]>
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 | * | |
20d29d7a AB |
34 | */ |
35 | struct macvtap_queue { | |
36 | struct sock sk; | |
37 | struct socket sock; | |
43815482 | 38 | struct socket_wq wq; |
55afbd08 | 39 | int vnet_hdr_sz; |
13707f9e | 40 | struct macvlan_dev __rcu *vlan; |
20d29d7a | 41 | struct file *file; |
b9fb9ee0 | 42 | unsigned int flags; |
376b1aab | 43 | u16 queue_index; |
815f236d JW |
44 | bool enabled; |
45 | struct list_head next; | |
20d29d7a AB |
46 | }; |
47 | ||
48 | static struct proto macvtap_proto = { | |
49 | .name = "macvtap", | |
50 | .owner = THIS_MODULE, | |
51 | .obj_size = sizeof (struct macvtap_queue), | |
52 | }; | |
53 | ||
54 | /* | |
e09eff7f | 55 | * Variables for dealing with macvtaps device numbers. |
20d29d7a | 56 | */ |
1ebed71a | 57 | static dev_t macvtap_major; |
e09eff7f EB |
58 | #define MACVTAP_NUM_DEVS (1U << MINORBITS) |
59 | static DEFINE_MUTEX(minor_lock); | |
60 | static DEFINE_IDR(minor_idr); | |
61 | ||
97bc3633 | 62 | #define GOODCOPY_LEN 128 |
20d29d7a AB |
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 | ||
2be5c767 VY |
68 | #define TUN_OFFLOADS (NETIF_F_HW_CSUM | NETIF_F_TSO_ECN | NETIF_F_TSO | \ |
69 | NETIF_F_TSO6 | NETIF_F_UFO) | |
70 | #define RX_OFFLOADS (NETIF_F_GRO | NETIF_F_LRO) | |
a567dd62 VY |
71 | #define TAP_FEATURES (NETIF_F_GSO | NETIF_F_SG) |
72 | ||
20d29d7a AB |
73 | /* |
74 | * RCU usage: | |
02df55d2 AB |
75 | * The macvtap_queue and the macvlan_dev are loosely coupled, the |
76 | * pointers from one to the other can only be read while rcu_read_lock | |
441ac0fc | 77 | * or rtnl is held. |
20d29d7a | 78 | * |
02df55d2 AB |
79 | * Both the file and the macvlan_dev hold a reference on the macvtap_queue |
80 | * through sock_hold(&q->sk). When the macvlan_dev goes away first, | |
81 | * q->vlan becomes inaccessible. When the files gets closed, | |
82 | * macvtap_get_queue() fails. | |
20d29d7a | 83 | * |
02df55d2 AB |
84 | * There may still be references to the struct sock inside of the |
85 | * queue from outbound SKBs, but these never reference back to the | |
86 | * file or the dev. The data structure is freed through __sk_free | |
87 | * when both our references and any pending SKBs are gone. | |
20d29d7a | 88 | */ |
20d29d7a | 89 | |
815f236d | 90 | static int macvtap_enable_queue(struct net_device *dev, struct file *file, |
20d29d7a | 91 | struct macvtap_queue *q) |
815f236d JW |
92 | { |
93 | struct macvlan_dev *vlan = netdev_priv(dev); | |
94 | int err = -EINVAL; | |
95 | ||
441ac0fc | 96 | ASSERT_RTNL(); |
815f236d JW |
97 | |
98 | if (q->enabled) | |
99 | goto out; | |
100 | ||
101 | err = 0; | |
102 | rcu_assign_pointer(vlan->taps[vlan->numvtaps], q); | |
103 | q->queue_index = vlan->numvtaps; | |
104 | q->enabled = true; | |
105 | ||
106 | vlan->numvtaps++; | |
107 | out: | |
815f236d JW |
108 | return err; |
109 | } | |
110 | ||
111 | static int macvtap_set_queue(struct net_device *dev, struct file *file, | |
112 | struct macvtap_queue *q) | |
20d29d7a AB |
113 | { |
114 | struct macvlan_dev *vlan = netdev_priv(dev); | |
115 | int err = -EBUSY; | |
116 | ||
441ac0fc | 117 | rtnl_lock(); |
815f236d | 118 | if (vlan->numqueues == MAX_MACVTAP_QUEUES) |
20d29d7a AB |
119 | goto out; |
120 | ||
121 | err = 0; | |
02df55d2 | 122 | rcu_assign_pointer(q->vlan, vlan); |
376b1aab | 123 | rcu_assign_pointer(vlan->taps[vlan->numvtaps], q); |
02df55d2 | 124 | sock_hold(&q->sk); |
20d29d7a AB |
125 | |
126 | q->file = file; | |
376b1aab | 127 | q->queue_index = vlan->numvtaps; |
815f236d | 128 | q->enabled = true; |
02df55d2 | 129 | file->private_data = q; |
815f236d | 130 | list_add_tail(&q->next, &vlan->queue_list); |
20d29d7a | 131 | |
1565c7c1 | 132 | vlan->numvtaps++; |
815f236d | 133 | vlan->numqueues++; |
1565c7c1 | 134 | |
20d29d7a | 135 | out: |
441ac0fc | 136 | rtnl_unlock(); |
20d29d7a AB |
137 | return err; |
138 | } | |
139 | ||
441ac0fc | 140 | static int macvtap_disable_queue(struct macvtap_queue *q) |
815f236d JW |
141 | { |
142 | struct macvlan_dev *vlan; | |
143 | struct macvtap_queue *nq; | |
144 | ||
441ac0fc | 145 | ASSERT_RTNL(); |
815f236d JW |
146 | if (!q->enabled) |
147 | return -EINVAL; | |
148 | ||
441ac0fc VY |
149 | vlan = rtnl_dereference(q->vlan); |
150 | ||
815f236d JW |
151 | if (vlan) { |
152 | int index = q->queue_index; | |
153 | BUG_ON(index >= vlan->numvtaps); | |
441ac0fc | 154 | nq = rtnl_dereference(vlan->taps[vlan->numvtaps - 1]); |
815f236d JW |
155 | nq->queue_index = index; |
156 | ||
157 | rcu_assign_pointer(vlan->taps[index], nq); | |
158 | RCU_INIT_POINTER(vlan->taps[vlan->numvtaps - 1], NULL); | |
159 | q->enabled = false; | |
160 | ||
161 | vlan->numvtaps--; | |
162 | } | |
163 | ||
164 | return 0; | |
165 | } | |
166 | ||
20d29d7a | 167 | /* |
02df55d2 AB |
168 | * The file owning the queue got closed, give up both |
169 | * the reference that the files holds as well as the | |
170 | * one from the macvlan_dev if that still exists. | |
20d29d7a AB |
171 | * |
172 | * Using the spinlock makes sure that we don't get | |
173 | * to the queue again after destroying it. | |
20d29d7a | 174 | */ |
02df55d2 | 175 | static void macvtap_put_queue(struct macvtap_queue *q) |
20d29d7a | 176 | { |
02df55d2 | 177 | struct macvlan_dev *vlan; |
20d29d7a | 178 | |
441ac0fc VY |
179 | rtnl_lock(); |
180 | vlan = rtnl_dereference(q->vlan); | |
181 | ||
02df55d2 | 182 | if (vlan) { |
815f236d | 183 | if (q->enabled) |
441ac0fc | 184 | BUG_ON(macvtap_disable_queue(q)); |
376b1aab | 185 | |
815f236d | 186 | vlan->numqueues--; |
2cfa5a04 | 187 | RCU_INIT_POINTER(q->vlan, NULL); |
02df55d2 | 188 | sock_put(&q->sk); |
815f236d | 189 | list_del_init(&q->next); |
20d29d7a AB |
190 | } |
191 | ||
441ac0fc | 192 | rtnl_unlock(); |
20d29d7a AB |
193 | |
194 | synchronize_rcu(); | |
195 | sock_put(&q->sk); | |
