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; | |
e09eff7f EB |
282 | |
283 | mutex_lock(&minor_lock); | |
ec09ebc1 TH |
284 | retval = idr_alloc(&minor_idr, vlan, 1, MACVTAP_NUM_DEVS, GFP_KERNEL); |
285 | if (retval >= 0) { | |
286 | vlan->minor = retval; | |
287 | } else if (retval == -ENOSPC) { | |
e09eff7f EB |
288 | printk(KERN_ERR "too many macvtap devices\n"); |
289 | retval = -EINVAL; | |
e09eff7f | 290 | } |
e09eff7f | 291 | mutex_unlock(&minor_lock); |
ec09ebc1 | 292 | return retval < 0 ? retval : 0; |
e09eff7f EB |
293 | } |
294 | ||
295 | static void macvtap_free_minor(struct macvlan_dev *vlan) | |
296 | { | |
297 | mutex_lock(&minor_lock); | |
298 | if (vlan->minor) { | |
299 | idr_remove(&minor_idr, vlan->minor); | |
300 | vlan->minor = 0; | |
301 | } | |
302 | mutex_unlock(&minor_lock); | |
303 | } | |
304 | ||
305 | static struct net_device *dev_get_by_macvtap_minor(int minor) | |
306 | { | |
307 | struct net_device *dev = NULL; | |
308 | struct macvlan_dev *vlan; | |
309 | ||
310 | mutex_lock(&minor_lock); | |
311 | vlan = idr_find(&minor_idr, minor); | |
312 | if (vlan) { | |
313 | dev = vlan->dev; | |
314 | dev_hold(dev); | |
315 | } | |
316 | mutex_unlock(&minor_lock); | |
317 | return dev; | |
318 | } | |
319 | ||
20d29d7a AB |
320 | static int macvtap_newlink(struct net *src_net, |
321 | struct net_device *dev, | |
322 | struct nlattr *tb[], | |
323 | struct nlattr *data[]) | |
324 | { | |
9bf1907f EB |
325 | /* Don't put anything that may fail after macvlan_common_newlink |
326 | * because we can't undo what it does. | |
327 | */ | |
328 | return macvlan_common_newlink(src_net, dev, tb, data, | |
329 | macvtap_receive, macvtap_forward); | |
20d29d7a AB |
330 | } |
331 | ||
332 | static void macvtap_dellink(struct net_device *dev, | |
333 | struct list_head *head) | |
334 | { | |
20d29d7a AB |
335 | macvtap_del_queues(dev); |
336 | macvlan_dellink(dev, head); | |
337 | } | |
338 | ||
8a35747a HX |
339 | static void macvtap_setup(struct net_device *dev) |
340 | { | |
341 | macvlan_common_setup(dev); | |
342 | dev->tx_queue_len = TUN_READQ_SIZE; | |
343 | } | |
344 | ||
20d29d7a AB |
345 | static struct rtnl_link_ops macvtap_link_ops __read_mostly = { |
346 | .kind = "macvtap", | |
8a35747a | 347 | .setup = macvtap_setup, |
20d29d7a AB |
348 | .newlink = macvtap_newlink, |
349 | .dellink = macvtap_dellink, | |
350 | }; | |
351 | ||
352 | ||
353 | static void macvtap_sock_write_space(struct sock *sk) | |
354 | { | |
43815482 ED |
355 | wait_queue_head_t *wqueue; |
356 | ||
20d29d7a AB |
357 | if (!sock_writeable(sk) || |
358 | !test_and_clear_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags)) | |
359 | return; | |
360 | ||
43815482 ED |
361 | wqueue = sk_sleep(sk); |
362 | if (wqueue && waitqueue_active(wqueue)) | |
363 | wake_up_interruptible_poll(wqueue, POLLOUT | POLLWRNORM | POLLWRBAND); | |
20d29d7a AB |
364 | } |
365 | ||
2259fef0 EB |
366 | static void macvtap_sock_destruct(struct sock *sk) |
367 | { | |
368 | skb_queue_purge(&sk->sk_receive_queue); | |
369 | } | |
370 | ||
20d29d7a AB |
371 | static int macvtap_open(struct inode *inode, struct file *file) |
372 | { | |
373 | struct net *net = current->nsproxy->net_ns; | |
e09eff7f | 374 | struct net_device *dev = dev_get_by_macvtap_minor(iminor(inode)); |
20d29d7a AB |
375 | struct macvtap_queue *q; |
376 | int err; | |
377 | ||
378 | err = -ENODEV; | |
379 | if (!dev) | |
380 | goto out; | |
381 | ||
20d29d7a AB |
382 | err = -ENOMEM; |
383 | q = (struct macvtap_queue *)sk_alloc(net, AF_UNSPEC, GFP_KERNEL, | |
384 | &macvtap_proto); | |
385 | if (!q) | |
386 | goto out; | |
387 | ||
43815482 ED |
388 | q->sock.wq = &q->wq; |
389 | init_waitqueue_head(&q->wq.wait); | |
20d29d7a AB |
390 | q->sock.type = SOCK_RAW; |
391 | q->sock.state = SS_CONNECTED; | |
501c774c AB |
392 | q->sock.file = file; |
