blob: 95394edd1ed528fc9cf94fb39aadeaa28c47fb26 [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)
Jason Wangf23d5382015-10-23 00:57:05 -0400140#define TAP_FEATURES (NETIF_F_GSO | NETIF_F_SG | NETIF_F_FRAGLIST)
Vlad Yasevicha567dd62013-08-16 15:25:00 -0400141
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 &&
Tom Herberta1882222015-12-14 11:19:43 -0800391 !(features & NETIF_F_CSUM_MASK) &&
Vlad Yasevichcbdb0422014-04-29 10:09:50 -0400392 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) ||
Eric Dumazet9cd3e072015-11-29 20:03:10 -0800501 !test_and_clear_bit(SOCKWQ_ASYNC_NOSPACE, &sk->sk_socket->flags))
Arnd Bergmann20d29d72010-01-30 12:24:26 +0000502 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) ||
Eric Dumazet9cd3e072015-11-29 20:03:10 -0800588 (!test_and_set_bit(SOCKWQ_ASYNC_NOSPACE, &q->sock.flags) &&
Arnd Bergmann20d29d72010-01-30 12:24:26 +0000589 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;
Willem de Bruijn8e2ad412016-03-08 15:18:54 -0500763 else if (copylen < ETH_HLEN)
764 copylen = ETH_HLEN;
Jason Wang61d46bf2013-07-10 13:43:28 +0800765 linear = copylen;
Al Virof5ff53b2014-06-19 15:36:49 -0400766 i = *from;
767 iov_iter_advance(&i, copylen);
768 if (iov_iter_npages(&i, INT_MAX) <= MAX_SKB_FRAGS)
Jason Wangece793f2013-07-18 10:55:16 +0800769 zerocopy = true;
770 }
771
772 if (!zerocopy) {
Shirley Ma97bc3632011-07-06 12:26:11 +0000773 copylen = len;
Willem de Bruijn8e2ad412016-03-08 15:18:54 -0500774 linear = macvtap16_to_cpu(q, vnet_hdr.hdr_len);
775 if (linear > good_linear)
Jason Wang16a3fa22013-11-13 14:00:40 +0800776 linear = good_linear;
Willem de Bruijn8e2ad412016-03-08 15:18:54 -0500777 else if (linear < ETH_HLEN)
778 linear = ETH_HLEN;
Jason Wang61d46bf2013-07-10 13:43:28 +0800779 }
Shirley Ma97bc3632011-07-06 12:26:11 +0000780
Eric Dumazet2f1d8b92015-02-27 18:35:35 -0800781 skb = macvtap_alloc_skb(&q->sk, MACVTAP_RESERVE, copylen,
Jason Wang61d46bf2013-07-10 13:43:28 +0800782 linear, noblock, &err);
Arnd Bergmann02df55d2010-02-18 05:45:36 +0000783 if (!skb)
784 goto err;
Arnd Bergmann20d29d72010-01-30 12:24:26 +0000785
Jason Wang01d66572012-05-02 11:42:06 +0800786 if (zerocopy)
Al Virof5ff53b2014-06-19 15:36:49 -0400787 err = zerocopy_sg_from_iter(skb, from);
Jason Wangece793f2013-07-18 10:55:16 +0800788 else {
Al Virof5ff53b2014-06-19 15:36:49 -0400789 err = skb_copy_datagram_from_iter(skb, 0, from, len);
Jason Wangece793f2013-07-18 10:55:16 +0800790 if (!err && m && m->msg_control) {
791 struct ubuf_info *uarg = m->msg_control;
792 uarg->callback(uarg, false);
793 }
794 }
795
Arnd Bergmann02df55d2010-02-18 05:45:36 +0000796 if (err)
Arnd Bergmannb9fb9ee2010-02-18 05:48:17 +0000797 goto err_kfree;
Arnd Bergmann20d29d72010-01-30 12:24:26 +0000798
799 skb_set_network_header(skb, ETH_HLEN);
Arnd Bergmannb9fb9ee2010-02-18 05:48:17 +0000800 skb_reset_mac_header(skb);
801 skb->protocol = eth_hdr(skb)->h_proto;
802
803 if (vnet_hdr_len) {
Michael S. Tsirkin6ae7feb2014-11-23 17:24:15 +0200804 err = macvtap_skb_from_vnet_hdr(q, skb, &vnet_hdr);
Arnd Bergmannb9fb9ee2010-02-18 05:48:17 +0000805 if (err)
806 goto err_kfree;
807 }
808
Jason Wang40893fd2013-03-26 23:11:22 +0000809 skb_probe_transport_header(skb, ETH_HLEN);
Jason Wang9b4d6692013-03-25 20:19:55 +0000810
Ivan Vecerac5c62f12015-07-23 16:37:43 +0200811 /* Move network header to the right position for VLAN tagged packets */
