]>
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; |
13707f9e | 42 | struct macvlan_dev __rcu *vlan; |
20d29d7a | 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 | /* | |
1565c7c1 KK |
87 | * get_slot: return a [unused/occupied] slot in vlan->taps[]: |
88 | * - if 'q' is NULL, return the first empty slot; | |
89 | * - otherwise, return the slot this pointer occupies. | |
20d29d7a | 90 | */ |
1565c7c1 KK |
91 | static int get_slot(struct macvlan_dev *vlan, struct macvtap_queue *q) |
92 | { | |
93 | int i; | |
94 | ||
95 | for (i = 0; i < MAX_MACVTAP_QUEUES; i++) { | |
96 | if (rcu_dereference(vlan->taps[i]) == q) | |
97 | return i; | |
98 | } | |
99 | ||
100 | /* Should never happen */ | |
101 | BUG_ON(1); | |
102 | } | |
103 | ||
20d29d7a AB |
104 | static int macvtap_set_queue(struct net_device *dev, struct file *file, |
105 | struct macvtap_queue *q) | |
106 | { | |
107 | struct macvlan_dev *vlan = netdev_priv(dev); | |
1565c7c1 | 108 | int index; |
20d29d7a AB |
109 | int err = -EBUSY; |
110 | ||
111 | spin_lock(&macvtap_lock); | |
1565c7c1 | 112 | if (vlan->numvtaps == MAX_MACVTAP_QUEUES) |
20d29d7a AB |
113 | goto out; |
114 | ||
115 | err = 0; | |
1565c7c1 | 116 | index = get_slot(vlan, NULL); |
02df55d2 | 117 | rcu_assign_pointer(q->vlan, vlan); |
1565c7c1 | 118 | rcu_assign_pointer(vlan->taps[index], q); |
02df55d2 | 119 | sock_hold(&q->sk); |
20d29d7a AB |
120 | |
121 | q->file = file; | |
02df55d2 | 122 | file->private_data = q; |
20d29d7a | 123 | |
1565c7c1 KK |
124 | vlan->numvtaps++; |
125 | ||
20d29d7a AB |
126 | out: |
127 | spin_unlock(&macvtap_lock); | |
128 | return err; | |
129 | } | |
130 | ||
131 | /* | |
02df55d2 AB |
132 | * The file owning the queue got closed, give up both |
133 | * the reference that the files holds as well as the | |
134 | * one from the macvlan_dev if that still exists. | |
20d29d7a AB |
135 | * |
136 | * Using the spinlock makes sure that we don't get | |
137 | * to the queue again after destroying it. | |
20d29d7a | 138 | */ |
02df55d2 | 139 | static void macvtap_put_queue(struct macvtap_queue *q) |
20d29d7a | 140 | { |
02df55d2 | 141 | struct macvlan_dev *vlan; |
20d29d7a AB |
142 | |
143 | spin_lock(&macvtap_lock); | |
13707f9e ED |
144 | vlan = rcu_dereference_protected(q->vlan, |
145 | lockdep_is_held(&macvtap_lock)); | |
02df55d2 | 146 | if (vlan) { |
1565c7c1 KK |
147 | int index = get_slot(vlan, q); |
148 | ||
149 | rcu_assign_pointer(vlan->taps[index], NULL); | |
02df55d2 AB |
150 | rcu_assign_pointer(q->vlan, NULL); |
151 | sock_put(&q->sk); | |
1565c7c1 | 152 | --vlan->numvtaps; |
20d29d7a AB |
153 | } |
154 | ||
20d29d7a AB |
155 | spin_unlock(&macvtap_lock); |
156 | ||
157 | synchronize_rcu(); | |
158 | sock_put(&q->sk); | |
159 | } | |
160 | ||
161 | /* | |
1565c7c1 KK |
162 | * Select a queue based on the rxq of the device on which this packet |
163 | * arrived. If the incoming device is not mq, calculate a flow hash | |
164 | * to select a queue. If all fails, find the first available queue. | |
165 | * Cache vlan->numvtaps since it can become zero during the execution | |
166 | * of this function. | |
20d29d7a AB |
167 | */ |
168 | static struct macvtap_queue *macvtap_get_queue(struct net_device *dev, | |
169 | struct sk_buff *skb) | |
170 | { | |
171 | struct macvlan_dev *vlan = netdev_priv(dev); | |
1565c7c1 KK |
172 | struct macvtap_queue *tap = NULL; |
173 | int numvtaps = vlan->numvtaps; | |
174 | __u32 rxq; | |
175 | ||
176 | if (!numvtaps) | |
177 | goto out; | |
178 | ||
179 | if (likely(skb_rx_queue_recorded(skb))) { | |