393 | q->sock.ops = &macvtap_socket_ops; | |
20d29d7a | 394 | sock_init_data(&q->sock, &q->sk); |
20d29d7a | 395 | q->sk.sk_write_space = macvtap_sock_write_space; |
2259fef0 | 396 | q->sk.sk_destruct = macvtap_sock_destruct; |
b9fb9ee0 | 397 | q->flags = IFF_VNET_HDR | IFF_NO_PI | IFF_TAP; |
55afbd08 | 398 | q->vnet_hdr_sz = sizeof(struct virtio_net_hdr); |
20d29d7a | 399 | |
97bc3633 SM |
400 | /* |
401 | * so far only KVM virtio_net uses macvtap, enable zero copy between | |
402 | * guest kernel and host kernel when lower device supports zerocopy | |
047af9cf EB |
403 | * |
404 | * The macvlan supports zerocopy iff the lower device supports zero | |
405 | * copy so we don't have to look at the lower device directly. | |
97bc3633 | 406 | */ |
047af9cf EB |
407 | if ((dev->features & NETIF_F_HIGHDMA) && (dev->features & NETIF_F_SG)) |
408 | sock_set_flag(&q->sk, SOCK_ZEROCOPY); | |
97bc3633 | 409 | |
20d29d7a AB |
410 | err = macvtap_set_queue(dev, file, q); |
411 | if (err) | |
412 | sock_put(&q->sk); | |
413 | ||
414 | out: | |
415 | if (dev) | |
416 | dev_put(dev); | |
417 | ||
418 | return err; | |
419 | } | |
420 | ||
421 | static int macvtap_release(struct inode *inode, struct file *file) | |
422 | { | |
02df55d2 AB |
423 | struct macvtap_queue *q = file->private_data; |
424 | macvtap_put_queue(q); | |
20d29d7a AB |
425 | return 0; |
426 | } | |
427 | ||
428 | static unsigned int macvtap_poll(struct file *file, poll_table * wait) | |
429 | { | |
02df55d2 | 430 | struct macvtap_queue *q = file->private_data; |
20d29d7a AB |
431 | unsigned int mask = POLLERR; |
432 | ||
433 | if (!q) | |
434 | goto out; | |
435 | ||
436 | mask = 0; | |
43815482 | 437 | poll_wait(file, &q->wq.wait, wait); |
20d29d7a AB |
438 | |
439 | if (!skb_queue_empty(&q->sk.sk_receive_queue)) | |
440 | mask |= POLLIN | POLLRDNORM; | |
441 | ||
442 | if (sock_writeable(&q->sk) || | |
443 | (!test_and_set_bit(SOCK_ASYNC_NOSPACE, &q->sock.flags) && | |
444 | sock_writeable(&q->sk))) | |
445 | mask |= POLLOUT | POLLWRNORM; | |
446 | ||
447 | out: | |
20d29d7a AB |
448 | return mask; |
449 | } | |
450 | ||
b9fb9ee0 AB |
451 | static inline struct sk_buff *macvtap_alloc_skb(struct sock *sk, size_t prepad, |
452 | size_t len, size_t linear, | |
453 | int noblock, int *err) | |
454 | { | |
455 | struct sk_buff *skb; | |
456 | ||
457 | /* Under a page? Don't bother with paged skb. */ | |
458 | if (prepad + len < PAGE_SIZE || !linear) | |
459 | linear = len; | |
460 | ||
461 | skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock, | |
462 | err); | |
463 | if (!skb) | |
464 | return NULL; | |
465 | ||
466 | skb_reserve(skb, prepad); | |
467 | skb_put(skb, linear); | |
468 | skb->data_len = len - linear; | |
469 | skb->len += len - linear; | |
470 | ||
471 | return skb; | |
472 | } | |
473 | ||
97bc3633 SM |
474 | /* set skb frags from iovec, this can move to core network code for reuse */ |
475 | static int zerocopy_sg_from_iovec(struct sk_buff *skb, const struct iovec *from, | |
476 | int offset, size_t count) | |
477 | { | |
478 | int len = iov_length(from, count) - offset; | |
479 | int copy = skb_headlen(skb); | |
480 | int size, offset1 = 0; | |
481 | int i = 0; | |
97bc3633 SM |
482 | |
483 | /* Skip over from offset */ | |
484 | while (count && (offset >= from->iov_len)) { | |
485 | offset -= from->iov_len; | |
486 | ++from; | |
487 | --count; | |
488 | } | |
489 | ||
490 | /* copy up to skb headlen */ | |
491 | while (count && (copy > 0)) { | |
492 | size = min_t(unsigned int, copy, from->iov_len - offset); | |
493 | if (copy_from_user(skb->data + offset1, from->iov_base + offset, | |
494 | size)) | |
495 | return -EFAULT; | |
496 | if (copy > size) { | |
497 | ++from; | |
498 | --count; | |
3afc9621 JW |
499 | offset = 0; |
500 | } else | |
501 | offset += size; | |
97bc3633 SM |
502 | copy -= size; |
503 | offset1 += size; | |
