Commit | Line | Data |
---|---|---|
20d29d7a AB |
1 | #include <linux/etherdevice.h> |
2 | #include <linux/if_macvlan.h> | |
f09e2249 | 3 | #include <linux/if_vlan.h> |
20d29d7a AB |
4 | #include <linux/interrupt.h> |
5 | #include <linux/nsproxy.h> | |
6 | #include <linux/compat.h> | |
7 | #include <linux/if_tun.h> | |
8 | #include <linux/module.h> | |
9 | #include <linux/skbuff.h> | |
10 | #include <linux/cache.h> | |
11 | #include <linux/sched.h> | |
12 | #include <linux/types.h> | |
5a0e3ad6 | 13 | #include <linux/slab.h> |
20d29d7a AB |
14 | #include <linux/init.h> |
15 | #include <linux/wait.h> | |
16 | #include <linux/cdev.h> | |
40401530 | 17 | #include <linux/idr.h> |
20d29d7a AB |
18 | #include <linux/fs.h> |
19 | ||
20 | #include <net/net_namespace.h> | |
21 | #include <net/rtnetlink.h> | |
22 | #include <net/sock.h> | |
b9fb9ee0 | 23 | #include <linux/virtio_net.h> |
20d29d7a AB |
24 | |
25 | /* | |
26 | * A macvtap queue is the central object of this driver, it connects | |
27 | * an open character device to a macvlan interface. There can be | |
28 | * multiple queues on one interface, which map back to queues | |
29 | * implemented in hardware on the underlying device. | |
30 | * | |
31 | * macvtap_proto is used to allocate queues through the sock allocation | |
32 | * mechanism. | |
33 | * | |
34 | * TODO: multiqueue support is currently not implemented, even though | |
35 | * macvtap is basically prepared for that. We will need to add this | |
36 | * here as well as in virtio-net and qemu to get line rate on 10gbit | |
37 | * adapters from a guest. | |
38 | */ | |
39 | struct macvtap_queue { | |
40 | struct sock sk; | |
41 | struct socket sock; | |
43815482 | 42 | struct socket_wq wq; |
55afbd08 | 43 | int vnet_hdr_sz; |
13707f9e | 44 | struct macvlan_dev __rcu *vlan; |
20d29d7a | 45 | struct file *file; |
b9fb9ee0 | 46 | unsigned int flags; |
20d29d7a AB |
47 | }; |
48 | ||
49 | static struct proto macvtap_proto = { | |
50 | .name = "macvtap", | |
51 | .owner = THIS_MODULE, | |
52 | .obj_size = sizeof (struct macvtap_queue), | |
53 | }; | |
54 | ||
55 | /* | |
e09eff7f | 56 | * Variables for dealing with macvtaps device numbers. |
20d29d7a | 57 | */ |
1ebed71a | 58 | static dev_t macvtap_major; |
e09eff7f EB |
59 | #define MACVTAP_NUM_DEVS (1U << MINORBITS) |
60 | static DEFINE_MUTEX(minor_lock); | |
61 | static DEFINE_IDR(minor_idr); | |
62 | ||
97bc3633 | 63 | #define GOODCOPY_LEN 128 |
20d29d7a AB |
64 | static struct class *macvtap_class; |
65 | static struct cdev macvtap_cdev; | |
66 | ||
501c774c AB |
67 | static const struct proto_ops macvtap_socket_ops; |
68 | ||
20d29d7a AB |
69 | /* |
70 | * RCU usage: | |
02df55d2 AB |
71 | * The macvtap_queue and the macvlan_dev are loosely coupled, the |
72 | * pointers from one to the other can only be read while rcu_read_lock | |
73 | * or macvtap_lock is held. | |
20d29d7a | 74 | * |
02df55d2 AB |
75 | * Both the file and the macvlan_dev hold a reference on the macvtap_queue |
76 | * through sock_hold(&q->sk). When the macvlan_dev goes away first, | |
77 | * q->vlan becomes inaccessible. When the files gets closed, | |
78 | * macvtap_get_queue() fails. | |
20d29d7a | 79 | * |
02df55d2 AB |
80 | * There may still be references to the struct sock inside of the |
81 | * queue from outbound SKBs, but these never reference back to the | |
82 | * file or the dev. The data structure is freed through __sk_free | |
83 | * when both our references and any pending SKBs are gone. | |
20d29d7a AB |
84 | */ |
85 | static DEFINE_SPINLOCK(macvtap_lock); | |
86 | ||
87 | /* | |
1565c7c1 KK |
88 | * get_slot: return a [unused/occupied] slot in vlan->taps[]: |
89 | * - if 'q' is NULL, return the first empty slot; | |
90 | * - otherwise, return the slot this pointer occupies. | |
20d29d7a | 91 | */ |
1565c7c1 KK |
92 | static int get_slot(struct macvlan_dev *vlan, struct macvtap_queue *q) |
93 | { | |
94 | int i; | |
95 | ||
96 | for (i = 0; i < MAX_MACVTAP_QUEUES; i++) { | |
3a7f8c34 DE |
97 | if (rcu_dereference_protected(vlan->taps[i], |
98 | lockdep_is_held(&macvtap_lock)) == q) | |
1565c7c1 KK |
99 | return i; |
100 | } | |
101 | ||
102 | /* Should never happen */ | |
103 | BUG_ON(1); | |
104 | } | |
105 | ||
20d29d7a AB |
106 | static int macvtap_set_queue(struct net_device *dev, struct file *file, |
107 | struct macvtap_queue *q) | |
108 | { | |
109 | struct macvlan_dev *vlan = netdev_priv(dev); | |
1565c7c1 | 110 | int index; |
20d29d7a AB |
111 | int err = -EBUSY; |
112 | ||
113 | spin_lock(&macvtap_lock); | |
1565c7c1 | 114 | if (vlan->numvtaps == MAX_MACVTAP_QUEUES) |
20d29d7a AB |
115 | goto out; |
116 | ||
117 | err = 0; | |
1565c7c1 | 118 | index = get_slot(vlan, NULL); |
02df55d2 | 119 | rcu_assign_pointer(q->vlan, vlan); |
1565c7c1 | 120 | rcu_assign_pointer(vlan->taps[index], q); |
02df55d2 | 121 | sock_hold(&q->sk); |
20d29d7a AB |
122 | |
123 | q->file = file; | |
02df55d2 | 124 | file->private_data = q; |
20d29d7a | 125 | |
1565c7c1 KK |
126 | vlan->numvtaps++; |
127 | ||
20d29d7a AB |
128 | out: |
129 | spin_unlock(&macvtap_lock); | |
130 | return err; | |
131 | } | |
132 | ||
133 | /* | |
02df55d2 AB |
