk);
Arnd Bergmann20d29d72010-01-30 12:24:26 +0000142
143 q->file = file;
Jason Wang376b1aa2013-06-05 23:54:38 +0000144 q->queue_index = vlan->numvtaps;
Jason Wang815f2362013-06-05 23:54:39 +0000145 q->enabled = true;
Arnd Bergmann02df55d2010-02-18 05:45:36 +0000146 file->private_data = q;
Jason Wang815f2362013-06-05 23:54:39 +0000147 list_add_tail(&q->next, &vlan->queue_list);
Arnd Bergmann20d29d72010-01-30 12:24:26 +0000148
Krishna Kumar1565c7c2010-08-04 06:15:59 +0000149 vlan->numvtaps++;
Jason Wang815f2362013-06-05 23:54:39 +0000150 vlan->numqueues++;
Krishna Kumar1565c7c2010-08-04 06:15:59 +0000151
Vlad Yasevich40b8fe42014-09-22 16:34:17 -0400152 return 0;
Arnd Bergmann20d29d72010-01-30 12:24:26 +0000153}
154
Vlad Yasevich441ac0f2013-06-25 16:04:19 -0400155static int macvtap_disable_queue(struct macvtap_queue *q)
Jason Wang815f2362013-06-05 23:54:39 +0000156{
157 struct macvlan_dev *vlan;
158 struct macvtap_queue *nq;
159
Vlad Yasevich441ac0f2013-06-25 16:04:19 -0400160 ASSERT_RTNL();
Jason Wang815f2362013-06-05 23:54:39 +0000161 if (!q->enabled)
162 return -EINVAL;
163
Vlad Yasevich441ac0f2013-06-25 16:04:19 -0400164 vlan = rtnl_dereference(q->vlan);
165
Jason Wang815f2362013-06-05 23:54:39 +0000166 if (vlan) {
167 int index = q->queue_index;
168 BUG_ON(index >= vlan->numvtaps);
Vlad Yasevich441ac0f2013-06-25 16:04:19 -0400169 nq = rtnl_dereference(vlan->taps[vlan->numvtaps - 1]);
Jason Wang815f2362013-06-05 23:54:39 +0000170 nq->queue_index = index;
171
172 rcu_assign_pointer(vlan->taps[index], nq);
173 RCU_INIT_POINTER(vlan->taps[vlan->numvtaps - 1], NULL);
174 q->enabled = false;
175
176 vlan->numvtaps--;
177 }
178
179 return 0;
180}
181
Arnd Bergmann20d29d72010-01-30 12:24:26 +0000182/*
Arnd Bergmann02df55d2010-02-18 05:45:36 +0000183 * The file owning the queue got closed, give up both
184 * the reference that the files holds as well as the
185 * one from the macvlan_dev if that still exists.
Arnd Bergmann20d29d72010-01-30 12:24:26 +0000186 *
187 * Using the spinlock makes sure that we don't get
188 * to the queue again after destroying it.
Arnd Bergmann20d29d72010-01-30 12:24:26 +0000189 */
Arnd Bergmann02df55d2010-02-18 05:45:36 +0000190static void macvtap_put_queue(struct macvtap_queue *q)
Arnd Bergmann20d29d72010-01-30 12:24:26 +0000191{
Arnd Bergmann02df55d2010-02-18 05:45:36 +0000192 struct macvlan_dev *vlan;
Arnd Bergmann20d29d72010-01-30 12:24:26 +0000193
Vlad Yasevich441ac0f2013-06-25 16:04:19 -0400194 rtnl_lock();
195 vlan = rtnl_dereference(q->vlan);
196
Arnd Bergmann02df55d2010-02-18 05:45:36 +0000197 if (vlan) {
Jason Wang815f2362013-06-05 23:54:39 +0000198 if (q->enabled)
Vlad Yasevich441ac0f2013-06-25 16:04:19 -0400199 BUG_ON(macvtap_disable_queue(q));
Krishna Kumar1565c7c2010-08-04 06:15:59 +0000200
Jason Wang815f2362013-06-05 23:54:39 +0000201 vlan->numqueues--;
Eric Dumazet2cfa5a02011-11-23 07:09:32 +0000202 RCU_INIT_POINTER(q->vlan, NULL);
Arnd Bergmann02df55d2010-02-18 05:45:36 +0000203 sock_put(&q->sk);
Jason Wang815f2362013-06-05 23:54:39 +0000204 list_del_init(&q->next);
Arnd Bergmann20d29d72010-01-30 12:24:26 +0000205 }
206
Vlad Yasevich441ac0f2013-06-25 16:04:19 -0400207 rtnl_unlock();
Arnd Bergmann20d29d72010-01-30 12:24:26 +0000208
209 synchronize_rcu();
210 sock_put(&q->sk);
211}
212
213/*
Krishna Kumar1565c7c2010-08-04 06:15:59 +0000214 * Select a queue based on the rxq of the device on which this packet
215 * arrived. If the incoming device is not mq, calculate a flow hash
216 * to select a queue. If all fails, find the first available queue.
217 * Cache vlan->numvtaps since it can become zero during the execution
218 * of this function.
Arnd Bergmann20d29d72010-01-30 12:24:26 +0000219 */
220static struct macvtap_queue *macvtap_get_queue(struct net_device *dev,
221 struct sk_buff *skb)
222{
223 struct macvlan_dev *vlan = netdev_priv(dev);
Krishna Kumar1565c7c2010-08-04 06:15:59 +0000224 struct macvtap_queue *tap = NULL;
Jason Wang815f2362013-06-05 23:54:39 +0000225 /* Access to taps array is protected by rcu, but access to numvtaps
226 * isn't. Below we use it to lookup a queue, but treat it as a hint
227 * and validate that the result isn't NULL - in case we are
228 * racing against queue removal.
229 */
Jason Wanged0483f2013-06-05 23:54:33 +0000230 int numvtaps = ACCESS_ONCE(vlan->numvtaps);
Krishna Kumar1565c7c2010-08-04 06:15:59 +0000231 __u32 rxq;
Arnd Bergmann20d29d72010-01-30 12:24:26 +0000232
Krishna Kumar1565c7c2010-08-04 06:15:59 +0000233 if (!numvtaps)
234 goto out;
235
Krishna Kumaref0002b2011-11-23 22:17:14 +0000236 /* Check if we can use flow to select a queue */
Tom Herbert3958afa1b2013-12-15 22:12:06 -0800237 rxq = skb_get_hash(skb);
Krishna Kumaref0002b2011-11-23 22:17:14 +0000238 if (rxq) {
239 tap = rcu_dereference(vlan->taps[rxq % numvtaps]);
Jason Wang376b1aa2013-06-05 23:54:38 +0000240 goto out;
Krishna Kumaref0002b2011-11-23 22:17:14 +0000241 }
242
Krishna Kumar1565c7c2010-08-04 06:15:59 +0000243 if (likely(skb_rx_queue_recorded(skb))) {
244 rxq = skb_get_rx_queue(skb);
245
246 while (unlikely(rxq >= numvtaps))
247 rxq -= numvtaps;
248
249 tap = rcu_dereference(vlan->taps[rxq]);
Jason Wang376b1aa2013-06-05 23:54:38 +0000250 goto out;
Krishna Kumar1565c7c2010-08-04 06:15:59 +0000251 }
252
Jason Wang376b1aa2013-06-05 23:54:38 +0000253 tap = rcu_dereference(vlan->taps[0]);
Krishna Kumar1565c7c2010-08-04 06:15:59 +0000254out:
255 return tap;
Arnd Bergmann20d29d72010-01-30 12:24:26 +0000256}
257
Arnd Bergmann02df55d2010-02-18 05:45:36 +0000258/*
259 * The net_device is going away, give up the reference
Krishna Kumar1565c7c2010-08-04 06:15:59 +0000260 * that it holds on all queues and safely set the pointer
261 * from the queues to NULL.
