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; | |
43815482 | 40 | struct socket_wq wq; |
20d29d7a AB |
41 | struct macvlan_dev *vlan; |
42 | struct file *file; | |
b9fb9ee0 | 43 | unsigned int flags; |
20d29d7a AB |
44 | }; |
45 | ||
46 | static struct proto macvtap_proto = { | |
47 | .name = "macvtap", | |
48 | .owner = THIS_MODULE, | |
49 | .obj_size = sizeof (struct macvtap_queue), | |
50 | }; | |
51 | ||
52 | /* | |
53 | * Minor number matches netdev->ifindex, so need a potentially | |
54 | * large value. This also makes it possible to split the | |
55 | * tap functionality out again in the future by offering it | |
56 | * from other drivers besides macvtap. As long as every device | |
57 | * only has one tap, the interface numbers assure that the | |
58 | * device nodes are unique. | |
59 | */ | |
60 | static unsigned int macvtap_major; | |
61 | #define MACVTAP_NUM_DEVS 65536 | |
62 | static struct class *macvtap_class; | |
63 | static struct cdev macvtap_cdev; | |
64 | ||
501c774c AB |
65 | static const struct proto_ops macvtap_socket_ops; |
66 | ||
20d29d7a AB |
67 | /* |
68 | * RCU usage: | |
02df55d2 AB |
69 | * The macvtap_queue and the macvlan_dev are loosely coupled, the |
70 | * pointers from one to the other can only be read while rcu_read_lock | |
71 | * or macvtap_lock is held. | |
20d29d7a | 72 | * |
02df55d2 AB |
73 | * Both the file and the macvlan_dev hold a reference on the macvtap_queue |
74 | * through sock_hold(&q->sk). When the macvlan_dev goes away first, | |
75 | * q->vlan becomes inaccessible. When the files gets closed, | |
76 | * macvtap_get_queue() fails. | |
20d29d7a | 77 | * |
02df55d2 AB |
78 | * There may still be references to the struct sock inside of the |
79 | * queue from outbound SKBs, but these never reference back to the | |
80 | * file or the dev. The data structure is freed through __sk_free | |
81 | * when both our references and any pending SKBs are gone. | |
20d29d7a AB |
82 | */ |
83 | static DEFINE_SPINLOCK(macvtap_lock); | |
84 | ||
85 | /* | |
86 | * Choose the next free queue, for now there is only one | |
87 | */ | |
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); | |
95 | if (rcu_dereference(vlan->tap)) | |
96 | goto out; | |
97 | ||
98 | err = 0; | |
02df55d2 | 99 | rcu_assign_pointer(q->vlan, vlan); |
20d29d7a | 100 | rcu_assign_pointer(vlan->tap, q); |
02df55d2 | 101 | sock_hold(&q->sk); |
20d29d7a AB |
102 | |
103 | q->file = file; | |
02df55d2 | 104 | file->private_data = q; |
20d29d7a AB |
105 | |
106 | out: | |
107 | spin_unlock(&macvtap_lock); | |
108 | return err; | |
109 | } | |
110 | ||
111 | /* | |
02df55d2 AB |
112 | * The file owning the queue got closed, give up both |
113 | * the reference that the files holds as well as the | |
114 | * one from the macvlan_dev if that still exists. | |
20d29d7a AB |
115 | * |
116 | * Using the spinlock makes sure that we don't get | |
117 | * to the queue again after destroying it. | |
20d29d7a | 118 | */ |
02df55d2 | 119 | static void macvtap_put_queue(struct macvtap_queue *q) |
20d29d7a | 120 | { |
02df55d2 | 121 | struct macvlan_dev *vlan; |
20d29d7a AB |
122 | |
123 | spin_lock(&macvtap_lock); | |
02df55d2 AB |
124 | vlan = rcu_dereference(q->vlan); |
125 | if (vlan) { | |
