]>
Commit | Line | Data |
---|---|---|
20d29d7a AB |
1 | #include <linux/etherdevice.h> |
2 | #include <linux/if_macvlan.h> | |
3 | #include <linux/interrupt.h> | |
4 | #include <linux/nsproxy.h> | |
5 | #include <linux/compat.h> | |
6 | #include <linux/if_tun.h> | |
7 | #include <linux/module.h> | |
8 | #include <linux/skbuff.h> | |
9 | #include <linux/cache.h> | |
10 | #include <linux/sched.h> | |
11 | #include <linux/types.h> | |
5a0e3ad6 | 12 | #include <linux/slab.h> |
20d29d7a AB |
13 | #include <linux/init.h> |
14 | #include <linux/wait.h> | |
15 | #include <linux/cdev.h> | |
16 | #include <linux/fs.h> | |
17 | ||
18 | #include <net/net_namespace.h> | |
19 | #include <net/rtnetlink.h> | |
20 | #include <net/sock.h> | |
b9fb9ee0 | 21 | #include <linux/virtio_net.h> |
20d29d7a AB |
22 | |
23 | /* | |
24 | * A macvtap queue is the central object of this driver, it connects | |
25 | * an open character device to a macvlan interface. There can be | |
26 | * multiple queues on one interface, which map back to queues | |
27 | * implemented in hardware on the underlying device. | |
28 | * | |
29 | * macvtap_proto is used to allocate queues through the sock allocation | |
30 | * mechanism. | |
31 | * | |
32 | * TODO: multiqueue support is currently not implemented, even though | |
33 | * macvtap is basically prepared for that. We will need to add this | |
34 | * here as well as in virtio-net and qemu to get line rate on 10gbit | |
35 | * adapters from a guest. | |
36 | */ | |
37 | struct macvtap_queue { | |
38 | struct sock sk; | |
39 | struct socket sock; | |
40 | struct macvlan_dev *vlan; | |
41 | struct file *file; | |
b9fb9ee0 | 42 | unsigned int flags; |
20d29d7a AB |
43 | }; |
44 | ||
45 | static struct proto macvtap_proto = { | |
46 | .name = "macvtap", | |
47 | .owner = THIS_MODULE, | |
48 | .obj_size = sizeof (struct macvtap_queue), | |
49 | }; | |
50 | ||
51 | /* | |
52 | * Minor number matches netdev->ifindex, so need a potentially | |
53 | * large value. This also makes it possible to split the | |
54 | * tap functionality out again in the future by offering it | |
55 | * from other drivers besides macvtap. As long as every device | |
56 | * only has one tap, the interface numbers assure that the | |
57 | * device nodes are unique. | |
58 | */ | |
59 | static unsigned int macvtap_major; | |
60 | #define MACVTAP_NUM_DEVS 65536 | |
61 | static struct class *macvtap_class; | |
62 | static struct cdev macvtap_cdev; | |
63 | ||
501c774c AB |
64 | static const struct proto_ops macvtap_socket_ops; |
65 | ||
20d29d7a AB |
66 | /* |
67 | * RCU usage: | |
02df55d2 AB |
68 | * The macvtap_queue and the macvlan_dev are loosely coupled, the |
69 | * pointers from one to the other can only be read while rcu_read_lock | |
70 | * or macvtap_lock is held. | |
20d29d7a | 71 | * |
02df55d2 AB |
72 | * Both the file and the macvlan_dev hold a reference on the macvtap_queue |
73 | * through sock_hold(&q->sk). When the macvlan_dev goes away first, | |
74 | * q->vlan becomes inaccessible. When the files gets closed, | |
75 | * macvtap_get_queue() fails. | |
20d29d7a | 76 | * |
02df55d2 AB |
77 | * There may still be references to the struct sock inside of the |
78 | * queue from outbound SKBs, but these never reference back to the | |
79 | * file or the dev. The data structure is freed through __sk_free | |
80 | * when both our references and any pending SKBs are gone. | |
20d29d7a AB |
81 | */ |