812 if ((skb->protocol == htons(ETH_P_8021Q) ||
813 skb->protocol == htons(ETH_P_8021AD)) &&
814 __vlan_get_protocol(skb, skb->protocol, &depth) != 0)
815 skb_set_network_header(skb, depth);
816
Vlad Yasevichac4e4af2013-06-25 16:04:20 -0400817 rcu_read_lock();
818 vlan = rcu_dereference(q->vlan);
Shirley Ma97bc3632011-07-06 12:26:11 +0000819 /* copy skb_ubuf_info for callback when skb has no error */
Jason Wang01d66572012-05-02 11:42:06 +0800820 if (zerocopy) {
Shirley Ma97bc3632011-07-06 12:26:11 +0000821 skb_shinfo(skb)->destructor_arg = m->msg_control;
Jason Wang01d66572012-05-02 11:42:06 +0800822 skb_shinfo(skb)->tx_flags |= SKBTX_DEV_ZEROCOPY;
Pravin B Shelarc9af6db2013-02-11 09:27:41 +0000823 skb_shinfo(skb)->tx_flags |= SKBTX_SHARED_FRAG;
Jason Wang01d66572012-05-02 11:42:06 +0800824 }
Eric Dumazet29d79192013-08-08 08:06:14 -0700825 if (vlan) {
Vlad Yasevich6acf54f2013-12-11 13:27:10 -0500826 skb->dev = vlan->dev;
827 dev_queue_xmit(skb);
Eric Dumazet29d79192013-08-08 08:06:14 -0700828 } else {
Arnd Bergmann02df55d2010-02-18 05:45:36 +0000829 kfree_skb(skb);
Eric Dumazet29d79192013-08-08 08:06:14 -0700830 }
Vlad Yasevichac4e4af2013-06-25 16:04:20 -0400831 rcu_read_unlock();
Arnd Bergmann20d29d72010-01-30 12:24:26 +0000832
Shirley Ma97bc3632011-07-06 12:26:11 +0000833 return total_len;
Arnd Bergmann02df55d2010-02-18 05:45:36 +0000834
Arnd Bergmannb9fb9ee2010-02-18 05:48:17 +0000835err_kfree:
836 kfree_skb(skb);
837
Arnd Bergmann02df55d2010-02-18 05:45:36 +0000838err:
Vlad Yasevichac4e4af2013-06-25 16:04:20 -0400839 rcu_read_lock();
840 vlan = rcu_dereference(q->vlan);
Arnd Bergmann02df55d2010-02-18 05:45:36 +0000841 if (vlan)
Jason Wangcd3e22b2013-11-25 17:19:04 +0800842 this_cpu_inc(vlan->pcpu_stats->tx_dropped);
Vlad Yasevichac4e4af2013-06-25 16:04:20 -0400843 rcu_read_unlock();
Arnd Bergmann02df55d2010-02-18 05:45:36 +0000844
Arnd Bergmann02df55d2010-02-18 05:45:36 +0000845 return err;
Arnd Bergmann20d29d72010-01-30 12:24:26 +0000846}
847
Al Virof5ff53b2014-06-19 15:36:49 -0400848static ssize_t macvtap_write_iter(struct kiocb *iocb, struct iov_iter *from)
Arnd Bergmann20d29d72010-01-30 12:24:26 +0000849{
850 struct file *file = iocb->ki_filp;
Arnd Bergmann02df55d2010-02-18 05:45:36 +0000851 struct macvtap_queue *q = file->private_data;
Arnd Bergmann20d29d72010-01-30 12:24:26 +0000852
Al Virof5ff53b2014-06-19 15:36:49 -0400853 return macvtap_get_user(q, NULL, from, file->f_flags & O_NONBLOCK);
Arnd Bergmann20d29d72010-01-30 12:24:26 +0000854}
855
856/* Put packet to the user space buffer */
857static ssize_t macvtap_put_user(struct macvtap_queue *q,
858 const struct sk_buff *skb,
Herbert Xu6c36d2e2014-11-07 21:22:25 +0800859 struct iov_iter *iter)
Arnd Bergmann20d29d72010-01-30 12:24:26 +0000860{
Arnd Bergmann20d29d72010-01-30 12:24:26 +0000861 int ret;
Arnd Bergmannb9fb9ee2010-02-18 05:48:17 +0000862 int vnet_hdr_len = 0;
Basil Gorf09e2242012-05-03 22:55:24 +0000863 int vlan_offset = 0;
Herbert Xu6c36d2e2014-11-07 21:22:25 +0800864 int total;
Arnd Bergmannb9fb9ee2010-02-18 05:48:17 +0000865
866 if (q->flags & IFF_VNET_HDR) {
867 struct virtio_net_hdr vnet_hdr;
Michael S. Tsirkin55afbd02010-04-29 13:50:48 +0300868 vnet_hdr_len = q->vnet_hdr_sz;
Herbert Xu6c36d2e2014-11-07 21:22:25 +0800869 if (iov_iter_count(iter) < vnet_hdr_len)
Arnd Bergmannb9fb9ee2010-02-18 05:48:17 +0000870 return -EINVAL;
