]>
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; |
55afbd08 | 41 | int vnet_hdr_sz; |
20d29d7a AB |
42 | struct macvlan_dev *vlan; |
43 | struct file *file; | |
b9fb9ee0 | 44 | unsigned int flags; |
20d29d7a AB |
45 | }; |
46 | ||
47 | static struct proto macvtap_proto = { | |
48 | .name = "macvtap", | |
49 | .owner = THIS_MODULE, | |
50 | .obj_size = sizeof (struct macvtap_queue), | |
51 | }; | |
52 | ||
53 | /* | |
54 | * Minor number matches netdev->ifindex, so need a potentially | |
55 | * large value. This also makes it possible to split the | |
56 | * tap functionality out again in the future by offering it | |
57 | * from other drivers besides macvtap. As long as every device | |
58 | * only has one tap, the interface numbers assure that the | |
59 | * device nodes are unique. | |
60 | */ | |
1ebed71a | 61 | static dev_t macvtap_major; |
20d29d7a AB |
62 | #define MACVTAP_NUM_DEVS 65536 |
63 | static struct class *macvtap_class; | |
64 | static struct cdev macvtap_cdev; | |
65 | ||
501c774c AB |
66 | static const struct proto_ops macvtap_socket_ops; |
67 | ||
20d29d7a AB |
68 | /* |
69 | * RCU usage: | |
02df55d2 AB |
70 | * The macvtap_queue and the macvlan_dev are loosely coupled, the |
71 | * pointers from one to the other can only be read while rcu_read_lock | |
72 | * or macvtap_lock is held. | |
20d29d7a | 73 | * |
02df55d2 AB |
74 | * Both the file and the macvlan_dev hold a reference on the macvtap_queue |
75 | * through sock_hold(&q->sk). When the macvlan_dev goes away first, | |
76 | * q->vlan becomes inaccessible. When the files gets closed, | |
77 | * macvtap_get_queue() fails. | |
20d29d7a | 78 | * |
02df55d2 AB |
79 | * There may still be references to the struct sock inside of the |
80 | * queue from outbound SKBs, but these never reference back to the | |
81 | * file or the dev. The data structure is freed through __sk_free | |
82 | * when both our references and any pending SKBs are gone. | |
20d29d7a AB |
83 | */ |
84 | static DEFINE_SPINLOCK(macvtap_lock); | |
85 | ||
86 | /* | |
87 | * Choose the next free queue, for now there is only one | |
88 | */ | |
89 | static int macvtap_set_queue(struct net_device *dev, struct file *file, | |
90 | struct macvtap_queue *q) | |
91 | { | |
92 | struct macvlan_dev *vlan = netdev_priv(dev); | |
93 | int err = -EBUSY; | |
94 | ||
95 | spin_lock(&macvtap_lock); | |
96 | if (rcu_dereference(vlan->tap)) | |
97 | goto out; | |
98 | ||
99 | err = 0; | |
02df55d2 | 100 | rcu_assign_pointer(q->vlan, vlan); |
20d29d7a | 101 | rcu_assign_pointer(vlan->tap, q); |
02df55d2 | 102 | sock_hold(&q->sk); |
20d29d7a AB |
103 | |
104 | q->file = file; | |
02df55d2 | 105 | file->private_data = q; |
20d29d7a AB |
106 | |
107 | out: | |
108 | spin_unlock(&macvtap_lock); | |
109 | return err; | |
110 | } | |
111 | ||
112 | /* | |
02df55d2 AB |
113 | * The file owning the queue got closed, give up both |
114 | * the reference that the files holds as well as the | |
115 | * one from the macvlan_dev if that still exists. | |
20d29d7a AB |
116 | * |
117 | * Using the spinlock makes sure that we don't get | |
118 | * to the queue again after destroying it. | |
20d29d7a | 119 | */ |
02df55d2 | 120 | static void macvtap_put_queue(struct macvtap_queue *q) |
20d29d7a | 121 | { |
02df55d2 | 122 | struct macvlan_dev *vlan; |
20d29d7a AB |
123 | |
124 | spin_lock(&macvtap_lock); | |
02df55d2 AB |
125 | vlan = rcu_dereference(q->vlan); |