82 | static DEFINE_SPINLOCK(macvtap_lock); | |
83 | ||
84 | /* | |
85 | * Choose the next free queue, for now there is only one | |
86 | */ | |
87 | static int macvtap_set_queue(struct net_device *dev, struct file *file, | |
88 | struct macvtap_queue *q) | |
89 | { | |
90 | struct macvlan_dev *vlan = netdev_priv(dev); | |
91 | int err = -EBUSY; | |
92 | ||
93 | spin_lock(&macvtap_lock); | |
94 | if (rcu_dereference(vlan->tap)) | |
95 | goto out; | |
96 | ||
97 | err = 0; | |
02df55d2 | 98 | rcu_assign_pointer(q->vlan, vlan); |
20d29d7a | 99 | rcu_assign_pointer(vlan->tap, q); |
02df55d2 | 100 | sock_hold(&q->sk); |
20d29d7a AB |
101 | |
102 | q->file = file; | |
02df55d2 | 103 | file->private_data = q; |
20d29d7a AB |
104 | |
105 | out: | |
106 | spin_unlock(&macvtap_lock); | |
107 | return err; | |
108 | } | |
109 | ||
110 | /* | |
02df55d2 AB |
111 | * The file owning the queue got closed, give up both |
112 | * the reference that the files holds as well as the | |
113 | * one from the macvlan_dev if that still exists. | |
20d29d7a AB |
114 | * |
115 | * Using the spinlock makes sure that we don't get | |
116 | * to the queue again after destroying it. | |
20d29d7a | 117 | */ |
02df55d2 | 118 | static void macvtap_put_queue(struct macvtap_queue *q) |
20d29d7a | 119 | { |
02df55d2 | 120 | struct macvlan_dev *vlan; |
20d29d7a AB |
121 | |
122 | spin_lock(&macvtap_lock); | |
02df55d2 AB |
123 | vlan = rcu_dereference(q->vlan); |
124 | if (vlan) { | |
125 | rcu_assign_pointer(vlan->tap, NULL); | |
126 | rcu_assign_pointer(q->vlan, NULL); | |
127 | sock_put(&q->sk); | |
20d29d7a AB |
128 | } |
129 | ||
20d29d7a AB |
130 | spin_unlock(&macvtap_lock); |
131 | ||
132 | synchronize_rcu(); | |
133 | sock_put(&q->sk); | |
134 | } | |
135 | ||
136 | /* | |
137 | * Since we only support one queue, just dereference the pointer. | |
138 | */ | |
139 | static struct macvtap_queue *macvtap_get_queue(struct net_device *dev, | |
140 | struct sk_buff *skb) | |
141 | { | |
142 | struct macvlan_dev *vlan = netdev_priv(dev); | |
143 | ||
144 | return rcu_dereference(vlan->tap); | |
145 | } | |
146 | ||
02df55d2 AB |
147 | /* |
148 | * The net_device is going away, give up the reference | |
149 | * that it holds on the queue (all the queues one day) | |
150 | * and safely set the pointer from the queues to NULL. | |
151 | */ | |
20d29d7a AB |
152 | static void macvtap_del_queues(struct net_device *dev) |
153 | { | |
154 | struct macvlan_dev *vlan = netdev_priv(dev); | |
564517e8 | 155 | struct macvtap_queue *q; |
02df55d2 AB |
156 | |
157 | spin_lock(&macvtap_lock); | |
158 | q = rcu_dereference(vlan->tap); | |
159 | if (!q) { | |
160 | spin_unlock(&macvtap_lock); | |
161 | return; | |
564517e8 | 162 | } |
20d29d7a | 163 | |
02df55d2 AB |
164 | rcu_assign_pointer(vlan->tap, NULL); |
165 | rcu_assign_pointer(q->vlan, NULL); | |
166 | spin_unlock(&macvtap_lock); | |
167 | ||
168 | synchronize_rcu(); | |
564517e8 | 169 | sock_put(&q->sk); |
20d29d7a AB |
170 | } |
171 | ||
172 | /* | |
173 | * Forward happens for data that gets sent from one macvlan | |
174 | * endpoint to another one in bridge mode. We just take | |
175 | * the skb and put it into the receive queue. | |
176 | */ | |
177 | static int macvtap_forward(struct net_device *dev, struct sk_buff *skb) | |
178 | { | |
179 | struct macvtap_queue *q = macvtap_get_queue(dev, skb); | |
180 | if (!q) | |
181 | return -ENOLINK; | |
182 | ||
183 | skb_queue_tail(&q->sk.sk_receive_queue, skb); | |