180 | rxq = skb_get_rx_queue(skb); | |
20d29d7a | 181 | |
1565c7c1 KK |
182 | while (unlikely(rxq >= numvtaps)) |
183 | rxq -= numvtaps; | |
184 | ||
185 | tap = rcu_dereference(vlan->taps[rxq]); | |
186 | if (tap) | |
187 | goto out; | |
188 | } | |
189 | ||
190 | /* Check if we can use flow to select a queue */ | |
191 | rxq = skb_get_rxhash(skb); | |
192 | if (rxq) { | |
193 | tap = rcu_dereference(vlan->taps[rxq % numvtaps]); | |
194 | if (tap) | |
195 | goto out; | |
196 | } | |
197 | ||
198 | /* Everything failed - find first available queue */ | |
199 | for (rxq = 0; rxq < MAX_MACVTAP_QUEUES; rxq++) { | |
200 | tap = rcu_dereference(vlan->taps[rxq]); | |
201 | if (tap) | |
202 | break; | |
203 | } | |
204 | ||
205 | out: | |
206 | return tap; | |
20d29d7a AB |
207 | } |
208 | ||
02df55d2 AB |
209 | /* |
210 | * The net_device is going away, give up the reference | |
1565c7c1 KK |
211 | * that it holds on all queues and safely set the pointer |
212 | * from the queues to NULL. | |
02df55d2 | 213 | */ |
20d29d7a AB |
214 | static void macvtap_del_queues(struct net_device *dev) |
215 | { | |
216 | struct macvlan_dev *vlan = netdev_priv(dev); | |
1565c7c1 KK |
217 | struct macvtap_queue *q, *qlist[MAX_MACVTAP_QUEUES]; |
218 | int i, j = 0; | |
02df55d2 | 219 | |
1565c7c1 | 220 | /* macvtap_put_queue can free some slots, so go through all slots */ |
02df55d2 | 221 | spin_lock(&macvtap_lock); |
1565c7c1 | 222 | for (i = 0; i < MAX_MACVTAP_QUEUES && vlan->numvtaps; i++) { |
13707f9e ED |
223 | q = rcu_dereference_protected(vlan->taps[i], |
224 | lockdep_is_held(&macvtap_lock)); | |
1565c7c1 KK |
225 | if (q) { |
226 | qlist[j++] = q; | |
227 | rcu_assign_pointer(vlan->taps[i], NULL); | |
228 | rcu_assign_pointer(q->vlan, NULL); | |
229 | vlan->numvtaps--; | |
230 | } | |
564517e8 | 231 | } |
1565c7c1 | 232 | BUG_ON(vlan->numvtaps != 0); |
02df55d2 AB |
233 | spin_unlock(&macvtap_lock); |
234 | ||
235 | synchronize_rcu(); | |
1565c7c1 KK |
236 | |
237 | for (--j; j >= 0; j--) | |
238 | sock_put(&qlist[j]->sk); | |
20d29d7a AB |
239 | } |
240 | ||
241 | /* | |
242 | * Forward happens for data that gets sent from one macvlan | |
243 | * endpoint to another one in bridge mode. We just take | |
244 | * the skb and put it into the receive queue. | |
245 | */ | |
246 | static int macvtap_forward(struct net_device *dev, struct sk_buff *skb) | |
247 | { | |
248 | struct macvtap_queue *q = macvtap_get_queue(dev, skb); | |
249 | if (!q) | |
8a35747a HX |
250 | goto drop; |
251 | ||
252 | if (skb_queue_len(&q->sk.sk_receive_queue) >= dev->tx_queue_len) | |
253 | goto drop; | |
20d29d7a AB |
254 | |
255 | skb_queue_tail(&q->sk.sk_receive_queue, skb); | |
4a4771a5 | 256 | wake_up_interruptible_poll(sk_sleep(&q->sk), POLLIN | POLLRDNORM | POLLRDBAND); |
8a35747a HX |
257 | return NET_RX_SUCCESS; |
258 | ||
259 | drop: | |
260 | kfree_skb(skb); | |
261 | return NET_RX_DROP; | |
20d29d7a AB |
262 | } |
263 | ||
264 | /* | |
265 | * Receive is for data from the external interface (lowerdev), | |
266 | * in case of macvtap, we can treat that the same way as | |
267 | * forward, which macvlan cannot. | |
268 | */ | |
269 | static int macvtap_receive(struct sk_buff *skb) | |
270 | { | |
271 | skb_push(skb, ETH_HLEN); | |
272 | return macvtap_forward(skb->dev, skb); | |
273 | } | |
274 | ||
275 | static int macvtap_newlink(struct net *src_net, | |
276 | struct net_device *dev, | |
277 | struct nlattr *tb[], | |
278 | struct nlattr *data[]) | |
279 | { | |
280 | struct device *classdev; | |