871
Michael S. Tsirkin6ae7feb2014-11-23 17:24:15 +0200872 macvtap_skb_to_vnet_hdr(q, skb, &vnet_hdr);
Arnd Bergmannb9fb9ee2010-02-18 05:48:17 +0000873
Herbert Xu6c36d2e2014-11-07 21:22:25 +0800874 if (copy_to_iter(&vnet_hdr, sizeof(vnet_hdr), iter) !=
875 sizeof(vnet_hdr))
Arnd Bergmannb9fb9ee2010-02-18 05:48:17 +0000876 return -EFAULT;
Jason Wang7cc76f52014-11-20 16:31:05 +0800877
878 iov_iter_advance(iter, vnet_hdr_len - sizeof(vnet_hdr));
Arnd Bergmannb9fb9ee2010-02-18 05:48:17 +0000879 }
Herbert Xu6c36d2e2014-11-07 21:22:25 +0800880 total = vnet_hdr_len;
Jason Wangce232ce2013-12-11 13:08:34 +0800881 total += skb->len;
Arnd Bergmann20d29d72010-01-30 12:24:26 +0000882
Jiri Pirkodf8a39d2015-01-13 17:13:44 +0100883 if (skb_vlan_tag_present(skb)) {
Basil Gorf09e2242012-05-03 22:55:24 +0000884 struct {
885 __be16 h_vlan_proto;
886 __be16 h_vlan_TCI;
887 } veth;
Jason Wang0fbe0d42013-07-16 13:36:34 +0800888 veth.h_vlan_proto = skb->vlan_proto;
Jiri Pirkodf8a39d2015-01-13 17:13:44 +0100889 veth.h_vlan_TCI = htons(skb_vlan_tag_get(skb));
Arnd Bergmann20d29d72010-01-30 12:24:26 +0000890
Basil Gorf09e2242012-05-03 22:55:24 +0000891 vlan_offset = offsetof(struct vlan_ethhdr, h_vlan_proto);
Jason Wangce232ce2013-12-11 13:08:34 +0800892 total += VLAN_HLEN;
Arnd Bergmann20d29d72010-01-30 12:24:26 +0000893
Herbert Xu6c36d2e2014-11-07 21:22:25 +0800894 ret = skb_copy_datagram_iter(skb, 0, iter, vlan_offset);
895 if (ret || !iov_iter_count(iter))
Basil Gorf09e2242012-05-03 22:55:24 +0000896 goto done;
897
Herbert Xu6c36d2e2014-11-07 21:22:25 +0800898 ret = copy_to_iter(&veth, sizeof(veth), iter);
899 if (ret != sizeof(veth) || !iov_iter_count(iter))
Basil Gorf09e2242012-05-03 22:55:24 +0000900 goto done;
901 }
902
Herbert Xu6c36d2e2014-11-07 21:22:25 +0800903 ret = skb_copy_datagram_iter(skb, vlan_offset, iter,
904 skb->len - vlan_offset);
Basil Gorf09e2242012-05-03 22:55:24 +0000905
906done:
Jason Wangce232ce2013-12-11 13:08:34 +0800907 return ret ? ret : total;
Arnd Bergmann20d29d72010-01-30 12:24:26 +0000908}
909
Zhi Yong Wu55ec8e22013-12-07 04:13:05 +0800910static ssize_t macvtap_do_read(struct macvtap_queue *q,
Al Viro3af0bfe2014-11-07 14:13:53 -0500911 struct iov_iter *to,
Arnd Bergmann501c7742010-02-18 05:46:50 +0000912 int noblock)
Arnd Bergmann20d29d72010-01-30 12:24:26 +0000913{
Hong zhi guoccf7e722012-06-06 22:36:27 +0000914 DEFINE_WAIT(wait);
Arnd Bergmann20d29d72010-01-30 12:24:26 +0000915 struct sk_buff *skb;
Arnd Bergmann501c7742010-02-18 05:46:50 +0000916 ssize_t ret = 0;
Arnd Bergmann20d29d72010-01-30 12:24:26 +0000917
Al Viro3af0bfe2014-11-07 14:13:53 -0500918 if (!iov_iter_count(to))
919 return 0;
920
921 while (1) {
Jason Wang89cee912013-06-05 23:54:34 +0000922 if (!noblock)
923 prepare_to_wait(sk_sleep(&q->sk), &wait,
924 TASK_INTERRUPTIBLE);
Arnd Bergmann20d29d72010-01-30 12:24:26 +0000925
926 /* Read frames from the queue */
927 skb = skb_dequeue(&q->sk.sk_receive_queue);
Al Viro3af0bfe2014-11-07 14:13:53 -0500928 if (skb)
929 break;
930 if (noblock) {
931 ret = -EAGAIN;
932 break;
Arnd Bergmann20d29d72010-01-30 12:24:26 +0000933 }
Al Viro3af0bfe2014-11-07 14:13:53 -0500934 if (signal_pending(current)) {
935 ret = -ERESTARTSYS;
936 break;
937 }
938 /* Nothing to read, let's sleep */
939 schedule();
Arnd Bergmann20d29d72010-01-30 12:24:26 +0000940 }
