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 | * | |
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 | ||
20d29d7a AB |
68 | /* |
69 | * RCU usage: | |
02df55d2 AB |
70 | * The macvtap_queue and the macvlan_dev are loosely coupled, the |
71 | * pointers from one to the other can only be read while rcu_read_lock | |
72 | * or macvtap_lock is held. | |
20d29d7a | 73 | * |
02df55d2 AB |
74 | * Both the file and the macvlan_dev hold a reference on the macvtap_queue |
75 | * through sock_hold(&q->sk). When the macvlan_dev goes away first, | |
76 | * q->vlan becomes inaccessible. When the files gets closed, | |
77 | * macvtap_get_queue() fails. | |
20d29d7a | 78 | * |
02df55d2 AB |
79 | * There may still be references to the struct sock inside of the |
80 | * queue from outbound SKBs, but these never reference back to the | |
81 | * file or the dev. The data structure is freed through __sk_free | |
82 | * when both our references and any pending SKBs are gone. | |
20d29d7a AB |
83 | */ |
84 | static DEFINE_SPINLOCK(macvtap_lock); | |
85 | ||
815f236d | 86 | static int macvtap_enable_queue(struct net_device *dev, struct file *file, |
20d29d7a | 87 | struct macvtap_queue *q) |
815f236d JW |
88 | { |
89 | struct macvlan_dev *vlan = netdev_priv(dev); | |
90 | int err = -EINVAL; | |
91 | ||
92 | spin_lock(&macvtap_lock); | |
93 | ||
94 | if (q->enabled) | |
95 | goto out; | |
96 | ||
97 | err = 0; | |
98 | rcu_assign_pointer(vlan->taps[vlan->numvtaps], q); | |
99 | q->queue_index = vlan->numvtaps; | |
100 | q->enabled = true; | |
101 | ||
102 | vlan->numvtaps++; | |
103 | out: | |
104 | spin_unlock(&macvtap_lock); | |
105 | return err; | |
106 | } | |
107 | ||
108 | static int macvtap_set_queue(struct net_device *dev, struct file *file, | |
109 | struct macvtap_queue *q) | |
20d29d7a AB |
110 | { |
111 | struct macvlan_dev *vlan = netdev_priv(dev); | |
112 | int err = -EBUSY; | |
113 | ||
114 | spin_lock(&macvtap_lock); | |
815f236d | 115 | if (vlan->numqueues == MAX_MACVTAP_QUEUES) |
20d29d7a AB |
116 | goto out; |
117 | ||
118 | err = 0; | |
02df55d2 | 119 | rcu_assign_pointer(q->vlan, vlan); |
376b1aab | 120 | rcu_assign_pointer(vlan->taps[vlan->numvtaps], q); |
02df55d2 | 121 | sock_hold(&q->sk); |
20d29d7a AB |
122 | |
123 | q->file = file; | |
376b1aab | 124 | q->queue_index = vlan->numvtaps; |
815f236d | 125 | q->enabled = true; |
02df55d2 | 126 | file->private_data = q; |
815f236d | 127 | list_add_tail(&q->next, &vlan->queue_list); |
20d29d7a | 128 | |
1565c7c1 | 129 | vlan->numvtaps++; |
815f236d | 130 | vlan->numqueues++; |
1565c7c1 | 131 | |
20d29d7a AB |
132 | out: |
133 | spin_unlock(&macvtap_lock); | |
134 | return err; | |
135 | } | |
136 | ||
815f236d JW |
137 | static int __macvtap_disable_queue(struct macvtap_queue *q) |
138 | { | |
139 | struct macvlan_dev *vlan; | |
140 | struct macvtap_queue *nq; | |
141 | ||
142 | vlan = rcu_dereference_protected(q->vlan, | |
143 | lockdep_is_held(&macvtap_lock)); | |
144 | ||
145 | if (!q->enabled) | |
146 | return -EINVAL; | |
147 | ||
148 | if (vlan) { | |
149 | int index = q->queue_index; | |
150 | BUG_ON(index >= vlan->numvtaps); | |
151 | nq = rcu_dereference_protected(vlan->taps[vlan->numvtaps - 1], | |
152 | lockdep_is_held(&macvtap_lock)); | |
153 | nq->queue_index = index; | |
154 | ||
155 | rcu_assign_pointer(vlan->taps[index], nq); | |
156 | RCU_INIT_POINTER(vlan->taps[vlan->numvtaps - 1], NULL); | |
157 | q->enabled = false; | |
158 | ||
159 | vlan->numvtaps--; | |
160 | } | |
161 | ||
162 | return 0; | |
163 | } | |
164 | ||
165 | static int macvtap_disable_queue(struct macvtap_queue *q) | |
166 | { | |
167 | int err; | |
168 | ||
169 | spin_lock(&macvtap_lock); | |
170 | err = __macvtap_disable_queue(q); | |
171 | spin_unlock(&macvtap_lock); | |
172 | ||
173 | return err; | |
174 | } | |
175 | ||
20d29d7a | 176 | /* |
02df55d2 AB |
177 | * The file owning the queue got closed, give up both |
178 | * the reference that the files holds as well as the | |
179 | * one from the macvlan_dev if that still exists. | |
20d29d7a AB |
180 | * |
181 | * Using the spinlock makes sure that we don't get | |
182 | * to the queue again after destroying it. | |
20d29d7a | 183 | */ |
02df55d2 | 184 | static void macvtap_put_queue(struct macvtap_queue *q) |
20d29d7a | 185 | { |
02df55d2 | 186 | struct macvlan_dev *vlan; |
20d29d7a AB |
187 | |
188 | spin_lock(&macvtap_lock); | |
13707f9e ED |
189 | vlan = rcu_dereference_protected(q->vlan, |
190 | lockdep_is_held(&macvtap_lock)); | |
02df55d2 | 191 | if (vlan) { |
815f236d JW |
192 | if (q->enabled) |
193 | BUG_ON(__macvtap_disable_queue(q)); | |
376b1aab | 194 | |
815f236d | 195 | vlan->numqueues--; |
2cfa5a04 | 196 | RCU_INIT_POINTER(q->vlan, NULL); |
02df55d2 | 197 | sock_put(&q->sk); |
815f236d | 198 | list_del_init(&q->next); |
20d29d7a AB |
199 | } |
200 | ||
20d29d7a AB |
201 | spin_unlock(&macvtap_lock); |
202 | ||
203 | synchronize_rcu(); | |
204 | sock_put(&q->sk); | |
205 | } | |
206 | ||
207 | /* | |
1565c7c1 KK |
208 | * Select a queue based on the rxq of the device on which this packet |
209 | * arrived. If the incoming device is not mq, calculate a flow hash | |
210 | * to select a queue. If all fails, find the first available queue. | |