126 | rcu_assign_pointer(vlan->tap, NULL); | |
127 | rcu_assign_pointer(q->vlan, NULL); | |
128 | sock_put(&q->sk); | |
20d29d7a AB |
129 | } |
130 | ||
20d29d7a AB |
131 | spin_unlock(&macvtap_lock); |
132 | ||
133 | synchronize_rcu(); | |
134 | sock_put(&q->sk); | |
135 | } | |
136 | ||
137 | /* | |
138 | * Since we only support one queue, just dereference the pointer. | |
139 | */ | |
140 | static struct macvtap_queue *macvtap_get_queue(struct net_device *dev, | |
141 | struct sk_buff *skb) | |
142 | { | |
143 | struct macvlan_dev *vlan = netdev_priv(dev); | |
144 | ||
145 | return rcu_dereference(vlan->tap); | |
146 | } | |
147 | ||
02df55d2 AB |
148 | /* |
149 | * The net_device is going away, give up the reference | |
150 | * that it holds on the queue (all the queues one day) | |
151 | * and safely set the pointer from the queues to NULL. | |
152 | */ | |
20d29d7a AB |
153 | static void macvtap_del_queues(struct net_device *dev) |
154 | { | |
155 | struct macvlan_dev *vlan = netdev_priv(dev); | |
564517e8 | 156 | struct macvtap_queue *q; |
02df55d2 AB |
157 | |
158 | spin_lock(&macvtap_lock); | |
159 | q = rcu_dereference(vlan->tap); | |
160 | if (!q) { | |
161 | spin_unlock(&macvtap_lock); | |
162 | return; | |
564517e8 | 163 | } |
20d29d7a | 164 | |
02df55d2 AB |
165 | rcu_assign_pointer(vlan->tap, NULL); |
166 | rcu_assign_pointer(q->vlan, NULL); | |
167 | spin_unlock(&macvtap_lock); | |
168 | ||
169 | synchronize_rcu(); | |
564517e8 | 170 | sock_put(&q->sk); |
20d29d7a AB |
171 | } |
172 | ||
173 | /* | |
174 | * Forward happens for data that gets sent from one macvlan | |
175 | * endpoint to another one in bridge mode. We just take | |
176 | * the skb and put it into the receive queue. | |
177 | */ | |
178 | static int macvtap_forward(struct net_device *dev, struct sk_buff *skb) | |
179 | { | |
180 | struct macvtap_queue *q = macvtap_get_queue(dev, skb); | |
181 | if (!q) | |
182 | return -ENOLINK; | |
183 | ||
184 | skb_queue_tail(&q->sk.sk_receive_queue, skb); | |
4a4771a5 | 185 | wake_up_interruptible_poll(sk_sleep(&q->sk), POLLIN | POLLRDNORM | POLLRDBAND); |
20d29d7a AB |
186 | return 0; |
187 | } | |
188 | ||
189 | /* | |
190 | * Receive is for data from the external interface (lowerdev), | |
191 | * in case of macvtap, we can treat that the same way as | |
192 | * forward, which macvlan cannot. | |
193 | */ | |
194 | static int macvtap_receive(struct sk_buff *skb) | |
195 | { | |
196 | skb_push(skb, ETH_HLEN); | |
197 | return macvtap_forward(skb->dev, skb); | |
198 | } | |
199 | ||
200 | static int macvtap_newlink(struct net *src_net, | |
201 | struct net_device *dev, | |
202 | struct nlattr *tb[], | |
203 | struct nlattr *data[]) | |
204 | { | |
205 | struct device *classdev; | |
206 | dev_t devt; | |
207 | int err; | |
208 | ||
209 | err = macvlan_common_newlink(src_net, dev, tb, data, | |
210 | macvtap_receive, macvtap_forward); | |
211 | if (err) | |
212 | goto out; | |
213 | ||
214 | devt = MKDEV(MAJOR(macvtap_major), dev->ifindex); | |
215 | ||
216 | classdev = device_create(macvtap_class, &dev->dev, devt, | |
217 | dev, "tap%d", dev->ifindex); | |
218 | if (IS_ERR(classdev)) { | |
219 | err = PTR_ERR(classdev); | |
220 | macvtap_del_queues(dev); | |
221 | } | |
222 | ||
223 | out: | |
224 | return err; | |
