blob: edd77342773a8d4ef0713717adfa5bcf6bdf44f7 [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
Greg Kurz8b8e6582015-04-24 14:50:36 +020051#define MACVTAP_VNET_BE 0x40000000
52
53#ifdef CONFIG_TUN_VNET_CROSS_LE
54static inline bool macvtap_legacy_is_little_endian(struct macvtap_queue *q)
55{
56 return q->flags & MACVTAP_VNET_BE ? false :
57 virtio_legacy_is_little_endian();
58}
59
60static long macvtap_get_vnet_be(struct macvtap_queue *q, int __user *sp)
61{
62 int s = !!(q->flags & MACVTAP_VNET_BE);
63
64 if (put_user(s, sp))
65 return -EFAULT;
66
67 return 0;
68}
69
70static long macvtap_set_vnet_be(struct macvtap_queue *q, int __user *sp)
71{
72 int s;
73
74 if (get_user(s, sp))
75 return -EFAULT;
76
77 if (s)
78 q->flags |= MACVTAP_VNET_BE;
79 else
80 q->flags &= ~MACVTAP_VNET_BE;
81
82 return 0;
83}
84#else
85static inline bool macvtap_legacy_is_little_endian(struct macvtap_queue *q)
86{
87 return virtio_legacy_is_little_endian();
88}
89
90static long macvtap_get_vnet_be(struct macvtap_queue *q, int __user *argp)
91{
92 return -EINVAL;
93}
94
95static long macvtap_set_vnet_be(struct macvtap_queue *q, int __user *argp)
96{
97 return -EINVAL;
98}
99#endif /* CONFIG_TUN_VNET_CROSS_LE */
Michael S. Tsirkin6ae7feb2014-11-23 17:24:15 +0200100
Greg Kurz5b11e152015-04-24 14:24:48 +0200101static inline bool macvtap_is_little_endian(struct macvtap_queue *q)
102{
Greg Kurz7d824102015-04-24 14:26:24 +0200103 return q->flags & MACVTAP_VNET_LE ||
Greg Kurz8b8e6582015-04-24 14:50:36 +0200104 macvtap_legacy_is_little_endian(q);
Greg Kurz5b11e152015-04-24 14:24:48 +0200105}
Michael S. Tsirkin6ae7feb2014-11-23 17:24:15 +0200106
107static inline u16 macvtap16_to_cpu(struct macvtap_queue *q, __virtio16 val)
108{
Greg Kurz5b11e152015-04-24 14:24:48 +0200109 return __virtio16_to_cpu(macvtap_is_little_endian(q), val);
Michael S. Tsirkin6ae7feb2014-11-23 17:24:15 +0200110}
111
112static inline __virtio16 cpu_to_macvtap16(struct macvtap_queue *q, u16 val)
113{
Greg Kurz5b11e152015-04-24 14:24:48 +0200114 return __cpu_to_virtio16(macvtap_is_little_endian(q), val);
Michael S. Tsirkin6ae7feb2014-11-23 17:24:15 +0200115}
116
Arnd Bergmann20d29d72010-01-30 12:24:26 +0000117static struct proto macvtap_proto = {
118 .name = "macvtap",
119 .owner = THIS_MODULE,
120 .obj_size = sizeof (struct macvtap_queue),
121};
122
123/*
Eric W. Biedermane09eff72011-10-20 04:29:24 +0000124 * Variables for dealing with macvtaps device numbers.
Arnd Bergmann20d29d72010-01-30 12:24:26 +0000125 */
David S. Miller1ebed712010-07-10 19:25:50 -0700126static dev_t macvtap_major;
Eric W. Biedermane09eff72011-10-20 04:29:24 +0000127#define MACVTAP_NUM_DEVS (1U << MINORBITS)
128static DEFINE_MUTEX(minor_lock);
129static DEFINE_IDR(minor_idr);
130
Shirley Ma97bc3632011-07-06 12:26:11 +0000131#define GOODCOPY_LEN 128
Arnd Bergmann20d29d72010-01-30 12:24:26 +0000132static struct class *macvtap_class;
133static struct cdev macvtap_cdev;
134
Arnd Bergmann501c7742010-02-18 05:46:50 +0000135static const struct proto_ops macvtap_socket_ops;
136
Vlad Yasevich2be5c762013-06-25 16:04:21 -0400137#define TUN_OFFLOADS (NETIF_F_HW_CSUM | NETIF_F_TSO_ECN | NETIF_F_TSO | \
Vlad Yaseviche3e3c422015-02-03 16:36:17 -0500138 NETIF_F_TSO6 | NETIF_F_UFO)
Vlad Yasevich2be5c762013-06-25 16:04:21 -0400139#define RX_OFFLOADS (NETIF_F_GRO | NETIF_F_LRO)
Vlad Yasevicha567dd62013-08-16 15:25:00 -0400140#define TAP_FEATURES (NETIF_F_GSO | NETIF_F_SG)
141
Vlad Yasevich6acf54f2013-12-11 13:27:10 -0500142static struct macvlan_dev *macvtap_get_vlan_rcu(const struct net_device *dev)
143{
144 return rcu_dereference(dev->rx_handler_data);
145}
146
Arnd Bergmann20d29d72010-01-30 12:24:26 +0000147/*
148 * RCU usage:
Arnd Bergmann02df55d2010-02-18 05:45:36 +0000149 * The macvtap_queue and the macvlan_dev are loosely coupled, the
150 * pointers from one to the other can only be read while rcu_read_lock
Vlad Yasevich441ac0f2013-06-25 16:04:19 -0400151 * or rtnl is held.
Arnd Bergmann20d29d72010-01-30 12:24:26 +0000152 *
Arnd Bergmann02df55d2010-02-18 05:45:36 +0000153 * Both the file and the macvlan_dev hold a reference on the macvtap_queue
154 * through sock_hold(&q->sk). When the macvlan_dev goes away first,
155 * q->vlan becomes inaccessible. When the files gets closed,
156 * macvtap_get_queue() fails.
Arnd Bergmann20d29d72010-01-30 12:24:26 +0000157 *
Arnd Bergmann02df55d2010-02-18 05:45:36 +0000158 * There may still be references to the struct sock inside of the
159 * queue from outbound SKBs, but these never reference back to the
160 * file or the dev. The data structure is freed through __sk_free
161 * when both our references and any pending SKBs are gone.
Arnd Bergmann20d29d72010-01-30 12:24:26 +0000162 */
Arnd Bergmann20d29d72010-01-30 12:24:26 +0000163
Jason Wang815f2362013-06-05 23:54:39 +0000164static int macvtap_enable_queue(struct net_device *dev, struct file *file,
Arnd Bergmann20d29d72010-01-30 12:24:26 +0000165 struct macvtap_queue *q)
166{
167 struct macvlan_dev *vlan = netdev_priv(dev);
Jason Wang815f2362013-06-05 23:54:39 +0000168 int err = -EINVAL;
169
Vlad Yasevich441ac0f2013-06-25 16:04:19 -0400170 ASSERT_RTNL();
Jason Wang815f2362013-06-05 23:54:39 +0000171
172 if (q->enabled)
173 goto out;
174
175 err = 0;
176 rcu_assign_pointer(vlan->taps[vlan->numvtaps], q);
177 q->queue_index = vlan->numvtaps;
178 q->enabled = true;
179
180 vlan->numvtaps++;
181out:
Jason Wang815f2362013-06-05 23:54:39 +0000182 return err;
183}
184
Vlad Yasevich40b8fe42014-09-22 16:34:17 -0400185/* Requires RTNL */
Jason Wang815f2362013-06-05 23:54:39 +0000186static int macvtap_set_queue(struct net_device *dev, struct file *file,
187 struct macvtap_queue *q)
188{
189 struct macvlan_dev *vlan = netdev_priv(dev);
Arnd Bergmann20d29d72010-01-30 12:24:26 +0000190
Jason Wang815f2362013-06-05 23:54:39 +0000191 if (vlan->numqueues == MAX_MACVTAP_QUEUES)
Vlad Yasevich40b8fe42014-09-22 16:34:17 -0400192 return -EBUSY;
Arnd Bergmann20d29d72010-01-30 12:24:26 +0000193
Arnd Bergmann02df55d2010-02-18 05:45:36 +0000194 rcu_assign_pointer(q->vlan, vlan);
Jason Wang376b1aa2013-06-05 23:54:38 +0000195 rcu_assign_pointer(vlan->taps[vlan->numvtaps], q);
Arnd Bergmann02df55d2010-02-18 05:45:36 +0000196 sock_hold(&q->sk);
Arnd Bergmann20d29d72010-01-30 12:24:26 +0000197
198 q->file = file;
Jason Wang376b1aa2013-06-05 23:54:38 +0000199 q->queue_index = vlan->numvtaps;
Jason Wang815f2362013-06-05 23:54:39 +0000200 q->enabled = true;
Arnd Bergmann02df55d2010-02-18 05:45:36 +0000201 file->private_data = q;
Jason Wang815f2362013-06-05 23:54:39 +0000202 list_add_tail(&q->next, &vlan->queue_list);
Arnd Bergmann20d29d72010-01-30 12:24:26 +0000203
Krishna Kumar1565c7c2010-08-04 06:15:59 +0000204 vlan->numvtaps++;
Jason Wang815f2362013-06-05 23:54:39 +0000205 vlan->numqueues++;
Krishna Kumar1565c7c2010-08-04 06:15:59 +0000206
Vlad Yasevich40b8fe42014-09-22 16:34:17 -0400207 return 0;
Arnd Bergmann20d29d72010-01-30 12:24:26 +0000208}
209
Vlad Yasevich441ac0f2013-06-25 16:04:19 -0400210static int macvtap_disable_queue(struct macvtap_queue *q)
Jason Wang815f2362013-06-05 23:54:39 +0000211{
212 struct macvlan_dev *vlan;
213 struct macvtap_queue *nq;
214
Vlad Yasevich441ac0f2013-06-25 16:04:19 -0400215 ASSERT_RTNL();
Jason Wang815f2362013-06-05 23:54:39 +0000216 if (!q->enabled)
217 return -EINVAL;
218
Vlad Yasevich441ac0f2013-06-25 16:04:19 -0400219 vlan = rtnl_dereference(q->vlan);
220
Jason Wang815f2362013-06-05 23:54:39 +0000221 if (vlan) {
222 int index = q->queue_index;
223 BUG_ON(index >= vlan->numvtaps);
Vlad Yasevich441ac0f2013-06-25 16:04:19 -0400224 nq = rtnl_dereference(vlan->taps[vlan->numvtaps - 1]);
Jason Wang815f2362013-06-05 23:54:39 +0000225 nq->queue_index = index;
226
227 rcu_assign_pointer(vlan->taps[index], nq);
228 RCU_INIT_POINTER(vlan->taps[vlan->numvtaps - 1], NULL);
229 q->enabled = false;
230
231 vlan->numvtaps--;
232 }
233
234 return 0;
235}
236
Arnd Bergmann20d29d72010-01-30 12:24:26 +0000237/*
Arnd Bergmann02df55d2010-02-18 05:45:36 +0000238 * The file owning the queue got closed, give up both
239 * the reference that the files holds as well as the
240 * one from the macvlan_dev if that still exists.