Arnd Bergmann02df55d2010-02-18 05:45:36 +0000262 */
Arnd Bergmann20d29d72010-01-30 12:24:26 +0000263static void macvtap_del_queues(struct net_device *dev)
264{
265 struct macvlan_dev *vlan = netdev_priv(dev);
Jason Wang815f2362013-06-05 23:54:39 +0000266 struct macvtap_queue *q, *tmp, *qlist[MAX_MACVTAP_QUEUES];
Krishna Kumar1565c7c2010-08-04 06:15:59 +0000267 int i, j = 0;
Arnd Bergmann20d29d72010-01-30 12:24:26 +0000268
Vlad Yasevich441ac0f2013-06-25 16:04:19 -0400269 ASSERT_RTNL();
Jason Wang815f2362013-06-05 23:54:39 +0000270 list_for_each_entry_safe(q, tmp, &vlan->queue_list, next) {
271 list_del_init(&q->next);
Jason Wang376b1aa2013-06-05 23:54:38 +0000272 qlist[j++] = q;
Jason Wang376b1aa2013-06-05 23:54:38 +0000273 RCU_INIT_POINTER(q->vlan, NULL);
Jason Wang815f2362013-06-05 23:54:39 +0000274 if (q->enabled)
275 vlan->numvtaps--;
276 vlan->numqueues--;
Arnd Bergmann02df55d2010-02-18 05:45:36 +0000277 }
Jason Wang815f2362013-06-05 23:54:39 +0000278 for (i = 0; i < vlan->numvtaps; i++)
279 RCU_INIT_POINTER(vlan->taps[i], NULL);
280 BUG_ON(vlan->numvtaps);
281 BUG_ON(vlan->numqueues);
Eric W. Biederman99f34b32011-10-20 04:26:01 +0000282 /* guarantee that any future macvtap_set_queue will fail */
283 vlan->numvtaps = MAX_MACVTAP_QUEUES;
Krishna Kumar1565c7c2010-08-04 06:15:59 +0000284
285 for (--j; j >= 0; j--)
286 sock_put(&qlist[j]->sk);
Arnd Bergmann20d29d72010-01-30 12:24:26 +0000287}
288
Vlad Yasevich6acf54f2013-12-11 13:27:10 -0500289static rx_handler_result_t macvtap_handle_frame(struct sk_buff **pskb)
Arnd Bergmann20d29d72010-01-30 12:24:26 +0000290{
Vlad Yasevich6acf54f2013-12-11 13:27:10 -0500291 struct sk_buff *skb = *pskb;
292 struct net_device *dev = skb->dev;
293 struct macvlan_dev *vlan;
294 struct macvtap_queue *q;
Vlad Yasevicha567dd62013-08-16 15:25:00 -0400295 netdev_features_t features = TAP_FEATURES;
296
Vlad Yasevich6acf54f2013-12-11 13:27:10 -0500297 vlan = macvtap_get_vlan_rcu(dev);
298 if (!vlan)
299 return RX_HANDLER_PASS;
300
301 q = macvtap_get_queue(dev, skb);
Arnd Bergmann20d29d72010-01-30 12:24:26 +0000302 if (!q)
Vlad Yasevich6acf54f2013-12-11 13:27:10 -0500303 return RX_HANDLER_PASS;
Herbert Xu8a357472010-07-21 21:44:31 +0000304
305 if (skb_queue_len(&q->sk.sk_receive_queue) >= dev->tx_queue_len)
306 goto drop;
Arnd Bergmann20d29d72010-01-30 12:24:26 +0000307
Vlad Yasevich6acf54f2013-12-11 13:27:10 -0500308 skb_push(skb, ETH_HLEN);
309
Vlad Yasevich3e4f8b72013-06-25 16:04:22 -0400310 /* Apply the forward feature mask so that we perform segmentation
Vlad Yaseviche5733322013-08-16 15:25:02 -0400311 * according to users wishes. This only works if VNET_HDR is
312 * enabled.
Vlad Yasevich3e4f8b72013-06-25 16:04:22 -0400313 */
Vlad Yaseviche5733322013-08-16 15:25:02 -0400314 if (q->flags & IFF_VNET_HDR)
315 features |= vlan->tap_features;
Tom Herbert04ffcb22014-10-14 15:19:06 -0700316 if (netif_needs_gso(dev, skb, features)) {
Vlad Yasevich3e4f8b72013-06-25 16:04:22 -0400317 struct sk_buff *segs = __skb_gso_segment(skb, features, false);
318
319 if (IS_ERR(segs))
320 goto drop;
321
322 if (!segs) {
323 skb_queue_tail(&q->sk.sk_receive_queue, skb);
324 goto wake_up;
325 }
326
327 kfree_skb(skb);
328 while (segs) {
329 struct sk_buff *nskb = segs->next;
330
331 segs->next = NULL;
332 skb_queue_tail(&q->sk.sk_receive_queue, segs);
333 segs = nskb;
334 }
335 } else {
Vlad Yasevichcbdb0422014-04-29 10:09:50 -0400336 /* If we receive a partial checksum and the tap side
337 * doesn't support checksum offload, compute the checksum.
338 * Note: it doesn't matter which checksum feature to
339 * check, we either support them all or none.
340 */
341 if (skb->ip_summed == CHECKSUM_PARTIAL &&
342 !(features & NETIF_F_ALL_CSUM) &&
343 skb_checksum_help(skb))
344 goto drop;
Vlad Yasevich3e4f8b72013-06-25 16:04:22 -0400345 skb_queue_tail(&q->sk.sk_receive_queue, skb);
346 }
347
348wake_up:
Eric Dumazet4a4771a2010-04-25 22:20:06 +0000349 wake_up_interruptible_poll(sk_sleep(&q->sk), POLLIN | POLLRDNORM | POLLRDBAND);
Vlad Yasevich6acf54f2013-12-11 13:27:10 -0500350 return RX_HANDLER_CONSUMED;
Herbert Xu8a357472010-07-21 21:44:31 +0000351
352drop:
Vlad Yasevich6acf54f2013-12-11 13:27:10 -0500353 /* Count errors/drops only here, thus don't care about args. */
354 macvlan_count_rx(vlan, 0, 0, 0);
Herbert Xu8a357472010-07-21 21:44:31 +0000355 kfree_skb(skb);
Vlad Yasevich6acf54f2013-12-11 13:27:10 -0500356 return RX_HANDLER_CONSUMED;
Arnd Bergmann20d29d72010-01-30 12:24:26 +0000357}
358
Eric W. Biedermane09eff72011-10-20 04:29:24 +0000359static int macvtap_get_minor(struct macvlan_dev *vlan)
360{
361 int retval = -ENOMEM;
Eric W. Biedermane09eff72011-10-20 04:29:24 +0000362
363 mutex_lock(&minor_lock);
Tejun Heoec09ebc2013-02-27 17:04:34 -0800364 retval = idr_alloc(&minor_idr, vlan, 1, MACVTAP_NUM_DEVS, GFP_KERNEL);
365 if (retval >= 0) {
366 vlan->minor = retval;
367 } else if (retval == -ENOSPC) {
Eric W. Biedermane09eff72011-10-20 04:29:24 +0000368 printk(KERN_ERR "too many macvtap devices\n");
369 retval = -EINVAL;
Eric W. Biedermane09eff72011-10-20 04:29:24 +0000370 }
Eric W. Biedermane09eff72011-10-20 04:29:24 +0000371 mutex_unlock(&minor_lock);
Tejun Heoec09ebc2013-02-27 17:04:34 -0800372 return retval < 0 ? retval : 0;
Eric W. Biedermane09eff72011-10-20 04:29:24 +0000373}
374
375static void macvtap_free_minor(struct macvlan_dev *vlan)
376{
377 mutex_lock(&minor_lock);
378 if (vlan->minor) {
379 idr_remove(&minor_idr, vlan->minor);
380 vlan->minor = 0;
381 }
382 mutex_unlock(&minor_lock);
383}
384
385static struct net_device *dev_get_by_macvtap_minor(int minor)
386{
387 struct net_device *dev = NULL;
388 struct macvlan_dev *vlan;
389
390 mutex_lock(&minor_lock);
391 vlan = idr_find(&minor_idr, minor);
392 if (vlan) {
393 dev = vlan->dev;
394 dev_hold(dev);
395 }
396 mutex_unlock(&minor_lock);
397 return dev;
398}
399
Arnd Bergmann20d29d72010-01-30 12:24:26 +0000400static int macvtap_newlink(struct net *src_net,
401 struct net_device *dev,
402 struct nlattr *tb[],
403 struct nlattr *data[])
404{
Jason Wang815f2362013-06-05 23:54:39 +0000405 struct macvlan_dev *vlan = netdev_priv(dev);
Vlad Yasevich6acf54f2013-12-11 13:27:10 -0500406 int err;
407
Jason Wang815f2362013-06-05 23:54:39 +0000408 INIT_LIST_HEAD(&vlan->queue_list);
409
Vlad Yasevich2be5c762013-06-25 16:04:21 -0400410 /* Since macvlan supports all offloads by default, make
411 * tap support all offloads also.
412 */
413 vlan->tap_features = TUN_OFFLOADS;
414
Vlad Yasevich6acf54f2013-12-11 13:27:10 -0500415 err = netdev_rx_handler_register(dev, macvtap_handle_frame, vlan);
416 if (err)
417 return err;
418
Eric W. Biederman9bf19072011-10-20 04:28:46 +0000419 /* Don't put anything that may fail after macvlan_common_newlink
420 * because we can't undo what it does.
