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/wait.h> |
15 | #include <linux/cdev.h> | |
40401530 | 16 | #include <linux/idr.h> |
20d29d7a | 17 | #include <linux/fs.h> |
6c36d2e2 | 18 | #include <linux/uio.h> |
20d29d7a | 19 | |
5188cd44 | 20 | #include <net/ipv6.h> |
20d29d7a AB |
21 | #include <net/net_namespace.h> |
22 | #include <net/rtnetlink.h> | |
23 | #include <net/sock.h> | |
b9fb9ee0 | 24 | #include <linux/virtio_net.h> |
20d29d7a AB |
25 | |
26 | /* | |
27 | * A macvtap queue is the central object of this driver, it connects | |
28 | * an open character device to a macvlan interface. There can be | |
29 | * multiple queues on one interface, which map back to queues | |
30 | * implemented in hardware on the underlying device. | |
31 | * | |
32 | * macvtap_proto is used to allocate queues through the sock allocation | |
33 | * mechanism. | |
34 | * | |
20d29d7a AB |
35 | */ |
36 | struct macvtap_queue { | |
37 | struct sock sk; | |
38 | struct socket sock; | |
43815482 | 39 | struct socket_wq wq; |
55afbd08 | 40 | int vnet_hdr_sz; |
13707f9e | 41 | struct macvlan_dev __rcu *vlan; |
20d29d7a | 42 | struct file *file; |
b9fb9ee0 | 43 | unsigned int flags; |
376b1aab | 44 | u16 queue_index; |
815f236d JW |
45 | bool enabled; |
46 | struct list_head next; | |
20d29d7a AB |
47 | }; |
48 | ||
49 | static struct proto macvtap_proto = { | |
50 | .name = "macvtap", | |
51 | .owner = THIS_MODULE, | |
52 | .obj_size = sizeof (struct macvtap_queue), | |
53 | }; | |
54 | ||
55 | /* | |
e09eff7f | 56 | * Variables for dealing with macvtaps device numbers. |
20d29d7a | 57 | */ |
1ebed71a | 58 | static dev_t macvtap_major; |
e09eff7f EB |
59 | #define MACVTAP_NUM_DEVS (1U << MINORBITS) |
60 | static DEFINE_MUTEX(minor_lock); | |
61 | static DEFINE_IDR(minor_idr); | |
62 | ||
97bc3633 | 63 | #define GOODCOPY_LEN 128 |
20d29d7a AB |
64 | static struct class *macvtap_class; |
65 | static struct cdev macvtap_cdev; | |
66 | ||
501c774c AB |
67 | static const struct proto_ops macvtap_socket_ops; |
68 | ||
2be5c767 | 69 | #define TUN_OFFLOADS (NETIF_F_HW_CSUM | NETIF_F_TSO_ECN | NETIF_F_TSO | \ |
3d0ad094 | 70 | NETIF_F_TSO6) |
2be5c767 | 71 | #define RX_OFFLOADS (NETIF_F_GRO | NETIF_F_LRO) |
a567dd62 VY |
72 | #define TAP_FEATURES (NETIF_F_GSO | NETIF_F_SG) |
73 | ||
6acf54f1 VY |
74 | static struct macvlan_dev *macvtap_get_vlan_rcu(const struct net_device *dev) |
75 | { | |
76 | return rcu_dereference(dev->rx_handler_data); | |
77 | } | |
78 | ||
20d29d7a AB |
79 | /* |
80 | * RCU usage: | |
02df55d2 AB |
81 | * The macvtap_queue and the macvlan_dev are loosely coupled, the |
82 | * pointers from one to the other can only be read while rcu_read_lock | |
441ac0fc | 83 | * or rtnl is held. |
20d29d7a | 84 | * |
02df55d2 AB |
85 | * Both the file and the macvlan_dev hold a reference on the macvtap_queue |
86 | * through sock_hold(&q->sk). When the macvlan_dev goes away first, | |
87 | * q->vlan becomes inaccessible. When the files gets closed, | |
88 | * macvtap_get_queue() fails. | |
20d29d7a | 89 | * |
02df55d2 AB |
90 | * There may still be references to the struct sock inside of the |
91 | * queue from outbound SKBs, but these never reference back to the | |
92 | * file or the dev. The data structure is freed through __sk_free | |
93 | * when both our references and any pending SKBs are gone. | |
20d29d7a | 94 | */ |
20d29d7a | 95 | |
815f236d | 96 | static int macvtap_enable_queue(struct net_device *dev, struct file *file, |
20d29d7a | 97 | struct macvtap_queue *q) |
815f236d JW |
98 | { |
99 | struct macvlan_dev *vlan = netdev_priv(dev); | |
100 | int err = -EINVAL; | |
101 | ||
441ac0fc | 102 | ASSERT_RTNL(); |
815f236d JW |
103 | |
104 | if (q->enabled) | |
105 | goto out; | |
106 | ||
107 | err = 0; | |
108 | rcu_assign_pointer(vlan->taps[vlan->numvtaps], q); | |
109 | q->queue_index = vlan->numvtaps; | |
110 | q->enabled = true; | |
111 | ||
112 | vlan->numvtaps++; | |
113 | out: | |
815f236d JW |
114 | return err; |
115 | } | |
116 | ||
40b8fe45 | 117 | /* Requires RTNL */ |
815f236d JW |
118 | static int macvtap_set_queue(struct net_device *dev, struct file *file, |
119 | struct macvtap_queue *q) | |
20d29d7a AB |
120 | { |
121 | struct macvlan_dev *vlan = netdev_priv(dev); | |
20d29d7a | 122 | |
815f236d | 123 | if (vlan->numqueues == MAX_MACVTAP_QUEUES) |
40b8fe45 | 124 | return -EBUSY; |
20d29d7a | 125 | |
02df55d2 | 126 | rcu_assign_pointer(q->vlan, vlan); |
376b1aab | 127 | rcu_assign_pointer(vlan->taps[vlan->numvtaps], q); |
02df55d2 | 128 | sock_hold(&q->sk); |
20d29d7a AB |
129 | |
130 | q->file = file; | |
376b1aab | 131 | q->queue_index = vlan->numvtaps; |
815f236d | 132 | q->enabled = true; |
02df55d2 | 133 | file->private_data = q; |
815f236d | 134 | list_add_tail(&q->next, &vlan->queue_list); |
20d29d7a | 135 | |
1565c7c1 | 136 | vlan->numvtaps++; |
815f236d | 137 | vlan->numqueues++; |
1565c7c1 | 138 | |
40b8fe45 | 139 | return 0; |
20d29d7a AB |
140 | } |
141 | ||
441ac0fc | 142 | static int macvtap_disable_queue(struct macvtap_queue *q) |
815f236d JW |
143 | { |
144 | struct macvlan_dev *vlan; | |
145 | struct macvtap_queue *nq; | |
146 | ||
441ac0fc | 147 | ASSERT_RTNL(); |
815f236d JW |
148 | if (!q->enabled) |
149 | return -EINVAL; | |
150 | ||
441ac0fc VY |
151 | vlan = rtnl_dereference(q->vlan); |
152 | ||
815f236d JW |
153 | if (vlan) { |
154 | int index = q->queue_index; | |
155 | BUG_ON(index >= vlan->numvtaps); | |
441ac0fc | 156 | nq = rtnl_dereference(vlan->taps[vlan->numvtaps - 1]); |
815f236d JW |
157 | nq->queue_index = index; |
158 | ||
159 | rcu_assign_pointer(vlan->taps[index], nq); | |