97bc3633 SM |
504 | } |
505 | ||
506 | if (len == offset1) | |
507 | return 0; | |
508 | ||
509 | while (count--) { | |
510 | struct page *page[MAX_SKB_FRAGS]; | |
511 | int num_pages; | |
512 | unsigned long base; | |
4ef67ebe | 513 | unsigned long truesize; |
97bc3633 | 514 | |
3afc9621 | 515 | len = from->iov_len - offset; |
97bc3633 | 516 | if (!len) { |
3afc9621 | 517 | offset = 0; |
97bc3633 SM |
518 | ++from; |
519 | continue; | |
520 | } | |
3afc9621 | 521 | base = (unsigned long)from->iov_base + offset; |
97bc3633 | 522 | size = ((base & ~PAGE_MASK) + len + ~PAGE_MASK) >> PAGE_SHIFT; |
b92946e2 JW |
523 | if (i + size > MAX_SKB_FRAGS) |
524 | return -EMSGSIZE; | |
97bc3633 | 525 | num_pages = get_user_pages_fast(base, size, 0, &page[i]); |
b92946e2 | 526 | if (num_pages != size) { |
02ce04bb JW |
527 | for (i = 0; i < num_pages; i++) |
528 | put_page(page[i]); | |
97bc3633 | 529 | return -EFAULT; |
02ce04bb | 530 | } |
4ef67ebe | 531 | truesize = size * PAGE_SIZE; |
97bc3633 SM |
532 | skb->data_len += len; |
533 | skb->len += len; | |
4ef67ebe JW |
534 | skb->truesize += truesize; |
535 | atomic_add(truesize, &skb->sk->sk_wmem_alloc); | |
97bc3633 | 536 | while (len) { |
653fc915 JW |
537 | int off = base & ~PAGE_MASK; |
538 | int size = min_t(int, len, PAGE_SIZE - off); | |
539 | __skb_fill_page_desc(skb, i, page[i], off, size); | |
97bc3633 SM |
540 | skb_shinfo(skb)->nr_frags++; |
541 | /* increase sk_wmem_alloc */ | |
653fc915 JW |
542 | base += size; |
543 | len -= size; | |
97bc3633 SM |
544 | i++; |
545 | } | |
3afc9621 | 546 | offset = 0; |
97bc3633 SM |
547 | ++from; |
548 | } | |
549 | return 0; | |
550 | } | |
551 | ||
b9fb9ee0 AB |
552 | /* |
553 | * macvtap_skb_from_vnet_hdr and macvtap_skb_to_vnet_hdr should | |
554 | * be shared with the tun/tap driver. | |
555 | */ | |
556 | static int macvtap_skb_from_vnet_hdr(struct sk_buff *skb, | |
557 | struct virtio_net_hdr *vnet_hdr) | |
558 | { | |
559 | unsigned short gso_type = 0; | |
560 | if (vnet_hdr->gso_type != VIRTIO_NET_HDR_GSO_NONE) { | |
561 | switch (vnet_hdr->gso_type & ~VIRTIO_NET_HDR_GSO_ECN) { | |
562 | case VIRTIO_NET_HDR_GSO_TCPV4: | |
563 | gso_type = SKB_GSO_TCPV4; | |
564 | break; | |
565 | case VIRTIO_NET_HDR_GSO_TCPV6: | |
566 | gso_type = SKB_GSO_TCPV6; | |
567 | break; | |
568 | case VIRTIO_NET_HDR_GSO_UDP: | |
569 | gso_type = SKB_GSO_UDP; | |
570 | break; | |
571 | default: | |
572 | return -EINVAL; | |
573 | } | |
574 | ||
575 | if (vnet_hdr->gso_type & VIRTIO_NET_HDR_GSO_ECN) | |
576 | gso_type |= SKB_GSO_TCP_ECN; | |
577 | ||
578 | if (vnet_hdr->gso_size == 0) | |
579 | return -EINVAL; | |
580 | } | |
581 | ||
582 | if (vnet_hdr->flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) { | |
583 | if (!skb_partial_csum_set(skb, vnet_hdr->csum_start, | |
584 | vnet_hdr->csum_offset)) | |
585 | return -EINVAL; | |
586 | } | |
587 | ||
588 | if (vnet_hdr->gso_type != VIRTIO_NET_HDR_GSO_NONE) { | |
589 | skb_shinfo(skb)->gso_size = vnet_hdr->gso_size; | |
c9af6db4 | 590 | skb_shinfo(skb)->gso_type = gso_type; |
b9fb9ee0 AB |
591 | |
592 | /* Header must be checked, and gso_segs computed. */ | |
593 | skb_shinfo(skb)->gso_type |= SKB_GSO_DODGY; | |
594 | skb_shinfo(skb)->gso_segs = 0; | |
595 | } | |
596 | return 0; | |
597 | } | |
598 | ||
599 | static int macvtap_skb_to_vnet_hdr(const struct sk_buff *skb, | |
600 | struct virtio_net_hdr *vnet_hdr) | |
601 | { | |
602 | memset(vnet_hdr, 0, sizeof(*vnet_hdr)); | |
603 | ||
604 | if (skb_is_gso(skb)) { | |
605 | struct skb_shared_info *sinfo = skb_shinfo(skb); | |
606 | ||
607 | /* This is a hint as to how much should be linear. */ | |
608 | vnet_hdr->hdr_len = skb_headlen(skb); | |
609 | vnet_hdr->gso_size = sinfo->gso_size; | |
610 | if (sinfo->gso_type & SKB_GSO_TCPV4) | |
611 | vnet_hdr->gso_type = VIRTIO_NET_HDR_GSO_TCPV4; | |