134 | * The file owning the queue got closed, give up both |
135 | * the reference that the files holds as well as the | |
136 | * one from the macvlan_dev if that still exists. | |
20d29d7a AB |
137 | * |
138 | * Using the spinlock makes sure that we don't get | |
139 | * to the queue again after destroying it. | |
20d29d7a | 140 | */ |
02df55d2 | 141 | static void macvtap_put_queue(struct macvtap_queue *q) |
20d29d7a | 142 | { |
02df55d2 | 143 | struct macvlan_dev *vlan; |
20d29d7a AB |
144 | |
145 | spin_lock(&macvtap_lock); | |
13707f9e ED |
146 | vlan = rcu_dereference_protected(q->vlan, |
147 | lockdep_is_held(&macvtap_lock)); | |
02df55d2 | 148 | if (vlan) { |
1565c7c1 KK |
149 | int index = get_slot(vlan, q); |
150 | ||
2cfa5a04 ED |
151 | RCU_INIT_POINTER(vlan->taps[index], NULL); |
152 | RCU_INIT_POINTER(q->vlan, NULL); | |
02df55d2 | 153 | sock_put(&q->sk); |
1565c7c1 | 154 | --vlan->numvtaps; |
20d29d7a AB |
155 | } |
156 | ||
20d29d7a AB |
157 | spin_unlock(&macvtap_lock); |
158 | ||
159 | synchronize_rcu(); | |
160 | sock_put(&q->sk); | |
161 | } | |
162 | ||
163 | /* | |
1565c7c1 KK |
164 | * Select a queue based on the rxq of the device on which this packet |
165 | * arrived. If the incoming device is not mq, calculate a flow hash | |
166 | * to select a queue. If all fails, find the first available queue. | |
167 | * Cache vlan->numvtaps since it can become zero during the execution | |
168 | * of this function. | |
20d29d7a AB |
169 | */ |
170 | static struct macvtap_queue *macvtap_get_queue(struct net_device *dev, | |
171 | struct sk_buff *skb) | |
172 | { | |
173 | struct macvlan_dev *vlan = netdev_priv(dev); | |
1565c7c1 KK |
174 | struct macvtap_queue *tap = NULL; |
175 | int numvtaps = vlan->numvtaps; | |
176 | __u32 rxq; | |
177 | ||
178 | if (!numvtaps) | |
179 | goto out; | |
180 | ||
ef0002b5 KK |
181 | /* Check if we can use flow to select a queue */ |
182 | rxq = skb_get_rxhash(skb); | |
183 | if (rxq) { | |
184 | tap = rcu_dereference(vlan->taps[rxq % numvtaps]); | |
185 | if (tap) | |
186 | goto out; | |
187 | } | |
188 | ||
1565c7c1 KK |
189 | if (likely(skb_rx_queue_recorded(skb))) { |
190 | rxq = skb_get_rx_queue(skb); | |
20d29d7a | 191 | |
1565c7c1 KK |
192 | while (unlikely(rxq >= numvtaps)) |
193 | rxq -= numvtaps; | |
194 | ||
195 | tap = rcu_dereference(vlan->taps[rxq]); | |
196 | if (tap) | |
197 | goto out; | |
198 | } | |
199 | ||
1565c7c1 KK |
200 | /* Everything failed - find first available queue */ |
201 | for (rxq = 0; rxq < MAX_MACVTAP_QUEUES; rxq++) { | |
202 | tap = rcu_dereference(vlan->taps[rxq]); | |
203 | if (tap) | |
204 | break; | |
205 | } | |
206 | ||
207 | out: | |
208 | return tap; | |
20d29d7a AB |
209 | } |
210 | ||
02df55d2 AB |
211 | /* |
212 | * The net_device is going away, give up the reference | |
1565c7c1 KK |
213 | * that it holds on all queues and safely set the pointer |
214 | * from the queues to NULL. | |
02df55d2 | 215 | */ |
20d29d7a AB |
216 | static void macvtap_del_queues(struct net_device *dev) |
217 | { | |
218 | struct macvlan_dev *vlan = netdev_priv(dev); | |
1565c7c1 KK |
219 | struct macvtap_queue *q, *qlist[MAX_MACVTAP_QUEUES]; |
220 | int i, j = 0; | |
02df55d2 | 221 | |
1565c7c1 | 222 | /* macvtap_put_queue can free some slots, so go through all slots */ |
02df55d2 | 223 | spin_lock(&macvtap_lock); |
1565c7c1 | 224 | for (i = 0; i < MAX_MACVTAP_QUEUES && vlan->numvtaps; i++) { |
13707f9e ED |
225 | q = rcu_dereference_protected(vlan->taps[i], |
226 | lockdep_is_held(&macvtap_lock)); | |
1565c7c1 KK |
227 | if (q) { |
228 | qlist[j++] = q; | |
2cfa5a04 ED |
229 | RCU_INIT_POINTER(vlan->taps[i], NULL); |
230 | RCU_INIT_POINTER(q->vlan, NULL); | |
1565c7c1 KK |
231 | vlan->numvtaps--; |
232 | } | |
564517e8 | 233 | } |
1565c7c1 | 234 | BUG_ON(vlan->numvtaps != 0); |
99f34b38 EB |
235 | /* guarantee that any future macvtap_set_queue will fail */ |
236 | vlan->numvtaps = MAX_MACVTAP_QUEUES; | |
02df55d2 AB |
237 | spin_unlock(&macvtap_lock); |
238 | ||
239 | synchronize_rcu(); | |
1565c7c1 KK |
240 | |
241 | for (--j; j >= 0; j--) | |
242 | sock_put(&qlist[j]->sk); | |
20d29d7a AB |
243 | } |
244 | ||
245 | /* | |
246 | * Forward happens for data that gets sent from one macvlan | |
247 | * endpoint to another one in bridge mode. We just take | |
248 | * the skb and put it into the receive queue. | |
249 | */ | |
250 | static int macvtap_forward(struct net_device *dev, struct sk_buff *skb) | |
251 | { | |
252 | struct macvtap_queue *q = macvtap_get_queue(dev, skb); | |
253 | if (!q) | |
8a35747a HX |
254 | goto drop; |
255 | ||
256 | if (skb_queue_len(&q->sk.sk_receive_queue) >= dev->tx_queue_len) | |
257 | goto drop; | |
20d29d7a AB |
258 | |
259 | skb_queue_tail(&q->sk.sk_receive_queue, skb); | |
4a4771a5 | 260 | wake_up_interruptible_poll(sk_sleep(&q->sk), POLLIN | POLLRDNORM | POLLRDBAND); |
8a35747a HX |
261 | return NET_RX_SUCCESS; |
262 | ||
263 | drop: | |
264 | kfree_skb(skb); | |
265 | return NET_RX_DROP; | |
20d29d7a AB |
266 | } |
267 | ||
268 | /* | |
269 | * Receive is for data from the external interface (lowerdev), | |
270 | * in case of macvtap, we can treat that the same way as | |
271 | * forward, which macvlan cannot. | |
