Commit | Line | Data |
---|---|---|
20d29d7a AB |
1 | #include <linux/etherdevice.h> |
2 | #include <linux/if_macvlan.h> | |
f09e2249 | 3 | #include <linux/if_vlan.h> |
20d29d7a AB |
4 | #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> | |
5a0e3ad6 | 13 | #include <linux/slab.h> |
20d29d7a AB |
14 | #include <linux/wait.h> |
15 | #include <linux/cdev.h> | |
40401530 | 16 | #include <linux/idr.h> |
20d29d7a AB |
17 | #include <linux/fs.h> |
18 | ||
5188cd44 | 19 | #include <net/ipv6.h> |
20d29d7a AB |
20 | #include <net/net_namespace.h> |
21 | #include <net/rtnetlink.h> | |
22 | #include <net/sock.h> | |
b9fb9ee0 | 23 | #include <linux/virtio_net.h> |
20d29d7a AB |
24 | |
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 | * | |
20d29d7a AB |
34 | */ |
35 | struct macvtap_queue { | |
36 | struct sock sk; | |
37 | struct socket sock; | |
43815482 | 38 | struct socket_wq wq; |
55afbd08 | 39 | int vnet_hdr_sz; |
13707f9e | 40 | struct macvlan_dev __rcu *vlan; |
20d29d7a | 41 | struct file *file; |
b9fb9ee0 | 42 | unsigned int flags; |
376b1aab | 43 | u16 queue_index; |
815f236d JW |
44 | bool enabled; |
45 | struct list_head next; | |
20d29d7a AB |
46 | }; |
47 | ||
48 | static struct proto macvtap_proto = { | |
49 | .name = "macvtap", | |
50 | .owner = THIS_MODULE, | |
51 | .obj_size = sizeof (struct macvtap_queue), | |
52 | }; | |
53 | ||
54 | /* | |
e09eff7f | 55 | * Variables for dealing with macvtaps device numbers. |
20d29d7a | 56 | */ |
1ebed71a | 57 | static dev_t macvtap_major; |
e09eff7f EB |
58 | #define MACVTAP_NUM_DEVS (1U << MINORBITS) |
59 | static DEFINE_MUTEX(minor_lock); | |
60 | static DEFINE_IDR(minor_idr); | |
61 | ||
97bc3633 | 62 | #define GOODCOPY_LEN 128 |
20d29d7a AB |
63 | static struct class *macvtap_class; |
64 | static struct cdev macvtap_cdev; | |
65 | ||
501c774c AB |
66 | static const struct proto_ops macvtap_socket_ops; |
67 | ||
2be5c767 | 68 | #define TUN_OFFLOADS (NETIF_F_HW_CSUM | NETIF_F_TSO_ECN | NETIF_F_TSO | \ |
3d0ad094 | 69 | NETIF_F_TSO6) |
2be5c767 | 70 | #define RX_OFFLOADS (NETIF_F_GRO | NETIF_F_LRO) |
a567dd62 VY |
71 | #define TAP_FEATURES (NETIF_F_GSO | NETIF_F_SG) |
72 | ||
6acf54f1 VY |
73 | static struct macvlan_dev *macvtap_get_vlan_rcu(const struct net_device *dev) |
74 | { | |
75 | return rcu_dereference(dev->rx_handler_data); | |
76 | } | |
77 | ||
20d29d7a AB |
78 | /* |
79 | * RCU usage: | |
02df55d2 AB |
80 | * The macvtap_queue and the macvlan_dev are loosely coupled, the |
81 | * pointers from one to the other can only be read while rcu_read_lock | |
441ac0fc | 82 | * or rtnl is held. |
20d29d7a | 83 | * |
02df55d2 AB |
84 | * Both the file and the macvlan_dev hold a reference on the macvtap_queue |
85 | * through sock_hold(&q->sk). When the macvlan_dev goes away first, | |
86 | * q->vlan becomes inaccessible. When the files gets closed, | |
87 | * macvtap_get_queue() fails. | |
20d29d7a | 88 | * |
02df55d2 AB |
89 | * There may still be references to the struct sock inside of the |
90 | * queue from outbound SKBs, but these never reference back to the | |
91 | * file or the dev. The data structure is freed through __sk_free | |
92 | * when both our references and any pending SKBs are gone. | |
20d29d7a | 93 | */ |
20d29d7a | 94 | |
815f236d | 95 | static int macvtap_enable_queue(struct net_device *dev, struct file *file, |
20d29d7a | 96 | struct macvtap_queue *q) |
815f236d JW |
97 | { |
98 | struct macvlan_dev *vlan = netdev_priv(dev); | |
99 | int err = -EINVAL; | |
100 | ||
441ac0fc | 101 | ASSERT_RTNL(); |
815f236d JW |
102 | |
103 | if (q->enabled) | |
104 | goto out; | |
105 | ||
106 | err = 0; | |
107 | rcu_assign_pointer(vlan->taps[vlan->numvtaps], q); | |
108 | q->queue_index = vlan->numvtaps; | |
109 | q->enabled = true; | |
110 | ||
111 | vlan->numvtaps++; | |
112 | out: | |
815f236d JW |
113 | return err; |
114 | } | |
115 | ||
40b8fe45 | 116 | /* Requires RTNL */ |
815f236d JW |
117 | static int macvtap_set_queue(struct net_device *dev, struct file *file, |
118 | struct macvtap_queue *q) | |
20d29d7a AB |
119 | { |
120 | struct macvlan_dev *vlan = netdev_priv(dev); | |
20d29d7a | 121 | |
815f236d | 122 | if (vlan->numqueues == MAX_MACVTAP_QUEUES) |
40b8fe45 | 123 | return -EBUSY; |
20d29d7a | 124 | |
02df55d2 | 125 | rcu_assign_pointer(q->vlan, vlan); |
376b1aab | 126 | rcu_assign_pointer(vlan->taps[vlan->numvtaps], q); |
02df55d2 | 127 | sock_hold(&q->sk); |
20d29d7a AB |
128 | |
129 | q->file = file; | |
376b1aab | 130 | q->queue_index = vlan->numvtaps; |
815f236d | 131 | q->enabled = true; |
02df55d2 | 132 | file->private_data = q; |
815f236d | 133 | list_add_tail(&q->next, &vlan->queue_list); |
20d29d7a | 134 | |
1565c7c1 | 135 | vlan->numvtaps++; |
815f236d | 136 | vlan->numqueues++; |
1565c7c1 | 137 | |
40b8fe45 | 138 | return 0; |
20d29d7a AB |
139 | } |
140 | ||
441ac0fc | 141 | static int macvtap_disable_queue(struct macvtap_queue *q) |
815f236d JW |
142 | { |
143 | struct macvlan_dev *vlan; | |
144 | struct macvtap_queue *nq; | |
145 | ||
441ac0fc | 146 | ASSERT_RTNL(); |
815f236d JW |
147 | if (!q->enabled) |
148 | return -EINVAL; | |
149 | ||
441ac0fc VY |
150 | vlan = rtnl_dereference(q->vlan); |
151 | ||
815f236d JW |
152 | if (vlan) { |
153 | int index = q->queue_index; | |
154 | BUG_ON(index >= vlan->numvtaps); | |
441ac0fc | 155 | nq = rtnl_dereference(vlan->taps[vlan->numvtaps - 1]); |
815f236d JW |
156 | nq->queue_index = index; |
157 | ||
158 | rcu_assign_pointer(vlan->taps[index], nq); | |
159 | RCU_INIT_POINTER(vlan->taps[vlan->numvtaps - 1], NULL); | |
160 | q->enabled = false; | |
161 | ||