211 | * Cache vlan->numvtaps since it can become zero during the execution | |
212 | * of this function. | |
20d29d7a AB |
213 | */ |
214 | static struct macvtap_queue *macvtap_get_queue(struct net_device *dev, | |
215 | struct sk_buff *skb) | |
216 | { | |
217 | struct macvlan_dev *vlan = netdev_priv(dev); | |
1565c7c1 | 218 | struct macvtap_queue *tap = NULL; |
815f236d JW |
219 | /* Access to taps array is protected by rcu, but access to numvtaps |
220 | * isn't. Below we use it to lookup a queue, but treat it as a hint | |
221 | * and validate that the result isn't NULL - in case we are | |
222 | * racing against queue removal. | |
223 | */ | |
ed0483fa | 224 | int numvtaps = ACCESS_ONCE(vlan->numvtaps); |
1565c7c1 KK |
225 | __u32 rxq; |
226 | ||
227 | if (!numvtaps) | |
228 | goto out; | |
229 | ||
ef0002b5 KK |
230 | /* Check if we can use flow to select a queue */ |
231 | rxq = skb_get_rxhash(skb); | |
232 | if (rxq) { | |
233 | tap = rcu_dereference(vlan->taps[rxq % numvtaps]); | |
376b1aab | 234 | goto out; |
ef0002b5 KK |
235 | } |
236 | ||
1565c7c1 KK |
237 | if (likely(skb_rx_queue_recorded(skb))) { |
238 | rxq = skb_get_rx_queue(skb); | |
20d29d7a | 239 | |
1565c7c1 KK |
240 | while (unlikely(rxq >= numvtaps)) |
241 | rxq -= numvtaps; | |
242 | ||
243 | tap = rcu_dereference(vlan->taps[rxq]); | |
376b1aab | 244 | goto out; |
1565c7c1 KK |
245 | } |
246 | ||
376b1aab | 247 | tap = rcu_dereference(vlan->taps[0]); |
1565c7c1 KK |
248 | out: |
249 | return tap; | |
20d29d7a AB |
250 | } |
251 | ||
02df55d2 AB |
252 | /* |
253 | * The net_device is going away, give up the reference | |
1565c7c1 KK |
254 | * that it holds on all queues and safely set the pointer |
255 | * from the queues to NULL. | |
02df55d2 | 256 | */ |
20d29d7a AB |
257 | static void macvtap_del_queues(struct net_device *dev) |
258 | { | |
259 | struct macvlan_dev *vlan = netdev_priv(dev); | |
815f236d | 260 | struct macvtap_queue *q, *tmp, *qlist[MAX_MACVTAP_QUEUES]; |
1565c7c1 | 261 | int i, j = 0; |
02df55d2 AB |
262 | |
263 | spin_lock(&macvtap_lock); | |
815f236d JW |
264 | list_for_each_entry_safe(q, tmp, &vlan->queue_list, next) { |
265 | list_del_init(&q->next); | |
376b1aab | 266 | qlist[j++] = q; |
376b1aab | 267 | RCU_INIT_POINTER(q->vlan, NULL); |
815f236d JW |
268 | if (q->enabled) |
269 | vlan->numvtaps--; | |
270 | vlan->numqueues--; | |
564517e8 | 271 | } |
815f236d JW |
272 | for (i = 0; i < vlan->numvtaps; i++) |
273 | RCU_INIT_POINTER(vlan->taps[i], NULL); | |
274 | BUG_ON(vlan->numvtaps); | |
275 | BUG_ON(vlan->numqueues); | |
99f34b38 EB |
276 | /* guarantee that any future macvtap_set_queue will fail */ |
277 | vlan->numvtaps = MAX_MACVTAP_QUEUES; | |
02df55d2 AB |
278 | spin_unlock(&macvtap_lock); |
279 | ||
280 | synchronize_rcu(); | |
1565c7c1 KK |
281 | |
282 | for (--j; j >= 0; j--) | |
283 | sock_put(&qlist[j]->sk); | |
20d29d7a AB |
284 | } |
285 | ||
286 | /* | |
287 | * Forward happens for data that gets sent from one macvlan | |
288 | * endpoint to another one in bridge mode. We just take | |
289 | * the skb and put it into the receive queue. | |
290 | */ | |
291 | static int macvtap_forward(struct net_device *dev, struct sk_buff *skb) | |
292 | { | |
293 | struct macvtap_queue *q = macvtap_get_queue(dev, skb); | |
294 | if (!q) | |
8a35747a HX |
295 | goto drop; |
296 | ||
297 | if (skb_queue_len(&q->sk.sk_receive_queue) >= dev->tx_queue_len) | |
298 | goto drop; | |
20d29d7a AB |
299 | |
300 | skb_queue_tail(&q->sk.sk_receive_queue, skb); | |
4a4771a5 | 301 | wake_up_interruptible_poll(sk_sleep(&q->sk), POLLIN | POLLRDNORM | POLLRDBAND); |
8a35747a HX |
302 | return NET_RX_SUCCESS; |
303 | ||
304 | drop: | |
305 | kfree_skb(skb); | |
306 | return NET_RX_DROP; | |
20d29d7a AB |
307 | } |
308 | ||
309 | /* | |
310 | * Receive is for data from the external interface (lowerdev), | |
311 | * in case of macvtap, we can treat that the same way as | |
312 | * forward, which macvlan cannot. | |
313 | */ | |
314 | static int macvtap_receive(struct sk_buff *skb) | |
315 | { | |
316 | skb_push(skb, ETH_HLEN); | |
317 | return macvtap_forward(skb->dev, skb); | |
318 | } | |
319 | ||
e09eff7f EB |
320 | static int macvtap_get_minor(struct macvlan_dev *vlan) |
321 | { | |
322 | int retval = -ENOMEM; | |
e09eff7f EB |
323 | |
324 | mutex_lock(&minor_lock); | |
ec09ebc1 TH |
325 | retval = idr_alloc(&minor_idr, vlan, 1, MACVTAP_NUM_DEVS, GFP_KERNEL); |
326 | if (retval >= 0) { | |
327 | vlan->minor = retval; | |
328 | } else if (retval == -ENOSPC) { | |
e09eff7f EB |
329 | printk(KERN_ERR "too many macvtap devices\n"); |
330 | retval = -EINVAL; | |
e09eff7f | 331 | } |
e09eff7f | 332 | mutex_unlock(&minor_lock); |
ec09ebc1 | 333 | return retval < 0 ? retval : 0; |
e09eff7f EB |
334 | } |
335 | ||
336 | static void macvtap_free_minor(struct macvlan_dev *vlan) | |
337 | { | |
338 | mutex_lock(&minor_lock); | |
339 | if (vlan->minor) { | |
340 | idr_remove(&minor_idr, vlan->minor); | |
341 | vlan->minor = 0; | |
342 | } | |
343 | mutex_unlock(&minor_lock); | |
344 | } | |
345 | ||
346 | static struct net_device *dev_get_by_macvtap_minor(int minor) | |
347 | { | |
348 | struct net_device *dev = NULL; | |
349 | struct macvlan_dev *vlan; | |
350 | ||
351 | mutex_lock(&minor_lock); | |