126 | if (vlan) { | |
127 | rcu_assign_pointer(vlan->tap, NULL); | |
128 | rcu_assign_pointer(q->vlan, NULL); | |
129 | sock_put(&q->sk); | |
20d29d7a AB |
130 | } |
131 | ||
20d29d7a AB |
132 | spin_unlock(&macvtap_lock); |
133 | ||
134 | synchronize_rcu(); | |
135 | sock_put(&q->sk); | |
136 | } | |
137 | ||
138 | /* | |
139 | * Since we only support one queue, just dereference the pointer. | |
140 | */ | |
141 | static struct macvtap_queue *macvtap_get_queue(struct net_device *dev, | |
142 | struct sk_buff *skb) | |
143 | { | |
144 | struct macvlan_dev *vlan = netdev_priv(dev); | |
145 | ||
146 | return rcu_dereference(vlan->tap); | |
147 | } | |
148 | ||
02df55d2 AB |
149 | /* |
150 | * The net_device is going away, give up the reference | |
151 | * that it holds on the queue (all the queues one day) | |
152 | * and safely set the pointer from the queues to NULL. | |
153 | */ | |
20d29d7a AB |
154 | static void macvtap_del_queues(struct net_device *dev) |
155 | { | |
156 | struct macvlan_dev *vlan = netdev_priv(dev); | |
564517e8 | 157 | struct macvtap_queue *q; |
02df55d2 AB |
158 | |
159 | spin_lock(&macvtap_lock); | |
160 | q = rcu_dereference(vlan->tap); | |
161 | if (!q) { | |
162 | spin_unlock(&macvtap_lock); | |
163 | return; | |
564517e8 | 164 | } |
20d29d7a | 165 | |
02df55d2 AB |
166 | rcu_assign_pointer(vlan->tap, NULL); |
167 | rcu_assign_pointer(q->vlan, NULL); | |
168 | spin_unlock(&macvtap_lock); | |
169 | ||
170 | synchronize_rcu(); | |
564517e8 | 171 | sock_put(&q->sk); |
20d29d7a AB |
172 | } |
173 | ||
174 | /* | |
175 | * Forward happens for data that gets sent from one macvlan | |
176 | * endpoint to another one in bridge mode. We just take | |
177 | * the skb and put it into the receive queue. | |
178 | */ | |
179 | static int macvtap_forward(struct net_device *dev, struct sk_buff *skb) | |
180 | { | |
181 | struct macvtap_queue *q = macvtap_get_queue(dev, skb); | |
182 | if (!q) | |
8a35747a HX |
183 | goto drop; |
184 | ||
185 | if (skb_queue_len(&q->sk.sk_receive_queue) >= dev->tx_queue_len) | |
186 | goto drop; | |
20d29d7a AB |
187 | |
188 | skb_queue_tail(&q->sk.sk_receive_queue, skb); | |
4a4771a5 | 189 | wake_up_interruptible_poll(sk_sleep(&q->sk), POLLIN | POLLRDNORM | POLLRDBAND); |
8a35747a HX |
190 | return NET_RX_SUCCESS; |
191 | ||
192 | drop: | |
193 | kfree_skb(skb); | |
194 | return NET_RX_DROP; | |
20d29d7a AB |
195 | } |
196 | ||
197 | /* | |
198 | * Receive is for data from the external interface (lowerdev), | |
199 | * in case of macvtap, we can treat that the same way as | |
200 | * forward, which macvlan cannot. | |
201 | */ | |
202 | static int macvtap_receive(struct sk_buff *skb) | |
203 | { | |
204 | skb_push(skb, ETH_HLEN); | |
205 | return macvtap_forward(skb->dev, skb); | |
206 | } | |
207 | ||
208 | static int macvtap_newlink(struct net *src_net, | |
209 | struct net_device *dev, | |
210 | struct nlattr *tb[], | |
211 | struct nlattr *data[]) | |
212 | { | |
213 | struct device *classdev; | |
214 | dev_t devt; | |
215 | int err; | |
216 | ||
217 | err = macvlan_common_newlink(src_net, dev, tb, data, | |
218 | macvtap_receive, macvtap_forward); | |
219 | if (err) | |
220 | goto out; | |
221 | ||
222 | devt = MKDEV(MAJOR(macvtap_major), dev->ifindex); | |
223 | ||
224 | classdev = device_create(macvtap_class, &dev->dev, devt, | |
225 | dev, "tap%d", dev->ifindex); | |
226 | if (IS_ERR(classdev)) { | |
227 | err = PTR_ERR(classdev); | |