281 | dev_t devt; | |
282 | int err; | |
283 | ||
284 | err = macvlan_common_newlink(src_net, dev, tb, data, | |
285 | macvtap_receive, macvtap_forward); | |
286 | if (err) | |
287 | goto out; | |
288 | ||
289 | devt = MKDEV(MAJOR(macvtap_major), dev->ifindex); | |
290 | ||
291 | classdev = device_create(macvtap_class, &dev->dev, devt, | |
292 | dev, "tap%d", dev->ifindex); | |
293 | if (IS_ERR(classdev)) { | |
294 | err = PTR_ERR(classdev); | |
295 | macvtap_del_queues(dev); | |
296 | } | |
297 | ||
298 | out: | |
299 | return err; | |
300 | } | |
301 | ||
302 | static void macvtap_dellink(struct net_device *dev, | |
303 | struct list_head *head) | |
304 | { | |
305 | device_destroy(macvtap_class, | |
306 | MKDEV(MAJOR(macvtap_major), dev->ifindex)); | |
307 | ||
308 | macvtap_del_queues(dev); | |
309 | macvlan_dellink(dev, head); | |
310 | } | |
311 | ||
8a35747a HX |
312 | static void macvtap_setup(struct net_device *dev) |
313 | { | |
314 | macvlan_common_setup(dev); | |
315 | dev->tx_queue_len = TUN_READQ_SIZE; | |
316 | } | |
317 | ||
20d29d7a AB |
318 | static struct rtnl_link_ops macvtap_link_ops __read_mostly = { |
319 | .kind = "macvtap", | |
8a35747a | 320 | .setup = macvtap_setup, |
20d29d7a AB |
321 | .newlink = macvtap_newlink, |
322 | .dellink = macvtap_dellink, | |
323 | }; | |
324 | ||
325 | ||
326 | static void macvtap_sock_write_space(struct sock *sk) | |
327 | { | |
43815482 ED |
328 | wait_queue_head_t *wqueue; |
329 | ||
20d29d7a AB |
330 | if (!sock_writeable(sk) || |
331 | !test_and_clear_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags)) | |
332 | return; | |
333 | ||
43815482 ED |
334 | wqueue = sk_sleep(sk); |
335 | if (wqueue && waitqueue_active(wqueue)) | |
336 | wake_up_interruptible_poll(wqueue, POLLOUT | POLLWRNORM | POLLWRBAND); | |
20d29d7a AB |
337 | } |
338 | ||
339 | static int macvtap_open(struct inode *inode, struct file *file) | |
340 | { | |
341 | struct net *net = current->nsproxy->net_ns; | |
342 | struct net_device *dev = dev_get_by_index(net, iminor(inode)); | |
343 | struct macvtap_queue *q; | |
344 | int err; | |
345 | ||
346 | err = -ENODEV; | |
347 | if (!dev) | |
348 | goto out; | |
349 | ||
350 | /* check if this is a macvtap device */ | |
351 | err = -EINVAL; | |
352 | if (dev->rtnl_link_ops != &macvtap_link_ops) | |
353 | goto out; | |
354 | ||
355 | err = -ENOMEM; | |
356 | q = (struct macvtap_queue *)sk_alloc(net, AF_UNSPEC, GFP_KERNEL, | |
357 | &macvtap_proto); | |
358 | if (!q) | |
359 | goto out; | |
360 | ||
43815482 ED |
361 | q->sock.wq = &q->wq; |
362 | init_waitqueue_head(&q->wq.wait); | |
20d29d7a AB |
363 | q->sock.type = SOCK_RAW; |
364 | q->sock.state = SS_CONNECTED; | |
501c774c AB |
365 | q->sock.file = file; |
366 | q->sock.ops = &macvtap_socket_ops; | |
20d29d7a | 367 | sock_init_data(&q->sock, &q->sk); |
20d29d7a | 368 | q->sk.sk_write_space = macvtap_sock_write_space; |
b9fb9ee0 | 369 | q->flags = IFF_VNET_HDR | IFF_NO_PI | IFF_TAP; |
55afbd08 | 370 | q->vnet_hdr_sz = sizeof(struct virtio_net_hdr); |
20d29d7a AB |
371 | |
372 | err = macvtap_set_queue(dev, file, q); | |
373 | if (err) | |
374 | sock_put(&q->sk); | |
375 | ||
376 | out: | |
377 | if (dev) | |
378 | dev_put(dev); | |
379 | ||
380 | return err; | |
381 | } | |
382 | ||
383 | static int macvtap_release(struct inode *inode, struct file *file) | |
384 | { | |
02df55d2 AB |
385 | struct macvtap_queue *q = file->private_data; |
386 | macvtap_put_queue(q); | |
20d29d7a AB |
387 | return 0; |
388 | } | |
389 | ||
390 | static unsigned int macvtap_poll(struct file *file, poll_table * wait) | |