160 | RCU_INIT_POINTER(vlan->taps[vlan->numvtaps - 1], NULL); | |
161 | q->enabled = false; | |
162 | ||
163 | vlan->numvtaps--; | |
164 | } | |
165 | ||
166 | return 0; | |
167 | } | |
168 | ||
20d29d7a | 169 | /* |
02df55d2 AB |
170 | * The file owning the queue got closed, give up both |
171 | * the reference that the files holds as well as the | |
172 | * one from the macvlan_dev if that still exists. | |
20d29d7a AB |
173 | * |
174 | * Using the spinlock makes sure that we don't get | |
175 | * to the queue again after destroying it. | |
20d29d7a | 176 | */ |
02df55d2 | 177 | static void macvtap_put_queue(struct macvtap_queue *q) |
20d29d7a | 178 | { |
02df55d2 | 179 | struct macvlan_dev *vlan; |
20d29d7a | 180 | |
441ac0fc VY |
181 | rtnl_lock(); |
182 | vlan = rtnl_dereference(q->vlan); | |
183 | ||
02df55d2 | 184 | if (vlan) { |
815f236d | 185 | if (q->enabled) |
441ac0fc | 186 | BUG_ON(macvtap_disable_queue(q)); |
376b1aab | 187 | |
815f236d | 188 | vlan->numqueues--; |
2cfa5a04 | 189 | RCU_INIT_POINTER(q->vlan, NULL); |
02df55d2 | 190 | sock_put(&q->sk); |
815f236d | 191 | list_del_init(&q->next); |
20d29d7a AB |
192 | } |
193 | ||
441ac0fc | 194 | rtnl_unlock(); |
20d29d7a AB |
195 | |
196 | synchronize_rcu(); | |
197 | sock_put(&q->sk); | |
198 | } | |
199 | ||
200 | /* | |
1565c7c1 KK |
201 | * Select a queue based on the rxq of the device on which this packet |
202 | * arrived. If the incoming device is not mq, calculate a flow hash | |
203 | * to select a queue. If all fails, find the first available queue. | |
204 | * Cache vlan->numvtaps since it can become zero during the execution | |
205 | * of this function. | |
20d29d7a AB |
206 | */ |
207 | static struct macvtap_queue *macvtap_get_queue(struct net_device *dev, | |
208 | struct sk_buff *skb) | |
209 | { | |
210 | struct macvlan_dev *vlan = netdev_priv(dev); | |
1565c7c1 | 211 | struct macvtap_queue *tap = NULL; |
815f236d JW |
212 | /* Access to taps array is protected by rcu, but access to numvtaps |
213 | * isn't. Below we use it to lookup a queue, but treat it as a hint | |
214 | * and validate that the result isn't NULL - in case we are | |
215 | * racing against queue removal. | |
216 | */ | |
ed0483fa | 217 | int numvtaps = ACCESS_ONCE(vlan->numvtaps); |
1565c7c1 KK |
218 | __u32 rxq; |
219 | ||
220 | if (!numvtaps) | |
221 | goto out; | |
222 | ||
ef0002b5 | 223 | /* Check if we can use flow to select a queue */ |
3958afa1 | 224 | rxq = skb_get_hash(skb); |
ef0002b5 KK |
225 | if (rxq) { |
226 | tap = rcu_dereference(vlan->taps[rxq % numvtaps]); | |
376b1aab | 227 | goto out; |
ef0002b5 KK |
228 | } |
229 | ||
1565c7c1 KK |
230 | if (likely(skb_rx_queue_recorded(skb))) { |
231 | rxq = skb_get_rx_queue(skb); | |
20d29d7a | 232 | |
1565c7c1 KK |
233 | while (unlikely(rxq >= numvtaps)) |
234 | rxq -= numvtaps; | |
235 | ||
236 | tap = rcu_dereference(vlan->taps[rxq]); | |
376b1aab | 237 | goto out; |
1565c7c1 KK |
238 | } |
239 | ||
376b1aab | 240 | tap = rcu_dereference(vlan->taps[0]); |
1565c7c1 KK |
241 | out: |
242 | return tap; | |
20d29d7a AB |
243 | } |
244 | ||
02df55d2 AB |
245 | /* |
246 | * The net_device is going away, give up the reference | |
1565c7c1 KK |
247 | * that it holds on all queues and safely set the pointer |
248 | * from the queues to NULL. | |
02df55d2 | 249 | */ |
20d29d7a AB |
250 | static void macvtap_del_queues(struct net_device *dev) |
251 | { | |
252 | struct macvlan_dev *vlan = netdev_priv(dev); | |
815f236d | 253 | struct macvtap_queue *q, *tmp, *qlist[MAX_MACVTAP_QUEUES]; |
1565c7c1 | 254 | int i, j = 0; |
02df55d2 | 255 | |
441ac0fc | 256 | ASSERT_RTNL(); |
815f236d JW |
257 | list_for_each_entry_safe(q, tmp, &vlan->queue_list, next) { |
258 | list_del_init(&q->next); | |
376b1aab | 259 | qlist[j++] = q; |
376b1aab | 260 | RCU_INIT_POINTER(q->vlan, NULL); |
815f236d JW |
261 | if (q->enabled) |
262 | vlan->numvtaps--; | |
263 | vlan->numqueues--; | |
564517e8 | 264 | } |
815f236d JW |
265 | for (i = 0; i < vlan->numvtaps; i++) |
266 | RCU_INIT_POINTER(vlan->taps[i], NULL); | |
267 | BUG_ON(vlan->numvtaps); | |
268 | BUG_ON(vlan->numqueues); | |
99f34b38 EB |
269 | /* guarantee that any future macvtap_set_queue will fail */ |
270 | vlan->numvtaps = MAX_MACVTAP_QUEUES; | |
1565c7c1 KK |
271 | |
272 | for (--j; j >= 0; j--) | |
273 | sock_put(&qlist[j]->sk); | |
20d29d7a AB |
274 | } |
275 | ||
6acf54f1 | 276 | static rx_handler_result_t macvtap_handle_frame(struct sk_buff **pskb) |
20d29d7a | 277 | { |
6acf54f1 VY |
278 | struct sk_buff *skb = *pskb; |
279 | struct net_device *dev = skb->dev; | |
280 | struct macvlan_dev *vlan; | |
281 | struct macvtap_queue *q; | |
a567dd62 VY |
282 | netdev_features_t features = TAP_FEATURES; |
283 | ||
6acf54f1 VY |
284 | vlan = macvtap_get_vlan_rcu(dev); |
285 | if (!vlan) | |
286 | return RX_HANDLER_PASS; | |
287 | ||
288 | q = macvtap_get_queue(dev, skb); | |
20d29d7a | 289 | if (!q) |
6acf54f1 | 290 | return RX_HANDLER_PASS; |
8a35747a HX |
291 | |
292 | if (skb_queue_len(&q->sk.sk_receive_queue) >= dev->tx_queue_len) | |
293 | goto drop; | |
20d29d7a | 294 | |
6acf54f1 VY |
295 | skb_push(skb, ETH_HLEN); |
296 | ||
3e4f8b78 | 297 | /* Apply the forward feature mask so that we perform segmentation |
e5733321 VY |
298 | * according to users wishes. This only works if VNET_HDR is |
299 | * enabled. | |
3e4f8b78 | 300 | */ |
e5733321 VY |
301 | if (q->flags & IFF_VNET_HDR) |
302 | features |= vlan->tap_features; | |
04ffcb25 | 303 | if (netif_needs_gso(dev, skb, features)) { |
3e4f8b78 VY |
304 | struct sk_buff *segs = __skb_gso_segment(skb, features, false); |
305 | ||
306 | if (IS_ERR(segs)) | |