162 | vlan->numvtaps--; | |
163 | } | |
164 | ||
165 | return 0; | |
166 | } | |
167 | ||
20d29d7a | 168 | /* |
02df55d2 AB |
169 | * The file owning the queue got closed, give up both |
170 | * the reference that the files holds as well as the | |
171 | * one from the macvlan_dev if that still exists. | |
20d29d7a AB |
172 | * |
173 | * Using the spinlock makes sure that we don't get | |
174 | * to the queue again after destroying it. | |
20d29d7a | 175 | */ |
02df55d2 | 176 | static void macvtap_put_queue(struct macvtap_queue *q) |
20d29d7a | 177 | { |
02df55d2 | 178 | struct macvlan_dev *vlan; |
20d29d7a | 179 | |
441ac0fc VY |
180 | rtnl_lock(); |
181 | vlan = rtnl_dereference(q->vlan); | |
182 | ||
02df55d2 | 183 | if (vlan) { |
815f236d | 184 | if (q->enabled) |
441ac0fc | 185 | BUG_ON(macvtap_disable_queue(q)); |
376b1aab | 186 | |
815f236d | 187 | vlan->numqueues--; |
2cfa5a04 | 188 | RCU_INIT_POINTER(q->vlan, NULL); |
02df55d2 | 189 | sock_put(&q->sk); |
815f236d | 190 | list_del_init(&q->next); |
20d29d7a AB |
191 | } |
192 | ||
441ac0fc | 193 | rtnl_unlock(); |
20d29d7a AB |
194 | |
195 | synchronize_rcu(); | |
196 | sock_put(&q->sk); | |
197 | } | |
198 | ||
199 | /* | |
1565c7c1 KK |
200 | * Select a queue based on the rxq of the device on which this packet |
201 | * arrived. If the incoming device is not mq, calculate a flow hash | |
202 | * to select a queue. If all fails, find the first available queue. | |
203 | * Cache vlan->numvtaps since it can become zero during the execution | |
204 | * of this function. | |
20d29d7a AB |
205 | */ |
206 | static struct macvtap_queue *macvtap_get_queue(struct net_device *dev, | |
207 | struct sk_buff *skb) | |
208 | { | |
209 | struct macvlan_dev *vlan = netdev_priv(dev); | |
1565c7c1 | 210 | struct macvtap_queue *tap = NULL; |
815f236d JW |
211 | /* Access to taps array is protected by rcu, but access to numvtaps |
212 | * isn't. Below we use it to lookup a queue, but treat it as a hint | |
213 | * and validate that the result isn't NULL - in case we are | |
214 | * racing against queue removal. | |
215 | */ | |
ed0483fa | 216 | int numvtaps = ACCESS_ONCE(vlan->numvtaps); |
1565c7c1 KK |
217 | __u32 rxq; |
218 | ||
219 | if (!numvtaps) | |
220 | goto out; | |
221 | ||
ef0002b5 | 222 | /* Check if we can use flow to select a queue */ |
3958afa1 | 223 | rxq = skb_get_hash(skb); |
ef0002b5 KK |
224 | if (rxq) { |
225 | tap = rcu_dereference(vlan->taps[rxq % numvtaps]); | |
376b1aab | 226 | goto out; |
ef0002b5 KK |
227 | } |
228 | ||
1565c7c1 KK |
229 | if (likely(skb_rx_queue_recorded(skb))) { |
230 | rxq = skb_get_rx_queue(skb); | |
20d29d7a | 231 | |
1565c7c1 KK |
232 | while (unlikely(rxq >= numvtaps)) |
233 | rxq -= numvtaps; | |
234 | ||
235 | tap = rcu_dereference(vlan->taps[rxq]); | |
376b1aab | 236 | goto out; |
1565c7c1 KK |
237 | } |
238 | ||
376b1aab | 239 | tap = rcu_dereference(vlan->taps[0]); |
1565c7c1 KK |
240 | out: |
241 | return tap; | |
20d29d7a AB |
242 | } |
243 | ||
02df55d2 AB |
244 | /* |
245 | * The net_device is going away, give up the reference | |
1565c7c1 KK |
246 | * that it holds on all queues and safely set the pointer |
247 | * from the queues to NULL. | |
02df55d2 | 248 | */ |
20d29d7a AB |
249 | static void macvtap_del_queues(struct net_device *dev) |
250 | { | |
251 | struct macvlan_dev *vlan = netdev_priv(dev); | |
815f236d | 252 | struct macvtap_queue *q, *tmp, *qlist[MAX_MACVTAP_QUEUES]; |
1565c7c1 | 253 | int i, j = 0; |
02df55d2 | 254 | |
441ac0fc | 255 | ASSERT_RTNL(); |
815f236d JW |
256 | list_for_each_entry_safe(q, tmp, &vlan->queue_list, next) { |
257 | list_del_init(&q->next); | |
376b1aab | 258 | qlist[j++] = q; |
376b1aab | 259 | RCU_INIT_POINTER(q->vlan, NULL); |
815f236d JW |
260 | if (q->enabled) |
261 | vlan->numvtaps--; | |
262 | vlan->numqueues--; | |
564517e8 | 263 | } |
815f236d JW |
264 | for (i = 0; i < vlan->numvtaps; i++) |
265 | RCU_INIT_POINTER(vlan->taps[i], NULL); | |
266 | BUG_ON(vlan->numvtaps); | |
267 | BUG_ON(vlan->numqueues); | |
99f34b38 EB |
268 | /* guarantee that any future macvtap_set_queue will fail */ |
269 | vlan->numvtaps = MAX_MACVTAP_QUEUES; | |
1565c7c1 KK |
270 | |
271 | for (--j; j >= 0; j--) | |
272 | sock_put(&qlist[j]->sk); | |
20d29d7a AB |
273 | } |
274 | ||
6acf54f1 | 275 | static rx_handler_result_t macvtap_handle_frame(struct sk_buff **pskb) |
20d29d7a | 276 | { |
6acf54f1 VY |
277 | struct sk_buff *skb = *pskb; |
278 | struct net_device *dev = skb->dev; | |
279 | struct macvlan_dev *vlan; | |
280 | struct macvtap_queue *q; | |
a567dd62 VY |
281 | netdev_features_t features = TAP_FEATURES; |
282 | ||
6acf54f1 VY |
283 | vlan = macvtap_get_vlan_rcu(dev); |
284 | if (!vlan) | |
285 | return RX_HANDLER_PASS; | |
286 | ||
287 | q = macvtap_get_queue(dev, skb); | |
20d29d7a | 288 | if (!q) |
6acf54f1 | 289 | return RX_HANDLER_PASS; |
8a35747a HX |
290 | |
291 | if (skb_queue_len(&q->sk.sk_receive_queue) >= dev->tx_queue_len) | |
292 | goto drop; | |
20d29d7a | 293 | |
6acf54f1 VY |
294 | skb_push(skb, ETH_HLEN); |
295 | ||
3e4f8b78 | 296 | /* Apply the forward feature mask so that we perform segmentation |
e5733321 VY |
297 | * according to users wishes. This only works if VNET_HDR is |
298 | * enabled. | |
3e4f8b78 | 299 | */ |
e5733321 VY |
300 | if (q->flags & IFF_VNET_HDR) |
301 | features |= vlan->tap_features; | |
04ffcb25 | 302 | if (netif_needs_gso(dev, skb, features)) { |
3e4f8b78 VY |
303 | struct sk_buff *segs = __skb_gso_segment(skb, features, false); |
304 | ||
305 | if (IS_ERR(segs)) | |
306 | goto drop; | |
307 | ||
308 | if (!segs) { | |