501c774c | 184 | wake_up_interruptible_poll(q->sk.sk_sleep, POLLIN | POLLRDNORM | POLLRDBAND); |
20d29d7a AB |
185 | return 0; |
186 | } | |
187 | ||
188 | /* | |
189 | * Receive is for data from the external interface (lowerdev), | |
190 | * in case of macvtap, we can treat that the same way as | |
191 | * forward, which macvlan cannot. | |
192 | */ | |
193 | static int macvtap_receive(struct sk_buff *skb) | |
194 | { | |
195 | skb_push(skb, ETH_HLEN); | |
196 | return macvtap_forward(skb->dev, skb); | |
197 | } | |
198 | ||
199 | static int macvtap_newlink(struct net *src_net, | |
200 | struct net_device *dev, | |
201 | struct nlattr *tb[], | |
202 | struct nlattr *data[]) | |
203 | { | |
204 | struct device *classdev; | |
205 | dev_t devt; | |
206 | int err; | |
207 | ||
208 | err = macvlan_common_newlink(src_net, dev, tb, data, | |
209 | macvtap_receive, macvtap_forward); | |
210 | if (err) | |
211 | goto out; | |
212 | ||
213 | devt = MKDEV(MAJOR(macvtap_major), dev->ifindex); | |
214 | ||
215 | classdev = device_create(macvtap_class, &dev->dev, devt, | |
216 | dev, "tap%d", dev->ifindex); | |
217 | if (IS_ERR(classdev)) { | |
218 | err = PTR_ERR(classdev); | |
219 | macvtap_del_queues(dev); | |
220 | } | |
221 | ||
222 | out: | |
223 | return err; | |
224 | } | |
225 | ||
226 | static void macvtap_dellink(struct net_device *dev, | |
227 | struct list_head *head) | |
228 | { | |
229 | device_destroy(macvtap_class, | |
230 | MKDEV(MAJOR(macvtap_major), dev->ifindex)); | |
231 | ||
232 | macvtap_del_queues(dev); | |
233 | macvlan_dellink(dev, head); | |
234 | } | |
235 | ||
236 | static struct rtnl_link_ops macvtap_link_ops __read_mostly = { | |
237 | .kind = "macvtap", | |
238 | .newlink = macvtap_newlink, | |
239 | .dellink = macvtap_dellink, | |
240 | }; | |
241 | ||
242 | ||
243 | static void macvtap_sock_write_space(struct sock *sk) | |
244 | { | |
245 | if (!sock_writeable(sk) || | |
246 | !test_and_clear_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags)) | |
247 | return; | |
248 | ||
249 | if (sk->sk_sleep && waitqueue_active(sk->sk_sleep)) | |
501c774c | 250 | wake_up_interruptible_poll(sk->sk_sleep, POLLOUT | POLLWRNORM | POLLWRBAND); |
20d29d7a AB |
251 | } |
252 | ||
253 | static int macvtap_open(struct inode *inode, struct file *file) | |
254 | { | |
255 | struct net *net = current->nsproxy->net_ns; | |
256 | struct net_device *dev = dev_get_by_index(net, iminor(inode)); | |
257 | struct macvtap_queue *q; | |
258 | int err; | |
259 | ||
260 | err = -ENODEV; | |
261 | if (!dev) | |
262 | goto out; | |
263 | ||
264 | /* check if this is a macvtap device */ | |
265 | err = -EINVAL; | |
266 | if (dev->rtnl_link_ops != &macvtap_link_ops) | |
267 | goto out; | |
268 | ||
269 | err = -ENOMEM; | |
270 | q = (struct macvtap_queue *)sk_alloc(net, AF_UNSPEC, GFP_KERNEL, | |
271 | &macvtap_proto); | |
272 | if (!q) | |
273 | goto out; | |
274 | ||
275 | init_waitqueue_head(&q->sock.wait); | |
276 | q->sock.type = SOCK_RAW; | |
277 | q->sock.state = SS_CONNECTED; | |
501c774c AB |
278 | q->sock.file = file; |
279 | q->sock.ops = &macvtap_socket_ops; | |
20d29d7a | 280 | sock_init_data(&q->sock, &q->sk); |
20d29d7a | 281 | q->sk.sk_write_space = macvtap_sock_write_space; |
b9fb9ee0 | 282 | q->flags = IFF_VNET_HDR | IFF_NO_PI | IFF_TAP; |
20d29d7a AB |
283 | |
284 | err = macvtap_set_queue(dev, file, q); | |
285 | if (err) | |
286 | sock_put(&q->sk); | |
287 | ||
288 | out: | |