391 | { | |
02df55d2 | 392 | struct macvtap_queue *q = file->private_data; |
20d29d7a AB |
393 | unsigned int mask = POLLERR; |
394 | ||
395 | if (!q) | |
396 | goto out; | |
397 | ||
398 | mask = 0; | |
43815482 | 399 | poll_wait(file, &q->wq.wait, wait); |
20d29d7a AB |
400 | |
401 | if (!skb_queue_empty(&q->sk.sk_receive_queue)) | |
402 | mask |= POLLIN | POLLRDNORM; | |
403 | ||
404 | if (sock_writeable(&q->sk) || | |
405 | (!test_and_set_bit(SOCK_ASYNC_NOSPACE, &q->sock.flags) && | |
406 | sock_writeable(&q->sk))) | |
407 | mask |= POLLOUT | POLLWRNORM; | |
408 | ||
409 | out: | |
20d29d7a AB |
410 | return mask; |
411 | } | |
412 | ||
b9fb9ee0 AB |
413 | static inline struct sk_buff *macvtap_alloc_skb(struct sock *sk, size_t prepad, |
414 | size_t len, size_t linear, | |
415 | int noblock, int *err) | |
416 | { | |
417 | struct sk_buff *skb; | |
418 | ||
419 | /* Under a page? Don't bother with paged skb. */ | |
420 | if (prepad + len < PAGE_SIZE || !linear) | |
421 | linear = len; | |
422 | ||
423 | skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock, | |
424 | err); | |
425 | if (!skb) | |
426 | return NULL; | |
427 | ||
428 | skb_reserve(skb, prepad); | |
429 | skb_put(skb, linear); | |
430 | skb->data_len = len - linear; | |
431 | skb->len += len - linear; | |
432 | ||
433 | return skb; | |
434 | } | |
435 | ||
436 | /* | |
437 | * macvtap_skb_from_vnet_hdr and macvtap_skb_to_vnet_hdr should | |
438 | * be shared with the tun/tap driver. | |
439 | */ | |
440 | static int macvtap_skb_from_vnet_hdr(struct sk_buff *skb, | |
441 | struct virtio_net_hdr *vnet_hdr) | |
442 | { | |
443 | unsigned short gso_type = 0; | |
444 | if (vnet_hdr->gso_type != VIRTIO_NET_HDR_GSO_NONE) { | |
445 | switch (vnet_hdr->gso_type & ~VIRTIO_NET_HDR_GSO_ECN) { | |
446 | case VIRTIO_NET_HDR_GSO_TCPV4: | |
447 | gso_type = SKB_GSO_TCPV4; | |
448 | break; | |
449 | case VIRTIO_NET_HDR_GSO_TCPV6: | |
450 | gso_type = SKB_GSO_TCPV6; | |
451 | break; | |
452 | case VIRTIO_NET_HDR_GSO_UDP: | |
453 | gso_type = SKB_GSO_UDP; | |
454 | break; | |
455 | default: | |
456 | return -EINVAL; | |
457 | } | |
458 | ||
459 | if (vnet_hdr->gso_type & VIRTIO_NET_HDR_GSO_ECN) | |
460 | gso_type |= SKB_GSO_TCP_ECN; | |
461 | ||
462 | if (vnet_hdr->gso_size == 0) | |
463 | return -EINVAL; | |
464 | } | |
465 | ||
466 | if (vnet_hdr->flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) { | |
467 | if (!skb_partial_csum_set(skb, vnet_hdr->csum_start, | |
468 | vnet_hdr->csum_offset)) | |
469 | return -EINVAL; | |
470 | } | |
471 | ||
472 | if (vnet_hdr->gso_type != VIRTIO_NET_HDR_GSO_NONE) { | |
473 | skb_shinfo(skb)->gso_size = vnet_hdr->gso_size; | |
474 | skb_shinfo(skb)->gso_type = gso_type; | |
475 | ||
476 | /* Header must be checked, and gso_segs computed. */ | |
477 | skb_shinfo(skb)->gso_type |= SKB_GSO_DODGY; | |
478 | skb_shinfo(skb)->gso_segs = 0; | |
479 | } | |
480 | return 0; | |
481 | } | |
482 | ||
483 | static int macvtap_skb_to_vnet_hdr(const struct sk_buff *skb, | |
484 | struct virtio_net_hdr *vnet_hdr) | |
485 | { | |
486 | memset(vnet_hdr, 0, sizeof(*vnet_hdr)); | |
487 | ||
488 | if (skb_is_gso(skb)) { | |
489 | struct skb_shared_info *sinfo = skb_shinfo(skb); | |
490 | ||
491 | /* This is a hint as to how much should be linear. */ | |
492 | vnet_hdr->hdr_len = skb_headlen(skb); | |
493 | vnet_hdr->gso_size = sinfo->gso_size; | |
494 | if (sinfo->gso_type & SKB_GSO_TCPV4) | |
495 | vnet_hdr->gso_type = VIRTIO_NET_HDR_GSO_TCPV4; | |
496 | else if (sinfo->gso_type & SKB_GSO_TCPV6) | |