612 | else if (sinfo->gso_type & SKB_GSO_TCPV6) | |
613 | vnet_hdr->gso_type = VIRTIO_NET_HDR_GSO_TCPV6; | |
614 | else if (sinfo->gso_type & SKB_GSO_UDP) | |
615 | vnet_hdr->gso_type = VIRTIO_NET_HDR_GSO_UDP; | |
616 | else | |
617 | BUG(); | |
618 | if (sinfo->gso_type & SKB_GSO_TCP_ECN) | |
619 | vnet_hdr->gso_type |= VIRTIO_NET_HDR_GSO_ECN; | |
620 | } else | |
621 | vnet_hdr->gso_type = VIRTIO_NET_HDR_GSO_NONE; | |
622 | ||
623 | if (skb->ip_summed == CHECKSUM_PARTIAL) { | |
624 | vnet_hdr->flags = VIRTIO_NET_HDR_F_NEEDS_CSUM; | |
55508d60 | 625 | vnet_hdr->csum_start = skb_checksum_start_offset(skb); |
b9fb9ee0 | 626 | vnet_hdr->csum_offset = skb->csum_offset; |
10a8d94a JW |
627 | } else if (skb->ip_summed == CHECKSUM_UNNECESSARY) { |
628 | vnet_hdr->flags = VIRTIO_NET_HDR_F_DATA_VALID; | |
b9fb9ee0 AB |
629 | } /* else everything is zero */ |
630 | ||
631 | return 0; | |
632 | } | |
633 | ||
634 | ||
20d29d7a | 635 | /* Get packet from user space buffer */ |
97bc3633 SM |
636 | static ssize_t macvtap_get_user(struct macvtap_queue *q, struct msghdr *m, |
637 | const struct iovec *iv, unsigned long total_len, | |
638 | size_t count, int noblock) | |
20d29d7a AB |
639 | { |
640 | struct sk_buff *skb; | |
02df55d2 | 641 | struct macvlan_dev *vlan; |
97bc3633 | 642 | unsigned long len = total_len; |
20d29d7a | 643 | int err; |
b9fb9ee0 AB |
644 | struct virtio_net_hdr vnet_hdr = { 0 }; |
645 | int vnet_hdr_len = 0; | |
b92946e2 | 646 | int copylen = 0; |
97bc3633 | 647 | bool zerocopy = false; |
b9fb9ee0 AB |
648 | |
649 | if (q->flags & IFF_VNET_HDR) { | |
55afbd08 | 650 | vnet_hdr_len = q->vnet_hdr_sz; |
b9fb9ee0 AB |
651 | |
652 | err = -EINVAL; | |
ce3c8692 | 653 | if (len < vnet_hdr_len) |
b9fb9ee0 | 654 | goto err; |
ce3c8692 | 655 | len -= vnet_hdr_len; |
b9fb9ee0 AB |
656 | |
657 | err = memcpy_fromiovecend((void *)&vnet_hdr, iv, 0, | |
55afbd08 | 658 | sizeof(vnet_hdr)); |
b9fb9ee0 AB |
659 | if (err < 0) |
660 | goto err; | |
661 | if ((vnet_hdr.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) && | |
662 | vnet_hdr.csum_start + vnet_hdr.csum_offset + 2 > | |
663 | vnet_hdr.hdr_len) | |
664 | vnet_hdr.hdr_len = vnet_hdr.csum_start + | |
665 | vnet_hdr.csum_offset + 2; | |
666 | err = -EINVAL; | |
667 | if (vnet_hdr.hdr_len > len) | |
668 | goto err; | |
669 | } | |
20d29d7a | 670 | |
b9fb9ee0 | 671 | err = -EINVAL; |
20d29d7a | 672 | if (unlikely(len < ETH_HLEN)) |
b9fb9ee0 | 673 | goto err; |
20d29d7a | 674 | |
b92946e2 JW |
675 | err = -EMSGSIZE; |
676 | if (unlikely(count > UIO_MAXIOV)) | |
677 | goto err; | |
678 | ||
97bc3633 SM |
679 | if (m && m->msg_control && sock_flag(&q->sk, SOCK_ZEROCOPY)) |
680 | zerocopy = true; | |
681 | ||
682 | if (zerocopy) { | |
b92946e2 JW |
683 | /* Userspace may produce vectors with count greater than |
684 | * MAX_SKB_FRAGS, so we need to linearize parts of the skb | |
685 | * to let the rest of data to be fit in the frags. | |
686 | */ | |
687 | if (count > MAX_SKB_FRAGS) { | |
688 | copylen = iov_length(iv, count - MAX_SKB_FRAGS); | |
689 | if (copylen < vnet_hdr_len) | |
690 | copylen = 0; | |
691 | else | |
692 | copylen -= vnet_hdr_len; | |
693 | } | |
97bc3633 SM |
694 | /* There are 256 bytes to be copied in skb, so there is enough |
695 | * room for skb expand head in case it is used. | |
696 | * The rest buffer is mapped from userspace. | |
697 | */ | |
b92946e2 JW |
698 | if (copylen < vnet_hdr.hdr_len) |
699 | copylen = vnet_hdr.hdr_len; | |
97bc3633 SM |
700 | if (!copylen) |
701 | copylen = GOODCOPY_LEN; | |
702 | } else | |
703 | copylen = len; | |
704 | ||
705 | skb = macvtap_alloc_skb(&q->sk, NET_IP_ALIGN, copylen, | |
706 | vnet_hdr.hdr_len, noblock, &err); | |
02df55d2 AB |
707 | if (!skb) |
708 | goto err; | |
20d29d7a | 709 | |
01d6657b | 710 | if (zerocopy) |
97bc3633 | 711 | err = zerocopy_sg_from_iovec(skb, iv, vnet_hdr_len, count); |