352 | vlan = idr_find(&minor_idr, minor); | |
353 | if (vlan) { | |
354 | dev = vlan->dev; | |
355 | dev_hold(dev); | |
356 | } | |
357 | mutex_unlock(&minor_lock); | |
358 | return dev; | |
359 | } | |
360 | ||
20d29d7a AB |
361 | static int macvtap_newlink(struct net *src_net, |
362 | struct net_device *dev, | |
363 | struct nlattr *tb[], | |
364 | struct nlattr *data[]) | |
365 | { | |
815f236d JW |
366 | struct macvlan_dev *vlan = netdev_priv(dev); |
367 | INIT_LIST_HEAD(&vlan->queue_list); | |
368 | ||
9bf1907f EB |
369 | /* Don't put anything that may fail after macvlan_common_newlink |
370 | * because we can't undo what it does. | |
371 | */ | |
372 | return macvlan_common_newlink(src_net, dev, tb, data, | |
373 | macvtap_receive, macvtap_forward); | |
20d29d7a AB |
374 | } |
375 | ||
376 | static void macvtap_dellink(struct net_device *dev, | |
377 | struct list_head *head) | |
378 | { | |
20d29d7a AB |
379 | macvtap_del_queues(dev); |
380 | macvlan_dellink(dev, head); | |
381 | } | |
382 | ||
8a35747a HX |
383 | static void macvtap_setup(struct net_device *dev) |
384 | { | |
385 | macvlan_common_setup(dev); | |
386 | dev->tx_queue_len = TUN_READQ_SIZE; | |
387 | } | |
388 | ||
20d29d7a AB |
389 | static struct rtnl_link_ops macvtap_link_ops __read_mostly = { |
390 | .kind = "macvtap", | |
8a35747a | 391 | .setup = macvtap_setup, |
20d29d7a AB |
392 | .newlink = macvtap_newlink, |
393 | .dellink = macvtap_dellink, | |
394 | }; | |
395 | ||
396 | ||
397 | static void macvtap_sock_write_space(struct sock *sk) | |
398 | { | |
43815482 ED |
399 | wait_queue_head_t *wqueue; |
400 | ||
20d29d7a AB |
401 | if (!sock_writeable(sk) || |
402 | !test_and_clear_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags)) | |
403 | return; | |
404 | ||
43815482 ED |
405 | wqueue = sk_sleep(sk); |
406 | if (wqueue && waitqueue_active(wqueue)) | |
407 | wake_up_interruptible_poll(wqueue, POLLOUT | POLLWRNORM | POLLWRBAND); | |
20d29d7a AB |
408 | } |
409 | ||
2259fef0 EB |
410 | static void macvtap_sock_destruct(struct sock *sk) |
411 | { | |
412 | skb_queue_purge(&sk->sk_receive_queue); | |
413 | } | |
414 | ||
20d29d7a AB |
415 | static int macvtap_open(struct inode *inode, struct file *file) |
416 | { | |
417 | struct net *net = current->nsproxy->net_ns; | |
e09eff7f | 418 | struct net_device *dev = dev_get_by_macvtap_minor(iminor(inode)); |
20d29d7a AB |
419 | struct macvtap_queue *q; |
420 | int err; | |
421 | ||
422 | err = -ENODEV; | |
423 | if (!dev) | |
424 | goto out; | |
425 | ||
20d29d7a AB |
426 | err = -ENOMEM; |
427 | q = (struct macvtap_queue *)sk_alloc(net, AF_UNSPEC, GFP_KERNEL, | |
428 | &macvtap_proto); | |
429 | if (!q) | |
430 | goto out; | |
431 | ||
d9a90a31 | 432 | RCU_INIT_POINTER(q->sock.wq, &q->wq); |
43815482 | 433 | init_waitqueue_head(&q->wq.wait); |
20d29d7a AB |
434 | q->sock.type = SOCK_RAW; |
435 | q->sock.state = SS_CONNECTED; | |
501c774c AB |
436 | q->sock.file = file; |
437 | q->sock.ops = &macvtap_socket_ops; | |
20d29d7a | 438 | sock_init_data(&q->sock, &q->sk); |
20d29d7a | 439 | q->sk.sk_write_space = macvtap_sock_write_space; |
2259fef0 | 440 | q->sk.sk_destruct = macvtap_sock_destruct; |
b9fb9ee0 | 441 | q->flags = IFF_VNET_HDR | IFF_NO_PI | IFF_TAP; |
55afbd08 | 442 | q->vnet_hdr_sz = sizeof(struct virtio_net_hdr); |
20d29d7a | 443 | |
97bc3633 SM |
444 | /* |
445 | * so far only KVM virtio_net uses macvtap, enable zero copy between | |
446 | * guest kernel and host kernel when lower device supports zerocopy | |
047af9cf EB |
447 | * |
448 | * The macvlan supports zerocopy iff the lower device supports zero | |
449 | * copy so we don't have to look at the lower device directly. | |
97bc3633 | 450 | */ |
047af9cf EB |
451 | if ((dev->features & NETIF_F_HIGHDMA) && (dev->features & NETIF_F_SG)) |
452 | sock_set_flag(&q->sk, SOCK_ZEROCOPY); | |
97bc3633 | 453 | |
20d29d7a AB |
454 | err = macvtap_set_queue(dev, file, q); |
455 | if (err) | |
456 | sock_put(&q->sk); | |
457 | ||
458 | out: | |
459 | if (dev) | |
460 | dev_put(dev); | |
461 | ||
462 | return err; | |
463 | } | |
464 | ||
465 | static int macvtap_release(struct inode *inode, struct file *file) | |
466 | { | |
02df55d2 AB |
467 | struct macvtap_queue *q = file->private_data; |
468 | macvtap_put_queue(q); | |
20d29d7a AB |
469 | return 0; |
470 | } | |
471 | ||
472 | static unsigned int macvtap_poll(struct file *file, poll_table * wait) | |
473 | { | |
02df55d2 | 474 | struct macvtap_queue *q = file->private_data; |
20d29d7a AB |
475 | unsigned int mask = POLLERR; |
476 | ||
477 | if (!q) | |
478 | goto out; | |
479 | ||
480 | mask = 0; | |
43815482 | 481 | poll_wait(file, &q->wq.wait, wait); |
20d29d7a AB |
482 | |
483 | if (!skb_queue_empty(&q->sk.sk_receive_queue)) | |
484 | mask |= POLLIN | POLLRDNORM; | |
485 | ||
486 | if (sock_writeable(&q->sk) || | |
487 | (!test_and_set_bit(SOCK_ASYNC_NOSPACE, &q->sock.flags) && | |
488 | sock_writeable(&q->sk))) | |
489 | mask |= POLLOUT | POLLWRNORM; | |
490 | ||
491 | out: | |
20d29d7a AB |
492 | return mask; |
493 | } | |
494 | ||
b9fb9ee0 AB |
495 | static inline struct sk_buff *macvtap_alloc_skb(struct sock *sk, size_t prepad, |
496 | size_t len, size_t linear, | |
497 | int noblock, int *err) | |
