]>
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/init.h> |
15 | #include <linux/wait.h> | |
16 | #include <linux/cdev.h> | |
40401530 | 17 | #include <linux/idr.h> |
20d29d7a AB |
18 | #include <linux/fs.h> |
19 | ||
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 | * | |
34 | * TODO: multiqueue support is currently not implemented, even though | |
35 | * macvtap is basically prepared for that. We will need to add this | |
36 | * here as well as in virtio-net and qemu to get line rate on 10gbit | |
37 | * adapters from a guest. | |
38 | */ | |
39 | struct macvtap_queue { | |
40 | struct sock sk; | |
41 | struct socket sock; | |
43815482 | 42 | struct socket_wq wq; |
55afbd08 | 43 | int vnet_hdr_sz; |
13707f9e | 44 | struct macvlan_dev __rcu *vlan; |
20d29d7a | 45 | struct file *file; |
b9fb9ee0 | 46 | unsigned int flags; |
376b1aab | 47 | u16 queue_index; |
815f236d JW |
48 | bool enabled; |
49 | struct list_head next; | |
20d29d7a AB |
50 | }; |
51 | ||
52 | static struct proto macvtap_proto = { | |
53 | .name = "macvtap", | |
54 | .owner = THIS_MODULE, | |
55 | .obj_size = sizeof (struct macvtap_queue), | |
56 | }; | |
57 | ||
58 | /* | |
e09eff7f | 59 | * Variables for dealing with macvtaps device numbers. |
20d29d7a | 60 | */ |
1ebed71a | 61 | static dev_t macvtap_major; |
e09eff7f EB |
62 | #define MACVTAP_NUM_DEVS (1U << MINORBITS) |
63 | static DEFINE_MUTEX(minor_lock); | |
64 | static DEFINE_IDR(minor_idr); | |
65 | ||
97bc3633 | 66 | #define GOODCOPY_LEN 128 |
20d29d7a AB |
67 | static struct class *macvtap_class; |
68 | static struct cdev macvtap_cdev; | |
69 | ||
501c774c AB |
70 | static const struct proto_ops macvtap_socket_ops; |
71 | ||
20d29d7a AB |
72 | /* |
73 | * RCU usage: | |
02df55d2 AB |
74 | * The macvtap_queue and the macvlan_dev are loosely coupled, the |
75 | * pointers from one to the other can only be read while rcu_read_lock | |
76 | * or macvtap_lock is held. | |
20d29d7a | 77 | * |
02df55d2 AB |
78 | * Both the file and the macvlan_dev hold a reference on the macvtap_queue |
79 | * through sock_hold(&q->sk). When the macvlan_dev goes away first, | |
80 | * q->vlan becomes inaccessible. When the files gets closed, | |
81 | * macvtap_get_queue() fails. | |
20d29d7a | 82 | * |
02df55d2 AB |
83 | * There may still be references to the struct sock inside of the |
84 | * queue from outbound SKBs, but these never reference back to the | |
85 | * file or the dev. The data structure is freed through __sk_free | |
86 | * when both our references and any pending SKBs are gone. | |
20d29d7a AB |
87 | */ |
88 | static DEFINE_SPINLOCK(macvtap_lock); | |
89 | ||
815f236d | 90 | static int macvtap_enable_queue(struct net_device *dev, struct file *file, |
20d29d7a | 91 | struct macvtap_queue *q) |
815f236d JW |
92 | { |
93 | struct macvlan_dev *vlan = netdev_priv(dev); | |
94 | int err = -EINVAL; | |
95 | ||
96 | spin_lock(&macvtap_lock); | |
97 | ||
98 | if (q->enabled) | |
99 | goto out; | |
100 | ||
101 | err = 0; | |
102 | rcu_assign_pointer(vlan->taps[vlan->numvtaps], q); | |
103 | q->queue_index = vlan->numvtaps; | |
104 | q->enabled = true; | |
105 | ||
106 | vlan->numvtaps++; | |
107 | out: | |
108 | spin_unlock(&macvtap_lock); | |
109 | return err; | |
110 | } | |
111 | ||
112 | static int macvtap_set_queue(struct net_device *dev, struct file *file, | |
113 | struct macvtap_queue *q) | |
20d29d7a AB |
114 | { |
115 | struct macvlan_dev *vlan = netdev_priv(dev); | |
116 | int err = -EBUSY; | |
117 | ||
118 | spin_lock(&macvtap_lock); | |
815f236d | 119 | if (vlan->numqueues == MAX_MACVTAP_QUEUES) |
20d29d7a AB |
120 | goto out; |
121 | ||
122 | err = 0; | |
02df55d2 | 123 | rcu_assign_pointer(q->vlan, vlan); |
376b1aab | 124 | rcu_assign_pointer(vlan->taps[vlan->numvtaps], q); |
02df55d2 | 125 | sock_hold(&q->sk); |
20d29d7a AB |
126 | |
127 | q->file = file; | |
376b1aab | 128 | q->queue_index = vlan->numvtaps; |
815f236d | 129 | q->enabled = true; |
02df55d2 | 130 | file->private_data = q; |
815f236d | 131 | list_add_tail(&q->next, &vlan->queue_list); |
20d29d7a | 132 | |
1565c7c1 | 133 | vlan->numvtaps++; |
815f236d | 134 | vlan->numqueues++; |
1565c7c1 | 135 | |
20d29d7a AB |
136 | out: |
137 | spin_unlock(&macvtap_lock); | |
138 | return err; | |
139 | } | |
140 | ||
815f236d JW |
141 | static int __macvtap_disable_queue(struct macvtap_queue *q) |
142 | { | |
143 | struct macvlan_dev *vlan; | |
144 | struct macvtap_queue *nq; | |
145 | ||
146 | vlan = rcu_dereference_protected(q->vlan, | |
147 | lockdep_is_held(&macvtap_lock)); | |
148 | ||
149 | if (!q->enabled) | |
150 | return -EINVAL; | |
151 | ||
152 | if (vlan) { | |
153 | int index = q->queue_index; | |
154 | BUG_ON(index >= vlan->numvtaps); | |
155 | nq = rcu_dereference_protected(vlan->taps[vlan->numvtaps - 1], | |
156 | lockdep_is_held(&macvtap_lock)); | |
157 | nq->queue_index = index; | |
158 | ||
159 | rcu_assign_pointer(vlan->taps[index], nq); | |
160 | RCU_INIT_POINTER(vlan->taps[vlan->numvtaps - 1], NULL); | |
161 | q->enabled = false; | |
162 | ||
163 | vlan->numvtaps--; | |
164 | } | |
165 | ||
166 | return 0; | |
167 | } | |
168 | ||
169 | static int macvtap_disable_queue(struct macvtap_queue *q) | |
170 | { | |
171 | int err; | |
172 | ||
173 | spin_lock(&macvtap_lock); | |
174 | err = __macvtap_disable_queue(q); | |
175 | spin_unlock(&macvtap_lock); | |
176 | ||
177 | return err; | |
178 | } | |
179 | ||
20d29d7a | 180 | /* |