307 | goto drop; | |
308 | ||
309 | if (!segs) { | |
310 | skb_queue_tail(&q->sk.sk_receive_queue, skb); | |
311 | goto wake_up; | |
312 | } | |
313 | ||
314 | kfree_skb(skb); | |
315 | while (segs) { | |
316 | struct sk_buff *nskb = segs->next; | |
317 | ||
318 | segs->next = NULL; | |
319 | skb_queue_tail(&q->sk.sk_receive_queue, segs); | |
320 | segs = nskb; | |
321 | } | |
322 | } else { | |
cbdb0427 VY |
323 | /* If we receive a partial checksum and the tap side |
324 | * doesn't support checksum offload, compute the checksum. | |
325 | * Note: it doesn't matter which checksum feature to | |
326 | * check, we either support them all or none. | |
327 | */ | |
328 | if (skb->ip_summed == CHECKSUM_PARTIAL && | |
329 | !(features & NETIF_F_ALL_CSUM) && | |
330 | skb_checksum_help(skb)) | |
331 | goto drop; | |
3e4f8b78 VY |
332 | skb_queue_tail(&q->sk.sk_receive_queue, skb); |
333 | } | |
334 | ||
335 | wake_up: | |
4a4771a5 | 336 | wake_up_interruptible_poll(sk_sleep(&q->sk), POLLIN | POLLRDNORM | POLLRDBAND); |
6acf54f1 | 337 | return RX_HANDLER_CONSUMED; |
8a35747a HX |
338 | |
339 | drop: | |
6acf54f1 VY |
340 | /* Count errors/drops only here, thus don't care about args. */ |
341 | macvlan_count_rx(vlan, 0, 0, 0); | |
8a35747a | 342 | kfree_skb(skb); |
6acf54f1 | 343 | return RX_HANDLER_CONSUMED; |
20d29d7a AB |
344 | } |
345 | ||
e09eff7f EB |
346 | static int macvtap_get_minor(struct macvlan_dev *vlan) |
347 | { | |
348 | int retval = -ENOMEM; | |
e09eff7f EB |
349 | |
350 | mutex_lock(&minor_lock); | |
ec09ebc1 TH |
351 | retval = idr_alloc(&minor_idr, vlan, 1, MACVTAP_NUM_DEVS, GFP_KERNEL); |
352 | if (retval >= 0) { | |
353 | vlan->minor = retval; | |
354 | } else if (retval == -ENOSPC) { | |
e09eff7f EB |
355 | printk(KERN_ERR "too many macvtap devices\n"); |
356 | retval = -EINVAL; | |
e09eff7f | 357 | } |
e09eff7f | 358 | mutex_unlock(&minor_lock); |
ec09ebc1 | 359 | return retval < 0 ? retval : 0; |
e09eff7f EB |
360 | } |
361 | ||
362 | static void macvtap_free_minor(struct macvlan_dev *vlan) | |
363 | { | |
364 | mutex_lock(&minor_lock); | |
365 | if (vlan->minor) { | |
366 | idr_remove(&minor_idr, vlan->minor); | |
367 | vlan->minor = 0; | |
368 | } | |
369 | mutex_unlock(&minor_lock); | |
370 | } | |
371 | ||
372 | static struct net_device *dev_get_by_macvtap_minor(int minor) | |
373 | { | |
374 | struct net_device *dev = NULL; | |
375 | struct macvlan_dev *vlan; | |
376 | ||
377 | mutex_lock(&minor_lock); | |
378 | vlan = idr_find(&minor_idr, minor); | |
379 | if (vlan) { | |
380 | dev = vlan->dev; | |
381 | dev_hold(dev); | |
382 | } | |
383 | mutex_unlock(&minor_lock); | |
384 | return dev; | |
385 | } | |
386 | ||
20d29d7a AB |
387 | static int macvtap_newlink(struct net *src_net, |
388 | struct net_device *dev, | |
389 | struct nlattr *tb[], | |
390 | struct nlattr *data[]) | |
391 | { | |
815f236d | 392 | struct macvlan_dev *vlan = netdev_priv(dev); |
6acf54f1 VY |
393 | int err; |
394 | ||
815f236d JW |
395 | INIT_LIST_HEAD(&vlan->queue_list); |
396 | ||
2be5c767 VY |
397 | /* Since macvlan supports all offloads by default, make |
398 | * tap support all offloads also. | |
399 | */ | |
400 | vlan->tap_features = TUN_OFFLOADS; | |
401 | ||
6acf54f1 VY |
402 | err = netdev_rx_handler_register(dev, macvtap_handle_frame, vlan); |
403 | if (err) | |
404 | return err; | |
405 | ||
9bf1907f EB |
406 | /* Don't put anything that may fail after macvlan_common_newlink |
407 | * because we can't undo what it does. | |
408 | */ | |
2f6a1b66 | 409 | return macvlan_common_newlink(src_net, dev, tb, data); |
20d29d7a AB |
410 | } |
411 | ||
412 | static void macvtap_dellink(struct net_device *dev, | |
413 | struct list_head *head) | |
414 | { | |
6acf54f1 | 415 | netdev_rx_handler_unregister(dev); |
20d29d7a AB |
416 | macvtap_del_queues(dev); |
417 | macvlan_dellink(dev, head); | |
418 | } | |
419 | ||
8a35747a HX |
420 | static void macvtap_setup(struct net_device *dev) |
421 | { | |
422 | macvlan_common_setup(dev); | |
423 | dev->tx_queue_len = TUN_READQ_SIZE; | |
424 | } | |
425 | ||
20d29d7a AB |
426 | static struct rtnl_link_ops macvtap_link_ops __read_mostly = { |
427 | .kind = "macvtap", | |
8a35747a | 428 | .setup = macvtap_setup, |
20d29d7a AB |
429 | .newlink = macvtap_newlink, |
430 | .dellink = macvtap_dellink, | |
431 | }; | |
432 | ||
433 | ||
434 | static void macvtap_sock_write_space(struct sock *sk) | |
435 | { | |
43815482 ED |
436 | wait_queue_head_t *wqueue; |
437 | ||
20d29d7a AB |
438 | if (!sock_writeable(sk) || |
439 | !test_and_clear_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags)) | |
440 | return; | |
441 | ||
43815482 ED |
442 | wqueue = sk_sleep(sk); |
443 | if (wqueue && waitqueue_active(wqueue)) | |
444 | wake_up_interruptible_poll(wqueue, POLLOUT | POLLWRNORM | POLLWRBAND); | |
20d29d7a AB |
445 | } |
446 | ||
2259fef0 EB |
447 | static void macvtap_sock_destruct(struct sock *sk) |
448 | { | |
449 | skb_queue_purge(&sk->sk_receive_queue); | |
450 | } | |
451 | ||
20d29d7a AB |
452 | static int macvtap_open(struct inode *inode, struct file *file) |
453 | { | |
454 | struct net *net = current->nsproxy->net_ns; | |
40b8fe45 | 455 | struct net_device *dev; |
20d29d7a | 456 | struct macvtap_queue *q; |
40b8fe45 | 457 | int err = -ENODEV; |
20d29d7a | 458 | |
40b8fe45 VY |
459 | rtnl_lock(); |
460 | dev = dev_get_by_macvtap_minor(iminor(inode)); | |
20d29d7a AB |
461 | if (!dev) |
462 | goto out; | |
463 | ||
20d29d7a AB |
464 | err = -ENOMEM; |
465 | q = (struct macvtap_queue *)sk_alloc(net, AF_UNSPEC, GFP_KERNEL, | |
466 | &macvtap_proto); | |
467 | if (!q) | |
468 | goto out; | |
469 | ||