498 | { | |
499 | struct sk_buff *skb; | |
500 | ||
501 | /* Under a page? Don't bother with paged skb. */ | |
502 | if (prepad + len < PAGE_SIZE || !linear) | |
503 | linear = len; | |
504 | ||
505 | skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock, | |
506 | err); | |
507 | if (!skb) | |
508 | return NULL; | |
509 | ||
510 | skb_reserve(skb, prepad); | |
511 | skb_put(skb, linear); | |
512 | skb->data_len = len - linear; | |
513 | skb->len += len - linear; | |
514 | ||
515 | return skb; | |
516 | } | |
517 | ||
97bc3633 SM |
518 | /* set skb frags from iovec, this can move to core network code for reuse */ |
519 | static int zerocopy_sg_from_iovec(struct sk_buff *skb, const struct iovec *from, | |
520 | int offset, size_t count) | |
521 | { | |
522 | int len = iov_length(from, count) - offset; | |
523 | int copy = skb_headlen(skb); | |
524 | int size, offset1 = 0; | |
525 | int i = 0; | |
97bc3633 SM |
526 | |
527 | /* Skip over from offset */ | |
528 | while (count && (offset >= from->iov_len)) { | |
529 | offset -= from->iov_len; | |
530 | ++from; | |
531 | --count; | |
532 | } | |
533 | ||
534 | /* copy up to skb headlen */ | |
535 | while (count && (copy > 0)) { | |
536 | size = min_t(unsigned int, copy, from->iov_len - offset); | |
537 | if (copy_from_user(skb->data + offset1, from->iov_base + offset, | |
538 | size)) | |
539 | return -EFAULT; | |
540 | if (copy > size) { | |
541 | ++from; | |
542 | --count; | |
3afc9621 JW |
543 | offset = 0; |
544 | } else | |
545 | offset += size; | |
97bc3633 SM |
546 | copy -= size; |
547 | offset1 += size; | |
97bc3633 SM |
548 | } |
549 | ||
550 | if (len == offset1) | |
551 | return 0; | |
552 | ||
553 | while (count--) { | |
554 | struct page *page[MAX_SKB_FRAGS]; | |
555 | int num_pages; | |
556 | unsigned long base; | |
4ef67ebe | 557 | unsigned long truesize; |
97bc3633 | 558 | |
3afc9621 | 559 | len = from->iov_len - offset; |
97bc3633 | 560 | if (!len) { |
3afc9621 | 561 | offset = 0; |
97bc3633 SM |
562 | ++from; |
563 | continue; | |
564 | } | |
3afc9621 | 565 | base = (unsigned long)from->iov_base + offset; |
97bc3633 | 566 | size = ((base & ~PAGE_MASK) + len + ~PAGE_MASK) >> PAGE_SHIFT; |
b92946e2 JW |
567 | if (i + size > MAX_SKB_FRAGS) |
568 | return -EMSGSIZE; | |
97bc3633 | 569 | num_pages = get_user_pages_fast(base, size, 0, &page[i]); |
b92946e2 | 570 | if (num_pages != size) { |
02ce04bb JW |
571 | for (i = 0; i < num_pages; i++) |
572 | put_page(page[i]); | |
97bc3633 | 573 | return -EFAULT; |
02ce04bb | 574 | } |
4ef67ebe | 575 | truesize = size * PAGE_SIZE; |
97bc3633 SM |
576 | skb->data_len += len; |
577 | skb->len += len; | |
4ef67ebe JW |
578 | skb->truesize += truesize; |
579 | atomic_add(truesize, &skb->sk->sk_wmem_alloc); | |
97bc3633 | 580 | while (len) { |
653fc915 JW |
581 | int off = base & ~PAGE_MASK; |
582 | int size = min_t(int, len, PAGE_SIZE - off); | |
583 | __skb_fill_page_desc(skb, i, page[i], off, size); | |
97bc3633 SM |
584 | skb_shinfo(skb)->nr_frags++; |
585 | /* increase sk_wmem_alloc */ | |
653fc915 JW |
586 | base += size; |
587 | len -= size; | |
97bc3633 SM |
588 | i++; |
589 | } | |
3afc9621 | 590 | offset = 0; |
97bc3633 SM |
591 | ++from; |
592 | } | |
593 | return 0; | |
594 | } | |
595 | ||
b9fb9ee0 AB |
596 | /* |
597 | * macvtap_skb_from_vnet_hdr and macvtap_skb_to_vnet_hdr should | |
598 | * be shared with the tun/tap driver. | |
599 | */ | |
600 | static int macvtap_skb_from_vnet_hdr(struct sk_buff *skb, | |
601 | struct virtio_net_hdr *vnet_hdr) | |
602 | { | |
603 | unsigned short gso_type = 0; | |
604 | if (vnet_hdr->gso_type != VIRTIO_NET_HDR_GSO_NONE) { | |
605 | switch (vnet_hdr->gso_type & ~VIRTIO_NET_HDR_GSO_ECN) { | |
606 | case VIRTIO_NET_HDR_GSO_TCPV4: | |
607 | gso_type = SKB_GSO_TCPV4; | |
608 | break; | |
609 | case VIRTIO_NET_HDR_GSO_TCPV6: | |
610 | gso_type = SKB_GSO_TCPV6; | |
611 | break; | |
612 | case VIRTIO_NET_HDR_GSO_UDP: | |
613 | gso_type = SKB_GSO_UDP; | |
614 | break; | |
615 | default: | |
616 | return -EINVAL; | |
617 | } | |
618 | ||
619 | if (vnet_hdr->gso_type & VIRTIO_NET_HDR_GSO_ECN) | |
620 | gso_type |= SKB_GSO_TCP_ECN; | |
621 | ||
622 | if (vnet_hdr->gso_size == 0) | |
623 | return -EINVAL; | |
624 | } | |
625 | ||
626 | if (vnet_hdr->flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) { | |
627 | if (!skb_partial_csum_set(skb, vnet_hdr->csum_start, | |
628 | vnet_hdr->csum_offset)) | |
629 | return -EINVAL; | |
630 | } | |
631 | ||
632 | if (vnet_hdr->gso_type != VIRTIO_NET_HDR_GSO_NONE) { | |
633 | skb_shinfo(skb)->gso_size = vnet_hdr->gso_size; | |
c9af6db4 | 634 | skb_shinfo(skb)->gso_type = gso_type; |
b9fb9ee0 AB |
635 | |
636 | /* Header must be checked, and gso_segs computed. */ | |
637 | skb_shinfo(skb)->gso_type |= SKB_GSO_DODGY; | |
638 | skb_shinfo(skb)->gso_segs = 0; | |
639 | } | |
640 | return 0; | |
641 | } | |
642 | ||
643 | static int macvtap_skb_to_vnet_hdr(const struct sk_buff *skb, | |
644 | struct virtio_net_hdr *vnet_hdr) | |
645 | { | |
646 | memset(vnet_hdr, 0, sizeof(*vnet_hdr)); | |
647 | ||
648 | if (skb_is_gso(skb)) { | |
649 | struct skb_shared_info *sinfo = skb_shinfo(skb); | |
650 | ||
651 | /* This is a hint as to how much should be linear. */ | |
652 | vnet_hdr->hdr_len = skb_headlen(skb); | |