421 */
Vlad Yasevich2f6a1b62013-12-11 13:27:11 -0500422 return macvlan_common_newlink(src_net, dev, tb, data);
Arnd Bergmann20d29d72010-01-30 12:24:26 +0000423}
424
425static void macvtap_dellink(struct net_device *dev,
426 struct list_head *head)
427{
Vlad Yasevich6acf54f2013-12-11 13:27:10 -0500428 netdev_rx_handler_unregister(dev);
Arnd Bergmann20d29d72010-01-30 12:24:26 +0000429 macvtap_del_queues(dev);
430 macvlan_dellink(dev, head);
431}
432
Herbert Xu8a357472010-07-21 21:44:31 +0000433static void macvtap_setup(struct net_device *dev)
434{
435 macvlan_common_setup(dev);
436 dev->tx_queue_len = TUN_READQ_SIZE;
437}
438
Arnd Bergmann20d29d72010-01-30 12:24:26 +0000439static struct rtnl_link_ops macvtap_link_ops __read_mostly = {
440 .kind = "macvtap",
Herbert Xu8a357472010-07-21 21:44:31 +0000441 .setup = macvtap_setup,
Arnd Bergmann20d29d72010-01-30 12:24:26 +0000442 .newlink = macvtap_newlink,
443 .dellink = macvtap_dellink,
444};
445
446
447static void macvtap_sock_write_space(struct sock *sk)
448{
Eric Dumazet43815482010-04-29 11:01:49 +0000449 wait_queue_head_t *wqueue;
450
Arnd Bergmann20d29d72010-01-30 12:24:26 +0000451 if (!sock_writeable(sk) ||
452 !test_and_clear_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags))
453 return;
454
Eric Dumazet43815482010-04-29 11:01:49 +0000455 wqueue = sk_sleep(sk);
456 if (wqueue && waitqueue_active(wqueue))
457 wake_up_interruptible_poll(wqueue, POLLOUT | POLLWRNORM | POLLWRBAND);
Arnd Bergmann20d29d72010-01-30 12:24:26 +0000458}
459
Eric W. Biederman2259fef2011-10-20 04:27:24 +0000460static void macvtap_sock_destruct(struct sock *sk)
461{
462 skb_queue_purge(&sk->sk_receive_queue);
463}
464
Arnd Bergmann20d29d72010-01-30 12:24:26 +0000465static int macvtap_open(struct inode *inode, struct file *file)
466{
467 struct net *net = current->nsproxy->net_ns;
Vlad Yasevich40b8fe42014-09-22 16:34:17 -0400468 struct net_device *dev;
Arnd Bergmann20d29d72010-01-30 12:24:26 +0000469 struct macvtap_queue *q;
Vlad Yasevich40b8fe42014-09-22 16:34:17 -0400470 int err = -ENODEV;
Arnd Bergmann20d29d72010-01-30 12:24:26 +0000471
Vlad Yasevich40b8fe42014-09-22 16:34:17 -0400472 rtnl_lock();
473 dev = dev_get_by_macvtap_minor(iminor(inode));
Arnd Bergmann20d29d72010-01-30 12:24:26 +0000474 if (!dev)
475 goto out;
476
Arnd Bergmann20d29d72010-01-30 12:24:26 +0000477 err = -ENOMEM;
478 q = (struct macvtap_queue *)sk_alloc(net, AF_UNSPEC, GFP_KERNEL,
479 &macvtap_proto);
480 if (!q)
481 goto out;
482
Jason Wangd9a90a32013-06-13 14:23:35 +0800483 RCU_INIT_POINTER(q->sock.wq, &q->wq);
Eric Dumazet43815482010-04-29 11:01:49 +0000484 init_waitqueue_head(&q->wq.wait);
Arnd Bergmann20d29d72010-01-30 12:24:26 +0000485 q->sock.type = SOCK_RAW;
486 q->sock.state = SS_CONNECTED;
Arnd Bergmann501c7742010-02-18 05:46:50 +0000487 q->sock.file = file;
488 q->sock.ops = &macvtap_socket_ops;
Arnd Bergmann20d29d72010-01-30 12:24:26 +0000489 sock_init_data(&q->sock, &q->sk);
Arnd Bergmann20d29d72010-01-30 12:24:26 +0000490 q->sk.sk_write_space = macvtap_sock_write_space;
Eric W. Biederman2259fef2011-10-20 04:27:24 +0000491 q->sk.sk_destruct = macvtap_sock_destruct;
Arnd Bergmannb9fb9ee2010-02-18 05:48:17 +0000492 q->flags = IFF_VNET_HDR | IFF_NO_PI | IFF_TAP;
Michael S. Tsirkin55afbd02010-04-29 13:50:48 +0300493 q->vnet_hdr_sz = sizeof(struct virtio_net_hdr);
Arnd Bergmann20d29d72010-01-30 12:24:26 +0000494
Shirley Ma97bc3632011-07-06 12:26:11 +0000495 /*
496 * so far only KVM virtio_net uses macvtap, enable zero copy between
497 * guest kernel and host kernel when lower device supports zerocopy
Eric W. Biederman047af9cf2011-10-20 04:26:39 +0000498 *
499 * The macvlan supports zerocopy iff the lower device supports zero
500 * copy so we don't have to look at the lower device directly.
Shirley Ma97bc3632011-07-06 12:26:11 +0000501 */
Eric W. Biederman047af9cf2011-10-20 04:26:39 +0000502 if ((dev->features & NETIF_F_HIGHDMA) && (dev->features & NETIF_F_SG))
503 sock_set_flag(&q->sk, SOCK_ZEROCOPY);
Shirley Ma97bc3632011-07-06 12:26:11 +0000504
Arnd Bergmann20d29d72010-01-30 12:24:26 +0000505 err = macvtap_set_queue(dev, file, q);
506 if (err)
507 sock_put(&q->sk);
508
509out:
510 if (dev)
511 dev_put(dev);
512
Vlad Yasevich40b8fe42014-09-22 16:34:17 -0400513 rtnl_unlock();
Arnd Bergmann20d29d72010-01-30 12:24:26 +0000514 return err;
515}
516
517static int macvtap_release(struct inode *inode, struct file *file)
518{
Arnd Bergmann02df55d2010-02-18 05:45:36 +0000519 struct macvtap_queue *q = file->private_data;
520 macvtap_put_queue(q);
Arnd Bergmann20d29d72010-01-30 12:24:26 +0000521 return 0;
522}
523
524static unsigned int macvtap_poll(struct file *file, poll_table * wait)
525{
Arnd Bergmann02df55d2010-02-18 05:45:36 +0000526 struct macvtap_queue *q = file->private_data;
Arnd Bergmann20d29d72010-01-30 12:24:26 +0000527 unsigned int mask = POLLERR;
528
529 if (!q)
530 goto out;
531
532 mask = 0;
Eric Dumazet43815482010-04-29 11:01:49 +0000533 poll_wait(file, &q->wq.wait, wait);
Arnd Bergmann20d29d72010-01-30 12:24:26 +0000534
535 if (!skb_queue_empty(&q->sk.sk_receive_queue))
536 mask |= POLLIN | POLLRDNORM;
537
538 if (sock_writeable(&q->sk) ||
539 (!test_and_set_bit(SOCK_ASYNC_NOSPACE, &q->sock.flags) &&
540 sock_writeable(&q->sk)))
541 mask |= POLLOUT | POLLWRNORM;
542
543out:
Arnd Bergmann20d29d72010-01-30 12:24:26 +0000544 return mask;
545}
546
Arnd Bergmannb9fb9ee2010-02-18 05:48:17 +0000547static inline struct sk_buff *macvtap_alloc_skb(struct sock *sk, size_t prepad,
548 size_t len, size_t linear,
549 int noblock, int *err)
550{
551 struct sk_buff *skb;
552
553 /* Under a page? Don't bother with paged skb. */
554 if (prepad + len < PAGE_SIZE || !linear)
555 linear = len;
556
557 skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock,
Eric Dumazet28d64272013-08-08 14:38:47 -0700558 err, 0);
Arnd Bergmannb9fb9ee2010-02-18 05:48:17 +0000559 if (!skb)
560 return NULL;
561
562 skb_reserve(skb, prepad);
563 skb_put(skb, linear);
564 skb->data_len = len - linear;
565 skb->len += len - linear;
566
567 return skb;
568}
569
570/*
571 * macvtap_skb_from_vnet_hdr and macvtap_skb_to_vnet_hdr should
572 * be shared with the tun/tap driver.