d9a90a31 | 470 | RCU_INIT_POINTER(q->sock.wq, &q->wq); |
43815482 | 471 | init_waitqueue_head(&q->wq.wait); |
20d29d7a AB |
472 | q->sock.type = SOCK_RAW; |
473 | q->sock.state = SS_CONNECTED; | |
501c774c AB |
474 | q->sock.file = file; |
475 | q->sock.ops = &macvtap_socket_ops; | |
20d29d7a | 476 | sock_init_data(&q->sock, &q->sk); |
20d29d7a | 477 | q->sk.sk_write_space = macvtap_sock_write_space; |
2259fef0 | 478 | q->sk.sk_destruct = macvtap_sock_destruct; |
b9fb9ee0 | 479 | q->flags = IFF_VNET_HDR | IFF_NO_PI | IFF_TAP; |
55afbd08 | 480 | q->vnet_hdr_sz = sizeof(struct virtio_net_hdr); |
20d29d7a | 481 | |
97bc3633 SM |
482 | /* |
483 | * so far only KVM virtio_net uses macvtap, enable zero copy between | |
484 | * guest kernel and host kernel when lower device supports zerocopy | |
047af9cf EB |
485 | * |
486 | * The macvlan supports zerocopy iff the lower device supports zero | |
487 | * copy so we don't have to look at the lower device directly. | |
97bc3633 | 488 | */ |
047af9cf EB |
489 | if ((dev->features & NETIF_F_HIGHDMA) && (dev->features & NETIF_F_SG)) |
490 | sock_set_flag(&q->sk, SOCK_ZEROCOPY); | |
97bc3633 | 491 | |
20d29d7a AB |
492 | err = macvtap_set_queue(dev, file, q); |
493 | if (err) | |
494 | sock_put(&q->sk); | |
495 | ||
496 | out: | |
497 | if (dev) | |
498 | dev_put(dev); | |
499 | ||
40b8fe45 | 500 | rtnl_unlock(); |
20d29d7a AB |
501 | return err; |
502 | } | |
503 | ||
504 | static int macvtap_release(struct inode *inode, struct file *file) | |
505 | { | |
02df55d2 AB |
506 | struct macvtap_queue *q = file->private_data; |
507 | macvtap_put_queue(q); | |
20d29d7a AB |
508 | return 0; |
509 | } | |
510 | ||
511 | static unsigned int macvtap_poll(struct file *file, poll_table * wait) | |
512 | { | |
02df55d2 | 513 | struct macvtap_queue *q = file->private_data; |
20d29d7a AB |
514 | unsigned int mask = POLLERR; |
515 | ||
516 | if (!q) | |
517 | goto out; | |
518 | ||
519 | mask = 0; | |
43815482 | 520 | poll_wait(file, &q->wq.wait, wait); |
20d29d7a AB |
521 | |
522 | if (!skb_queue_empty(&q->sk.sk_receive_queue)) | |
523 | mask |= POLLIN | POLLRDNORM; | |
524 | ||
525 | if (sock_writeable(&q->sk) || | |
526 | (!test_and_set_bit(SOCK_ASYNC_NOSPACE, &q->sock.flags) && | |
527 | sock_writeable(&q->sk))) | |
528 | mask |= POLLOUT | POLLWRNORM; | |
529 | ||
530 | out: | |
20d29d7a AB |
531 | return mask; |
532 | } | |
533 | ||
b9fb9ee0 AB |
534 | static inline struct sk_buff *macvtap_alloc_skb(struct sock *sk, size_t prepad, |
535 | size_t len, size_t linear, | |
536 | int noblock, int *err) | |
537 | { | |
538 | struct sk_buff *skb; | |
539 | ||
540 | /* Under a page? Don't bother with paged skb. */ | |
541 | if (prepad + len < PAGE_SIZE || !linear) | |
542 | linear = len; | |
543 | ||
544 | skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock, | |
28d64271 | 545 | err, 0); |
b9fb9ee0 AB |
546 | if (!skb) |
547 | return NULL; | |
548 | ||
549 | skb_reserve(skb, prepad); | |
550 | skb_put(skb, linear); | |
551 | skb->data_len = len - linear; | |
552 | skb->len += len - linear; | |
553 | ||
554 | return skb; | |
555 | } | |
556 | ||
557 | /* | |
558 | * macvtap_skb_from_vnet_hdr and macvtap_skb_to_vnet_hdr should | |
559 | * be shared with the tun/tap driver. | |
560 | */ | |
561 | static int macvtap_skb_from_vnet_hdr(struct sk_buff *skb, | |
562 | struct virtio_net_hdr *vnet_hdr) | |
563 | { | |
564 | unsigned short gso_type = 0; | |
565 | if (vnet_hdr->gso_type != VIRTIO_NET_HDR_GSO_NONE) { | |
566 | switch (vnet_hdr->gso_type & ~VIRTIO_NET_HDR_GSO_ECN) { | |
567 | case VIRTIO_NET_HDR_GSO_TCPV4: | |
568 | gso_type = SKB_GSO_TCPV4; | |
569 | break; | |
570 | case VIRTIO_NET_HDR_GSO_TCPV6: | |
571 | gso_type = SKB_GSO_TCPV6; | |
572 | break; | |
573 | case VIRTIO_NET_HDR_GSO_UDP: | |
3d0ad094 BH |
574 | pr_warn_once("macvtap: %s: using disabled UFO feature; please fix this program\n", |
575 | current->comm); | |
b9fb9ee0 | 576 | gso_type = SKB_GSO_UDP; |
5188cd44 BH |
577 | if (skb->protocol == htons(ETH_P_IPV6)) |
578 | ipv6_proxy_select_ident(skb); | |
b9fb9ee0 AB |
579 | break; |
580 | default: | |
581 | return -EINVAL; | |
582 | } | |
583 | ||
584 | if (vnet_hdr->gso_type & VIRTIO_NET_HDR_GSO_ECN) | |
585 | gso_type |= SKB_GSO_TCP_ECN; | |
586 | ||
587 | if (vnet_hdr->gso_size == 0) | |
588 | return -EINVAL; | |
589 | } | |
590 | ||
591 | if (vnet_hdr->flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) { | |
592 | if (!skb_partial_csum_set(skb, vnet_hdr->csum_start, | |
593 | vnet_hdr->csum_offset)) | |
594 | return -EINVAL; | |
595 | } | |
596 | ||
597 | if (vnet_hdr->gso_type != VIRTIO_NET_HDR_GSO_NONE) { | |
598 | skb_shinfo(skb)->gso_size = vnet_hdr->gso_size; | |
c9af6db4 | 599 | skb_shinfo(skb)->gso_type = gso_type; |
b9fb9ee0 AB |
600 | |
601 | /* Header must be checked, and gso_segs computed. */ | |
602 | skb_shinfo(skb)->gso_type |= SKB_GSO_DODGY; | |
603 | skb_shinfo(skb)->gso_segs = 0; | |
604 | } | |
605 | return 0; | |
606 | } | |
607 | ||
359d44d7 | 608 | static void macvtap_skb_to_vnet_hdr(const struct sk_buff *skb, |
b9fb9ee0 AB |
609 | struct virtio_net_hdr *vnet_hdr) |
610 | { | |
611 | memset(vnet_hdr, 0, sizeof(*vnet_hdr)); | |
612 | ||
613 | if (skb_is_gso(skb)) { | |
614 | struct skb_shared_info *sinfo = skb_shinfo(skb); | |
615 | ||
616 | /* This is a hint as to how much should be linear. */ | |
617 | vnet_hdr->hdr_len = skb_headlen(skb); | |
618 | vnet_hdr->gso_size = sinfo->gso_size; | |
619 | if (sinfo->gso_type & SKB_GSO_TCPV4) | |
620 | vnet_hdr->gso_type = VIRTIO_NET_HDR_GSO_TCPV4; | |
621 | else if (sinfo->gso_type & SKB_GSO_TCPV6) | |
622 | vnet_hdr->gso_type = VIRTIO_NET_HDR_GSO_TCPV6; | |