272 | */ | |
273 | static int macvtap_receive(struct sk_buff *skb) | |
274 | { | |
275 | skb_push(skb, ETH_HLEN); | |
276 | return macvtap_forward(skb->dev, skb); | |
277 | } | |
278 | ||
e09eff7f EB |
279 | static int macvtap_get_minor(struct macvlan_dev *vlan) |
280 | { | |
281 | int retval = -ENOMEM; | |
282 | int id; | |
283 | ||
284 | mutex_lock(&minor_lock); | |
285 | if (idr_pre_get(&minor_idr, GFP_KERNEL) == 0) | |
286 | goto exit; | |
287 | ||
288 | retval = idr_get_new_above(&minor_idr, vlan, 1, &id); | |
289 | if (retval < 0) { | |
290 | if (retval == -EAGAIN) | |
291 | retval = -ENOMEM; | |
292 | goto exit; | |
293 | } | |
294 | if (id < MACVTAP_NUM_DEVS) { | |
295 | vlan->minor = id; | |
296 | } else { | |
297 | printk(KERN_ERR "too many macvtap devices\n"); | |
298 | retval = -EINVAL; | |
299 | idr_remove(&minor_idr, id); | |
300 | } | |
301 | exit: | |
302 | mutex_unlock(&minor_lock); | |
303 | return retval; | |
304 | } | |
305 | ||
306 | static void macvtap_free_minor(struct macvlan_dev *vlan) | |
307 | { | |
308 | mutex_lock(&minor_lock); | |
309 | if (vlan->minor) { | |
310 | idr_remove(&minor_idr, vlan->minor); | |
311 | vlan->minor = 0; | |
312 | } | |
313 | mutex_unlock(&minor_lock); | |
314 | } | |
315 | ||
316 | static struct net_device *dev_get_by_macvtap_minor(int minor) | |
317 | { | |
318 | struct net_device *dev = NULL; | |
319 | struct macvlan_dev *vlan; | |
320 | ||
321 | mutex_lock(&minor_lock); | |
322 | vlan = idr_find(&minor_idr, minor); | |
323 | if (vlan) { | |
324 | dev = vlan->dev; | |
325 | dev_hold(dev); | |
326 | } | |
327 | mutex_unlock(&minor_lock); | |
328 | return dev; | |
329 | } | |
330 | ||
20d29d7a AB |
331 | static int macvtap_newlink(struct net *src_net, |
332 | struct net_device *dev, | |
333 | struct nlattr *tb[], | |
334 | struct nlattr *data[]) | |
335 | { | |
9bf1907f EB |
336 | /* Don't put anything that may fail after macvlan_common_newlink |
337 | * because we can't undo what it does. | |
338 | */ | |
339 | return macvlan_common_newlink(src_net, dev, tb, data, | |
340 | macvtap_receive, macvtap_forward); | |
20d29d7a AB |
341 | } |
342 | ||
343 | static void macvtap_dellink(struct net_device *dev, | |
344 | struct list_head *head) | |
345 | { | |
20d29d7a AB |
346 | macvtap_del_queues(dev); |
347 | macvlan_dellink(dev, head); | |
348 | } | |
349 | ||
8a35747a HX |
350 | static void macvtap_setup(struct net_device *dev) |
351 | { | |
352 | macvlan_common_setup(dev); | |
353 | dev->tx_queue_len = TUN_READQ_SIZE; | |
354 | } | |
355 | ||
20d29d7a AB |
356 | static struct rtnl_link_ops macvtap_link_ops __read_mostly = { |
357 | .kind = "macvtap", | |
8a35747a | 358 | .setup = macvtap_setup, |
20d29d7a AB |
359 | .newlink = macvtap_newlink, |
360 | .dellink = macvtap_dellink, | |
361 | }; | |
362 | ||
363 | ||
364 | static void macvtap_sock_write_space(struct sock *sk) | |
365 | { | |
43815482 ED |
366 | wait_queue_head_t *wqueue; |
367 | ||
20d29d7a AB |
368 | if (!sock_writeable(sk) || |
369 | !test_and_clear_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags)) | |
370 | return; | |
371 | ||
43815482 ED |
372 | wqueue = sk_sleep(sk); |
373 | if (wqueue && waitqueue_active(wqueue)) | |
374 | wake_up_interruptible_poll(wqueue, POLLOUT | POLLWRNORM | POLLWRBAND); | |
20d29d7a AB |
375 | } |
376 | ||
2259fef0 EB |
377 | static void macvtap_sock_destruct(struct sock *sk) |
378 | { | |
379 | skb_queue_purge(&sk->sk_receive_queue); | |
380 | } | |
381 | ||
20d29d7a AB |
382 | static int macvtap_open(struct inode *inode, struct file *file) |
383 | { | |
384 | struct net *net = current->nsproxy->net_ns; | |
e09eff7f | 385 | struct net_device *dev = dev_get_by_macvtap_minor(iminor(inode)); |
20d29d7a AB |
386 | struct macvtap_queue *q; |
387 | int err; | |
388 | ||
389 | err = -ENODEV; | |
390 | if (!dev) | |
391 | goto out; | |
392 | ||
20d29d7a AB |
393 | err = -ENOMEM; |
394 | q = (struct macvtap_queue *)sk_alloc(net, AF_UNSPEC, GFP_KERNEL, | |
395 | &macvtap_proto); | |
396 | if (!q) | |
397 | goto out; | |
398 | ||
43815482 ED |
399 | q->sock.wq = &q->wq; |
400 | init_waitqueue_head(&q->wq.wait); | |
20d29d7a AB |
401 | q->sock.type = SOCK_RAW; |
402 | q->sock.state = SS_CONNECTED; | |
501c774c AB |
403 | q->sock.file = file; |
404 | q->sock.ops = &macvtap_socket_ops; | |
20d29d7a | 405 | sock_init_data(&q->sock, &q->sk); |
20d29d7a | 406 | q->sk.sk_write_space = macvtap_sock_write_space; |
2259fef0 | 407 | q->sk.sk_destruct = macvtap_sock_destruct; |
b9fb9ee0 | 408 | q->flags = IFF_VNET_HDR | IFF_NO_PI | IFF_TAP; |
55afbd08 | 409 | q->vnet_hdr_sz = sizeof(struct virtio_net_hdr); |
20d29d7a | 410 | |
97bc3633 SM |
411 | /* |
412 | * so far only KVM virtio_net uses macvtap, enable zero copy between | |
413 | * guest kernel and host kernel when lower device supports zerocopy | |
047af9cf EB |
414 | * |
415 | * The macvlan supports zerocopy iff the lower device supports zero | |
416 | * copy so we don't have to look at the lower device directly. | |
97bc3633 | 417 | */ |
047af9cf EB |
418 | if ((dev->features & NETIF_F_HIGHDMA) && (dev->features & NETIF_F_SG)) |
419 | sock_set_flag(&q->sk, SOCK_ZEROCOPY); | |
97bc3633 | 420 | |
20d29d7a AB |