02df55d2 AB |
181 | * The file owning the queue got closed, give up both |
182 | * the reference that the files holds as well as the | |
183 | * one from the macvlan_dev if that still exists. | |
20d29d7a AB |
184 | * |
185 | * Using the spinlock makes sure that we don't get | |
186 | * to the queue again after destroying it. | |
20d29d7a | 187 | */ |
02df55d2 | 188 | static void macvtap_put_queue(struct macvtap_queue *q) |
20d29d7a | 189 | { |
02df55d2 | 190 | struct macvlan_dev *vlan; |
20d29d7a AB |
191 | |
192 | spin_lock(&macvtap_lock); | |
13707f9e ED |
193 | vlan = rcu_dereference_protected(q->vlan, |
194 | lockdep_is_held(&macvtap_lock)); | |
02df55d2 | 195 | if (vlan) { |
815f236d JW |
196 | if (q->enabled) |
197 | BUG_ON(__macvtap_disable_queue(q)); | |
376b1aab | 198 | |
815f236d | 199 | vlan->numqueues--; |
2cfa5a04 | 200 | RCU_INIT_POINTER(q->vlan, NULL); |
02df55d2 | 201 | sock_put(&q->sk); |
815f236d | 202 | list_del_init(&q->next); |
20d29d7a AB |
203 | } |
204 | ||
20d29d7a AB |
205 | spin_unlock(&macvtap_lock); |
206 | ||
207 | synchronize_rcu(); | |
208 | sock_put(&q->sk); | |
209 | } | |
210 | ||
211 | /* | |
1565c7c1 KK |
212 | * Select a queue based on the rxq of the device on which this packet |
213 | * arrived. If the incoming device is not mq, calculate a flow hash | |
214 | * to select a queue. If all fails, find the first available queue. | |
215 | * Cache vlan->numvtaps since it can become zero during the execution | |
216 | * of this function. | |
20d29d7a AB |
217 | */ |
218 | static struct macvtap_queue *macvtap_get_queue(struct net_device *dev, | |
219 | struct sk_buff *skb) | |
220 | { | |
221 | struct macvlan_dev *vlan = netdev_priv(dev); | |
1565c7c1 | 222 | struct macvtap_queue *tap = NULL; |
815f236d JW |
223 | /* Access to taps array is protected by rcu, but access to numvtaps |
224 | * isn't. Below we use it to lookup a queue, but treat it as a hint | |
225 | * and validate that the result isn't NULL - in case we are | |
226 | * racing against queue removal. | |
227 | */ | |
ed0483fa | 228 | int numvtaps = ACCESS_ONCE(vlan->numvtaps); |
1565c7c1 KK |
229 | __u32 rxq; |
230 | ||
231 | if (!numvtaps) | |
232 | goto out; | |
233 | ||
ef0002b5 KK |
234 | /* Check if we can use flow to select a queue */ |
235 | rxq = skb_get_rxhash(skb); | |
236 | if (rxq) { | |
237 | tap = rcu_dereference(vlan->taps[rxq % numvtaps]); | |
376b1aab | 238 | goto out; |
ef0002b5 KK |
239 | } |
240 | ||
1565c7c1 KK |
241 | if (likely(skb_rx_queue_recorded(skb))) { |
242 | rxq = skb_get_rx_queue(skb); | |
20d29d7a | 243 | |
1565c7c1 KK |
244 | while (unlikely(rxq >= numvtaps)) |
245 | rxq -= numvtaps; | |
246 | ||
247 | tap = rcu_dereference(vlan->taps[rxq]); | |
376b1aab | 248 | goto out; |
1565c7c1 KK |
249 | } |
250 | ||
376b1aab | 251 | tap = rcu_dereference(vlan->taps[0]); |
1565c7c1 KK |
252 | out: |
253 | return tap; | |
20d29d7a AB |
254 | } |
255 | ||
02df55d2 AB |
256 | /* |
257 | * The net_device is going away, give up the reference | |
1565c7c1 KK |
258 | * that it holds on all queues and safely set the pointer |
259 | * from the queues to NULL. | |
02df55d2 | 260 | */ |
20d29d7a AB |
261 | static void macvtap_del_queues(struct net_device *dev) |
262 | { | |
263 | struct macvlan_dev *vlan = netdev_priv(dev); | |
815f236d | 264 | struct macvtap_queue *q, *tmp, *qlist[MAX_MACVTAP_QUEUES]; |
1565c7c1 | 265 | int i, j = 0; |
02df55d2 AB |
266 | |
267 | spin_lock(&macvtap_lock); | |
815f236d JW |
268 | list_for_each_entry_safe(q, tmp, &vlan->queue_list, next) { |
269 | list_del_init(&q->next); | |
376b1aab | 270 | qlist[j++] = q; |
376b1aab | 271 | RCU_INIT_POINTER(q->vlan, NULL); |
815f236d JW |
272 | if (q->enabled) |
273 | vlan->numvtaps--; | |
274 | vlan->numqueues--; | |
564517e8 | 275 | } |
815f236d JW |
276 | for (i = 0; i < vlan->numvtaps; i++) |
277 | RCU_INIT_POINTER(vlan->taps[i], NULL); | |
278 | BUG_ON(vlan->numvtaps); | |
279 | BUG_ON(vlan->numqueues); | |
99f34b38 EB |
280 | /* guarantee that any future macvtap_set_queue will fail */ |
281 | vlan->numvtaps = MAX_MACVTAP_QUEUES; | |
02df55d2 AB |
282 | spin_unlock(&macvtap_lock); |
283 | ||
284 | synchronize_rcu(); | |
1565c7c1 KK |
285 | |
286 | for (--j; j >= 0; j--) | |
287 | sock_put(&qlist[j]->sk); | |
20d29d7a AB |
288 | } |
289 | ||
290 | /* | |
291 | * Forward happens for data that gets sent from one macvlan | |
292 | * endpoint to another one in bridge mode. We just take | |
293 | * the skb and put it into the receive queue. | |
294 | */ | |
295 | static int macvtap_forward(struct net_device *dev, struct sk_buff *skb) | |
296 | { | |
297 | struct macvtap_queue *q = macvtap_get_queue(dev, skb); | |
298 | if (!q) | |
8a35747a HX |
299 | goto drop; |
300 | ||
301 | if (skb_queue_len(&q->sk.sk_receive_queue) >= dev->tx_queue_len) | |
302 | goto drop; | |
20d29d7a AB |
303 | |
304 | skb_queue_tail(&q->sk.sk_receive_queue, skb); | |
4a4771a5 | 305 | wake_up_interruptible_poll(sk_sleep(&q->sk), POLLIN | POLLRDNORM | POLLRDBAND); |
8a35747a HX |
306 | return NET_RX_SUCCESS; |
307 | ||
308 | drop: | |
309 | kfree_skb(skb); | |
310 | return NET_RX_DROP; | |
20d29d7a AB |
311 | } |
312 | ||
313 | /* | |
314 | * Receive is for data from the external interface (lowerdev), | |
315 | * in case of macvtap, we can treat that the same way as | |
316 | * forward, which macvlan cannot. | |
317 | */ | |
318 | static int macvtap_receive(struct sk_buff *skb) | |