653 | vnet_hdr->gso_size = sinfo->gso_size; | |
654 | if (sinfo->gso_type & SKB_GSO_TCPV4) | |
655 | vnet_hdr->gso_type = VIRTIO_NET_HDR_GSO_TCPV4; | |
656 | else if (sinfo->gso_type & SKB_GSO_TCPV6) | |
657 | vnet_hdr->gso_type = VIRTIO_NET_HDR_GSO_TCPV6; | |
658 | else if (sinfo->gso_type & SKB_GSO_UDP) | |
659 | vnet_hdr->gso_type = VIRTIO_NET_HDR_GSO_UDP; | |
660 | else | |
661 | BUG(); | |
662 | if (sinfo->gso_type & SKB_GSO_TCP_ECN) | |
663 | vnet_hdr->gso_type |= VIRTIO_NET_HDR_GSO_ECN; | |
664 | } else | |
665 | vnet_hdr->gso_type = VIRTIO_NET_HDR_GSO_NONE; | |
666 | ||
667 | if (skb->ip_summed == CHECKSUM_PARTIAL) { | |
668 | vnet_hdr->flags = VIRTIO_NET_HDR_F_NEEDS_CSUM; | |
55508d60 | 669 | vnet_hdr->csum_start = skb_checksum_start_offset(skb); |
b9fb9ee0 | 670 | vnet_hdr->csum_offset = skb->csum_offset; |
10a8d94a JW |
671 | } else if (skb->ip_summed == CHECKSUM_UNNECESSARY) { |
672 | vnet_hdr->flags = VIRTIO_NET_HDR_F_DATA_VALID; | |
b9fb9ee0 AB |
673 | } /* else everything is zero */ |
674 | ||
675 | return 0; | |
676 | } | |
677 | ||
678 | ||
20d29d7a | 679 | /* Get packet from user space buffer */ |
97bc3633 SM |
680 | static ssize_t macvtap_get_user(struct macvtap_queue *q, struct msghdr *m, |
681 | const struct iovec *iv, unsigned long total_len, | |
682 | size_t count, int noblock) | |
20d29d7a AB |
683 | { |
684 | struct sk_buff *skb; | |
02df55d2 | 685 | struct macvlan_dev *vlan; |
97bc3633 | 686 | unsigned long len = total_len; |
20d29d7a | 687 | int err; |
b9fb9ee0 AB |
688 | struct virtio_net_hdr vnet_hdr = { 0 }; |
689 | int vnet_hdr_len = 0; | |
b92946e2 | 690 | int copylen = 0; |
97bc3633 | 691 | bool zerocopy = false; |
b9fb9ee0 AB |
692 | |
693 | if (q->flags & IFF_VNET_HDR) { | |
55afbd08 | 694 | vnet_hdr_len = q->vnet_hdr_sz; |
b9fb9ee0 AB |
695 | |
696 | err = -EINVAL; | |
ce3c8692 | 697 | if (len < vnet_hdr_len) |
b9fb9ee0 | 698 | goto err; |
ce3c8692 | 699 | len -= vnet_hdr_len; |
b9fb9ee0 AB |
700 | |
701 | err = memcpy_fromiovecend((void *)&vnet_hdr, iv, 0, | |
55afbd08 | 702 | sizeof(vnet_hdr)); |
b9fb9ee0 AB |
703 | if (err < 0) |
704 | goto err; | |
705 | if ((vnet_hdr.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) && | |
706 | vnet_hdr.csum_start + vnet_hdr.csum_offset + 2 > | |
707 | vnet_hdr.hdr_len) | |
708 | vnet_hdr.hdr_len = vnet_hdr.csum_start + | |
709 | vnet_hdr.csum_offset + 2; | |
710 | err = -EINVAL; | |
711 | if (vnet_hdr.hdr_len > len) | |
712 | goto err; | |
713 | } | |
20d29d7a | 714 | |
b9fb9ee0 | 715 | err = -EINVAL; |
20d29d7a | 716 | if (unlikely(len < ETH_HLEN)) |
b9fb9ee0 | 717 | goto err; |
20d29d7a | 718 | |
b92946e2 JW |
719 | err = -EMSGSIZE; |
720 | if (unlikely(count > UIO_MAXIOV)) | |
721 | goto err; | |
722 | ||
97bc3633 SM |
723 | if (m && m->msg_control && sock_flag(&q->sk, SOCK_ZEROCOPY)) |
724 | zerocopy = true; | |
725 | ||
726 | if (zerocopy) { | |
b92946e2 JW |
727 | /* Userspace may produce vectors with count greater than |
728 | * MAX_SKB_FRAGS, so we need to linearize parts of the skb | |
729 | * to let the rest of data to be fit in the frags. | |
730 | */ | |
731 | if (count > MAX_SKB_FRAGS) { | |
732 | copylen = iov_length(iv, count - MAX_SKB_FRAGS); | |
733 | if (copylen < vnet_hdr_len) | |
734 | copylen = 0; | |
735 | else | |
736 | copylen -= vnet_hdr_len; | |
737 | } | |
97bc3633 SM |
738 | /* There are 256 bytes to be copied in skb, so there is enough |
739 | * room for skb expand head in case it is used. | |
740 | * The rest buffer is mapped from userspace. | |
741 | */ | |
b92946e2 JW |
742 | if (copylen < vnet_hdr.hdr_len) |
743 | copylen = vnet_hdr.hdr_len; | |
97bc3633 SM |
744 | if (!copylen) |
745 | copylen = GOODCOPY_LEN; | |
746 | } else | |
747 | copylen = len; | |
748 | ||
749 | skb = macvtap_alloc_skb(&q->sk, NET_IP_ALIGN, copylen, | |
750 | vnet_hdr.hdr_len, noblock, &err); | |
02df55d2 AB |
751 | if (!skb) |
752 | goto err; | |
20d29d7a | 753 | |
01d6657b | 754 | if (zerocopy) |
97bc3633 | 755 | err = zerocopy_sg_from_iovec(skb, iv, vnet_hdr_len, count); |
01d6657b | 756 | else |
97bc3633 SM |
757 | err = skb_copy_datagram_from_iovec(skb, 0, iv, vnet_hdr_len, |
758 | len); | |
02df55d2 | 759 | if (err) |
b9fb9ee0 | 760 | goto err_kfree; |
20d29d7a AB |
761 | |
762 | skb_set_network_header(skb, ETH_HLEN); | |
b9fb9ee0 AB |
763 | skb_reset_mac_header(skb); |
764 | skb->protocol = eth_hdr(skb)->h_proto; | |
765 | ||
766 | if (vnet_hdr_len) { | |
767 | err = macvtap_skb_from_vnet_hdr(skb, &vnet_hdr); | |
768 | if (err) | |
769 | goto err_kfree; | |
770 | } | |
771 | ||
40893fd0 | 772 | skb_probe_transport_header(skb, ETH_HLEN); |
9b4d669b | 773 | |
02df55d2 | 774 | rcu_read_lock_bh(); |
13707f9e | 775 | vlan = rcu_dereference_bh(q->vlan); |
97bc3633 | 776 | /* copy skb_ubuf_info for callback when skb has no error */ |
01d6657b | 777 | if (zerocopy) { |
97bc3633 | 778 | skb_shinfo(skb)->destructor_arg = m->msg_control; |
01d6657b | 779 | skb_shinfo(skb)->tx_flags |= SKBTX_DEV_ZEROCOPY; |
c9af6db4 | 780 | skb_shinfo(skb)->tx_flags |= SKBTX_SHARED_FRAG; |
01d6657b | 781 | } |
02df55d2 AB |
782 | if (vlan) |
783 | macvlan_start_xmit(skb, vlan->dev); | |
784 | else | |
785 | kfree_skb(skb); | |
786 | rcu_read_unlock_bh(); | |
20d29d7a | 787 | |