421 | err = macvtap_set_queue(dev, file, q); |
422 | if (err) | |
423 | sock_put(&q->sk); | |
424 | ||
425 | out: | |
426 | if (dev) | |
427 | dev_put(dev); | |
428 | ||
429 | return err; | |
430 | } | |
431 | ||
432 | static int macvtap_release(struct inode *inode, struct file *file) | |
433 | { | |
02df55d2 AB |
434 | struct macvtap_queue *q = file->private_data; |
435 | macvtap_put_queue(q); | |
20d29d7a AB |
436 | return 0; |
437 | } | |
438 | ||
439 | static unsigned int macvtap_poll(struct file *file, poll_table * wait) | |
440 | { | |
02df55d2 | 441 | struct macvtap_queue *q = file->private_data; |
20d29d7a AB |
442 | unsigned int mask = POLLERR; |
443 | ||
444 | if (!q) | |
445 | goto out; | |
446 | ||
447 | mask = 0; | |
43815482 | 448 | poll_wait(file, &q->wq.wait, wait); |
20d29d7a AB |
449 | |
450 | if (!skb_queue_empty(&q->sk.sk_receive_queue)) | |
451 | mask |= POLLIN | POLLRDNORM; | |
452 | ||
453 | if (sock_writeable(&q->sk) || | |
454 | (!test_and_set_bit(SOCK_ASYNC_NOSPACE, &q->sock.flags) && | |
455 | sock_writeable(&q->sk))) | |
456 | mask |= POLLOUT | POLLWRNORM; | |
457 | ||
458 | out: | |
20d29d7a AB |
459 | return mask; |
460 | } | |
461 | ||
b9fb9ee0 AB |
462 | static inline struct sk_buff *macvtap_alloc_skb(struct sock *sk, size_t prepad, |
463 | size_t len, size_t linear, | |
464 | int noblock, int *err) | |
465 | { | |
466 | struct sk_buff *skb; | |
467 | ||
468 | /* Under a page? Don't bother with paged skb. */ | |
469 | if (prepad + len < PAGE_SIZE || !linear) | |
470 | linear = len; | |
471 | ||
472 | skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock, | |
473 | err); | |
474 | if (!skb) | |
475 | return NULL; | |
476 | ||
477 | skb_reserve(skb, prepad); | |
478 | skb_put(skb, linear); | |
479 | skb->data_len = len - linear; | |
480 | skb->len += len - linear; | |
481 | ||
482 | return skb; | |
483 | } | |
484 | ||
97bc3633 SM |
485 | /* set skb frags from iovec, this can move to core network code for reuse */ |
486 | static int zerocopy_sg_from_iovec(struct sk_buff *skb, const struct iovec *from, | |
487 | int offset, size_t count) | |
488 | { | |
489 | int len = iov_length(from, count) - offset; | |
490 | int copy = skb_headlen(skb); | |
491 | int size, offset1 = 0; | |
492 | int i = 0; | |
97bc3633 SM |
493 | |
494 | /* Skip over from offset */ | |
495 | while (count && (offset >= from->iov_len)) { | |
496 | offset -= from->iov_len; | |
497 | ++from; | |
498 | --count; | |
499 | } | |
500 | ||
501 | /* copy up to skb headlen */ | |
502 | while (count && (copy > 0)) { | |
503 | size = min_t(unsigned int, copy, from->iov_len - offset); | |
504 | if (copy_from_user(skb->data + offset1, from->iov_base + offset, | |
505 | size)) | |
506 | return -EFAULT; | |
507 | if (copy > size) { | |
508 | ++from; | |
509 | --count; | |
3afc9621 JW |
510 | offset = 0; |
511 | } else | |
512 | offset += size; | |
97bc3633 SM |
513 | copy -= size; |
514 | offset1 += size; | |
97bc3633 SM |
515 | } |
516 | ||
517 | if (len == offset1) | |
518 | return 0; | |
519 | ||
520 | while (count--) { | |
521 | struct page *page[MAX_SKB_FRAGS]; | |
522 | int num_pages; | |
523 | unsigned long base; | |
4ef67ebe | 524 | unsigned long truesize; |
97bc3633 | 525 | |
3afc9621 | 526 | len = from->iov_len - offset; |
97bc3633 | 527 | if (!len) { |
3afc9621 | 528 | offset = 0; |
97bc3633 SM |
529 | ++from; |
530 | continue; | |
531 | } | |
3afc9621 | 532 | base = (unsigned long)from->iov_base + offset; |
97bc3633 | 533 | size = ((base & ~PAGE_MASK) + len + ~PAGE_MASK) >> PAGE_SHIFT; |
b92946e2 JW |
534 | if (i + size > MAX_SKB_FRAGS) |
535 | return -EMSGSIZE; | |
97bc3633 | 536 | num_pages = get_user_pages_fast(base, size, 0, &page[i]); |
b92946e2 | 537 | if (num_pages != size) { |
02ce04bb JW |
538 | for (i = 0; i < num_pages; i++) |
539 | put_page(page[i]); | |
97bc3633 | 540 | return -EFAULT; |
02ce04bb | 541 | } |
4ef67ebe | 542 | truesize = size * PAGE_SIZE; |
97bc3633 SM |
543 | skb->data_len += len; |
544 | skb->len += len; | |
4ef67ebe JW |
545 | skb->truesize += truesize; |
546 | atomic_add(truesize, &skb->sk->sk_wmem_alloc); | |
97bc3633 | 547 | while (len) { |
653fc915 JW |
548 | int off = base & ~PAGE_MASK; |
549 | int size = min_t(int, len, PAGE_SIZE - off); | |
550 | __skb_fill_page_desc(skb, i, page[i], off, size); | |
97bc3633 SM |
551 | skb_shinfo(skb)->nr_frags++; |
552 | /* increase sk_wmem_alloc */ | |
653fc915 JW |
553 | base += size; |
554 | len -= size; | |
97bc3633 SM |
555 | i++; |
556 | } | |
3afc9621 | 557 | offset = 0; |
97bc3633 SM |
558 | ++from; |
559 | } | |
560 | return 0; | |
561 | } | |
562 | ||
b9fb9ee0 AB |
563 | /* |
564 | * macvtap_skb_from_vnet_hdr and macvtap_skb_to_vnet_hdr should | |
565 | * be shared with the tun/tap driver. | |
566 | */ | |
567 | static int macvtap_skb_from_vnet_hdr(struct sk_buff *skb, | |
568 | struct virtio_net_hdr *vnet_hdr) | |
569 | { | |
570 | unsigned short gso_type = 0; | |
571 | if (vnet_hdr->gso_type != VIRTIO_NET_HDR_GSO_NONE) { | |
572 | switch (vnet_hdr->gso_type & ~VIRTIO_NET_HDR_GSO_ECN) { | |
573 | case VIRTIO_NET_HDR_GSO_TCPV4: | |
574 | gso_type = SKB_GSO_TCPV4; | |
575 | break; | |
576 | case VIRTIO_NET_HDR_GSO_TCPV6: | |