319 | { | |
320 | skb_push(skb, ETH_HLEN); | |
321 | return macvtap_forward(skb->dev, skb); | |
322 | } | |
323 | ||
e09eff7f EB |
324 | static int macvtap_get_minor(struct macvlan_dev *vlan) |
325 | { | |
326 | int retval = -ENOMEM; | |
e09eff7f EB |
327 | |
328 | mutex_lock(&minor_lock); | |
ec09ebc1 TH |
329 | retval = idr_alloc(&minor_idr, vlan, 1, MACVTAP_NUM_DEVS, GFP_KERNEL); |
330 | if (retval >= 0) { | |
331 | vlan->minor = retval; | |
332 | } else if (retval == -ENOSPC) { | |
e09eff7f EB |
333 | printk(KERN_ERR "too many macvtap devices\n"); |
334 | retval = -EINVAL; | |
e09eff7f | 335 | } |
e09eff7f | 336 | mutex_unlock(&minor_lock); |
ec09ebc1 | 337 | return retval < 0 ? retval : 0; |
e09eff7f EB |
338 | } |
339 | ||
340 | static void macvtap_free_minor(struct macvlan_dev *vlan) | |
341 | { | |
342 | mutex_lock(&minor_lock); | |
343 | if (vlan->minor) { | |
344 | idr_remove(&minor_idr, vlan->minor); | |
345 | vlan->minor = 0; | |
346 | } | |
347 | mutex_unlock(&minor_lock); | |
348 | } | |
349 | ||
350 | static struct net_device *dev_get_by_macvtap_minor(int minor) | |
351 | { | |
352 | struct net_device *dev = NULL; | |
353 | struct macvlan_dev *vlan; | |
354 | ||
355 | mutex_lock(&minor_lock); | |
356 | vlan = idr_find(&minor_idr, minor); | |
357 | if (vlan) { | |
358 | dev = vlan->dev; | |
359 | dev_hold(dev); | |
360 | } | |
361 | mutex_unlock(&minor_lock); | |
362 | return dev; | |
363 | } | |
364 | ||
20d29d7a AB |
365 | static int macvtap_newlink(struct net *src_net, |
366 | struct net_device *dev, | |
367 | struct nlattr *tb[], | |
368 | struct nlattr *data[]) | |
369 | { | |
815f236d JW |
370 | struct macvlan_dev *vlan = netdev_priv(dev); |
371 | INIT_LIST_HEAD(&vlan->queue_list); | |
372 | ||
9bf1907f EB |
373 | /* Don't put anything that may fail after macvlan_common_newlink |
374 | * because we can't undo what it does. | |
375 | */ | |
376 | return macvlan_common_newlink(src_net, dev, tb, data, | |
377 | macvtap_receive, macvtap_forward); | |
20d29d7a AB |
378 | } |
379 | ||
380 | static void macvtap_dellink(struct net_device *dev, | |
381 | struct list_head *head) | |
382 | { | |
20d29d7a AB |
383 | macvtap_del_queues(dev); |
384 | macvlan_dellink(dev, head); | |
385 | } | |
386 | ||
8a35747a HX |
387 | static void macvtap_setup(struct net_device *dev) |
388 | { | |
389 | macvlan_common_setup(dev); | |
390 | dev->tx_queue_len = TUN_READQ_SIZE; | |
391 | } | |
392 | ||
20d29d7a AB |
393 | static struct rtnl_link_ops macvtap_link_ops __read_mostly = { |
394 | .kind = "macvtap", | |
8a35747a | 395 | .setup = macvtap_setup, |
20d29d7a AB |
396 | .newlink = macvtap_newlink, |
397 | .dellink = macvtap_dellink, | |
398 | }; | |
399 | ||
400 | ||
401 | static void macvtap_sock_write_space(struct sock *sk) | |
402 | { | |
43815482 ED |
403 | wait_queue_head_t *wqueue; |
404 | ||
20d29d7a AB |
405 | if (!sock_writeable(sk) || |
406 | !test_and_clear_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags)) | |
407 | return; | |
408 | ||
43815482 ED |
409 | wqueue = sk_sleep(sk); |
410 | if (wqueue && waitqueue_active(wqueue)) | |
411 | wake_up_interruptible_poll(wqueue, POLLOUT | POLLWRNORM | POLLWRBAND); | |
20d29d7a AB |
412 | } |
413 | ||
2259fef0 EB |
414 | static void macvtap_sock_destruct(struct sock *sk) |
415 | { | |
416 | skb_queue_purge(&sk->sk_receive_queue); | |
417 | } | |
418 | ||
20d29d7a AB |
419 | static int macvtap_open(struct inode *inode, struct file *file) |
420 | { | |
421 | struct net *net = current->nsproxy->net_ns; | |
e09eff7f | 422 | struct net_device *dev = dev_get_by_macvtap_minor(iminor(inode)); |
20d29d7a AB |
423 | struct macvtap_queue *q; |
424 | int err; | |
425 | ||
426 | err = -ENODEV; | |
427 | if (!dev) | |
428 | goto out; | |
429 | ||
20d29d7a AB |
430 | err = -ENOMEM; |
431 | q = (struct macvtap_queue *)sk_alloc(net, AF_UNSPEC, GFP_KERNEL, | |
432 | &macvtap_proto); | |
433 | if (!q) | |
434 | goto out; | |
435 | ||
43815482 ED |
436 | q->sock.wq = &q->wq; |
437 | init_waitqueue_head(&q->wq.wait); | |
20d29d7a AB |
438 | q->sock.type = SOCK_RAW; |
439 | q->sock.state = SS_CONNECTED; | |
501c774c AB |
440 | q->sock.file = file; |
441 | q->sock.ops = &macvtap_socket_ops; | |
20d29d7a | 442 | sock_init_data(&q->sock, &q->sk); |
20d29d7a | 443 | q->sk.sk_write_space = macvtap_sock_write_space; |
2259fef0 | 444 | q->sk.sk_destruct = macvtap_sock_destruct; |
b9fb9ee0 | 445 | q->flags = IFF_VNET_HDR | IFF_NO_PI | IFF_TAP; |
55afbd08 | 446 | q->vnet_hdr_sz = sizeof(struct virtio_net_hdr); |
20d29d7a | 447 | |
97bc3633 SM |
448 | /* |
449 | * so far only KVM virtio_net uses macvtap, enable zero copy between | |
450 | * guest kernel and host kernel when lower device supports zerocopy | |
047af9cf EB |
451 | * |
452 | * The macvlan supports zerocopy iff the lower device supports zero | |
453 | * copy so we don't have to look at the lower device directly. | |
97bc3633 | 454 | */ |
047af9cf EB |
455 | if ((dev->features & NETIF_F_HIGHDMA) && (dev->features & NETIF_F_SG)) |
456 | sock_set_flag(&q->sk, SOCK_ZEROCOPY); | |
97bc3633 | 457 | |
20d29d7a AB |
458 | err = macvtap_set_queue(dev, file, q); |
459 | if (err) | |
460 | sock_put(&q->sk); | |
461 | ||
462 | out: | |
463 | if (dev) | |
464 | dev_put(dev); | |
465 | ||
466 | return err; | |
467 | } | |
468 | ||
469 | static int macvtap_release(struct inode *inode, struct file *file) | |
470 | { | |
02df55d2 AB |
471 | struct macvtap_queue *q = file->private_data; |