196 | } | |
197 | ||
198 | /* | |
1565c7c1 KK |
199 | * Select a queue based on the rxq of the device on which this packet |
200 | * arrived. If the incoming device is not mq, calculate a flow hash | |
201 | * to select a queue. If all fails, find the first available queue. | |
202 | * Cache vlan->numvtaps since it can become zero during the execution | |
203 | * of this function. | |
20d29d7a AB |
204 | */ |
205 | static struct macvtap_queue *macvtap_get_queue(struct net_device *dev, | |
206 | struct sk_buff *skb) | |
207 | { | |
208 | struct macvlan_dev *vlan = netdev_priv(dev); | |
1565c7c1 | 209 | struct macvtap_queue *tap = NULL; |
815f236d JW |
210 | /* Access to taps array is protected by rcu, but access to numvtaps |
211 | * isn't. Below we use it to lookup a queue, but treat it as a hint | |
212 | * and validate that the result isn't NULL - in case we are | |
213 | * racing against queue removal. | |
214 | */ | |
ed0483fa | 215 | int numvtaps = ACCESS_ONCE(vlan->numvtaps); |
1565c7c1 KK |
216 | __u32 rxq; |
217 | ||
218 | if (!numvtaps) | |
219 | goto out; | |
220 | ||
ef0002b5 KK |
221 | /* Check if we can use flow to select a queue */ |
222 | rxq = skb_get_rxhash(skb); | |
223 | if (rxq) { | |
224 | tap = rcu_dereference(vlan->taps[rxq % numvtaps]); | |
376b1aab | 225 | goto out; |
ef0002b5 KK |
226 | } |
227 | ||
1565c7c1 KK |
228 | if (likely(skb_rx_queue_recorded(skb))) { |
229 | rxq = skb_get_rx_queue(skb); | |
20d29d7a | 230 | |
1565c7c1 KK |
231 | while (unlikely(rxq >= numvtaps)) |
232 | rxq -= numvtaps; | |
233 | ||
234 | tap = rcu_dereference(vlan->taps[rxq]); | |
376b1aab | 235 | goto out; |
1565c7c1 KK |
236 | } |
237 | ||
376b1aab | 238 | tap = rcu_dereference(vlan->taps[0]); |
1565c7c1 KK |
239 | out: |
240 | return tap; | |
20d29d7a AB |
241 | } |
242 | ||
02df55d2 AB |
243 | /* |
244 | * The net_device is going away, give up the reference | |
1565c7c1 KK |
245 | * that it holds on all queues and safely set the pointer |
246 | * from the queues to NULL. | |
02df55d2 | 247 | */ |
20d29d7a AB |
248 | static void macvtap_del_queues(struct net_device *dev) |
249 | { | |
250 | struct macvlan_dev *vlan = netdev_priv(dev); | |
815f236d | 251 | struct macvtap_queue *q, *tmp, *qlist[MAX_MACVTAP_QUEUES]; |
1565c7c1 | 252 | int i, j = 0; |
02df55d2 | 253 | |
441ac0fc | 254 | ASSERT_RTNL(); |
815f236d JW |
255 | list_for_each_entry_safe(q, tmp, &vlan->queue_list, next) { |
256 | list_del_init(&q->next); | |
376b1aab | 257 | qlist[j++] = q; |
376b1aab | 258 | RCU_INIT_POINTER(q->vlan, NULL); |
815f236d JW |
259 | if (q->enabled) |
260 | vlan->numvtaps--; | |
261 | vlan->numqueues--; | |
564517e8 | 262 | } |
815f236d JW |
263 | for (i = 0; i < vlan->numvtaps; i++) |
264 | RCU_INIT_POINTER(vlan->taps[i], NULL); | |
265 | BUG_ON(vlan->numvtaps); | |
266 | BUG_ON(vlan->numqueues); | |
99f34b38 EB |
267 | /* guarantee that any future macvtap_set_queue will fail */ |
268 | vlan->numvtaps = MAX_MACVTAP_QUEUES; | |
1565c7c1 KK |
269 | |
270 | for (--j; j >= 0; j--) | |
271 | sock_put(&qlist[j]->sk); | |
20d29d7a AB |
272 | } |
273 | ||
274 | /* | |
275 | * Forward happens for data that gets sent from one macvlan | |
276 | * endpoint to another one in bridge mode. We just take | |
277 | * the skb and put it into the receive queue. | |
278 | */ | |
279 | static int macvtap_forward(struct net_device *dev, struct sk_buff *skb) | |
280 | { | |
3e4f8b78 | 281 | struct macvlan_dev *vlan = netdev_priv(dev); |
20d29d7a | 282 | struct macvtap_queue *q = macvtap_get_queue(dev, skb); |
a567dd62 VY |
283 | netdev_features_t features = TAP_FEATURES; |
284 | ||
20d29d7a | 285 | if (!q) |
8a35747a HX |
286 | goto drop; |
287 | ||
288 | if (skb_queue_len(&q->sk.sk_receive_queue) >= dev->tx_queue_len) | |
289 | goto drop; | |
20d29d7a | 290 | |
3e4f8b78 VY |
291 | skb->dev = dev; |
292 | /* Apply the forward feature mask so that we perform segmentation | |
293 | * according to users wishes. | |
294 | */ | |
a567dd62 | 295 | features |= vlan->tap_features; |
3e4f8b78 VY |
296 | if (netif_needs_gso(skb, features)) { |
297 | struct sk_buff *segs = __skb_gso_segment(skb, features, false); | |
298 | ||
299 | if (IS_ERR(segs)) | |
300 | goto drop; | |
301 | ||
302 | if (!segs) { | |
303 | skb_queue_tail(&q->sk.sk_receive_queue, skb); | |
304 | goto wake_up; | |
305 | } | |
306 | ||
307 | kfree_skb(skb); | |
308 | while (segs) { | |
309 | struct sk_buff *nskb = segs->next; | |
310 | ||
311 | segs->next = NULL; | |
312 | skb_queue_tail(&q->sk.sk_receive_queue, segs); | |
313 | segs = nskb; | |
314 | } | |
315 | } else { | |
316 | skb_queue_tail(&q->sk.sk_receive_queue, skb); | |
317 | } | |
318 | ||
319 | wake_up: | |
4a4771a5 | 320 | wake_up_interruptible_poll(sk_sleep(&q->sk), POLLIN | POLLRDNORM | POLLRDBAND); |
8a35747a HX |
321 | return NET_RX_SUCCESS; |
322 | ||
323 | drop: | |
324 | kfree_skb(skb); | |
325 | return NET_RX_DROP; | |
20d29d7a AB |
326 | } |
327 | ||
328 | /* | |
329 | * Receive is for data from the external interface (lowerdev), | |
330 | * in case of macvtap, we can treat that the same way as | |
331 | * forward, which macvlan cannot. | |
332 | */ | |
333 | static int macvtap_receive(struct sk_buff *skb) | |
334 | { | |
335 | skb_push(skb, ETH_HLEN); | |
336 | return macvtap_forward(skb->dev, skb); | |
337 | } | |
338 | ||
e09eff7f EB |
339 | static int macvtap_get_minor(struct macvlan_dev *vlan) |
340 | { | |
341 | int retval = -ENOMEM; | |
e09eff7f EB |
342 | |
343 | mutex_lock(&minor_lock); | |
ec09ebc1 TH |
344 | retval = idr_alloc(&minor_idr, vlan, 1, MACVTAP_NUM_DEVS, GFP_KERNEL); |
345 | if (retval >= 0) { | |
346 | vlan->minor = retval; | |
347 | } else if (retval == -ENOSPC) { | |
e09eff7f EB |
348 | printk(KERN_ERR "too many macvtap devices\n"); |
349 | retval = -EINVAL; | |
e09eff7f | 350 | } |