228 | macvtap_del_queues(dev); | |
229 | } | |
230 | ||
231 | out: | |
232 | return err; | |
233 | } | |
234 | ||
235 | static void macvtap_dellink(struct net_device *dev, | |
236 | struct list_head *head) | |
237 | { | |
238 | device_destroy(macvtap_class, | |
239 | MKDEV(MAJOR(macvtap_major), dev->ifindex)); | |
240 | ||
241 | macvtap_del_queues(dev); | |
242 | macvlan_dellink(dev, head); | |
243 | } | |
244 | ||
8a35747a HX |
245 | static void macvtap_setup(struct net_device *dev) |
246 | { | |
247 | macvlan_common_setup(dev); | |
248 | dev->tx_queue_len = TUN_READQ_SIZE; | |
249 | } | |
250 | ||
20d29d7a AB |
251 | static struct rtnl_link_ops macvtap_link_ops __read_mostly = { |
252 | .kind = "macvtap", | |
8a35747a | 253 | .setup = macvtap_setup, |
20d29d7a AB |
254 | .newlink = macvtap_newlink, |
255 | .dellink = macvtap_dellink, | |
256 | }; | |
257 | ||
258 | ||
259 | static void macvtap_sock_write_space(struct sock *sk) | |
260 | { | |
43815482 ED |
261 | wait_queue_head_t *wqueue; |
262 | ||
20d29d7a AB |
263 | if (!sock_writeable(sk) || |
264 | !test_and_clear_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags)) | |
265 | return; | |
266 | ||
43815482 ED |
267 | wqueue = sk_sleep(sk); |
268 | if (wqueue && waitqueue_active(wqueue)) | |
269 | wake_up_interruptible_poll(wqueue, POLLOUT | POLLWRNORM | POLLWRBAND); | |
20d29d7a AB |
270 | } |
271 | ||
272 | static int macvtap_open(struct inode *inode, struct file *file) | |
273 | { | |
274 | struct net *net = current->nsproxy->net_ns; | |
275 | struct net_device *dev = dev_get_by_index(net, iminor(inode)); | |
276 | struct macvtap_queue *q; | |
277 | int err; | |
278 | ||
279 | err = -ENODEV; | |
280 | if (!dev) | |
281 | goto out; | |
282 | ||
283 | /* check if this is a macvtap device */ | |
284 | err = -EINVAL; | |
285 | if (dev->rtnl_link_ops != &macvtap_link_ops) | |
286 | goto out; | |
287 | ||
288 | err = -ENOMEM; | |
289 | q = (struct macvtap_queue *)sk_alloc(net, AF_UNSPEC, GFP_KERNEL, | |
290 | &macvtap_proto); | |
291 | if (!q) | |
292 | goto out; | |
293 | ||
43815482 ED |
294 | q->sock.wq = &q->wq; |
295 | init_waitqueue_head(&q->wq.wait); | |
20d29d7a AB |
296 | q->sock.type = SOCK_RAW; |
297 | q->sock.state = SS_CONNECTED; | |
501c774c AB |
298 | q->sock.file = file; |
299 | q->sock.ops = &macvtap_socket_ops; | |
20d29d7a | 300 | sock_init_data(&q->sock, &q->sk); |
20d29d7a | 301 | q->sk.sk_write_space = macvtap_sock_write_space; |
b9fb9ee0 | 302 | q->flags = IFF_VNET_HDR | IFF_NO_PI | IFF_TAP; |
55afbd08 | 303 | q->vnet_hdr_sz = sizeof(struct virtio_net_hdr); |
20d29d7a AB |
304 | |
305 | err = macvtap_set_queue(dev, file, q); | |
306 | if (err) | |
307 | sock_put(&q->sk); | |
308 | ||
309 | out: | |
310 | if (dev) | |
311 | dev_put(dev); | |
312 | ||
313 | return err; | |
314 | } | |
315 | ||
316 | static int macvtap_release(struct inode *inode, struct file *file) | |
317 | { | |
02df55d2 AB |
318 | struct macvtap_queue *q = file->private_data; |
319 | macvtap_put_queue(q); | |
20d29d7a AB |
320 | return 0; |
321 | } | |
322 | ||
323 | static unsigned int macvtap_poll(struct file *file, poll_table * wait) | |
324 | { | |
02df55d2 | 325 | struct macvtap_queue *q = file->private_data; |
20d29d7a AB |
326 | unsigned int mask = POLLERR; |
327 | ||
328 | if (!q) | |
329 | goto out; | |
330 | ||
331 | mask = 0; | |
43815482 | 332 | poll_wait(file, &q->wq.wait, wait); |
20d29d7a AB |
333 | |