472 | macvtap_put_queue(q); | |
20d29d7a AB |
473 | return 0; |
474 | } | |
475 | ||
476 | static unsigned int macvtap_poll(struct file *file, poll_table * wait) | |
477 | { | |
02df55d2 | 478 | struct macvtap_queue *q = file->private_data; |
20d29d7a AB |
479 | unsigned int mask = POLLERR; |
480 | ||
481 | if (!q) | |
482 | goto out; | |
483 | ||
484 | mask = 0; | |
43815482 | 485 | poll_wait(file, &q->wq.wait, wait); |
20d29d7a AB |
486 | |
487 | if (!skb_queue_empty(&q->sk.sk_receive_queue)) | |
488 | mask |= POLLIN | POLLRDNORM; | |
489 | ||
490 | if (sock_writeable(&q->sk) || | |
491 | (!test_and_set_bit(SOCK_ASYNC_NOSPACE, &q->sock.flags) && | |
492 | sock_writeable(&q->sk))) | |
493 | mask |= POLLOUT | POLLWRNORM; | |
494 | ||
495 | out: | |
20d29d7a AB |
496 | return mask; |
497 | } | |
498 | ||
b9fb9ee0 AB |
499 | static inline struct sk_buff *macvtap_alloc_skb(struct sock *sk, size_t prepad, |
500 | size_t len, size_t linear, | |
501 | int noblock, int *err) | |
502 | { | |
503 | struct sk_buff *skb; | |
504 | ||
505 | /* Under a page? Don't bother with paged skb. */ | |
506 | if (prepad + len < PAGE_SIZE || !linear) | |
507 | linear = len; | |
508 | ||
509 | skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock, | |
510 | err); | |
511 | if (!skb) | |
512 | return NULL; | |
513 | ||
514 | skb_reserve(skb, prepad); | |
515 | skb_put(skb, linear); | |
516 | skb->data_len = len - linear; | |
517 | skb->len += len - linear; | |
518 | ||
519 | return skb; | |
520 | } | |
521 | ||
97bc3633 SM |
522 | /* set skb frags from iovec, this can move to core network code for reuse */ |
523 | static int zerocopy_sg_from_iovec(struct sk_buff *skb, const struct iovec *from, | |
524 | int offset, size_t count) | |
525 | { | |
526 | int len = iov_length(from, count) - offset; | |
527 | int copy = skb_headlen(skb); | |
528 | int size, offset1 = 0; | |
529 | int i = 0; | |
97bc3633 SM |
530 | |
531 | /* Skip over from offset */ | |
532 | while (count && (offset >= from->iov_len)) { | |
533 | offset -= from->iov_len; | |
534 | ++from; | |
535 | --count; | |
536 | } | |
537 | ||
538 | /* copy up to skb headlen */ | |
539 | while (count && (copy > 0)) { | |
540 | size = min_t(unsigned int, copy, from->iov_len - offset); | |
541 | if (copy_from_user(skb->data + offset1, from->iov_base + offset, | |
542 | size)) | |
543 | return -EFAULT; | |
544 | if (copy > size) { | |
545 | ++from; | |
546 | --count; | |
3afc9621 JW |
547 | offset = 0; |
548 | } else | |
549 | offset += size; | |
97bc3633 SM |
550 | copy -= size; |
551 | offset1 += size; | |
97bc3633 SM |
552 | } |
553 | ||
554 | if (len == offset1) | |
555 | return 0; | |
556 | ||
557 | while (count--) { | |
558 | struct page *page[MAX_SKB_FRAGS]; | |
559 | int num_pages; | |
560 | unsigned long base; | |
4ef67ebe | 561 | unsigned long truesize; |
97bc3633 | 562 | |
3afc9621 | 563 | len = from->iov_len - offset; |
97bc3633 | 564 | if (!len) { |
3afc9621 | 565 | offset = 0; |
97bc3633 SM |
566 | ++from; |
567 | continue; | |
568 | } | |
3afc9621 | 569 | base = (unsigned long)from->iov_base + offset; |
97bc3633 | 570 | size = ((base & ~PAGE_MASK) + len + ~PAGE_MASK) >> PAGE_SHIFT; |
b92946e2 JW |
571 | if (i + size > MAX_SKB_FRAGS) |
572 | return -EMSGSIZE; | |
97bc3633 | 573 | num_pages = get_user_pages_fast(base, size, 0, &page[i]); |
b92946e2 | 574 | if (num_pages != size) { |
02ce04bb JW |
575 | for (i = 0; i < num_pages; i++) |
576 | put_page(page[i]); | |
97bc3633 | 577 | return -EFAULT; |
02ce04bb | 578 | } |
4ef67ebe | 579 | truesize = size * PAGE_SIZE; |
97bc3633 SM |
580 | skb->data_len += len; |
581 | skb->len += len; | |
4ef67ebe JW |
582 | skb->truesize += truesize; |
583 | atomic_add(truesize, &skb->sk->sk_wmem_alloc); | |
97bc3633 | 584 | while (len) { |
653fc915 JW |
585 | int off = base & ~PAGE_MASK; |
586 | int size = min_t(int, len, PAGE_SIZE - off); | |
587 | __skb_fill_page_desc(skb, i, page[i], off, size); | |
97bc3633 SM |
588 | skb_shinfo(skb)->nr_frags++; |
589 | /* increase sk_wmem_alloc */ | |
653fc915 JW |
590 | base += size; |
591 | len -= size; | |
97bc3633 SM |
592 | i++; |
593 | } | |
3afc9621 | 594 | offset = 0; |
97bc3633 SM |
595 | ++from; |
596 | } | |
597 | return 0; | |
598 | } | |
599 | ||
b9fb9ee0 AB |
600 | /* |
601 | * macvtap_skb_from_vnet_hdr and macvtap_skb_to_vnet_hdr should | |
602 | * be shared with the tun/tap driver. | |
603 | */ | |
604 | static int macvtap_skb_from_vnet_hdr(struct sk_buff *skb, | |
605 | struct virtio_net_hdr *vnet_hdr) | |
606 | { | |
607 | unsigned short gso_type = 0; | |
608 | if (vnet_hdr->gso_type != VIRTIO_NET_HDR_GSO_NONE) { | |
609 | switch (vnet_hdr->gso_type & ~VIRTIO_NET_HDR_GSO_ECN) { | |
610 | case VIRTIO_NET_HDR_GSO_TCPV4: | |
611 | gso_type = SKB_GSO_TCPV4; | |
612 | break; | |
613 | case VIRTIO_NET_HDR_GSO_TCPV6: | |
614 | gso_type = SKB_GSO_TCPV6; | |
615 | break; | |
616 | case VIRTIO_NET_HDR_GSO_UDP: | |
617 | gso_type = SKB_GSO_UDP; | |
618 | break; | |
619 | default: | |
620 | return -EINVAL; | |
621 | } | |
622 | ||
623 | if (vnet_hdr->gso_type & VIRTIO_NET_HDR_GSO_ECN) | |
624 | gso_type |= SKB_GSO_TCP_ECN; | |
625 | ||
626 | if (vnet_hdr->gso_size == 0) | |
627 | return -EINVAL; | |
628 | } | |
629 | ||
630 | if (vnet_hdr->flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) { | |
631 | if (!skb_partial_csum_set(skb, vnet_hdr->csum_start, | |
632 | vnet_hdr->csum_offset)) | |
