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