e09eff7f | 351 | mutex_unlock(&minor_lock); |
ec09ebc1 | 352 | return retval < 0 ? retval : 0; |
e09eff7f EB |
353 | } |
354 | ||
355 | static void macvtap_free_minor(struct macvlan_dev *vlan) | |
356 | { | |
357 | mutex_lock(&minor_lock); | |
358 | if (vlan->minor) { | |
359 | idr_remove(&minor_idr, vlan->minor); | |
360 | vlan->minor = 0; | |
361 | } | |
362 | mutex_unlock(&minor_lock); | |
363 | } | |
364 | ||
365 | static struct net_device *dev_get_by_macvtap_minor(int minor) | |
366 | { | |
367 | struct net_device *dev = NULL; | |
368 | struct macvlan_dev *vlan; | |
369 | ||
370 | mutex_lock(&minor_lock); | |
371 | vlan = idr_find(&minor_idr, minor); | |
372 | if (vlan) { | |
373 | dev = vlan->dev; | |
374 | dev_hold(dev); | |
375 | } | |
376 | mutex_unlock(&minor_lock); | |
377 | return dev; | |
378 | } | |
379 | ||
20d29d7a AB |
380 | static int macvtap_newlink(struct net *src_net, |
381 | struct net_device *dev, | |
382 | struct nlattr *tb[], | |
383 | struct nlattr *data[]) | |
384 | { | |
815f236d JW |
385 | struct macvlan_dev *vlan = netdev_priv(dev); |
386 | INIT_LIST_HEAD(&vlan->queue_list); | |
387 | ||
2be5c767 VY |
388 | /* Since macvlan supports all offloads by default, make |
389 | * tap support all offloads also. | |
390 | */ | |
391 | vlan->tap_features = TUN_OFFLOADS; | |
392 | ||
9bf1907f EB |
393 | /* Don't put anything that may fail after macvlan_common_newlink |
394 | * because we can't undo what it does. | |
395 | */ | |
396 | return macvlan_common_newlink(src_net, dev, tb, data, | |
397 | macvtap_receive, macvtap_forward); | |
20d29d7a AB |
398 | } |
399 | ||
400 | static void macvtap_dellink(struct net_device *dev, | |
401 | struct list_head *head) | |
402 | { | |
20d29d7a AB |
403 | macvtap_del_queues(dev); |
404 | macvlan_dellink(dev, head); | |
405 | } | |
406 | ||
8a35747a HX |
407 | static void macvtap_setup(struct net_device *dev) |
408 | { | |
409 | macvlan_common_setup(dev); | |
410 | dev->tx_queue_len = TUN_READQ_SIZE; | |
411 | } | |
412 | ||
20d29d7a AB |
413 | static struct rtnl_link_ops macvtap_link_ops __read_mostly = { |
414 | .kind = "macvtap", | |
8a35747a | 415 | .setup = macvtap_setup, |
20d29d7a AB |
416 | .newlink = macvtap_newlink, |
417 | .dellink = macvtap_dellink, | |
418 | }; | |
419 | ||
420 | ||
421 | static void macvtap_sock_write_space(struct sock *sk) | |
422 | { | |
43815482 ED |
423 | wait_queue_head_t *wqueue; |
424 | ||
20d29d7a AB |
425 | if (!sock_writeable(sk) || |
426 | !test_and_clear_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags)) | |
427 | return; | |
428 | ||
43815482 ED |
429 | wqueue = sk_sleep(sk); |
430 | if (wqueue && waitqueue_active(wqueue)) | |
431 | wake_up_interruptible_poll(wqueue, POLLOUT | POLLWRNORM | POLLWRBAND); | |
20d29d7a AB |
432 | } |
433 | ||
2259fef0 EB |
434 | static void macvtap_sock_destruct(struct sock *sk) |
435 | { | |
436 | skb_queue_purge(&sk->sk_receive_queue); | |
437 | } | |
438 | ||
20d29d7a AB |
439 | static int macvtap_open(struct inode *inode, struct file *file) |
440 | { | |
441 | struct net *net = current->nsproxy->net_ns; | |
e09eff7f | 442 | struct net_device *dev = dev_get_by_macvtap_minor(iminor(inode)); |
20d29d7a AB |
443 | struct macvtap_queue *q; |
444 | int err; | |
445 | ||
446 | err = -ENODEV; | |
447 | if (!dev) | |
448 | goto out; | |
449 | ||
20d29d7a AB |
450 | err = -ENOMEM; |
451 | q = (struct macvtap_queue *)sk_alloc(net, AF_UNSPEC, GFP_KERNEL, | |
452 | &macvtap_proto); | |
453 | if (!q) | |
454 | goto out; | |
455 | ||
d9a90a31 | 456 | RCU_INIT_POINTER(q->sock.wq, &q->wq); |
43815482 | 457 | init_waitqueue_head(&q->wq.wait); |
20d29d7a AB |
458 | q->sock.type = SOCK_RAW; |
459 | q->sock.state = SS_CONNECTED; | |
501c774c AB |
460 | q->sock.file = file; |
461 | q->sock.ops = &macvtap_socket_ops; | |
20d29d7a | 462 | sock_init_data(&q->sock, &q->sk); |
20d29d7a | 463 | q->sk.sk_write_space = macvtap_sock_write_space; |
2259fef0 | 464 | q->sk.sk_destruct = macvtap_sock_destruct; |
b9fb9ee0 | 465 | q->flags = IFF_VNET_HDR | IFF_NO_PI | IFF_TAP; |
55afbd08 | 466 | q->vnet_hdr_sz = sizeof(struct virtio_net_hdr); |
20d29d7a | 467 | |
97bc3633 SM |
468 | /* |
469 | * so far only KVM virtio_net uses macvtap, enable zero copy between | |
470 | * guest kernel and host kernel when lower device supports zerocopy | |
047af9cf EB |
471 | * |
472 | * The macvlan supports zerocopy iff the lower device supports zero | |
473 | * copy so we don't have to look at the lower device directly. | |
97bc3633 | 474 | */ |
047af9cf EB |
475 | if ((dev->features & NETIF_F_HIGHDMA) && (dev->features & NETIF_F_SG)) |
476 | sock_set_flag(&q->sk, SOCK_ZEROCOPY); | |
97bc3633 | 477 | |
20d29d7a AB |
478 | err = macvtap_set_queue(dev, file, q); |
479 | if (err) | |
480 | sock_put(&q->sk); | |
481 | ||
482 | out: | |
483 | if (dev) | |
484 | dev_put(dev); | |
485 | ||
486 | return err; | |
487 | } | |
488 | ||
489 | static int macvtap_release(struct inode *inode, struct file *file) | |
490 | { | |
02df55d2 AB |
491 | struct macvtap_queue *q = file->private_data; |
492 | macvtap_put_queue(q); | |
20d29d7a AB |
493 | return 0; |
494 | } | |
495 | ||
496 | static unsigned int macvtap_poll(struct file *file, poll_table * wait) | |
497 | { | |
02df55d2 | 498 | struct macvtap_queue *q = file->private_data; |
20d29d7a AB |
499 | unsigned int mask = POLLERR; |
500 | ||
501 | if (!q) | |
502 | goto out; | |
503 | ||
504 | mask = 0; | |
43815482 | 505 | poll_wait(file, &q->wq.wait, wait); |
20d29d7a AB |
506 | |
507 | if (!skb_queue_empty(&q->sk.sk_receive_queue)) | |
508 | mask |= POLLIN | POLLRDNORM; | |
509 | ||
510 | if (sock_writeable(&q->sk) || | |
511 | (!test_and_set_bit(SOCK_ASYNC_NOSPACE, &q->sock.flags) && | |
512 | sock_writeable(&q->sk))) | |
513 | mask |= POLLOUT | POLLWRNORM; | |
514 | ||
515 | out: | |
20d29d7a AB |