334 | if (!skb_queue_empty(&q->sk.sk_receive_queue)) | |
335 | mask |= POLLIN | POLLRDNORM; | |
336 | ||
337 | if (sock_writeable(&q->sk) || | |
338 | (!test_and_set_bit(SOCK_ASYNC_NOSPACE, &q->sock.flags) && | |
339 | sock_writeable(&q->sk))) | |
340 | mask |= POLLOUT | POLLWRNORM; | |
341 | ||
342 | out: | |
20d29d7a AB |
343 | return mask; |
344 | } | |
345 | ||
b9fb9ee0 AB |
346 | static inline struct sk_buff *macvtap_alloc_skb(struct sock *sk, size_t prepad, |
347 | size_t len, size_t linear, | |
348 | int noblock, int *err) | |
349 | { | |
350 | struct sk_buff *skb; | |
351 | ||
352 | /* Under a page? Don't bother with paged skb. */ | |
353 | if (prepad + len < PAGE_SIZE || !linear) | |
354 | linear = len; | |
355 | ||
356 | skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock, | |
357 | err); | |
358 | if (!skb) | |
359 | return NULL; | |
360 | ||
361 | skb_reserve(skb, prepad); | |
362 | skb_put(skb, linear); | |
363 | skb->data_len = len - linear; | |
364 | skb->len += len - linear; | |
365 | ||
366 | return skb; | |
367 | } | |
368 | ||
369 | /* | |
370 | * macvtap_skb_from_vnet_hdr and macvtap_skb_to_vnet_hdr should | |
371 | * be shared with the tun/tap driver. | |
372 | */ | |
373 | static int macvtap_skb_from_vnet_hdr(struct sk_buff *skb, | |
374 | struct virtio_net_hdr *vnet_hdr) | |
375 | { | |
376 | unsigned short gso_type = 0; | |
377 | if (vnet_hdr->gso_type != VIRTIO_NET_HDR_GSO_NONE) { | |
378 | switch (vnet_hdr->gso_type & ~VIRTIO_NET_HDR_GSO_ECN) { | |
379 | case VIRTIO_NET_HDR_GSO_TCPV4: | |
380 | gso_type = SKB_GSO_TCPV4; | |
381 | break; | |
382 | case VIRTIO_NET_HDR_GSO_TCPV6: | |
383 | gso_type = SKB_GSO_TCPV6; | |
384 | break; | |
385 | case VIRTIO_NET_HDR_GSO_UDP: | |
386 | gso_type = SKB_GSO_UDP; | |
387 | break; | |
388 | default: | |
389 | return -EINVAL; | |
390 | } | |
391 | ||
392 | if (vnet_hdr->gso_type & VIRTIO_NET_HDR_GSO_ECN) | |
393 | gso_type |= SKB_GSO_TCP_ECN; | |
394 | ||
395 | if (vnet_hdr->gso_size == 0) | |
396 | return -EINVAL; | |
397 | } | |
398 | ||
399 | if (vnet_hdr->flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) { | |
400 | if (!skb_partial_csum_set(skb, vnet_hdr->csum_start, | |
401 | vnet_hdr->csum_offset)) | |
402 | return -EINVAL; | |
403 | } | |
404 | ||
405 | if (vnet_hdr->gso_type != VIRTIO_NET_HDR_GSO_NONE) { | |
406 | skb_shinfo(skb)->gso_size = vnet_hdr->gso_size; | |
407 | skb_shinfo(skb)->gso_type = gso_type; | |
408 | ||
409 | /* Header must be checked, and gso_segs computed. */ | |
410 | skb_shinfo(skb)->gso_type |= SKB_GSO_DODGY; | |
411 | skb_shinfo(skb)->gso_segs = 0; | |
412 | } | |
413 | return 0; | |
414 | } | |
415 | ||
416 | static int macvtap_skb_to_vnet_hdr(const struct sk_buff *skb, | |
417 | struct virtio_net_hdr *vnet_hdr) | |
418 | { | |
419 | memset(vnet_hdr, 0, sizeof(*vnet_hdr)); | |
420 | ||
421 | if (skb_is_gso(skb)) { | |
422 | struct skb_shared_info *sinfo = skb_shinfo(skb); | |
423 | ||
424 | /* This is a hint as to how much should be linear. */ | |
425 | vnet_hdr->hdr_len = skb_headlen(skb); | |
426 | vnet_hdr->gso_size = sinfo->gso_size; | |
427 | if (sinfo->gso_type & SKB_GSO_TCPV4) | |
428 | vnet_hdr->gso_type = VIRTIO_NET_HDR_GSO_TCPV4; | |
429 | else if (sinfo->gso_type & SKB_GSO_TCPV6) | |
430 | vnet_hdr->gso_type = VIRTIO_NET_HDR_GSO_TCPV6; | |
431 | else if (sinfo->gso_type & SKB_GSO_UDP) | |
432 | vnet_hdr->gso_type = VIRTIO_NET_HDR_GSO_UDP; | |
433 | else | |