Vlad Yasevicha499a2e2015-11-09 09:14:17 -0500941 if (!noblock)
942 finish_wait(sk_sleep(&q->sk), &wait);
943
Al Viro3af0bfe2014-11-07 14:13:53 -0500944 if (skb) {
945 ret = macvtap_put_user(q, skb, to);
Jason Wangf51a5e82014-12-01 16:53:15 +0800946 if (unlikely(ret < 0))
947 kfree_skb(skb);
948 else
949 consume_skb(skb);
Al Viro3af0bfe2014-11-07 14:13:53 -0500950 }
Arnd Bergmann501c7742010-02-18 05:46:50 +0000951 return ret;
952}
Arnd Bergmann20d29d72010-01-30 12:24:26 +0000953
Al Viro3af0bfe2014-11-07 14:13:53 -0500954static ssize_t macvtap_read_iter(struct kiocb *iocb, struct iov_iter *to)
Arnd Bergmann501c7742010-02-18 05:46:50 +0000955{
956 struct file *file = iocb->ki_filp;
957 struct macvtap_queue *q = file->private_data;
Al Viro3af0bfe2014-11-07 14:13:53 -0500958 ssize_t len = iov_iter_count(to), ret;
Arnd Bergmann501c7742010-02-18 05:46:50 +0000959
Al Viro3af0bfe2014-11-07 14:13:53 -0500960 ret = macvtap_do_read(q, to, file->f_flags & O_NONBLOCK);
Jason Wangce232ce2013-12-11 13:08:34 +0800961 ret = min_t(ssize_t, ret, len);
Zhi Yong Wue6ebc7f2013-12-06 14:16:50 +0800962 if (ret > 0)
963 iocb->ki_pos = ret;
Arnd Bergmann20d29d72010-01-30 12:24:26 +0000964 return ret;
965}
966
Jason Wang8f475a32013-06-05 23:54:36 +0000967static struct macvlan_dev *macvtap_get_vlan(struct macvtap_queue *q)
968{
969 struct macvlan_dev *vlan;
970
Vlad Yasevich441ac0f2013-06-25 16:04:19 -0400971 ASSERT_RTNL();
972 vlan = rtnl_dereference(q->vlan);
Jason Wang8f475a32013-06-05 23:54:36 +0000973 if (vlan)
974 dev_hold(vlan->dev);
Jason Wang8f475a32013-06-05 23:54:36 +0000975
976 return vlan;
977}
978
979static void macvtap_put_vlan(struct macvlan_dev *vlan)
980{
981 dev_put(vlan->dev);
982}
983
Jason Wang815f2362013-06-05 23:54:39 +0000984static int macvtap_ioctl_set_queue(struct file *file, unsigned int flags)
985{
986 struct macvtap_queue *q = file->private_data;
987 struct macvlan_dev *vlan;
988 int ret;
989
990 vlan = macvtap_get_vlan(q);
991 if (!vlan)
992 return -EINVAL;
993
994 if (flags & IFF_ATTACH_QUEUE)
995 ret = macvtap_enable_queue(vlan->dev, file, q);
996 else if (flags & IFF_DETACH_QUEUE)
997 ret = macvtap_disable_queue(q);
Jason Wangf57855a2013-06-13 14:23:36 +0800998 else
999 ret = -EINVAL;
Jason Wang815f2362013-06-05 23:54:39 +00001000
1001 macvtap_put_vlan(vlan);
1002 return ret;
1003}
1004
Vlad Yasevich2be5c762013-06-25 16:04:21 -04001005static int set_offload(struct macvtap_queue *q, unsigned long arg)
1006{
1007 struct macvlan_dev *vlan;
1008 netdev_features_t features;
1009 netdev_features_t feature_mask = 0;
1010
1011 vlan = rtnl_dereference(q->vlan);
1012 if (!vlan)
1013 return -ENOLINK;
1014
1015 features = vlan->dev->features;
1016
1017 if (arg & TUN_F_CSUM) {
1018 feature_mask = NETIF_F_HW_CSUM;
1019
1020 if (arg & (TUN_F_TSO4 | TUN_F_TSO6)) {
1021 if (arg & TUN_F_TSO_ECN)
1022 feature_mask |= NETIF_F_TSO_ECN;
1023 if (arg & TUN_F_TSO4)
1024 feature_mask |= NETIF_F_TSO;
1025 if (arg & TUN_F_TSO6)
1026 feature_mask |= NETIF_F_TSO6;
1027 }
Vlad Yaseviche3e3c422015-02-03 16:36:17 -05001028
1029 if (arg & TUN_F_UFO)
1030 feature_mask |= NETIF_F_UFO;
Vlad Yasevich2be5c762013-06-25 16:04:21 -04001031 }
1032
1033 /* tun/tap driver inverts the usage for TSO offloads, where
1034 * setting the TSO bit means that the userspace wants to
1035 * accept TSO frames and turning it off means that user space
1036 * does not support TSO.