225 | } | |
226 | ||
227 | static void macvtap_dellink(struct net_device *dev, | |
228 | struct list_head *head) | |
229 | { | |
230 | device_destroy(macvtap_class, | |
231 | MKDEV(MAJOR(macvtap_major), dev->ifindex)); | |
232 | ||
233 | macvtap_del_queues(dev); | |
234 | macvlan_dellink(dev, head); | |
235 | } | |
236 | ||
237 | static struct rtnl_link_ops macvtap_link_ops __read_mostly = { | |
238 | .kind = "macvtap", | |
239 | .newlink = macvtap_newlink, | |
240 | .dellink = macvtap_dellink, | |
241 | }; | |
242 | ||
243 | ||
244 | static void macvtap_sock_write_space(struct sock *sk) | |
245 | { | |
43815482 ED |
246 | wait_queue_head_t *wqueue; |
247 | ||
20d29d7a AB |
248 | if (!sock_writeable(sk) || |
249 | !test_and_clear_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags)) | |
250 | return; | |
251 | ||
43815482 ED |
252 | wqueue = sk_sleep(sk); |
253 | if (wqueue && waitqueue_active(wqueue)) | |
254 | wake_up_interruptible_poll(wqueue, POLLOUT | POLLWRNORM | POLLWRBAND); | |
20d29d7a AB |
255 | } |
256 | ||
257 | static int macvtap_open(struct inode *inode, struct file *file) | |
258 | { | |
259 | struct net *net = current->nsproxy->net_ns; | |
260 | struct net_device *dev = dev_get_by_index(net, iminor(inode)); | |
261 | struct macvtap_queue *q; | |
262 | int err; | |
263 | ||
264 | err = -ENODEV; | |
265 | if (!dev) | |
266 | goto out; | |
267 | ||
268 | /* check if this is a macvtap device */ | |
269 | err = -EINVAL; | |
270 | if (dev->rtnl_link_ops != &macvtap_link_ops) | |
271 | goto out; | |
272 | ||
273 | err = -ENOMEM; | |
274 | q = (struct macvtap_queue *)sk_alloc(net, AF_UNSPEC, GFP_KERNEL, | |
275 | &macvtap_proto); | |
276 | if (!q) | |
277 | goto out; | |
278 | ||
43815482 ED |
279 | q->sock.wq = &q->wq; |
280 | init_waitqueue_head(&q->wq.wait); | |
20d29d7a AB |
281 | q->sock.type = SOCK_RAW; |
282 | q->sock.state = SS_CONNECTED; | |
501c774c AB |
283 | q->sock.file = file; |
284 | q->sock.ops = &macvtap_socket_ops; | |
20d29d7a | 285 | sock_init_data(&q->sock, &q->sk); |
20d29d7a | 286 | q->sk.sk_write_space = macvtap_sock_write_space; |
b9fb9ee0 | 287 | q->flags = IFF_VNET_HDR | IFF_NO_PI | IFF_TAP; |
20d29d7a AB |
288 | |
289 | err = macvtap_set_queue(dev, file, q); | |
290 | if (err) | |
291 | sock_put(&q->sk); | |
292 | ||
293 | out: | |
294 | if (dev) | |
295 | dev_put(dev); | |
296 | ||
297 | return err; | |
298 | } | |
299 | ||
300 | static int macvtap_release(struct inode *inode, struct file *file) | |
301 | { | |
02df55d2 AB |
302 | struct macvtap_queue *q = file->private_data; |
303 | macvtap_put_queue(q); | |
20d29d7a AB |
304 | return 0; |
305 | } | |
306 | ||
307 | static unsigned int macvtap_poll(struct file *file, poll_table * wait) | |
308 | { | |
02df55d2 | 309 | struct macvtap_queue *q = file->private_data; |
20d29d7a AB |
310 | unsigned int mask = POLLERR; |
311 | ||
312 | if (!q) | |
313 | goto out; | |
314 | ||
315 | mask = 0; | |
43815482 | 316 | poll_wait(file, &q->wq.wait, wait); |
20d29d7a AB |
317 | |
318 | if (!skb_queue_empty(&q->sk.sk_receive_queue)) | |
319 | mask |= POLLIN | POLLRDNORM; | |
320 | ||
321 | if (sock_writeable(&q->sk) || | |
322 | (!test_and_set_bit(SOCK_ASYNC_NOSPACE, &q->sock.flags) && | |
323 | sock_writeable(&q->sk))) | |