309 | skb_queue_tail(&q->sk.sk_receive_queue, skb); | |
310 | goto wake_up; | |
311 | } | |
312 | ||
313 | kfree_skb(skb); | |
314 | while (segs) { | |
315 | struct sk_buff *nskb = segs->next; | |
316 | ||
317 | segs->next = NULL; | |
318 | skb_queue_tail(&q->sk.sk_receive_queue, segs); | |
319 | segs = nskb; | |
320 | } | |
321 | } else { | |
cbdb0427 VY |
322 | /* If we receive a partial checksum and the tap side |
323 | * doesn't support checksum offload, compute the checksum. | |
324 | * Note: it doesn't matter which checksum feature to | |
325 | * check, we either support them all or none. | |
326 | */ | |
327 | if (skb->ip_summed == CHECKSUM_PARTIAL && | |
328 | !(features & NETIF_F_ALL_CSUM) && | |
329 | skb_checksum_help(skb)) | |
330 | goto drop; | |
3e4f8b78 VY |
331 | skb_queue_tail(&q->sk.sk_receive_queue, skb); |
332 | } | |
333 | ||
334 | wake_up: | |
4a4771a5 | 335 | wake_up_interruptible_poll(sk_sleep(&q->sk), POLLIN | POLLRDNORM | POLLRDBAND); |
6acf54f1 | 336 | return RX_HANDLER_CONSUMED; |
8a35747a HX |
337 | |
338 | drop: | |
6acf54f1 VY |
339 | /* Count errors/drops only here, thus don't care about args. */ |
340 | macvlan_count_rx(vlan, 0, 0, 0); | |
8a35747a | 341 | kfree_skb(skb); |
6acf54f1 | 342 | return RX_HANDLER_CONSUMED; |
20d29d7a AB |
343 | } |
344 | ||
e09eff7f EB |
345 | static int macvtap_get_minor(struct macvlan_dev *vlan) |
346 | { | |
347 | int retval = -ENOMEM; | |
e09eff7f EB |
348 | |
349 | mutex_lock(&minor_lock); | |
ec09ebc1 TH |
350 | retval = idr_alloc(&minor_idr, vlan, 1, MACVTAP_NUM_DEVS, GFP_KERNEL); |
351 | if (retval >= 0) { | |
352 | vlan->minor = retval; | |
353 | } else if (retval == -ENOSPC) { | |
e09eff7f EB |
354 | printk(KERN_ERR "too many macvtap devices\n"); |
355 | retval = -EINVAL; | |
e09eff7f | 356 | } |
e09eff7f | 357 | mutex_unlock(&minor_lock); |
ec09ebc1 | 358 | return retval < 0 ? retval : 0; |
e09eff7f EB |
359 | } |
360 | ||
361 | static void macvtap_free_minor(struct macvlan_dev *vlan) | |
362 | { | |
363 | mutex_lock(&minor_lock); | |
364 | if (vlan->minor) { | |
365 | idr_remove(&minor_idr, vlan->minor); | |
366 | vlan->minor = 0; | |
367 | } | |
368 | mutex_unlock(&minor_lock); | |
369 | } | |
370 | ||
371 | static struct net_device *dev_get_by_macvtap_minor(int minor) | |
372 | { | |
373 | struct net_device *dev = NULL; | |
374 | struct macvlan_dev *vlan; | |
375 | ||
376 | mutex_lock(&minor_lock); | |
377 | vlan = idr_find(&minor_idr, minor); | |
378 | if (vlan) { | |
379 | dev = vlan->dev; | |
380 | dev_hold(dev); | |
381 | } | |
382 | mutex_unlock(&minor_lock); | |
383 | return dev; | |
384 | } | |
385 | ||
20d29d7a AB |
386 | static int macvtap_newlink(struct net *src_net, |
387 | struct net_device *dev, | |
388 | struct nlattr *tb[], | |
389 | struct nlattr *data[]) | |
390 | { | |
815f236d | 391 | struct macvlan_dev *vlan = netdev_priv(dev); |
6acf54f1 VY |
392 | int err; |
393 | ||
815f236d JW |
394 | INIT_LIST_HEAD(&vlan->queue_list); |
395 | ||
2be5c767 VY |
396 | /* Since macvlan supports all offloads by default, make |
397 | * tap support all offloads also. | |
398 | */ | |
399 | vlan->tap_features = TUN_OFFLOADS; | |
400 | ||
6acf54f1 VY |
401 | err = netdev_rx_handler_register(dev, macvtap_handle_frame, vlan); |
402 | if (err) | |
403 | return err; | |
404 | ||
9bf1907f EB |
405 | /* Don't put anything that may fail after macvlan_common_newlink |
406 | * because we can't undo what it does. | |
407 | */ | |
2f6a1b66 | 408 | return macvlan_common_newlink(src_net, dev, tb, data); |
20d29d7a AB |
409 | } |
410 | ||
411 | static void macvtap_dellink(struct net_device *dev, | |
412 | struct list_head *head) | |
413 | { | |
6acf54f1 | 414 | netdev_rx_handler_unregister(dev); |
20d29d7a AB |
415 | macvtap_del_queues(dev); |
416 | macvlan_dellink(dev, head); | |
417 | } | |
418 | ||
8a35747a HX |
419 | static void macvtap_setup(struct net_device *dev) |
420 | { | |
421 | macvlan_common_setup(dev); | |
422 | dev->tx_queue_len = TUN_READQ_SIZE; | |
423 | } | |
424 | ||
20d29d7a AB |
425 | static struct rtnl_link_ops macvtap_link_ops __read_mostly = { |
426 | .kind = "macvtap", | |
8a35747a | 427 | .setup = macvtap_setup, |
20d29d7a AB |
428 | .newlink = macvtap_newlink, |
429 | .dellink = macvtap_dellink, | |
430 | }; | |
431 | ||
432 | ||
433 | static void macvtap_sock_write_space(struct sock *sk) | |
434 | { | |
43815482 ED |
435 | wait_queue_head_t *wqueue; |
436 | ||
20d29d7a AB |
437 | if (!sock_writeable(sk) || |
438 | !test_and_clear_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags)) | |
439 | return; | |
440 | ||
43815482 ED |
441 | wqueue = sk_sleep(sk); |
442 | if (wqueue && waitqueue_active(wqueue)) | |
443 | wake_up_interruptible_poll(wqueue, POLLOUT | POLLWRNORM | POLLWRBAND); | |
20d29d7a AB |
444 | } |
445 | ||
2259fef0 EB |
446 | static void macvtap_sock_destruct(struct sock *sk) |
447 | { | |
448 | skb_queue_purge(&sk->sk_receive_queue); | |
449 | } | |
450 | ||
20d29d7a AB |
451 | static int macvtap_open(struct inode *inode, struct file *file) |
452 | { | |
453 | struct net *net = current->nsproxy->net_ns; | |
40b8fe45 | 454 | struct net_device *dev; |
20d29d7a | 455 | struct macvtap_queue *q; |
40b8fe45 | 456 | int err = -ENODEV; |
20d29d7a | 457 | |
40b8fe45 VY |
458 | rtnl_lock(); |
459 | dev = dev_get_by_macvtap_minor(iminor(inode)); | |
20d29d7a AB |
460 | if (!dev) |
461 | goto out; | |
462 | ||
20d29d7a AB |
463 | err = -ENOMEM; |
464 | q = (struct macvtap_queue *)sk_alloc(net, AF_UNSPEC, GFP_KERNEL, | |
465 | &macvtap_proto); | |
466 | if (!q) | |
467 | goto out; | |
468 | ||
d9a90a31 | 469 | RCU_INIT_POINTER(q->sock.wq, &q->wq); |