Arnd Bergmann20d29d72010-01-30 12:24:26 +0000241 *
242 * Using the spinlock makes sure that we don't get
243 * to the queue again after destroying it.
Arnd Bergmann20d29d72010-01-30 12:24:26 +0000244 */
Arnd Bergmann02df55d2010-02-18 05:45:36 +0000245static void macvtap_put_queue(struct macvtap_queue *q)
Arnd Bergmann20d29d72010-01-30 12:24:26 +0000246{
Arnd Bergmann02df55d2010-02-18 05:45:36 +0000247 struct macvlan_dev *vlan;
Arnd Bergmann20d29d72010-01-30 12:24:26 +0000248
Vlad Yasevich441ac0f2013-06-25 16:04:19 -0400249 rtnl_lock();
250 vlan = rtnl_dereference(q->vlan);
251
Arnd Bergmann02df55d2010-02-18 05:45:36 +0000252 if (vlan) {
Jason Wang815f2362013-06-05 23:54:39 +0000253 if (q->enabled)
Vlad Yasevich441ac0f2013-06-25 16:04:19 -0400254 BUG_ON(macvtap_disable_queue(q));
Krishna Kumar1565c7c2010-08-04 06:15:59 +0000255
Jason Wang815f2362013-06-05 23:54:39 +0000256 vlan->numqueues--;
Eric Dumazet2cfa5a02011-11-23 07:09:32 +0000257 RCU_INIT_POINTER(q->vlan, NULL);
Arnd Bergmann02df55d2010-02-18 05:45:36 +0000258 sock_put(&q->sk);
Jason Wang815f2362013-06-05 23:54:39 +0000259 list_del_init(&q->next);
Arnd Bergmann20d29d72010-01-30 12:24:26 +0000260 }
261
Vlad Yasevich441ac0f2013-06-25 16:04:19 -0400262 rtnl_unlock();
Arnd Bergmann20d29d72010-01-30 12:24:26 +0000263
264 synchronize_rcu();
265 sock_put(&q->sk);
266}
267
268/*
Krishna Kumar1565c7c2010-08-04 06:15:59 +0000269 * Select a queue based on the rxq of the device on which this packet
270 * arrived. If the incoming device is not mq, calculate a flow hash
271 * to select a queue. If all fails, find the first available queue.
272 * Cache vlan->numvtaps since it can become zero during the execution
273 * of this function.
Arnd Bergmann20d29d72010-01-30 12:24:26 +0000274 */
275static struct macvtap_queue *macvtap_get_queue(struct net_device *dev,
276 struct sk_buff *skb)
277{
278 struct macvlan_dev *vlan = netdev_priv(dev);
Krishna Kumar1565c7c2010-08-04 06:15:59 +0000279 struct macvtap_queue *tap = NULL;
Jason Wang815f2362013-06-05 23:54:39 +0000280 /* Access to taps array is protected by rcu, but access to numvtaps
281 * isn't. Below we use it to lookup a queue, but treat it as a hint
282 * and validate that the result isn't NULL - in case we are
283 * racing against queue removal.
284 */
Jason Wanged0483f2013-06-05 23:54:33 +0000285 int numvtaps = ACCESS_ONCE(vlan->numvtaps);
Krishna Kumar1565c7c2010-08-04 06:15:59 +0000286 __u32 rxq;
Arnd Bergmann20d29d72010-01-30 12:24:26 +0000287
Krishna Kumar1565c7c2010-08-04 06:15:59 +0000288 if (!numvtaps)
289 goto out;
290
Krishna Kumaref0002b2011-11-23 22:17:14 +0000291 /* Check if we can use flow to select a queue */
Tom Herbert3958afa1b2013-12-15 22:12:06 -0800292 rxq = skb_get_hash(skb);
Krishna Kumaref0002b2011-11-23 22:17:14 +0000293 if (rxq) {
294 tap = rcu_dereference(vlan->taps[rxq % numvtaps]);
Jason Wang376b1aa2013-06-05 23:54:38 +0000295 goto out;
Krishna Kumaref0002b2011-11-23 22:17:14 +0000296 }
297
Krishna Kumar1565c7c2010-08-04 06:15:59 +0000298 if (likely(skb_rx_queue_recorded(skb))) {
299 rxq = skb_get_rx_queue(skb);
300
301 while (unlikely(rxq >= numvtaps))
302 rxq -= numvtaps;
303
304 tap = rcu_dereference(vlan->taps[rxq]);
Jason Wang376b1aa2013-06-05 23:54:38 +0000305 goto out;
Krishna Kumar1565c7c2010-08-04 06:15:59 +0000306 }
307
Jason Wang376b1aa2013-06-05 23:54:38 +0000308 tap = rcu_dereference(vlan->taps[0]);
Krishna Kumar1565c7c2010-08-04 06:15:59 +0000309out:
310 return tap;
Arnd Bergmann20d29d72010-01-30 12:24:26 +0000311}
312
Arnd Bergmann02df55d2010-02-18 05:45:36 +0000313/*
314 * The net_device is going away, give up the reference
Krishna Kumar1565c7c2010-08-04 06:15:59 +0000315 * that it holds on all queues and safely set the pointer
316 * from the queues to NULL.
Arnd Bergmann02df55d2010-02-18 05:45:36 +0000317 */
Arnd Bergmann20d29d72010-01-30 12:24:26 +0000318static void macvtap_del_queues(struct net_device *dev)
319{
320 struct macvlan_dev *vlan = netdev_priv(dev);
Pankaj Guptadfe816c2015-06-19 19:47:53 +0530321 struct macvtap_queue *q, *tmp;
Arnd Bergmann20d29d72010-01-30 12:24:26 +0000322
Vlad Yasevich441ac0f2013-06-25 16:04:19 -0400323 ASSERT_RTNL();
Jason Wang815f2362013-06-05 23:54:39 +0000324 list_for_each_entry_safe(q, tmp, &vlan->queue_list, next) {
325 list_del_init(&q->next);
Jason Wang376b1aa2013-06-05 23:54:38 +0000326 RCU_INIT_POINTER(q->vlan, NULL);
Jason Wang815f2362013-06-05 23:54:39 +0000327 if (q->enabled)
328 vlan->numvtaps--;
329 vlan->numqueues--;
Pankaj Guptadfe816c2015-06-19 19:47:53 +0530330 sock_put(&q->sk);
Arnd Bergmann02df55d2010-02-18 05:45:36 +0000331 }
Jason Wang815f2362013-06-05 23:54:39 +0000332 BUG_ON(vlan->numvtaps);
333 BUG_ON(vlan->numqueues);
Eric W. Biederman99f34b32011-10-20 04:26:01 +0000334 /* guarantee that any future macvtap_set_queue will fail */
335 vlan->numvtaps = MAX_MACVTAP_QUEUES;
Arnd Bergmann20d29d72010-01-30 12:24:26 +0000336}
337
Vlad Yasevich6acf54f2013-12-11 13:27:10 -0500338static rx_handler_result_t macvtap_handle_frame(struct sk_buff **pskb)
Arnd Bergmann20d29d72010-01-30 12:24:26 +0000339{
Vlad Yasevich6acf54f2013-12-11 13:27:10 -0500340 struct sk_buff *skb = *pskb;
341 struct net_device *dev = skb->dev;
342 struct macvlan_dev *vlan;
343 struct macvtap_queue *q;
Vlad Yasevicha567dd62013-08-16 15:25:00 -0400344 netdev_features_t features = TAP_FEATURES;
345
Vlad Yasevich6acf54f2013-12-11 13:27:10 -0500346 vlan = macvtap_get_vlan_rcu(dev);
347 if (!vlan)
348 return RX_HANDLER_PASS;
349
350 q = macvtap_get_queue(dev, skb);
Arnd Bergmann20d29d72010-01-30 12:24:26 +0000351 if (!q)
Vlad Yasevich6acf54f2013-12-11 13:27:10 -0500352 return RX_HANDLER_PASS;
Herbert Xu8a357472010-07-21 21:44:31 +0000353
354 if (skb_queue_len(&q->sk.sk_receive_queue) >= dev->tx_queue_len)
355 goto drop;
Arnd Bergmann20d29d72010-01-30 12:24:26 +0000356
Vlad Yasevich6acf54f2013-12-11 13:27:10 -0500357 skb_push(skb, ETH_HLEN);
358
Vlad Yasevich3e4f8b72013-06-25 16:04:22 -0400359 /* Apply the forward feature mask so that we perform segmentation
Vlad Yaseviche5733322013-08-16 15:25:02 -0400360 * according to users wishes. This only works if VNET_HDR is
361 * enabled.
Vlad Yasevich3e4f8b72013-06-25 16:04:22 -0400362 */
Vlad Yaseviche5733322013-08-16 15:25:02 -0400363 if (q->flags & IFF_VNET_HDR)
364 features |= vlan->tap_features;
Johannes Berg8b86a612015-04-17 15:45:04 +0200365 if (netif_needs_gso(skb, features)) {
Vlad Yasevich3e4f8b72013-06-25 16:04:22 -0400366 struct sk_buff *segs = __skb_gso_segment(skb, features, false);
367
368 if (IS_ERR(segs))
369 goto drop;
370
371 if (!segs) {
372 skb_queue_tail(&q->sk.sk_receive_queue, skb);
373 goto wake_up;
374 }
375
376 kfree_skb(skb);
377 while (segs) {
378 struct sk_buff *nskb = segs->next;
379
380 segs->next = NULL;
381 skb_queue_tail(&q->sk.sk_receive_queue, segs);
382 segs = nskb;
383 }
384 } else {
Vlad Yasevichcbdb0422014-04-29 10:09:50 -0400385 /* If we receive a partial checksum and the tap side
386 * doesn't support checksum offload, compute the checksum.