289 | if (dev) | |
290 | dev_put(dev); | |
291 | ||
292 | return err; | |
293 | } | |
294 | ||
295 | static int macvtap_release(struct inode *inode, struct file *file) | |
296 | { | |
02df55d2 AB |
297 | struct macvtap_queue *q = file->private_data; |
298 | macvtap_put_queue(q); | |
20d29d7a AB |
299 | return 0; |
300 | } | |
301 | ||
302 | static unsigned int macvtap_poll(struct file *file, poll_table * wait) | |
303 | { | |
02df55d2 | 304 | struct macvtap_queue *q = file->private_data; |
20d29d7a AB |
305 | unsigned int mask = POLLERR; |
306 | ||
307 | if (!q) | |
308 | goto out; | |
309 | ||
310 | mask = 0; | |
311 | poll_wait(file, &q->sock.wait, wait); | |
312 | ||
313 | if (!skb_queue_empty(&q->sk.sk_receive_queue)) | |
314 | mask |= POLLIN | POLLRDNORM; | |
315 | ||
316 | if (sock_writeable(&q->sk) || | |
317 | (!test_and_set_bit(SOCK_ASYNC_NOSPACE, &q->sock.flags) && | |
318 | sock_writeable(&q->sk))) | |
319 | mask |= POLLOUT | POLLWRNORM; | |
320 | ||
321 | out: | |
20d29d7a AB |
322 | return mask; |
323 | } | |
324 | ||
b9fb9ee0 AB |
325 | static inline struct sk_buff *macvtap_alloc_skb(struct sock *sk, size_t prepad, |
326 | size_t len, size_t linear, | |
327 | int noblock, int *err) | |
328 | { | |
329 | struct sk_buff *skb; | |
330 | ||
331 | /* Under a page? Don't bother with paged skb. */ | |
332 | if (prepad + len < PAGE_SIZE || !linear) | |
333 | linear = len; | |
334 | ||
335 | skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock, | |
336 | err); | |
337 | if (!skb) | |
338 | return NULL; | |
339 | ||
340 | skb_reserve(skb, prepad); | |
341 | skb_put(skb, linear); | |
342 | skb->data_len = len - linear; | |
343 | skb->len += len - linear; | |
344 | ||
345 | return skb; | |
346 | } | |
347 | ||
348 | /* | |
349 | * macvtap_skb_from_vnet_hdr and macvtap_skb_to_vnet_hdr should | |
350 | * be shared with the tun/tap driver. | |
351 | */ | |
352 | static int macvtap_skb_from_vnet_hdr(struct sk_buff *skb, | |
353 | struct virtio_net_hdr *vnet_hdr) | |
354 | { | |
355 | unsigned short gso_type = 0; | |
356 | if (vnet_hdr->gso_type != VIRTIO_NET_HDR_GSO_NONE) { | |
357 | switch (vnet_hdr->gso_type & ~VIRTIO_NET_HDR_GSO_ECN) { | |
358 | case VIRTIO_NET_HDR_GSO_TCPV4: | |
359 | gso_type = SKB_GSO_TCPV4; | |
360 | break; | |
361 | case VIRTIO_NET_HDR_GSO_TCPV6: | |
362 | gso_type = SKB_GSO_TCPV6; | |
363 | break; | |
364 | case VIRTIO_NET_HDR_GSO_UDP: | |
365 | gso_type = SKB_GSO_UDP; | |
366 | break; | |
367 | default: | |
368 | return -EINVAL; | |
369 | } | |
370 | ||
371 | if (vnet_hdr->gso_type & VIRTIO_NET_HDR_GSO_ECN) | |
372 | gso_type |= SKB_GSO_TCP_ECN; | |
373 | ||
374 | if (vnet_hdr->gso_size == 0) | |
375 | return -EINVAL; | |
376 | } | |
377 | ||
378 | if (vnet_hdr->flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) { | |
379 | if (!skb_partial_csum_set(skb, vnet_hdr->csum_start, | |
380 | vnet_hdr->csum_offset)) | |
381 | return -EINVAL; | |
382 | } | |
383 | ||
384 | if (vnet_hdr->gso_type != VIRTIO_NET_HDR_GSO_NONE) { | |
385 | skb_shinfo(skb)->gso_size = vnet_hdr->gso_size; | |
386 | skb_shinfo(skb)->gso_type = gso_type; | |
387 | ||
388 | /* Header must be checked, and gso_segs computed. */ | |
389 | skb_shinfo(skb)->gso_type |= SKB_GSO_DODGY; | |
390 | skb_shinfo(skb)->gso_segs = 0; | |
391 | } | |
392 | return 0; | |
393 | } | |
394 | ||
395 | static int macvtap_skb_to_vnet_hdr(const struct sk_buff *skb, | |
396 | struct virtio_net_hdr *vnet_hdr) | |