434 | BUG(); | |
435 | if (sinfo->gso_type & SKB_GSO_TCP_ECN) | |
436 | vnet_hdr->gso_type |= VIRTIO_NET_HDR_GSO_ECN; | |
437 | } else | |
438 | vnet_hdr->gso_type = VIRTIO_NET_HDR_GSO_NONE; | |
439 | ||
440 | if (skb->ip_summed == CHECKSUM_PARTIAL) { | |
441 | vnet_hdr->flags = VIRTIO_NET_HDR_F_NEEDS_CSUM; | |
442 | vnet_hdr->csum_start = skb->csum_start - | |
443 | skb_headroom(skb); | |
444 | vnet_hdr->csum_offset = skb->csum_offset; | |
445 | } /* else everything is zero */ | |
446 | ||
447 | return 0; | |
448 | } | |
449 | ||
450 | ||
20d29d7a AB |
451 | /* Get packet from user space buffer */ |
452 | static ssize_t macvtap_get_user(struct macvtap_queue *q, | |
453 | const struct iovec *iv, size_t count, | |
454 | int noblock) | |
455 | { | |
456 | struct sk_buff *skb; | |
02df55d2 | 457 | struct macvlan_dev *vlan; |
20d29d7a AB |
458 | size_t len = count; |
459 | int err; | |
b9fb9ee0 AB |
460 | struct virtio_net_hdr vnet_hdr = { 0 }; |
461 | int vnet_hdr_len = 0; | |
462 | ||
463 | if (q->flags & IFF_VNET_HDR) { | |
55afbd08 | 464 | vnet_hdr_len = q->vnet_hdr_sz; |
b9fb9ee0 AB |
465 | |
466 | err = -EINVAL; | |
467 | if ((len -= vnet_hdr_len) < 0) | |
468 | goto err; | |
469 | ||
470 | err = memcpy_fromiovecend((void *)&vnet_hdr, iv, 0, | |
55afbd08 | 471 | sizeof(vnet_hdr)); |
b9fb9ee0 AB |
472 | if (err < 0) |
473 | goto err; | |
474 | if ((vnet_hdr.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) && | |
475 | vnet_hdr.csum_start + vnet_hdr.csum_offset + 2 > | |
476 | vnet_hdr.hdr_len) | |
477 | vnet_hdr.hdr_len = vnet_hdr.csum_start + | |
478 | vnet_hdr.csum_offset + 2; | |
479 | err = -EINVAL; | |
480 | if (vnet_hdr.hdr_len > len) | |
481 | goto err; | |
482 | } | |
20d29d7a | 483 | |
b9fb9ee0 | 484 | err = -EINVAL; |
20d29d7a | 485 | if (unlikely(len < ETH_HLEN)) |
b9fb9ee0 | 486 | goto err; |
20d29d7a | 487 | |
b9fb9ee0 AB |
488 | skb = macvtap_alloc_skb(&q->sk, NET_IP_ALIGN, len, vnet_hdr.hdr_len, |
489 | noblock, &err); | |
02df55d2 AB |
490 | if (!skb) |
491 | goto err; | |
20d29d7a | 492 | |
b9fb9ee0 | 493 | err = skb_copy_datagram_from_iovec(skb, 0, iv, vnet_hdr_len, len); |
02df55d2 | 494 | if (err) |
b9fb9ee0 | 495 | goto err_kfree; |
20d29d7a AB |
496 | |
497 | skb_set_network_header(skb, ETH_HLEN); | |
b9fb9ee0 AB |
498 | skb_reset_mac_header(skb); |
499 | skb->protocol = eth_hdr(skb)->h_proto; | |
500 | ||
501 | if (vnet_hdr_len) { | |
502 | err = macvtap_skb_from_vnet_hdr(skb, &vnet_hdr); | |
503 | if (err) | |
504 | goto err_kfree; | |
505 | } | |
506 | ||
02df55d2 AB |
507 | rcu_read_lock_bh(); |
508 | vlan = rcu_dereference(q->vlan); | |
509 | if (vlan) | |
510 | macvlan_start_xmit(skb, vlan->dev); | |
511 | else | |
512 | kfree_skb(skb); | |
513 | rcu_read_unlock_bh(); | |
20d29d7a AB |
514 | |
515 | return count; | |
02df55d2 | 516 | |
b9fb9ee0 AB |
517 | err_kfree: |
518 | kfree_skb(skb); | |
519 | ||
02df55d2 AB |
520 | err: |
521 | rcu_read_lock_bh(); | |
522 | vlan = rcu_dereference(q->vlan); | |
523 | if (vlan) | |
b9fb9ee0 | 524 | netdev_get_tx_queue(vlan->dev, 0)->tx_dropped++; |
02df55d2 AB |
525 | rcu_read_unlock_bh(); |
526 | ||
02df55d2 | 527 | return err; |
20d29d7a AB |
528 | } |
529 | ||
530 | static ssize_t macvtap_aio_write(struct kiocb *iocb, const struct iovec *iv, | |
531 | unsigned long count, loff_t pos) | |
532 | { | |
533 | struct file *file = iocb->ki_filp; | |
534 | ssize_t result = -ENOLINK; | |