387 * Note: it doesn't matter which checksum feature to
388 * check, we either support them all or none.
389 */
390 if (skb->ip_summed == CHECKSUM_PARTIAL &&
391 !(features & NETIF_F_ALL_CSUM) &&
392 skb_checksum_help(skb))
393 goto drop;
Vlad Yasevich3e4f8b72013-06-25 16:04:22 -0400394 skb_queue_tail(&q->sk.sk_receive_queue, skb);
395 }
396
397wake_up:
Eric Dumazet4a4771a2010-04-25 22:20:06 +0000398 wake_up_interruptible_poll(sk_sleep(&q->sk), POLLIN | POLLRDNORM | POLLRDBAND);
Vlad Yasevich6acf54f2013-12-11 13:27:10 -0500399 return RX_HANDLER_CONSUMED;
Herbert Xu8a357472010-07-21 21:44:31 +0000400
401drop:
Vlad Yasevich6acf54f2013-12-11 13:27:10 -0500402 /* Count errors/drops only here, thus don't care about args. */
403 macvlan_count_rx(vlan, 0, 0, 0);
Herbert Xu8a357472010-07-21 21:44:31 +0000404 kfree_skb(skb);
Vlad Yasevich6acf54f2013-12-11 13:27:10 -0500405 return RX_HANDLER_CONSUMED;
Arnd Bergmann20d29d72010-01-30 12:24:26 +0000406}
407
Eric W. Biedermane09eff72011-10-20 04:29:24 +0000408static int macvtap_get_minor(struct macvlan_dev *vlan)
409{
410 int retval = -ENOMEM;
Eric W. Biedermane09eff72011-10-20 04:29:24 +0000411
412 mutex_lock(&minor_lock);
Tejun Heoec09ebc2013-02-27 17:04:34 -0800413 retval = idr_alloc(&minor_idr, vlan, 1, MACVTAP_NUM_DEVS, GFP_KERNEL);
414 if (retval >= 0) {
415 vlan->minor = retval;
416 } else if (retval == -ENOSPC) {
Eric W. Biedermane09eff72011-10-20 04:29:24 +0000417 printk(KERN_ERR "too many macvtap devices\n");
418 retval = -EINVAL;
Eric W. Biedermane09eff72011-10-20 04:29:24 +0000419 }
Eric W. Biedermane09eff72011-10-20 04:29:24 +0000420 mutex_unlock(&minor_lock);
Tejun Heoec09ebc2013-02-27 17:04:34 -0800421 return retval < 0 ? retval : 0;
Eric W. Biedermane09eff72011-10-20 04:29:24 +0000422}
423
424static void macvtap_free_minor(struct macvlan_dev *vlan)
425{
426 mutex_lock(&minor_lock);
427 if (vlan->minor) {
428 idr_remove(&minor_idr, vlan->minor);
429 vlan->minor = 0;
430 }
431 mutex_unlock(&minor_lock);
432}
433
434static struct net_device *dev_get_by_macvtap_minor(int minor)
435{
436 struct net_device *dev = NULL;
437 struct macvlan_dev *vlan;
438
439 mutex_lock(&minor_lock);
440 vlan = idr_find(&minor_idr, minor);
441 if (vlan) {
442 dev = vlan->dev;
443 dev_hold(dev);
444 }
445 mutex_unlock(&minor_lock);
446 return dev;
447}
448
Arnd Bergmann20d29d72010-01-30 12:24:26 +0000449static int macvtap_newlink(struct net *src_net,
450 struct net_device *dev,
451 struct nlattr *tb[],
452 struct nlattr *data[])
453{
Jason Wang815f2362013-06-05 23:54:39 +0000454 struct macvlan_dev *vlan = netdev_priv(dev);
Vlad Yasevich6acf54f2013-12-11 13:27:10 -0500455 int err;
456
Jason Wang815f2362013-06-05 23:54:39 +0000457 INIT_LIST_HEAD(&vlan->queue_list);
458
Vlad Yasevich2be5c762013-06-25 16:04:21 -0400459 /* Since macvlan supports all offloads by default, make
460 * tap support all offloads also.
461 */
462 vlan->tap_features = TUN_OFFLOADS;
463
Vlad Yasevich6acf54f2013-12-11 13:27:10 -0500464 err = netdev_rx_handler_register(dev, macvtap_handle_frame, vlan);
465 if (err)
466 return err;
467
Eric W. Biederman9bf19072011-10-20 04:28:46 +0000468 /* Don't put anything that may fail after macvlan_common_newlink
469 * because we can't undo what it does.
470 */
Vlad Yasevich2f6a1b62013-12-11 13:27:11 -0500471 return macvlan_common_newlink(src_net, dev, tb, data);
Arnd Bergmann20d29d72010-01-30 12:24:26 +0000472}
473
474static void macvtap_dellink(struct net_device *dev,
475 struct list_head *head)
476{
Vlad Yasevich6acf54f2013-12-11 13:27:10 -0500477 netdev_rx_handler_unregister(dev);
Arnd Bergmann20d29d72010-01-30 12:24:26 +0000478 macvtap_del_queues(dev);
479 macvlan_dellink(dev, head);
480}
481
Herbert Xu8a357472010-07-21 21:44:31 +0000482static void macvtap_setup(struct net_device *dev)
483{
484 macvlan_common_setup(dev);
485 dev->tx_queue_len = TUN_READQ_SIZE;
486}
487
Arnd Bergmann20d29d72010-01-30 12:24:26 +0000488static struct rtnl_link_ops macvtap_link_ops __read_mostly = {
489 .kind = "macvtap",
Herbert Xu8a357472010-07-21 21:44:31 +0000490 .setup = macvtap_setup,
Arnd Bergmann20d29d72010-01-30 12:24:26 +0000491 .newlink = macvtap_newlink,
492 .dellink = macvtap_dellink,
493};
494
495
496static void macvtap_sock_write_space(struct sock *sk)
497{
Eric Dumazet43815482010-04-29 11:01:49 +0000498 wait_queue_head_t *wqueue;
499
Arnd Bergmann20d29d72010-01-30 12:24:26 +0000500 if (!sock_writeable(sk) ||
501 !test_and_clear_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags))
502 return;
503
Eric Dumazet43815482010-04-29 11:01:49 +0000504 wqueue = sk_sleep(sk);
505 if (wqueue && waitqueue_active(wqueue))
506 wake_up_interruptible_poll(wqueue, POLLOUT | POLLWRNORM | POLLWRBAND);
Arnd Bergmann20d29d72010-01-30 12:24:26 +0000507}
508
Eric W. Biederman2259fef2011-10-20 04:27:24 +0000509static void macvtap_sock_destruct(struct sock *sk)
510{
511 skb_queue_purge(&sk->sk_receive_queue);
512}
513
Arnd Bergmann20d29d72010-01-30 12:24:26 +0000514static int macvtap_open(struct inode *inode, struct file *file)
515{
516 struct net *net = current->nsproxy->net_ns;
Vlad Yasevich40b8fe42014-09-22 16:34:17 -0400517 struct net_device *dev;
Arnd Bergmann20d29d72010-01-30 12:24:26 +0000518 struct macvtap_queue *q;
Vlad Yasevich40b8fe42014-09-22 16:34:17 -0400519 int err = -ENODEV;
Arnd Bergmann20d29d72010-01-30 12:24:26 +0000520
Vlad Yasevich40b8fe42014-09-22 16:34:17 -0400521 rtnl_lock();
522 dev = dev_get_by_macvtap_minor(iminor(inode));
Arnd Bergmann20d29d72010-01-30 12:24:26 +0000523 if (!dev)
524 goto out;
525
Arnd Bergmann20d29d72010-01-30 12:24:26 +0000526 err = -ENOMEM;
527 q = (struct macvtap_queue *)sk_alloc(net, AF_UNSPEC, GFP_KERNEL,
Eric W. Biederman11aa9c22015-05-08 21:09:13 -0500528 &macvtap_proto, 0);
Arnd Bergmann20d29d72010-01-30 12:24:26 +0000529 if (!q)
530 goto out;
531
Jason Wangd9a90a32013-06-13 14:23:35 +0800532 RCU_INIT_POINTER(q->sock.wq, &q->wq);
Eric Dumazet43815482010-04-29 11:01:49 +0000533 init_waitqueue_head(&q->wq.wait);
Arnd Bergmann20d29d72010-01-30 12:24:26 +0000534 q->sock.type = SOCK_RAW;
535 q->sock.state = SS_CONNECTED;
Arnd Bergmann501c7742010-02-18 05:46:50 +0000536 q->sock.file = file;
537 q->sock.ops = &macvtap_socket_ops;
Arnd Bergmann20d29d72010-01-30 12:24:26 +0000538 sock_init_data(&q->sock, &q->sk);
Arnd Bergmann20d29d72010-01-30 12:24:26 +0000539 q->sk.sk_write_space = macvtap_sock_write_space;
Eric W. Biederman2259fef2011-10-20 04:27:24 +0000540 q->sk.sk_destruct = macvtap_sock_destruct;
Arnd Bergmannb9fb9ee2010-02-18 05:48:17 +0000541 q->flags = IFF_VNET_HDR | IFF_NO_PI | IFF_TAP;
Michael S. Tsirkin55afbd02010-04-29 13:50:48 +0300542 q->vnet_hdr_sz = sizeof(struct virtio_net_hdr);
Arnd Bergmann20d29d72010-01-30 12:24:26 +0000543
Shirley Ma97bc3632011-07-06 12:26:11 +0000544 /*
545 * so far only KVM virtio_net uses macvtap, enable zero copy between
546 * guest kernel and host kernel when lower device supports zerocopy
Eric W. Biederman047af9cf2011-10-20 04:26:39 +0000547 *
548 * The macvlan supports zerocopy iff the lower device supports zero
549 * copy so we don't have to look at the lower device directly.