397 | { | |
398 | memset(vnet_hdr, 0, sizeof(*vnet_hdr)); | |
399 | ||
400 | if (skb_is_gso(skb)) { | |
401 | struct skb_shared_info *sinfo = skb_shinfo(skb); | |
402 | ||
403 | /* This is a hint as to how much should be linear. */ | |
404 | vnet_hdr->hdr_len = skb_headlen(skb); | |
405 | vnet_hdr->gso_size = sinfo->gso_size; | |
406 | if (sinfo->gso_type & SKB_GSO_TCPV4) | |
407 | vnet_hdr->gso_type = VIRTIO_NET_HDR_GSO_TCPV4; | |
408 | else if (sinfo->gso_type & SKB_GSO_TCPV6) | |
409 | vnet_hdr->gso_type = VIRTIO_NET_HDR_GSO_TCPV6; | |
410 | else if (sinfo->gso_type & SKB_GSO_UDP) | |
411 | vnet_hdr->gso_type = VIRTIO_NET_HDR_GSO_UDP; | |
412 | else | |
413 | BUG(); | |
414 | if (sinfo->gso_type & SKB_GSO_TCP_ECN) | |
415 | vnet_hdr->gso_type |= VIRTIO_NET_HDR_GSO_ECN; | |
416 | } else | |
417 | vnet_hdr->gso_type = VIRTIO_NET_HDR_GSO_NONE; | |
418 | ||
419 | if (skb->ip_summed == CHECKSUM_PARTIAL) { | |
420 | vnet_hdr->flags = VIRTIO_NET_HDR_F_NEEDS_CSUM; | |
421 | vnet_hdr->csum_start = skb->csum_start - | |
422 | skb_headroom(skb); | |
423 | vnet_hdr->csum_offset = skb->csum_offset; | |
424 | } /* else everything is zero */ | |
425 | ||
426 | return 0; | |
427 | } | |
428 | ||
429 | ||
20d29d7a AB |
430 | /* Get packet from user space buffer */ |
431 | static ssize_t macvtap_get_user(struct macvtap_queue *q, | |
432 | const struct iovec *iv, size_t count, | |
433 | int noblock) | |
434 | { | |
435 | struct sk_buff *skb; | |
02df55d2 | 436 | struct macvlan_dev *vlan; |
20d29d7a AB |
437 | size_t len = count; |
438 | int err; | |
b9fb9ee0 AB |
439 | struct virtio_net_hdr vnet_hdr = { 0 }; |
440 | int vnet_hdr_len = 0; | |
441 | ||
442 | if (q->flags & IFF_VNET_HDR) { | |
443 | vnet_hdr_len = sizeof(vnet_hdr); | |
444 | ||
445 | err = -EINVAL; | |
446 | if ((len -= vnet_hdr_len) < 0) | |
447 | goto err; | |
448 | ||
449 | err = memcpy_fromiovecend((void *)&vnet_hdr, iv, 0, | |
450 | vnet_hdr_len); | |
451 | if (err < 0) | |
452 | goto err; | |
453 | if ((vnet_hdr.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) && | |
454 | vnet_hdr.csum_start + vnet_hdr.csum_offset + 2 > | |
455 | vnet_hdr.hdr_len) | |
456 | vnet_hdr.hdr_len = vnet_hdr.csum_start + | |
457 | vnet_hdr.csum_offset + 2; | |
458 | err = -EINVAL; | |
459 | if (vnet_hdr.hdr_len > len) | |
460 | goto err; | |
461 | } | |
20d29d7a | 462 | |
b9fb9ee0 | 463 | err = -EINVAL; |
20d29d7a | 464 | if (unlikely(len < ETH_HLEN)) |
b9fb9ee0 | 465 | goto err; |
20d29d7a | 466 | |
b9fb9ee0 AB |
467 | skb = macvtap_alloc_skb(&q->sk, NET_IP_ALIGN, len, vnet_hdr.hdr_len, |
468 | noblock, &err); | |
02df55d2 AB |
469 | if (!skb) |
470 | goto err; | |
20d29d7a | 471 | |
b9fb9ee0 | 472 | err = skb_copy_datagram_from_iovec(skb, 0, iv, vnet_hdr_len, len); |
02df55d2 | 473 | if (err) |
b9fb9ee0 | 474 | goto err_kfree; |
20d29d7a AB |
475 | |
476 | skb_set_network_header(skb, ETH_HLEN); | |
b9fb9ee0 AB |
477 | skb_reset_mac_header(skb); |
478 | skb->protocol = eth_hdr(skb)->h_proto; | |
479 | ||
480 | if (vnet_hdr_len) { | |
481 | err = macvtap_skb_from_vnet_hdr(skb, &vnet_hdr); | |
482 | if (err) | |
483 | goto err_kfree; | |
484 | } | |
485 | ||
02df55d2 AB |
486 | rcu_read_lock_bh(); |
487 | vlan = rcu_dereference(q->vlan); | |
488 | if (vlan) | |
489 | macvlan_start_xmit(skb, vlan->dev); | |
490 | else | |
491 | kfree_skb(skb); | |
492 | rcu_read_unlock_bh(); | |
20d29d7a AB |
493 | |
494 | return count; | |
02df55d2 | 495 | |
b9fb9ee0 AB |