b9fb9ee0 AB |
623 | else |
624 | BUG(); | |
625 | if (sinfo->gso_type & SKB_GSO_TCP_ECN) | |
626 | vnet_hdr->gso_type |= VIRTIO_NET_HDR_GSO_ECN; | |
627 | } else | |
628 | vnet_hdr->gso_type = VIRTIO_NET_HDR_GSO_NONE; | |
629 | ||
630 | if (skb->ip_summed == CHECKSUM_PARTIAL) { | |
631 | vnet_hdr->flags = VIRTIO_NET_HDR_F_NEEDS_CSUM; | |
55508d60 | 632 | vnet_hdr->csum_start = skb_checksum_start_offset(skb); |
3ce9b20f HX |
633 | if (vlan_tx_tag_present(skb)) |
634 | vnet_hdr->csum_start += VLAN_HLEN; | |
b9fb9ee0 | 635 | vnet_hdr->csum_offset = skb->csum_offset; |
10a8d94a JW |
636 | } else if (skb->ip_summed == CHECKSUM_UNNECESSARY) { |
637 | vnet_hdr->flags = VIRTIO_NET_HDR_F_DATA_VALID; | |
b9fb9ee0 | 638 | } /* else everything is zero */ |
b9fb9ee0 AB |
639 | } |
640 | ||
20d29d7a | 641 | /* Get packet from user space buffer */ |
97bc3633 | 642 | static ssize_t macvtap_get_user(struct macvtap_queue *q, struct msghdr *m, |
f5ff53b4 | 643 | struct iov_iter *from, int noblock) |
20d29d7a | 644 | { |
16a3fa28 | 645 | int good_linear = SKB_MAX_HEAD(NET_IP_ALIGN); |
20d29d7a | 646 | struct sk_buff *skb; |
02df55d2 | 647 | struct macvlan_dev *vlan; |
f5ff53b4 | 648 | unsigned long total_len = iov_iter_count(from); |
97bc3633 | 649 | unsigned long len = total_len; |
20d29d7a | 650 | int err; |
b9fb9ee0 AB |
651 | struct virtio_net_hdr vnet_hdr = { 0 }; |
652 | int vnet_hdr_len = 0; | |
b92946e2 | 653 | int copylen = 0; |
97bc3633 | 654 | bool zerocopy = false; |
61d46bf9 | 655 | size_t linear; |
f5ff53b4 | 656 | ssize_t n; |
b9fb9ee0 AB |
657 | |
658 | if (q->flags & IFF_VNET_HDR) { | |
55afbd08 | 659 | vnet_hdr_len = q->vnet_hdr_sz; |
b9fb9ee0 AB |
660 | |
661 | err = -EINVAL; | |
ce3c8692 | 662 | if (len < vnet_hdr_len) |
b9fb9ee0 | 663 | goto err; |
ce3c8692 | 664 | len -= vnet_hdr_len; |
b9fb9ee0 | 665 | |
f5ff53b4 AV |
666 | err = -EFAULT; |
667 | n = copy_from_iter(&vnet_hdr, sizeof(vnet_hdr), from); | |
668 | if (n != sizeof(vnet_hdr)) | |
b9fb9ee0 | 669 | goto err; |
f5ff53b4 | 670 | iov_iter_advance(from, vnet_hdr_len - sizeof(vnet_hdr)); |
b9fb9ee0 AB |
671 | if ((vnet_hdr.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) && |
672 | vnet_hdr.csum_start + vnet_hdr.csum_offset + 2 > | |
673 | vnet_hdr.hdr_len) | |
674 | vnet_hdr.hdr_len = vnet_hdr.csum_start + | |
675 | vnet_hdr.csum_offset + 2; | |
676 | err = -EINVAL; | |
677 | if (vnet_hdr.hdr_len > len) | |
678 | goto err; | |
679 | } | |
20d29d7a | 680 | |
b9fb9ee0 | 681 | err = -EINVAL; |
20d29d7a | 682 | if (unlikely(len < ETH_HLEN)) |
b9fb9ee0 | 683 | goto err; |
20d29d7a | 684 | |
ece793fc | 685 | if (m && m->msg_control && sock_flag(&q->sk, SOCK_ZEROCOPY)) { |
f5ff53b4 AV |
686 | struct iov_iter i; |
687 | ||
ece793fc | 688 | copylen = vnet_hdr.hdr_len ? vnet_hdr.hdr_len : GOODCOPY_LEN; |
16a3fa28 JW |
689 | if (copylen > good_linear) |
690 | copylen = good_linear; | |
61d46bf9 | 691 | linear = copylen; |
f5ff53b4 AV |
692 | i = *from; |
693 | iov_iter_advance(&i, copylen); | |
694 | if (iov_iter_npages(&i, INT_MAX) <= MAX_SKB_FRAGS) | |
ece793fc JW |
695 | zerocopy = true; |
696 | } | |
697 | ||
698 | if (!zerocopy) { | |
97bc3633 | 699 | copylen = len; |
16a3fa28 JW |
700 | if (vnet_hdr.hdr_len > good_linear) |
701 | linear = good_linear; | |
702 | else | |
703 | linear = vnet_hdr.hdr_len; | |
61d46bf9 | 704 | } |
97bc3633 SM |
705 | |
706 | skb = macvtap_alloc_skb(&q->sk, NET_IP_ALIGN, copylen, | |
61d46bf9 | 707 | linear, noblock, &err); |
02df55d2 AB |
708 | if (!skb) |
709 | goto err; | |
20d29d7a | 710 | |
01d6657b | 711 | if (zerocopy) |
f5ff53b4 | 712 | err = zerocopy_sg_from_iter(skb, from); |
ece793fc | 713 | else { |
f5ff53b4 | 714 | err = skb_copy_datagram_from_iter(skb, 0, from, len); |
ece793fc JW |
715 | if (!err && m && m->msg_control) { |
716 | struct ubuf_info *uarg = m->msg_control; | |
717 | uarg->callback(uarg, false); | |
718 | } | |
719 | } | |
720 | ||
02df55d2 | 721 | if (err) |
b9fb9ee0 | 722 | goto err_kfree; |
20d29d7a AB |
723 | |
724 | skb_set_network_header(skb, ETH_HLEN); | |
b9fb9ee0 AB |
725 | skb_reset_mac_header(skb); |
726 | skb->protocol = eth_hdr(skb)->h_proto; | |
727 | ||
728 | if (vnet_hdr_len) { | |
729 | err = macvtap_skb_from_vnet_hdr(skb, &vnet_hdr); | |
730 | if (err) | |
731 | goto err_kfree; | |
732 | } | |
733 | ||
40893fd0 | 734 | skb_probe_transport_header(skb, ETH_HLEN); |
9b4d669b | 735 | |
ac4e4af1 VY |
736 | rcu_read_lock(); |
737 | vlan = rcu_dereference(q->vlan); | |
97bc3633 | 738 | /* copy skb_ubuf_info for callback when skb has no error */ |
01d6657b | 739 | if (zerocopy) { |
97bc3633 | 740 | skb_shinfo(skb)->destructor_arg = m->msg_control; |
01d6657b | 741 | skb_shinfo(skb)->tx_flags |= SKBTX_DEV_ZEROCOPY; |
c9af6db4 | 742 | skb_shinfo(skb)->tx_flags |= SKBTX_SHARED_FRAG; |
01d6657b | 743 | } |
29d79196 | 744 | if (vlan) { |
6acf54f1 VY |
745 | skb->dev = vlan->dev; |
746 | dev_queue_xmit(skb); | |
29d79196 | 747 | } else { |
02df55d2 | 748 | kfree_skb(skb); |
29d79196 | 749 | } |
ac4e4af1 | 750 | rcu_read_unlock(); |
20d29d7a | 751 | |
97bc3633 | 752 | return total_len; |
02df55d2 | 753 | |
b9fb9ee0 AB |
754 | err_kfree: |
755 | kfree_skb(skb); | |
756 | ||
02df55d2 | 757 | err: |
ac4e4af1 VY |
758 | rcu_read_lock(); |
759 | vlan = rcu_dereference(q->vlan); | |
02df55d2 | 760 | if (vlan) |
cd3e22b7 | 761 | this_cpu_inc(vlan->pcpu_stats->tx_dropped); |
ac4e4af1 | 762 | rcu_read_unlock(); |
02df55d2 | 763 | |
02df55d2 | 764 | return err; |
20d29d7a AB |
765 | } |
766 | ||
f5ff53b4 | 767 | static ssize_t macvtap_write_iter(struct kiocb *iocb, struct iov_iter *from) |
20d29d7a AB |
768 | { |