577 | gso_type = SKB_GSO_TCPV6; | |
578 | break; | |
579 | case VIRTIO_NET_HDR_GSO_UDP: | |
580 | gso_type = SKB_GSO_UDP; | |
581 | break; | |
582 | default: | |
583 | return -EINVAL; | |
584 | } | |
585 | ||
586 | if (vnet_hdr->gso_type & VIRTIO_NET_HDR_GSO_ECN) | |
587 | gso_type |= SKB_GSO_TCP_ECN; | |
588 | ||
589 | if (vnet_hdr->gso_size == 0) | |
590 | return -EINVAL; | |
591 | } | |
592 | ||
593 | if (vnet_hdr->flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) { | |
594 | if (!skb_partial_csum_set(skb, vnet_hdr->csum_start, | |
595 | vnet_hdr->csum_offset)) | |
596 | return -EINVAL; | |
597 | } | |
598 | ||
599 | if (vnet_hdr->gso_type != VIRTIO_NET_HDR_GSO_NONE) { | |
600 | skb_shinfo(skb)->gso_size = vnet_hdr->gso_size; | |
601 | skb_shinfo(skb)->gso_type = gso_type; | |
602 | ||
603 | /* Header must be checked, and gso_segs computed. */ | |
604 | skb_shinfo(skb)->gso_type |= SKB_GSO_DODGY; | |
605 | skb_shinfo(skb)->gso_segs = 0; | |
606 | } | |
607 | return 0; | |
608 | } | |
609 | ||
610 | static int macvtap_skb_to_vnet_hdr(const struct sk_buff *skb, | |
611 | struct virtio_net_hdr *vnet_hdr) | |
612 | { | |
613 | memset(vnet_hdr, 0, sizeof(*vnet_hdr)); | |
614 | ||
615 | if (skb_is_gso(skb)) { | |
616 | struct skb_shared_info *sinfo = skb_shinfo(skb); | |
617 | ||
618 | /* This is a hint as to how much should be linear. */ | |
619 | vnet_hdr->hdr_len = skb_headlen(skb); | |
620 | vnet_hdr->gso_size = sinfo->gso_size; | |
621 | if (sinfo->gso_type & SKB_GSO_TCPV4) | |
622 | vnet_hdr->gso_type = VIRTIO_NET_HDR_GSO_TCPV4; | |
623 | else if (sinfo->gso_type & SKB_GSO_TCPV6) | |
624 | vnet_hdr->gso_type = VIRTIO_NET_HDR_GSO_TCPV6; | |
625 | else if (sinfo->gso_type & SKB_GSO_UDP) | |
626 | vnet_hdr->gso_type = VIRTIO_NET_HDR_GSO_UDP; | |
627 | else | |
628 | BUG(); | |
629 | if (sinfo->gso_type & SKB_GSO_TCP_ECN) | |
630 | vnet_hdr->gso_type |= VIRTIO_NET_HDR_GSO_ECN; | |
631 | } else | |
632 | vnet_hdr->gso_type = VIRTIO_NET_HDR_GSO_NONE; | |
633 | ||
634 | if (skb->ip_summed == CHECKSUM_PARTIAL) { | |
635 | vnet_hdr->flags = VIRTIO_NET_HDR_F_NEEDS_CSUM; | |
55508d60 | 636 | vnet_hdr->csum_start = skb_checksum_start_offset(skb); |
b9fb9ee0 | 637 | vnet_hdr->csum_offset = skb->csum_offset; |
10a8d94a JW |
638 | } else if (skb->ip_summed == CHECKSUM_UNNECESSARY) { |
639 | vnet_hdr->flags = VIRTIO_NET_HDR_F_DATA_VALID; | |
b9fb9ee0 AB |
640 | } /* else everything is zero */ |
641 | ||
642 | return 0; | |
643 | } | |
644 | ||
645 | ||
20d29d7a | 646 | /* Get packet from user space buffer */ |
97bc3633 SM |
647 | static ssize_t macvtap_get_user(struct macvtap_queue *q, struct msghdr *m, |
648 | const struct iovec *iv, unsigned long total_len, | |
649 | size_t count, int noblock) | |
20d29d7a AB |
650 | { |
651 | struct sk_buff *skb; | |
02df55d2 | 652 | struct macvlan_dev *vlan; |
97bc3633 | 653 | unsigned long len = total_len; |
20d29d7a | 654 | int err; |
b9fb9ee0 AB |
655 | struct virtio_net_hdr vnet_hdr = { 0 }; |
656 | int vnet_hdr_len = 0; | |
b92946e2 | 657 | int copylen = 0; |
97bc3633 | 658 | bool zerocopy = false; |
b9fb9ee0 AB |
659 | |
660 | if (q->flags & IFF_VNET_HDR) { | |
55afbd08 | 661 | vnet_hdr_len = q->vnet_hdr_sz; |
b9fb9ee0 AB |
662 | |
663 | err = -EINVAL; | |
ce3c8692 | 664 | if (len < vnet_hdr_len) |
b9fb9ee0 | 665 | goto err; |
ce3c8692 | 666 | len -= vnet_hdr_len; |
b9fb9ee0 AB |
667 | |
668 | err = memcpy_fromiovecend((void *)&vnet_hdr, iv, 0, | |
55afbd08 | 669 | sizeof(vnet_hdr)); |
b9fb9ee0 AB |
670 | if (err < 0) |
671 | goto err; | |
672 | if ((vnet_hdr.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) && | |
673 | vnet_hdr.csum_start + vnet_hdr.csum_offset + 2 > | |
674 | vnet_hdr.hdr_len) | |
675 | vnet_hdr.hdr_len = vnet_hdr.csum_start + | |
676 | vnet_hdr.csum_offset + 2; | |
677 | err = -EINVAL; | |
678 | if (vnet_hdr.hdr_len > len) | |
679 | goto err; | |
680 | } | |
20d29d7a | 681 | |
b9fb9ee0 | 682 | err = -EINVAL; |
20d29d7a | 683 | if (unlikely(len < ETH_HLEN)) |
b9fb9ee0 | 684 | goto err; |
20d29d7a | 685 | |
b92946e2 JW |
686 | err = -EMSGSIZE; |
687 | if (unlikely(count > UIO_MAXIOV)) | |
688 | goto err; | |
689 | ||
97bc3633 SM |
690 | if (m && m->msg_control && sock_flag(&q->sk, SOCK_ZEROCOPY)) |
691 | zerocopy = true; | |
692 | ||
693 | if (zerocopy) { | |
b92946e2 JW |
694 | /* Userspace may produce vectors with count greater than |
695 | * MAX_SKB_FRAGS, so we need to linearize parts of the skb | |
696 | * to let the rest of data to be fit in the frags. | |
697 | */ | |
698 | if (count > MAX_SKB_FRAGS) { | |
699 | copylen = iov_length(iv, count - MAX_SKB_FRAGS); | |
700 | if (copylen < vnet_hdr_len) | |
701 | copylen = 0; | |
702 | else | |
703 | copylen -= vnet_hdr_len; | |
704 | } | |
97bc3633 SM |
705 | /* There are 256 bytes to be copied in skb, so there is enough |
706 | * room for skb expand head in case it is used. | |
707 | * The rest buffer is mapped from userspace. | |
708 | */ | |
b92946e2 JW |
709 | if (copylen < vnet_hdr.hdr_len) |
710 | copylen = vnet_hdr.hdr_len; | |
97bc3633 SM |
711 | if (!copylen) |
712 | copylen = GOODCOPY_LEN; | |
713 | } else | |
714 | copylen = len; | |
715 | ||
716 | skb = macvtap_alloc_skb(&q->sk, NET_IP_ALIGN, copylen, | |