97bc3633 | 788 | return total_len; |
02df55d2 | 789 | |
b9fb9ee0 AB |
790 | err_kfree: |
791 | kfree_skb(skb); | |
792 | ||
02df55d2 AB |
793 | err: |
794 | rcu_read_lock_bh(); | |
13707f9e | 795 | vlan = rcu_dereference_bh(q->vlan); |
02df55d2 | 796 | if (vlan) |
1ac9ad13 | 797 | vlan->dev->stats.tx_dropped++; |
02df55d2 AB |
798 | rcu_read_unlock_bh(); |
799 | ||
02df55d2 | 800 | return err; |
20d29d7a AB |
801 | } |
802 | ||
803 | static ssize_t macvtap_aio_write(struct kiocb *iocb, const struct iovec *iv, | |
804 | unsigned long count, loff_t pos) | |
805 | { | |
806 | struct file *file = iocb->ki_filp; | |
807 | ssize_t result = -ENOLINK; | |
02df55d2 | 808 | struct macvtap_queue *q = file->private_data; |
20d29d7a | 809 | |
97bc3633 SM |
810 | result = macvtap_get_user(q, NULL, iv, iov_length(iv, count), count, |
811 | file->f_flags & O_NONBLOCK); | |
20d29d7a AB |
812 | return result; |
813 | } | |
814 | ||
815 | /* Put packet to the user space buffer */ | |
816 | static ssize_t macvtap_put_user(struct macvtap_queue *q, | |
817 | const struct sk_buff *skb, | |
818 | const struct iovec *iv, int len) | |
819 | { | |
02df55d2 | 820 | struct macvlan_dev *vlan; |
20d29d7a | 821 | int ret; |
b9fb9ee0 | 822 | int vnet_hdr_len = 0; |
f09e2249 BG |
823 | int vlan_offset = 0; |
824 | int copied; | |
b9fb9ee0 AB |
825 | |
826 | if (q->flags & IFF_VNET_HDR) { | |
827 | struct virtio_net_hdr vnet_hdr; | |
55afbd08 | 828 | vnet_hdr_len = q->vnet_hdr_sz; |
b9fb9ee0 AB |
829 | if ((len -= vnet_hdr_len) < 0) |
830 | return -EINVAL; | |
831 | ||
832 | ret = macvtap_skb_to_vnet_hdr(skb, &vnet_hdr); | |
833 | if (ret) | |
834 | return ret; | |
835 | ||
55afbd08 | 836 | if (memcpy_toiovecend(iv, (void *)&vnet_hdr, 0, sizeof(vnet_hdr))) |
b9fb9ee0 AB |
837 | return -EFAULT; |
838 | } | |
f09e2249 BG |
839 | copied = vnet_hdr_len; |
840 | ||
841 | if (!vlan_tx_tag_present(skb)) | |
842 | len = min_t(int, skb->len, len); | |
843 | else { | |
844 | int copy; | |
845 | struct { | |
846 | __be16 h_vlan_proto; | |
847 | __be16 h_vlan_TCI; | |
848 | } veth; | |
849 | veth.h_vlan_proto = htons(ETH_P_8021Q); | |
850 | veth.h_vlan_TCI = htons(vlan_tx_tag_get(skb)); | |
851 | ||
852 | vlan_offset = offsetof(struct vlan_ethhdr, h_vlan_proto); | |
853 | len = min_t(int, skb->len + VLAN_HLEN, len); | |
854 | ||
855 | copy = min_t(int, vlan_offset, len); | |
856 | ret = skb_copy_datagram_const_iovec(skb, 0, iv, copied, copy); | |
857 | len -= copy; | |
858 | copied += copy; | |
859 | if (ret || !len) | |
860 | goto done; | |
861 | ||
862 | copy = min_t(int, sizeof(veth), len); | |
863 | ret = memcpy_toiovecend(iv, (void *)&veth, copied, copy); | |
864 | len -= copy; | |
865 | copied += copy; | |
866 | if (ret || !len) | |
867 | goto done; | |
868 | } | |
20d29d7a | 869 | |
f09e2249 BG |
870 | ret = skb_copy_datagram_const_iovec(skb, vlan_offset, iv, copied, len); |
871 | copied += len; | |
20d29d7a | 872 | |
f09e2249 | 873 | done: |
02df55d2 | 874 | rcu_read_lock_bh(); |
13707f9e | 875 | vlan = rcu_dereference_bh(q->vlan); |
501c774c | 876 | if (vlan) |
f09e2249 | 877 | macvlan_count_rx(vlan, copied - vnet_hdr_len, ret == 0, 0); |
02df55d2 | 878 | rcu_read_unlock_bh(); |
20d29d7a | 879 | |
f09e2249 | 880 | return ret ? ret : copied; |
20d29d7a AB |
881 | } |
882 | ||
501c774c AB |
883 | static ssize_t macvtap_do_read(struct macvtap_queue *q, struct kiocb *iocb, |
884 | const struct iovec *iv, unsigned long len, | |
885 | int noblock) | |
20d29d7a | 886 | { |
ccf7e72b | 887 | DEFINE_WAIT(wait); |
20d29d7a | 888 | struct sk_buff *skb; |
501c774c | 889 | ssize_t ret = 0; |
20d29d7a | 890 | |
20d29d7a | 891 | while (len) { |
89cee917 JW |
892 | if (!noblock) |
893 | prepare_to_wait(sk_sleep(&q->sk), &wait, | |
894 | TASK_INTERRUPTIBLE); | |
20d29d7a AB |
895 | |
896 | /* Read frames from the queue */ | |
897 | skb = skb_dequeue(&q->sk.sk_receive_queue); | |
898 | if (!skb) { | |
501c774c | 899 | if (noblock) { |
20d29d7a AB |
900 | ret = -EAGAIN; |
901 | break; | |
902 | } | |
903 | if (signal_pending(current)) { | |
904 | ret = -ERESTARTSYS; | |
905 | break; | |
906 | } | |
907 | /* Nothing to read, let's sleep */ | |
908 | schedule(); | |
909 | continue; | |
910 | } | |
911 | ret = macvtap_put_user(q, skb, iv, len); | |
912 | kfree_skb(skb); | |
913 | break; | |
914 | } | |
915 | ||
89cee917 JW |
916 | if (!noblock) |
917 | finish_wait(sk_sleep(&q->sk), &wait); | |
501c774c AB |
918 | return ret; |
919 | } | |
920 | ||
921 | static ssize_t macvtap_aio_read(struct kiocb *iocb, const struct iovec *iv, | |
922 | unsigned long count, loff_t pos) | |
923 | { | |
924 | struct file *file = iocb->ki_filp; | |
925 | struct macvtap_queue *q = file->private_data; | |
926 | ssize_t len, ret = 0; | |
20d29d7a | 927 | |
501c774c AB |
928 | len = iov_length(iv, count); |
929 | if (len < 0) { | |
930 | ret = -EINVAL; | |
931 | goto out; | |
932 | } | |
933 | ||
934 | ret = macvtap_do_read(q, iocb, iv, len, file->f_flags & O_NONBLOCK); | |
935 | ret = min_t(ssize_t, ret, len); /* XXX copied from tun.c. Why? */ | |
20d29d7a | 936 | out: |
20d29d7a AB |
937 | return ret; |
938 | } | |
939 | ||
8f475a31 JW |
940 | static struct macvlan_dev *macvtap_get_vlan(struct macvtap_queue *q) |
941 | { | |