02df55d2 | 535 | struct macvtap_queue *q = file->private_data; |
20d29d7a AB |
536 | |
537 | result = macvtap_get_user(q, iv, iov_length(iv, count), | |
538 | file->f_flags & O_NONBLOCK); | |
20d29d7a AB |
539 | return result; |
540 | } | |
541 | ||
542 | /* Put packet to the user space buffer */ | |
543 | static ssize_t macvtap_put_user(struct macvtap_queue *q, | |
544 | const struct sk_buff *skb, | |
545 | const struct iovec *iv, int len) | |
546 | { | |
02df55d2 | 547 | struct macvlan_dev *vlan; |
20d29d7a | 548 | int ret; |
b9fb9ee0 AB |
549 | int vnet_hdr_len = 0; |
550 | ||
551 | if (q->flags & IFF_VNET_HDR) { | |
552 | struct virtio_net_hdr vnet_hdr; | |
55afbd08 | 553 | vnet_hdr_len = q->vnet_hdr_sz; |
b9fb9ee0 AB |
554 | if ((len -= vnet_hdr_len) < 0) |
555 | return -EINVAL; | |
556 | ||
557 | ret = macvtap_skb_to_vnet_hdr(skb, &vnet_hdr); | |
558 | if (ret) | |
559 | return ret; | |
560 | ||
55afbd08 | 561 | if (memcpy_toiovecend(iv, (void *)&vnet_hdr, 0, sizeof(vnet_hdr))) |
b9fb9ee0 AB |
562 | return -EFAULT; |
563 | } | |
20d29d7a AB |
564 | |
565 | len = min_t(int, skb->len, len); | |
566 | ||
b9fb9ee0 | 567 | ret = skb_copy_datagram_const_iovec(skb, 0, iv, vnet_hdr_len, len); |
20d29d7a | 568 | |
02df55d2 AB |
569 | rcu_read_lock_bh(); |
570 | vlan = rcu_dereference(q->vlan); | |
501c774c AB |
571 | if (vlan) |
572 | macvlan_count_rx(vlan, len, ret == 0, 0); | |
02df55d2 | 573 | rcu_read_unlock_bh(); |
20d29d7a | 574 | |
b9fb9ee0 | 575 | return ret ? ret : (len + vnet_hdr_len); |
20d29d7a AB |
576 | } |
577 | ||
501c774c AB |
578 | static ssize_t macvtap_do_read(struct macvtap_queue *q, struct kiocb *iocb, |
579 | const struct iovec *iv, unsigned long len, | |
580 | int noblock) | |
20d29d7a | 581 | { |
20d29d7a AB |
582 | DECLARE_WAITQUEUE(wait, current); |
583 | struct sk_buff *skb; | |
501c774c | 584 | ssize_t ret = 0; |
20d29d7a | 585 | |
4a4771a5 | 586 | add_wait_queue(sk_sleep(&q->sk), &wait); |
20d29d7a AB |
587 | while (len) { |
588 | current->state = TASK_INTERRUPTIBLE; | |
589 | ||
590 | /* Read frames from the queue */ | |
591 | skb = skb_dequeue(&q->sk.sk_receive_queue); | |
592 | if (!skb) { | |
501c774c | 593 | if (noblock) { |
20d29d7a AB |
594 | ret = -EAGAIN; |
595 | break; | |
596 | } | |
597 | if (signal_pending(current)) { | |
598 | ret = -ERESTARTSYS; | |
599 | break; | |
600 | } | |
601 | /* Nothing to read, let's sleep */ | |
602 | schedule(); | |
603 | continue; | |
604 | } | |
605 | ret = macvtap_put_user(q, skb, iv, len); | |
606 | kfree_skb(skb); | |
607 | break; | |
608 | } | |
609 | ||
610 | current->state = TASK_RUNNING; | |
4a4771a5 | 611 | remove_wait_queue(sk_sleep(&q->sk), &wait); |
501c774c AB |
612 | return ret; |
613 | } | |
614 | ||
615 | static ssize_t macvtap_aio_read(struct kiocb *iocb, const struct iovec *iv, | |
616 | unsigned long count, loff_t pos) | |
617 | { | |
618 | struct file *file = iocb->ki_filp; | |
619 | struct macvtap_queue *q = file->private_data; | |
620 | ssize_t len, ret = 0; | |
20d29d7a | 621 | |
501c774c AB |
622 | len = iov_length(iv, count); |
623 | if (len < 0) { | |
624 | ret = -EINVAL; | |
625 | goto out; | |
626 | } | |
627 | ||
628 | ret = macvtap_do_read(q, iocb, iv, len, file->f_flags & O_NONBLOCK); | |
629 | ret = min_t(ssize_t, ret, len); /* XXX copied from tun.c. Why? */ | |
20d29d7a | 630 | out: |
20d29d7a AB |
631 | return ret; |
632 | } | |
633 | ||
634 | /* | |
635 | * provide compatibility with generic tun/tap interface | |