573 */
Michael S. Tsirkin6ae7feb2014-11-23 17:24:15 +0200574static int macvtap_skb_from_vnet_hdr(struct macvtap_queue *q,
575 struct sk_buff *skb,
Arnd Bergmannb9fb9ee2010-02-18 05:48:17 +0000576 struct virtio_net_hdr *vnet_hdr)
577{
578 unsigned short gso_type = 0;
579 if (vnet_hdr->gso_type != VIRTIO_NET_HDR_GSO_NONE) {
580 switch (vnet_hdr->gso_type & ~VIRTIO_NET_HDR_GSO_ECN) {
581 case VIRTIO_NET_HDR_GSO_TCPV4:
582 gso_type = SKB_GSO_TCPV4;
583 break;
584 case VIRTIO_NET_HDR_GSO_TCPV6:
585 gso_type = SKB_GSO_TCPV6;
586 break;
587 case VIRTIO_NET_HDR_GSO_UDP:
588 gso_type = SKB_GSO_UDP;
589 break;
590 default:
591 return -EINVAL;
592 }
593
594 if (vnet_hdr->gso_type & VIRTIO_NET_HDR_GSO_ECN)
595 gso_type |= SKB_GSO_TCP_ECN;
596
597 if (vnet_hdr->gso_size == 0)
598 return -EINVAL;
599 }
600
601 if (vnet_hdr->flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) {
Michael S. Tsirkin6ae7feb2014-11-23 17:24:15 +0200602 if (!skb_partial_csum_set(skb, macvtap16_to_cpu(q, vnet_hdr->csum_start),
603 macvtap16_to_cpu(q, vnet_hdr->csum_offset)))
Arnd Bergmannb9fb9ee2010-02-18 05:48:17 +0000604 return -EINVAL;
605 }
606
607 if (vnet_hdr->gso_type != VIRTIO_NET_HDR_GSO_NONE) {
Michael S. Tsirkin6ae7feb2014-11-23 17:24:15 +0200608 skb_shinfo(skb)->gso_size = macvtap16_to_cpu(q, vnet_hdr->gso_size);
Pravin B Shelarc9af6db2013-02-11 09:27:41 +0000609 skb_shinfo(skb)->gso_type = gso_type;
Arnd Bergmannb9fb9ee2010-02-18 05:48:17 +0000610
611 /* Header must be checked, and gso_segs computed. */
612 skb_shinfo(skb)->gso_type |= SKB_GSO_DODGY;
613 skb_shinfo(skb)->gso_segs = 0;
614 }
615 return 0;
616}
617
Michael S. Tsirkin6ae7feb2014-11-23 17:24:15 +0200618static void macvtap_skb_to_vnet_hdr(struct macvtap_queue *q,
619 const struct sk_buff *skb,
620 struct virtio_net_hdr *vnet_hdr)
Arnd Bergmannb9fb9ee2010-02-18 05:48:17 +0000621{
622 memset(vnet_hdr, 0, sizeof(*vnet_hdr));
623
624 if (skb_is_gso(skb)) {
625 struct skb_shared_info *sinfo = skb_shinfo(skb);
626
627 /* This is a hint as to how much should be linear. */
Michael S. Tsirkin6ae7feb2014-11-23 17:24:15 +0200628 vnet_hdr->hdr_len = cpu_to_macvtap16(q, skb_headlen(skb));
629 vnet_hdr->gso_size = cpu_to_macvtap16(q, sinfo->gso_size);
Arnd Bergmannb9fb9ee2010-02-18 05:48:17 +0000630 if (sinfo->gso_type & SKB_GSO_TCPV4)
631 vnet_hdr->gso_type = VIRTIO_NET_HDR_GSO_TCPV4;
632 else if (sinfo->gso_type & SKB_GSO_TCPV6)
633 vnet_hdr->gso_type = VIRTIO_NET_HDR_GSO_TCPV6;
Vlad Yaseviche3e3c422015-02-03 16:36:17 -0500634 else if (sinfo->gso_type & SKB_GSO_UDP)
635 vnet_hdr->gso_type = VIRTIO_NET_HDR_GSO_UDP;
Arnd Bergmannb9fb9ee2010-02-18 05:48:17 +0000636 else
637 BUG();
638 if (sinfo->gso_type & SKB_GSO_TCP_ECN)
639 vnet_hdr->gso_type |= VIRTIO_NET_HDR_GSO_ECN;
640 } else
641 vnet_hdr->gso_type = VIRTIO_NET_HDR_GSO_NONE;
642
643 if (skb->ip_summed == CHECKSUM_PARTIAL) {
644 vnet_hdr->flags = VIRTIO_NET_HDR_F_NEEDS_CSUM;
Jiri Pirkodf8a39d2015-01-13 17:13:44 +0100645 if (skb_vlan_tag_present(skb))
Michael S. Tsirkin6ae7feb2014-11-23 17:24:15 +0200646 vnet_hdr->csum_start = cpu_to_macvtap16(q,
647 skb_checksum_start_offset(skb) + VLAN_HLEN);
648 else
649 vnet_hdr->csum_start = cpu_to_macvtap16(q,
650 skb_checksum_start_offset(skb));
651 vnet_hdr->csum_offset = cpu_to_macvtap16(q, skb->csum_offset);
Jason Wang10a8d942011-06-10 00:56:17 +0000652 } else if (skb->ip_summed == CHECKSUM_UNNECESSARY) {
653 vnet_hdr->flags = VIRTIO_NET_HDR_F_DATA_VALID;
Arnd Bergmannb9fb9ee2010-02-18 05:48:17 +0000654 } /* else everything is zero */
Arnd Bergmannb9fb9ee2010-02-18 05:48:17 +0000655}
656
Eric Dumazet2f1d8b92015-02-27 18:35:35 -0800657/* Neighbour code has some assumptions on HH_DATA_MOD alignment */
658#define MACVTAP_RESERVE HH_DATA_OFF(ETH_HLEN)
659
Arnd Bergmann20d29d72010-01-30 12:24:26 +0000660/* Get packet from user space buffer */
Shirley Ma97bc3632011-07-06 12:26:11 +0000661static ssize_t macvtap_get_user(struct macvtap_queue *q, struct msghdr *m,
Al Virof5ff53b2014-06-19 15:36:49 -0400662 struct iov_iter *from, int noblock)
Arnd Bergmann20d29d72010-01-30 12:24:26 +0000663{
Eric Dumazet2f1d8b92015-02-27 18:35:35 -0800664 int good_linear = SKB_MAX_HEAD(MACVTAP_RESERVE);
Arnd Bergmann20d29d72010-01-30 12:24:26 +0000665 struct sk_buff *skb;
Arnd Bergmann02df55d2010-02-18 05:45:36 +0000666 struct macvlan_dev *vlan;
Al Virof5ff53b2014-06-19 15:36:49 -0400667 unsigned long total_len = iov_iter_count(from);
Shirley Ma97bc3632011-07-06 12:26:11 +0000668 unsigned long len = total_len;
Arnd Bergmann20d29d72010-01-30 12:24:26 +0000669 int err;
Arnd Bergmannb9fb9ee2010-02-18 05:48:17 +0000670 struct virtio_net_hdr vnet_hdr = { 0 };
671 int vnet_hdr_len = 0;
Jason Wangb92946e2012-05-02 11:42:15 +0800672 int copylen = 0;
Shirley Ma97bc3632011-07-06 12:26:11 +0000673 bool zerocopy = false;
Jason Wang61d46bf2013-07-10 13:43:28 +0800674 size_t linear;
Al Virof5ff53b2014-06-19 15:36:49 -0400675 ssize_t n;
Arnd Bergmann20d29d72010-01-30 12:24:26 +0000676
Arnd Bergmannb9fb9ee2010-02-18 05:48:17 +0000677 if (q->flags & IFF_VNET_HDR) {
Michael S. Tsirkin55afbd02010-04-29 13:50:48 +0300678 vnet_hdr_len = q->vnet_hdr_sz;
Arnd Bergmannb9fb9ee2010-02-18 05:48:17 +0000679
680 err = -EINVAL;
Nicolas Kaiserce3c8692011-03-04 13:49:41 +0000681 if (len < vnet_hdr_len)
Arnd Bergmannb9fb9ee2010-02-18 05:48:17 +0000682 goto err;
Nicolas Kaiserce3c8692011-03-04 13:49:41 +0000683 len -= vnet_hdr_len;
Arnd Bergmannb9fb9ee2010-02-18 05:48:17 +0000684
Al Virof5ff53b2014-06-19 15:36:49 -0400685 err = -EFAULT;
686 n = copy_from_iter(&vnet_hdr, sizeof(vnet_hdr), from);
687 if (n != sizeof(vnet_hdr))
Arnd Bergmannb9fb9ee2010-02-18 05:48:17 +0000688 goto err;
Al Virof5ff53b2014-06-19 15:36:49 -0400689 iov_iter_advance(from, vnet_hdr_len - sizeof(vnet_hdr));
Arnd Bergmannb9fb9ee2010-02-18 05:48:17 +0000690 if ((vnet_hdr.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) &&
Michael S. Tsirkin6ae7feb2014-11-23 17:24:15 +0200691 macvtap16_to_cpu(q, vnet_hdr.csum_start) +
692 macvtap16_to_cpu(q, vnet_hdr.csum_offset) + 2 >
693 macvtap16_to_cpu(q, vnet_hdr.hdr_len))
694 vnet_hdr.hdr_len = cpu_to_macvtap16(q,
695 macvtap16_to_cpu(q, vnet_hdr.csum_start) +
696 macvtap16_to_cpu(q, vnet_hdr.csum_offset) + 2);
Arnd Bergmannb9fb9ee2010-02-18 05:48:17 +0000697 err = -EINVAL;
Michael S. Tsirkin6ae7feb2014-11-23 17:24:15 +0200698 if (macvtap16_to_cpu(q, vnet_hdr.hdr_len) > len)
Arnd Bergmannb9fb9ee2010-02-18 05:48:17 +0000699 goto err;
700 }
701
702 err = -EINVAL;
Arnd Bergmann20d29d72010-01-30 12:24:26 +0000703 if (unlikely(len < ETH_HLEN))
Arnd Bergmannb9fb9ee2010-02-18 05:48:17 +0000704 goto err;
Arnd Bergmann20d29d72010-01-30 12:24:26 +0000705
Jason Wangece793f2013-07-18 10:55:16 +0800706 if (m && m->msg_control && sock_flag(&q->sk, SOCK_ZEROCOPY)) {
Al Virof5ff53b2014-06-19 15:36:49 -0400707 struct iov_iter i;
708
Michael S. Tsirkin6ae7feb2014-11-23 17:24:15 +0200709 copylen = vnet_hdr.hdr_len ?