633 | return -EINVAL; | |
634 | } | |
635 | ||
636 | if (vnet_hdr->gso_type != VIRTIO_NET_HDR_GSO_NONE) { | |
637 | skb_shinfo(skb)->gso_size = vnet_hdr->gso_size; | |
c9af6db4 | 638 | skb_shinfo(skb)->gso_type = gso_type; |
b9fb9ee0 AB |
639 | |
640 | /* Header must be checked, and gso_segs computed. */ | |
641 | skb_shinfo(skb)->gso_type |= SKB_GSO_DODGY; | |
642 | skb_shinfo(skb)->gso_segs = 0; | |
643 | } | |
644 | return 0; | |
645 | } | |
646 | ||
647 | static int macvtap_skb_to_vnet_hdr(const struct sk_buff *skb, | |
648 | struct virtio_net_hdr *vnet_hdr) | |
649 | { | |
650 | memset(vnet_hdr, 0, sizeof(*vnet_hdr)); | |
651 | ||
652 | if (skb_is_gso(skb)) { | |
653 | struct skb_shared_info *sinfo = skb_shinfo(skb); | |
654 | ||
655 | /* This is a hint as to how much should be linear. */ | |
656 | vnet_hdr->hdr_len = skb_headlen(skb); | |
657 | vnet_hdr->gso_size = sinfo->gso_size; | |
658 | if (sinfo->gso_type & SKB_GSO_TCPV4) | |
659 | vnet_hdr->gso_type = VIRTIO_NET_HDR_GSO_TCPV4; | |
660 | else if (sinfo->gso_type & SKB_GSO_TCPV6) | |
661 | vnet_hdr->gso_type = VIRTIO_NET_HDR_GSO_TCPV6; | |
662 | else if (sinfo->gso_type & SKB_GSO_UDP) | |
663 | vnet_hdr->gso_type = VIRTIO_NET_HDR_GSO_UDP; | |
664 | else | |
665 | BUG(); | |
666 | if (sinfo->gso_type & SKB_GSO_TCP_ECN) | |
667 | vnet_hdr->gso_type |= VIRTIO_NET_HDR_GSO_ECN; | |
668 | } else | |
669 | vnet_hdr->gso_type = VIRTIO_NET_HDR_GSO_NONE; | |
670 | ||
671 | if (skb->ip_summed == CHECKSUM_PARTIAL) { | |
672 | vnet_hdr->flags = VIRTIO_NET_HDR_F_NEEDS_CSUM; | |
55508d60 | 673 | vnet_hdr->csum_start = skb_checksum_start_offset(skb); |
b9fb9ee0 | 674 | vnet_hdr->csum_offset = skb->csum_offset; |
10a8d94a JW |
675 | } else if (skb->ip_summed == CHECKSUM_UNNECESSARY) { |
676 | vnet_hdr->flags = VIRTIO_NET_HDR_F_DATA_VALID; | |
b9fb9ee0 AB |
677 | } /* else everything is zero */ |
678 | ||
679 | return 0; | |
680 | } | |
681 | ||
682 | ||
20d29d7a | 683 | /* Get packet from user space buffer */ |
97bc3633 SM |
684 | static ssize_t macvtap_get_user(struct macvtap_queue *q, struct msghdr *m, |
685 | const struct iovec *iv, unsigned long total_len, | |
686 | size_t count, int noblock) | |
20d29d7a AB |
687 | { |
688 | struct sk_buff *skb; | |
02df55d2 | 689 | struct macvlan_dev *vlan; |
97bc3633 | 690 | unsigned long len = total_len; |
20d29d7a | 691 | int err; |
b9fb9ee0 AB |
692 | struct virtio_net_hdr vnet_hdr = { 0 }; |
693 | int vnet_hdr_len = 0; | |
b92946e2 | 694 | int copylen = 0; |
97bc3633 | 695 | bool zerocopy = false; |
b9fb9ee0 AB |
696 | |
697 | if (q->flags & IFF_VNET_HDR) { | |
55afbd08 | 698 | vnet_hdr_len = q->vnet_hdr_sz; |
b9fb9ee0 AB |
699 | |
700 | err = -EINVAL; | |
ce3c8692 | 701 | if (len < vnet_hdr_len) |
b9fb9ee0 | 702 | goto err; |
ce3c8692 | 703 | len -= vnet_hdr_len; |
b9fb9ee0 AB |
704 | |
705 | err = memcpy_fromiovecend((void *)&vnet_hdr, iv, 0, | |
55afbd08 | 706 | sizeof(vnet_hdr)); |
b9fb9ee0 AB |
707 | if (err < 0) |
708 | goto err; | |
709 | if ((vnet_hdr.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) && | |
710 | vnet_hdr.csum_start + vnet_hdr.csum_offset + 2 > | |
711 | vnet_hdr.hdr_len) | |
712 | vnet_hdr.hdr_len = vnet_hdr.csum_start + | |
713 | vnet_hdr.csum_offset + 2; | |
714 | err = -EINVAL; | |
715 | if (vnet_hdr.hdr_len > len) | |
716 | goto err; | |
717 | } | |
20d29d7a | 718 | |
b9fb9ee0 | 719 | err = -EINVAL; |
20d29d7a | 720 | if (unlikely(len < ETH_HLEN)) |
b9fb9ee0 | 721 | goto err; |
20d29d7a | 722 | |
b92946e2 JW |
723 | err = -EMSGSIZE; |
724 | if (unlikely(count > UIO_MAXIOV)) | |
725 | goto err; | |
726 | ||
97bc3633 SM |
727 | if (m && m->msg_control && sock_flag(&q->sk, SOCK_ZEROCOPY)) |
728 | zerocopy = true; | |
729 | ||
730 | if (zerocopy) { | |
b92946e2 JW |
731 | /* Userspace may produce vectors with count greater than |
732 | * MAX_SKB_FRAGS, so we need to linearize parts of the skb | |
733 | * to let the rest of data to be fit in the frags. | |
734 | */ | |
735 | if (count > MAX_SKB_FRAGS) { | |
736 | copylen = iov_length(iv, count - MAX_SKB_FRAGS); | |
737 | if (copylen < vnet_hdr_len) | |
738 | copylen = 0; | |
739 | else | |
740 | copylen -= vnet_hdr_len; | |
741 | } | |
97bc3633 SM |
742 | /* There are 256 bytes to be copied in skb, so there is enough |
743 | * room for skb expand head in case it is used. | |
744 | * The rest buffer is mapped from userspace. | |
745 | */ | |
b92946e2 JW |
746 | if (copylen < vnet_hdr.hdr_len) |
747 | copylen = vnet_hdr.hdr_len; | |
97bc3633 SM |
748 | if (!copylen) |
749 | copylen = GOODCOPY_LEN; | |
750 | } else | |
751 | copylen = len; | |
752 | ||
753 | skb = macvtap_alloc_skb(&q->sk, NET_IP_ALIGN, copylen, | |
754 | vnet_hdr.hdr_len, noblock, &err); | |
02df55d2 AB |
755 | if (!skb) |
756 | goto err; | |
20d29d7a | 757 | |
01d6657b | 758 | if (zerocopy) |
97bc3633 | 759 | err = zerocopy_sg_from_iovec(skb, iv, vnet_hdr_len, count); |
01d6657b | 760 | else |
97bc3633 SM |
761 | err = skb_copy_datagram_from_iovec(skb, 0, iv, vnet_hdr_len, |
762 | len); | |
02df55d2 | 763 | if (err) |
b9fb9ee0 | 764 | goto err_kfree; |
20d29d7a AB |
765 | |
766 | skb_set_network_header(skb, ETH_HLEN); | |
b9fb9ee0 AB |
767 | skb_reset_mac_header(skb); |
768 | skb->protocol = eth_hdr(skb)->h_proto; | |
769 | ||
770 | if (vnet_hdr_len) { | |
771 | err = macvtap_skb_from_vnet_hdr(skb, &vnet_hdr); | |
772 | if (err) | |
773 | goto err_kfree; | |
774 | } | |
775 | ||
40893fd0 | 776 | skb_probe_transport_header(skb, ETH_HLEN); |
