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