01d6657b | 712 | else |
97bc3633 SM |
713 | err = skb_copy_datagram_from_iovec(skb, 0, iv, vnet_hdr_len, |
714 | len); | |
02df55d2 | 715 | if (err) |
b9fb9ee0 | 716 | goto err_kfree; |
20d29d7a AB |
717 | |
718 | skb_set_network_header(skb, ETH_HLEN); | |
b9fb9ee0 AB |
719 | skb_reset_mac_header(skb); |
720 | skb->protocol = eth_hdr(skb)->h_proto; | |
721 | ||
722 | if (vnet_hdr_len) { | |
723 | err = macvtap_skb_from_vnet_hdr(skb, &vnet_hdr); | |
724 | if (err) | |
725 | goto err_kfree; | |
726 | } | |
727 | ||
40893fd0 | 728 | skb_probe_transport_header(skb, ETH_HLEN); |
9b4d669b | 729 | |
02df55d2 | 730 | rcu_read_lock_bh(); |
13707f9e | 731 | vlan = rcu_dereference_bh(q->vlan); |
97bc3633 | 732 | /* copy skb_ubuf_info for callback when skb has no error */ |
01d6657b | 733 | if (zerocopy) { |
97bc3633 | 734 | skb_shinfo(skb)->destructor_arg = m->msg_control; |
01d6657b | 735 | skb_shinfo(skb)->tx_flags |= SKBTX_DEV_ZEROCOPY; |
c9af6db4 | 736 | skb_shinfo(skb)->tx_flags |= SKBTX_SHARED_FRAG; |
01d6657b | 737 | } |
02df55d2 AB |
738 | if (vlan) |
739 | macvlan_start_xmit(skb, vlan->dev); | |
740 | else | |
741 | kfree_skb(skb); | |
742 | rcu_read_unlock_bh(); | |
20d29d7a | 743 | |
97bc3633 | 744 | return total_len; |
02df55d2 | 745 | |
b9fb9ee0 AB |
746 | err_kfree: |
747 | kfree_skb(skb); | |
748 | ||
02df55d2 AB |
749 | err: |
750 | rcu_read_lock_bh(); | |
13707f9e | 751 | vlan = rcu_dereference_bh(q->vlan); |
02df55d2 | 752 | if (vlan) |
1ac9ad13 | 753 | vlan->dev->stats.tx_dropped++; |
02df55d2 AB |
754 | rcu_read_unlock_bh(); |
755 | ||
02df55d2 | 756 | return err; |
20d29d7a AB |
757 | } |
758 | ||
759 | static ssize_t macvtap_aio_write(struct kiocb *iocb, const struct iovec *iv, | |
760 | unsigned long count, loff_t pos) | |
761 | { | |
762 | struct file *file = iocb->ki_filp; | |
763 | ssize_t result = -ENOLINK; | |
02df55d2 | 764 | struct macvtap_queue *q = file->private_data; |
20d29d7a | 765 | |
97bc3633 SM |
766 | result = macvtap_get_user(q, NULL, iv, iov_length(iv, count), count, |
767 | file->f_flags & O_NONBLOCK); | |
20d29d7a AB |
768 | return result; |
769 | } | |
770 | ||
771 | /* Put packet to the user space buffer */ | |
772 | static ssize_t macvtap_put_user(struct macvtap_queue *q, | |
773 | const struct sk_buff *skb, | |
774 | const struct iovec *iv, int len) | |
775 | { | |
02df55d2 | 776 | struct macvlan_dev *vlan; |
20d29d7a | 777 | int ret; |
b9fb9ee0 | 778 | int vnet_hdr_len = 0; |
f09e2249 BG |
779 | int vlan_offset = 0; |
780 | int copied; | |
b9fb9ee0 AB |
781 | |
782 | if (q->flags & IFF_VNET_HDR) { | |
783 | struct virtio_net_hdr vnet_hdr; | |
55afbd08 | 784 | vnet_hdr_len = q->vnet_hdr_sz; |
b9fb9ee0 AB |
785 | if ((len -= vnet_hdr_len) < 0) |
786 | return -EINVAL; | |
787 | ||
788 | ret = macvtap_skb_to_vnet_hdr(skb, &vnet_hdr); | |
789 | if (ret) | |
790 | return ret; | |
791 | ||
55afbd08 | 792 | if (memcpy_toiovecend(iv, (void *)&vnet_hdr, 0, sizeof(vnet_hdr))) |
b9fb9ee0 AB |
793 | return -EFAULT; |
794 | } | |
f09e2249 BG |
795 | copied = vnet_hdr_len; |
796 | ||
797 | if (!vlan_tx_tag_present(skb)) | |
798 | len = min_t(int, skb->len, len); | |
799 | else { | |
800 | int copy; | |
801 | struct { | |
802 | __be16 h_vlan_proto; | |
803 | __be16 h_vlan_TCI; | |
804 | } veth; | |
805 | veth.h_vlan_proto = htons(ETH_P_8021Q); | |
806 | veth.h_vlan_TCI = htons(vlan_tx_tag_get(skb)); | |
807 | ||
808 | vlan_offset = offsetof(struct vlan_ethhdr, h_vlan_proto); | |
809 | len = min_t(int, skb->len + VLAN_HLEN, len); | |
810 | ||
811 | copy = min_t(int, vlan_offset, len); | |
812 | ret = skb_copy_datagram_const_iovec(skb, 0, iv, copied, copy); | |
813 | len -= copy; | |
814 | copied += copy; | |
815 | if (ret || !len) | |
816 | goto done; | |
817 | ||
818 | copy = min_t(int, sizeof(veth), len); | |
819 | ret = memcpy_toiovecend(iv, (void *)&veth, copied, copy); | |