9b4d669b | 777 | |
02df55d2 | 778 | rcu_read_lock_bh(); |
13707f9e | 779 | vlan = rcu_dereference_bh(q->vlan); |
97bc3633 | 780 | /* copy skb_ubuf_info for callback when skb has no error */ |
01d6657b | 781 | if (zerocopy) { |
97bc3633 | 782 | skb_shinfo(skb)->destructor_arg = m->msg_control; |
01d6657b | 783 | skb_shinfo(skb)->tx_flags |= SKBTX_DEV_ZEROCOPY; |
c9af6db4 | 784 | skb_shinfo(skb)->tx_flags |= SKBTX_SHARED_FRAG; |
01d6657b | 785 | } |
02df55d2 AB |
786 | if (vlan) |
787 | macvlan_start_xmit(skb, vlan->dev); | |
788 | else | |
789 | kfree_skb(skb); | |
790 | rcu_read_unlock_bh(); | |
20d29d7a | 791 | |
97bc3633 | 792 | return total_len; |
02df55d2 | 793 | |
b9fb9ee0 AB |
794 | err_kfree: |
795 | kfree_skb(skb); | |
796 | ||
02df55d2 AB |
797 | err: |
798 | rcu_read_lock_bh(); | |
13707f9e | 799 | vlan = rcu_dereference_bh(q->vlan); |
02df55d2 | 800 | if (vlan) |
1ac9ad13 | 801 | vlan->dev->stats.tx_dropped++; |
02df55d2 AB |
802 | rcu_read_unlock_bh(); |
803 | ||
02df55d2 | 804 | return err; |
20d29d7a AB |
805 | } |
806 | ||
807 | static ssize_t macvtap_aio_write(struct kiocb *iocb, const struct iovec *iv, | |
808 | unsigned long count, loff_t pos) | |
809 | { | |
810 | struct file *file = iocb->ki_filp; | |
811 | ssize_t result = -ENOLINK; | |
02df55d2 | 812 | struct macvtap_queue *q = file->private_data; |
20d29d7a | 813 | |
97bc3633 SM |
814 | result = macvtap_get_user(q, NULL, iv, iov_length(iv, count), count, |
815 | file->f_flags & O_NONBLOCK); | |
20d29d7a AB |
816 | return result; |
817 | } | |
818 | ||
819 | /* Put packet to the user space buffer */ | |
820 | static ssize_t macvtap_put_user(struct macvtap_queue *q, | |
821 | const struct sk_buff *skb, | |
822 | const struct iovec *iv, int len) | |
823 | { | |
02df55d2 | 824 | struct macvlan_dev *vlan; |
20d29d7a | 825 | int ret; |
b9fb9ee0 | 826 | int vnet_hdr_len = 0; |
f09e2249 BG |
827 | int vlan_offset = 0; |
828 | int copied; | |
b9fb9ee0 AB |
829 | |
830 | if (q->flags & IFF_VNET_HDR) { | |
831 | struct virtio_net_hdr vnet_hdr; | |
55afbd08 | 832 | vnet_hdr_len = q->vnet_hdr_sz; |
b9fb9ee0 AB |
833 | if ((len -= vnet_hdr_len) < 0) |
834 | return -EINVAL; | |
835 | ||
836 | ret = macvtap_skb_to_vnet_hdr(skb, &vnet_hdr); | |
837 | if (ret) | |
838 | return ret; | |
839 | ||
55afbd08 | 840 | if (memcpy_toiovecend(iv, (void *)&vnet_hdr, 0, sizeof(vnet_hdr))) |
b9fb9ee0 AB |
841 | return -EFAULT; |
842 | } | |
f09e2249 BG |
843 | copied = vnet_hdr_len; |
844 | ||
845 | if (!vlan_tx_tag_present(skb)) | |
846 | len = min_t(int, skb->len, len); | |
847 | else { | |
848 | int copy; | |
849 | struct { | |
850 | __be16 h_vlan_proto; | |
851 | __be16 h_vlan_TCI; | |
852 | } veth; | |
853 | veth.h_vlan_proto = htons(ETH_P_8021Q); | |
854 | veth.h_vlan_TCI = htons(vlan_tx_tag_get(skb)); | |
855 | ||
856 | vlan_offset = offsetof(struct vlan_ethhdr, h_vlan_proto); | |
857 | len = min_t(int, skb->len + VLAN_HLEN, len); | |
858 | ||
859 | copy = min_t(int, vlan_offset, len); | |
860 | ret = skb_copy_datagram_const_iovec(skb, 0, iv, copied, copy); | |
861 | len -= copy; | |
862 | copied += copy; | |
863 | if (ret || !len) | |
864 | goto done; | |
865 | ||
866 | copy = min_t(int, sizeof(veth), len); | |
867 | ret = memcpy_toiovecend(iv, (void *)&veth, copied, copy); | |
868 | len -= copy; | |
869 | copied += copy; | |
870 | if (ret || !len) | |
871 | goto done; | |
872 | } | |
20d29d7a | 873 | |
f09e2249 BG |
874 | ret = skb_copy_datagram_const_iovec(skb, vlan_offset, iv, copied, len); |
875 | copied += len; | |
20d29d7a | 876 | |
f09e2249 | 877 | done: |
02df55d2 | 878 | rcu_read_lock_bh(); |
13707f9e | 879 | vlan = rcu_dereference_bh(q->vlan); |
501c774c | 880 | if (vlan) |
f09e2249 | 881 | macvlan_count_rx(vlan, copied - vnet_hdr_len, ret == 0, 0); |
02df55d2 | 882 | rcu_read_unlock_bh(); |
20d29d7a | 883 | |
f09e2249 | 884 | return ret ? ret : copied; |
20d29d7a AB |
885 | } |
886 | ||
501c774c AB |
887 | static ssize_t macvtap_do_read(struct macvtap_queue *q, struct kiocb *iocb, |
888 | const struct iovec *iv, unsigned long len, | |
889 | int noblock) | |
20d29d7a | 890 | { |
ccf7e72b | 891 | DEFINE_WAIT(wait); |
20d29d7a | 892 | struct sk_buff *skb; |
501c774c | 893 | ssize_t ret = 0; |
20d29d7a | 894 | |
20d29d7a | 895 | while (len) { |
89cee917 JW |
896 | if (!noblock) |
897 | prepare_to_wait(sk_sleep(&q->sk), &wait, | |
898 | TASK_INTERRUPTIBLE); | |
20d29d7a AB |
899 | |
900 | /* Read frames from the queue */ | |
901 | skb = skb_dequeue(&q->sk.sk_receive_queue); | |
902 | if (!skb) { | |
501c774c | 903 | if (noblock) { |
20d29d7a AB |
904 | ret = -EAGAIN; |
905 | break; | |
906 | } | |
907 | if (signal_pending(current)) { | |
908 | ret = -ERESTARTSYS; | |
909 | break; | |
910 | } | |
911 | /* Nothing to read, let's sleep */ | |
912 | schedule(); | |
913 | continue; | |
914 | } | |
915 | ret = macvtap_put_user(q, skb, iv, len); | |
916 | kfree_skb(skb); | |
917 | break; | |
918 | } | |
919 | ||
89cee917 JW |
920 | if (!noblock) |
921 | finish_wait(sk_sleep(&q->sk), &wait); | |
501c774c AB |
922 | return ret; |
923 | } | |
924 | ||
925 | static ssize_t macvtap_aio_read(struct kiocb *iocb, const struct iovec *iv, | |
926 | unsigned long count, loff_t pos) | |
927 | { | |
928 | struct file *file = iocb->ki_filp; | |
929 | struct macvtap_queue *q = file->private_data; | |
930 | ssize_t len, ret = 0; | |
20d29d7a | 931 | |
501c774c AB |