1037 * For macvtap, we have to invert it to mean the same thing.
1038 * When user space turns off TSO, we turn off GSO/LRO so that
1039 * user-space will not receive TSO frames.
1040 */
Vlad Yaseviche3e3c422015-02-03 16:36:17 -05001041 if (feature_mask & (NETIF_F_TSO | NETIF_F_TSO6 | NETIF_F_UFO))
Vlad Yasevich2be5c762013-06-25 16:04:21 -04001042 features |= RX_OFFLOADS;
1043 else
1044 features &= ~RX_OFFLOADS;
1045
1046 /* tap_features are the same as features on tun/tap and
1047 * reflect user expectations.
1048 */
Vlad Yasevicha567dd62013-08-16 15:25:00 -04001049 vlan->tap_features = feature_mask;
Vlad Yasevich2be5c762013-06-25 16:04:21 -04001050 vlan->set_features = features;
1051 netdev_update_features(vlan->dev);
1052
1053 return 0;
1054}
1055
Arnd Bergmann20d29d72010-01-30 12:24:26 +00001056/*
1057 * provide compatibility with generic tun/tap interface
1058 */
1059static long macvtap_ioctl(struct file *file, unsigned int cmd,
1060 unsigned long arg)
1061{
Arnd Bergmann02df55d2010-02-18 05:45:36 +00001062 struct macvtap_queue *q = file->private_data;
1063 struct macvlan_dev *vlan;
Arnd Bergmann20d29d72010-01-30 12:24:26 +00001064 void __user *argp = (void __user *)arg;
1065 struct ifreq __user *ifr = argp;
1066 unsigned int __user *up = argp;
Michael S. Tsirkin39ec7de2014-12-16 15:04:56 +02001067 unsigned short u;
Michael S. Tsirkin55afbd02010-04-29 13:50:48 +03001068 int __user *sp = argp;
Justin Cormack7f460d32015-05-13 19:19:02 +01001069 struct sockaddr sa;
Michael S. Tsirkin55afbd02010-04-29 13:50:48 +03001070 int s;
Arnd Bergmann02df55d2010-02-18 05:45:36 +00001071 int ret;
Arnd Bergmann20d29d72010-01-30 12:24:26 +00001072
1073 switch (cmd) {
1074 case TUNSETIFF:
1075 /* ignore the name, just look at flags */
1076 if (get_user(u, &ifr->ifr_flags))
1077 return -EFAULT;
Arnd Bergmannb9fb9ee2010-02-18 05:48:17 +00001078
1079 ret = 0;
Michael S. Tsirkin6ae7feb2014-11-23 17:24:15 +02001080 if ((u & ~MACVTAP_FEATURES) != (IFF_NO_PI | IFF_TAP))
Arnd Bergmannb9fb9ee2010-02-18 05:48:17 +00001081 ret = -EINVAL;
1082 else
Michael S. Tsirkin39ec7de2014-12-16 15:04:56 +02001083 q->flags = (q->flags & ~MACVTAP_FEATURES) | u;
Arnd Bergmannb9fb9ee2010-02-18 05:48:17 +00001084
1085 return ret;
Arnd Bergmann20d29d72010-01-30 12:24:26 +00001086
1087 case TUNGETIFF:
Vlad Yasevich441ac0f2013-06-25 16:04:19 -04001088 rtnl_lock();
Jason Wang8f475a32013-06-05 23:54:36 +00001089 vlan = macvtap_get_vlan(q);
Vlad Yasevich441ac0f2013-06-25 16:04:19 -04001090 if (!vlan) {
1091 rtnl_unlock();
Arnd Bergmann02df55d2010-02-18 05:45:36 +00001092 return -ENOLINK;
Vlad Yasevich441ac0f2013-06-25 16:04:19 -04001093 }
Arnd Bergmann02df55d2010-02-18 05:45:36 +00001094
1095 ret = 0;
Michael S. Tsirkin39ec7de2014-12-16 15:04:56 +02001096 u = q->flags;
Eric Dumazet13707f92011-01-26 19:28:23 +00001097 if (copy_to_user(&ifr->ifr_name, vlan->dev->name, IFNAMSIZ) ||
Michael S. Tsirkin39ec7de2014-12-16 15:04:56 +02001098 put_user(u, &ifr->ifr_flags))
Arnd Bergmann02df55d2010-02-18 05:45:36 +00001099 ret = -EFAULT;
Jason Wang8f475a32013-06-05 23:54:36 +00001100 macvtap_put_vlan(vlan);
Vlad Yasevich441ac0f2013-06-25 16:04:19 -04001101 rtnl_unlock();
Arnd Bergmann02df55d2010-02-18 05:45:36 +00001102 return ret;
Arnd Bergmann20d29d72010-01-30 12:24:26 +00001103
Jason Wang815f2362013-06-05 23:54:39 +00001104 case TUNSETQUEUE:
1105 if (get_user(u, &ifr->ifr_flags))
1106 return -EFAULT;
Vlad Yasevich441ac0f2013-06-25 16:04:19 -04001107 rtnl_lock();