710 macvtap16_to_cpu(q, vnet_hdr.hdr_len) : GOODCOPY_LEN;
Jason Wang16a3fa22013-11-13 14:00:40 +0800711 if (copylen > good_linear)
712 copylen = good_linear;
Jason Wang61d46bf2013-07-10 13:43:28 +0800713 linear = copylen;
Al Virof5ff53b2014-06-19 15:36:49 -0400714 i = *from;
715 iov_iter_advance(&i, copylen);
716 if (iov_iter_npages(&i, INT_MAX) <= MAX_SKB_FRAGS)
Jason Wangece793f2013-07-18 10:55:16 +0800717 zerocopy = true;
718 }
719
720 if (!zerocopy) {
Shirley Ma97bc3632011-07-06 12:26:11 +0000721 copylen = len;
Michael S. Tsirkin6ae7feb2014-11-23 17:24:15 +0200722 if (macvtap16_to_cpu(q, vnet_hdr.hdr_len) > good_linear)
Jason Wang16a3fa22013-11-13 14:00:40 +0800723 linear = good_linear;
724 else
Michael S. Tsirkin6ae7feb2014-11-23 17:24:15 +0200725 linear = macvtap16_to_cpu(q, vnet_hdr.hdr_len);
Jason Wang61d46bf2013-07-10 13:43:28 +0800726 }
Shirley Ma97bc3632011-07-06 12:26:11 +0000727
Eric Dumazet2f1d8b92015-02-27 18:35:35 -0800728 skb = macvtap_alloc_skb(&q->sk, MACVTAP_RESERVE, copylen,
Jason Wang61d46bf2013-07-10 13:43:28 +0800729 linear, noblock, &err);
Arnd Bergmann02df55d2010-02-18 05:45:36 +0000730 if (!skb)
731 goto err;
Arnd Bergmann20d29d72010-01-30 12:24:26 +0000732
Jason Wang01d66572012-05-02 11:42:06 +0800733 if (zerocopy)
Al Virof5ff53b2014-06-19 15:36:49 -0400734 err = zerocopy_sg_from_iter(skb, from);
Jason Wangece793f2013-07-18 10:55:16 +0800735 else {
Al Virof5ff53b2014-06-19 15:36:49 -0400736 err = skb_copy_datagram_from_iter(skb, 0, from, len);
Jason Wangece793f2013-07-18 10:55:16 +0800737 if (!err && m && m->msg_control) {
738 struct ubuf_info *uarg = m->msg_control;
739 uarg->callback(uarg, false);
740 }
741 }
742
Arnd Bergmann02df55d2010-02-18 05:45:36 +0000743 if (err)
Arnd Bergmannb9fb9ee2010-02-18 05:48:17 +0000744 goto err_kfree;
Arnd Bergmann20d29d72010-01-30 12:24:26 +0000745
746 skb_set_network_header(skb, ETH_HLEN);
Arnd Bergmannb9fb9ee2010-02-18 05:48:17 +0000747 skb_reset_mac_header(skb);
748 skb->protocol = eth_hdr(skb)->h_proto;
749
750 if (vnet_hdr_len) {
Michael S. Tsirkin6ae7feb2014-11-23 17:24:15 +0200751 err = macvtap_skb_from_vnet_hdr(q, skb, &vnet_hdr);
Arnd Bergmannb9fb9ee2010-02-18 05:48:17 +0000752 if (err)
753 goto err_kfree;
754 }
755
Jason Wang40893fd2013-03-26 23:11:22 +0000756 skb_probe_transport_header(skb, ETH_HLEN);
Jason Wang9b4d6692013-03-25 20:19:55 +0000757
Vlad Yasevichac4e4af2013-06-25 16:04:20 -0400758 rcu_read_lock();
759 vlan = rcu_dereference(q->vlan);
Shirley Ma97bc3632011-07-06 12:26:11 +0000760 /* copy skb_ubuf_info for callback when skb has no error */
Jason Wang01d66572012-05-02 11:42:06 +0800761 if (zerocopy) {
Shirley Ma97bc3632011-07-06 12:26:11 +0000762 skb_shinfo(skb)->destructor_arg = m->msg_control;
Jason Wang01d66572012-05-02 11:42:06 +0800763 skb_shinfo(skb)->tx_flags |= SKBTX_DEV_ZEROCOPY;
Pravin B Shelarc9af6db2013-02-11 09:27:41 +0000764 skb_shinfo(skb)->tx_flags |= SKBTX_SHARED_FRAG;
Jason Wang01d66572012-05-02 11:42:06 +0800765 }
Eric Dumazet29d79192013-08-08 08:06:14 -0700766 if (vlan) {
Vlad Yasevich6acf54f2013-12-11 13:27:10 -0500767 skb->dev = vlan->dev;
768 dev_queue_xmit(skb);
Eric Dumazet29d79192013-08-08 08:06:14 -0700769 } else {
Arnd Bergmann02df55d2010-02-18 05:45:36 +0000770 kfree_skb(skb);
Eric Dumazet29d79192013-08-08 08:06:14 -0700771 }
Vlad Yasevichac4e4af2013-06-25 16:04:20 -0400772 rcu_read_unlock();
Arnd Bergmann20d29d72010-01-30 12:24:26 +0000773
Shirley Ma97bc3632011-07-06 12:26:11 +0000774 return total_len;
Arnd Bergmann02df55d2010-02-18 05:45:36 +0000775
Arnd Bergmannb9fb9ee2010-02-18 05:48:17 +0000776err_kfree:
777 kfree_skb(skb);
778
Arnd Bergmann02df55d2010-02-18 05:45:36 +0000779err:
Vlad Yasevichac4e4af2013-06-25 16:04:20 -0400780 rcu_read_lock();
781 vlan = rcu_dereference(q->vlan);
Arnd Bergmann02df55d2010-02-18 05:45:36 +0000782 if (vlan)
Jason Wangcd3e22b2013-11-25 17:19:04 +0800783 this_cpu_inc(vlan->pcpu_stats->tx_dropped);
Vlad Yasevichac4e4af2013-06-25 16:04:20 -0400784 rcu_read_unlock();
Arnd Bergmann02df55d2010-02-18 05:45:36 +0000785
Arnd Bergmann02df55d2010-02-18 05:45:36 +0000786 return err;
Arnd Bergmann20d29d72010-01-30 12:24:26 +0000787}
788
Al Virof5ff53b2014-06-19 15:36:49 -0400789static ssize_t macvtap_write_iter(struct kiocb *iocb, struct iov_iter *from)
Arnd Bergmann20d29d72010-01-30 12:24:26 +0000790{
791 struct file *file = iocb->ki_filp;
Arnd Bergmann02df55d2010-02-18 05:45:36 +0000792 struct macvtap_queue *q = file->private_data;
Arnd Bergmann20d29d72010-01-30 12:24:26 +0000793
Al Virof5ff53b2014-06-19 15:36:49 -0400794 return macvtap_get_user(q, NULL, from, file->f_flags & O_NONBLOCK);
Arnd Bergmann20d29d72010-01-30 12:24:26 +0000795}
796
797/* Put packet to the user space buffer */
798static ssize_t macvtap_put_user(struct macvtap_queue *q,
799 const struct sk_buff *skb,
Herbert Xu6c36d2e2014-11-07 21:22:25 +0800800 struct iov_iter *iter)
Arnd Bergmann20d29d72010-01-30 12:24:26 +0000801{
Arnd Bergmann20d29d72010-01-30 12:24:26 +0000802 int ret;
Arnd Bergmannb9fb9ee2010-02-18 05:48:17 +0000803 int vnet_hdr_len = 0;
Basil Gorf09e2242012-05-03 22:55:24 +0000804 int vlan_offset = 0;
Herbert Xu6c36d2e2014-11-07 21:22:25 +0800805 int total;
Arnd Bergmannb9fb9ee2010-02-18 05:48:17 +0000806
807 if (q->flags & IFF_VNET_HDR) {
808 struct virtio_net_hdr vnet_hdr;
Michael S. Tsirkin55afbd02010-04-29 13:50:48 +0300809 vnet_hdr_len = q->vnet_hdr_sz;
Herbert Xu6c36d2e2014-11-07 21:22:25 +0800810 if (iov_iter_count(iter) < vnet_hdr_len)
Arnd Bergmannb9fb9ee2010-02-18 05:48:17 +0000811 return -EINVAL;
812
Michael S. Tsirkin6ae7feb2014-11-23 17:24:15 +0200813 macvtap_skb_to_vnet_hdr(q, skb, &vnet_hdr);
Arnd Bergmannb9fb9ee2010-02-18 05:48:17 +0000814
Herbert Xu6c36d2e2014-11-07 21:22:25 +0800815 if (copy_to_iter(&vnet_hdr, sizeof(vnet_hdr), iter) !=
816 sizeof(vnet_hdr))
Arnd Bergmannb9fb9ee2010-02-18 05:48:17 +0000817 return -EFAULT;
Jason Wang7cc76f52014-11-20 16:31:05 +0800818
819 iov_iter_advance(iter, vnet_hdr_len - sizeof(vnet_hdr));
Arnd Bergmannb9fb9ee2010-02-18 05:48:17 +0000820 }
Herbert Xu6c36d2e2014-11-07 21:22:25 +0800821 total = vnet_hdr_len;