717 | vnet_hdr.hdr_len, noblock, &err); | |
02df55d2 AB |
718 | if (!skb) |
719 | goto err; | |
20d29d7a | 720 | |
01d6657b | 721 | if (zerocopy) |
97bc3633 | 722 | err = zerocopy_sg_from_iovec(skb, iv, vnet_hdr_len, count); |
01d6657b | 723 | else |
97bc3633 SM |
724 | err = skb_copy_datagram_from_iovec(skb, 0, iv, vnet_hdr_len, |
725 | len); | |
02df55d2 | 726 | if (err) |
b9fb9ee0 | 727 | goto err_kfree; |
20d29d7a AB |
728 | |
729 | skb_set_network_header(skb, ETH_HLEN); | |
b9fb9ee0 AB |
730 | skb_reset_mac_header(skb); |
731 | skb->protocol = eth_hdr(skb)->h_proto; | |
732 | ||
733 | if (vnet_hdr_len) { | |
734 | err = macvtap_skb_from_vnet_hdr(skb, &vnet_hdr); | |
735 | if (err) | |
736 | goto err_kfree; | |
737 | } | |
738 | ||
02df55d2 | 739 | rcu_read_lock_bh(); |
13707f9e | 740 | vlan = rcu_dereference_bh(q->vlan); |
97bc3633 | 741 | /* copy skb_ubuf_info for callback when skb has no error */ |
01d6657b | 742 | if (zerocopy) { |
97bc3633 | 743 | skb_shinfo(skb)->destructor_arg = m->msg_control; |
01d6657b JW |
744 | skb_shinfo(skb)->tx_flags |= SKBTX_DEV_ZEROCOPY; |
745 | } | |
02df55d2 AB |
746 | if (vlan) |
747 | macvlan_start_xmit(skb, vlan->dev); | |
748 | else | |
749 | kfree_skb(skb); | |
750 | rcu_read_unlock_bh(); | |
20d29d7a | 751 | |
97bc3633 | 752 | return total_len; |
02df55d2 | 753 | |
b9fb9ee0 AB |
754 | err_kfree: |
755 | kfree_skb(skb); | |
756 | ||
02df55d2 AB |
757 | err: |
758 | rcu_read_lock_bh(); | |
13707f9e | 759 | vlan = rcu_dereference_bh(q->vlan); |
02df55d2 | 760 | if (vlan) |
1ac9ad13 | 761 | vlan->dev->stats.tx_dropped++; |
02df55d2 AB |
762 | rcu_read_unlock_bh(); |
763 | ||
02df55d2 | 764 | return err; |
20d29d7a AB |
765 | } |
766 | ||
767 | static ssize_t macvtap_aio_write(struct kiocb *iocb, const struct iovec *iv, | |
768 | unsigned long count, loff_t pos) | |
769 | { | |
770 | struct file *file = iocb->ki_filp; | |
771 | ssize_t result = -ENOLINK; | |
02df55d2 | 772 | struct macvtap_queue *q = file->private_data; |
20d29d7a | 773 | |
97bc3633 SM |
774 | result = macvtap_get_user(q, NULL, iv, iov_length(iv, count), count, |
775 | file->f_flags & O_NONBLOCK); | |
20d29d7a AB |
776 | return result; |
777 | } | |
778 | ||
779 | /* Put packet to the user space buffer */ | |
780 | static ssize_t macvtap_put_user(struct macvtap_queue *q, | |
781 | const struct sk_buff *skb, | |
782 | const struct iovec *iv, int len) | |
783 | { | |
02df55d2 | 784 | struct macvlan_dev *vlan; |
20d29d7a | 785 | int ret; |
b9fb9ee0 | 786 | int vnet_hdr_len = 0; |
f09e2249 BG |
787 | int vlan_offset = 0; |
788 | int copied; | |
b9fb9ee0 AB |
789 | |
790 | if (q->flags & IFF_VNET_HDR) { | |
791 | struct virtio_net_hdr vnet_hdr; | |
55afbd08 | 792 | vnet_hdr_len = q->vnet_hdr_sz; |
b9fb9ee0 AB |
793 | if ((len -= vnet_hdr_len) < 0) |
794 | return -EINVAL; | |
795 | ||
796 | ret = macvtap_skb_to_vnet_hdr(skb, &vnet_hdr); | |
797 | if (ret) | |
798 | return ret; | |
799 | ||
55afbd08 | 800 | if (memcpy_toiovecend(iv, (void *)&vnet_hdr, 0, sizeof(vnet_hdr))) |
b9fb9ee0 AB |
801 | return -EFAULT; |
802 | } | |
f09e2249 BG |
803 | copied = vnet_hdr_len; |
804 | ||
805 | if (!vlan_tx_tag_present(skb)) | |
806 | len = min_t(int, skb->len, len); | |
807 | else { | |
808 | int copy; | |
809 | struct { | |
810 | __be16 h_vlan_proto; | |
811 | __be16 h_vlan_TCI; | |
812 | } veth; | |
813 | veth.h_vlan_proto = htons(ETH_P_8021Q); | |
814 | veth.h_vlan_TCI = htons(vlan_tx_tag_get(skb)); | |
815 | ||
816 | vlan_offset = offsetof(struct vlan_ethhdr, h_vlan_proto); | |
817 | len = min_t(int, skb->len + VLAN_HLEN, len); | |
818 | ||
819 | copy = min_t(int, vlan_offset, len); | |
820 | ret = skb_copy_datagram_const_iovec(skb, 0, iv, copied, copy); | |
821 | len -= copy; | |
822 | copied += copy; | |
823 | if (ret || !len) | |
824 | goto done; | |
825 | ||
826 | copy = min_t(int, sizeof(veth), len); | |
827 | ret = memcpy_toiovecend(iv, (void *)&veth, copied, copy); | |
828 | len -= copy; | |
829 | copied += copy; | |
830 | if (ret || !len) | |
831 | goto done; | |
832 | } | |
20d29d7a | 833 | |
f09e2249 BG |
834 | ret = skb_copy_datagram_const_iovec(skb, vlan_offset, iv, copied, len); |
835 | copied += len; | |
20d29d7a | 836 | |
f09e2249 | 837 | done: |
02df55d2 | 838 | rcu_read_lock_bh(); |
13707f9e | 839 | vlan = rcu_dereference_bh(q->vlan); |
501c774c | 840 | if (vlan) |
f09e2249 | 841 | macvlan_count_rx(vlan, copied - vnet_hdr_len, ret == 0, 0); |
02df55d2 | 842 | rcu_read_unlock_bh(); |
20d29d7a | 843 | |
f09e2249 | 844 | return ret ? ret : copied; |
20d29d7a AB |
845 | } |
846 | ||
501c774c AB |
847 | static ssize_t macvtap_do_read(struct macvtap_queue *q, struct kiocb *iocb, |
848 | const struct iovec *iv, unsigned long len, | |
849 | int noblock) | |
20d29d7a | 850 | { |
ccf7e72b | 851 | DEFINE_WAIT(wait); |
20d29d7a | 852 | struct sk_buff *skb; |
501c774c | 853 | ssize_t ret = 0; |
20d29d7a | 854 | |
20d29d7a | 855 | while (len) { |
ccf7e72b | 856 | prepare_to_wait(sk_sleep(&q->sk), &wait, TASK_INTERRUPTIBLE); |
20d29d7a AB |
857 | |
858 | /* Read frames from the queue */ | |
859 | skb = skb_dequeue(&q->sk.sk_receive_queue); | |
860 | if (!skb) { | |