324 | mask |= POLLOUT | POLLWRNORM; | |
325 | ||
326 | out: | |
20d29d7a AB |
327 | return mask; |
328 | } | |
329 | ||
b9fb9ee0 AB |
330 | static inline struct sk_buff *macvtap_alloc_skb(struct sock *sk, size_t prepad, |
331 | size_t len, size_t linear, | |
332 | int noblock, int *err) | |
333 | { | |
334 | struct sk_buff *skb; | |
335 | ||
336 | /* Under a page? Don't bother with paged skb. */ | |
337 | if (prepad + len < PAGE_SIZE || !linear) | |
338 | linear = len; | |
339 | ||
340 | skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock, | |
341 | err); | |
342 | if (!skb) | |
343 | return NULL; | |
344 | ||
345 | skb_reserve(skb, prepad); | |
346 | skb_put(skb, linear); | |
347 | skb->data_len = len - linear; | |
348 | skb->len += len - linear; | |
349 | ||
350 | return skb; | |
351 | } | |
352 | ||
353 | /* | |
354 | * macvtap_skb_from_vnet_hdr and macvtap_skb_to_vnet_hdr should | |
355 | * be shared with the tun/tap driver. | |
356 | */ | |
357 | static int macvtap_skb_from_vnet_hdr(struct sk_buff *skb, | |
358 | struct virtio_net_hdr *vnet_hdr) | |
359 | { | |
360 | unsigned short gso_type = 0; | |
361 | if (vnet_hdr->gso_type != VIRTIO_NET_HDR_GSO_NONE) { | |
362 | switch (vnet_hdr->gso_type & ~VIRTIO_NET_HDR_GSO_ECN) { | |
363 | case VIRTIO_NET_HDR_GSO_TCPV4: | |
364 | gso_type = SKB_GSO_TCPV4; | |
365 | break; | |
366 | case VIRTIO_NET_HDR_GSO_TCPV6: | |
367 | gso_type = SKB_GSO_TCPV6; | |
368 | break; | |
369 | case VIRTIO_NET_HDR_GSO_UDP: | |
370 | gso_type = SKB_GSO_UDP; | |
371 | break; | |
372 | default: | |
373 | return -EINVAL; | |
374 | } | |
375 | ||
376 | if (vnet_hdr->gso_type & VIRTIO_NET_HDR_GSO_ECN) | |
377 | gso_type |= SKB_GSO_TCP_ECN; | |
378 | ||
379 | if (vnet_hdr->gso_size == 0) | |
380 | return -EINVAL; | |
381 | } | |
382 | ||
383 | if (vnet_hdr->flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) { | |
384 | if (!skb_partial_csum_set(skb, vnet_hdr->csum_start, | |
385 | vnet_hdr->csum_offset)) | |
386 | return -EINVAL; | |
387 | } | |
388 | ||
389 | if (vnet_hdr->gso_type != VIRTIO_NET_HDR_GSO_NONE) { | |
390 | skb_shinfo(skb)->gso_size = vnet_hdr->gso_size; | |
391 | skb_shinfo(skb)->gso_type = gso_type; | |
392 | ||
393 | /* Header must be checked, and gso_segs computed. */ | |
394 | skb_shinfo(skb)->gso_type |= SKB_GSO_DODGY; | |
395 | skb_shinfo(skb)->gso_segs = 0; | |
396 | } | |
397 | return 0; | |
398 | } | |
399 | ||
400 | static int macvtap_skb_to_vnet_hdr(const struct sk_buff *skb, | |
401 | struct virtio_net_hdr *vnet_hdr) | |
402 | { | |
403 | memset(vnet_hdr, 0, sizeof(*vnet_hdr)); | |
404 | ||
405 | if (skb_is_gso(skb)) { | |
406 | struct skb_shared_info *sinfo = skb_shinfo(skb); | |
407 | ||
408 | /* This is a hint as to how much should be linear. */ | |
409 | vnet_hdr->hdr_len = skb_headlen(skb); | |
410 | vnet_hdr->gso_size = sinfo->gso_size; | |
411 | if (sinfo->gso_type & SKB_GSO_TCPV4) | |
412 | vnet_hdr->gso_type = VIRTIO_NET_HDR_GSO_TCPV4; | |
413 | else if (sinfo->gso_type & SKB_GSO_TCPV6) | |
414 | vnet_hdr->gso_type = VIRTIO_NET_HDR_GSO_TCPV6; | |
415 | else if (sinfo->gso_type & SKB_GSO_UDP) | |
416 | vnet_hdr->gso_type = VIRTIO_NET_HDR_GSO_UDP; | |
417 | else | |
418 | BUG(); | |
419 | if (sinfo->gso_type & SKB_GSO_TCP_ECN) | |
420 | vnet_hdr->gso_type |= VIRTIO_NET_HDR_GSO_ECN; | |