932 | len = iov_length(iv, count); |
933 | if (len < 0) { | |
934 | ret = -EINVAL; | |
935 | goto out; | |
936 | } | |
937 | ||
938 | ret = macvtap_do_read(q, iocb, iv, len, file->f_flags & O_NONBLOCK); | |
939 | ret = min_t(ssize_t, ret, len); /* XXX copied from tun.c. Why? */ | |
20d29d7a | 940 | out: |
20d29d7a AB |
941 | return ret; |
942 | } | |
943 | ||
8f475a31 JW |
944 | static struct macvlan_dev *macvtap_get_vlan(struct macvtap_queue *q) |
945 | { | |
946 | struct macvlan_dev *vlan; | |
947 | ||
948 | rcu_read_lock_bh(); | |
949 | vlan = rcu_dereference_bh(q->vlan); | |
950 | if (vlan) | |
951 | dev_hold(vlan->dev); | |
952 | rcu_read_unlock_bh(); | |
953 | ||
954 | return vlan; | |
955 | } | |
956 | ||
957 | static void macvtap_put_vlan(struct macvlan_dev *vlan) | |
958 | { | |
959 | dev_put(vlan->dev); | |
960 | } | |
961 | ||
815f236d JW |
962 | static int macvtap_ioctl_set_queue(struct file *file, unsigned int flags) |
963 | { | |
964 | struct macvtap_queue *q = file->private_data; | |
965 | struct macvlan_dev *vlan; | |
966 | int ret; | |
967 | ||
968 | vlan = macvtap_get_vlan(q); | |
969 | if (!vlan) | |
970 | return -EINVAL; | |
971 | ||
972 | if (flags & IFF_ATTACH_QUEUE) | |
973 | ret = macvtap_enable_queue(vlan->dev, file, q); | |
974 | else if (flags & IFF_DETACH_QUEUE) | |
975 | ret = macvtap_disable_queue(q); | |
976 | ||
977 | macvtap_put_vlan(vlan); | |
978 | return ret; | |
979 | } | |
980 | ||
20d29d7a AB |
981 | /* |
982 | * provide compatibility with generic tun/tap interface | |
983 | */ | |
984 | static long macvtap_ioctl(struct file *file, unsigned int cmd, | |
985 | unsigned long arg) | |
986 | { | |
02df55d2 AB |
987 | struct macvtap_queue *q = file->private_data; |
988 | struct macvlan_dev *vlan; | |
20d29d7a AB |
989 | void __user *argp = (void __user *)arg; |
990 | struct ifreq __user *ifr = argp; | |
991 | unsigned int __user *up = argp; | |
992 | unsigned int u; | |
55afbd08 MT |
993 | int __user *sp = argp; |
994 | int s; | |
02df55d2 | 995 | int ret; |
20d29d7a AB |
996 | |
997 | switch (cmd) { | |
998 | case TUNSETIFF: | |
999 | /* ignore the name, just look at flags */ | |
1000 | if (get_user(u, &ifr->ifr_flags)) | |
1001 | return -EFAULT; | |
b9fb9ee0 AB |
1002 | |
1003 | ret = 0; | |
815f236d JW |
1004 | if ((u & ~(IFF_VNET_HDR | IFF_MULTI_QUEUE)) != |
1005 | (IFF_NO_PI | IFF_TAP)) | |
b9fb9ee0 AB |
1006 | ret = -EINVAL; |
1007 | else | |
1008 | q->flags = u; | |
1009 | ||
1010 | return ret; | |
20d29d7a AB |
1011 | |
1012 | case TUNGETIFF: | |
8f475a31 | 1013 | vlan = macvtap_get_vlan(q); |
02df55d2 | 1014 | if (!vlan) |
20d29d7a | 1015 | return -ENOLINK; |
20d29d7a | 1016 | |
02df55d2 | 1017 | ret = 0; |
13707f9e | 1018 | if (copy_to_user(&ifr->ifr_name, vlan->dev->name, IFNAMSIZ) || |
b9fb9ee0 | 1019 | put_user(q->flags, &ifr->ifr_flags)) |
02df55d2 | 1020 | ret = -EFAULT; |
8f475a31 | 1021 | macvtap_put_vlan(vlan); |
02df55d2 | 1022 | return ret; |
20d29d7a | 1023 | |
815f236d JW |
1024 | case TUNSETQUEUE: |
1025 | if (get_user(u, &ifr->ifr_flags)) | |
1026 | return -EFAULT; | |
1027 | return macvtap_ioctl_set_queue(file, u); | |
1028 | ||
20d29d7a | 1029 | case TUNGETFEATURES: |
b9fb9ee0 | 1030 | if (put_user(IFF_TAP | IFF_NO_PI | IFF_VNET_HDR, up)) |
20d29d7a AB |
1031 | return -EFAULT; |
1032 | return 0; | |
1033 | ||
1034 | case TUNSETSNDBUF: | |
1035 | if (get_user(u, up)) | |
1036 | return -EFAULT; | |
1037 | ||
20d29d7a | 1038 | q->sk.sk_sndbuf = u; |
20d29d7a AB |
1039 | return 0; |
1040 | ||
55afbd08 MT |
1041 | case TUNGETVNETHDRSZ: |
1042 | s = q->vnet_hdr_sz; | |
1043 | if (put_user(s, sp)) | |
1044 | return -EFAULT; | |
1045 | return 0; | |
1046 | ||
1047 | case TUNSETVNETHDRSZ: | |
1048 | if (get_user(s, sp)) | |
1049 | return -EFAULT; | |
1050 | if (s < (int)sizeof(struct virtio_net_hdr)) | |
1051 | return -EINVAL; | |
1052 | ||
1053 | q->vnet_hdr_sz = s; | |
1054 | return 0; | |
1055 | ||
20d29d7a AB |
1056 | case TUNSETOFFLOAD: |
1057 | /* let the user check for future flags */ | |
1058 | if (arg & ~(TUN_F_CSUM | TUN_F_TSO4 | TUN_F_TSO6 | | |
b9fb9ee0 | 1059 | TUN_F_TSO_ECN | TUN_F_UFO)) |
20d29d7a AB |
1060 | return -EINVAL; |
1061 | ||
b9fb9ee0 AB |
1062 | /* TODO: only accept frames with the features that |
1063 | got enabled for forwarded frames */ | |
1064 | if (!(q->flags & IFF_VNET_HDR)) | |
1065 | return -EINVAL; | |
20d29d7a AB |
1066 | return 0; |
1067 | ||
1068 | default: | |
1069 | return -EINVAL; | |
1070 | } | |
1071 | } | |
1072 | ||
1073 | #ifdef CONFIG_COMPAT | |
1074 | static long macvtap_compat_ioctl(struct file *file, unsigned int cmd, | |
1075 | unsigned long arg) | |
1076 | { | |
1077 | return macvtap_ioctl(file, cmd, (unsigned long)compat_ptr(arg)); | |
1078 | } | |
1079 | #endif | |
1080 | ||
1081 | static const struct file_operations macvtap_fops = { | |
1082 | .owner = THIS_MODULE, | |
1083 | .open = macvtap_open, | |
1084 | .release = macvtap_release, | |
1085 | .aio_read = macvtap_aio_read, | |
1086 | .aio_write = macvtap_aio_write, | |
1087 | .poll = macvtap_poll, | |
1088 | .llseek = no_llseek, | |
1089 | .unlocked_ioctl = macvtap_ioctl, | |
1090 | #ifdef CONFIG_COMPAT | |
1091 | .compat_ioctl = macvtap_compat_ioctl, | |
1092 | #endif | |
1093 | }; | |
1094 | ||
501c774c AB |
1095 | static int macvtap_sendmsg(struct kiocb *iocb, struct socket *sock, |
1096 | struct msghdr *m, size_t total_len) | |
1097 | { | |
1098 | struct macvtap_queue *q = container_of(sock, struct macvtap_queue, sock); | |