769 | struct file *file = iocb->ki_filp; | |
02df55d2 | 770 | struct macvtap_queue *q = file->private_data; |
20d29d7a | 771 | |
f5ff53b4 | 772 | return macvtap_get_user(q, NULL, from, file->f_flags & O_NONBLOCK); |
20d29d7a AB |
773 | } |
774 | ||
775 | /* Put packet to the user space buffer */ | |
776 | static ssize_t macvtap_put_user(struct macvtap_queue *q, | |
777 | const struct sk_buff *skb, | |
6c36d2e2 | 778 | struct iov_iter *iter) |
20d29d7a | 779 | { |
20d29d7a | 780 | int ret; |
b9fb9ee0 | 781 | int vnet_hdr_len = 0; |
f09e2249 | 782 | int vlan_offset = 0; |
6c36d2e2 | 783 | int total; |
b9fb9ee0 AB |
784 | |
785 | if (q->flags & IFF_VNET_HDR) { | |
786 | struct virtio_net_hdr vnet_hdr; | |
55afbd08 | 787 | vnet_hdr_len = q->vnet_hdr_sz; |
6c36d2e2 | 788 | if (iov_iter_count(iter) < vnet_hdr_len) |
b9fb9ee0 AB |
789 | return -EINVAL; |
790 | ||
359d44d7 | 791 | macvtap_skb_to_vnet_hdr(skb, &vnet_hdr); |
b9fb9ee0 | 792 | |
6c36d2e2 HX |
793 | if (copy_to_iter(&vnet_hdr, sizeof(vnet_hdr), iter) != |
794 | sizeof(vnet_hdr)) | |
b9fb9ee0 | 795 | return -EFAULT; |
7cc76f51 JW |
796 | |
797 | iov_iter_advance(iter, vnet_hdr_len - sizeof(vnet_hdr)); | |
b9fb9ee0 | 798 | } |
6c36d2e2 | 799 | total = vnet_hdr_len; |
ce232ce0 | 800 | total += skb->len; |
f09e2249 | 801 | |
6c36d2e2 | 802 | if (vlan_tx_tag_present(skb)) { |
f09e2249 BG |
803 | struct { |
804 | __be16 h_vlan_proto; | |
805 | __be16 h_vlan_TCI; | |
806 | } veth; | |
0fbe0d47 | 807 | veth.h_vlan_proto = skb->vlan_proto; |
f09e2249 BG |
808 | veth.h_vlan_TCI = htons(vlan_tx_tag_get(skb)); |
809 | ||
810 | vlan_offset = offsetof(struct vlan_ethhdr, h_vlan_proto); | |
ce232ce0 | 811 | total += VLAN_HLEN; |
f09e2249 | 812 | |
6c36d2e2 HX |
813 | ret = skb_copy_datagram_iter(skb, 0, iter, vlan_offset); |
814 | if (ret || !iov_iter_count(iter)) | |
f09e2249 BG |
815 | goto done; |
816 | ||
6c36d2e2 HX |
817 | ret = copy_to_iter(&veth, sizeof(veth), iter); |
818 | if (ret != sizeof(veth) || !iov_iter_count(iter)) | |
f09e2249 BG |
819 | goto done; |
820 | } | |
20d29d7a | 821 | |
6c36d2e2 HX |
822 | ret = skb_copy_datagram_iter(skb, vlan_offset, iter, |
823 | skb->len - vlan_offset); | |
20d29d7a | 824 | |
f09e2249 | 825 | done: |
ce232ce0 | 826 | return ret ? ret : total; |
20d29d7a AB |
827 | } |
828 | ||
55ec8e25 | 829 | static ssize_t macvtap_do_read(struct macvtap_queue *q, |
3af0bfe5 | 830 | struct iov_iter *to, |
501c774c | 831 | int noblock) |
20d29d7a | 832 | { |
ccf7e72b | 833 | DEFINE_WAIT(wait); |
20d29d7a | 834 | struct sk_buff *skb; |
501c774c | 835 | ssize_t ret = 0; |
20d29d7a | 836 | |
3af0bfe5 AV |
837 | if (!iov_iter_count(to)) |
838 | return 0; | |
839 | ||
840 | while (1) { | |
89cee917 JW |
841 | if (!noblock) |
842 | prepare_to_wait(sk_sleep(&q->sk), &wait, | |
843 | TASK_INTERRUPTIBLE); | |
20d29d7a AB |
844 | |
845 | /* Read frames from the queue */ | |
846 | skb = skb_dequeue(&q->sk.sk_receive_queue); | |
3af0bfe5 AV |
847 | if (skb) |
848 | break; | |
849 | if (noblock) { | |
850 | ret = -EAGAIN; | |
851 | break; | |
20d29d7a | 852 | } |
3af0bfe5 AV |
853 | if (signal_pending(current)) { |
854 | ret = -ERESTARTSYS; | |
855 | break; | |
856 | } | |
857 | /* Nothing to read, let's sleep */ | |
858 | schedule(); | |
859 | } | |
860 | if (skb) { | |
861 | ret = macvtap_put_user(q, skb, to); | |
f51a5e82 JW |
862 | if (unlikely(ret < 0)) |
863 | kfree_skb(skb); | |
864 | else | |
865 | consume_skb(skb); | |
20d29d7a | 866 | } |
89cee917 JW |
867 | if (!noblock) |
868 | finish_wait(sk_sleep(&q->sk), &wait); | |
501c774c AB |
869 | return ret; |
870 | } | |
871 | ||
3af0bfe5 | 872 | static ssize_t macvtap_read_iter(struct kiocb *iocb, struct iov_iter *to) |
501c774c AB |
873 | { |
874 | struct file *file = iocb->ki_filp; | |
875 | struct macvtap_queue *q = file->private_data; | |
3af0bfe5 | 876 | ssize_t len = iov_iter_count(to), ret; |
501c774c | 877 | |
3af0bfe5 | 878 | ret = macvtap_do_read(q, to, file->f_flags & O_NONBLOCK); |
ce232ce0 | 879 | ret = min_t(ssize_t, ret, len); |
e6ebc7f1 ZYW |
880 | if (ret > 0) |
881 | iocb->ki_pos = ret; | |
20d29d7a AB |
882 | return ret; |
883 | } | |
884 | ||
8f475a31 JW |
885 | static struct macvlan_dev *macvtap_get_vlan(struct macvtap_queue *q) |
886 | { | |
887 | struct macvlan_dev *vlan; | |
888 | ||
441ac0fc VY |
889 | ASSERT_RTNL(); |
890 | vlan = rtnl_dereference(q->vlan); | |
8f475a31 JW |
891 | if (vlan) |
892 | dev_hold(vlan->dev); | |
8f475a31 JW |
893 | |
894 | return vlan; | |
895 | } | |
896 | ||
897 | static void macvtap_put_vlan(struct macvlan_dev *vlan) | |
898 | { | |
899 | dev_put(vlan->dev); | |
900 | } | |
901 | ||
815f236d JW |
902 | static int macvtap_ioctl_set_queue(struct file *file, unsigned int flags) |
903 | { | |
904 | struct macvtap_queue *q = file->private_data; | |
905 | struct macvlan_dev *vlan; | |
906 | int ret; | |
907 | ||
908 | vlan = macvtap_get_vlan(q); | |
909 | if (!vlan) | |
910 | return -EINVAL; | |
911 | ||
912 | if (flags & IFF_ATTACH_QUEUE) | |
913 | ret = macvtap_enable_queue(vlan->dev, file, q); | |
914 | else if (flags & IFF_DETACH_QUEUE) | |
915 | ret = macvtap_disable_queue(q); | |
f57855a5 JW |
916 | else |
917 | ret = -EINVAL; | |
815f236d JW |
918 | |
919 | macvtap_put_vlan(vlan); | |
920 | return ret; | |
921 | } | |
922 | ||
2be5c767 VY |
923 | static int set_offload(struct macvtap_queue *q, unsigned long arg) |
924 | { | |
925 | struct macvlan_dev *vlan; | |
926 | netdev_features_t features; | |
927 | netdev_features_t feature_mask = 0; | |
928 | ||
929 | vlan = rtnl_dereference(q->vlan); | |
930 | if (!vlan) | |
931 | return -ENOLINK; | |
932 | ||