421 | } else | |
422 | vnet_hdr->gso_type = VIRTIO_NET_HDR_GSO_NONE; | |
423 | ||
424 | if (skb->ip_summed == CHECKSUM_PARTIAL) { | |
425 | vnet_hdr->flags = VIRTIO_NET_HDR_F_NEEDS_CSUM; | |
426 | vnet_hdr->csum_start = skb->csum_start - | |
427 | skb_headroom(skb); | |
428 | vnet_hdr->csum_offset = skb->csum_offset; | |
429 | } /* else everything is zero */ | |
430 | ||
431 | return 0; | |
432 | } | |
433 | ||
434 | ||
20d29d7a AB |
435 | /* Get packet from user space buffer */ |
436 | static ssize_t macvtap_get_user(struct macvtap_queue *q, | |
437 | const struct iovec *iv, size_t count, | |
438 | int noblock) | |
439 | { | |
440 | struct sk_buff *skb; | |
02df55d2 | 441 | struct macvlan_dev *vlan; |
20d29d7a AB |
442 | size_t len = count; |
443 | int err; | |
b9fb9ee0 AB |
444 | struct virtio_net_hdr vnet_hdr = { 0 }; |
445 | int vnet_hdr_len = 0; | |
446 | ||
447 | if (q->flags & IFF_VNET_HDR) { | |
448 | vnet_hdr_len = sizeof(vnet_hdr); | |
449 | ||
450 | err = -EINVAL; | |
451 | if ((len -= vnet_hdr_len) < 0) | |
452 | goto err; | |
453 | ||
454 | err = memcpy_fromiovecend((void *)&vnet_hdr, iv, 0, | |
455 | vnet_hdr_len); | |
456 | if (err < 0) | |
457 | goto err; | |
458 | if ((vnet_hdr.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) && | |
459 | vnet_hdr.csum_start + vnet_hdr.csum_offset + 2 > | |
460 | vnet_hdr.hdr_len) | |
461 | vnet_hdr.hdr_len = vnet_hdr.csum_start + | |
462 | vnet_hdr.csum_offset + 2; | |
463 | err = -EINVAL; | |
464 | if (vnet_hdr.hdr_len > len) | |
465 | goto err; | |
466 | } | |
20d29d7a | 467 | |
b9fb9ee0 | 468 | err = -EINVAL; |
20d29d7a | 469 | if (unlikely(len < ETH_HLEN)) |
b9fb9ee0 | 470 | goto err; |
20d29d7a | 471 | |
b9fb9ee0 AB |
472 | skb = macvtap_alloc_skb(&q->sk, NET_IP_ALIGN, len, vnet_hdr.hdr_len, |
473 | noblock, &err); | |
02df55d2 AB |
474 | if (!skb) |
475 | goto err; | |
20d29d7a | 476 | |
b9fb9ee0 | 477 | err = skb_copy_datagram_from_iovec(skb, 0, iv, vnet_hdr_len, len); |
02df55d2 | 478 | if (err) |
b9fb9ee0 | 479 | goto err_kfree; |
20d29d7a AB |
480 | |
481 | skb_set_network_header(skb, ETH_HLEN); | |
b9fb9ee0 AB |
482 | skb_reset_mac_header(skb); |
483 | skb->protocol = eth_hdr(skb)->h_proto; | |
484 | ||
485 | if (vnet_hdr_len) { | |
486 | err = macvtap_skb_from_vnet_hdr(skb, &vnet_hdr); | |
487 | if (err) | |
488 | goto err_kfree; | |
489 | } | |
490 | ||
02df55d2 AB |
491 | rcu_read_lock_bh(); |
492 | vlan = rcu_dereference(q->vlan); | |
493 | if (vlan) | |
494 | macvlan_start_xmit(skb, vlan->dev); | |
495 | else | |
496 | kfree_skb(skb); | |
497 | rcu_read_unlock_bh(); | |
20d29d7a AB |
498 | |
499 | return count; | |
02df55d2 | 500 | |
b9fb9ee0 AB |
501 | err_kfree: |
502 | kfree_skb(skb); | |
503 | ||
02df55d2 AB |
504 | err: |
505 | rcu_read_lock_bh(); | |
506 | vlan = rcu_dereference(q->vlan); | |
507 | if (vlan) | |
b9fb9ee0 | 508 | netdev_get_tx_queue(vlan->dev, 0)->tx_dropped++; |
02df55d2 AB |
509 | rcu_read_unlock_bh(); |
510 | ||
02df55d2 | 511 | return err; |
20d29d7a AB |
512 | } |
513 | ||
514 | static ssize_t macvtap_aio_write(struct kiocb *iocb, const struct iovec *iv, | |
515 | unsigned long count, loff_t pos) | |
516 | { | |
517 | struct file *file = iocb->ki_filp; | |
518 | ssize_t result = -ENOLINK; | |