516 | return mask; |
517 | } | |
518 | ||
b9fb9ee0 AB |
519 | static inline struct sk_buff *macvtap_alloc_skb(struct sock *sk, size_t prepad, |
520 | size_t len, size_t linear, | |
521 | int noblock, int *err) | |
522 | { | |
523 | struct sk_buff *skb; | |
524 | ||
525 | /* Under a page? Don't bother with paged skb. */ | |
526 | if (prepad + len < PAGE_SIZE || !linear) | |
527 | linear = len; | |
528 | ||
529 | skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock, | |
530 | err); | |
531 | if (!skb) | |
532 | return NULL; | |
533 | ||
534 | skb_reserve(skb, prepad); | |
535 | skb_put(skb, linear); | |
536 | skb->data_len = len - linear; | |
537 | skb->len += len - linear; | |
538 | ||
539 | return skb; | |
540 | } | |
541 | ||
97bc3633 SM |
542 | /* set skb frags from iovec, this can move to core network code for reuse */ |
543 | static int zerocopy_sg_from_iovec(struct sk_buff *skb, const struct iovec *from, | |
544 | int offset, size_t count) | |
545 | { | |
546 | int len = iov_length(from, count) - offset; | |
547 | int copy = skb_headlen(skb); | |
548 | int size, offset1 = 0; | |
549 | int i = 0; | |
97bc3633 SM |
550 | |
551 | /* Skip over from offset */ | |
552 | while (count && (offset >= from->iov_len)) { | |
553 | offset -= from->iov_len; | |
554 | ++from; | |
555 | --count; | |
556 | } | |
557 | ||
558 | /* copy up to skb headlen */ | |
559 | while (count && (copy > 0)) { | |
560 | size = min_t(unsigned int, copy, from->iov_len - offset); | |
561 | if (copy_from_user(skb->data + offset1, from->iov_base + offset, | |
562 | size)) | |
563 | return -EFAULT; | |
564 | if (copy > size) { | |
565 | ++from; | |
566 | --count; | |
3afc9621 JW |
567 | offset = 0; |
568 | } else | |
569 | offset += size; | |
97bc3633 SM |
570 | copy -= size; |
571 | offset1 += size; | |
97bc3633 SM |
572 | } |
573 | ||
574 | if (len == offset1) | |
575 | return 0; | |
576 | ||
577 | while (count--) { | |
578 | struct page *page[MAX_SKB_FRAGS]; | |
579 | int num_pages; | |
580 | unsigned long base; | |
4ef67ebe | 581 | unsigned long truesize; |
97bc3633 | 582 | |
3afc9621 | 583 | len = from->iov_len - offset; |
97bc3633 | 584 | if (!len) { |
3afc9621 | 585 | offset = 0; |
97bc3633 SM |
586 | ++from; |
587 | continue; | |
588 | } | |
3afc9621 | 589 | base = (unsigned long)from->iov_base + offset; |
97bc3633 | 590 | size = ((base & ~PAGE_MASK) + len + ~PAGE_MASK) >> PAGE_SHIFT; |
b92946e2 JW |
591 | if (i + size > MAX_SKB_FRAGS) |
592 | return -EMSGSIZE; | |
97bc3633 | 593 | num_pages = get_user_pages_fast(base, size, 0, &page[i]); |
b92946e2 | 594 | if (num_pages != size) { |
4c7ab054 MT |
595 | int j; |
596 | ||
597 | for (j = 0; j < num_pages; j++) | |
598 | put_page(page[i + j]); | |
97bc3633 | 599 | return -EFAULT; |
02ce04bb | 600 | } |
4ef67ebe | 601 | truesize = size * PAGE_SIZE; |
97bc3633 SM |
602 | skb->data_len += len; |
603 | skb->len += len; | |
4ef67ebe JW |
604 | skb->truesize += truesize; |
605 | atomic_add(truesize, &skb->sk->sk_wmem_alloc); | |
97bc3633 | 606 | while (len) { |
653fc915 JW |
607 | int off = base & ~PAGE_MASK; |
608 | int size = min_t(int, len, PAGE_SIZE - off); | |
609 | __skb_fill_page_desc(skb, i, page[i], off, size); | |
97bc3633 SM |
610 | skb_shinfo(skb)->nr_frags++; |
611 | /* increase sk_wmem_alloc */ | |
653fc915 JW |
612 | base += size; |
613 | len -= size; | |
97bc3633 SM |
614 | i++; |
615 | } | |
3afc9621 | 616 | offset = 0; |
97bc3633 SM |
617 | ++from; |
618 | } | |
619 | return 0; | |
620 | } | |
621 | ||
b9fb9ee0 AB |
622 | /* |
623 | * macvtap_skb_from_vnet_hdr and macvtap_skb_to_vnet_hdr should | |
624 | * be shared with the tun/tap driver. | |
625 | */ | |
626 | static int macvtap_skb_from_vnet_hdr(struct sk_buff *skb, | |
627 | struct virtio_net_hdr *vnet_hdr) | |
628 | { | |
629 | unsigned short gso_type = 0; | |
630 | if (vnet_hdr->gso_type != VIRTIO_NET_HDR_GSO_NONE) { | |
631 | switch (vnet_hdr->gso_type & ~VIRTIO_NET_HDR_GSO_ECN) { | |
632 | case VIRTIO_NET_HDR_GSO_TCPV4: | |
633 | gso_type = SKB_GSO_TCPV4; | |
634 | break; | |
635 | case VIRTIO_NET_HDR_GSO_TCPV6: | |
636 | gso_type = SKB_GSO_TCPV6; | |
637 | break; | |
638 | case VIRTIO_NET_HDR_GSO_UDP: | |
639 | gso_type = SKB_GSO_UDP; | |
640 | break; | |
641 | default: | |
642 | return -EINVAL; | |
643 | } | |
644 | ||
645 | if (vnet_hdr->gso_type & VIRTIO_NET_HDR_GSO_ECN) | |
646 | gso_type |= SKB_GSO_TCP_ECN; | |
647 | ||
648 | if (vnet_hdr->gso_size == 0) | |
649 | return -EINVAL; | |
650 | } | |
651 | ||
652 | if (vnet_hdr->flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) { | |
653 | if (!skb_partial_csum_set(skb, vnet_hdr->csum_start, | |
654 | vnet_hdr->csum_offset)) | |
655 | return -EINVAL; | |
656 | } | |
657 | ||
658 | if (vnet_hdr->gso_type != VIRTIO_NET_HDR_GSO_NONE) { | |
659 | skb_shinfo(skb)->gso_size = vnet_hdr->gso_size; | |
c9af6db4 | 660 | skb_shinfo(skb)->gso_type = gso_type; |
b9fb9ee0 AB |
661 | |
662 | /* Header must be checked, and gso_segs computed. */ | |
663 | skb_shinfo(skb)->gso_type |= SKB_GSO_DODGY; | |
664 | skb_shinfo(skb)->gso_segs = 0; | |
665 | } | |
666 | return 0; | |
667 | } | |
668 | ||
669 | static int macvtap_skb_to_vnet_hdr(const struct sk_buff *skb, | |
670 | struct virtio_net_hdr *vnet_hdr) | |
671 | { | |
672 | memset(vnet_hdr, 0, sizeof(*vnet_hdr)); | |
673 | ||
674 | if (skb_is_gso(skb)) { | |
675 | struct skb_shared_info *sinfo = skb_shinfo(skb); | |
676 | ||
677 | /* This is a hint as to how much should be linear. */ | |
678 | vnet_hdr->hdr_len = skb_headlen(skb); | |
679 | vnet_hdr->gso_size = sinfo->gso_size; | |
680 | if (sinfo->gso_type & SKB_GSO_TCPV4) | |
681 | vnet_hdr->gso_type = VIRTIO_NET_HDR_GSO_TCPV4; | |
682 | else if (sinfo->gso_type & SKB_GSO_TCPV6) | |
683 | vnet_hdr->gso_type = VIRTIO_NET_HDR_GSO_TCPV6; | |
684 | else if (sinfo->gso_type & SKB_GSO_UDP) | |