97bc3633 | 1099 | return macvtap_get_user(q, m, m->msg_iov, total_len, m->msg_iovlen, |
501c774c AB |
1100 | m->msg_flags & MSG_DONTWAIT); |
1101 | } | |
1102 | ||
1103 | static int macvtap_recvmsg(struct kiocb *iocb, struct socket *sock, | |
1104 | struct msghdr *m, size_t total_len, | |
1105 | int flags) | |
1106 | { | |
1107 | struct macvtap_queue *q = container_of(sock, struct macvtap_queue, sock); | |
1108 | int ret; | |
1109 | if (flags & ~(MSG_DONTWAIT|MSG_TRUNC)) | |
1110 | return -EINVAL; | |
1111 | ret = macvtap_do_read(q, iocb, m->msg_iov, total_len, | |
1112 | flags & MSG_DONTWAIT); | |
1113 | if (ret > total_len) { | |
1114 | m->msg_flags |= MSG_TRUNC; | |
1115 | ret = flags & MSG_TRUNC ? ret : total_len; | |
1116 | } | |
1117 | return ret; | |
1118 | } | |
1119 | ||
1120 | /* Ops structure to mimic raw sockets with tun */ | |
1121 | static const struct proto_ops macvtap_socket_ops = { | |
1122 | .sendmsg = macvtap_sendmsg, | |
1123 | .recvmsg = macvtap_recvmsg, | |
1124 | }; | |
1125 | ||
1126 | /* Get an underlying socket object from tun file. Returns error unless file is | |
1127 | * attached to a device. The returned object works like a packet socket, it | |
1128 | * can be used for sock_sendmsg/sock_recvmsg. The caller is responsible for | |
1129 | * holding a reference to the file for as long as the socket is in use. */ | |
1130 | struct socket *macvtap_get_socket(struct file *file) | |
1131 | { | |
1132 | struct macvtap_queue *q; | |
1133 | if (file->f_op != &macvtap_fops) | |
1134 | return ERR_PTR(-EINVAL); | |
1135 | q = file->private_data; | |
1136 | if (!q) | |
1137 | return ERR_PTR(-EBADFD); | |
1138 | return &q->sock; | |
1139 | } | |
1140 | EXPORT_SYMBOL_GPL(macvtap_get_socket); | |
1141 | ||
9bf1907f EB |
1142 | static int macvtap_device_event(struct notifier_block *unused, |
1143 | unsigned long event, void *ptr) | |
1144 | { | |
351638e7 | 1145 | struct net_device *dev = netdev_notifier_info_to_dev(ptr); |
e09eff7f | 1146 | struct macvlan_dev *vlan; |
9bf1907f EB |
1147 | struct device *classdev; |
1148 | dev_t devt; | |
e09eff7f | 1149 | int err; |
9bf1907f EB |
1150 | |
1151 | if (dev->rtnl_link_ops != &macvtap_link_ops) | |
1152 | return NOTIFY_DONE; | |
1153 | ||
e09eff7f | 1154 | vlan = netdev_priv(dev); |
9bf1907f EB |
1155 | |
1156 | switch (event) { | |
1157 | case NETDEV_REGISTER: | |
1158 | /* Create the device node here after the network device has | |
1159 | * been registered but before register_netdevice has | |
1160 | * finished running. | |
1161 | */ | |
e09eff7f EB |
1162 | err = macvtap_get_minor(vlan); |
1163 | if (err) | |
1164 | return notifier_from_errno(err); | |
1165 | ||
1166 | devt = MKDEV(MAJOR(macvtap_major), vlan->minor); | |
9bf1907f EB |
1167 | classdev = device_create(macvtap_class, &dev->dev, devt, |
1168 | dev, "tap%d", dev->ifindex); | |
e09eff7f EB |
1169 | if (IS_ERR(classdev)) { |
1170 | macvtap_free_minor(vlan); | |
9bf1907f | 1171 | return notifier_from_errno(PTR_ERR(classdev)); |
e09eff7f | 1172 | } |
9bf1907f EB |
1173 | break; |
1174 | case NETDEV_UNREGISTER: | |
e09eff7f | 1175 | devt = MKDEV(MAJOR(macvtap_major), vlan->minor); |
9bf1907f | 1176 | device_destroy(macvtap_class, devt); |
e09eff7f | 1177 | macvtap_free_minor(vlan); |
9bf1907f EB |
1178 | break; |
1179 | } | |
1180 | ||
1181 | return NOTIFY_DONE; | |
1182 | } | |
1183 | ||
1184 | static struct notifier_block macvtap_notifier_block __read_mostly = { | |
1185 | .notifier_call = macvtap_device_event, | |
1186 | }; | |
1187 | ||
20d29d7a AB |
1188 | static int macvtap_init(void) |
1189 | { | |
1190 | int err; | |
1191 | ||
1192 | err = alloc_chrdev_region(&macvtap_major, 0, | |
1193 | MACVTAP_NUM_DEVS, "macvtap"); | |
1194 | if (err) | |
1195 | goto out1; | |
1196 | ||
1197 | cdev_init(&macvtap_cdev, &macvtap_fops); | |
1198 | err = cdev_add(&macvtap_cdev, macvtap_major, MACVTAP_NUM_DEVS); | |
1199 | if (err) | |
1200 | goto out2; | |
1201 | ||
1202 | macvtap_class = class_create(THIS_MODULE, "macvtap"); | |
1203 | if (IS_ERR(macvtap_class)) { | |
1204 | err = PTR_ERR(macvtap_class); | |
1205 | goto out3; | |
1206 | } | |
1207 | ||
9bf1907f | 1208 | err = register_netdevice_notifier(&macvtap_notifier_block); |
20d29d7a AB |
1209 | if (err) |
1210 | goto out4; | |
1211 | ||
9bf1907f EB |
1212 | err = macvlan_link_register(&macvtap_link_ops); |
1213 | if (err) | |
1214 | goto out5; | |
1215 | ||
20d29d7a AB |
1216 | return 0; |
1217 | ||
9bf1907f EB |
1218 | out5: |
1219 | unregister_netdevice_notifier(&macvtap_notifier_block); | |
20d29d7a AB |
1220 | out4: |
1221 | class_unregister(macvtap_class); | |
1222 | out3: | |
1223 | cdev_del(&macvtap_cdev); | |
1224 | out2: | |
1225 | unregister_chrdev_region(macvtap_major, MACVTAP_NUM_DEVS); | |
1226 | out1: | |
1227 | return err; | |
1228 | } | |
1229 | module_init(macvtap_init); | |
1230 | ||
1231 | static void macvtap_exit(void) | |
1232 | { | |
1233 | rtnl_link_unregister(&macvtap_link_ops); | |
9bf1907f | 1234 | unregister_netdevice_notifier(&macvtap_notifier_block); |
20d29d7a AB |
1235 | class_unregister(macvtap_class); |
1236 | cdev_del(&macvtap_cdev); | |
1237 | unregister_chrdev_region(macvtap_major, MACVTAP_NUM_DEVS); | |
1238 | } | |
1239 | module_exit(macvtap_exit); | |
1240 | ||
1241 | MODULE_ALIAS_RTNL_LINK("macvtap"); | |
1242 | MODULE_AUTHOR("Arnd Bergmann <arnd@arndb.de>"); | |
1243 | MODULE_LICENSE("GPL"); |