43815482 | 470 | init_waitqueue_head(&q->wq.wait); |
20d29d7a AB |
471 | q->sock.type = SOCK_RAW; |
472 | q->sock.state = SS_CONNECTED; | |
501c774c AB |
473 | q->sock.file = file; |
474 | q->sock.ops = &macvtap_socket_ops; | |
20d29d7a | 475 | sock_init_data(&q->sock, &q->sk); |
20d29d7a | 476 | q->sk.sk_write_space = macvtap_sock_write_space; |
2259fef0 | 477 | q->sk.sk_destruct = macvtap_sock_destruct; |
b9fb9ee0 | 478 | q->flags = IFF_VNET_HDR | IFF_NO_PI | IFF_TAP; |
55afbd08 | 479 | q->vnet_hdr_sz = sizeof(struct virtio_net_hdr); |
20d29d7a | 480 | |
97bc3633 SM |
481 | /* |
482 | * so far only KVM virtio_net uses macvtap, enable zero copy between | |
483 | * guest kernel and host kernel when lower device supports zerocopy | |
047af9cf EB |
484 | * |
485 | * The macvlan supports zerocopy iff the lower device supports zero | |
486 | * copy so we don't have to look at the lower device directly. | |
97bc3633 | 487 | */ |
047af9cf EB |
488 | if ((dev->features & NETIF_F_HIGHDMA) && (dev->features & NETIF_F_SG)) |
489 | sock_set_flag(&q->sk, SOCK_ZEROCOPY); | |
97bc3633 | 490 | |
20d29d7a AB |
491 | err = macvtap_set_queue(dev, file, q); |
492 | if (err) | |
493 | sock_put(&q->sk); | |
494 | ||
495 | out: | |
496 | if (dev) | |
497 | dev_put(dev); | |
498 | ||
40b8fe45 | 499 | rtnl_unlock(); |
20d29d7a AB |
500 | return err; |
501 | } | |
502 | ||
503 | static int macvtap_release(struct inode *inode, struct file *file) | |
504 | { | |
02df55d2 AB |
505 | struct macvtap_queue *q = file->private_data; |
506 | macvtap_put_queue(q); | |
20d29d7a AB |
507 | return 0; |
508 | } | |
509 | ||
510 | static unsigned int macvtap_poll(struct file *file, poll_table * wait) | |
511 | { | |
02df55d2 | 512 | struct macvtap_queue *q = file->private_data; |
20d29d7a AB |
513 | unsigned int mask = POLLERR; |
514 | ||
515 | if (!q) | |
516 | goto out; | |
517 | ||
518 | mask = 0; | |
43815482 | 519 | poll_wait(file, &q->wq.wait, wait); |
20d29d7a AB |
520 | |
521 | if (!skb_queue_empty(&q->sk.sk_receive_queue)) | |
522 | mask |= POLLIN | POLLRDNORM; | |
523 | ||
524 | if (sock_writeable(&q->sk) || | |
525 | (!test_and_set_bit(SOCK_ASYNC_NOSPACE, &q->sock.flags) && | |
526 | sock_writeable(&q->sk))) | |
527 | mask |= POLLOUT | POLLWRNORM; | |
528 | ||
529 | out: | |
20d29d7a AB |
530 | return mask; |
531 | } | |
532 | ||
b9fb9ee0 AB |
533 | static inline struct sk_buff *macvtap_alloc_skb(struct sock *sk, size_t prepad, |
534 | size_t len, size_t linear, | |
535 | int noblock, int *err) | |
536 | { | |
537 | struct sk_buff *skb; | |
538 | ||
539 | /* Under a page? Don't bother with paged skb. */ | |
540 | if (prepad + len < PAGE_SIZE || !linear) | |
541 | linear = len; | |
542 | ||
543 | skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock, | |
28d64271 | 544 | err, 0); |
b9fb9ee0 AB |
545 | if (!skb) |
546 | return NULL; | |
547 | ||
548 | skb_reserve(skb, prepad); | |
549 | skb_put(skb, linear); | |
550 | skb->data_len = len - linear; | |
551 | skb->len += len - linear; | |
552 | ||
553 | return skb; | |
554 | } | |
555 | ||
556 | /* | |
557 | * macvtap_skb_from_vnet_hdr and macvtap_skb_to_vnet_hdr should | |
558 | * be shared with the tun/tap driver. | |
559 | */ | |
560 | static int macvtap_skb_from_vnet_hdr(struct sk_buff *skb, | |
561 | struct virtio_net_hdr *vnet_hdr) | |
562 | { | |
563 | unsigned short gso_type = 0; | |
564 | if (vnet_hdr->gso_type != VIRTIO_NET_HDR_GSO_NONE) { | |
565 | switch (vnet_hdr->gso_type & ~VIRTIO_NET_HDR_GSO_ECN) { | |
566 | case VIRTIO_NET_HDR_GSO_TCPV4: | |
567 | gso_type = SKB_GSO_TCPV4; | |
568 | break; | |
569 | case VIRTIO_NET_HDR_GSO_TCPV6: | |
570 | gso_type = SKB_GSO_TCPV6; | |
571 | break; | |
572 | case VIRTIO_NET_HDR_GSO_UDP: | |
3d0ad094 BH |
573 | pr_warn_once("macvtap: %s: using disabled UFO feature; please fix this program\n", |
574 | current->comm); | |
b9fb9ee0 | 575 | gso_type = SKB_GSO_UDP; |
5188cd44 BH |
576 | if (skb->protocol == htons(ETH_P_IPV6)) |
577 | ipv6_proxy_select_ident(skb); | |
b9fb9ee0 AB |
578 | break; |
579 | default: | |
580 | return -EINVAL; | |
581 | } | |
582 | ||
583 | if (vnet_hdr->gso_type & VIRTIO_NET_HDR_GSO_ECN) | |
584 | gso_type |= SKB_GSO_TCP_ECN; | |
585 | ||
586 | if (vnet_hdr->gso_size == 0) | |
587 | return -EINVAL; | |
588 | } | |
589 | ||
590 | if (vnet_hdr->flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) { | |
591 | if (!skb_partial_csum_set(skb, vnet_hdr->csum_start, | |
592 | vnet_hdr->csum_offset)) | |
593 | return -EINVAL; | |
594 | } | |
595 | ||
596 | if (vnet_hdr->gso_type != VIRTIO_NET_HDR_GSO_NONE) { | |
597 | skb_shinfo(skb)->gso_size = vnet_hdr->gso_size; | |
c9af6db4 | 598 | skb_shinfo(skb)->gso_type = gso_type; |
b9fb9ee0 AB |
599 | |
600 | /* Header must be checked, and gso_segs computed. */ | |
601 | skb_shinfo(skb)->gso_type |= SKB_GSO_DODGY; | |
602 | skb_shinfo(skb)->gso_segs = 0; | |
603 | } | |
604 | return 0; | |
605 | } | |
606 | ||
359d44d7 | 607 | static void macvtap_skb_to_vnet_hdr(const struct sk_buff *skb, |
b9fb9ee0 AB |
608 | struct virtio_net_hdr *vnet_hdr) |
609 | { | |
610 | memset(vnet_hdr, 0, sizeof(*vnet_hdr)); | |
611 | ||
612 | if (skb_is_gso(skb)) { | |
613 | struct skb_shared_info *sinfo = skb_shinfo(skb); | |
614 | ||
615 | /* This is a hint as to how much should be linear. */ | |
616 | vnet_hdr->hdr_len = skb_headlen(skb); | |
617 | vnet_hdr->gso_size = sinfo->gso_size; | |
618 | if (sinfo->gso_type & SKB_GSO_TCPV4) | |
619 | vnet_hdr->gso_type = VIRTIO_NET_HDR_GSO_TCPV4; | |
620 | else if (sinfo->gso_type & SKB_GSO_TCPV6) | |
621 | vnet_hdr->gso_type = VIRTIO_NET_HDR_GSO_TCPV6; | |