685 | vnet_hdr->gso_type = VIRTIO_NET_HDR_GSO_UDP; | |
686 | else | |
687 | BUG(); | |
688 | if (sinfo->gso_type & SKB_GSO_TCP_ECN) | |
689 | vnet_hdr->gso_type |= VIRTIO_NET_HDR_GSO_ECN; | |
690 | } else | |
691 | vnet_hdr->gso_type = VIRTIO_NET_HDR_GSO_NONE; | |
692 | ||
693 | if (skb->ip_summed == CHECKSUM_PARTIAL) { | |
694 | vnet_hdr->flags = VIRTIO_NET_HDR_F_NEEDS_CSUM; | |
55508d60 | 695 | vnet_hdr->csum_start = skb_checksum_start_offset(skb); |
b9fb9ee0 | 696 | vnet_hdr->csum_offset = skb->csum_offset; |
10a8d94a JW |
697 | } else if (skb->ip_summed == CHECKSUM_UNNECESSARY) { |
698 | vnet_hdr->flags = VIRTIO_NET_HDR_F_DATA_VALID; | |
b9fb9ee0 AB |
699 | } /* else everything is zero */ |
700 | ||
701 | return 0; | |
702 | } | |
703 | ||
ece793fc JW |
704 | static unsigned long iov_pages(const struct iovec *iv, int offset, |
705 | unsigned long nr_segs) | |
706 | { | |
707 | unsigned long seg, base; | |
708 | int pages = 0, len, size; | |
709 | ||
710 | while (nr_segs && (offset >= iv->iov_len)) { | |
711 | offset -= iv->iov_len; | |
712 | ++iv; | |
713 | --nr_segs; | |
714 | } | |
715 | ||
716 | for (seg = 0; seg < nr_segs; seg++) { | |
717 | base = (unsigned long)iv[seg].iov_base + offset; | |
718 | len = iv[seg].iov_len - offset; | |
719 | size = ((base & ~PAGE_MASK) + len + ~PAGE_MASK) >> PAGE_SHIFT; | |
720 | pages += size; | |
721 | offset = 0; | |
722 | } | |
723 | ||
724 | return pages; | |
725 | } | |
b9fb9ee0 | 726 | |
20d29d7a | 727 | /* Get packet from user space buffer */ |
97bc3633 SM |
728 | static ssize_t macvtap_get_user(struct macvtap_queue *q, struct msghdr *m, |
729 | const struct iovec *iv, unsigned long total_len, | |
730 | size_t count, int noblock) | |
20d29d7a AB |
731 | { |
732 | struct sk_buff *skb; | |
02df55d2 | 733 | struct macvlan_dev *vlan; |
97bc3633 | 734 | unsigned long len = total_len; |
20d29d7a | 735 | int err; |
b9fb9ee0 AB |
736 | struct virtio_net_hdr vnet_hdr = { 0 }; |
737 | int vnet_hdr_len = 0; | |
b92946e2 | 738 | int copylen = 0; |
97bc3633 | 739 | bool zerocopy = false; |
61d46bf9 | 740 | size_t linear; |
b9fb9ee0 AB |
741 | |
742 | if (q->flags & IFF_VNET_HDR) { | |
55afbd08 | 743 | vnet_hdr_len = q->vnet_hdr_sz; |
b9fb9ee0 AB |
744 | |
745 | err = -EINVAL; | |
ce3c8692 | 746 | if (len < vnet_hdr_len) |
b9fb9ee0 | 747 | goto err; |
ce3c8692 | 748 | len -= vnet_hdr_len; |
b9fb9ee0 AB |
749 | |
750 | err = memcpy_fromiovecend((void *)&vnet_hdr, iv, 0, | |
55afbd08 | 751 | sizeof(vnet_hdr)); |
b9fb9ee0 AB |
752 | if (err < 0) |
753 | goto err; | |
754 | if ((vnet_hdr.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) && | |
755 | vnet_hdr.csum_start + vnet_hdr.csum_offset + 2 > | |
756 | vnet_hdr.hdr_len) | |
757 | vnet_hdr.hdr_len = vnet_hdr.csum_start + | |
758 | vnet_hdr.csum_offset + 2; | |
759 | err = -EINVAL; | |
760 | if (vnet_hdr.hdr_len > len) | |
761 | goto err; | |
762 | } | |
20d29d7a | 763 | |
b9fb9ee0 | 764 | err = -EINVAL; |
20d29d7a | 765 | if (unlikely(len < ETH_HLEN)) |
b9fb9ee0 | 766 | goto err; |
20d29d7a | 767 | |
b92946e2 JW |
768 | err = -EMSGSIZE; |
769 | if (unlikely(count > UIO_MAXIOV)) | |
770 | goto err; | |
771 | ||
ece793fc JW |
772 | if (m && m->msg_control && sock_flag(&q->sk, SOCK_ZEROCOPY)) { |
773 | copylen = vnet_hdr.hdr_len ? vnet_hdr.hdr_len : GOODCOPY_LEN; | |
61d46bf9 | 774 | linear = copylen; |
ece793fc JW |
775 | if (iov_pages(iv, vnet_hdr_len + copylen, count) |
776 | <= MAX_SKB_FRAGS) | |
777 | zerocopy = true; | |
778 | } | |
779 | ||
780 | if (!zerocopy) { | |
97bc3633 | 781 | copylen = len; |
61d46bf9 JW |
782 | linear = vnet_hdr.hdr_len; |
783 | } | |
97bc3633 SM |
784 | |
785 | skb = macvtap_alloc_skb(&q->sk, NET_IP_ALIGN, copylen, | |
61d46bf9 | 786 | linear, noblock, &err); |
02df55d2 AB |
787 | if (!skb) |
788 | goto err; | |
20d29d7a | 789 | |
01d6657b | 790 | if (zerocopy) |
97bc3633 | 791 | err = zerocopy_sg_from_iovec(skb, iv, vnet_hdr_len, count); |
ece793fc | 792 | else { |
97bc3633 SM |
793 | err = skb_copy_datagram_from_iovec(skb, 0, iv, vnet_hdr_len, |
794 | len); | |
ece793fc JW |
795 | if (!err && m && m->msg_control) { |
796 | struct ubuf_info *uarg = m->msg_control; | |
797 | uarg->callback(uarg, false); | |
798 | } | |
799 | } | |
800 | ||
02df55d2 | 801 | if (err) |
b9fb9ee0 | 802 | goto err_kfree; |
20d29d7a AB |
803 | |
804 | skb_set_network_header(skb, ETH_HLEN); | |
b9fb9ee0 AB |
805 | skb_reset_mac_header(skb); |
806 | skb->protocol = eth_hdr(skb)->h_proto; | |
807 | ||
808 | if (vnet_hdr_len) { | |
809 | err = macvtap_skb_from_vnet_hdr(skb, &vnet_hdr); | |
810 | if (err) | |
811 | goto err_kfree; | |
812 | } | |
813 | ||
40893fd0 | 814 | skb_probe_transport_header(skb, ETH_HLEN); |
9b4d669b | 815 | |
ac4e4af1 VY |
816 | rcu_read_lock(); |
817 | vlan = rcu_dereference(q->vlan); | |
97bc3633 | 818 | /* copy skb_ubuf_info for callback when skb has no error */ |
01d6657b | 819 | if (zerocopy) { |
97bc3633 | 820 | skb_shinfo(skb)->destructor_arg = m->msg_control; |
01d6657b | 821 | skb_shinfo(skb)->tx_flags |= SKBTX_DEV_ZEROCOPY; |
c9af6db4 | 822 | skb_shinfo(skb)->tx_flags |= SKBTX_SHARED_FRAG; |
01d6657b | 823 | } |
29d79196 ED |
824 | if (vlan) { |
825 | local_bh_disable(); | |
02df55d2 | 826 | macvlan_start_xmit(skb, vlan->dev); |
29d79196 ED |
827 | local_bh_enable(); |
828 | } else { | |
02df55d2 | 829 | kfree_skb(skb); |
29d79196 | 830 | } |
ac4e4af1 | 831 | rcu_read_unlock(); |
20d29d7a | 832 | |
97bc3633 | 833 | return total_len; |
02df55d2 | 834 | |
b9fb9ee0 AB |
835 | err_kfree: |
836 | kfree_skb(skb); | |
837 | ||
02df55d2 | 838 | err: |
ac4e4af1 VY |
839 | rcu_read_lock(); |
840 | vlan = rcu_dereference(q->vlan); | |
02df55d2 | 841 | if (vlan) |
1ac9ad13 | 842 | vlan->dev->stats.tx_dropped++; |
ac4e4af1 | 843 | rcu_read_unlock(); |
02df55d2 | 844 | |
02df55d2 | 845 | return err; |