636 | */ | |
637 | static long macvtap_ioctl(struct file *file, unsigned int cmd, | |
638 | unsigned long arg) | |
639 | { | |
02df55d2 AB |
640 | struct macvtap_queue *q = file->private_data; |
641 | struct macvlan_dev *vlan; | |
20d29d7a AB |
642 | void __user *argp = (void __user *)arg; |
643 | struct ifreq __user *ifr = argp; | |
644 | unsigned int __user *up = argp; | |
645 | unsigned int u; | |
55afbd08 MT |
646 | int __user *sp = argp; |
647 | int s; | |
02df55d2 | 648 | int ret; |
20d29d7a AB |
649 | |
650 | switch (cmd) { | |
651 | case TUNSETIFF: | |
652 | /* ignore the name, just look at flags */ | |
653 | if (get_user(u, &ifr->ifr_flags)) | |
654 | return -EFAULT; | |
b9fb9ee0 AB |
655 | |
656 | ret = 0; | |
657 | if ((u & ~IFF_VNET_HDR) != (IFF_NO_PI | IFF_TAP)) | |
658 | ret = -EINVAL; | |
659 | else | |
660 | q->flags = u; | |
661 | ||
662 | return ret; | |
20d29d7a AB |
663 | |
664 | case TUNGETIFF: | |
02df55d2 AB |
665 | rcu_read_lock_bh(); |
666 | vlan = rcu_dereference(q->vlan); | |
667 | if (vlan) | |
668 | dev_hold(vlan->dev); | |
669 | rcu_read_unlock_bh(); | |
670 | ||
671 | if (!vlan) | |
20d29d7a | 672 | return -ENOLINK; |
20d29d7a | 673 | |
02df55d2 | 674 | ret = 0; |
20d29d7a | 675 | if (copy_to_user(&ifr->ifr_name, q->vlan->dev->name, IFNAMSIZ) || |
b9fb9ee0 | 676 | put_user(q->flags, &ifr->ifr_flags)) |
02df55d2 AB |
677 | ret = -EFAULT; |
678 | dev_put(vlan->dev); | |
679 | return ret; | |
20d29d7a AB |
680 | |
681 | case TUNGETFEATURES: | |
b9fb9ee0 | 682 | if (put_user(IFF_TAP | IFF_NO_PI | IFF_VNET_HDR, up)) |
20d29d7a AB |
683 | return -EFAULT; |
684 | return 0; | |
685 | ||
686 | case TUNSETSNDBUF: | |
687 | if (get_user(u, up)) | |
688 | return -EFAULT; | |
689 | ||
20d29d7a | 690 | q->sk.sk_sndbuf = u; |
20d29d7a AB |
691 | return 0; |
692 | ||
55afbd08 MT |
693 | case TUNGETVNETHDRSZ: |
694 | s = q->vnet_hdr_sz; | |
695 | if (put_user(s, sp)) | |
696 | return -EFAULT; | |
697 | return 0; | |
698 | ||
699 | case TUNSETVNETHDRSZ: | |
700 | if (get_user(s, sp)) | |
701 | return -EFAULT; | |
702 | if (s < (int)sizeof(struct virtio_net_hdr)) | |
703 | return -EINVAL; | |
704 | ||
705 | q->vnet_hdr_sz = s; | |
706 | return 0; | |
707 | ||
20d29d7a AB |
708 | case TUNSETOFFLOAD: |
709 | /* let the user check for future flags */ | |
710 | if (arg & ~(TUN_F_CSUM | TUN_F_TSO4 | TUN_F_TSO6 | | |
b9fb9ee0 | 711 | TUN_F_TSO_ECN | TUN_F_UFO)) |
20d29d7a AB |
712 | return -EINVAL; |
713 | ||
b9fb9ee0 AB |
714 | /* TODO: only accept frames with the features that |
715 | got enabled for forwarded frames */ | |
716 | if (!(q->flags & IFF_VNET_HDR)) | |
717 | return -EINVAL; | |
20d29d7a AB |
718 | return 0; |
719 | ||
720 | default: | |
721 | return -EINVAL; | |
722 | } | |
723 | } | |
724 | ||
725 | #ifdef CONFIG_COMPAT | |
726 | static long macvtap_compat_ioctl(struct file *file, unsigned int cmd, | |
727 | unsigned long arg) | |
728 | { | |
729 | return macvtap_ioctl(file, cmd, (unsigned long)compat_ptr(arg)); | |
730 | } | |
731 | #endif | |
732 | ||
733 | static const struct file_operations macvtap_fops = { | |
734 | .owner = THIS_MODULE, | |
735 | .open = macvtap_open, | |
736 | .release = macvtap_release, | |
737 | .aio_read = macvtap_aio_read, | |
738 | .aio_write = macvtap_aio_write, | |
739 | .poll = macvtap_poll, | |
740 | .llseek = no_llseek, | |
741 | .unlocked_ioctl = macvtap_ioctl, | |
742 | #ifdef CONFIG_COMPAT | |