820 | len -= copy; | |
821 | copied += copy; | |
822 | if (ret || !len) | |
823 | goto done; | |
824 | } | |
20d29d7a | 825 | |
f09e2249 BG |
826 | ret = skb_copy_datagram_const_iovec(skb, vlan_offset, iv, copied, len); |
827 | copied += len; | |
20d29d7a | 828 | |
f09e2249 | 829 | done: |
02df55d2 | 830 | rcu_read_lock_bh(); |
13707f9e | 831 | vlan = rcu_dereference_bh(q->vlan); |
501c774c | 832 | if (vlan) |
f09e2249 | 833 | macvlan_count_rx(vlan, copied - vnet_hdr_len, ret == 0, 0); |
02df55d2 | 834 | rcu_read_unlock_bh(); |
20d29d7a | 835 | |
f09e2249 | 836 | return ret ? ret : copied; |
20d29d7a AB |
837 | } |
838 | ||
501c774c AB |
839 | static ssize_t macvtap_do_read(struct macvtap_queue *q, struct kiocb *iocb, |
840 | const struct iovec *iv, unsigned long len, | |
841 | int noblock) | |
20d29d7a | 842 | { |
ccf7e72b | 843 | DEFINE_WAIT(wait); |
20d29d7a | 844 | struct sk_buff *skb; |
501c774c | 845 | ssize_t ret = 0; |
20d29d7a | 846 | |
20d29d7a | 847 | while (len) { |
ccf7e72b | 848 | prepare_to_wait(sk_sleep(&q->sk), &wait, TASK_INTERRUPTIBLE); |
20d29d7a AB |
849 | |
850 | /* Read frames from the queue */ | |
851 | skb = skb_dequeue(&q->sk.sk_receive_queue); | |
852 | if (!skb) { | |
501c774c | 853 | if (noblock) { |
20d29d7a AB |
854 | ret = -EAGAIN; |
855 | break; | |
856 | } | |
857 | if (signal_pending(current)) { | |
858 | ret = -ERESTARTSYS; | |
859 | break; | |
860 | } | |
861 | /* Nothing to read, let's sleep */ | |
862 | schedule(); | |
863 | continue; | |
864 | } | |
865 | ret = macvtap_put_user(q, skb, iv, len); | |
866 | kfree_skb(skb); | |
867 | break; | |
868 | } | |
869 | ||
ccf7e72b | 870 | finish_wait(sk_sleep(&q->sk), &wait); |
501c774c AB |
871 | return ret; |
872 | } | |
873 | ||
874 | static ssize_t macvtap_aio_read(struct kiocb *iocb, const struct iovec *iv, | |
875 | unsigned long count, loff_t pos) | |
876 | { | |
877 | struct file *file = iocb->ki_filp; | |
878 | struct macvtap_queue *q = file->private_data; | |
879 | ssize_t len, ret = 0; | |
20d29d7a | 880 | |
501c774c AB |
881 | len = iov_length(iv, count); |
882 | if (len < 0) { | |
883 | ret = -EINVAL; | |
884 | goto out; | |
885 | } | |
886 | ||
887 | ret = macvtap_do_read(q, iocb, iv, len, file->f_flags & O_NONBLOCK); | |
888 | ret = min_t(ssize_t, ret, len); /* XXX copied from tun.c. Why? */ | |
20d29d7a | 889 | out: |
20d29d7a AB |
890 | return ret; |
891 | } | |
892 | ||
893 | /* | |
894 | * provide compatibility with generic tun/tap interface | |
895 | */ | |
896 | static long macvtap_ioctl(struct file *file, unsigned int cmd, | |
897 | unsigned long arg) | |
898 | { | |
02df55d2 AB |
899 | struct macvtap_queue *q = file->private_data; |
900 | struct macvlan_dev *vlan; | |
20d29d7a AB |
901 | void __user *argp = (void __user *)arg; |
902 | struct ifreq __user *ifr = argp; | |
903 | unsigned int __user *up = argp; | |
904 | unsigned int u; | |
55afbd08 MT |
905 | int __user *sp = argp; |
906 | int s; | |
02df55d2 | 907 | int ret; |
20d29d7a AB |
908 | |
909 | switch (cmd) { | |
910 | case TUNSETIFF: | |
911 | /* ignore the name, just look at flags */ | |
912 | if (get_user(u, &ifr->ifr_flags)) | |
913 | return -EFAULT; | |
b9fb9ee0 AB |
914 | |
915 | ret = 0; | |
916 | if ((u & ~IFF_VNET_HDR) != (IFF_NO_PI | IFF_TAP)) | |
917 | ret = -EINVAL; | |
918 | else | |
919 | q->flags = u; | |
920 | ||
921 | return ret; | |
20d29d7a AB |
922 | |
923 | case TUNGETIFF: | |
02df55d2 | 924 | rcu_read_lock_bh(); |
13707f9e | 925 | vlan = rcu_dereference_bh(q->vlan); |
02df55d2 AB |
926 | if (vlan) |
927 | dev_hold(vlan->dev); | |
928 | rcu_read_unlock_bh(); | |
929 | ||
930 | if (!vlan) | |
20d29d7a | 931 | return -ENOLINK; |
20d29d7a | 932 | |
02df55d2 | 933 | ret = 0; |
13707f9e | 934 | if (copy_to_user(&ifr->ifr_name, vlan->dev->name, IFNAMSIZ) || |
b9fb9ee0 | 935 | put_user(q->flags, &ifr->ifr_flags)) |
02df55d2 AB |