1108 ret = macvtap_ioctl_set_queue(file, u);
1109 rtnl_unlock();
Jason Wang82a19eb2013-07-16 13:36:33 +08001110 return ret;
Jason Wang815f2362013-06-05 23:54:39 +00001111
Arnd Bergmann20d29d72010-01-30 12:24:26 +00001112 case TUNGETFEATURES:
Michael S. Tsirkin6ae7feb2014-11-23 17:24:15 +02001113 if (put_user(IFF_TAP | IFF_NO_PI | MACVTAP_FEATURES, up))
Arnd Bergmann20d29d72010-01-30 12:24:26 +00001114 return -EFAULT;
1115 return 0;
1116
1117 case TUNSETSNDBUF:
Michael S. Tsirkin3ea79242015-09-18 13:41:09 +03001118 if (get_user(s, sp))
Arnd Bergmann20d29d72010-01-30 12:24:26 +00001119 return -EFAULT;
1120
Michael S. Tsirkin3ea79242015-09-18 13:41:09 +03001121 q->sk.sk_sndbuf = s;
Arnd Bergmann20d29d72010-01-30 12:24:26 +00001122 return 0;
1123
Michael S. Tsirkin55afbd02010-04-29 13:50:48 +03001124 case TUNGETVNETHDRSZ:
1125 s = q->vnet_hdr_sz;
1126 if (put_user(s, sp))
1127 return -EFAULT;
1128 return 0;
1129
1130 case TUNSETVNETHDRSZ:
1131 if (get_user(s, sp))
1132 return -EFAULT;
1133 if (s < (int)sizeof(struct virtio_net_hdr))
1134 return -EINVAL;
1135
1136 q->vnet_hdr_sz = s;
1137 return 0;
1138
Michael S. Tsirkin01b07fb2014-12-16 15:05:10 +02001139 case TUNGETVNETLE:
1140 s = !!(q->flags & MACVTAP_VNET_LE);
1141 if (put_user(s, sp))
1142 return -EFAULT;
1143 return 0;
1144
1145 case TUNSETVNETLE:
1146 if (get_user(s, sp))
1147 return -EFAULT;
1148 if (s)
1149 q->flags |= MACVTAP_VNET_LE;
1150 else
1151 q->flags &= ~MACVTAP_VNET_LE;
1152 return 0;
1153
Greg Kurz8b8e6582015-04-24 14:50:36 +02001154 case TUNGETVNETBE:
1155 return macvtap_get_vnet_be(q, sp);
1156
1157 case TUNSETVNETBE:
1158 return macvtap_set_vnet_be(q, sp);
1159
Arnd Bergmann20d29d72010-01-30 12:24:26 +00001160 case TUNSETOFFLOAD:
1161 /* let the user check for future flags */
1162 if (arg & ~(TUN_F_CSUM | TUN_F_TSO4 | TUN_F_TSO6 |
Vlad Yaseviche3e3c422015-02-03 16:36:17 -05001163 TUN_F_TSO_ECN | TUN_F_UFO))
Arnd Bergmann20d29d72010-01-30 12:24:26 +00001164 return -EINVAL;
1165
Vlad Yasevich2be5c762013-06-25 16:04:21 -04001166 rtnl_lock();
1167 ret = set_offload(q, arg);
1168 rtnl_unlock();
1169 return ret;
Arnd Bergmann20d29d72010-01-30 12:24:26 +00001170
Justin Cormackb5082082015-05-11 20:00:10 +01001171 case SIOCGIFHWADDR:
1172 rtnl_lock();
1173 vlan = macvtap_get_vlan(q);
1174 if (!vlan) {
1175 rtnl_unlock();
1176 return -ENOLINK;
1177 }
1178 ret = 0;
1179 u = vlan->dev->type;
1180 if (copy_to_user(&ifr->ifr_name, vlan->dev->name, IFNAMSIZ) ||
1181 copy_to_user(&ifr->ifr_hwaddr.sa_data, vlan->dev->dev_addr, ETH_ALEN) ||
1182 put_user(u, &ifr->ifr_hwaddr.sa_family))
1183 ret = -EFAULT;
1184 macvtap_put_vlan(vlan);
1185 rtnl_unlock();
1186 return ret;
1187
1188 case SIOCSIFHWADDR:
Justin Cormack7f460d32015-05-13 19:19:02 +01001189 if (copy_from_user(&sa, &ifr->ifr_hwaddr, sizeof(sa)))
1190 return -EFAULT;
Justin Cormackb5082082015-05-11 20:00:10 +01001191 rtnl_lock();
1192 vlan = macvtap_get_vlan(q);
1193 if (!vlan) {
1194 rtnl_unlock();
1195 return -ENOLINK;
1196 }
Justin Cormack7f460d32015-05-13 19:19:02 +01001197 ret = dev_set_mac_address(vlan->dev, &sa);
Justin Cormackb5082082015-05-11 20:00:10 +01001198 macvtap_put_vlan(vlan);
1199 rtnl_unlock();
1200 return ret;
1201
Arnd Bergmann20d29d72010-01-30 12:24:26 +00001202 default:
1203 return -EINVAL;
1204 }
1205}
1206
1207#ifdef CONFIG_COMPAT
1208static long macvtap_compat_ioctl(struct file *file, unsigned int cmd,
1209 unsigned long arg)
1210{
1211 return macvtap_ioctl(file, cmd, (unsigned long)compat_ptr(arg));