743 | .compat_ioctl = macvtap_compat_ioctl, | |
744 | #endif | |
745 | }; | |
746 | ||
501c774c AB |
747 | static int macvtap_sendmsg(struct kiocb *iocb, struct socket *sock, |
748 | struct msghdr *m, size_t total_len) | |
749 | { | |
750 | struct macvtap_queue *q = container_of(sock, struct macvtap_queue, sock); | |
751 | return macvtap_get_user(q, m->msg_iov, total_len, | |
752 | m->msg_flags & MSG_DONTWAIT); | |
753 | } | |
754 | ||
755 | static int macvtap_recvmsg(struct kiocb *iocb, struct socket *sock, | |
756 | struct msghdr *m, size_t total_len, | |
757 | int flags) | |
758 | { | |
759 | struct macvtap_queue *q = container_of(sock, struct macvtap_queue, sock); | |
760 | int ret; | |
761 | if (flags & ~(MSG_DONTWAIT|MSG_TRUNC)) | |
762 | return -EINVAL; | |
763 | ret = macvtap_do_read(q, iocb, m->msg_iov, total_len, | |
764 | flags & MSG_DONTWAIT); | |
765 | if (ret > total_len) { | |
766 | m->msg_flags |= MSG_TRUNC; | |
767 | ret = flags & MSG_TRUNC ? ret : total_len; | |
768 | } | |
769 | return ret; | |
770 | } | |
771 | ||
772 | /* Ops structure to mimic raw sockets with tun */ | |
773 | static const struct proto_ops macvtap_socket_ops = { | |
774 | .sendmsg = macvtap_sendmsg, | |
775 | .recvmsg = macvtap_recvmsg, | |
776 | }; | |
777 | ||
778 | /* Get an underlying socket object from tun file. Returns error unless file is | |
779 | * attached to a device. The returned object works like a packet socket, it | |
780 | * can be used for sock_sendmsg/sock_recvmsg. The caller is responsible for | |
781 | * holding a reference to the file for as long as the socket is in use. */ | |
782 | struct socket *macvtap_get_socket(struct file *file) | |
783 | { | |
784 | struct macvtap_queue *q; | |
785 | if (file->f_op != &macvtap_fops) | |
786 | return ERR_PTR(-EINVAL); | |
787 | q = file->private_data; | |
788 | if (!q) | |
789 | return ERR_PTR(-EBADFD); | |
790 | return &q->sock; | |
791 | } | |
792 | EXPORT_SYMBOL_GPL(macvtap_get_socket); | |
793 | ||
20d29d7a AB |
794 | static int macvtap_init(void) |
795 | { | |
796 | int err; | |
797 | ||
798 | err = alloc_chrdev_region(&macvtap_major, 0, | |
799 | MACVTAP_NUM_DEVS, "macvtap"); | |
800 | if (err) | |
801 | goto out1; | |
802 | ||
803 | cdev_init(&macvtap_cdev, &macvtap_fops); | |
804 | err = cdev_add(&macvtap_cdev, macvtap_major, MACVTAP_NUM_DEVS); | |
805 | if (err) | |
806 | goto out2; | |
807 | ||
808 | macvtap_class = class_create(THIS_MODULE, "macvtap"); | |
809 | if (IS_ERR(macvtap_class)) { | |
810 | err = PTR_ERR(macvtap_class); | |
811 | goto out3; | |
812 | } | |
813 | ||
814 | err = macvlan_link_register(&macvtap_link_ops); | |
815 | if (err) | |
816 | goto out4; | |
817 | ||
818 | return 0; | |
819 | ||
820 | out4: | |
821 | class_unregister(macvtap_class); | |
822 | out3: | |
823 | cdev_del(&macvtap_cdev); | |
824 | out2: | |
825 | unregister_chrdev_region(macvtap_major, MACVTAP_NUM_DEVS); | |
826 | out1: | |
827 | return err; | |
828 | } | |
829 | module_init(macvtap_init); | |
830 | ||
831 | static void macvtap_exit(void) | |
832 | { | |
833 | rtnl_link_unregister(&macvtap_link_ops); | |
834 | class_unregister(macvtap_class); | |
835 | cdev_del(&macvtap_cdev); | |
836 | unregister_chrdev_region(macvtap_major, MACVTAP_NUM_DEVS); | |
837 | } | |
838 | module_exit(macvtap_exit); | |
839 | ||
840 | MODULE_ALIAS_RTNL_LINK("macvtap"); | |
841 | MODULE_AUTHOR("Arnd Bergmann <arnd@arndb.de>"); | |
842 | MODULE_LICENSE("GPL"); |