20d29d7a AB |
846 | } |
847 | ||
848 | static ssize_t macvtap_aio_write(struct kiocb *iocb, const struct iovec *iv, | |
849 | unsigned long count, loff_t pos) | |
850 | { | |
851 | struct file *file = iocb->ki_filp; | |
852 | ssize_t result = -ENOLINK; | |
02df55d2 | 853 | struct macvtap_queue *q = file->private_data; |
20d29d7a | 854 | |
97bc3633 SM |
855 | result = macvtap_get_user(q, NULL, iv, iov_length(iv, count), count, |
856 | file->f_flags & O_NONBLOCK); | |
20d29d7a AB |
857 | return result; |
858 | } | |
859 | ||
860 | /* Put packet to the user space buffer */ | |
861 | static ssize_t macvtap_put_user(struct macvtap_queue *q, | |
862 | const struct sk_buff *skb, | |
863 | const struct iovec *iv, int len) | |
864 | { | |
02df55d2 | 865 | struct macvlan_dev *vlan; |
20d29d7a | 866 | int ret; |
b9fb9ee0 | 867 | int vnet_hdr_len = 0; |
f09e2249 BG |
868 | int vlan_offset = 0; |
869 | int copied; | |
b9fb9ee0 AB |
870 | |
871 | if (q->flags & IFF_VNET_HDR) { | |
872 | struct virtio_net_hdr vnet_hdr; | |
55afbd08 | 873 | vnet_hdr_len = q->vnet_hdr_sz; |
b9fb9ee0 AB |
874 | if ((len -= vnet_hdr_len) < 0) |
875 | return -EINVAL; | |
876 | ||
877 | ret = macvtap_skb_to_vnet_hdr(skb, &vnet_hdr); | |
878 | if (ret) | |
879 | return ret; | |
880 | ||
55afbd08 | 881 | if (memcpy_toiovecend(iv, (void *)&vnet_hdr, 0, sizeof(vnet_hdr))) |
b9fb9ee0 AB |
882 | return -EFAULT; |
883 | } | |
f09e2249 BG |
884 | copied = vnet_hdr_len; |
885 | ||
886 | if (!vlan_tx_tag_present(skb)) | |
887 | len = min_t(int, skb->len, len); | |
888 | else { | |
889 | int copy; | |
890 | struct { | |
891 | __be16 h_vlan_proto; | |
892 | __be16 h_vlan_TCI; | |
893 | } veth; | |
0fbe0d47 | 894 | veth.h_vlan_proto = skb->vlan_proto; |
f09e2249 BG |
895 | veth.h_vlan_TCI = htons(vlan_tx_tag_get(skb)); |
896 | ||
897 | vlan_offset = offsetof(struct vlan_ethhdr, h_vlan_proto); | |
898 | len = min_t(int, skb->len + VLAN_HLEN, len); | |
899 | ||
900 | copy = min_t(int, vlan_offset, len); | |
901 | ret = skb_copy_datagram_const_iovec(skb, 0, iv, copied, copy); | |
902 | len -= copy; | |
903 | copied += copy; | |
904 | if (ret || !len) | |
905 | goto done; | |
906 | ||
907 | copy = min_t(int, sizeof(veth), len); | |
908 | ret = memcpy_toiovecend(iv, (void *)&veth, copied, copy); | |
909 | len -= copy; | |
910 | copied += copy; | |
911 | if (ret || !len) | |
912 | goto done; | |
913 | } | |
20d29d7a | 914 | |
f09e2249 BG |
915 | ret = skb_copy_datagram_const_iovec(skb, vlan_offset, iv, copied, len); |
916 | copied += len; | |
20d29d7a | 917 | |
f09e2249 | 918 | done: |
ac4e4af1 VY |
919 | rcu_read_lock(); |
920 | vlan = rcu_dereference(q->vlan); | |
29d79196 ED |
921 | if (vlan) { |
922 | preempt_disable(); | |
f09e2249 | 923 | macvlan_count_rx(vlan, copied - vnet_hdr_len, ret == 0, 0); |
29d79196 ED |
924 | preempt_enable(); |
925 | } | |
ac4e4af1 | 926 | rcu_read_unlock(); |
20d29d7a | 927 | |
f09e2249 | 928 | return ret ? ret : copied; |
20d29d7a AB |
929 | } |
930 | ||
501c774c AB |
931 | static ssize_t macvtap_do_read(struct macvtap_queue *q, struct kiocb *iocb, |
932 | const struct iovec *iv, unsigned long len, | |
933 | int noblock) | |
20d29d7a | 934 | { |
ccf7e72b | 935 | DEFINE_WAIT(wait); |
20d29d7a | 936 | struct sk_buff *skb; |
501c774c | 937 | ssize_t ret = 0; |
20d29d7a | 938 | |
20d29d7a | 939 | while (len) { |
89cee917 JW |
940 | if (!noblock) |
941 | prepare_to_wait(sk_sleep(&q->sk), &wait, | |
942 | TASK_INTERRUPTIBLE); | |
20d29d7a AB |
943 | |
944 | /* Read frames from the queue */ | |
945 | skb = skb_dequeue(&q->sk.sk_receive_queue); | |
946 | if (!skb) { | |
501c774c | 947 | if (noblock) { |
20d29d7a AB |
948 | ret = -EAGAIN; |
949 | break; | |
950 | } | |
951 | if (signal_pending(current)) { | |
952 | ret = -ERESTARTSYS; | |
953 | break; | |
954 | } | |
955 | /* Nothing to read, let's sleep */ | |
956 | schedule(); | |
957 | continue; | |
958 | } | |
959 | ret = macvtap_put_user(q, skb, iv, len); | |
960 | kfree_skb(skb); | |
961 | break; | |
962 | } | |
963 | ||
89cee917 JW |
964 | if (!noblock) |
965 | finish_wait(sk_sleep(&q->sk), &wait); | |
501c774c AB |
966 | return ret; |
967 | } | |
968 | ||
969 | static ssize_t macvtap_aio_read(struct kiocb *iocb, const struct iovec *iv, | |
970 | unsigned long count, loff_t pos) | |
971 | { | |
972 | struct file *file = iocb->ki_filp; | |
973 | struct macvtap_queue *q = file->private_data; | |
974 | ssize_t len, ret = 0; | |
20d29d7a | 975 | |
501c774c AB |
976 | len = iov_length(iv, count); |
977 | if (len < 0) { | |
978 | ret = -EINVAL; | |
979 | goto out; | |
980 | } | |
981 | ||
982 | ret = macvtap_do_read(q, iocb, iv, len, file->f_flags & O_NONBLOCK); | |
983 | ret = min_t(ssize_t, ret, len); /* XXX copied from tun.c. Why? */ | |
20d29d7a | 984 | out: |
20d29d7a AB |
985 | return ret; |
986 | } | |
987 | ||
8f475a31 JW |
988 | static struct macvlan_dev *macvtap_get_vlan(struct macvtap_queue *q) |
989 | { | |
990 | struct macvlan_dev *vlan; | |
991 | ||
441ac0fc VY |
992 | ASSERT_RTNL(); |
993 | vlan = rtnl_dereference(q->vlan); | |
8f475a31 JW |
994 | if (vlan) |
995 | dev_hold(vlan->dev); | |
8f475a31 JW |
996 | |
997 | return vlan; | |
998 | } | |
999 | ||
1000 | static void macvtap_put_vlan(struct macvlan_dev *vlan) | |
1001 | { | |
1002 | dev_put(vlan->dev); | |
1003 | } | |
1004 | ||
815f236d JW |
1005 | static int macvtap_ioctl_set_queue(struct file *file, unsigned int flags) |
1006 | { | |
1007 | struct macvtap_queue *q = file->private_data; | |
1008 | struct macvlan_dev *vlan; | |
1009 | int ret; | |
1010 | ||
1011 | vlan = macvtap_get_vlan(q); | |
1012 | if (!vlan) | |
1013 | return -EINVAL; | |
1014 | ||
1015 | if (flags & IFF_ATTACH_QUEUE) | |
1016 | ret = macvtap_enable_queue(vlan->dev, file, q); | |
1017 | else if (flags & IFF_DETACH_QUEUE) | |