936 | ret = -EFAULT; |
937 | dev_put(vlan->dev); | |
938 | return ret; | |
20d29d7a AB |
939 | |
940 | case TUNGETFEATURES: | |
b9fb9ee0 | 941 | if (put_user(IFF_TAP | IFF_NO_PI | IFF_VNET_HDR, up)) |
20d29d7a AB |
942 | return -EFAULT; |
943 | return 0; | |
944 | ||
945 | case TUNSETSNDBUF: | |
946 | if (get_user(u, up)) | |
947 | return -EFAULT; | |
948 | ||
20d29d7a | 949 | q->sk.sk_sndbuf = u; |
20d29d7a AB |
950 | return 0; |
951 | ||
55afbd08 MT |
952 | case TUNGETVNETHDRSZ: |
953 | s = q->vnet_hdr_sz; | |
954 | if (put_user(s, sp)) | |
955 | return -EFAULT; | |
956 | return 0; | |
957 | ||
958 | case TUNSETVNETHDRSZ: | |
959 | if (get_user(s, sp)) | |
960 | return -EFAULT; | |
961 | if (s < (int)sizeof(struct virtio_net_hdr)) | |
962 | return -EINVAL; | |
963 | ||
964 | q->vnet_hdr_sz = s; | |
965 | return 0; | |
966 | ||
20d29d7a AB |
967 | case TUNSETOFFLOAD: |
968 | /* let the user check for future flags */ | |
969 | if (arg & ~(TUN_F_CSUM | TUN_F_TSO4 | TUN_F_TSO6 | | |
b9fb9ee0 | 970 | TUN_F_TSO_ECN | TUN_F_UFO)) |
20d29d7a AB |
971 | return -EINVAL; |
972 | ||
b9fb9ee0 AB |
973 | /* TODO: only accept frames with the features that |
974 | got enabled for forwarded frames */ | |
975 | if (!(q->flags & IFF_VNET_HDR)) | |
976 | return -EINVAL; | |
20d29d7a AB |
977 | return 0; |
978 | ||
979 | default: | |
980 | return -EINVAL; | |
981 | } | |
982 | } | |
983 | ||
984 | #ifdef CONFIG_COMPAT | |
985 | static long macvtap_compat_ioctl(struct file *file, unsigned int cmd, | |
986 | unsigned long arg) | |
987 | { | |
988 | return macvtap_ioctl(file, cmd, (unsigned long)compat_ptr(arg)); | |
989 | } | |
990 | #endif | |
991 | ||
992 | static const struct file_operations macvtap_fops = { | |
993 | .owner = THIS_MODULE, | |
994 | .open = macvtap_open, | |
995 | .release = macvtap_release, | |
996 | .aio_read = macvtap_aio_read, | |
997 | .aio_write = macvtap_aio_write, | |
998 | .poll = macvtap_poll, | |
999 | .llseek = no_llseek, | |
1000 | .unlocked_ioctl = macvtap_ioctl, | |
1001 | #ifdef CONFIG_COMPAT | |
1002 | .compat_ioctl = macvtap_compat_ioctl, | |
1003 | #endif | |
1004 | }; | |
1005 | ||
501c774c AB |
1006 | static int macvtap_sendmsg(struct kiocb *iocb, struct socket *sock, |
1007 | struct msghdr *m, size_t total_len) | |
1008 | { | |
1009 | struct macvtap_queue *q = container_of(sock, struct macvtap_queue, sock); | |
97bc3633 | 1010 | return macvtap_get_user(q, m, m->msg_iov, total_len, m->msg_iovlen, |
501c774c AB |
1011 | m->msg_flags & MSG_DONTWAIT); |
1012 | } | |
1013 | ||
1014 | static int macvtap_recvmsg(struct kiocb *iocb, struct socket *sock, | |
1015 | struct msghdr *m, size_t total_len, | |
1016 | int flags) | |
1017 | { | |
1018 | struct macvtap_queue *q = container_of(sock, struct macvtap_queue, sock); | |
1019 | int ret; | |
1020 | if (flags & ~(MSG_DONTWAIT|MSG_TRUNC)) | |
1021 | return -EINVAL; | |
1022 | ret = macvtap_do_read(q, iocb, m->msg_iov, total_len, | |
1023 | flags & MSG_DONTWAIT); | |
1024 | if (ret > total_len) { | |
1025 | m->msg_flags |= MSG_TRUNC; | |
1026 | ret = flags & MSG_TRUNC ? ret : total_len; | |
1027 | } | |
1028 | return ret; | |
1029 | } | |
1030 | ||
1031 | /* Ops structure to mimic raw sockets with tun */ | |
1032 | static const struct proto_ops macvtap_socket_ops = { | |
1033 | .sendmsg = macvtap_sendmsg, | |
1034 | .recvmsg = macvtap_recvmsg, | |
1035 | }; | |
1036 | ||
1037 | /* Get an underlying socket object from tun file. Returns error unless file is | |
1038 | * attached to a device. The returned object works like a packet socket, it | |
1039 | * can be used for sock_sendmsg/sock_recvmsg. The caller is responsible for | |
1040 | * holding a reference to the file for as long as the socket is in use. */ | |
1041 | struct socket *macvtap_get_socket(struct file *file) | |
1042 | { | |
1043 | struct macvtap_queue *q; | |