Jason Wangce232ce2013-12-11 13:08:34 +0800822 total += skb->len;
Arnd Bergmann20d29d72010-01-30 12:24:26 +0000823
Jiri Pirkodf8a39d2015-01-13 17:13:44 +0100824 if (skb_vlan_tag_present(skb)) {
Basil Gorf09e2242012-05-03 22:55:24 +0000825 struct {
826 __be16 h_vlan_proto;
827 __be16 h_vlan_TCI;
828 } veth;
Jason Wang0fbe0d42013-07-16 13:36:34 +0800829 veth.h_vlan_proto = skb->vlan_proto;
Jiri Pirkodf8a39d2015-01-13 17:13:44 +0100830 veth.h_vlan_TCI = htons(skb_vlan_tag_get(skb));
Arnd Bergmann20d29d72010-01-30 12:24:26 +0000831
Basil Gorf09e2242012-05-03 22:55:24 +0000832 vlan_offset = offsetof(struct vlan_ethhdr, h_vlan_proto);
Jason Wangce232ce2013-12-11 13:08:34 +0800833 total += VLAN_HLEN;
Arnd Bergmann20d29d72010-01-30 12:24:26 +0000834
Herbert Xu6c36d2e2014-11-07 21:22:25 +0800835 ret = skb_copy_datagram_iter(skb, 0, iter, vlan_offset);
836 if (ret || !iov_iter_count(iter))
Basil Gorf09e2242012-05-03 22:55:24 +0000837 goto done;
838
Herbert Xu6c36d2e2014-11-07 21:22:25 +0800839 ret = copy_to_iter(&veth, sizeof(veth), iter);
840 if (ret != sizeof(veth) || !iov_iter_count(iter))
Basil Gorf09e2242012-05-03 22:55:24 +0000841 goto done;
842 }
843
Herbert Xu6c36d2e2014-11-07 21:22:25 +0800844 ret = skb_copy_datagram_iter(skb, vlan_offset, iter,
845 skb->len - vlan_offset);
Basil Gorf09e2242012-05-03 22:55:24 +0000846
847done:
Jason Wangce232ce2013-12-11 13:08:34 +0800848 return ret ? ret : total;
Arnd Bergmann20d29d72010-01-30 12:24:26 +0000849}
850
Zhi Yong Wu55ec8e22013-12-07 04:13:05 +0800851static ssize_t macvtap_do_read(struct macvtap_queue *q,
Al Viro3af0bfe2014-11-07 14:13:53 -0500852 struct iov_iter *to,
Arnd Bergmann501c7742010-02-18 05:46:50 +0000853 int noblock)
Arnd Bergmann20d29d72010-01-30 12:24:26 +0000854{
Hong zhi guoccf7e722012-06-06 22:36:27 +0000855 DEFINE_WAIT(wait);
Arnd Bergmann20d29d72010-01-30 12:24:26 +0000856 struct sk_buff *skb;
Arnd Bergmann501c7742010-02-18 05:46:50 +0000857 ssize_t ret = 0;
Arnd Bergmann20d29d72010-01-30 12:24:26 +0000858
Al Viro3af0bfe2014-11-07 14:13:53 -0500859 if (!iov_iter_count(to))
860 return 0;
861
862 while (1) {
Jason Wang89cee912013-06-05 23:54:34 +0000863 if (!noblock)
864 prepare_to_wait(sk_sleep(&q->sk), &wait,
865 TASK_INTERRUPTIBLE);
Arnd Bergmann20d29d72010-01-30 12:24:26 +0000866
867 /* Read frames from the queue */
868 skb = skb_dequeue(&q->sk.sk_receive_queue);
Al Viro3af0bfe2014-11-07 14:13:53 -0500869 if (skb)
870 break;
871 if (noblock) {
872 ret = -EAGAIN;
873 break;
Arnd Bergmann20d29d72010-01-30 12:24:26 +0000874 }
Al Viro3af0bfe2014-11-07 14:13:53 -0500875 if (signal_pending(current)) {
876 ret = -ERESTARTSYS;
877 break;
878 }
879 /* Nothing to read, let's sleep */
880 schedule();
Arnd Bergmann20d29d72010-01-30 12:24:26 +0000881 }
Al Viro3af0bfe2014-11-07 14:13:53 -0500882 if (skb) {
883 ret = macvtap_put_user(q, skb, to);
Jason Wangf51a5e82014-12-01 16:53:15 +0800884 if (unlikely(ret < 0))
885 kfree_skb(skb);
886 else
887 consume_skb(skb);
Al Viro3af0bfe2014-11-07 14:13:53 -0500888 }
Jason Wang89cee912013-06-05 23:54:34 +0000889 if (!noblock)
890 finish_wait(sk_sleep(&q->sk), &wait);
Arnd Bergmann501c7742010-02-18 05:46:50 +0000891 return ret;
892}
Arnd Bergmann20d29d72010-01-30 12:24:26 +0000893
Al Viro3af0bfe2014-11-07 14:13:53 -0500894static ssize_t macvtap_read_iter(struct kiocb *iocb, struct iov_iter *to)
Arnd Bergmann501c7742010-02-18 05:46:50 +0000895{
896 struct file *file = iocb->ki_filp;
897 struct macvtap_queue *q = file->private_data;
Al Viro3af0bfe2014-11-07 14:13:53 -0500898 ssize_t len = iov_iter_count(to), ret;
Arnd Bergmann501c7742010-02-18 05:46:50 +0000899
Al Viro3af0bfe2014-11-07 14:13:53 -0500900 ret = macvtap_do_read(q, to, file->f_flags & O_NONBLOCK);
Jason Wangce232ce2013-12-11 13:08:34 +0800901 ret = min_t(ssize_t, ret, len);
Zhi Yong Wue6ebc7f2013-12-06 14:16:50 +0800902 if (ret > 0)
903 iocb->ki_pos = ret;
Arnd Bergmann20d29d72010-01-30 12:24:26 +0000904 return ret;
905}
906
Jason Wang8f475a32013-06-05 23:54:36 +0000907static struct macvlan_dev *macvtap_get_vlan(struct macvtap_queue *q)
908{
909 struct macvlan_dev *vlan;
910
Vlad Yasevich441ac0f2013-06-25 16:04:19 -0400911 ASSERT_RTNL();
912 vlan = rtnl_dereference(q->vlan);
Jason Wang8f475a32013-06-05 23:54:36 +0000913 if (vlan)
914 dev_hold(vlan->dev);
Jason Wang8f475a32013-06-05 23:54:36 +0000915
916 return vlan;
917}
918
919static void macvtap_put_vlan(struct macvlan_dev *vlan)
920{
921 dev_put(vlan->dev);
922}
923
Jason Wang815f2362013-06-05 23:54:39 +0000924static int macvtap_ioctl_set_queue(struct file *file, unsigned int flags)
925{
926 struct macvtap_queue *q = file->private_data;
927 struct macvlan_dev *vlan;
928 int ret;
929
930 vlan = macvtap_get_vlan(q);
931 if (!vlan)
932 return -EINVAL;
933
934 if (flags & IFF_ATTACH_QUEUE)
935 ret = macvtap_enable_queue(vlan->dev, file, q);
936 else if (flags & IFF_DETACH_QUEUE)
937 ret = macvtap_disable_queue(q);
Jason Wangf57855a2013-06-13 14:23:36 +0800938 else
939 ret = -EINVAL;
Jason Wang815f2362013-06-05 23:54:39 +0000940
941 macvtap_put_vlan(vlan);
942 return ret;
943}
944
Vlad Yasevich2be5c762013-06-25 16:04:21 -0400945static int set_offload(struct macvtap_queue *q, unsigned long arg)
946{
947 struct macvlan_dev *vlan;
948 netdev_features_t features;
949 netdev_features_t feature_mask = 0;
950
951 vlan = rtnl_dereference(q->vlan);
952 if (!vlan)
953 return -ENOLINK;
954
955 features = vlan->dev->features;
956
957 if (arg & TUN_F_CSUM) {
958 feature_mask = NETIF_F_HW_CSUM;
959
960 if (arg & (TUN_F_TSO4 | TUN_F_TSO6)) {
961 if (arg & TUN_F_TSO_ECN)
962 feature_mask |= NETIF_F_TSO_ECN;
963 if (arg & TUN_F_TSO4)
964 feature_mask |= NETIF_F_TSO;
965 if (arg & TUN_F_TSO6)
966 feature_mask |= NETIF_F_TSO6;
967 }
Vlad Yaseviche3e3c422015-02-03 16:36:17 -0500968
969 if (arg & TUN_F_UFO)
970 feature_mask |= NETIF_F_UFO;
Vlad Yasevich2be5c762013-06-25 16:04:21 -0400971 }
972
973 /* tun/tap driver inverts the usage for TSO offloads, where
974 * setting the TSO bit means that the userspace wants to
975 * accept TSO frames and turning it off means that user space
976 * does not support TSO.