501c774c | 861 | if (noblock) { |
20d29d7a AB |
862 | ret = -EAGAIN; |
863 | break; | |
864 | } | |
865 | if (signal_pending(current)) { | |
866 | ret = -ERESTARTSYS; | |
867 | break; | |
868 | } | |
869 | /* Nothing to read, let's sleep */ | |
870 | schedule(); | |
871 | continue; | |
872 | } | |
873 | ret = macvtap_put_user(q, skb, iv, len); | |
874 | kfree_skb(skb); | |
875 | break; | |
876 | } | |
877 | ||
ccf7e72b | 878 | finish_wait(sk_sleep(&q->sk), &wait); |
501c774c AB |
879 | return ret; |
880 | } | |
881 | ||
882 | static ssize_t macvtap_aio_read(struct kiocb *iocb, const struct iovec *iv, | |
883 | unsigned long count, loff_t pos) | |
884 | { | |
885 | struct file *file = iocb->ki_filp; | |
886 | struct macvtap_queue *q = file->private_data; | |
887 | ssize_t len, ret = 0; | |
20d29d7a | 888 | |
501c774c AB |
889 | len = iov_length(iv, count); |
890 | if (len < 0) { | |
891 | ret = -EINVAL; | |
892 | goto out; | |
893 | } | |
894 | ||
895 | ret = macvtap_do_read(q, iocb, iv, len, file->f_flags & O_NONBLOCK); | |
896 | ret = min_t(ssize_t, ret, len); /* XXX copied from tun.c. Why? */ | |
20d29d7a | 897 | out: |
20d29d7a AB |
898 | return ret; |
899 | } | |
900 | ||
901 | /* | |
902 | * provide compatibility with generic tun/tap interface | |
903 | */ | |
904 | static long macvtap_ioctl(struct file *file, unsigned int cmd, | |
905 | unsigned long arg) | |
906 | { | |
02df55d2 AB |
907 | struct macvtap_queue *q = file->private_data; |
908 | struct macvlan_dev *vlan; | |
20d29d7a AB |
909 | void __user *argp = (void __user *)arg; |
910 | struct ifreq __user *ifr = argp; | |
911 | unsigned int __user *up = argp; | |
912 | unsigned int u; | |
55afbd08 MT |
913 | int __user *sp = argp; |
914 | int s; | |
02df55d2 | 915 | int ret; |
20d29d7a AB |
916 | |
917 | switch (cmd) { | |
918 | case TUNSETIFF: | |
919 | /* ignore the name, just look at flags */ | |
920 | if (get_user(u, &ifr->ifr_flags)) | |
921 | return -EFAULT; | |
b9fb9ee0 AB |
922 | |
923 | ret = 0; | |
924 | if ((u & ~IFF_VNET_HDR) != (IFF_NO_PI | IFF_TAP)) | |
925 | ret = -EINVAL; | |
926 | else | |
927 | q->flags = u; | |
928 | ||
929 | return ret; | |
20d29d7a AB |
930 | |
931 | case TUNGETIFF: | |
02df55d2 | 932 | rcu_read_lock_bh(); |
13707f9e | 933 | vlan = rcu_dereference_bh(q->vlan); |
02df55d2 AB |
934 | if (vlan) |
935 | dev_hold(vlan->dev); | |
936 | rcu_read_unlock_bh(); | |
937 | ||
938 | if (!vlan) | |
20d29d7a | 939 | return -ENOLINK; |
20d29d7a | 940 | |
02df55d2 | 941 | ret = 0; |
13707f9e | 942 | if (copy_to_user(&ifr->ifr_name, vlan->dev->name, IFNAMSIZ) || |
b9fb9ee0 | 943 | put_user(q->flags, &ifr->ifr_flags)) |
02df55d2 AB |
944 | ret = -EFAULT; |
945 | dev_put(vlan->dev); | |
946 | return ret; | |
20d29d7a AB |
947 | |
948 | case TUNGETFEATURES: | |
b9fb9ee0 | 949 | if (put_user(IFF_TAP | IFF_NO_PI | IFF_VNET_HDR, up)) |
20d29d7a AB |
950 | return -EFAULT; |
951 | return 0; | |
952 | ||
953 | case TUNSETSNDBUF: | |
954 | if (get_user(u, up)) | |
955 | return -EFAULT; | |
956 | ||
20d29d7a | 957 | q->sk.sk_sndbuf = u; |
20d29d7a AB |
958 | return 0; |
959 | ||
55afbd08 MT |
960 | case TUNGETVNETHDRSZ: |
961 | s = q->vnet_hdr_sz; | |
962 | if (put_user(s, sp)) | |
963 | return -EFAULT; | |
964 | return 0; | |
965 | ||
966 | case TUNSETVNETHDRSZ: | |
967 | if (get_user(s, sp)) | |
968 | return -EFAULT; | |
969 | if (s < (int)sizeof(struct virtio_net_hdr)) | |
970 | return -EINVAL; | |
971 | ||
972 | q->vnet_hdr_sz = s; | |
973 | return 0; | |
974 | ||
20d29d7a AB |
975 | case TUNSETOFFLOAD: |
976 | /* let the user check for future flags */ | |
977 | if (arg & ~(TUN_F_CSUM | TUN_F_TSO4 | TUN_F_TSO6 | | |
b9fb9ee0 | 978 | TUN_F_TSO_ECN | TUN_F_UFO)) |
20d29d7a AB |
979 | return -EINVAL; |
980 | ||
b9fb9ee0 AB |
981 | /* TODO: only accept frames with the features that |
982 | got enabled for forwarded frames */ | |
983 | if (!(q->flags & IFF_VNET_HDR)) | |
984 | return -EINVAL; | |
20d29d7a AB |
985 | return 0; |
986 | ||
987 | default: | |
988 | return -EINVAL; | |
989 | } | |
990 | } | |
991 | ||
992 | #ifdef CONFIG_COMPAT | |
993 | static long macvtap_compat_ioctl(struct file *file, unsigned int cmd, | |
994 | unsigned long arg) | |
995 | { | |
996 | return macvtap_ioctl(file, cmd, (unsigned long)compat_ptr(arg)); | |
997 | } | |
998 | #endif | |
999 | ||
1000 | static const struct file_operations macvtap_fops = { | |
1001 | .owner = THIS_MODULE, | |
1002 | .open = macvtap_open, | |
1003 | .release = macvtap_release, | |
1004 | .aio_read = macvtap_aio_read, | |
1005 | .aio_write = macvtap_aio_write, | |
1006 | .poll = macvtap_poll, | |
1007 | .llseek = no_llseek, | |
1008 | .unlocked_ioctl = macvtap_ioctl, | |
1009 | #ifdef CONFIG_COMPAT | |
1010 | .compat_ioctl = macvtap_compat_ioctl, | |
1011 | #endif | |
1012 | }; | |
1013 | ||
501c774c AB |
1014 | static int macvtap_sendmsg(struct kiocb *iocb, struct socket *sock, |
1015 | struct msghdr *m, size_t total_len) | |
1016 | { | |
1017 | struct macvtap_queue *q = container_of(sock, struct macvtap_queue, sock); | |
97bc3633 | 1018 | return macvtap_get_user(q, m, m->msg_iov, total_len, m->msg_iovlen, |
501c774c AB |
1019 | m->msg_flags & MSG_DONTWAIT); |
1020 | } | |
1021 | ||