1044 | if (file->f_op != &macvtap_fops) | |
1045 | return ERR_PTR(-EINVAL); | |
1046 | q = file->private_data; | |
1047 | if (!q) | |
1048 | return ERR_PTR(-EBADFD); | |
1049 | return &q->sock; | |
1050 | } | |
1051 | EXPORT_SYMBOL_GPL(macvtap_get_socket); | |
1052 | ||
9bf1907f EB |
1053 | static int macvtap_device_event(struct notifier_block *unused, |
1054 | unsigned long event, void *ptr) | |
1055 | { | |
1056 | struct net_device *dev = ptr; | |
e09eff7f | 1057 | struct macvlan_dev *vlan; |
9bf1907f EB |
1058 | struct device *classdev; |
1059 | dev_t devt; | |
e09eff7f | 1060 | int err; |
9bf1907f EB |
1061 | |
1062 | if (dev->rtnl_link_ops != &macvtap_link_ops) | |
1063 | return NOTIFY_DONE; | |
1064 | ||
e09eff7f | 1065 | vlan = netdev_priv(dev); |
9bf1907f EB |
1066 | |
1067 | switch (event) { | |
1068 | case NETDEV_REGISTER: | |
1069 | /* Create the device node here after the network device has | |
1070 | * been registered but before register_netdevice has | |
1071 | * finished running. | |
1072 | */ | |
e09eff7f EB |
1073 | err = macvtap_get_minor(vlan); |
1074 | if (err) | |
1075 | return notifier_from_errno(err); | |
1076 | ||
1077 | devt = MKDEV(MAJOR(macvtap_major), vlan->minor); | |
9bf1907f EB |
1078 | classdev = device_create(macvtap_class, &dev->dev, devt, |
1079 | dev, "tap%d", dev->ifindex); | |
e09eff7f EB |
1080 | if (IS_ERR(classdev)) { |
1081 | macvtap_free_minor(vlan); | |
9bf1907f | 1082 | return notifier_from_errno(PTR_ERR(classdev)); |
e09eff7f | 1083 | } |
9bf1907f EB |
1084 | break; |
1085 | case NETDEV_UNREGISTER: | |
e09eff7f | 1086 | devt = MKDEV(MAJOR(macvtap_major), vlan->minor); |
9bf1907f | 1087 | device_destroy(macvtap_class, devt); |
e09eff7f | 1088 | macvtap_free_minor(vlan); |
9bf1907f EB |
1089 | break; |
1090 | } | |
1091 | ||
1092 | return NOTIFY_DONE; | |
1093 | } | |
1094 | ||
1095 | static struct notifier_block macvtap_notifier_block __read_mostly = { | |
1096 | .notifier_call = macvtap_device_event, | |
1097 | }; | |
1098 | ||
20d29d7a AB |
1099 | static int macvtap_init(void) |
1100 | { | |
1101 | int err; | |
1102 | ||
1103 | err = alloc_chrdev_region(&macvtap_major, 0, | |
1104 | MACVTAP_NUM_DEVS, "macvtap"); | |
1105 | if (err) | |
1106 | goto out1; | |
1107 | ||
1108 | cdev_init(&macvtap_cdev, &macvtap_fops); | |
1109 | err = cdev_add(&macvtap_cdev, macvtap_major, MACVTAP_NUM_DEVS); | |
1110 | if (err) | |
1111 | goto out2; | |
1112 | ||
1113 | macvtap_class = class_create(THIS_MODULE, "macvtap"); | |
1114 | if (IS_ERR(macvtap_class)) { | |
1115 | err = PTR_ERR(macvtap_class); | |
1116 | goto out3; | |
1117 | } | |
1118 | ||
9bf1907f | 1119 | err = register_netdevice_notifier(&macvtap_notifier_block); |
20d29d7a AB |
1120 | if (err) |
1121 | goto out4; | |
1122 | ||
9bf1907f EB |
1123 | err = macvlan_link_register(&macvtap_link_ops); |
1124 | if (err) | |
1125 | goto out5; | |
1126 | ||
20d29d7a AB |
1127 | return 0; |
1128 | ||
9bf1907f EB |
1129 | out5: |
1130 | unregister_netdevice_notifier(&macvtap_notifier_block); | |
20d29d7a AB |
1131 | out4: |
1132 | class_unregister(macvtap_class); | |
1133 | out3: | |
1134 | cdev_del(&macvtap_cdev); | |
1135 | out2: | |
1136 | unregister_chrdev_region(macvtap_major, MACVTAP_NUM_DEVS); | |
1137 | out1: | |
1138 | return err; | |
1139 | } | |
1140 | module_init(macvtap_init); | |
1141 | ||
1142 | static void macvtap_exit(void) | |
1143 | { | |
1144 | rtnl_link_unregister(&macvtap_link_ops); | |
9bf1907f | 1145 | unregister_netdevice_notifier(&macvtap_notifier_block); |
20d29d7a AB |
1146 | class_unregister(macvtap_class); |
1147 | cdev_del(&macvtap_cdev); | |
1148 | unregister_chrdev_region(macvtap_major, MACVTAP_NUM_DEVS); | |
1149 | } | |
1150 | module_exit(macvtap_exit); | |
1151 | ||
1152 | MODULE_ALIAS_RTNL_LINK("macvtap"); | |
1153 | MODULE_AUTHOR("Arnd Bergmann <arnd@arndb.de>"); | |
1154 | MODULE_LICENSE("GPL"); |