497 | vnet_hdr->gso_type = VIRTIO_NET_HDR_GSO_TCPV6; | |
498 | else if (sinfo->gso_type & SKB_GSO_UDP) | |
499 | vnet_hdr->gso_type = VIRTIO_NET_HDR_GSO_UDP; | |
500 | else | |
501 | BUG(); | |
502 | if (sinfo->gso_type & SKB_GSO_TCP_ECN) | |
503 | vnet_hdr->gso_type |= VIRTIO_NET_HDR_GSO_ECN; | |
504 | } else | |
505 | vnet_hdr->gso_type = VIRTIO_NET_HDR_GSO_NONE; | |
506 | ||
507 | if (skb->ip_summed == CHECKSUM_PARTIAL) { | |
508 | vnet_hdr->flags = VIRTIO_NET_HDR_F_NEEDS_CSUM; | |
55508d60 | 509 | vnet_hdr->csum_start = skb_checksum_start_offset(skb); |
b9fb9ee0 | 510 | vnet_hdr->csum_offset = skb->csum_offset; |
10a8d94a JW |
511 | } else if (skb->ip_summed == CHECKSUM_UNNECESSARY) { |
512 | vnet_hdr->flags = VIRTIO_NET_HDR_F_DATA_VALID; | |
b9fb9ee0 AB |
513 | } /* else everything is zero */ |
514 | ||
515 | return 0; | |
516 | } | |
517 | ||
518 | ||
20d29d7a AB |
519 | /* Get packet from user space buffer */ |
520 | static ssize_t macvtap_get_user(struct macvtap_queue *q, | |
521 | const struct iovec *iv, size_t count, | |
522 | int noblock) | |
523 | { | |
524 | struct sk_buff *skb; | |
02df55d2 | 525 | struct macvlan_dev *vlan; |
20d29d7a AB |
526 | size_t len = count; |
527 | int err; | |
b9fb9ee0 AB |
528 | struct virtio_net_hdr vnet_hdr = { 0 }; |
529 | int vnet_hdr_len = 0; | |
530 | ||
531 | if (q->flags & IFF_VNET_HDR) { | |
55afbd08 | 532 | vnet_hdr_len = q->vnet_hdr_sz; |
b9fb9ee0 AB |
533 | |
534 | err = -EINVAL; | |
ce3c8692 | 535 | if (len < vnet_hdr_len) |
b9fb9ee0 | 536 | goto err; |
ce3c8692 | 537 | len -= vnet_hdr_len; |
b9fb9ee0 AB |
538 | |
539 | err = memcpy_fromiovecend((void *)&vnet_hdr, iv, 0, | |
55afbd08 | 540 | sizeof(vnet_hdr)); |
b9fb9ee0 AB |
541 | if (err < 0) |
542 | goto err; | |
543 | if ((vnet_hdr.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) && | |
544 | vnet_hdr.csum_start + vnet_hdr.csum_offset + 2 > | |
545 | vnet_hdr.hdr_len) | |
546 | vnet_hdr.hdr_len = vnet_hdr.csum_start + | |
547 | vnet_hdr.csum_offset + 2; | |
548 | err = -EINVAL; | |
549 | if (vnet_hdr.hdr_len > len) | |
550 | goto err; | |
551 | } | |
20d29d7a | 552 | |
b9fb9ee0 | 553 | err = -EINVAL; |
20d29d7a | 554 | if (unlikely(len < ETH_HLEN)) |
b9fb9ee0 | 555 | goto err; |
20d29d7a | 556 | |
b9fb9ee0 AB |
557 | skb = macvtap_alloc_skb(&q->sk, NET_IP_ALIGN, len, vnet_hdr.hdr_len, |
558 | noblock, &err); | |
02df55d2 AB |
559 | if (!skb) |
560 | goto err; | |
20d29d7a | 561 | |
b9fb9ee0 | 562 | err = skb_copy_datagram_from_iovec(skb, 0, iv, vnet_hdr_len, len); |
02df55d2 | 563 | if (err) |
b9fb9ee0 | 564 | goto err_kfree; |
20d29d7a AB |
565 | |
566 | skb_set_network_header(skb, ETH_HLEN); | |
b9fb9ee0 AB |
567 | skb_reset_mac_header(skb); |
568 | skb->protocol = eth_hdr(skb)->h_proto; | |
569 | ||
570 | if (vnet_hdr_len) { | |
571 | err = macvtap_skb_from_vnet_hdr(skb, &vnet_hdr); | |
572 | if (err) | |
573 | goto err_kfree; | |
574 | } | |
575 | ||
02df55d2 | 576 | rcu_read_lock_bh(); |
13707f9e | 577 | vlan = rcu_dereference_bh(q->vlan); |
02df55d2 AB |
578 | if (vlan) |
579 | macvlan_start_xmit(skb, vlan->dev); | |
580 | else | |
581 | kfree_skb(skb); | |
582 | rcu_read_unlock_bh(); | |
20d29d7a AB |
583 | |
584 | return count; | |
02df55d2 | 585 | |
b9fb9ee0 AB |
586 | err_kfree: |
587 | kfree_skb(skb); | |
588 | ||
02df55d2 AB |
589 | err: |
590 | rcu_read_lock_bh(); | |
13707f9e | 591 | vlan = rcu_dereference_bh(q->vlan); |
02df55d2 | 592 | if (vlan) |
1ac9ad13 | 593 | vlan->dev->stats.tx_dropped++; |
02df55d2 AB |
594 | rcu_read_unlock_bh(); |