02df55d2 | 519 | struct macvtap_queue *q = file->private_data; |
20d29d7a AB |
520 | |
521 | result = macvtap_get_user(q, iv, iov_length(iv, count), | |
522 | file->f_flags & O_NONBLOCK); | |
20d29d7a AB |
523 | return result; |
524 | } | |
525 | ||
526 | /* Put packet to the user space buffer */ | |
527 | static ssize_t macvtap_put_user(struct macvtap_queue *q, | |
528 | const struct sk_buff *skb, | |
529 | const struct iovec *iv, int len) | |
530 | { | |
02df55d2 | 531 | struct macvlan_dev *vlan; |
20d29d7a | 532 | int ret; |
b9fb9ee0 AB |
533 | int vnet_hdr_len = 0; |
534 | ||
535 | if (q->flags & IFF_VNET_HDR) { | |
536 | struct virtio_net_hdr vnet_hdr; | |
537 | vnet_hdr_len = sizeof (vnet_hdr); | |
538 | if ((len -= vnet_hdr_len) < 0) | |
539 | return -EINVAL; | |
540 | ||
541 | ret = macvtap_skb_to_vnet_hdr(skb, &vnet_hdr); | |
542 | if (ret) | |
543 | return ret; | |
544 | ||
545 | if (memcpy_toiovecend(iv, (void *)&vnet_hdr, 0, vnet_hdr_len)) | |
546 | return -EFAULT; | |
547 | } | |
20d29d7a AB |
548 | |
549 | len = min_t(int, skb->len, len); | |
550 | ||
b9fb9ee0 | 551 | ret = skb_copy_datagram_const_iovec(skb, 0, iv, vnet_hdr_len, len); |
20d29d7a | 552 | |
02df55d2 AB |
553 | rcu_read_lock_bh(); |
554 | vlan = rcu_dereference(q->vlan); | |
501c774c AB |
555 | if (vlan) |
556 | macvlan_count_rx(vlan, len, ret == 0, 0); | |
02df55d2 | 557 | rcu_read_unlock_bh(); |
20d29d7a | 558 | |
b9fb9ee0 | 559 | return ret ? ret : (len + vnet_hdr_len); |
20d29d7a AB |
560 | } |
561 | ||
501c774c AB |
562 | static ssize_t macvtap_do_read(struct macvtap_queue *q, struct kiocb *iocb, |
563 | const struct iovec *iv, unsigned long len, | |
564 | int noblock) | |
20d29d7a | 565 | { |
20d29d7a AB |
566 | DECLARE_WAITQUEUE(wait, current); |
567 | struct sk_buff *skb; | |
501c774c | 568 | ssize_t ret = 0; |
20d29d7a | 569 | |
4a4771a5 | 570 | add_wait_queue(sk_sleep(&q->sk), &wait); |
20d29d7a AB |
571 | while (len) { |
572 | current->state = TASK_INTERRUPTIBLE; | |
573 | ||
574 | /* Read frames from the queue */ | |
575 | skb = skb_dequeue(&q->sk.sk_receive_queue); | |
576 | if (!skb) { | |
501c774c | 577 | if (noblock) { |
20d29d7a AB |
578 | ret = -EAGAIN; |
579 | break; | |
580 | } | |
581 | if (signal_pending(current)) { | |
582 | ret = -ERESTARTSYS; | |
583 | break; | |
584 | } | |
585 | /* Nothing to read, let's sleep */ | |
586 | schedule(); | |
587 | continue; | |
588 | } | |
589 | ret = macvtap_put_user(q, skb, iv, len); | |
590 | kfree_skb(skb); | |
591 | break; | |
592 | } | |
593 | ||
594 | current->state = TASK_RUNNING; | |
4a4771a5 | 595 | remove_wait_queue(sk_sleep(&q->sk), &wait); |
501c774c AB |
596 | return ret; |
597 | } | |
598 | ||
599 | static ssize_t macvtap_aio_read(struct kiocb *iocb, const struct iovec *iv, | |
600 | unsigned long count, loff_t pos) | |
601 | { | |
602 | struct file *file = iocb->ki_filp; | |
603 | struct macvtap_queue *q = file->private_data; | |
604 | ssize_t len, ret = 0; | |
20d29d7a | 605 | |
501c774c AB |
606 | len = iov_length(iv, count); |
607 | if (len < 0) { | |
608 | ret = -EINVAL; | |
609 | goto out; | |
610 | } | |
611 | ||
612 | ret = macvtap_do_read(q, iocb, iv, len, file->f_flags & O_NONBLOCK); | |
613 | ret = min_t(ssize_t, ret, len); /* XXX copied from tun.c. Why? */ | |