1022 | static int macvtap_recvmsg(struct kiocb *iocb, struct socket *sock, | |
1023 | struct msghdr *m, size_t total_len, | |
1024 | int flags) | |
1025 | { | |
1026 | struct macvtap_queue *q = container_of(sock, struct macvtap_queue, sock); | |
1027 | int ret; | |
1028 | if (flags & ~(MSG_DONTWAIT|MSG_TRUNC)) | |
1029 | return -EINVAL; | |
1030 | ret = macvtap_do_read(q, iocb, m->msg_iov, total_len, | |
1031 | flags & MSG_DONTWAIT); | |
1032 | if (ret > total_len) { | |
1033 | m->msg_flags |= MSG_TRUNC; | |
1034 | ret = flags & MSG_TRUNC ? ret : total_len; | |
1035 | } | |
1036 | return ret; | |
1037 | } | |
1038 | ||
1039 | /* Ops structure to mimic raw sockets with tun */ | |
1040 | static const struct proto_ops macvtap_socket_ops = { | |
1041 | .sendmsg = macvtap_sendmsg, | |
1042 | .recvmsg = macvtap_recvmsg, | |
1043 | }; | |
1044 | ||
1045 | /* Get an underlying socket object from tun file. Returns error unless file is | |
1046 | * attached to a device. The returned object works like a packet socket, it | |
1047 | * can be used for sock_sendmsg/sock_recvmsg. The caller is responsible for | |
1048 | * holding a reference to the file for as long as the socket is in use. */ | |
1049 | struct socket *macvtap_get_socket(struct file *file) | |
1050 | { | |
1051 | struct macvtap_queue *q; | |
1052 | if (file->f_op != &macvtap_fops) | |
1053 | return ERR_PTR(-EINVAL); | |
1054 | q = file->private_data; | |
1055 | if (!q) | |
1056 | return ERR_PTR(-EBADFD); | |
1057 | return &q->sock; | |
1058 | } | |
1059 | EXPORT_SYMBOL_GPL(macvtap_get_socket); | |
1060 | ||
9bf1907f EB |
1061 | static int macvtap_device_event(struct notifier_block *unused, |
1062 | unsigned long event, void *ptr) | |
1063 | { | |
1064 | struct net_device *dev = ptr; | |
e09eff7f | 1065 | struct macvlan_dev *vlan; |
9bf1907f EB |
1066 | struct device *classdev; |
1067 | dev_t devt; | |
e09eff7f | 1068 | int err; |
9bf1907f EB |
1069 | |
1070 | if (dev->rtnl_link_ops != &macvtap_link_ops) | |
1071 | return NOTIFY_DONE; | |
1072 | ||
e09eff7f | 1073 | vlan = netdev_priv(dev); |
9bf1907f EB |
1074 | |
1075 | switch (event) { | |
1076 | case NETDEV_REGISTER: | |
1077 | /* Create the device node here after the network device has | |
1078 | * been registered but before register_netdevice has | |
1079 | * finished running. | |
1080 | */ | |
e09eff7f EB |
1081 | err = macvtap_get_minor(vlan); |
1082 | if (err) | |
1083 | return notifier_from_errno(err); | |
1084 | ||
1085 | devt = MKDEV(MAJOR(macvtap_major), vlan->minor); | |
9bf1907f EB |
1086 | classdev = device_create(macvtap_class, &dev->dev, devt, |
1087 | dev, "tap%d", dev->ifindex); | |
e09eff7f EB |
1088 | if (IS_ERR(classdev)) { |
1089 | macvtap_free_minor(vlan); | |
9bf1907f | 1090 | return notifier_from_errno(PTR_ERR(classdev)); |
e09eff7f | 1091 | } |
9bf1907f EB |
1092 | break; |
1093 | case NETDEV_UNREGISTER: | |
e09eff7f | 1094 | devt = MKDEV(MAJOR(macvtap_major), vlan->minor); |
9bf1907f | 1095 | device_destroy(macvtap_class, devt); |
e09eff7f | 1096 | macvtap_free_minor(vlan); |
9bf1907f EB |
1097 | break; |
1098 | } | |
1099 | ||
1100 | return NOTIFY_DONE; | |
1101 | } | |
1102 | ||
1103 | static struct notifier_block macvtap_notifier_block __read_mostly = { | |
1104 | .notifier_call = macvtap_device_event, | |
1105 | }; | |
1106 | ||
20d29d7a AB |
1107 | static int macvtap_init(void) |
1108 | { | |
1109 | int err; | |
1110 | ||
1111 | err = alloc_chrdev_region(&macvtap_major, 0, | |
1112 | MACVTAP_NUM_DEVS, "macvtap"); | |
1113 | if (err) | |
1114 | goto out1; | |
1115 | ||
1116 | cdev_init(&macvtap_cdev, &macvtap_fops); | |
1117 | err = cdev_add(&macvtap_cdev, macvtap_major, MACVTAP_NUM_DEVS); | |
1118 | if (err) | |
1119 | goto out2; | |
1120 | ||
1121 | macvtap_class = class_create(THIS_MODULE, "macvtap"); | |
1122 | if (IS_ERR(macvtap_class)) { | |
1123 | err = PTR_ERR(macvtap_class); | |
1124 | goto out3; | |
1125 | } | |
1126 | ||
9bf1907f | 1127 | err = register_netdevice_notifier(&macvtap_notifier_block); |
20d29d7a AB |
1128 | if (err) |
1129 | goto out4; | |
1130 | ||
9bf1907f EB |
1131 | err = macvlan_link_register(&macvtap_link_ops); |
1132 | if (err) | |
1133 | goto out5; | |
1134 | ||
20d29d7a AB |
1135 | return 0; |
1136 | ||
9bf1907f EB |
1137 | out5: |
1138 | unregister_netdevice_notifier(&macvtap_notifier_block); | |
20d29d7a AB |
1139 | out4: |
1140 | class_unregister(macvtap_class); | |
1141 | out3: | |
1142 | cdev_del(&macvtap_cdev); | |
1143 | out2: | |
1144 | unregister_chrdev_region(macvtap_major, MACVTAP_NUM_DEVS); | |
1145 | out1: | |
1146 | return err; | |
1147 | } | |
1148 | module_init(macvtap_init); | |
1149 | ||
1150 | static void macvtap_exit(void) | |
1151 | { | |
1152 | rtnl_link_unregister(&macvtap_link_ops); | |
9bf1907f | 1153 | unregister_netdevice_notifier(&macvtap_notifier_block); |
20d29d7a AB |
1154 | class_unregister(macvtap_class); |
1155 | cdev_del(&macvtap_cdev); | |
1156 | unregister_chrdev_region(macvtap_major, MACVTAP_NUM_DEVS); | |
1157 | } | |
1158 | module_exit(macvtap_exit); | |
1159 | ||
1160 | MODULE_ALIAS_RTNL_LINK("macvtap"); | |
1161 | MODULE_AUTHOR("Arnd Bergmann <arnd@arndb.de>"); | |
1162 | MODULE_LICENSE("GPL"); |