942 | struct macvlan_dev *vlan; | |
943 | ||
944 | rcu_read_lock_bh(); | |
945 | vlan = rcu_dereference_bh(q->vlan); | |
946 | if (vlan) | |
947 | dev_hold(vlan->dev); | |
948 | rcu_read_unlock_bh(); | |
949 | ||
950 | return vlan; | |
951 | } | |
952 | ||
953 | static void macvtap_put_vlan(struct macvlan_dev *vlan) | |
954 | { | |
955 | dev_put(vlan->dev); | |
956 | } | |
957 | ||
815f236d JW |
958 | static int macvtap_ioctl_set_queue(struct file *file, unsigned int flags) |
959 | { | |
960 | struct macvtap_queue *q = file->private_data; | |
961 | struct macvlan_dev *vlan; | |
962 | int ret; | |
963 | ||
964 | vlan = macvtap_get_vlan(q); | |
965 | if (!vlan) | |
966 | return -EINVAL; | |
967 | ||
968 | if (flags & IFF_ATTACH_QUEUE) | |
969 | ret = macvtap_enable_queue(vlan->dev, file, q); | |
970 | else if (flags & IFF_DETACH_QUEUE) | |
971 | ret = macvtap_disable_queue(q); | |
f57855a5 JW |
972 | else |
973 | ret = -EINVAL; | |
815f236d JW |
974 | |
975 | macvtap_put_vlan(vlan); | |
976 | return ret; | |
977 | } | |
978 | ||
20d29d7a AB |
979 | /* |
980 | * provide compatibility with generic tun/tap interface | |
981 | */ | |
982 | static long macvtap_ioctl(struct file *file, unsigned int cmd, | |
983 | unsigned long arg) | |
984 | { | |
02df55d2 AB |
985 | struct macvtap_queue *q = file->private_data; |
986 | struct macvlan_dev *vlan; | |
20d29d7a AB |
987 | void __user *argp = (void __user *)arg; |
988 | struct ifreq __user *ifr = argp; | |
989 | unsigned int __user *up = argp; | |
990 | unsigned int u; | |
55afbd08 MT |
991 | int __user *sp = argp; |
992 | int s; | |
02df55d2 | 993 | int ret; |
20d29d7a AB |
994 | |
995 | switch (cmd) { | |
996 | case TUNSETIFF: | |
997 | /* ignore the name, just look at flags */ | |
998 | if (get_user(u, &ifr->ifr_flags)) | |
999 | return -EFAULT; | |
b9fb9ee0 AB |
1000 | |
1001 | ret = 0; | |
815f236d JW |
1002 | if ((u & ~(IFF_VNET_HDR | IFF_MULTI_QUEUE)) != |
1003 | (IFF_NO_PI | IFF_TAP)) | |
b9fb9ee0 AB |
1004 | ret = -EINVAL; |
1005 | else | |
1006 | q->flags = u; | |
1007 | ||
1008 | return ret; | |
20d29d7a AB |
1009 | |
1010 | case TUNGETIFF: | |
8f475a31 | 1011 | vlan = macvtap_get_vlan(q); |
02df55d2 | 1012 | if (!vlan) |
20d29d7a | 1013 | return -ENOLINK; |
20d29d7a | 1014 | |
02df55d2 | 1015 | ret = 0; |
13707f9e | 1016 | if (copy_to_user(&ifr->ifr_name, vlan->dev->name, IFNAMSIZ) || |
b9fb9ee0 | 1017 | put_user(q->flags, &ifr->ifr_flags)) |
02df55d2 | 1018 | ret = -EFAULT; |
8f475a31 | 1019 | macvtap_put_vlan(vlan); |
02df55d2 | 1020 | return ret; |
20d29d7a | 1021 | |
815f236d JW |
1022 | case TUNSETQUEUE: |
1023 | if (get_user(u, &ifr->ifr_flags)) | |
1024 | return -EFAULT; | |
1025 | return macvtap_ioctl_set_queue(file, u); | |
1026 | ||
20d29d7a | 1027 | case TUNGETFEATURES: |
df09b36f JW |
1028 | if (put_user(IFF_TAP | IFF_NO_PI | IFF_VNET_HDR | |
1029 | IFF_MULTI_QUEUE, up)) | |
20d29d7a AB |
1030 | return -EFAULT; |
1031 | return 0; | |
1032 | ||
1033 | case TUNSETSNDBUF: | |
1034 | if (get_user(u, up)) | |
1035 | return -EFAULT; | |
1036 | ||
20d29d7a | 1037 | q->sk.sk_sndbuf = u; |
20d29d7a AB |
1038 | return 0; |
1039 | ||
55afbd08 MT |
1040 | case TUNGETVNETHDRSZ: |
1041 | s = q->vnet_hdr_sz; | |
1042 | if (put_user(s, sp)) | |
1043 | return -EFAULT; | |
1044 | return 0; | |
1045 | ||
1046 | case TUNSETVNETHDRSZ: | |
1047 | if (get_user(s, sp)) | |
1048 | return -EFAULT; | |
1049 | if (s < (int)sizeof(struct virtio_net_hdr)) | |
1050 | return -EINVAL; | |
1051 | ||
1052 | q->vnet_hdr_sz = s; | |
1053 | return 0; | |
1054 | ||
20d29d7a AB |
1055 | case TUNSETOFFLOAD: |
1056 | /* let the user check for future flags */ | |
1057 | if (arg & ~(TUN_F_CSUM | TUN_F_TSO4 | TUN_F_TSO6 | | |
b9fb9ee0 | 1058 | TUN_F_TSO_ECN | TUN_F_UFO)) |
20d29d7a AB |
1059 | return -EINVAL; |
1060 | ||
b9fb9ee0 AB |
1061 | /* TODO: only accept frames with the features that |
1062 | got enabled for forwarded frames */ | |
1063 | if (!(q->flags & IFF_VNET_HDR)) | |
1064 | return -EINVAL; | |
20d29d7a AB |
1065 | return 0; |
1066 | ||
1067 | default: | |
1068 | return -EINVAL; | |
1069 | } | |
1070 | } | |
1071 | ||
1072 | #ifdef CONFIG_COMPAT | |
1073 | static long macvtap_compat_ioctl(struct file *file, unsigned int cmd, | |
1074 | unsigned long arg) | |
1075 | { | |
1076 | return macvtap_ioctl(file, cmd, (unsigned long)compat_ptr(arg)); | |
1077 | } | |
1078 | #endif | |
1079 | ||
1080 | static const struct file_operations macvtap_fops = { | |
1081 | .owner = THIS_MODULE, | |
1082 | .open = macvtap_open, | |
1083 | .release = macvtap_release, | |
1084 | .aio_read = macvtap_aio_read, | |
1085 | .aio_write = macvtap_aio_write, | |
1086 | .poll = macvtap_poll, | |
1087 | .llseek = no_llseek, | |
1088 | .unlocked_ioctl = macvtap_ioctl, | |
1089 | #ifdef CONFIG_COMPAT | |
1090 | .compat_ioctl = macvtap_compat_ioctl, | |
1091 | #endif | |
1092 | }; | |
1093 | ||
501c774c AB |
1094 | static int macvtap_sendmsg(struct kiocb *iocb, struct socket *sock, |
1095 | struct msghdr *m, size_t total_len) | |
1096 | { | |
1097 | struct macvtap_queue *q = container_of(sock, struct macvtap_queue, sock); | |
97bc3633 | 1098 | return macvtap_get_user(q, m, m->msg_iov, total_len, m->msg_iovlen, |
501c774c AB |
1099 | m->msg_flags & MSG_DONTWAIT); |