496 | err_kfree: |
497 | kfree_skb(skb); | |
498 | ||
02df55d2 AB |
499 | err: |
500 | rcu_read_lock_bh(); | |
501 | vlan = rcu_dereference(q->vlan); | |
502 | if (vlan) | |
b9fb9ee0 | 503 | netdev_get_tx_queue(vlan->dev, 0)->tx_dropped++; |
02df55d2 AB |
504 | rcu_read_unlock_bh(); |
505 | ||
02df55d2 | 506 | return err; |
20d29d7a AB |
507 | } |
508 | ||
509 | static ssize_t macvtap_aio_write(struct kiocb *iocb, const struct iovec *iv, | |
510 | unsigned long count, loff_t pos) | |
511 | { | |
512 | struct file *file = iocb->ki_filp; | |
513 | ssize_t result = -ENOLINK; | |
02df55d2 | 514 | struct macvtap_queue *q = file->private_data; |
20d29d7a AB |
515 | |
516 | result = macvtap_get_user(q, iv, iov_length(iv, count), | |
517 | file->f_flags & O_NONBLOCK); | |
20d29d7a AB |
518 | return result; |
519 | } | |
520 | ||
521 | /* Put packet to the user space buffer */ | |
522 | static ssize_t macvtap_put_user(struct macvtap_queue *q, | |
523 | const struct sk_buff *skb, | |
524 | const struct iovec *iv, int len) | |
525 | { | |
02df55d2 | 526 | struct macvlan_dev *vlan; |
20d29d7a | 527 | int ret; |
b9fb9ee0 AB |
528 | int vnet_hdr_len = 0; |
529 | ||
530 | if (q->flags & IFF_VNET_HDR) { | |
531 | struct virtio_net_hdr vnet_hdr; | |
532 | vnet_hdr_len = sizeof (vnet_hdr); | |
533 | if ((len -= vnet_hdr_len) < 0) | |
534 | return -EINVAL; | |
535 | ||
536 | ret = macvtap_skb_to_vnet_hdr(skb, &vnet_hdr); | |
537 | if (ret) | |
538 | return ret; | |
539 | ||
540 | if (memcpy_toiovecend(iv, (void *)&vnet_hdr, 0, vnet_hdr_len)) | |
541 | return -EFAULT; | |
542 | } | |
20d29d7a AB |
543 | |
544 | len = min_t(int, skb->len, len); | |
545 | ||
b9fb9ee0 | 546 | ret = skb_copy_datagram_const_iovec(skb, 0, iv, vnet_hdr_len, len); |
20d29d7a | 547 | |
02df55d2 AB |
548 | rcu_read_lock_bh(); |
549 | vlan = rcu_dereference(q->vlan); | |
501c774c AB |
550 | if (vlan) |
551 | macvlan_count_rx(vlan, len, ret == 0, 0); | |
02df55d2 | 552 | rcu_read_unlock_bh(); |
20d29d7a | 553 | |
b9fb9ee0 | 554 | return ret ? ret : (len + vnet_hdr_len); |
20d29d7a AB |
555 | } |
556 | ||
501c774c AB |
557 | static ssize_t macvtap_do_read(struct macvtap_queue *q, struct kiocb *iocb, |
558 | const struct iovec *iv, unsigned long len, | |
559 | int noblock) | |
20d29d7a | 560 | { |
20d29d7a AB |
561 | DECLARE_WAITQUEUE(wait, current); |
562 | struct sk_buff *skb; | |
501c774c | 563 | ssize_t ret = 0; |
20d29d7a AB |
564 | |
565 | add_wait_queue(q->sk.sk_sleep, &wait); | |
566 | while (len) { | |
567 | current->state = TASK_INTERRUPTIBLE; | |
568 | ||
569 | /* Read frames from the queue */ | |
570 | skb = skb_dequeue(&q->sk.sk_receive_queue); | |
571 | if (!skb) { | |
501c774c | 572 | if (noblock) { |
20d29d7a AB |
573 | ret = -EAGAIN; |
574 | break; | |
575 | } | |
576 | if (signal_pending(current)) { | |
577 | ret = -ERESTARTSYS; | |
578 | break; | |
579 | } | |
580 | /* Nothing to read, let's sleep */ | |
581 | schedule(); | |
582 | continue; | |
583 | } | |
584 | ret = macvtap_put_user(q, skb, iv, len); | |
585 | kfree_skb(skb); | |
586 | break; | |
587 | } | |
588 | ||
589 | current->state = TASK_RUNNING; | |
590 | remove_wait_queue(q->sk.sk_sleep, &wait); | |
501c774c AB |
591 | return ret; |
592 | } | |
593 | ||
594 | static ssize_t macvtap_aio_read(struct kiocb *iocb, const struct iovec *iv, | |
595 | unsigned long count, loff_t pos) | |
596 | { | |
597 | struct file *file = iocb->ki_filp; | |
598 | struct macvtap_queue *q = file->private_data; | |