Shirley Ma97bc3632011-07-06 12:26:11 +0000550 */
Eric W. Biederman047af9cf2011-10-20 04:26:39 +0000551 if ((dev->features & NETIF_F_HIGHDMA) && (dev->features & NETIF_F_SG))
552 sock_set_flag(&q->sk, SOCK_ZEROCOPY);
Shirley Ma97bc3632011-07-06 12:26:11 +0000553
Arnd Bergmann20d29d72010-01-30 12:24:26 +0000554 err = macvtap_set_queue(dev, file, q);
555 if (err)
556 sock_put(&q->sk);
557
558out:
559 if (dev)
560 dev_put(dev);
561
Vlad Yasevich40b8fe42014-09-22 16:34:17 -0400562 rtnl_unlock();
Arnd Bergmann20d29d72010-01-30 12:24:26 +0000563 return err;
564}
565
566static int macvtap_release(struct inode *inode, struct file *file)
567{
Arnd Bergmann02df55d2010-02-18 05:45:36 +0000568 struct macvtap_queue *q = file->private_data;
569 macvtap_put_queue(q);
Arnd Bergmann20d29d72010-01-30 12:24:26 +0000570 return 0;
571}
572
573static unsigned int macvtap_poll(struct file *file, poll_table * wait)
574{
Arnd Bergmann02df55d2010-02-18 05:45:36 +0000575 struct macvtap_queue *q = file->private_data;
Arnd Bergmann20d29d72010-01-30 12:24:26 +0000576 unsigned int mask = POLLERR;
577
578 if (!q)
579 goto out;
580
581 mask = 0;
Eric Dumazet43815482010-04-29 11:01:49 +0000582 poll_wait(file, &q->wq.wait, wait);
Arnd Bergmann20d29d72010-01-30 12:24:26 +0000583
584 if (!skb_queue_empty(&q->sk.sk_receive_queue))
585 mask |= POLLIN | POLLRDNORM;
586
587 if (sock_writeable(&q->sk) ||
588 (!test_and_set_bit(SOCK_ASYNC_NOSPACE, &q->sock.flags) &&
589 sock_writeable(&q->sk)))
590 mask |= POLLOUT | POLLWRNORM;
591
592out:
Arnd Bergmann20d29d72010-01-30 12:24:26 +0000593 return mask;
594}
595
Arnd Bergmannb9fb9ee2010-02-18 05:48:17 +0000596static inline struct sk_buff *macvtap_alloc_skb(struct sock *sk, size_t prepad,
597 size_t len, size_t linear,
598 int noblock, int *err)
599{
600 struct sk_buff *skb;
601
602 /* Under a page? Don't bother with paged skb. */
603 if (prepad + len < PAGE_SIZE || !linear)
604 linear = len;
605
606 skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock,
Eric Dumazet28d64272013-08-08 14:38:47 -0700607 err, 0);
Arnd Bergmannb9fb9ee2010-02-18 05:48:17 +0000608 if (!skb)
609 return NULL;
610
611 skb_reserve(skb, prepad);
612 skb_put(skb, linear);
613 skb->data_len = len - linear;
614 skb->len += len - linear;
615
616 return skb;
617}
618
619/*
620 * macvtap_skb_from_vnet_hdr and macvtap_skb_to_vnet_hdr should
621 * be shared with the tun/tap driver.
622 */
Michael S. Tsirkin6ae7feb2014-11-23 17:24:15 +0200623static int macvtap_skb_from_vnet_hdr(struct macvtap_queue *q,
624 struct sk_buff *skb,
Arnd Bergmannb9fb9ee2010-02-18 05:48:17 +0000625 struct virtio_net_hdr *vnet_hdr)
626{
627 unsigned short gso_type = 0;
628 if (vnet_hdr->gso_type != VIRTIO_NET_HDR_GSO_NONE) {
629 switch (vnet_hdr->gso_type & ~VIRTIO_NET_HDR_GSO_ECN) {
630 case VIRTIO_NET_HDR_GSO_TCPV4:
631 gso_type = SKB_GSO_TCPV4;
632 break;
633 case VIRTIO_NET_HDR_GSO_TCPV6:
634 gso_type = SKB_GSO_TCPV6;
635 break;
636 case VIRTIO_NET_HDR_GSO_UDP:
637 gso_type = SKB_GSO_UDP;
638 break;
639 default:
640 return -EINVAL;
641 }
642
643 if (vnet_hdr->gso_type & VIRTIO_NET_HDR_GSO_ECN)
644 gso_type |= SKB_GSO_TCP_ECN;
645
646 if (vnet_hdr->gso_size == 0)
647 return -EINVAL;
648 }
649
650 if (vnet_hdr->flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) {
Michael S. Tsirkin6ae7feb2014-11-23 17:24:15 +0200651 if (!skb_partial_csum_set(skb, macvtap16_to_cpu(q, vnet_hdr->csum_start),
652 macvtap16_to_cpu(q, vnet_hdr->csum_offset)))
Arnd Bergmannb9fb9ee2010-02-18 05:48:17 +0000653 return -EINVAL;
654 }
655
656 if (vnet_hdr->gso_type != VIRTIO_NET_HDR_GSO_NONE) {
Michael S. Tsirkin6ae7feb2014-11-23 17:24:15 +0200657 skb_shinfo(skb)->gso_size = macvtap16_to_cpu(q, vnet_hdr->gso_size);
Pravin B Shelarc9af6db2013-02-11 09:27:41 +0000658 skb_shinfo(skb)->gso_type = gso_type;
Arnd Bergmannb9fb9ee2010-02-18 05:48:17 +0000659
660 /* Header must be checked, and gso_segs computed. */
661 skb_shinfo(skb)->gso_type |= SKB_GSO_DODGY;
662 skb_shinfo(skb)->gso_segs = 0;
663 }
664 return 0;
665}
666
Michael S. Tsirkin6ae7feb2014-11-23 17:24:15 +0200667static void macvtap_skb_to_vnet_hdr(struct macvtap_queue *q,
668 const struct sk_buff *skb,
669 struct virtio_net_hdr *vnet_hdr)
Arnd Bergmannb9fb9ee2010-02-18 05:48:17 +0000670{
671 memset(vnet_hdr, 0, sizeof(*vnet_hdr));
672
673 if (skb_is_gso(skb)) {
674 struct skb_shared_info *sinfo = skb_shinfo(skb);
675
676 /* This is a hint as to how much should be linear. */
Michael S. Tsirkin6ae7feb2014-11-23 17:24:15 +0200677 vnet_hdr->hdr_len = cpu_to_macvtap16(q, skb_headlen(skb));
678 vnet_hdr->gso_size = cpu_to_macvtap16(q, sinfo->gso_size);
Arnd Bergmannb9fb9ee2010-02-18 05:48:17 +0000679 if (sinfo->gso_type & SKB_GSO_TCPV4)
680 vnet_hdr->gso_type = VIRTIO_NET_HDR_GSO_TCPV4;
681 else if (sinfo->gso_type & SKB_GSO_TCPV6)
682 vnet_hdr->gso_type = VIRTIO_NET_HDR_GSO_TCPV6;
Vlad Yaseviche3e3c422015-02-03 16:36:17 -0500683 else if (sinfo->gso_type & SKB_GSO_UDP)
684 vnet_hdr->gso_type = VIRTIO_NET_HDR_GSO_UDP;
Arnd Bergmannb9fb9ee2010-02-18 05:48:17 +0000685 else
686 BUG();
687 if (sinfo->gso_type & SKB_GSO_TCP_ECN)
688 vnet_hdr->gso_type |= VIRTIO_NET_HDR_GSO_ECN;
689 } else
690 vnet_hdr->gso_type = VIRTIO_NET_HDR_GSO_NONE;
691
692 if (skb->ip_summed == CHECKSUM_PARTIAL) {
693 vnet_hdr->flags = VIRTIO_NET_HDR_F_NEEDS_CSUM;
Jiri Pirkodf8a39d2015-01-13 17:13:44 +0100694 if (skb_vlan_tag_present(skb))
Michael S. Tsirkin6ae7feb2014-11-23 17:24:15 +0200695 vnet_hdr->csum_start = cpu_to_macvtap16(q,
696 skb_checksum_start_offset(skb) + VLAN_HLEN);
697 else
698 vnet_hdr->csum_start = cpu_to_macvtap16(q,
699 skb_checksum_start_offset(skb));
700 vnet_hdr->csum_offset = cpu_to_macvtap16(q, skb->csum_offset);
Jason Wang10a8d942011-06-10 00:56:17 +0000701 } else if (skb->ip_summed == CHECKSUM_UNNECESSARY) {
702 vnet_hdr->flags = VIRTIO_NET_HDR_F_DATA_VALID;
Arnd Bergmannb9fb9ee2010-02-18 05:48:17 +0000703 } /* else everything is zero */
Arnd Bergmannb9fb9ee2010-02-18 05:48:17 +0000704}
705
Eric Dumazet2f1d8b92015-02-27 18:35:35 -0800706/* Neighbour code has some assumptions on HH_DATA_MOD alignment */
707#define MACVTAP_RESERVE HH_DATA_OFF(ETH_HLEN)
708
Arnd Bergmann20d29d72010-01-30 12:24:26 +0000709/* Get packet from user space buffer */
Shirley Ma97bc3632011-07-06 12:26:11 +0000710static ssize_t macvtap_get_user(struct macvtap_queue *q, struct msghdr *m,
Al Virof5ff53b2014-06-19 15:36:49 -0400711 struct iov_iter *from, int noblock)
Arnd Bergmann20d29d72010-01-30 12:24:26 +0000712{
Eric Dumazet2f1d8b92015-02-27 18:35:35 -0800713 int good_linear = SKB_MAX_HEAD(MACVTAP_RESERVE);
Arnd Bergmann20d29d72010-01-30 12:24:26 +0000714 struct sk_buff *skb;
Arnd Bergmann02df55d2010-02-18 05:45:36 +0000715 struct macvlan_dev *vlan;
Al Virof5ff53b2014-06-19 15:36:49 -0400716 unsigned long total_len = iov_iter_count(from);
Shirley Ma97bc3632011-07-06 12:26:11 +0000717 unsigned long len = total_len;
Arnd Bergmann20d29d72010-01-30 12:24:26 +0000718 int err;
Arnd Bergmannb9fb9ee2010-02-18 05:48:17 +0000719 struct virtio_net_hdr vnet_hdr = { 0 };
720 int vnet_hdr_len = 0;
Jason Wangb92946e2012-05-02 11:42:15 +0800721 int copylen = 0;
Ivan Vecerac5c62f12015-07-23 16:37:43 +0200722 int depth;
Shirley Ma97bc3632011-07-06 12:26:11 +0000723 bool zerocopy = false;
Jason Wang61d46bf2013-07-10 13:43:28 +0800724 size_t linear;