1018 | ret = macvtap_disable_queue(q); | |
f57855a5 JW |
1019 | else |
1020 | ret = -EINVAL; | |
815f236d JW |
1021 | |
1022 | macvtap_put_vlan(vlan); | |
1023 | return ret; | |
1024 | } | |
1025 | ||
2be5c767 VY |
1026 | static int set_offload(struct macvtap_queue *q, unsigned long arg) |
1027 | { | |
1028 | struct macvlan_dev *vlan; | |
1029 | netdev_features_t features; | |
1030 | netdev_features_t feature_mask = 0; | |
1031 | ||
1032 | vlan = rtnl_dereference(q->vlan); | |
1033 | if (!vlan) | |
1034 | return -ENOLINK; | |
1035 | ||
1036 | features = vlan->dev->features; | |
1037 | ||
1038 | if (arg & TUN_F_CSUM) { | |
1039 | feature_mask = NETIF_F_HW_CSUM; | |
1040 | ||
1041 | if (arg & (TUN_F_TSO4 | TUN_F_TSO6)) { | |
1042 | if (arg & TUN_F_TSO_ECN) | |
1043 | feature_mask |= NETIF_F_TSO_ECN; | |
1044 | if (arg & TUN_F_TSO4) | |
1045 | feature_mask |= NETIF_F_TSO; | |
1046 | if (arg & TUN_F_TSO6) | |
1047 | feature_mask |= NETIF_F_TSO6; | |
1048 | } | |
1049 | ||
1050 | if (arg & TUN_F_UFO) | |
1051 | feature_mask |= NETIF_F_UFO; | |
1052 | } | |
1053 | ||
1054 | /* tun/tap driver inverts the usage for TSO offloads, where | |
1055 | * setting the TSO bit means that the userspace wants to | |
1056 | * accept TSO frames and turning it off means that user space | |
1057 | * does not support TSO. | |
1058 | * For macvtap, we have to invert it to mean the same thing. | |
1059 | * When user space turns off TSO, we turn off GSO/LRO so that | |
1060 | * user-space will not receive TSO frames. | |
1061 | */ | |
1062 | if (feature_mask & (NETIF_F_TSO | NETIF_F_TSO6 | NETIF_F_UFO)) | |
1063 | features |= RX_OFFLOADS; | |
1064 | else | |
1065 | features &= ~RX_OFFLOADS; | |
1066 | ||
1067 | /* tap_features are the same as features on tun/tap and | |
1068 | * reflect user expectations. | |
1069 | */ | |
a567dd62 | 1070 | vlan->tap_features = feature_mask; |
2be5c767 VY |
1071 | vlan->set_features = features; |
1072 | netdev_update_features(vlan->dev); | |
1073 | ||
1074 | return 0; | |
1075 | } | |
1076 | ||
20d29d7a AB |
1077 | /* |
1078 | * provide compatibility with generic tun/tap interface | |
1079 | */ | |
1080 | static long macvtap_ioctl(struct file *file, unsigned int cmd, | |
1081 | unsigned long arg) | |
1082 | { | |
02df55d2 AB |
1083 | struct macvtap_queue *q = file->private_data; |
1084 | struct macvlan_dev *vlan; | |
20d29d7a AB |
1085 | void __user *argp = (void __user *)arg; |
1086 | struct ifreq __user *ifr = argp; | |
1087 | unsigned int __user *up = argp; | |
1088 | unsigned int u; | |
55afbd08 MT |
1089 | int __user *sp = argp; |
1090 | int s; | |
02df55d2 | 1091 | int ret; |
20d29d7a AB |
1092 | |
1093 | switch (cmd) { | |
1094 | case TUNSETIFF: | |
1095 | /* ignore the name, just look at flags */ | |
1096 | if (get_user(u, &ifr->ifr_flags)) | |
1097 | return -EFAULT; | |
b9fb9ee0 AB |
1098 | |
1099 | ret = 0; | |
815f236d JW |
1100 | if ((u & ~(IFF_VNET_HDR | IFF_MULTI_QUEUE)) != |
1101 | (IFF_NO_PI | IFF_TAP)) | |
b9fb9ee0 AB |
1102 | ret = -EINVAL; |
1103 | else | |
1104 | q->flags = u; | |
1105 | ||
1106 | return ret; | |
20d29d7a AB |
1107 | |
1108 | case TUNGETIFF: | |
441ac0fc | 1109 | rtnl_lock(); |
8f475a31 | 1110 | vlan = macvtap_get_vlan(q); |
441ac0fc VY |
1111 | if (!vlan) { |
1112 | rtnl_unlock(); | |
20d29d7a | 1113 | return -ENOLINK; |
441ac0fc | 1114 | } |
20d29d7a | 1115 | |
02df55d2 | 1116 | ret = 0; |
13707f9e | 1117 | if (copy_to_user(&ifr->ifr_name, vlan->dev->name, IFNAMSIZ) || |
b9fb9ee0 | 1118 | put_user(q->flags, &ifr->ifr_flags)) |
02df55d2 | 1119 | ret = -EFAULT; |
8f475a31 | 1120 | macvtap_put_vlan(vlan); |
441ac0fc | 1121 | rtnl_unlock(); |
02df55d2 | 1122 | return ret; |
20d29d7a | 1123 | |
815f236d JW |
1124 | case TUNSETQUEUE: |
1125 | if (get_user(u, &ifr->ifr_flags)) | |
1126 | return -EFAULT; | |
441ac0fc VY |
1127 | rtnl_lock(); |
1128 | ret = macvtap_ioctl_set_queue(file, u); | |
1129 | rtnl_unlock(); | |
82a19eb8 | 1130 | return ret; |
815f236d | 1131 | |
20d29d7a | 1132 | case TUNGETFEATURES: |
df09b36f JW |
1133 | if (put_user(IFF_TAP | IFF_NO_PI | IFF_VNET_HDR | |
1134 | IFF_MULTI_QUEUE, up)) | |
20d29d7a AB |
1135 | return -EFAULT; |
1136 | return 0; | |
1137 | ||
1138 | case TUNSETSNDBUF: | |
1139 | if (get_user(u, up)) | |
1140 | return -EFAULT; | |
1141 | ||
20d29d7a | 1142 | q->sk.sk_sndbuf = u; |
20d29d7a AB |
1143 | return 0; |
1144 | ||
55afbd08 MT |
1145 | case TUNGETVNETHDRSZ: |
1146 | s = q->vnet_hdr_sz; | |
1147 | if (put_user(s, sp)) | |
1148 | return -EFAULT; | |
1149 | return 0; | |
1150 | ||
1151 | case TUNSETVNETHDRSZ: | |
1152 | if (get_user(s, sp)) | |
1153 | return -EFAULT; | |
1154 | if (s < (int)sizeof(struct virtio_net_hdr)) | |
1155 | return -EINVAL; | |
1156 | ||
1157 | q->vnet_hdr_sz = s; | |
1158 | return 0; | |
1159 | ||
20d29d7a AB |
1160 | case TUNSETOFFLOAD: |
1161 | /* let the user check for future flags */ | |
1162 | if (arg & ~(TUN_F_CSUM | TUN_F_TSO4 | TUN_F_TSO6 | | |
b9fb9ee0 | 1163 | TUN_F_TSO_ECN | TUN_F_UFO)) |
20d29d7a AB |
1164 | return -EINVAL; |
1165 | ||
2be5c767 VY |
1166 | rtnl_lock(); |
1167 | ret = set_offload(q, arg); | |
1168 | rtnl_unlock(); | |
1169 | return ret; | |
20d29d7a AB |
1170 | |
1171 | default: | |
1172 | return -EINVAL; | |
1173 | } | |
1174 | } | |
1175 | ||
1176 | #ifdef CONFIG_COMPAT | |
1177 | static long macvtap_compat_ioctl(struct file *file, unsigned int cmd, | |
1178 | unsigned long arg) | |
1179 | { | |
1180 | return macvtap_ioctl(file, cmd, (unsigned long)compat_ptr(arg)); | |
1181 | } | |
1182 | #endif | |
1183 | ||
1184 | static const struct file_operations macvtap_fops = { | |
1185 | .owner = THIS_MODULE, | |
1186 | .open = macvtap_open, | |
1187 | .release = macvtap_release, | |
1188 | .aio_read = macvtap_aio_read, | |
1189 | .aio_write = macvtap_aio_write, | |
1190 | .poll = macvtap_poll, | |