b9fb9ee0 AB |
622 | else |
623 | BUG(); | |
624 | if (sinfo->gso_type & SKB_GSO_TCP_ECN) | |
625 | vnet_hdr->gso_type |= VIRTIO_NET_HDR_GSO_ECN; | |
626 | } else | |
627 | vnet_hdr->gso_type = VIRTIO_NET_HDR_GSO_NONE; | |
628 | ||
629 | if (skb->ip_summed == CHECKSUM_PARTIAL) { | |
630 | vnet_hdr->flags = VIRTIO_NET_HDR_F_NEEDS_CSUM; | |
55508d60 | 631 | vnet_hdr->csum_start = skb_checksum_start_offset(skb); |
3ce9b20f HX |
632 | if (vlan_tx_tag_present(skb)) |
633 | vnet_hdr->csum_start += VLAN_HLEN; | |
b9fb9ee0 | 634 | vnet_hdr->csum_offset = skb->csum_offset; |
10a8d94a JW |
635 | } else if (skb->ip_summed == CHECKSUM_UNNECESSARY) { |
636 | vnet_hdr->flags = VIRTIO_NET_HDR_F_DATA_VALID; | |
b9fb9ee0 | 637 | } /* else everything is zero */ |
b9fb9ee0 AB |
638 | } |
639 | ||
20d29d7a | 640 | /* Get packet from user space buffer */ |
97bc3633 SM |
641 | static ssize_t macvtap_get_user(struct macvtap_queue *q, struct msghdr *m, |
642 | const struct iovec *iv, unsigned long total_len, | |
643 | size_t count, int noblock) | |
20d29d7a | 644 | { |
16a3fa28 | 645 | int good_linear = SKB_MAX_HEAD(NET_IP_ALIGN); |
20d29d7a | 646 | struct sk_buff *skb; |
02df55d2 | 647 | struct macvlan_dev *vlan; |
97bc3633 | 648 | unsigned long len = total_len; |
20d29d7a | 649 | int err; |
b9fb9ee0 AB |
650 | struct virtio_net_hdr vnet_hdr = { 0 }; |
651 | int vnet_hdr_len = 0; | |
b92946e2 | 652 | int copylen = 0; |
97bc3633 | 653 | bool zerocopy = false; |
61d46bf9 | 654 | size_t linear; |
b9fb9ee0 AB |
655 | |
656 | if (q->flags & IFF_VNET_HDR) { | |
55afbd08 | 657 | vnet_hdr_len = q->vnet_hdr_sz; |
b9fb9ee0 AB |
658 | |
659 | err = -EINVAL; | |
ce3c8692 | 660 | if (len < vnet_hdr_len) |
b9fb9ee0 | 661 | goto err; |
ce3c8692 | 662 | len -= vnet_hdr_len; |
b9fb9ee0 AB |
663 | |
664 | err = memcpy_fromiovecend((void *)&vnet_hdr, iv, 0, | |
55afbd08 | 665 | sizeof(vnet_hdr)); |
b9fb9ee0 AB |
666 | if (err < 0) |
667 | goto err; | |
668 | if ((vnet_hdr.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) && | |
669 | vnet_hdr.csum_start + vnet_hdr.csum_offset + 2 > | |
670 | vnet_hdr.hdr_len) | |
671 | vnet_hdr.hdr_len = vnet_hdr.csum_start + | |
672 | vnet_hdr.csum_offset + 2; | |
673 | err = -EINVAL; | |
674 | if (vnet_hdr.hdr_len > len) | |
675 | goto err; | |
676 | } | |
20d29d7a | 677 | |
b9fb9ee0 | 678 | err = -EINVAL; |
20d29d7a | 679 | if (unlikely(len < ETH_HLEN)) |
b9fb9ee0 | 680 | goto err; |
20d29d7a | 681 | |
b92946e2 JW |
682 | err = -EMSGSIZE; |
683 | if (unlikely(count > UIO_MAXIOV)) | |
684 | goto err; | |
685 | ||
ece793fc JW |
686 | if (m && m->msg_control && sock_flag(&q->sk, SOCK_ZEROCOPY)) { |
687 | copylen = vnet_hdr.hdr_len ? vnet_hdr.hdr_len : GOODCOPY_LEN; | |
16a3fa28 JW |
688 | if (copylen > good_linear) |
689 | copylen = good_linear; | |
61d46bf9 | 690 | linear = copylen; |
ece793fc JW |
691 | if (iov_pages(iv, vnet_hdr_len + copylen, count) |
692 | <= MAX_SKB_FRAGS) | |
693 | zerocopy = true; | |
694 | } | |
695 | ||
696 | if (!zerocopy) { | |
97bc3633 | 697 | copylen = len; |
16a3fa28 JW |
698 | if (vnet_hdr.hdr_len > good_linear) |
699 | linear = good_linear; | |
700 | else | |
701 | linear = vnet_hdr.hdr_len; | |
61d46bf9 | 702 | } |
97bc3633 SM |
703 | |
704 | skb = macvtap_alloc_skb(&q->sk, NET_IP_ALIGN, copylen, | |
61d46bf9 | 705 | linear, noblock, &err); |
02df55d2 AB |
706 | if (!skb) |
707 | goto err; | |
20d29d7a | 708 | |
01d6657b | 709 | if (zerocopy) |
97bc3633 | 710 | err = zerocopy_sg_from_iovec(skb, iv, vnet_hdr_len, count); |
ece793fc | 711 | else { |
97bc3633 SM |
712 | err = skb_copy_datagram_from_iovec(skb, 0, iv, vnet_hdr_len, |
713 | len); | |
ece793fc JW |
714 | if (!err && m && m->msg_control) { |
715 | struct ubuf_info *uarg = m->msg_control; | |
716 | uarg->callback(uarg, false); | |
717 | } | |
718 | } | |
719 | ||
02df55d2 | 720 | if (err) |
b9fb9ee0 | 721 | goto err_kfree; |
20d29d7a AB |
722 | |
723 | skb_set_network_header(skb, ETH_HLEN); | |
b9fb9ee0 AB |
724 | skb_reset_mac_header(skb); |
725 | skb->protocol = eth_hdr(skb)->h_proto; | |
726 | ||
727 | if (vnet_hdr_len) { | |
728 | err = macvtap_skb_from_vnet_hdr(skb, &vnet_hdr); | |
729 | if (err) | |
730 | goto err_kfree; | |
731 | } | |
732 | ||
40893fd0 | 733 | skb_probe_transport_header(skb, ETH_HLEN); |
9b4d669b | 734 | |
ac4e4af1 VY |
735 | rcu_read_lock(); |
736 | vlan = rcu_dereference(q->vlan); | |
97bc3633 | 737 | /* copy skb_ubuf_info for callback when skb has no error */ |
01d6657b | 738 | if (zerocopy) { |
97bc3633 | 739 | skb_shinfo(skb)->destructor_arg = m->msg_control; |
01d6657b | 740 | skb_shinfo(skb)->tx_flags |= SKBTX_DEV_ZEROCOPY; |
c9af6db4 | 741 | skb_shinfo(skb)->tx_flags |= SKBTX_SHARED_FRAG; |
01d6657b | 742 | } |
29d79196 | 743 | if (vlan) { |
6acf54f1 VY |
744 | skb->dev = vlan->dev; |
745 | dev_queue_xmit(skb); | |
29d79196 | 746 | } else { |
02df55d2 | 747 | kfree_skb(skb); |
29d79196 | 748 | } |
ac4e4af1 | 749 | rcu_read_unlock(); |
20d29d7a | 750 | |
97bc3633 | 751 | return total_len; |
02df55d2 | 752 | |
b9fb9ee0 AB |
753 | err_kfree: |
754 | kfree_skb(skb); | |
755 | ||
02df55d2 | 756 | err: |
ac4e4af1 VY |
757 | rcu_read_lock(); |
758 | vlan = rcu_dereference(q->vlan); | |
02df55d2 | 759 | if (vlan) |
cd3e22b7 | 760 | this_cpu_inc(vlan->pcpu_stats->tx_dropped); |
ac4e4af1 | 761 | rcu_read_unlock(); |
02df55d2 | 762 | |
02df55d2 | 763 | return err; |
20d29d7a AB |
764 | } |
765 | ||
766 | static ssize_t macvtap_aio_write(struct kiocb *iocb, const struct iovec *iv, | |
767 | unsigned long count, loff_t pos) | |
768 | { | |
769 | struct file *file = iocb->ki_filp; | |