Al Virof5ff53b2014-06-19 15:36:49 -0400725 ssize_t n;
Arnd Bergmann20d29d72010-01-30 12:24:26 +0000726
Arnd Bergmannb9fb9ee2010-02-18 05:48:17 +0000727 if (q->flags & IFF_VNET_HDR) {
Michael S. Tsirkin55afbd02010-04-29 13:50:48 +0300728 vnet_hdr_len = q->vnet_hdr_sz;
Arnd Bergmannb9fb9ee2010-02-18 05:48:17 +0000729
730 err = -EINVAL;
Nicolas Kaiserce3c8692011-03-04 13:49:41 +0000731 if (len < vnet_hdr_len)
Arnd Bergmannb9fb9ee2010-02-18 05:48:17 +0000732 goto err;
Nicolas Kaiserce3c8692011-03-04 13:49:41 +0000733 len -= vnet_hdr_len;
Arnd Bergmannb9fb9ee2010-02-18 05:48:17 +0000734
Al Virof5ff53b2014-06-19 15:36:49 -0400735 err = -EFAULT;
736 n = copy_from_iter(&vnet_hdr, sizeof(vnet_hdr), from);
737 if (n != sizeof(vnet_hdr))
Arnd Bergmannb9fb9ee2010-02-18 05:48:17 +0000738 goto err;
Al Virof5ff53b2014-06-19 15:36:49 -0400739 iov_iter_advance(from, vnet_hdr_len - sizeof(vnet_hdr));
Arnd Bergmannb9fb9ee2010-02-18 05:48:17 +0000740 if ((vnet_hdr.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) &&
Michael S. Tsirkin6ae7feb2014-11-23 17:24:15 +0200741 macvtap16_to_cpu(q, vnet_hdr.csum_start) +
742 macvtap16_to_cpu(q, vnet_hdr.csum_offset) + 2 >
743 macvtap16_to_cpu(q, vnet_hdr.hdr_len))
744 vnet_hdr.hdr_len = cpu_to_macvtap16(q,
745 macvtap16_to_cpu(q, vnet_hdr.csum_start) +
746 macvtap16_to_cpu(q, vnet_hdr.csum_offset) + 2);
Arnd Bergmannb9fb9ee2010-02-18 05:48:17 +0000747 err = -EINVAL;
Michael S. Tsirkin6ae7feb2014-11-23 17:24:15 +0200748 if (macvtap16_to_cpu(q, vnet_hdr.hdr_len) > len)
Arnd Bergmannb9fb9ee2010-02-18 05:48:17 +0000749 goto err;
750 }
751
752 err = -EINVAL;
Arnd Bergmann20d29d72010-01-30 12:24:26 +0000753 if (unlikely(len < ETH_HLEN))
Arnd Bergmannb9fb9ee2010-02-18 05:48:17 +0000754 goto err;
Arnd Bergmann20d29d72010-01-30 12:24:26 +0000755
Jason Wangece793f2013-07-18 10:55:16 +0800756 if (m && m->msg_control && sock_flag(&q->sk, SOCK_ZEROCOPY)) {
Al Virof5ff53b2014-06-19 15:36:49 -0400757 struct iov_iter i;
758
Michael S. Tsirkin6ae7feb2014-11-23 17:24:15 +0200759 copylen = vnet_hdr.hdr_len ?
760 macvtap16_to_cpu(q, vnet_hdr.hdr_len) : GOODCOPY_LEN;
Jason Wang16a3fa22013-11-13 14:00:40 +0800761 if (copylen > good_linear)
762 copylen = good_linear;
Jason Wang61d46bf2013-07-10 13:43:28 +0800763 linear = copylen;
Al Virof5ff53b2014-06-19 15:36:49 -0400764 i = *from;
765 iov_iter_advance(&i, copylen);
766 if (iov_iter_npages(&i, INT_MAX) <= MAX_SKB_FRAGS)
Jason Wangece793f2013-07-18 10:55:16 +0800767 zerocopy = true;
768 }
769
770 if (!zerocopy) {
Shirley Ma97bc3632011-07-06 12:26:11 +0000771 copylen = len;
Michael S. Tsirkin6ae7feb2014-11-23 17:24:15 +0200772 if (macvtap16_to_cpu(q, vnet_hdr.hdr_len) > good_linear)
Jason Wang16a3fa22013-11-13 14:00:40 +0800773 linear = good_linear;
774 else
Michael S. Tsirkin6ae7feb2014-11-23 17:24:15 +0200775 linear = macvtap16_to_cpu(q, vnet_hdr.hdr_len);
Jason Wang61d46bf2013-07-10 13:43:28 +0800776 }
Shirley Ma97bc3632011-07-06 12:26:11 +0000777
Eric Dumazet2f1d8b92015-02-27 18:35:35 -0800778 skb = macvtap_alloc_skb(&q->sk, MACVTAP_RESERVE, copylen,
Jason Wang61d46bf2013-07-10 13:43:28 +0800779 linear, noblock, &err);
Arnd Bergmann02df55d2010-02-18 05:45:36 +0000780 if (!skb)
781 goto err;
Arnd Bergmann20d29d72010-01-30 12:24:26 +0000782
Jason Wang01d66572012-05-02 11:42:06 +0800783 if (zerocopy)
Al Virof5ff53b2014-06-19 15:36:49 -0400784 err = zerocopy_sg_from_iter(skb, from);
Jason Wangece793f2013-07-18 10:55:16 +0800785 else {
Al Virof5ff53b2014-06-19 15:36:49 -0400786 err = skb_copy_datagram_from_iter(skb, 0, from, len);
Jason Wangece793f2013-07-18 10:55:16 +0800787 if (!err && m && m->msg_control) {
788 struct ubuf_info *uarg = m->msg_control;
789 uarg->callback(uarg, false);
790 }
791 }
792
Arnd Bergmann02df55d2010-02-18 05:45:36 +0000793 if (err)
Arnd Bergmannb9fb9ee2010-02-18 05:48:17 +0000794 goto err_kfree;
Arnd Bergmann20d29d72010-01-30 12:24:26 +0000795
796 skb_set_network_header(skb, ETH_HLEN);
Arnd Bergmannb9fb9ee2010-02-18 05:48:17 +0000797 skb_reset_mac_header(skb);
798 skb->protocol = eth_hdr(skb)->h_proto;
799
800 if (vnet_hdr_len) {
Michael S. Tsirkin6ae7feb2014-11-23 17:24:15 +0200801 err = macvtap_skb_from_vnet_hdr(q, skb, &vnet_hdr);
Arnd Bergmannb9fb9ee2010-02-18 05:48:17 +0000802 if (err)
803 goto err_kfree;
804 }
805
Jason Wang40893fd2013-03-26 23:11:22 +0000806 skb_probe_transport_header(skb, ETH_HLEN);
Jason Wang9b4d6692013-03-25 20:19:55 +0000807
Ivan Vecerac5c62f12015-07-23 16:37:43 +0200808 /* Move network header to the right position for VLAN tagged packets */
809 if ((skb->protocol == htons(ETH_P_8021Q) ||
810 skb->protocol == htons(ETH_P_8021AD)) &&
811 __vlan_get_protocol(skb, skb->protocol, &depth) != 0)
812 skb_set_network_header(skb, depth);
813
Vlad Yasevichac4e4af2013-06-25 16:04:20 -0400814 rcu_read_lock();
815 vlan = rcu_dereference(q->vlan);
Shirley Ma97bc3632011-07-06 12:26:11 +0000816 /* copy skb_ubuf_info for callback when skb has no error */
Jason Wang01d66572012-05-02 11:42:06 +0800817 if (zerocopy) {
Shirley Ma97bc3632011-07-06 12:26:11 +0000818 skb_shinfo(skb)->destructor_arg = m->msg_control;
Jason Wang01d66572012-05-02 11:42:06 +0800819 skb_shinfo(skb)->tx_flags |= SKBTX_DEV_ZEROCOPY;
Pravin B Shelarc9af6db2013-02-11 09:27:41 +0000820 skb_shinfo(skb)->tx_flags |= SKBTX_SHARED_FRAG;
Jason Wang01d66572012-05-02 11:42:06 +0800821 }
Eric Dumazet29d79192013-08-08 08:06:14 -0700822 if (vlan) {
Vlad Yasevich6acf54f2013-12-11 13:27:10 -0500823 skb->dev = vlan->dev;
824 dev_queue_xmit(skb);
Eric Dumazet29d79192013-08-08 08:06:14 -0700825 } else {
Arnd Bergmann02df55d2010-02-18 05:45:36 +0000826 kfree_skb(skb);
Eric Dumazet29d79192013-08-08 08:06:14 -0700827 }
Vlad Yasevichac4e4af2013-06-25 16:04:20 -0400828 rcu_read_unlock();
Arnd Bergmann20d29d72010-01-30 12:24:26 +0000829
Shirley Ma97bc3632011-07-06 12:26:11 +0000830 return total_len;
Arnd Bergmann02df55d2010-02-18 05:45:36 +0000831
Arnd Bergmannb9fb9ee2010-02-18 05:48:17 +0000832err_kfree:
833 kfree_skb(skb);
834
Arnd Bergmann02df55d2010-02-18 05:45:36 +0000835err:
Vlad Yasevichac4e4af2013-06-25 16:04:20 -0400836 rcu_read_lock();
837 vlan = rcu_dereference(q->vlan);
Arnd Bergmann02df55d2010-02-18 05:45:36 +0000838 if (vlan)
Jason Wangcd3e22b2013-11-25 17:19:04 +0800839 this_cpu_inc(vlan->pcpu_stats->tx_dropped);
Vlad Yasevichac4e4af2013-06-25 16:04:20 -0400840 rcu_read_unlock();
Arnd Bergmann02df55d2010-02-18 05:45:36 +0000841
Arnd Bergmann02df55d2010-02-18 05:45:36 +0000842 return err;
Arnd Bergmann20d29d72010-01-30 12:24:26 +0000843}
844
Al Virof5ff53b2014-06-19 15:36:49 -0400845static ssize_t macvtap_write_iter(struct kiocb *iocb, struct iov_iter *from)
Arnd Bergmann20d29d72010-01-30 12:24:26 +0000846{
847 struct file *file = iocb->ki_filp;
Arnd Bergmann02df55d2010-02-18 05:45:36 +0000848 struct macvtap_queue *q = file->private_data;
Arnd Bergmann20d29d72010-01-30 12:24:26 +0000849
Al Virof5ff53b2014-06-19 15:36:49 -0400850 return macvtap_get_user(q, NULL, from, file->f_flags & O_NONBLOCK);
Arnd Bergmann20d29d72010-01-30 12:24:26 +0000851}
852
853/* Put packet to the user space buffer */
854static ssize_t macvtap_put_user(struct macvtap_queue *q,
855 const struct sk_buff *skb,
Herbert Xu6c36d2e2014-11-07 21:22:25 +0800856 struct iov_iter *iter)
Arnd Bergmann20d29d72010-01-30 12:24:26 +0000857{