1100 | } | |
1101 | ||
1102 | static int macvtap_recvmsg(struct kiocb *iocb, struct socket *sock, | |
1103 | struct msghdr *m, size_t total_len, | |
1104 | int flags) | |
1105 | { | |
1106 | struct macvtap_queue *q = container_of(sock, struct macvtap_queue, sock); | |
1107 | int ret; | |
1108 | if (flags & ~(MSG_DONTWAIT|MSG_TRUNC)) | |
1109 | return -EINVAL; | |
1110 | ret = macvtap_do_read(q, iocb, m->msg_iov, total_len, | |
1111 | flags & MSG_DONTWAIT); | |
1112 | if (ret > total_len) { | |
1113 | m->msg_flags |= MSG_TRUNC; | |
1114 | ret = flags & MSG_TRUNC ? ret : total_len; | |
1115 | } | |
1116 | return ret; | |
1117 | } | |
1118 | ||
1119 | /* Ops structure to mimic raw sockets with tun */ | |
1120 | static const struct proto_ops macvtap_socket_ops = { | |
1121 | .sendmsg = macvtap_sendmsg, | |
1122 | .recvmsg = macvtap_recvmsg, | |
1123 | }; | |
1124 | ||
1125 | /* Get an underlying socket object from tun file. Returns error unless file is | |
1126 | * attached to a device. The returned object works like a packet socket, it | |
1127 | * can be used for sock_sendmsg/sock_recvmsg. The caller is responsible for | |
1128 | * holding a reference to the file for as long as the socket is in use. */ | |
1129 | struct socket *macvtap_get_socket(struct file *file) | |
1130 | { | |
1131 | struct macvtap_queue *q; | |
1132 | if (file->f_op != &macvtap_fops) | |
1133 | return ERR_PTR(-EINVAL); | |
1134 | q = file->private_data; | |
1135 | if (!q) | |
1136 | return ERR_PTR(-EBADFD); | |
1137 | return &q->sock; | |
1138 | } | |
1139 | EXPORT_SYMBOL_GPL(macvtap_get_socket); | |
1140 | ||
9bf1907f EB |
1141 | static int macvtap_device_event(struct notifier_block *unused, |
1142 | unsigned long event, void *ptr) | |
1143 | { | |
351638e7 | 1144 | struct net_device *dev = netdev_notifier_info_to_dev(ptr); |
e09eff7f | 1145 | struct macvlan_dev *vlan; |
9bf1907f EB |
1146 | struct device *classdev; |
1147 | dev_t devt; | |
e09eff7f | 1148 | int err; |
9bf1907f EB |
1149 | |
1150 | if (dev->rtnl_link_ops != &macvtap_link_ops) | |
1151 | return NOTIFY_DONE; | |
1152 | ||
e09eff7f | 1153 | vlan = netdev_priv(dev); |
9bf1907f EB |
1154 | |
1155 | switch (event) { | |
1156 | case NETDEV_REGISTER: | |
1157 | /* Create the device node here after the network device has | |
1158 | * been registered but before register_netdevice has | |
1159 | * finished running. | |
1160 | */ | |
e09eff7f EB |
1161 | err = macvtap_get_minor(vlan); |
1162 | if (err) | |
1163 | return notifier_from_errno(err); | |
1164 | ||
1165 | devt = MKDEV(MAJOR(macvtap_major), vlan->minor); | |
9bf1907f EB |
1166 | classdev = device_create(macvtap_class, &dev->dev, devt, |
1167 | dev, "tap%d", dev->ifindex); | |
e09eff7f EB |
1168 | if (IS_ERR(classdev)) { |
1169 | macvtap_free_minor(vlan); | |
9bf1907f | 1170 | return notifier_from_errno(PTR_ERR(classdev)); |
e09eff7f | 1171 | } |
9bf1907f EB |
1172 | break; |
1173 | case NETDEV_UNREGISTER: | |
e09eff7f | 1174 | devt = MKDEV(MAJOR(macvtap_major), vlan->minor); |
9bf1907f | 1175 | device_destroy(macvtap_class, devt); |
e09eff7f | 1176 | macvtap_free_minor(vlan); |
9bf1907f EB |
1177 | break; |
1178 | } | |
1179 | ||
1180 | return NOTIFY_DONE; | |
1181 | } | |
1182 | ||
1183 | static struct notifier_block macvtap_notifier_block __read_mostly = { | |
1184 | .notifier_call = macvtap_device_event, | |
1185 | }; | |
1186 | ||
20d29d7a AB |
1187 | static int macvtap_init(void) |
1188 | { | |
1189 | int err; | |
1190 | ||
1191 | err = alloc_chrdev_region(&macvtap_major, 0, | |
1192 | MACVTAP_NUM_DEVS, "macvtap"); | |
1193 | if (err) | |
1194 | goto out1; | |
1195 | ||
1196 | cdev_init(&macvtap_cdev, &macvtap_fops); | |
1197 | err = cdev_add(&macvtap_cdev, macvtap_major, MACVTAP_NUM_DEVS); | |
1198 | if (err) | |
1199 | goto out2; | |
1200 | ||
1201 | macvtap_class = class_create(THIS_MODULE, "macvtap"); | |
1202 | if (IS_ERR(macvtap_class)) { | |
1203 | err = PTR_ERR(macvtap_class); | |
1204 | goto out3; | |
1205 | } | |
1206 | ||
9bf1907f | 1207 | err = register_netdevice_notifier(&macvtap_notifier_block); |
20d29d7a AB |
1208 | if (err) |
1209 | goto out4; | |
1210 | ||
9bf1907f EB |
1211 | err = macvlan_link_register(&macvtap_link_ops); |
1212 | if (err) | |
1213 | goto out5; | |
1214 | ||
20d29d7a AB |
1215 | return 0; |
1216 | ||
9bf1907f EB |
1217 | out5: |
1218 | unregister_netdevice_notifier(&macvtap_notifier_block); | |
20d29d7a AB |
1219 | out4: |
1220 | class_unregister(macvtap_class); | |
1221 | out3: | |
1222 | cdev_del(&macvtap_cdev); | |
1223 | out2: | |
1224 | unregister_chrdev_region(macvtap_major, MACVTAP_NUM_DEVS); | |
1225 | out1: | |
1226 | return err; | |
1227 | } | |
1228 | module_init(macvtap_init); | |
1229 | ||
1230 | static void macvtap_exit(void) | |
1231 | { | |
1232 | rtnl_link_unregister(&macvtap_link_ops); | |
9bf1907f | 1233 | unregister_netdevice_notifier(&macvtap_notifier_block); |
20d29d7a AB |
1234 | class_unregister(macvtap_class); |
1235 | cdev_del(&macvtap_cdev); | |
1236 | unregister_chrdev_region(macvtap_major, MACVTAP_NUM_DEVS); | |
1237 | } | |
1238 | module_exit(macvtap_exit); | |
1239 | ||
1240 | MODULE_ALIAS_RTNL_LINK("macvtap"); | |
1241 | MODULE_AUTHOR("Arnd Bergmann <arnd@arndb.de>"); | |
1242 | MODULE_LICENSE("GPL"); |