770 | ssize_t result = -ENOLINK; | |
02df55d2 | 771 | struct macvtap_queue *q = file->private_data; |
20d29d7a | 772 | |
97bc3633 SM |
773 | result = macvtap_get_user(q, NULL, iv, iov_length(iv, count), count, |
774 | file->f_flags & O_NONBLOCK); | |
20d29d7a AB |
775 | return result; |
776 | } | |
777 | ||
778 | /* Put packet to the user space buffer */ | |
779 | static ssize_t macvtap_put_user(struct macvtap_queue *q, | |
780 | const struct sk_buff *skb, | |
781 | const struct iovec *iv, int len) | |
782 | { | |
20d29d7a | 783 | int ret; |
b9fb9ee0 | 784 | int vnet_hdr_len = 0; |
f09e2249 | 785 | int vlan_offset = 0; |
ce232ce0 | 786 | int copied, total; |
b9fb9ee0 AB |
787 | |
788 | if (q->flags & IFF_VNET_HDR) { | |
789 | struct virtio_net_hdr vnet_hdr; | |
55afbd08 | 790 | vnet_hdr_len = q->vnet_hdr_sz; |
b9fb9ee0 AB |
791 | if ((len -= vnet_hdr_len) < 0) |
792 | return -EINVAL; | |
793 | ||
359d44d7 | 794 | macvtap_skb_to_vnet_hdr(skb, &vnet_hdr); |
b9fb9ee0 | 795 | |
55afbd08 | 796 | if (memcpy_toiovecend(iv, (void *)&vnet_hdr, 0, sizeof(vnet_hdr))) |
b9fb9ee0 AB |
797 | return -EFAULT; |
798 | } | |
ce232ce0 JW |
799 | total = copied = vnet_hdr_len; |
800 | total += skb->len; | |
f09e2249 BG |
801 | |
802 | if (!vlan_tx_tag_present(skb)) | |
803 | len = min_t(int, skb->len, len); | |
804 | else { | |
805 | int copy; | |
806 | struct { | |
807 | __be16 h_vlan_proto; | |
808 | __be16 h_vlan_TCI; | |
809 | } veth; | |
0fbe0d47 | 810 | veth.h_vlan_proto = skb->vlan_proto; |
f09e2249 BG |
811 | veth.h_vlan_TCI = htons(vlan_tx_tag_get(skb)); |
812 | ||
813 | vlan_offset = offsetof(struct vlan_ethhdr, h_vlan_proto); | |
814 | len = min_t(int, skb->len + VLAN_HLEN, len); | |
ce232ce0 | 815 | total += VLAN_HLEN; |
f09e2249 BG |
816 | |
817 | copy = min_t(int, vlan_offset, len); | |
818 | ret = skb_copy_datagram_const_iovec(skb, 0, iv, copied, copy); | |
819 | len -= copy; | |
820 | copied += copy; | |
821 | if (ret || !len) | |
822 | goto done; | |
823 | ||
824 | copy = min_t(int, sizeof(veth), len); | |
825 | ret = memcpy_toiovecend(iv, (void *)&veth, copied, copy); | |
826 | len -= copy; | |
827 | copied += copy; | |
828 | if (ret || !len) | |
829 | goto done; | |
830 | } | |
20d29d7a | 831 | |
f09e2249 | 832 | ret = skb_copy_datagram_const_iovec(skb, vlan_offset, iv, copied, len); |
20d29d7a | 833 | |
f09e2249 | 834 | done: |
ce232ce0 | 835 | return ret ? ret : total; |
20d29d7a AB |
836 | } |
837 | ||
55ec8e25 | 838 | static ssize_t macvtap_do_read(struct macvtap_queue *q, |
501c774c AB |
839 | const struct iovec *iv, unsigned long len, |
840 | int noblock) | |
20d29d7a | 841 | { |
ccf7e72b | 842 | DEFINE_WAIT(wait); |
20d29d7a | 843 | struct sk_buff *skb; |
501c774c | 844 | ssize_t ret = 0; |
20d29d7a | 845 | |
20d29d7a | 846 | while (len) { |
89cee917 JW |
847 | if (!noblock) |
848 | prepare_to_wait(sk_sleep(&q->sk), &wait, | |
849 | TASK_INTERRUPTIBLE); | |
20d29d7a AB |
850 | |
851 | /* Read frames from the queue */ | |
852 | skb = skb_dequeue(&q->sk.sk_receive_queue); | |
853 | if (!skb) { | |
501c774c | 854 | if (noblock) { |
20d29d7a AB |
855 | ret = -EAGAIN; |
856 | break; | |
857 | } | |
858 | if (signal_pending(current)) { | |
859 | ret = -ERESTARTSYS; | |
860 | break; | |
861 | } | |
862 | /* Nothing to read, let's sleep */ | |
863 | schedule(); | |
864 | continue; | |
865 | } | |
866 | ret = macvtap_put_user(q, skb, iv, len); | |
867 | kfree_skb(skb); | |
868 | break; | |
869 | } | |
870 | ||
89cee917 JW |
871 | if (!noblock) |
872 | finish_wait(sk_sleep(&q->sk), &wait); | |
501c774c AB |
873 | return ret; |
874 | } | |
875 | ||
876 | static ssize_t macvtap_aio_read(struct kiocb *iocb, const struct iovec *iv, | |
877 | unsigned long count, loff_t pos) | |
878 | { | |
879 | struct file *file = iocb->ki_filp; | |
880 | struct macvtap_queue *q = file->private_data; | |
881 | ssize_t len, ret = 0; | |
20d29d7a | 882 | |
501c774c AB |
883 | len = iov_length(iv, count); |
884 | if (len < 0) { | |
885 | ret = -EINVAL; | |
886 | goto out; | |
887 | } | |
888 | ||
55ec8e25 | 889 | ret = macvtap_do_read(q, iv, len, file->f_flags & O_NONBLOCK); |
ce232ce0 | 890 | ret = min_t(ssize_t, ret, len); |
e6ebc7f1 ZYW |
891 | if (ret > 0) |
892 | iocb->ki_pos = ret; | |
20d29d7a | 893 | out: |
20d29d7a AB |
894 | return ret; |
895 | } | |
896 | ||
8f475a31 JW |
897 | static struct macvlan_dev *macvtap_get_vlan(struct macvtap_queue *q) |
898 | { | |
899 | struct macvlan_dev *vlan; | |
900 | ||
441ac0fc VY |
901 | ASSERT_RTNL(); |
902 | vlan = rtnl_dereference(q->vlan); | |
8f475a31 JW |
903 | if (vlan) |
904 | dev_hold(vlan->dev); | |
8f475a31 JW |
905 | |
906 | return vlan; | |
907 | } | |
908 | ||
909 | static void macvtap_put_vlan(struct macvlan_dev *vlan) | |
910 | { | |
911 | dev_put(vlan->dev); | |
912 | } | |
913 | ||
815f236d JW |
914 | static int macvtap_ioctl_set_queue(struct file *file, unsigned int flags) |
915 | { | |
916 | struct macvtap_queue *q = file->private_data; | |
917 | struct macvlan_dev *vlan; | |
918 | int ret; | |
919 | ||
920 | vlan = macvtap_get_vlan(q); | |
921 | if (!vlan) | |
922 | return -EINVAL; | |
923 | ||
924 | if (flags & IFF_ATTACH_QUEUE) | |
925 | ret = macvtap_enable_queue(vlan->dev, file, q); | |
926 | else if (flags & IFF_DETACH_QUEUE) | |
927 | ret = macvtap_disable_queue(q); | |
f57855a5 JW |
928 | else |
929 | ret = -EINVAL; | |
815f236d JW |
930 | |
931 | macvtap_put_vlan(vlan); | |
932 | return ret; | |
933 | } | |
934 | ||
2be5c767 VY |
935 | static int set_offload(struct macvtap_queue *q, unsigned long arg) |
936 | { | |
937 | struct macvlan_dev *vlan; | |