Arnd Bergmann20d29d72010-01-30 12:24:26 +0000858 int ret;
Arnd Bergmannb9fb9ee2010-02-18 05:48:17 +0000859 int vnet_hdr_len = 0;
Basil Gorf09e2242012-05-03 22:55:24 +0000860 int vlan_offset = 0;
Herbert Xu6c36d2e2014-11-07 21:22:25 +0800861 int total;
Arnd Bergmannb9fb9ee2010-02-18 05:48:17 +0000862
863 if (q->flags & IFF_VNET_HDR) {
864 struct virtio_net_hdr vnet_hdr;
Michael S. Tsirkin55afbd02010-04-29 13:50:48 +0300865 vnet_hdr_len = q->vnet_hdr_sz;
Herbert Xu6c36d2e2014-11-07 21:22:25 +0800866 if (iov_iter_count(iter) < vnet_hdr_len)
Arnd Bergmannb9fb9ee2010-02-18 05:48:17 +0000867 return -EINVAL;
868
Michael S. Tsirkin6ae7feb2014-11-23 17:24:15 +0200869 macvtap_skb_to_vnet_hdr(q, skb, &vnet_hdr);
Arnd Bergmannb9fb9ee2010-02-18 05:48:17 +0000870
Herbert Xu6c36d2e2014-11-07 21:22:25 +0800871 if (copy_to_iter(&vnet_hdr, sizeof(vnet_hdr), iter) !=
872 sizeof(vnet_hdr))
Arnd Bergmannb9fb9ee2010-02-18 05:48:17 +0000873 return -EFAULT;
Jason Wang7cc76f52014-11-20 16:31:05 +0800874
875 iov_iter_advance(iter, vnet_hdr_len - sizeof(vnet_hdr));
Arnd Bergmannb9fb9ee2010-02-18 05:48:17 +0000876 }
Herbert Xu6c36d2e2014-11-07 21:22:25 +0800877 total = vnet_hdr_len;
Jason Wangce232ce2013-12-11 13:08:34 +0800878 total += skb->len;
Arnd Bergmann20d29d72010-01-30 12:24:26 +0000879
Jiri Pirkodf8a39d2015-01-13 17:13:44 +0100880 if (skb_vlan_tag_present(skb)) {
Basil Gorf09e2242012-05-03 22:55:24 +0000881 struct {
882 __be16 h_vlan_proto;
883 __be16 h_vlan_TCI;
884 } veth;
Jason Wang0fbe0d42013-07-16 13:36:34 +0800885 veth.h_vlan_proto = skb->vlan_proto;
Jiri Pirkodf8a39d2015-01-13 17:13:44 +0100886 veth.h_vlan_TCI = htons(skb_vlan_tag_get(skb));
Arnd Bergmann20d29d72010-01-30 12:24:26 +0000887
Basil Gorf09e2242012-05-03 22:55:24 +0000888 vlan_offset = offsetof(struct vlan_ethhdr, h_vlan_proto);
Jason Wangce232ce2013-12-11 13:08:34 +0800889 total += VLAN_HLEN;
Arnd Bergmann20d29d72010-01-30 12:24:26 +0000890
Herbert Xu6c36d2e2014-11-07 21:22:25 +0800891 ret = skb_copy_datagram_iter(skb, 0, iter, vlan_offset);
892 if (ret || !iov_iter_count(iter))
Basil Gorf09e2242012-05-03 22:55:24 +0000893 goto done;
894
Herbert Xu6c36d2e2014-11-07 21:22:25 +0800895 ret = copy_to_iter(&veth, sizeof(veth), iter);
896 if (ret != sizeof(veth) || !iov_iter_count(iter))
Basil Gorf09e2242012-05-03 22:55:24 +0000897 goto done;
898 }
899
Herbert Xu6c36d2e2014-11-07 21:22:25 +0800900 ret = skb_copy_datagram_iter(skb, vlan_offset, iter,
901 skb->len - vlan_offset);
Basil Gorf09e2242012-05-03 22:55:24 +0000902
903done:
Jason Wangce232ce2013-12-11 13:08:34 +0800904 return ret ? ret : total;
Arnd Bergmann20d29d72010-01-30 12:24:26 +0000905}
906
Zhi Yong Wu55ec8e22013-12-07 04:13:05 +0800907static ssize_t macvtap_do_read(struct macvtap_queue *q,
Al Viro3af0bfe2014-11-07 14:13:53 -0500908 struct iov_iter *to,
Arnd Bergmann501c7742010-02-18 05:46:50 +0000909 int noblock)
Arnd Bergmann20d29d72010-01-30 12:24:26 +0000910{
Hong zhi guoccf7e722012-06-06 22:36:27 +0000911 DEFINE_WAIT(wait);
Arnd Bergmann20d29d72010-01-30 12:24:26 +0000912 struct sk_buff *skb;
Arnd Bergmann501c7742010-02-18 05:46:50 +0000913 ssize_t ret = 0;
Arnd Bergmann20d29d72010-01-30 12:24:26 +0000914
Al Viro3af0bfe2014-11-07 14:13:53 -0500915 if (!iov_iter_count(to))
916 return 0;
917
918 while (1) {
Jason Wang89cee912013-06-05 23:54:34 +0000919 if (!noblock)
920 prepare_to_wait(sk_sleep(&q->sk), &wait,
921 TASK_INTERRUPTIBLE);
Arnd Bergmann20d29d72010-01-30 12:24:26 +0000922
923 /* Read frames from the queue */
924 skb = skb_dequeue(&q->sk.sk_receive_queue);
Al Viro3af0bfe2014-11-07 14:13:53 -0500925 if (skb)
926 break;
927 if (noblock) {
928 ret = -EAGAIN;
929 break;
Arnd Bergmann20d29d72010-01-30 12:24:26 +0000930 }
Al Viro3af0bfe2014-11-07 14:13:53 -0500931 if (signal_pending(current)) {
932 ret = -ERESTARTSYS;
933 break;
934 }
935 /* Nothing to read, let's sleep */
936 schedule();
Arnd Bergmann20d29d72010-01-30 12:24:26 +0000937 }
Al Viro3af0bfe2014-11-07 14:13:53 -0500938 if (skb) {
939 ret = macvtap_put_user(q, skb, to);
Jason Wangf51a5e82014-12-01 16:53:15 +0800940 if (unlikely(ret < 0))
941 kfree_skb(skb);
942 else
943 consume_skb(skb);
Al Viro3af0bfe2014-11-07 14:13:53 -0500944 }
Jason Wang89cee912013-06-05 23:54:34 +0000945 if (!noblock)
946 finish_wait(sk_sleep(&q->sk), &wait);
Arnd Bergmann501c7742010-02-18 05:46:50 +0000947 return ret;
948}
Arnd Bergmann20d29d72010-01-30 12:24:26 +0000949
Al Viro3af0bfe2014-11-07 14:13:53 -0500950static ssize_t macvtap_read_iter(struct kiocb *iocb, struct iov_iter *to)
Arnd Bergmann501c7742010-02-18 05:46:50 +0000951{
952 struct file *file = iocb->ki_filp;
953 struct macvtap_queue *q = file->private_data;
Al Viro3af0bfe2014-11-07 14:13:53 -0500954 ssize_t len = iov_iter_count(to), ret;
Arnd Bergmann501c7742010-02-18 05:46:50 +0000955
Al Viro3af0bfe2014-11-07 14:13:53 -0500956 ret = macvtap_do_read(q, to, file->f_flags & O_NONBLOCK);
Jason Wangce232ce2013-12-11 13:08:34 +0800957 ret = min_t(ssize_t, ret, len);
Zhi Yong Wue6ebc7f2013-12-06 14:16:50 +0800958 if (ret > 0)
959 iocb->ki_pos = ret;
Arnd Bergmann20d29d72010-01-30 12:24:26 +0000960 return ret;
961}
962
Jason Wang8f475a32013-06-05 23:54:36 +0000963static struct macvlan_dev *macvtap_get_vlan(struct macvtap_queue *q)
964{
965 struct macvlan_dev *vlan;
966
Vlad Yasevich441ac0f2013-06-25 16:04:19 -0400967 ASSERT_RTNL();
968 vlan = rtnl_dereference(q->vlan);
Jason Wang8f475a32013-06-05 23:54:36 +0000969 if (vlan)
970 dev_hold(vlan->dev);
Jason Wang8f475a32013-06-05 23:54:36 +0000971
972 return vlan;
973}
974
975static void macvtap_put_vlan(struct macvlan_dev *vlan)
976{
977 dev_put(vlan->dev);
978}
979
Jason Wang815f2362013-06-05 23:54:39 +0000980static int macvtap_ioctl_set_queue(struct file *file, unsigned int flags)
981{
982 struct macvtap_queue *q = file->private_data;
983 struct macvlan_dev *vlan;
984 int ret;
985
986 vlan = macvtap_get_vlan(q);
987 if (!vlan)
988 return -EINVAL;
989
990 if (flags & IFF_ATTACH_QUEUE)
991 ret = macvtap_enable_queue(vlan->dev, file, q);
992 else if (flags & IFF_DETACH_QUEUE)
993 ret = macvtap_disable_queue(q);
Jason Wangf57855a2013-06-13 14:23:36 +0800994 else
995 ret = -EINVAL;
Jason Wang815f2362013-06-05 23:54:39 +0000996
997 macvtap_put_vlan(vlan);
998 return ret;
999}
1000
Vlad Yasevich2be5c762013-06-25 16:04:21 -04001001static int set_offload(struct macvtap_queue *q, unsigned long arg)
1002{
1003 struct macvlan_dev *vlan;
1004 netdev_features_t features;
1005 netdev_features_t feature_mask = 0;
1006
1007 vlan = rtnl_dereference(q->vlan);
1008 if (!vlan)
1009 return -ENOLINK;
1010
1011 features = vlan->dev->features;
1012
1013 if (arg & TUN_F_CSUM) {
1014 feature_mask = NETIF_F_HW_CSUM;
1015
1016 if (arg & (TUN_F_TSO4 | TUN_F_TSO6)) {
1017 if (arg & TUN_F_TSO_ECN)
1018 feature_mask |= NETIF_F_TSO_ECN;
1019 if (arg & TUN_F_TSO4)
1020 feature_mask |= NETIF_F_TSO;
1021 if (arg & TUN_F_TSO6)
1022 feature_mask |= NETIF_F_TSO6;
1023 }
Vlad Yaseviche3e3c422015-02-03 16:36:17 -05001024
1025 if (arg & TUN_F_UFO)
1026 feature_mask |= NETIF_F_UFO;
Vlad Yasevich2be5c762013-06-25 16:04:21 -04001027 }
1028
1029 /* tun/tap driver inverts the usage for TSO offloads, where
1030 * setting the TSO bit means that the userspace wants to
1031 * accept TSO frames and turning it off means that user space
1032 * does not support TSO.