599 | ssize_t len, ret = 0; | |
20d29d7a | 600 | |
501c774c AB |
601 | len = iov_length(iv, count); |
602 | if (len < 0) { | |
603 | ret = -EINVAL; | |
604 | goto out; | |
605 | } | |
606 | ||
607 | ret = macvtap_do_read(q, iocb, iv, len, file->f_flags & O_NONBLOCK); | |
608 | ret = min_t(ssize_t, ret, len); /* XXX copied from tun.c. Why? */ | |
20d29d7a | 609 | out: |
20d29d7a AB |
610 | return ret; |
611 | } | |
612 | ||
613 | /* | |
614 | * provide compatibility with generic tun/tap interface | |
615 | */ | |
616 | static long macvtap_ioctl(struct file *file, unsigned int cmd, | |
617 | unsigned long arg) | |
618 | { | |
02df55d2 AB |
619 | struct macvtap_queue *q = file->private_data; |
620 | struct macvlan_dev *vlan; | |
20d29d7a AB |
621 | void __user *argp = (void __user *)arg; |
622 | struct ifreq __user *ifr = argp; | |
623 | unsigned int __user *up = argp; | |
624 | unsigned int u; | |
02df55d2 | 625 | int ret; |
20d29d7a AB |
626 | |
627 | switch (cmd) { | |
628 | case TUNSETIFF: | |
629 | /* ignore the name, just look at flags */ | |
630 | if (get_user(u, &ifr->ifr_flags)) | |
631 | return -EFAULT; | |
b9fb9ee0 AB |
632 | |
633 | ret = 0; | |
634 | if ((u & ~IFF_VNET_HDR) != (IFF_NO_PI | IFF_TAP)) | |
635 | ret = -EINVAL; | |
636 | else | |
637 | q->flags = u; | |
638 | ||
639 | return ret; | |
20d29d7a AB |
640 | |
641 | case TUNGETIFF: | |
02df55d2 AB |
642 | rcu_read_lock_bh(); |
643 | vlan = rcu_dereference(q->vlan); | |
644 | if (vlan) | |
645 | dev_hold(vlan->dev); | |
646 | rcu_read_unlock_bh(); | |
647 | ||
648 | if (!vlan) | |
20d29d7a | 649 | return -ENOLINK; |
20d29d7a | 650 | |
02df55d2 | 651 | ret = 0; |
20d29d7a | 652 | if (copy_to_user(&ifr->ifr_name, q->vlan->dev->name, IFNAMSIZ) || |
b9fb9ee0 | 653 | put_user(q->flags, &ifr->ifr_flags)) |
02df55d2 AB |
654 | ret = -EFAULT; |
655 | dev_put(vlan->dev); | |
656 | return ret; | |
20d29d7a AB |
657 | |
658 | case TUNGETFEATURES: | |
b9fb9ee0 | 659 | if (put_user(IFF_TAP | IFF_NO_PI | IFF_VNET_HDR, up)) |
20d29d7a AB |
660 | return -EFAULT; |
661 | return 0; | |
662 | ||
663 | case TUNSETSNDBUF: | |
664 | if (get_user(u, up)) | |
665 | return -EFAULT; | |
666 | ||
20d29d7a | 667 | q->sk.sk_sndbuf = u; |
20d29d7a AB |
668 | return 0; |
669 | ||
670 | case TUNSETOFFLOAD: | |
671 | /* let the user check for future flags */ | |
672 | if (arg & ~(TUN_F_CSUM | TUN_F_TSO4 | TUN_F_TSO6 | | |
b9fb9ee0 | 673 | TUN_F_TSO_ECN | TUN_F_UFO)) |
20d29d7a AB |
674 | return -EINVAL; |
675 | ||
b9fb9ee0 AB |
676 | /* TODO: only accept frames with the features that |
677 | got enabled for forwarded frames */ | |
678 | if (!(q->flags & IFF_VNET_HDR)) | |
679 | return -EINVAL; | |
20d29d7a AB |
680 | return 0; |
681 | ||
682 | default: | |
683 | return -EINVAL; | |
684 | } | |
685 | } | |
686 | ||
687 | #ifdef CONFIG_COMPAT | |
688 | static long macvtap_compat_ioctl(struct file *file, unsigned int cmd, | |
689 | unsigned long arg) | |
690 | { | |
691 | return macvtap_ioctl(file, cmd, (unsigned long)compat_ptr(arg)); | |
692 | } | |
693 | #endif | |
694 | ||
695 | static const struct file_operations macvtap_fops = { | |
696 | .owner = THIS_MODULE, | |
697 | .open = macvtap_open, | |
698 | .release = macvtap_release, | |
699 | .aio_read = macvtap_aio_read, | |
700 | .aio_write = macvtap_aio_write, | |
701 | .poll = macvtap_poll, | |
702 | .llseek = no_llseek, | |
703 | .unlocked_ioctl = macvtap_ioctl, | |
704 | #ifdef CONFIG_COMPAT | |