933 | features = vlan->dev->features; | |
934 | ||
935 | if (arg & TUN_F_CSUM) { | |
936 | feature_mask = NETIF_F_HW_CSUM; | |
937 | ||
938 | if (arg & (TUN_F_TSO4 | TUN_F_TSO6)) { | |
939 | if (arg & TUN_F_TSO_ECN) | |
940 | feature_mask |= NETIF_F_TSO_ECN; | |
941 | if (arg & TUN_F_TSO4) | |
942 | feature_mask |= NETIF_F_TSO; | |
943 | if (arg & TUN_F_TSO6) | |
944 | feature_mask |= NETIF_F_TSO6; | |
945 | } | |
2be5c767 VY |
946 | } |
947 | ||
948 | /* tun/tap driver inverts the usage for TSO offloads, where | |
949 | * setting the TSO bit means that the userspace wants to | |
950 | * accept TSO frames and turning it off means that user space | |
951 | * does not support TSO. | |
952 | * For macvtap, we have to invert it to mean the same thing. | |
953 | * When user space turns off TSO, we turn off GSO/LRO so that | |
954 | * user-space will not receive TSO frames. | |
955 | */ | |
3d0ad094 | 956 | if (feature_mask & (NETIF_F_TSO | NETIF_F_TSO6)) |
2be5c767 VY |
957 | features |= RX_OFFLOADS; |
958 | else | |
959 | features &= ~RX_OFFLOADS; | |
960 | ||
961 | /* tap_features are the same as features on tun/tap and | |
962 | * reflect user expectations. | |
963 | */ | |
a567dd62 | 964 | vlan->tap_features = feature_mask; |
2be5c767 VY |
965 | vlan->set_features = features; |
966 | netdev_update_features(vlan->dev); | |
967 | ||
968 | return 0; | |
969 | } | |
970 | ||
20d29d7a AB |
971 | /* |
972 | * provide compatibility with generic tun/tap interface | |
973 | */ | |
974 | static long macvtap_ioctl(struct file *file, unsigned int cmd, | |
975 | unsigned long arg) | |
976 | { | |
02df55d2 AB |
977 | struct macvtap_queue *q = file->private_data; |
978 | struct macvlan_dev *vlan; | |
20d29d7a AB |
979 | void __user *argp = (void __user *)arg; |
980 | struct ifreq __user *ifr = argp; | |
981 | unsigned int __user *up = argp; | |
982 | unsigned int u; | |
55afbd08 MT |
983 | int __user *sp = argp; |
984 | int s; | |
02df55d2 | 985 | int ret; |
20d29d7a AB |
986 | |
987 | switch (cmd) { | |
988 | case TUNSETIFF: | |
989 | /* ignore the name, just look at flags */ | |
990 | if (get_user(u, &ifr->ifr_flags)) | |
991 | return -EFAULT; | |
b9fb9ee0 AB |
992 | |
993 | ret = 0; | |
815f236d JW |
994 | if ((u & ~(IFF_VNET_HDR | IFF_MULTI_QUEUE)) != |
995 | (IFF_NO_PI | IFF_TAP)) | |
b9fb9ee0 AB |
996 | ret = -EINVAL; |
997 | else | |
998 | q->flags = u; | |
999 | ||
1000 | return ret; | |
20d29d7a AB |
1001 | |
1002 | case TUNGETIFF: | |
441ac0fc | 1003 | rtnl_lock(); |
8f475a31 | 1004 | vlan = macvtap_get_vlan(q); |
441ac0fc VY |
1005 | if (!vlan) { |
1006 | rtnl_unlock(); | |
20d29d7a | 1007 | return -ENOLINK; |
441ac0fc | 1008 | } |
20d29d7a | 1009 | |
02df55d2 | 1010 | ret = 0; |
13707f9e | 1011 | if (copy_to_user(&ifr->ifr_name, vlan->dev->name, IFNAMSIZ) || |
b9fb9ee0 | 1012 | put_user(q->flags, &ifr->ifr_flags)) |
02df55d2 | 1013 | ret = -EFAULT; |
8f475a31 | 1014 | macvtap_put_vlan(vlan); |
441ac0fc | 1015 | rtnl_unlock(); |
02df55d2 | 1016 | return ret; |
20d29d7a | 1017 | |
815f236d JW |
1018 | case TUNSETQUEUE: |
1019 | if (get_user(u, &ifr->ifr_flags)) | |
1020 | return -EFAULT; | |
441ac0fc VY |
1021 | rtnl_lock(); |
1022 | ret = macvtap_ioctl_set_queue(file, u); | |
1023 | rtnl_unlock(); | |
82a19eb8 | 1024 | return ret; |
815f236d | 1025 | |
20d29d7a | 1026 | case TUNGETFEATURES: |
df09b36f JW |
1027 | if (put_user(IFF_TAP | IFF_NO_PI | IFF_VNET_HDR | |
1028 | IFF_MULTI_QUEUE, up)) | |
20d29d7a AB |
1029 | return -EFAULT; |
1030 | return 0; | |
1031 | ||
1032 | case TUNSETSNDBUF: | |
1033 | if (get_user(u, up)) | |
1034 | return -EFAULT; | |
1035 | ||
20d29d7a | 1036 | q->sk.sk_sndbuf = u; |
20d29d7a AB |
1037 | return 0; |
1038 | ||
55afbd08 MT |
1039 | case TUNGETVNETHDRSZ: |
1040 | s = q->vnet_hdr_sz; | |
1041 | if (put_user(s, sp)) | |
1042 | return -EFAULT; | |
1043 | return 0; | |
1044 | ||
1045 | case TUNSETVNETHDRSZ: | |
1046 | if (get_user(s, sp)) | |
1047 | return -EFAULT; | |
1048 | if (s < (int)sizeof(struct virtio_net_hdr)) | |
1049 | return -EINVAL; | |
1050 | ||
1051 | q->vnet_hdr_sz = s; | |
1052 | return 0; | |
1053 | ||
20d29d7a AB |
1054 | case TUNSETOFFLOAD: |
1055 | /* let the user check for future flags */ | |
1056 | if (arg & ~(TUN_F_CSUM | TUN_F_TSO4 | TUN_F_TSO6 | | |
3d0ad094 | 1057 | TUN_F_TSO_ECN)) |
20d29d7a AB |
1058 | return -EINVAL; |
1059 | ||
2be5c767 VY |
1060 | rtnl_lock(); |
1061 | ret = set_offload(q, arg); | |
1062 | rtnl_unlock(); | |
1063 | return ret; | |
20d29d7a AB |
1064 | |
1065 | default: | |
1066 | return -EINVAL; | |
1067 | } | |
1068 | } | |
1069 | ||
1070 | #ifdef CONFIG_COMPAT | |
1071 | static long macvtap_compat_ioctl(struct file *file, unsigned int cmd, | |
1072 | unsigned long arg) | |
1073 | { | |
1074 | return macvtap_ioctl(file, cmd, (unsigned long)compat_ptr(arg)); | |
1075 | } | |
1076 | #endif | |
1077 | ||
1078 | static const struct file_operations macvtap_fops = { | |
1079 | .owner = THIS_MODULE, | |
1080 | .open = macvtap_open, | |
1081 | .release = macvtap_release, | |
3af0bfe5 | 1082 | .read = new_sync_read, |
f5ff53b4 | 1083 | .write = new_sync_write, |
3af0bfe5 | 1084 | .read_iter = macvtap_read_iter, |
f5ff53b4 | 1085 | .write_iter = macvtap_write_iter, |
20d29d7a AB |
1086 | .poll = macvtap_poll, |
1087 | .llseek = no_llseek, | |
1088 | .unlocked_ioctl = macvtap_ioctl, | |
1089 | #ifdef CONFIG_COMPAT | |
1090 | .compat_ioctl = macvtap_compat_ioctl, | |
1091 | #endif | |
1092 | }; | |
1093 | ||
501c774c AB |
1094 | static int macvtap_sendmsg(struct kiocb *iocb, struct socket *sock, |
1095 | struct msghdr *m, size_t total_len) | |
1096 | { | |