20d29d7a | 614 | out: |
20d29d7a AB |
615 | return ret; |
616 | } | |
617 | ||
618 | /* | |
619 | * provide compatibility with generic tun/tap interface | |
620 | */ | |
621 | static long macvtap_ioctl(struct file *file, unsigned int cmd, | |
622 | unsigned long arg) | |
623 | { | |
02df55d2 AB |
624 | struct macvtap_queue *q = file->private_data; |
625 | struct macvlan_dev *vlan; | |
20d29d7a AB |
626 | void __user *argp = (void __user *)arg; |
627 | struct ifreq __user *ifr = argp; | |
628 | unsigned int __user *up = argp; | |
629 | unsigned int u; | |
02df55d2 | 630 | int ret; |
20d29d7a AB |
631 | |
632 | switch (cmd) { | |
633 | case TUNSETIFF: | |
634 | /* ignore the name, just look at flags */ | |
635 | if (get_user(u, &ifr->ifr_flags)) | |
636 | return -EFAULT; | |
b9fb9ee0 AB |
637 | |
638 | ret = 0; | |
639 | if ((u & ~IFF_VNET_HDR) != (IFF_NO_PI | IFF_TAP)) | |
640 | ret = -EINVAL; | |
641 | else | |
642 | q->flags = u; | |
643 | ||
644 | return ret; | |
20d29d7a AB |
645 | |
646 | case TUNGETIFF: | |
02df55d2 AB |
647 | rcu_read_lock_bh(); |
648 | vlan = rcu_dereference(q->vlan); | |
649 | if (vlan) | |
650 | dev_hold(vlan->dev); | |
651 | rcu_read_unlock_bh(); | |
652 | ||
653 | if (!vlan) | |
20d29d7a | 654 | return -ENOLINK; |
20d29d7a | 655 | |
02df55d2 | 656 | ret = 0; |
20d29d7a | 657 | if (copy_to_user(&ifr->ifr_name, q->vlan->dev->name, IFNAMSIZ) || |
b9fb9ee0 | 658 | put_user(q->flags, &ifr->ifr_flags)) |
02df55d2 AB |
659 | ret = -EFAULT; |
660 | dev_put(vlan->dev); | |
661 | return ret; | |
20d29d7a AB |
662 | |
663 | case TUNGETFEATURES: | |
b9fb9ee0 | 664 | if (put_user(IFF_TAP | IFF_NO_PI | IFF_VNET_HDR, up)) |
20d29d7a AB |
665 | return -EFAULT; |
666 | return 0; | |
667 | ||
668 | case TUNSETSNDBUF: | |
669 | if (get_user(u, up)) | |
670 | return -EFAULT; | |
671 | ||
20d29d7a | 672 | q->sk.sk_sndbuf = u; |
20d29d7a AB |
673 | return 0; |
674 | ||
675 | case TUNSETOFFLOAD: | |
676 | /* let the user check for future flags */ | |
677 | if (arg & ~(TUN_F_CSUM | TUN_F_TSO4 | TUN_F_TSO6 | | |
b9fb9ee0 | 678 | TUN_F_TSO_ECN | TUN_F_UFO)) |
20d29d7a AB |
679 | return -EINVAL; |
680 | ||
b9fb9ee0 AB |
681 | /* TODO: only accept frames with the features that |
682 | got enabled for forwarded frames */ | |
683 | if (!(q->flags & IFF_VNET_HDR)) | |
684 | return -EINVAL; | |
20d29d7a AB |
685 | return 0; |
686 | ||
687 | default: | |
688 | return -EINVAL; | |
689 | } | |
690 | } | |
691 | ||
692 | #ifdef CONFIG_COMPAT | |
693 | static long macvtap_compat_ioctl(struct file *file, unsigned int cmd, | |
694 | unsigned long arg) | |
695 | { | |
696 | return macvtap_ioctl(file, cmd, (unsigned long)compat_ptr(arg)); | |
697 | } | |
698 | #endif | |
699 | ||
700 | static const struct file_operations macvtap_fops = { | |
701 | .owner = THIS_MODULE, | |
702 | .open = macvtap_open, | |
703 | .release = macvtap_release, | |
704 | .aio_read = macvtap_aio_read, | |
705 | .aio_write = macvtap_aio_write, | |
706 | .poll = macvtap_poll, | |
707 | .llseek = no_llseek, | |
708 | .unlocked_ioctl = macvtap_ioctl, | |
709 | #ifdef CONFIG_COMPAT | |
710 | .compat_ioctl = macvtap_compat_ioctl, | |
711 | #endif | |
712 | }; | |
713 | ||
501c774c AB |
714 | static int macvtap_sendmsg(struct kiocb *iocb, struct socket *sock, |