705 | .compat_ioctl = macvtap_compat_ioctl, | |
706 | #endif | |
707 | }; | |
708 | ||
501c774c AB |
709 | static int macvtap_sendmsg(struct kiocb *iocb, struct socket *sock, |
710 | struct msghdr *m, size_t total_len) | |
711 | { | |
712 | struct macvtap_queue *q = container_of(sock, struct macvtap_queue, sock); | |
713 | return macvtap_get_user(q, m->msg_iov, total_len, | |
714 | m->msg_flags & MSG_DONTWAIT); | |
715 | } | |
716 | ||
717 | static int macvtap_recvmsg(struct kiocb *iocb, struct socket *sock, | |
718 | struct msghdr *m, size_t total_len, | |
719 | int flags) | |
720 | { | |
721 | struct macvtap_queue *q = container_of(sock, struct macvtap_queue, sock); | |
722 | int ret; | |
723 | if (flags & ~(MSG_DONTWAIT|MSG_TRUNC)) | |
724 | return -EINVAL; | |
725 | ret = macvtap_do_read(q, iocb, m->msg_iov, total_len, | |
726 | flags & MSG_DONTWAIT); | |
727 | if (ret > total_len) { | |
728 | m->msg_flags |= MSG_TRUNC; | |
729 | ret = flags & MSG_TRUNC ? ret : total_len; | |
730 | } | |
731 | return ret; | |
732 | } | |
733 | ||
734 | /* Ops structure to mimic raw sockets with tun */ | |
735 | static const struct proto_ops macvtap_socket_ops = { | |
736 | .sendmsg = macvtap_sendmsg, | |
737 | .recvmsg = macvtap_recvmsg, | |
738 | }; | |
739 | ||
740 | /* Get an underlying socket object from tun file. Returns error unless file is | |
741 | * attached to a device. The returned object works like a packet socket, it | |
742 | * can be used for sock_sendmsg/sock_recvmsg. The caller is responsible for | |
743 | * holding a reference to the file for as long as the socket is in use. */ | |
744 | struct socket *macvtap_get_socket(struct file *file) | |
745 | { | |
746 | struct macvtap_queue *q; | |
747 | if (file->f_op != &macvtap_fops) | |
748 | return ERR_PTR(-EINVAL); | |
749 | q = file->private_data; | |
750 | if (!q) | |
751 | return ERR_PTR(-EBADFD); | |
752 | return &q->sock; | |
753 | } | |
754 | EXPORT_SYMBOL_GPL(macvtap_get_socket); | |
755 | ||
20d29d7a AB |
756 | static int macvtap_init(void) |
757 | { | |
758 | int err; | |
759 | ||
760 | err = alloc_chrdev_region(&macvtap_major, 0, | |
761 | MACVTAP_NUM_DEVS, "macvtap"); | |
762 | if (err) | |
763 | goto out1; | |
764 | ||
765 | cdev_init(&macvtap_cdev, &macvtap_fops); | |
766 | err = cdev_add(&macvtap_cdev, macvtap_major, MACVTAP_NUM_DEVS); | |
767 | if (err) | |
768 | goto out2; | |
769 | ||
770 | macvtap_class = class_create(THIS_MODULE, "macvtap"); | |
771 | if (IS_ERR(macvtap_class)) { | |
772 | err = PTR_ERR(macvtap_class); | |
773 | goto out3; | |
774 | } | |
775 | ||
776 | err = macvlan_link_register(&macvtap_link_ops); | |
777 | if (err) | |
778 | goto out4; | |
779 | ||
780 | return 0; | |
781 | ||
782 | out4: | |
783 | class_unregister(macvtap_class); | |
784 | out3: | |
785 | cdev_del(&macvtap_cdev); | |
786 | out2: | |
787 | unregister_chrdev_region(macvtap_major, MACVTAP_NUM_DEVS); | |
788 | out1: | |
789 | return err; | |
790 | } | |
791 | module_init(macvtap_init); | |
792 | ||
793 | static void macvtap_exit(void) | |
794 | { | |
795 | rtnl_link_unregister(&macvtap_link_ops); | |
796 | class_unregister(macvtap_class); | |
797 | cdev_del(&macvtap_cdev); | |
798 | unregister_chrdev_region(macvtap_major, MACVTAP_NUM_DEVS); | |
799 | } | |
800 | module_exit(macvtap_exit); | |
801 | ||
802 | MODULE_ALIAS_RTNL_LINK("macvtap"); | |
803 | MODULE_AUTHOR("Arnd Bergmann <arnd@arndb.de>"); | |
804 | MODULE_LICENSE("GPL"); |