1191 | .llseek = no_llseek, | |
1192 | .unlocked_ioctl = macvtap_ioctl, | |
1193 | #ifdef CONFIG_COMPAT | |
1194 | .compat_ioctl = macvtap_compat_ioctl, | |
1195 | #endif | |
1196 | }; | |
1197 | ||
501c774c AB |
1198 | static int macvtap_sendmsg(struct kiocb *iocb, struct socket *sock, |
1199 | struct msghdr *m, size_t total_len) | |
1200 | { | |
1201 | struct macvtap_queue *q = container_of(sock, struct macvtap_queue, sock); | |
97bc3633 | 1202 | return macvtap_get_user(q, m, m->msg_iov, total_len, m->msg_iovlen, |
501c774c AB |
1203 | m->msg_flags & MSG_DONTWAIT); |
1204 | } | |
1205 | ||
1206 | static int macvtap_recvmsg(struct kiocb *iocb, struct socket *sock, | |
1207 | struct msghdr *m, size_t total_len, | |
1208 | int flags) | |
1209 | { | |
1210 | struct macvtap_queue *q = container_of(sock, struct macvtap_queue, sock); | |
1211 | int ret; | |
1212 | if (flags & ~(MSG_DONTWAIT|MSG_TRUNC)) | |
1213 | return -EINVAL; | |
1214 | ret = macvtap_do_read(q, iocb, m->msg_iov, total_len, | |
1215 | flags & MSG_DONTWAIT); | |
1216 | if (ret > total_len) { | |
1217 | m->msg_flags |= MSG_TRUNC; | |
1218 | ret = flags & MSG_TRUNC ? ret : total_len; | |
1219 | } | |
1220 | return ret; | |
1221 | } | |
1222 | ||
1223 | /* Ops structure to mimic raw sockets with tun */ | |
1224 | static const struct proto_ops macvtap_socket_ops = { | |
1225 | .sendmsg = macvtap_sendmsg, | |
1226 | .recvmsg = macvtap_recvmsg, | |
1227 | }; | |
1228 | ||
1229 | /* Get an underlying socket object from tun file. Returns error unless file is | |
1230 | * attached to a device. The returned object works like a packet socket, it | |
1231 | * can be used for sock_sendmsg/sock_recvmsg. The caller is responsible for | |
1232 | * holding a reference to the file for as long as the socket is in use. */ | |
1233 | struct socket *macvtap_get_socket(struct file *file) | |
1234 | { | |
1235 | struct macvtap_queue *q; | |
1236 | if (file->f_op != &macvtap_fops) | |
1237 | return ERR_PTR(-EINVAL); | |
1238 | q = file->private_data; | |
1239 | if (!q) | |
1240 | return ERR_PTR(-EBADFD); | |
1241 | return &q->sock; | |
1242 | } | |
1243 | EXPORT_SYMBOL_GPL(macvtap_get_socket); | |
1244 | ||
9bf1907f EB |
1245 | static int macvtap_device_event(struct notifier_block *unused, |
1246 | unsigned long event, void *ptr) | |
1247 | { | |
351638e7 | 1248 | struct net_device *dev = netdev_notifier_info_to_dev(ptr); |
e09eff7f | 1249 | struct macvlan_dev *vlan; |
9bf1907f EB |
1250 | struct device *classdev; |
1251 | dev_t devt; | |
e09eff7f | 1252 | int err; |
9bf1907f EB |
1253 | |
1254 | if (dev->rtnl_link_ops != &macvtap_link_ops) | |
1255 | return NOTIFY_DONE; | |
1256 | ||
e09eff7f | 1257 | vlan = netdev_priv(dev); |
9bf1907f EB |
1258 | |
1259 | switch (event) { | |
1260 | case NETDEV_REGISTER: | |
1261 | /* Create the device node here after the network device has | |
1262 | * been registered but before register_netdevice has | |
1263 | * finished running. | |
1264 | */ | |
e09eff7f EB |
1265 | err = macvtap_get_minor(vlan); |
1266 | if (err) | |
1267 | return notifier_from_errno(err); | |
1268 | ||
1269 | devt = MKDEV(MAJOR(macvtap_major), vlan->minor); | |
9bf1907f EB |
1270 | classdev = device_create(macvtap_class, &dev->dev, devt, |
1271 | dev, "tap%d", dev->ifindex); | |
e09eff7f EB |
1272 | if (IS_ERR(classdev)) { |
1273 | macvtap_free_minor(vlan); | |
9bf1907f | 1274 | return notifier_from_errno(PTR_ERR(classdev)); |
e09eff7f | 1275 | } |
9bf1907f EB |
1276 | break; |
1277 | case NETDEV_UNREGISTER: | |
e09eff7f | 1278 | devt = MKDEV(MAJOR(macvtap_major), vlan->minor); |
9bf1907f | 1279 | device_destroy(macvtap_class, devt); |
e09eff7f | 1280 | macvtap_free_minor(vlan); |
9bf1907f EB |
1281 | break; |
1282 | } | |
1283 | ||
1284 | return NOTIFY_DONE; | |
1285 | } | |
1286 | ||
1287 | static struct notifier_block macvtap_notifier_block __read_mostly = { | |
1288 | .notifier_call = macvtap_device_event, | |
1289 | }; | |
1290 | ||
20d29d7a AB |
1291 | static int macvtap_init(void) |
1292 | { | |
1293 | int err; | |
1294 | ||
1295 | err = alloc_chrdev_region(&macvtap_major, 0, | |
1296 | MACVTAP_NUM_DEVS, "macvtap"); | |
1297 | if (err) | |
1298 | goto out1; | |
1299 | ||
1300 | cdev_init(&macvtap_cdev, &macvtap_fops); | |
1301 | err = cdev_add(&macvtap_cdev, macvtap_major, MACVTAP_NUM_DEVS); | |
1302 | if (err) | |
1303 | goto out2; | |
1304 | ||
1305 | macvtap_class = class_create(THIS_MODULE, "macvtap"); | |
1306 | if (IS_ERR(macvtap_class)) { | |
1307 | err = PTR_ERR(macvtap_class); | |
1308 | goto out3; | |
1309 | } | |
1310 | ||
9bf1907f | 1311 | err = register_netdevice_notifier(&macvtap_notifier_block); |
20d29d7a AB |
1312 | if (err) |
1313 | goto out4; | |
1314 | ||
9bf1907f EB |
1315 | err = macvlan_link_register(&macvtap_link_ops); |
1316 | if (err) | |
1317 | goto out5; | |
1318 | ||
20d29d7a AB |
1319 | return 0; |
1320 | ||
9bf1907f EB |
1321 | out5: |
1322 | unregister_netdevice_notifier(&macvtap_notifier_block); | |
20d29d7a AB |
1323 | out4: |
1324 | class_unregister(macvtap_class); | |
1325 | out3: | |
1326 | cdev_del(&macvtap_cdev); | |
1327 | out2: | |
1328 | unregister_chrdev_region(macvtap_major, MACVTAP_NUM_DEVS); | |
1329 | out1: | |
1330 | return err; | |
1331 | } | |
1332 | module_init(macvtap_init); | |
1333 | ||
1334 | static void macvtap_exit(void) | |
1335 | { | |
1336 | rtnl_link_unregister(&macvtap_link_ops); | |
9bf1907f | 1337 | unregister_netdevice_notifier(&macvtap_notifier_block); |
20d29d7a AB |
1338 | class_unregister(macvtap_class); |
1339 | cdev_del(&macvtap_cdev); | |
1340 | unregister_chrdev_region(macvtap_major, MACVTAP_NUM_DEVS); | |
1341 | } | |
1342 | module_exit(macvtap_exit); | |
1343 | ||
1344 | MODULE_ALIAS_RTNL_LINK("macvtap"); | |
1345 | MODULE_AUTHOR("Arnd Bergmann <arnd@arndb.de>"); | |
1346 | MODULE_LICENSE("GPL"); |