938 | netdev_features_t features; | |
939 | netdev_features_t feature_mask = 0; | |
940 | ||
941 | vlan = rtnl_dereference(q->vlan); | |
942 | if (!vlan) | |
943 | return -ENOLINK; | |
944 | ||
945 | features = vlan->dev->features; | |
946 | ||
947 | if (arg & TUN_F_CSUM) { | |
948 | feature_mask = NETIF_F_HW_CSUM; | |
949 | ||
950 | if (arg & (TUN_F_TSO4 | TUN_F_TSO6)) { | |
951 | if (arg & TUN_F_TSO_ECN) | |
952 | feature_mask |= NETIF_F_TSO_ECN; | |
953 | if (arg & TUN_F_TSO4) | |
954 | feature_mask |= NETIF_F_TSO; | |
955 | if (arg & TUN_F_TSO6) | |
956 | feature_mask |= NETIF_F_TSO6; | |
957 | } | |
2be5c767 VY |
958 | } |
959 | ||
960 | /* tun/tap driver inverts the usage for TSO offloads, where | |
961 | * setting the TSO bit means that the userspace wants to | |
962 | * accept TSO frames and turning it off means that user space | |
963 | * does not support TSO. | |
964 | * For macvtap, we have to invert it to mean the same thing. | |
965 | * When user space turns off TSO, we turn off GSO/LRO so that | |
966 | * user-space will not receive TSO frames. | |
967 | */ | |
3d0ad094 | 968 | if (feature_mask & (NETIF_F_TSO | NETIF_F_TSO6)) |
2be5c767 VY |
969 | features |= RX_OFFLOADS; |
970 | else | |
971 | features &= ~RX_OFFLOADS; | |
972 | ||
973 | /* tap_features are the same as features on tun/tap and | |
974 | * reflect user expectations. | |
975 | */ | |
a567dd62 | 976 | vlan->tap_features = feature_mask; |
2be5c767 VY |
977 | vlan->set_features = features; |
978 | netdev_update_features(vlan->dev); | |
979 | ||
980 | return 0; | |
981 | } | |
982 | ||
20d29d7a AB |
983 | /* |
984 | * provide compatibility with generic tun/tap interface | |
985 | */ | |
986 | static long macvtap_ioctl(struct file *file, unsigned int cmd, | |
987 | unsigned long arg) | |
988 | { | |
02df55d2 AB |
989 | struct macvtap_queue *q = file->private_data; |
990 | struct macvlan_dev *vlan; | |
20d29d7a AB |
991 | void __user *argp = (void __user *)arg; |
992 | struct ifreq __user *ifr = argp; | |
993 | unsigned int __user *up = argp; | |
994 | unsigned int u; | |
55afbd08 MT |
995 | int __user *sp = argp; |
996 | int s; | |
02df55d2 | 997 | int ret; |
20d29d7a AB |
998 | |
999 | switch (cmd) { | |
1000 | case TUNSETIFF: | |
1001 | /* ignore the name, just look at flags */ | |
1002 | if (get_user(u, &ifr->ifr_flags)) | |
1003 | return -EFAULT; | |
b9fb9ee0 AB |
1004 | |
1005 | ret = 0; | |
815f236d JW |
1006 | if ((u & ~(IFF_VNET_HDR | IFF_MULTI_QUEUE)) != |
1007 | (IFF_NO_PI | IFF_TAP)) | |
b9fb9ee0 AB |
1008 | ret = -EINVAL; |
1009 | else | |
1010 | q->flags = u; | |
1011 | ||
1012 | return ret; | |
20d29d7a AB |
1013 | |
1014 | case TUNGETIFF: | |
441ac0fc | 1015 | rtnl_lock(); |
8f475a31 | 1016 | vlan = macvtap_get_vlan(q); |
441ac0fc VY |
1017 | if (!vlan) { |
1018 | rtnl_unlock(); | |
20d29d7a | 1019 | return -ENOLINK; |
441ac0fc | 1020 | } |
20d29d7a | 1021 | |
02df55d2 | 1022 | ret = 0; |
13707f9e | 1023 | if (copy_to_user(&ifr->ifr_name, vlan->dev->name, IFNAMSIZ) || |
b9fb9ee0 | 1024 | put_user(q->flags, &ifr->ifr_flags)) |
02df55d2 | 1025 | ret = -EFAULT; |
8f475a31 | 1026 | macvtap_put_vlan(vlan); |
441ac0fc | 1027 | rtnl_unlock(); |
02df55d2 | 1028 | return ret; |
20d29d7a | 1029 | |
815f236d JW |
1030 | case TUNSETQUEUE: |
1031 | if (get_user(u, &ifr->ifr_flags)) | |
1032 | return -EFAULT; | |
441ac0fc VY |
1033 | rtnl_lock(); |
1034 | ret = macvtap_ioctl_set_queue(file, u); | |
1035 | rtnl_unlock(); | |
82a19eb8 | 1036 | return ret; |
815f236d | 1037 | |
20d29d7a | 1038 | case TUNGETFEATURES: |
df09b36f JW |
1039 | if (put_user(IFF_TAP | IFF_NO_PI | IFF_VNET_HDR | |
1040 | IFF_MULTI_QUEUE, up)) | |
20d29d7a AB |
1041 | return -EFAULT; |
1042 | return 0; | |
1043 | ||
1044 | case TUNSETSNDBUF: | |
1045 | if (get_user(u, up)) | |
1046 | return -EFAULT; | |
1047 | ||
20d29d7a | 1048 | q->sk.sk_sndbuf = u; |
20d29d7a AB |
1049 | return 0; |
1050 | ||
55afbd08 MT |
1051 | case TUNGETVNETHDRSZ: |
1052 | s = q->vnet_hdr_sz; | |
1053 | if (put_user(s, sp)) | |
1054 | return -EFAULT; | |
1055 | return 0; | |
1056 | ||
1057 | case TUNSETVNETHDRSZ: | |
1058 | if (get_user(s, sp)) | |
1059 | return -EFAULT; | |
1060 | if (s < (int)sizeof(struct virtio_net_hdr)) | |
1061 | return -EINVAL; | |
1062 | ||
1063 | q->vnet_hdr_sz = s; | |
1064 | return 0; | |
1065 | ||
20d29d7a AB |
1066 | case TUNSETOFFLOAD: |
1067 | /* let the user check for future flags */ | |
1068 | if (arg & ~(TUN_F_CSUM | TUN_F_TSO4 | TUN_F_TSO6 | | |
3d0ad094 | 1069 | TUN_F_TSO_ECN)) |
20d29d7a AB |
1070 | return -EINVAL; |
1071 | ||
2be5c767 VY |
1072 | rtnl_lock(); |
1073 | ret = set_offload(q, arg); | |
1074 | rtnl_unlock(); | |
1075 | return ret; | |
20d29d7a AB |
1076 | |
1077 | default: | |
1078 | return -EINVAL; | |
1079 | } | |
1080 | } | |
1081 | ||
1082 | #ifdef CONFIG_COMPAT | |
1083 | static long macvtap_compat_ioctl(struct file *file, unsigned int cmd, | |
1084 | unsigned long arg) | |
1085 | { | |
1086 | return macvtap_ioctl(file, cmd, (unsigned long)compat_ptr(arg)); | |
1087 | } | |
1088 | #endif | |
1089 | ||
1090 | static const struct file_operations macvtap_fops = { | |
1091 | .owner = THIS_MODULE, | |
1092 | .open = macvtap_open, | |
1093 | .release = macvtap_release, | |
1094 | .aio_read = macvtap_aio_read, | |
1095 | .aio_write = macvtap_aio_write, | |
1096 | .poll = macvtap_poll, | |
1097 | .llseek = no_llseek, | |
1098 | .unlocked_ioctl = macvtap_ioctl, | |
1099 | #ifdef CONFIG_COMPAT | |
1100 | .compat_ioctl = macvtap_compat_ioctl, | |
1101 | #endif | |
1102 | }; | |
1103 | ||
501c774c AB |
1104 | static int macvtap_sendmsg(struct kiocb *iocb, struct socket *sock, |
