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