595 | ||
02df55d2 | 596 | return err; |
20d29d7a AB |
597 | } |
598 | ||
599 | static ssize_t macvtap_aio_write(struct kiocb *iocb, const struct iovec *iv, | |
600 | unsigned long count, loff_t pos) | |
601 | { | |
602 | struct file *file = iocb->ki_filp; | |
603 | ssize_t result = -ENOLINK; | |
02df55d2 | 604 | struct macvtap_queue *q = file->private_data; |
20d29d7a AB |
605 | |
606 | result = macvtap_get_user(q, iv, iov_length(iv, count), | |
607 | file->f_flags & O_NONBLOCK); | |
20d29d7a AB |
608 | return result; |
609 | } | |
610 | ||
611 | /* Put packet to the user space buffer */ | |
612 | static ssize_t macvtap_put_user(struct macvtap_queue *q, | |
613 | const struct sk_buff *skb, | |
614 | const struct iovec *iv, int len) | |
615 | { | |
02df55d2 | 616 | struct macvlan_dev *vlan; |
20d29d7a | 617 | int ret; |
b9fb9ee0 AB |
618 | int vnet_hdr_len = 0; |
619 | ||
620 | if (q->flags & IFF_VNET_HDR) { | |
621 | struct virtio_net_hdr vnet_hdr; | |
55afbd08 | 622 | vnet_hdr_len = q->vnet_hdr_sz; |
b9fb9ee0 AB |
623 | if ((len -= vnet_hdr_len) < 0) |
624 | return -EINVAL; | |
625 | ||
626 | ret = macvtap_skb_to_vnet_hdr(skb, &vnet_hdr); | |
627 | if (ret) | |
628 | return ret; | |
629 | ||
55afbd08 | 630 | if (memcpy_toiovecend(iv, (void *)&vnet_hdr, 0, sizeof(vnet_hdr))) |
b9fb9ee0 AB |
631 | return -EFAULT; |
632 | } | |
20d29d7a AB |
633 | |
634 | len = min_t(int, skb->len, len); | |
635 | ||
b9fb9ee0 | 636 | ret = skb_copy_datagram_const_iovec(skb, 0, iv, vnet_hdr_len, len); |
20d29d7a | 637 | |
02df55d2 | 638 | rcu_read_lock_bh(); |
13707f9e | 639 | vlan = rcu_dereference_bh(q->vlan); |
501c774c AB |
640 | if (vlan) |
641 | macvlan_count_rx(vlan, len, ret == 0, 0); | |
02df55d2 | 642 | rcu_read_unlock_bh(); |
20d29d7a | 643 | |
b9fb9ee0 | 644 | return ret ? ret : (len + vnet_hdr_len); |
20d29d7a AB |
645 | } |
646 | ||
501c774c AB |
647 | static ssize_t macvtap_do_read(struct macvtap_queue *q, struct kiocb *iocb, |
648 | const struct iovec *iv, unsigned long len, | |
649 | int noblock) | |
20d29d7a | 650 | { |
20d29d7a AB |
651 | DECLARE_WAITQUEUE(wait, current); |
652 | struct sk_buff *skb; | |
501c774c | 653 | ssize_t ret = 0; |
20d29d7a | 654 | |
4a4771a5 | 655 | add_wait_queue(sk_sleep(&q->sk), &wait); |
20d29d7a AB |
656 | while (len) { |
657 | current->state = TASK_INTERRUPTIBLE; | |
658 | ||
659 | /* Read frames from the queue */ | |
660 | skb = skb_dequeue(&q->sk.sk_receive_queue); | |
661 | if (!skb) { | |
501c774c | 662 | if (noblock) { |
20d29d7a AB |
663 | ret = -EAGAIN; |
664 | break; | |
665 | } | |
666 | if (signal_pending(current)) { | |
667 | ret = -ERESTARTSYS; | |
668 | break; | |
669 | } | |
670 | /* Nothing to read, let's sleep */ | |
671 | schedule(); | |
672 | continue; | |
673 | } | |
674 | ret = macvtap_put_user(q, skb, iv, len); | |
675 | kfree_skb(skb); | |
676 | break; | |
677 | } | |
678 | ||
679 | current->state = TASK_RUNNING; | |
4a4771a5 | 680 | remove_wait_queue(sk_sleep(&q->sk), &wait); |
501c774c AB |
681 | return ret; |
682 | } | |
683 | ||
684 | static ssize_t macvtap_aio_read(struct kiocb *iocb, const struct iovec *iv, | |
685 | unsigned long count, loff_t pos) | |
686 | { | |
687 | struct file *file = iocb->ki_filp; | |
688 | struct macvtap_queue *q = file->private_data; | |
689 | ssize_t len, ret = 0; | |
20d29d7a | 690 | |
501c774c AB |
691 | len = iov_length(iv, count); |
692 | if (len < 0) { | |
693 | ret = -EINVAL; | |
694 | goto out; | |
695 | } | |
696 | ||
697 | ret = macvtap_do_read(q, iocb, iv, len, file->f_flags & O_NONBLOCK); | |