1033 * For macvtap, we have to invert it to mean the same thing.
1034 * When user space turns off TSO, we turn off GSO/LRO so that
1035 * user-space will not receive TSO frames.
1036 */
Vlad Yaseviche3e3c422015-02-03 16:36:17 -05001037 if (feature_mask & (NETIF_F_TSO | NETIF_F_TSO6 | NETIF_F_UFO))
Vlad Yasevich2be5c762013-06-25 16:04:21 -04001038 features |= RX_OFFLOADS;
1039 else
1040 features &= ~RX_OFFLOADS;
1041
1042 /* tap_features are the same as features on tun/tap and
1043 * reflect user expectations.
1044 */
Vlad Yasevicha567dd62013-08-16 15:25:00 -04001045 vlan->tap_features = feature_mask;
Vlad Yasevich2be5c762013-06-25 16:04:21 -04001046 vlan->set_features = features;
1047 netdev_update_features(vlan->dev);
1048
1049 return 0;
1050}
1051
Arnd Bergmann20d29d72010-01-30 12:24:26 +00001052/*
1053 * provide compatibility with generic tun/tap interface
1054 */
1055static long macvtap_ioctl(struct file *file, unsigned int cmd,
1056 unsigned long arg)
1057{
Arnd Bergmann02df55d2010-02-18 05:45:36 +00001058 struct macvtap_queue *q = file->private_data;
1059 struct macvlan_dev *vlan;
Arnd Bergmann20d29d72010-01-30 12:24:26 +00001060 void __user *argp = (void __user *)arg;
1061 struct ifreq __user *ifr = argp;
1062 unsigned int __user *up = argp;
Michael S. Tsirkin39ec7de2014-12-16 15:04:56 +02001063 unsigned short u;
Michael S. Tsirkin55afbd02010-04-29 13:50:48 +03001064 int __user *sp = argp;
Justin Cormack7f460d32015-05-13 19:19:02 +01001065 struct sockaddr sa;
Michael S. Tsirkin55afbd02010-04-29 13:50:48 +03001066 int s;
Arnd Bergmann02df55d2010-02-18 05:45:36 +00001067 int ret;
Arnd Bergmann20d29d72010-01-30 12:24:26 +00001068
1069 switch (cmd) {
1070 case TUNSETIFF:
1071 /* ignore the name, just look at flags */
1072 if (get_user(u, &ifr->ifr_flags))
1073 return -EFAULT;
Arnd Bergmannb9fb9ee2010-02-18 05:48:17 +00001074
1075 ret = 0;
Michael S. Tsirkin6ae7feb2014-11-23 17:24:15 +02001076 if ((u & ~MACVTAP_FEATURES) != (IFF_NO_PI | IFF_TAP))
Arnd Bergmannb9fb9ee2010-02-18 05:48:17 +00001077 ret = -EINVAL;
1078 else
Michael S. Tsirkin39ec7de2014-12-16 15:04:56 +02001079 q->flags = (q->flags & ~MACVTAP_FEATURES) | u;
Arnd Bergmannb9fb9ee2010-02-18 05:48:17 +00001080
1081 return ret;
Arnd Bergmann20d29d72010-01-30 12:24:26 +00001082
1083 case TUNGETIFF:
Vlad Yasevich441ac0f2013-06-25 16:04:19 -04001084 rtnl_lock();
Jason Wang8f475a32013-06-05 23:54:36 +00001085 vlan = macvtap_get_vlan(q);
Vlad Yasevich441ac0f2013-06-25 16:04:19 -04001086 if (!vlan) {
1087 rtnl_unlock();
Arnd Bergmann02df55d2010-02-18 05:45:36 +00001088 return -ENOLINK;
Vlad Yasevich441ac0f2013-06-25 16:04:19 -04001089 }
Arnd Bergmann02df55d2010-02-18 05:45:36 +00001090
1091 ret = 0;
Michael S. Tsirkin39ec7de2014-12-16 15:04:56 +02001092 u = q->flags;
Eric Dumazet13707f92011-01-26 19:28:23 +00001093 if (copy_to_user(&ifr->ifr_name, vlan->dev->name, IFNAMSIZ) ||
Michael S. Tsirkin39ec7de2014-12-16 15:04:56 +02001094 put_user(u, &ifr->ifr_flags))
Arnd Bergmann02df55d2010-02-18 05:45:36 +00001095 ret = -EFAULT;
Jason Wang8f475a32013-06-05 23:54:36 +00001096 macvtap_put_vlan(vlan);
Vlad Yasevich441ac0f2013-06-25 16:04:19 -04001097 rtnl_unlock();
Arnd Bergmann02df55d2010-02-18 05:45:36 +00001098 return ret;
Arnd Bergmann20d29d72010-01-30 12:24:26 +00001099
Jason Wang815f2362013-06-05 23:54:39 +00001100 case TUNSETQUEUE:
1101 if (get_user(u, &ifr->ifr_flags))
1102 return -EFAULT;
Vlad Yasevich441ac0f2013-06-25 16:04:19 -04001103 rtnl_lock();
1104 ret = macvtap_ioctl_set_queue(file, u);
1105 rtnl_unlock();
Jason Wang82a19eb2013-07-16 13:36:33 +08001106 return ret;
Jason Wang815f2362013-06-05 23:54:39 +00001107
Arnd Bergmann20d29d72010-01-30 12:24:26 +00001108 case TUNGETFEATURES:
Michael S. Tsirkin6ae7feb2014-11-23 17:24:15 +02001109 if (put_user(IFF_TAP | IFF_NO_PI | MACVTAP_FEATURES, up))
Arnd Bergmann20d29d72010-01-30 12:24:26 +00001110 return -EFAULT;
1111 return 0;
1112
1113 case TUNSETSNDBUF:
1114 if (get_user(u, up))
1115 return -EFAULT;
1116
Arnd Bergmann20d29d72010-01-30 12:24:26 +00001117 q->sk.sk_sndbuf = u;
Arnd Bergmann20d29d72010-01-30 12:24:26 +00001118 return 0;
1119
Michael S. Tsirkin55afbd02010-04-29 13:50:48 +03001120 case TUNGETVNETHDRSZ:
1121 s = q->vnet_hdr_sz;
1122 if (put_user(s, sp))
1123 return -EFAULT;
1124 return 0;
1125
1126 case TUNSETVNETHDRSZ:
1127 if (get_user(s, sp))
1128 return -EFAULT;
1129 if (s < (int)sizeof(struct virtio_net_hdr))
1130 return -EINVAL;
1131
1132 q->vnet_hdr_sz = s;
1133 return 0;
1134
Michael S. Tsirkin01b07fb2014-12-16 15:05:10 +02001135 case TUNGETVNETLE:
1136 s = !!(q->flags & MACVTAP_VNET_LE);
1137 if (put_user(s, sp))
1138 return -EFAULT;
1139 return 0;
1140
1141 case TUNSETVNETLE:
1142 if (get_user(s, sp))
1143 return -EFAULT;
1144 if (s)
1145 q->flags |= MACVTAP_VNET_LE;
1146 else
1147 q->flags &= ~MACVTAP_VNET_LE;
1148 return 0;
1149
Greg Kurz8b8e6582015-04-24 14:50:36 +02001150 case TUNGETVNETBE:
1151 return macvtap_get_vnet_be(q, sp);
1152
1153 case TUNSETVNETBE:
1154 return macvtap_set_vnet_be(q, sp);
1155
Arnd Bergmann20d29d72010-01-30 12:24:26 +00001156 case TUNSETOFFLOAD:
1157 /* let the user check for future flags */
1158 if (arg & ~(TUN_F_CSUM | TUN_F_TSO4 | TUN_F_TSO6 |
Vlad Yaseviche3e3c422015-02-03 16:36:17 -05001159 TUN_F_TSO_ECN | TUN_F_UFO))
Arnd Bergmann20d29d72010-01-30 12:24:26 +00001160 return -EINVAL;
1161
Vlad Yasevich2be5c762013-06-25 16:04:21 -04001162 rtnl_lock();
1163 ret = set_offload(q, arg);
1164 rtnl_unlock();
1165 return ret;
Arnd Bergmann20d29d72010-01-30 12:24:26 +00001166
Justin Cormackb5082082015-05-11 20:00:10 +01001167 case SIOCGIFHWADDR:
1168 rtnl_lock();
1169 vlan = macvtap_get_vlan(q);
1170 if (!vlan) {
1171 rtnl_unlock();
1172 return -ENOLINK;
1173 }
1174 ret = 0;
1175 u = vlan->dev->type;
1176 if (copy_to_user(&ifr->ifr_name, vlan->dev->name, IFNAMSIZ) ||
1177 copy_to_user(&ifr->ifr_hwaddr.sa_data, vlan->dev->dev_addr, ETH_ALEN) ||
1178 put_user(u, &ifr->ifr_hwaddr.sa_family))
1179 ret = -EFAULT;
1180 macvtap_put_vlan(vlan);
1181 rtnl_unlock();
1182 return ret;
1183
1184 case SIOCSIFHWADDR:
Justin Cormack7f460d32015-05-13 19:19:02 +01001185 if (copy_from_user(&sa, &ifr->ifr_hwaddr, sizeof(sa)))
1186 return -EFAULT;
Justin Cormackb5082082015-05-11 20:00:10 +01001187 rtnl_lock();
1188 vlan = macvtap_get_vlan(q);
1189 if (!vlan) {
1190 rtnl_unlock();
1191 return -ENOLINK;
1192 }
Justin Cormack7f460d32015-05-13 19:19:02 +01001193 ret = dev_set_mac_address(vlan->dev, &sa);
Justin Cormackb5082082015-05-11 20:00:10 +01001194 macvtap_put_vlan(vlan);
1195 rtnl_unlock();
1196 return ret;
1197
Arnd Bergmann20d29d72010-01-30 12:24:26 +00001198 default:
1199 return -EINVAL;
1200 }
1201}
1202
1203#ifdef CONFIG_COMPAT
1204static long macvtap_compat_ioctl(struct file *file, unsigned int cmd,