1097 | struct macvtap_queue *q = container_of(sock, struct macvtap_queue, sock); | |
f5ff53b4 AV |
1098 | struct iov_iter from; |
1099 | iov_iter_init(&from, WRITE, m->msg_iov, m->msg_iovlen, total_len); | |
1100 | return macvtap_get_user(q, m, &from, m->msg_flags & MSG_DONTWAIT); | |
501c774c AB |
1101 | } |
1102 | ||
1103 | static int macvtap_recvmsg(struct kiocb *iocb, struct socket *sock, | |
1104 | struct msghdr *m, size_t total_len, | |
1105 | int flags) | |
1106 | { | |
1107 | struct macvtap_queue *q = container_of(sock, struct macvtap_queue, sock); | |
3af0bfe5 | 1108 | struct iov_iter to; |
501c774c AB |
1109 | int ret; |
1110 | if (flags & ~(MSG_DONTWAIT|MSG_TRUNC)) | |
1111 | return -EINVAL; | |
3af0bfe5 AV |
1112 | iov_iter_init(&to, READ, m->msg_iov, m->msg_iovlen, total_len); |
1113 | ret = macvtap_do_read(q, &to, flags & MSG_DONTWAIT); | |
de2aa476 DM |
1114 | if (ret > total_len) { |
1115 | m->msg_flags |= MSG_TRUNC; | |
1116 | ret = flags & MSG_TRUNC ? ret : total_len; | |
1117 | } | |
501c774c AB |
1118 | return ret; |
1119 | } | |
1120 | ||
1121 | /* Ops structure to mimic raw sockets with tun */ | |
1122 | static const struct proto_ops macvtap_socket_ops = { | |
1123 | .sendmsg = macvtap_sendmsg, | |
1124 | .recvmsg = macvtap_recvmsg, | |
1125 | }; | |
1126 | ||
1127 | /* Get an underlying socket object from tun file. Returns error unless file is | |
1128 | * attached to a device. The returned object works like a packet socket, it | |
1129 | * can be used for sock_sendmsg/sock_recvmsg. The caller is responsible for | |
1130 | * holding a reference to the file for as long as the socket is in use. */ | |
1131 | struct socket *macvtap_get_socket(struct file *file) | |
1132 | { | |
1133 | struct macvtap_queue *q; | |
1134 | if (file->f_op != &macvtap_fops) | |
1135 | return ERR_PTR(-EINVAL); | |
1136 | q = file->private_data; | |
1137 | if (!q) | |
1138 | return ERR_PTR(-EBADFD); | |
1139 | return &q->sock; | |
1140 | } | |
1141 | EXPORT_SYMBOL_GPL(macvtap_get_socket); | |
1142 | ||
9bf1907f EB |
1143 | static int macvtap_device_event(struct notifier_block *unused, |
1144 | unsigned long event, void *ptr) | |
1145 | { | |
351638e7 | 1146 | struct net_device *dev = netdev_notifier_info_to_dev(ptr); |
e09eff7f | 1147 | struct macvlan_dev *vlan; |
9bf1907f EB |
1148 | struct device *classdev; |
1149 | dev_t devt; | |
e09eff7f | 1150 | int err; |
9bf1907f EB |
1151 | |
1152 | if (dev->rtnl_link_ops != &macvtap_link_ops) | |
1153 | return NOTIFY_DONE; | |
1154 | ||
e09eff7f | 1155 | vlan = netdev_priv(dev); |
9bf1907f EB |
1156 | |
1157 | switch (event) { | |
1158 | case NETDEV_REGISTER: | |
1159 | /* Create the device node here after the network device has | |
1160 | * been registered but before register_netdevice has | |
1161 | * finished running. | |
1162 | */ | |
e09eff7f EB |
1163 | err = macvtap_get_minor(vlan); |
1164 | if (err) | |
1165 | return notifier_from_errno(err); | |
1166 | ||
1167 | devt = MKDEV(MAJOR(macvtap_major), vlan->minor); | |
9bf1907f EB |
1168 | classdev = device_create(macvtap_class, &dev->dev, devt, |
1169 | dev, "tap%d", dev->ifindex); | |
e09eff7f EB |
1170 | if (IS_ERR(classdev)) { |
1171 | macvtap_free_minor(vlan); | |
9bf1907f | 1172 | return notifier_from_errno(PTR_ERR(classdev)); |
e09eff7f | 1173 | } |
9bf1907f EB |
1174 | break; |
1175 | case NETDEV_UNREGISTER: | |
e09eff7f | 1176 | devt = MKDEV(MAJOR(macvtap_major), vlan->minor); |
9bf1907f | 1177 | device_destroy(macvtap_class, devt); |
e09eff7f | 1178 | macvtap_free_minor(vlan); |
9bf1907f EB |
1179 | break; |
1180 | } | |
1181 | ||
1182 | return NOTIFY_DONE; | |
1183 | } | |
1184 | ||
1185 | static struct notifier_block macvtap_notifier_block __read_mostly = { | |
1186 | .notifier_call = macvtap_device_event, | |
1187 | }; | |
1188 | ||
20d29d7a AB |
1189 | static int macvtap_init(void) |
1190 | { | |
1191 | int err; | |
1192 | ||
1193 | err = alloc_chrdev_region(&macvtap_major, 0, | |
1194 | MACVTAP_NUM_DEVS, "macvtap"); | |
1195 | if (err) | |
1196 | goto out1; | |
1197 | ||
1198 | cdev_init(&macvtap_cdev, &macvtap_fops); | |
1199 | err = cdev_add(&macvtap_cdev, macvtap_major, MACVTAP_NUM_DEVS); | |
1200 | if (err) | |
1201 | goto out2; | |
1202 | ||
1203 | macvtap_class = class_create(THIS_MODULE, "macvtap"); | |
1204 | if (IS_ERR(macvtap_class)) { | |
1205 | err = PTR_ERR(macvtap_class); | |
1206 | goto out3; | |
1207 | } | |
1208 | ||
9bf1907f | 1209 | err = register_netdevice_notifier(&macvtap_notifier_block); |
20d29d7a AB |
1210 | if (err) |
1211 | goto out4; | |
1212 | ||
9bf1907f EB |
1213 | err = macvlan_link_register(&macvtap_link_ops); |
1214 | if (err) | |
1215 | goto out5; | |
1216 | ||
20d29d7a AB |
1217 | return 0; |
1218 | ||
9bf1907f EB |
1219 | out5: |
1220 | unregister_netdevice_notifier(&macvtap_notifier_block); | |
20d29d7a AB |
1221 | out4: |
1222 | class_unregister(macvtap_class); | |
1223 | out3: | |
1224 | cdev_del(&macvtap_cdev); | |
1225 | out2: | |
1226 | unregister_chrdev_region(macvtap_major, MACVTAP_NUM_DEVS); | |
1227 | out1: | |
1228 | return err; | |
1229 | } | |
1230 | module_init(macvtap_init); | |
1231 | ||
1232 | static void macvtap_exit(void) | |
1233 | { | |
1234 | rtnl_link_unregister(&macvtap_link_ops); | |
9bf1907f | 1235 | unregister_netdevice_notifier(&macvtap_notifier_block); |
20d29d7a AB |
1236 | class_unregister(macvtap_class); |
1237 | cdev_del(&macvtap_cdev); | |
1238 | unregister_chrdev_region(macvtap_major, MACVTAP_NUM_DEVS); | |
1239 | } | |
1240 | module_exit(macvtap_exit); | |
1241 | ||
1242 | MODULE_ALIAS_RTNL_LINK("macvtap"); | |
1243 | MODULE_AUTHOR("Arnd Bergmann <arnd@arndb.de>"); | |
1244 | MODULE_LICENSE("GPL"); |