698 | ret = min_t(ssize_t, ret, len); /* XXX copied from tun.c. Why? */ | |
20d29d7a | 699 | out: |
20d29d7a AB |
700 | return ret; |
701 | } | |
702 | ||
703 | /* | |
704 | * provide compatibility with generic tun/tap interface | |
705 | */ | |
706 | static long macvtap_ioctl(struct file *file, unsigned int cmd, | |
707 | unsigned long arg) | |
708 | { | |
02df55d2 AB |
709 | struct macvtap_queue *q = file->private_data; |
710 | struct macvlan_dev *vlan; | |
20d29d7a AB |
711 | void __user *argp = (void __user *)arg; |
712 | struct ifreq __user *ifr = argp; | |
713 | unsigned int __user *up = argp; | |
714 | unsigned int u; | |
55afbd08 MT |
715 | int __user *sp = argp; |
716 | int s; | |
02df55d2 | 717 | int ret; |
20d29d7a AB |
718 | |
719 | switch (cmd) { | |
720 | case TUNSETIFF: | |
721 | /* ignore the name, just look at flags */ | |
722 | if (get_user(u, &ifr->ifr_flags)) | |
723 | return -EFAULT; | |
b9fb9ee0 AB |
724 | |
725 | ret = 0; | |
726 | if ((u & ~IFF_VNET_HDR) != (IFF_NO_PI | IFF_TAP)) | |
727 | ret = -EINVAL; | |
728 | else | |
729 | q->flags = u; | |
730 | ||
731 | return ret; | |
20d29d7a AB |
732 | |
733 | case TUNGETIFF: | |
02df55d2 | 734 | rcu_read_lock_bh(); |
13707f9e | 735 | vlan = rcu_dereference_bh(q->vlan); |
02df55d2 AB |
736 | if (vlan) |
737 | dev_hold(vlan->dev); | |
738 | rcu_read_unlock_bh(); | |
739 | ||
740 | if (!vlan) | |
20d29d7a | 741 | return -ENOLINK; |
20d29d7a | 742 | |
02df55d2 | 743 | ret = 0; |
13707f9e | 744 | if (copy_to_user(&ifr->ifr_name, vlan->dev->name, IFNAMSIZ) || |
b9fb9ee0 | 745 | put_user(q->flags, &ifr->ifr_flags)) |
02df55d2 AB |
746 | ret = -EFAULT; |
747 | dev_put(vlan->dev); | |
748 | return ret; | |
20d29d7a AB |
749 | |
750 | case TUNGETFEATURES: | |
b9fb9ee0 | 751 | if (put_user(IFF_TAP | IFF_NO_PI | IFF_VNET_HDR, up)) |
20d29d7a AB |
752 | return -EFAULT; |
753 | return 0; | |
754 | ||
755 | case TUNSETSNDBUF: | |
756 | if (get_user(u, up)) | |
757 | return -EFAULT; | |
758 | ||
20d29d7a | 759 | q->sk.sk_sndbuf = u; |
20d29d7a AB |
760 | return 0; |
761 | ||
55afbd08 MT |
762 | case TUNGETVNETHDRSZ: |
763 | s = q->vnet_hdr_sz; | |
764 | if (put_user(s, sp)) | |
765 | return -EFAULT; | |
766 | return 0; | |
767 | ||
768 | case TUNSETVNETHDRSZ: | |
769 | if (get_user(s, sp)) | |
770 | return -EFAULT; | |
771 | if (s < (int)sizeof(struct virtio_net_hdr)) | |
772 | return -EINVAL; | |
773 | ||
774 | q->vnet_hdr_sz = s; | |
775 | return 0; | |
776 | ||
20d29d7a AB |
777 | case TUNSETOFFLOAD: |
778 | /* let the user check for future flags */ | |
779 | if (arg & ~(TUN_F_CSUM | TUN_F_TSO4 | TUN_F_TSO6 | | |
b9fb9ee0 | 780 | TUN_F_TSO_ECN | TUN_F_UFO)) |
20d29d7a AB |
781 | return -EINVAL; |
782 | ||
b9fb9ee0 AB |
783 | /* TODO: only accept frames with the features that |
784 | got enabled for forwarded frames */ | |
785 | if (!(q->flags & IFF_VNET_HDR)) | |
786 | return -EINVAL; | |
20d29d7a AB |
787 | return 0; |
788 | ||
789 | default: | |
790 | return -EINVAL; | |
791 | } | |
792 | } | |
793 | ||
794 | #ifdef CONFIG_COMPAT | |
795 | static long macvtap_compat_ioctl(struct file *file, unsigned int cmd, | |
796 | unsigned long arg) | |
797 | { | |
798 | return macvtap_ioctl(file, cmd, (unsigned long)compat_ptr(arg)); | |
799 | } | |
800 | #endif | |
801 | ||
802 | static const struct file_operations macvtap_fops = { | |
803 | .owner = THIS_MODULE, | |
804 | .open = macvtap_open, | |
805 | .release = macvtap_release, | |
806 | .aio_read = macvtap_aio_read, | |
807 | .aio_write = macvtap_aio_write, | |
808 | .poll = macvtap_poll, | |