715 | struct msghdr *m, size_t total_len) | |
716 | { | |
717 | struct macvtap_queue *q = container_of(sock, struct macvtap_queue, sock); | |
718 | return macvtap_get_user(q, m->msg_iov, total_len, | |
719 | m->msg_flags & MSG_DONTWAIT); | |
720 | } | |
721 | ||
722 | static int macvtap_recvmsg(struct kiocb *iocb, struct socket *sock, | |
723 | struct msghdr *m, size_t total_len, | |
724 | int flags) | |
725 | { | |
726 | struct macvtap_queue *q = container_of(sock, struct macvtap_queue, sock); | |
727 | int ret; | |
728 | if (flags & ~(MSG_DONTWAIT|MSG_TRUNC)) | |
729 | return -EINVAL; | |
730 | ret = macvtap_do_read(q, iocb, m->msg_iov, total_len, | |
731 | flags & MSG_DONTWAIT); | |
732 | if (ret > total_len) { | |
733 | m->msg_flags |= MSG_TRUNC; | |
734 | ret = flags & MSG_TRUNC ? ret : total_len; | |
735 | } | |
736 | return ret; | |
737 | } | |
738 | ||
739 | /* Ops structure to mimic raw sockets with tun */ | |
740 | static const struct proto_ops macvtap_socket_ops = { | |
741 | .sendmsg = macvtap_sendmsg, | |
742 | .recvmsg = macvtap_recvmsg, | |
743 | }; | |
744 | ||
745 | /* Get an underlying socket object from tun file. Returns error unless file is | |
746 | * attached to a device. The returned object works like a packet socket, it | |
747 | * can be used for sock_sendmsg/sock_recvmsg. The caller is responsible for | |
748 | * holding a reference to the file for as long as the socket is in use. */ | |
749 | struct socket *macvtap_get_socket(struct file *file) | |
750 | { | |
751 | struct macvtap_queue *q; | |
752 | if (file->f_op != &macvtap_fops) | |
753 | return ERR_PTR(-EINVAL); | |
754 | q = file->private_data; | |
755 | if (!q) | |
756 | return ERR_PTR(-EBADFD); | |
757 | return &q->sock; | |
758 | } | |
759 | EXPORT_SYMBOL_GPL(macvtap_get_socket); | |
760 | ||
20d29d7a AB |
761 | static int macvtap_init(void) |
762 | { | |
763 | int err; | |
764 | ||
765 | err = alloc_chrdev_region(&macvtap_major, 0, | |
766 | MACVTAP_NUM_DEVS, "macvtap"); | |
767 | if (err) | |
768 | goto out1; | |
769 | ||
770 | cdev_init(&macvtap_cdev, &macvtap_fops); | |
771 | err = cdev_add(&macvtap_cdev, macvtap_major, MACVTAP_NUM_DEVS); | |
772 | if (err) | |
773 | goto out2; | |
774 | ||
775 | macvtap_class = class_create(THIS_MODULE, "macvtap"); | |
776 | if (IS_ERR(macvtap_class)) { | |
777 | err = PTR_ERR(macvtap_class); | |
778 | goto out3; | |
779 | } | |
780 | ||
781 | err = macvlan_link_register(&macvtap_link_ops); | |
782 | if (err) | |
783 | goto out4; | |
784 | ||
785 | return 0; | |
786 | ||
787 | out4: | |
788 | class_unregister(macvtap_class); | |
789 | out3: | |
790 | cdev_del(&macvtap_cdev); | |
791 | out2: | |
792 | unregister_chrdev_region(macvtap_major, MACVTAP_NUM_DEVS); | |
793 | out1: | |
794 | return err; | |
795 | } | |
796 | module_init(macvtap_init); | |
797 | ||
798 | static void macvtap_exit(void) | |
799 | { | |
800 | rtnl_link_unregister(&macvtap_link_ops); | |
801 | class_unregister(macvtap_class); | |
802 | cdev_del(&macvtap_cdev); | |
803 | unregister_chrdev_region(macvtap_major, MACVTAP_NUM_DEVS); | |
804 | } | |
805 | module_exit(macvtap_exit); | |
806 | ||
807 | MODULE_ALIAS_RTNL_LINK("macvtap"); | |
808 | MODULE_AUTHOR("Arnd Bergmann <arnd@arndb.de>"); | |
809 | MODULE_LICENSE("GPL"); |