1105 | struct msghdr *m, size_t total_len) | |
1106 | { | |
1107 | struct macvtap_queue *q = container_of(sock, struct macvtap_queue, sock); | |
97bc3633 | 1108 | return macvtap_get_user(q, m, m->msg_iov, total_len, m->msg_iovlen, |
501c774c AB |
1109 | m->msg_flags & MSG_DONTWAIT); |
1110 | } | |
1111 | ||
1112 | static int macvtap_recvmsg(struct kiocb *iocb, struct socket *sock, | |
1113 | struct msghdr *m, size_t total_len, | |
1114 | int flags) | |
1115 | { | |
1116 | struct macvtap_queue *q = container_of(sock, struct macvtap_queue, sock); | |
1117 | int ret; | |
1118 | if (flags & ~(MSG_DONTWAIT|MSG_TRUNC)) | |
1119 | return -EINVAL; | |
55ec8e25 | 1120 | ret = macvtap_do_read(q, m->msg_iov, total_len, |
501c774c | 1121 | flags & MSG_DONTWAIT); |
de2aa476 DM |
1122 | if (ret > total_len) { |
1123 | m->msg_flags |= MSG_TRUNC; | |
1124 | ret = flags & MSG_TRUNC ? ret : total_len; | |
1125 | } | |
501c774c AB |
1126 | return ret; |
1127 | } | |
1128 | ||
1129 | /* Ops structure to mimic raw sockets with tun */ | |
1130 | static const struct proto_ops macvtap_socket_ops = { | |
1131 | .sendmsg = macvtap_sendmsg, | |
1132 | .recvmsg = macvtap_recvmsg, | |
1133 | }; | |
1134 | ||
1135 | /* Get an underlying socket object from tun file. Returns error unless file is | |
1136 | * attached to a device. The returned object works like a packet socket, it | |
1137 | * can be used for sock_sendmsg/sock_recvmsg. The caller is responsible for | |
1138 | * holding a reference to the file for as long as the socket is in use. */ | |
1139 | struct socket *macvtap_get_socket(struct file *file) | |
1140 | { | |
1141 | struct macvtap_queue *q; | |
1142 | if (file->f_op != &macvtap_fops) | |
1143 | return ERR_PTR(-EINVAL); | |
1144 | q = file->private_data; | |
1145 | if (!q) | |
1146 | return ERR_PTR(-EBADFD); | |
1147 | return &q->sock; | |
1148 | } | |
1149 | EXPORT_SYMBOL_GPL(macvtap_get_socket); | |
1150 | ||
9bf1907f EB |
1151 | static int macvtap_device_event(struct notifier_block *unused, |
1152 | unsigned long event, void *ptr) | |
1153 | { | |
351638e7 | 1154 | struct net_device *dev = netdev_notifier_info_to_dev(ptr); |
e09eff7f | 1155 | struct macvlan_dev *vlan; |
9bf1907f EB |
1156 | struct device *classdev; |
1157 | dev_t devt; | |
e09eff7f | 1158 | int err; |
9bf1907f EB |
1159 | |
1160 | if (dev->rtnl_link_ops != &macvtap_link_ops) | |
1161 | return NOTIFY_DONE; | |
1162 | ||
e09eff7f | 1163 | vlan = netdev_priv(dev); |
9bf1907f EB |
1164 | |
1165 | switch (event) { | |
1166 | case NETDEV_REGISTER: | |
1167 | /* Create the device node here after the network device has | |
1168 | * been registered but before register_netdevice has | |
1169 | * finished running. | |
1170 | */ | |
e09eff7f EB |
1171 | err = macvtap_get_minor(vlan); |
1172 | if (err) | |
1173 | return notifier_from_errno(err); | |
1174 | ||
1175 | devt = MKDEV(MAJOR(macvtap_major), vlan->minor); | |
9bf1907f EB |
1176 | classdev = device_create(macvtap_class, &dev->dev, devt, |
1177 | dev, "tap%d", dev->ifindex); | |
e09eff7f EB |
1178 | if (IS_ERR(classdev)) { |
1179 | macvtap_free_minor(vlan); | |
9bf1907f | 1180 | return notifier_from_errno(PTR_ERR(classdev)); |
e09eff7f | 1181 | } |
9bf1907f EB |
1182 | break; |
1183 | case NETDEV_UNREGISTER: | |
e09eff7f | 1184 | devt = MKDEV(MAJOR(macvtap_major), vlan->minor); |
9bf1907f | 1185 | device_destroy(macvtap_class, devt); |
e09eff7f | 1186 | macvtap_free_minor(vlan); |
9bf1907f EB |
1187 | break; |
1188 | } | |
1189 | ||
1190 | return NOTIFY_DONE; | |
1191 | } | |
1192 | ||
1193 | static struct notifier_block macvtap_notifier_block __read_mostly = { | |
1194 | .notifier_call = macvtap_device_event, | |
1195 | }; | |
1196 | ||
20d29d7a AB |
1197 | static int macvtap_init(void) |
1198 | { | |
1199 | int err; | |
1200 | ||
1201 | err = alloc_chrdev_region(&macvtap_major, 0, | |
1202 | MACVTAP_NUM_DEVS, "macvtap"); | |
1203 | if (err) | |
1204 | goto out1; | |
1205 | ||
1206 | cdev_init(&macvtap_cdev, &macvtap_fops); | |
1207 | err = cdev_add(&macvtap_cdev, macvtap_major, MACVTAP_NUM_DEVS); | |
1208 | if (err) | |
1209 | goto out2; | |
1210 | ||
1211 | macvtap_class = class_create(THIS_MODULE, "macvtap"); | |
1212 | if (IS_ERR(macvtap_class)) { | |
1213 | err = PTR_ERR(macvtap_class); | |
1214 | goto out3; | |
1215 | } | |
1216 | ||
9bf1907f | 1217 | err = register_netdevice_notifier(&macvtap_notifier_block); |
20d29d7a AB |
1218 | if (err) |
1219 | goto out4; | |
1220 | ||
9bf1907f EB |
1221 | err = macvlan_link_register(&macvtap_link_ops); |
1222 | if (err) | |
1223 | goto out5; | |
1224 | ||
20d29d7a AB |
1225 | return 0; |
1226 | ||
9bf1907f EB |
1227 | out5: |
1228 | unregister_netdevice_notifier(&macvtap_notifier_block); | |
20d29d7a AB |
1229 | out4: |
1230 | class_unregister(macvtap_class); | |
1231 | out3: | |
1232 | cdev_del(&macvtap_cdev); | |
1233 | out2: | |
1234 | unregister_chrdev_region(macvtap_major, MACVTAP_NUM_DEVS); | |
1235 | out1: | |
1236 | return err; | |
1237 | } | |
1238 | module_init(macvtap_init); | |
1239 | ||
1240 | static void macvtap_exit(void) | |
1241 | { | |
1242 | rtnl_link_unregister(&macvtap_link_ops); | |
9bf1907f | 1243 | unregister_netdevice_notifier(&macvtap_notifier_block); |
20d29d7a AB |
1244 | class_unregister(macvtap_class); |
1245 | cdev_del(&macvtap_cdev); | |
1246 | unregister_chrdev_region(macvtap_major, MACVTAP_NUM_DEVS); | |
1247 | } | |
1248 | module_exit(macvtap_exit); | |
1249 | ||
1250 | MODULE_ALIAS_RTNL_LINK("macvtap"); | |
1251 | MODULE_AUTHOR("Arnd Bergmann <arnd@arndb.de>"); | |
1252 | MODULE_LICENSE("GPL"); |