809 | .llseek = no_llseek, | |
810 | .unlocked_ioctl = macvtap_ioctl, | |
811 | #ifdef CONFIG_COMPAT | |
812 | .compat_ioctl = macvtap_compat_ioctl, | |
813 | #endif | |
814 | }; | |
815 | ||
501c774c AB |
816 | static int macvtap_sendmsg(struct kiocb *iocb, struct socket *sock, |
817 | struct msghdr *m, size_t total_len) | |
818 | { | |
819 | struct macvtap_queue *q = container_of(sock, struct macvtap_queue, sock); | |
820 | return macvtap_get_user(q, m->msg_iov, total_len, | |
821 | m->msg_flags & MSG_DONTWAIT); | |
822 | } | |
823 | ||
824 | static int macvtap_recvmsg(struct kiocb *iocb, struct socket *sock, | |
825 | struct msghdr *m, size_t total_len, | |
826 | int flags) | |
827 | { | |
828 | struct macvtap_queue *q = container_of(sock, struct macvtap_queue, sock); | |
829 | int ret; | |
830 | if (flags & ~(MSG_DONTWAIT|MSG_TRUNC)) | |
831 | return -EINVAL; | |
832 | ret = macvtap_do_read(q, iocb, m->msg_iov, total_len, | |
833 | flags & MSG_DONTWAIT); | |
834 | if (ret > total_len) { | |
835 | m->msg_flags |= MSG_TRUNC; | |
836 | ret = flags & MSG_TRUNC ? ret : total_len; | |
837 | } | |
838 | return ret; | |
839 | } | |
840 | ||
841 | /* Ops structure to mimic raw sockets with tun */ | |
842 | static const struct proto_ops macvtap_socket_ops = { | |
843 | .sendmsg = macvtap_sendmsg, | |
844 | .recvmsg = macvtap_recvmsg, | |
845 | }; | |
846 | ||
847 | /* Get an underlying socket object from tun file. Returns error unless file is | |
848 | * attached to a device. The returned object works like a packet socket, it | |
849 | * can be used for sock_sendmsg/sock_recvmsg. The caller is responsible for | |
850 | * holding a reference to the file for as long as the socket is in use. */ | |
851 | struct socket *macvtap_get_socket(struct file *file) | |
852 | { | |
853 | struct macvtap_queue *q; | |
854 | if (file->f_op != &macvtap_fops) | |
855 | return ERR_PTR(-EINVAL); | |
856 | q = file->private_data; | |
857 | if (!q) | |
858 | return ERR_PTR(-EBADFD); | |
859 | return &q->sock; | |
860 | } | |
861 | EXPORT_SYMBOL_GPL(macvtap_get_socket); | |
862 | ||
20d29d7a AB |
863 | static int macvtap_init(void) |
864 | { | |
865 | int err; | |
866 | ||
867 | err = alloc_chrdev_region(&macvtap_major, 0, | |
868 | MACVTAP_NUM_DEVS, "macvtap"); | |
869 | if (err) | |
870 | goto out1; | |
871 | ||
872 | cdev_init(&macvtap_cdev, &macvtap_fops); | |
873 | err = cdev_add(&macvtap_cdev, macvtap_major, MACVTAP_NUM_DEVS); | |
874 | if (err) | |
875 | goto out2; | |
876 | ||
877 | macvtap_class = class_create(THIS_MODULE, "macvtap"); | |
878 | if (IS_ERR(macvtap_class)) { | |
879 | err = PTR_ERR(macvtap_class); | |
880 | goto out3; | |
881 | } | |
882 | ||
883 | err = macvlan_link_register(&macvtap_link_ops); | |
884 | if (err) | |
885 | goto out4; | |
886 | ||
887 | return 0; | |
888 | ||
889 | out4: | |
890 | class_unregister(macvtap_class); | |
891 | out3: | |
892 | cdev_del(&macvtap_cdev); | |
893 | out2: | |
894 | unregister_chrdev_region(macvtap_major, MACVTAP_NUM_DEVS); | |
895 | out1: | |
896 | return err; | |
897 | } | |
898 | module_init(macvtap_init); | |
899 | ||
900 | static void macvtap_exit(void) | |
901 | { | |
902 | rtnl_link_unregister(&macvtap_link_ops); | |
903 | class_unregister(macvtap_class); | |
904 | cdev_del(&macvtap_cdev); | |
905 | unregister_chrdev_region(macvtap_major, MACVTAP_NUM_DEVS); | |
906 | } | |
907 | module_exit(macvtap_exit); | |
908 | ||
909 | MODULE_ALIAS_RTNL_LINK("macvtap"); | |
910 | MODULE_AUTHOR("Arnd Bergmann <arnd@arndb.de>"); | |
911 | MODULE_LICENSE("GPL"); |