977 * For macvtap, we have to invert it to mean the same thing.
978 * When user space turns off TSO, we turn off GSO/LRO so that
979 * user-space will not receive TSO frames.
980 */
Vlad Yaseviche3e3c422015-02-03 16:36:17 -0500981 if (feature_mask & (NETIF_F_TSO | NETIF_F_TSO6 | NETIF_F_UFO))
Vlad Yasevich2be5c762013-06-25 16:04:21 -0400982 features |= RX_OFFLOADS;
983 else
984 features &= ~RX_OFFLOADS;
985
986 /* tap_features are the same as features on tun/tap and
987 * reflect user expectations.
988 */
Vlad Yasevicha567dd62013-08-16 15:25:00 -0400989 vlan->tap_features = feature_mask;
Vlad Yasevich2be5c762013-06-25 16:04:21 -0400990 vlan->set_features = features;
991 netdev_update_features(vlan->dev);
992
993 return 0;
994}
995
Arnd Bergmann20d29d72010-01-30 12:24:26 +0000996/*
997 * provide compatibility with generic tun/tap interface
998 */
999static long macvtap_ioctl(struct file *file, unsigned int cmd,
1000 unsigned long arg)
1001{
Arnd Bergmann02df55d2010-02-18 05:45:36 +00001002 struct macvtap_queue *q = file->private_data;
1003 struct macvlan_dev *vlan;
Arnd Bergmann20d29d72010-01-30 12:24:26 +00001004 void __user *argp = (void __user *)arg;
1005 struct ifreq __user *ifr = argp;
1006 unsigned int __user *up = argp;
Michael S. Tsirkin39ec7de2014-12-16 15:04:56 +02001007 unsigned short u;
Michael S. Tsirkin55afbd02010-04-29 13:50:48 +03001008 int __user *sp = argp;
1009 int s;
Arnd Bergmann02df55d2010-02-18 05:45:36 +00001010 int ret;
Arnd Bergmann20d29d72010-01-30 12:24:26 +00001011
1012 switch (cmd) {
1013 case TUNSETIFF:
1014 /* ignore the name, just look at flags */
1015 if (get_user(u, &ifr->ifr_flags))
1016 return -EFAULT;
Arnd Bergmannb9fb9ee2010-02-18 05:48:17 +00001017
1018 ret = 0;
Michael S. Tsirkin6ae7feb2014-11-23 17:24:15 +02001019 if ((u & ~MACVTAP_FEATURES) != (IFF_NO_PI | IFF_TAP))
Arnd Bergmannb9fb9ee2010-02-18 05:48:17 +00001020 ret = -EINVAL;
1021 else
Michael S. Tsirkin39ec7de2014-12-16 15:04:56 +02001022 q->flags = (q->flags & ~MACVTAP_FEATURES) | u;
Arnd Bergmannb9fb9ee2010-02-18 05:48:17 +00001023
1024 return ret;
Arnd Bergmann20d29d72010-01-30 12:24:26 +00001025
1026 case TUNGETIFF:
Vlad Yasevich441ac0f2013-06-25 16:04:19 -04001027 rtnl_lock();
Jason Wang8f475a32013-06-05 23:54:36 +00001028 vlan = macvtap_get_vlan(q);
Vlad Yasevich441ac0f2013-06-25 16:04:19 -04001029 if (!vlan) {
1030 rtnl_unlock();
Arnd Bergmann02df55d2010-02-18 05:45:36 +00001031 return -ENOLINK;
Vlad Yasevich441ac0f2013-06-25 16:04:19 -04001032 }
Arnd Bergmann02df55d2010-02-18 05:45:36 +00001033
1034 ret = 0;
Michael S. Tsirkin39ec7de2014-12-16 15:04:56 +02001035 u = q->flags;
Eric Dumazet13707f92011-01-26 19:28:23 +00001036 if (copy_to_user(&ifr->ifr_name, vlan->dev->name, IFNAMSIZ) ||
Michael S. Tsirkin39ec7de2014-12-16 15:04:56 +02001037 put_user(u, &ifr->ifr_flags))
Arnd Bergmann02df55d2010-02-18 05:45:36 +00001038 ret = -EFAULT;
Jason Wang8f475a32013-06-05 23:54:36 +00001039 macvtap_put_vlan(vlan);
Vlad Yasevich441ac0f2013-06-25 16:04:19 -04001040 rtnl_unlock();
Arnd Bergmann02df55d2010-02-18 05:45:36 +00001041 return ret;
Arnd Bergmann20d29d72010-01-30 12:24:26 +00001042
Jason Wang815f2362013-06-05 23:54:39 +00001043 case TUNSETQUEUE:
1044 if (get_user(u, &ifr->ifr_flags))
1045 return -EFAULT;
Vlad Yasevich441ac0f2013-06-25 16:04:19 -04001046 rtnl_lock();
1047 ret = macvtap_ioctl_set_queue(file, u);
1048 rtnl_unlock();
Jason Wang82a19eb2013-07-16 13:36:33 +08001049 return ret;
Jason Wang815f2362013-06-05 23:54:39 +00001050
Arnd Bergmann20d29d72010-01-30 12:24:26 +00001051 case TUNGETFEATURES:
Michael S. Tsirkin6ae7feb2014-11-23 17:24:15 +02001052 if (put_user(IFF_TAP | IFF_NO_PI | MACVTAP_FEATURES, up))
Arnd Bergmann20d29d72010-01-30 12:24:26 +00001053 return -EFAULT;
1054 return 0;
1055
1056 case TUNSETSNDBUF:
1057 if (get_user(u, up))
1058 return -EFAULT;
1059
Arnd Bergmann20d29d72010-01-30 12:24:26 +00001060 q->sk.sk_sndbuf = u;
Arnd Bergmann20d29d72010-01-30 12:24:26 +00001061 return 0;
1062
Michael S. Tsirkin55afbd02010-04-29 13:50:48 +03001063 case TUNGETVNETHDRSZ:
1064 s = q->vnet_hdr_sz;
1065 if (put_user(s, sp))
1066 return -EFAULT;
1067 return 0;
1068
1069 case TUNSETVNETHDRSZ:
1070 if (get_user(s, sp))
1071 return -EFAULT;
1072 if (s < (int)sizeof(struct virtio_net_hdr))
1073 return -EINVAL;
1074
1075 q->vnet_hdr_sz = s;
1076 return 0;
1077
Michael S. Tsirkin01b07fb2014-12-16 15:05:10 +02001078 case TUNGETVNETLE:
1079 s = !!(q->flags & MACVTAP_VNET_LE);
1080 if (put_user(s, sp))
1081 return -EFAULT;
1082 return 0;
1083
1084 case TUNSETVNETLE:
1085 if (get_user(s, sp))
1086 return -EFAULT;
1087 if (s)
1088 q->flags |= MACVTAP_VNET_LE;
1089 else
1090 q->flags &= ~MACVTAP_VNET_LE;
1091 return 0;
1092
Arnd Bergmann20d29d72010-01-30 12:24:26 +00001093 case TUNSETOFFLOAD:
1094 /* let the user check for future flags */
1095 if (arg & ~(TUN_F_CSUM | TUN_F_TSO4 | TUN_F_TSO6 |
Vlad Yaseviche3e3c422015-02-03 16:36:17 -05001096 TUN_F_TSO_ECN | TUN_F_UFO))
Arnd Bergmann20d29d72010-01-30 12:24:26 +00001097 return -EINVAL;
1098
Vlad Yasevich2be5c762013-06-25 16:04:21 -04001099 rtnl_lock();
1100 ret = set_offload(q, arg);
1101 rtnl_unlock();
1102 return ret;
Arnd Bergmann20d29d72010-01-30 12:24:26 +00001103
1104 default:
1105 return -EINVAL;
1106 }
1107}
1108
1109#ifdef CONFIG_COMPAT
1110static long macvtap_compat_ioctl(struct file *file, unsigned int cmd,
1111 unsigned long arg)
1112{
1113 return macvtap_ioctl(file, cmd, (unsigned long)compat_ptr(arg));
1114}
1115#endif
1116
1117static const struct file_operations macvtap_fops = {
1118 .owner = THIS_MODULE,
1119 .open = macvtap_open,
1120 .release = macvtap_release,
Al Viro3af0bfe2014-11-07 14:13:53 -05001121 .read = new_sync_read,
Al Virof5ff53b2014-06-19 15:36:49 -04001122 .write = new_sync_write,
Al Viro3af0bfe2014-11-07 14:13:53 -05001123 .read_iter = macvtap_read_iter,
Al Virof5ff53b2014-06-19 15:36:49 -04001124 .write_iter = macvtap_write_iter,
Arnd Bergmann20d29d72010-01-30 12:24:26 +00001125 .poll = macvtap_poll,
1126 .llseek = no_llseek,
1127 .unlocked_ioctl = macvtap_ioctl,
1128#ifdef CONFIG_COMPAT