1212}
1213#endif
1214
1215static const struct file_operations macvtap_fops = {
1216 .owner = THIS_MODULE,
1217 .open = macvtap_open,
1218 .release = macvtap_release,
Al Viro3af0bfe2014-11-07 14:13:53 -05001219 .read_iter = macvtap_read_iter,
Al Virof5ff53b2014-06-19 15:36:49 -04001220 .write_iter = macvtap_write_iter,
Arnd Bergmann20d29d72010-01-30 12:24:26 +00001221 .poll = macvtap_poll,
1222 .llseek = no_llseek,
1223 .unlocked_ioctl = macvtap_ioctl,
1224#ifdef CONFIG_COMPAT
1225 .compat_ioctl = macvtap_compat_ioctl,
1226#endif
1227};
1228
Ying Xue1b784142015-03-02 15:37:48 +08001229static int macvtap_sendmsg(struct socket *sock, struct msghdr *m,
1230 size_t total_len)
Arnd Bergmann501c7742010-02-18 05:46:50 +00001231{
1232 struct macvtap_queue *q = container_of(sock, struct macvtap_queue, sock);
Al Viroc0371da2014-11-24 10:42:55 -05001233 return macvtap_get_user(q, m, &m->msg_iter, m->msg_flags & MSG_DONTWAIT);
Arnd Bergmann501c7742010-02-18 05:46:50 +00001234}
1235
Ying Xue1b784142015-03-02 15:37:48 +08001236static int macvtap_recvmsg(struct socket *sock, struct msghdr *m,
1237 size_t total_len, int flags)
Arnd Bergmann501c7742010-02-18 05:46:50 +00001238{
1239 struct macvtap_queue *q = container_of(sock, struct macvtap_queue, sock);
1240 int ret;
1241 if (flags & ~(MSG_DONTWAIT|MSG_TRUNC))
1242 return -EINVAL;
Al Viroc0371da2014-11-24 10:42:55 -05001243 ret = macvtap_do_read(q, &m->msg_iter, flags & MSG_DONTWAIT);
David S. Millerde2aa472013-12-10 22:06:18 -05001244 if (ret > total_len) {
1245 m->msg_flags |= MSG_TRUNC;
1246 ret = flags & MSG_TRUNC ? ret : total_len;
1247 }
Arnd Bergmann501c7742010-02-18 05:46:50 +00001248 return ret;
1249}
1250
1251/* Ops structure to mimic raw sockets with tun */
1252static const struct proto_ops macvtap_socket_ops = {
1253 .sendmsg = macvtap_sendmsg,
1254 .recvmsg = macvtap_recvmsg,
1255};
1256
1257/* Get an underlying socket object from tun file. Returns error unless file is
1258 * attached to a device. The returned object works like a packet socket, it
1259 * can be used for sock_sendmsg/sock_recvmsg. The caller is responsible for
1260 * holding a reference to the file for as long as the socket is in use. */
1261struct socket *macvtap_get_socket(struct file *file)
1262{
1263 struct macvtap_queue *q;
1264 if (file->f_op != &macvtap_fops)
1265 return ERR_PTR(-EINVAL);
1266 q = file->private_data;
1267 if (!q)
1268 return ERR_PTR(-EBADFD);
1269 return &q->sock;
1270}
1271EXPORT_SYMBOL_GPL(macvtap_get_socket);
1272
Eric W. Biederman9bf19072011-10-20 04:28:46 +00001273static int macvtap_device_event(struct notifier_block *unused,
1274 unsigned long event, void *ptr)
1275{
Jiri Pirko351638e2013-05-28 01:30:21 +00001276 struct net_device *dev = netdev_notifier_info_to_dev(ptr);
Eric W. Biedermane09eff72011-10-20 04:29:24 +00001277 struct macvlan_dev *vlan;
Eric W. Biederman9bf19072011-10-20 04:28:46 +00001278 struct device *classdev;
1279 dev_t devt;
Eric W. Biedermane09eff72011-10-20 04:29:24 +00001280 int err;
Eric W. Biederman9bf19072011-10-20 04:28:46 +00001281
1282 if (dev->rtnl_link_ops != &macvtap_link_ops)
1283 return NOTIFY_DONE;
1284
Eric W. Biedermane09eff72011-10-20 04:29:24 +00001285 vlan = netdev_priv(dev);
Eric W. Biederman9bf19072011-10-20 04:28:46 +00001286
1287 switch (event) {
1288 case NETDEV_REGISTER:
1289 /* Create the device node here after the network device has
1290 * been registered but before register_netdevice has
1291 * finished running.