1205 unsigned long arg)
1206{
1207 return macvtap_ioctl(file, cmd, (unsigned long)compat_ptr(arg));
1208}
1209#endif
1210
1211static const struct file_operations macvtap_fops = {
1212 .owner = THIS_MODULE,
1213 .open = macvtap_open,
1214 .release = macvtap_release,
Al Viro3af0bfe2014-11-07 14:13:53 -05001215 .read_iter = macvtap_read_iter,
Al Virof5ff53b2014-06-19 15:36:49 -04001216 .write_iter = macvtap_write_iter,
Arnd Bergmann20d29d72010-01-30 12:24:26 +00001217 .poll = macvtap_poll,
1218 .llseek = no_llseek,
1219 .unlocked_ioctl = macvtap_ioctl,
1220#ifdef CONFIG_COMPAT
1221 .compat_ioctl = macvtap_compat_ioctl,
1222#endif
1223};
1224
Ying Xue1b784142015-03-02 15:37:48 +08001225static int macvtap_sendmsg(struct socket *sock, struct msghdr *m,
1226 size_t total_len)
Arnd Bergmann501c7742010-02-18 05:46:50 +00001227{
1228 struct macvtap_queue *q = container_of(sock, struct macvtap_queue, sock);
Al Viroc0371da2014-11-24 10:42:55 -05001229 return macvtap_get_user(q, m, &m->msg_iter, m->msg_flags & MSG_DONTWAIT);
Arnd Bergmann501c7742010-02-18 05:46:50 +00001230}
1231
Ying Xue1b784142015-03-02 15:37:48 +08001232static int macvtap_recvmsg(struct socket *sock, struct msghdr *m,
1233 size_t total_len, int flags)
Arnd Bergmann501c7742010-02-18 05:46:50 +00001234{
1235 struct macvtap_queue *q = container_of(sock, struct macvtap_queue, sock);
1236 int ret;
1237 if (flags & ~(MSG_DONTWAIT|MSG_TRUNC))
1238 return -EINVAL;
Al Viroc0371da2014-11-24 10:42:55 -05001239 ret = macvtap_do_read(q, &m->msg_iter, flags & MSG_DONTWAIT);
David S. Millerde2aa472013-12-10 22:06:18 -05001240 if (ret > total_len) {
1241 m->msg_flags |= MSG_TRUNC;
1242 ret = flags & MSG_TRUNC ? ret : total_len;
1243 }
Arnd Bergmann501c7742010-02-18 05:46:50 +00001244 return ret;
1245}
1246
1247/* Ops structure to mimic raw sockets with tun */
1248static const struct proto_ops macvtap_socket_ops = {
1249 .sendmsg = macvtap_sendmsg,
1250 .recvmsg = macvtap_recvmsg,
1251};
1252
1253/* Get an underlying socket object from tun file. Returns error unless file is
1254 * attached to a device. The returned object works like a packet socket, it
1255 * can be used for sock_sendmsg/sock_recvmsg. The caller is responsible for
1256 * holding a reference to the file for as long as the socket is in use. */
1257struct socket *macvtap_get_socket(struct file *file)
1258{
1259 struct macvtap_queue *q;
1260 if (file->f_op != &macvtap_fops)
1261 return ERR_PTR(-EINVAL);
1262 q = file->private_data;
1263 if (!q)
1264 return ERR_PTR(-EBADFD);
1265 return &q->sock;
1266}
1267EXPORT_SYMBOL_GPL(macvtap_get_socket);
1268
Eric W. Biederman9bf19072011-10-20 04:28:46 +00001269static int macvtap_device_event(struct notifier_block *unused,
1270 unsigned long event, void *ptr)
1271{
Jiri Pirko351638e2013-05-28 01:30:21 +00001272 struct net_device *dev = netdev_notifier_info_to_dev(ptr);
Eric W. Biedermane09eff72011-10-20 04:29:24 +00001273 struct macvlan_dev *vlan;
Eric W. Biederman9bf19072011-10-20 04:28:46 +00001274 struct device *classdev;
1275 dev_t devt;
Eric W. Biedermane09eff72011-10-20 04:29:24 +00001276 int err;
Eric W. Biederman9bf19072011-10-20 04:28:46 +00001277
1278 if (dev->rtnl_link_ops != &macvtap_link_ops)
1279 return NOTIFY_DONE;
1280
Eric W. Biedermane09eff72011-10-20 04:29:24 +00001281 vlan = netdev_priv(dev);
Eric W. Biederman9bf19072011-10-20 04:28:46 +00001282
1283 switch (event) {
1284 case NETDEV_REGISTER:
1285 /* Create the device node here after the network device has
1286 * been registered but before register_netdevice has
1287 * finished running.
1288 */
Eric W. Biedermane09eff72011-10-20 04:29:24 +00001289 err = macvtap_get_minor(vlan);
1290 if (err)
1291 return notifier_from_errno(err);
1292
1293 devt = MKDEV(MAJOR(macvtap_major), vlan->minor);
Eric W. Biederman9bf19072011-10-20 04:28:46 +00001294 classdev = device_create(macvtap_class, &dev->dev, devt,
1295 dev, "tap%d", dev->ifindex);
Eric W. Biedermane09eff72011-10-20 04:29:24 +00001296 if (IS_ERR(classdev)) {
1297 macvtap_free_minor(vlan);
Eric W. Biederman9bf19072011-10-20 04:28:46 +00001298 return notifier_from_errno(PTR_ERR(classdev));
Eric W. Biedermane09eff72011-10-20 04:29:24 +00001299 }
Eric W. Biederman9bf19072011-10-20 04:28:46 +00001300 break;
1301 case NETDEV_UNREGISTER:
Eric W. Biedermane09eff72011-10-20 04:29:24 +00001302 devt = MKDEV(MAJOR(macvtap_major), vlan->minor);
Eric W. Biederman9bf19072011-10-20 04:28:46 +00001303 device_destroy(macvtap_class, devt);
Eric W. Biedermane09eff72011-10-20 04:29:24 +00001304 macvtap_free_minor(vlan);
Eric W. Biederman9bf19072011-10-20 04:28:46 +00001305 break;
1306 }
1307
1308 return NOTIFY_DONE;
1309}
1310
1311static struct notifier_block macvtap_notifier_block __read_mostly = {
1312 .notifier_call = macvtap_device_event,
1313};
1314
Arnd Bergmann20d29d72010-01-30 12:24:26 +00001315static int macvtap_init(void)
1316{
1317 int err;
1318
1319 err = alloc_chrdev_region(&macvtap_major, 0,
1320 MACVTAP_NUM_DEVS, "macvtap");
1321 if (err)
1322 goto out1;
1323
1324 cdev_init(&macvtap_cdev, &macvtap_fops);
1325 err = cdev_add(&macvtap_cdev, macvtap_major, MACVTAP_NUM_DEVS);
1326 if (err)
1327 goto out2;
1328
1329 macvtap_class = class_create(THIS_MODULE, "macvtap");
1330 if (IS_ERR(macvtap_class)) {
1331 err = PTR_ERR(macvtap_class);
1332 goto out3;
1333 }
1334
Eric W. Biederman9bf19072011-10-20 04:28:46 +00001335 err = register_netdevice_notifier(&macvtap_notifier_block);
Arnd Bergmann20d29d72010-01-30 12:24:26 +00001336 if (err)
1337 goto out4;
1338
Eric W. Biederman9bf19072011-10-20 04:28:46 +00001339 err = macvlan_link_register(&macvtap_link_ops);
1340 if (err)
1341 goto out5;
1342
Arnd Bergmann20d29d72010-01-30 12:24:26 +00001343 return 0;
1344
Eric W. Biederman9bf19072011-10-20 04:28:46 +00001345out5:
1346 unregister_netdevice_notifier(&macvtap_notifier_block);
Arnd Bergmann20d29d72010-01-30 12:24:26 +00001347out4:
1348 class_unregister(macvtap_class);
1349out3:
1350 cdev_del(&macvtap_cdev);
1351out2:
1352 unregister_chrdev_region(macvtap_major, MACVTAP_NUM_DEVS);
1353out1:
1354 return err;
1355}
1356module_init(macvtap_init);
1357
1358static void macvtap_exit(void)
1359{
1360 rtnl_link_unregister(&macvtap_link_ops);
Eric W. Biederman9bf19072011-10-20 04:28:46 +00001361 unregister_netdevice_notifier(&macvtap_notifier_block);
Arnd Bergmann20d29d72010-01-30 12:24:26 +00001362 class_unregister(macvtap_class);
1363 cdev_del(&macvtap_cdev);
1364 unregister_chrdev_region(macvtap_major, MACVTAP_NUM_DEVS);
Johannes Thumshirnd5de1982015-07-08 17:16:49 +02001365 idr_destroy(&minor_idr);
Arnd Bergmann20d29d72010-01-30 12:24:26 +00001366}
1367module_exit(macvtap_exit);
1368
1369MODULE_ALIAS_RTNL_LINK("macvtap");
1370MODULE_AUTHOR("Arnd Bergmann <arnd@arndb.de>");
1371MODULE_LICENSE("GPL");