1129 .compat_ioctl = macvtap_compat_ioctl,
1130#endif
1131};
1132
Ying Xue1b784142015-03-02 15:37:48 +08001133static int macvtap_sendmsg(struct socket *sock, struct msghdr *m,
1134 size_t total_len)
Arnd Bergmann501c7742010-02-18 05:46:50 +00001135{
1136 struct macvtap_queue *q = container_of(sock, struct macvtap_queue, sock);
Al Viroc0371da2014-11-24 10:42:55 -05001137 return macvtap_get_user(q, m, &m->msg_iter, m->msg_flags & MSG_DONTWAIT);
Arnd Bergmann501c7742010-02-18 05:46:50 +00001138}
1139
Ying Xue1b784142015-03-02 15:37:48 +08001140static int macvtap_recvmsg(struct socket *sock, struct msghdr *m,
1141 size_t total_len, int flags)
Arnd Bergmann501c7742010-02-18 05:46:50 +00001142{
1143 struct macvtap_queue *q = container_of(sock, struct macvtap_queue, sock);
1144 int ret;
1145 if (flags & ~(MSG_DONTWAIT|MSG_TRUNC))
1146 return -EINVAL;
Al Viroc0371da2014-11-24 10:42:55 -05001147 ret = macvtap_do_read(q, &m->msg_iter, flags & MSG_DONTWAIT);
David S. Millerde2aa472013-12-10 22:06:18 -05001148 if (ret > total_len) {
1149 m->msg_flags |= MSG_TRUNC;
1150 ret = flags & MSG_TRUNC ? ret : total_len;
1151 }
Arnd Bergmann501c7742010-02-18 05:46:50 +00001152 return ret;
1153}
1154
1155/* Ops structure to mimic raw sockets with tun */
1156static const struct proto_ops macvtap_socket_ops = {
1157 .sendmsg = macvtap_sendmsg,
1158 .recvmsg = macvtap_recvmsg,
1159};
1160
1161/* Get an underlying socket object from tun file. Returns error unless file is
1162 * attached to a device. The returned object works like a packet socket, it
1163 * can be used for sock_sendmsg/sock_recvmsg. The caller is responsible for
1164 * holding a reference to the file for as long as the socket is in use. */
1165struct socket *macvtap_get_socket(struct file *file)
1166{
1167 struct macvtap_queue *q;
1168 if (file->f_op != &macvtap_fops)
1169 return ERR_PTR(-EINVAL);
1170 q = file->private_data;
1171 if (!q)
1172 return ERR_PTR(-EBADFD);
1173 return &q->sock;
1174}
1175EXPORT_SYMBOL_GPL(macvtap_get_socket);
1176
Eric W. Biederman9bf19072011-10-20 04:28:46 +00001177static int macvtap_device_event(struct notifier_block *unused,
1178 unsigned long event, void *ptr)
1179{
Jiri Pirko351638e2013-05-28 01:30:21 +00001180 struct net_device *dev = netdev_notifier_info_to_dev(ptr);
Eric W. Biedermane09eff72011-10-20 04:29:24 +00001181 struct macvlan_dev *vlan;
Eric W. Biederman9bf19072011-10-20 04:28:46 +00001182 struct device *classdev;
1183 dev_t devt;
Eric W. Biedermane09eff72011-10-20 04:29:24 +00001184 int err;
Eric W. Biederman9bf19072011-10-20 04:28:46 +00001185
1186 if (dev->rtnl_link_ops != &macvtap_link_ops)
1187 return NOTIFY_DONE;
1188
Eric W. Biedermane09eff72011-10-20 04:29:24 +00001189 vlan = netdev_priv(dev);
Eric W. Biederman9bf19072011-10-20 04:28:46 +00001190
1191 switch (event) {
1192 case NETDEV_REGISTER:
1193 /* Create the device node here after the network device has
1194 * been registered but before register_netdevice has
1195 * finished running.
1196 */
Eric W. Biedermane09eff72011-10-20 04:29:24 +00001197 err = macvtap_get_minor(vlan);
1198 if (err)
1199 return notifier_from_errno(err);
1200
1201 devt = MKDEV(MAJOR(macvtap_major), vlan->minor);
Eric W. Biederman9bf19072011-10-20 04:28:46 +00001202 classdev = device_create(macvtap_class, &dev->dev, devt,
1203 dev, "tap%d", dev->ifindex);
Eric W. Biedermane09eff72011-10-20 04:29:24 +00001204 if (IS_ERR(classdev)) {
1205 macvtap_free_minor(vlan);
Eric W. Biederman9bf19072011-10-20 04:28:46 +00001206 return notifier_from_errno(PTR_ERR(classdev));
Eric W. Biedermane09eff72011-10-20 04:29:24 +00001207 }
Eric W. Biederman9bf19072011-10-20 04:28:46 +00001208 break;
1209 case NETDEV_UNREGISTER:
Eric W. Biedermane09eff72011-10-20 04:29:24 +00001210 devt = MKDEV(MAJOR(macvtap_major), vlan->minor);
Eric W. Biederman9bf19072011-10-20 04:28:46 +00001211 device_destroy(macvtap_class, devt);
Eric W. Biedermane09eff72011-10-20 04:29:24 +00001212 macvtap_free_minor(vlan);
Eric W. Biederman9bf19072011-10-20 04:28:46 +00001213 break;
1214 }
1215
1216 return NOTIFY_DONE;
1217}
1218
1219static struct notifier_block macvtap_notifier_block __read_mostly = {
1220 .notifier_call = macvtap_device_event,
1221};
1222
Arnd Bergmann20d29d72010-01-30 12:24:26 +00001223static int macvtap_init(void)
1224{
1225 int err;
1226
1227 err = alloc_chrdev_region(&macvtap_major, 0,
1228 MACVTAP_NUM_DEVS, "macvtap");
1229 if (err)
1230 goto out1;
1231
1232 cdev_init(&macvtap_cdev, &macvtap_fops);
1233 err = cdev_add(&macvtap_cdev, macvtap_major, MACVTAP_NUM_DEVS);
1234 if (err)
1235 goto out2;
1236
1237 macvtap_class = class_create(THIS_MODULE, "macvtap");
1238 if (IS_ERR(macvtap_class)) {
1239 err = PTR_ERR(macvtap_class);
1240 goto out3;
1241 }
1242
Eric W. Biederman9bf19072011-10-20 04:28:46 +00001243 err = register_netdevice_notifier(&macvtap_notifier_block);
Arnd Bergmann20d29d72010-01-30 12:24:26 +00001244 if (err)
1245 goto out4;
1246
Eric W. Biederman9bf19072011-10-20 04:28:46 +00001247 err = macvlan_link_register(&macvtap_link_ops);
1248 if (err)
1249 goto out5;
1250
Arnd Bergmann20d29d72010-01-30 12:24:26 +00001251 return 0;
1252
Eric W. Biederman9bf19072011-10-20 04:28:46 +00001253out5:
1254 unregister_netdevice_notifier(&macvtap_notifier_block);
Arnd Bergmann20d29d72010-01-30 12:24:26 +00001255out4:
1256 class_unregister(macvtap_class);
1257out3:
1258 cdev_del(&macvtap_cdev);
1259out2:
1260 unregister_chrdev_region(macvtap_major, MACVTAP_NUM_DEVS);
1261out1:
1262 return err;
1263}
1264module_init(macvtap_init);
1265
1266static void macvtap_exit(void)
1267{
1268 rtnl_link_unregister(&macvtap_link_ops);
Eric W. Biederman9bf19072011-10-20 04:28:46 +00001269 unregister_netdevice_notifier(&macvtap_notifier_block);
Arnd Bergmann20d29d72010-01-30 12:24:26 +00001270 class_unregister(macvtap_class);
1271 cdev_del(&macvtap_cdev);
1272 unregister_chrdev_region(macvtap_major, MACVTAP_NUM_DEVS);
1273}
1274module_exit(macvtap_exit);
1275
1276MODULE_ALIAS_RTNL_LINK("macvtap");
1277MODULE_AUTHOR("Arnd Bergmann <arnd@arndb.de>");
1278MODULE_LICENSE("GPL");