1292 */
Eric W. Biedermane09eff72011-10-20 04:29:24 +00001293 err = macvtap_get_minor(vlan);
1294 if (err)
1295 return notifier_from_errno(err);
1296
1297 devt = MKDEV(MAJOR(macvtap_major), vlan->minor);
Eric W. Biederman9bf19072011-10-20 04:28:46 +00001298 classdev = device_create(macvtap_class, &dev->dev, devt,
1299 dev, "tap%d", dev->ifindex);
Eric W. Biedermane09eff72011-10-20 04:29:24 +00001300 if (IS_ERR(classdev)) {
1301 macvtap_free_minor(vlan);
Eric W. Biederman9bf19072011-10-20 04:28:46 +00001302 return notifier_from_errno(PTR_ERR(classdev));
Eric W. Biedermane09eff72011-10-20 04:29:24 +00001303 }
Eric W. Biederman9bf19072011-10-20 04:28:46 +00001304 break;
1305 case NETDEV_UNREGISTER:
Eric W. Biedermane09eff72011-10-20 04:29:24 +00001306 devt = MKDEV(MAJOR(macvtap_major), vlan->minor);
Eric W. Biederman9bf19072011-10-20 04:28:46 +00001307 device_destroy(macvtap_class, devt);
Eric W. Biedermane09eff72011-10-20 04:29:24 +00001308 macvtap_free_minor(vlan);
Eric W. Biederman9bf19072011-10-20 04:28:46 +00001309 break;
1310 }
1311
1312 return NOTIFY_DONE;
1313}
1314
1315static struct notifier_block macvtap_notifier_block __read_mostly = {
1316 .notifier_call = macvtap_device_event,
1317};
1318
Arnd Bergmann20d29d72010-01-30 12:24:26 +00001319static int macvtap_init(void)
1320{
1321 int err;
1322
1323 err = alloc_chrdev_region(&macvtap_major, 0,
1324 MACVTAP_NUM_DEVS, "macvtap");
1325 if (err)
1326 goto out1;
1327
1328 cdev_init(&macvtap_cdev, &macvtap_fops);
1329 err = cdev_add(&macvtap_cdev, macvtap_major, MACVTAP_NUM_DEVS);
1330 if (err)
1331 goto out2;
1332
1333 macvtap_class = class_create(THIS_MODULE, "macvtap");
1334 if (IS_ERR(macvtap_class)) {
1335 err = PTR_ERR(macvtap_class);
1336 goto out3;
1337 }
1338
Eric W. Biederman9bf19072011-10-20 04:28:46 +00001339 err = register_netdevice_notifier(&macvtap_notifier_block);
Arnd Bergmann20d29d72010-01-30 12:24:26 +00001340 if (err)
1341 goto out4;
1342
Eric W. Biederman9bf19072011-10-20 04:28:46 +00001343 err = macvlan_link_register(&macvtap_link_ops);
1344 if (err)
1345 goto out5;
1346
Arnd Bergmann20d29d72010-01-30 12:24:26 +00001347 return 0;
1348
Eric W. Biederman9bf19072011-10-20 04:28:46 +00001349out5:
1350 unregister_netdevice_notifier(&macvtap_notifier_block);
Arnd Bergmann20d29d72010-01-30 12:24:26 +00001351out4:
1352 class_unregister(macvtap_class);
1353out3:
1354 cdev_del(&macvtap_cdev);
1355out2:
1356 unregister_chrdev_region(macvtap_major, MACVTAP_NUM_DEVS);
1357out1:
1358 return err;
1359}
1360module_init(macvtap_init);
1361
1362static void macvtap_exit(void)
1363{
1364 rtnl_link_unregister(&macvtap_link_ops);
Eric W. Biederman9bf19072011-10-20 04:28:46 +00001365 unregister_netdevice_notifier(&macvtap_notifier_block);
Arnd Bergmann20d29d72010-01-30 12:24:26 +00001366 class_unregister(macvtap_class);
1367 cdev_del(&macvtap_cdev);
1368 unregister_chrdev_region(macvtap_major, MACVTAP_NUM_DEVS);
Johannes Thumshirnd5de1982015-07-08 17:16:49 +02001369 idr_destroy(&minor_idr);
Arnd Bergmann20d29d72010-01-30 12:24:26 +00001370}
1371module_exit(macvtap_exit);
1372
1373MODULE_ALIAS_RTNL_LINK("macvtap");
1374MODULE_AUTHOR("Arnd Bergmann <arnd@arndb.de>");
1375MODULE_LICENSE("GPL");