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 AB |
17 | #include <linux/fs.h> |
18 | ||
5188cd44 | 19 | #include <net/ipv6.h> |
20d29d7a AB |
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 | ||
6ae7feb3 MT |
48 | #define MACVTAP_FEATURES (IFF_VNET_HDR | IFF_VNET_LE | IFF_MULTI_QUEUE) |
49 | ||
50 | static inline u16 macvtap16_to_cpu(struct macvtap_queue *q, __virtio16 val) | |
51 | { | |
52 | return __virtio16_to_cpu(q->flags & IFF_VNET_LE, val); | |
53 | } | |
54 | ||
55 | static inline __virtio16 cpu_to_macvtap16(struct macvtap_queue *q, u16 val) | |
56 | { | |
57 | return __cpu_to_virtio16(q->flags & IFF_VNET_LE, val); | |
58 | } | |
59 | ||
20d29d7a AB |
60 | static struct proto macvtap_proto = { |
61 | .name = "macvtap", | |
62 | .owner = THIS_MODULE, | |
63 | .obj_size = sizeof (struct macvtap_queue), | |
64 | }; | |
65 | ||
66 | /* | |
e09eff7f | 67 | * Variables for dealing with macvtaps device numbers. |
20d29d7a | 68 | */ |
1ebed71a | 69 | static dev_t macvtap_major; |
e09eff7f EB |
70 | #define MACVTAP_NUM_DEVS (1U << MINORBITS) |
71 | static DEFINE_MUTEX(minor_lock); | |
72 | static DEFINE_IDR(minor_idr); | |
73 | ||
97bc3633 | 74 | #define GOODCOPY_LEN 128 |
20d29d7a AB |
75 | static struct class *macvtap_class; |
76 | static struct cdev macvtap_cdev; | |
77 | ||
501c774c AB |
78 | static const struct proto_ops macvtap_socket_ops; |
79 | ||
2be5c767 | 80 | #define TUN_OFFLOADS (NETIF_F_HW_CSUM | NETIF_F_TSO_ECN | NETIF_F_TSO | \ |
3d0ad094 | 81 | NETIF_F_TSO6) |
2be5c767 | 82 | #define RX_OFFLOADS (NETIF_F_GRO | NETIF_F_LRO) |
a567dd62 VY |
83 | #define TAP_FEATURES (NETIF_F_GSO | NETIF_F_SG) |
84 | ||
6acf54f1 VY |
85 | static struct macvlan_dev *macvtap_get_vlan_rcu(const struct net_device *dev) |
86 | { | |
87 | return rcu_dereference(dev->rx_handler_data); | |
88 | } | |
89 | ||
20d29d7a AB |
90 | /* |
91 | * RCU usage: | |
02df55d2 AB |
92 | * The macvtap_queue and the macvlan_dev are loosely coupled, the |
93 | * pointers from one to the other can only be read while rcu_read_lock | |
441ac0fc | 94 | * or rtnl is held. |
20d29d7a | 95 | * |
02df55d2 AB |
96 | * Both the file and the macvlan_dev hold a reference on the macvtap_queue |
97 | * through sock_hold(&q->sk). When the macvlan_dev goes away first, | |
98 | * q->vlan becomes inaccessible. When the files gets closed, | |
99 | * macvtap_get_queue() fails. | |
20d29d7a | 100 | * |
02df55d2 AB |
101 | * There may still be references to the struct sock inside of the |
102 | * queue from outbound SKBs, but these never reference back to the | |
103 | * file or the dev. The data structure is freed through __sk_free | |
104 | * when both our references and any pending SKBs are gone. | |
20d29d7a | 105 | */ |
20d29d7a | 106 | |
815f236d | 107 | static int macvtap_enable_queue(struct net_device *dev, struct file *file, |
20d29d7a | 108 | struct macvtap_queue *q) |
815f236d JW |
109 | { |
110 | struct macvlan_dev *vlan = netdev_priv(dev); | |
111 | int err = -EINVAL; | |
112 | ||
441ac0fc | 113 | ASSERT_RTNL(); |
815f236d JW |
114 | |
115 | if (q->enabled) | |
116 | goto out; | |
117 | ||
118 | err = 0; | |
119 | rcu_assign_pointer(vlan->taps[vlan->numvtaps], q); | |
120 | q->queue_index = vlan->numvtaps; | |
121 | q->enabled = true; | |
122 | ||
123 | vlan->numvtaps++; | |
124 | out: | |
815f236d JW |
125 | return err; |
126 | } | |
127 | ||
40b8fe45 | 128 | /* Requires RTNL */ |
815f236d JW |
129 | static int macvtap_set_queue(struct net_device *dev, struct file *file, |
130 | struct macvtap_queue *q) | |
20d29d7a AB |
131 | { |
132 | struct macvlan_dev *vlan = netdev_priv(dev); | |
20d29d7a | 133 | |
815f236d | 134 | if (vlan->numqueues == MAX_MACVTAP_QUEUES) |
40b8fe45 | 135 | return -EBUSY; |
20d29d7a | 136 | |
02df55d2 | 137 | rcu_assign_pointer(q->vlan, vlan); |
376b1aab | 138 | rcu_assign_pointer(vlan->taps[vlan->numvtaps], q); |
02df55d2 | 139 | sock_hold(&q->sk); |
20d29d7a AB |
140 | |
141 | q->file = file; | |
376b1aab | 142 | q->queue_index = vlan->numvtaps; |
815f236d | 143 | q->enabled = true; |
02df55d2 | 144 | file->private_data = q; |
815f236d | 145 | list_add_tail(&q->next, &vlan->queue_list); |
20d29d7a | 146 | |
1565c7c1 | 147 | vlan->numvtaps++; |
815f236d | 148 | vlan->numqueues++; |
1565c7c1 | 149 | |
40b8fe45 | 150 | return 0; |
20d29d7a AB |
151 | } |
152 | ||
441ac0fc | 153 | static int macvtap_disable_queue(struct macvtap_queue *q) |
815f236d JW |
154 | { |
155 | struct macvlan_dev *vlan; | |
156 | struct macvtap_queue *nq; | |
157 | ||
441ac0fc | 158 | ASSERT_RTNL(); |
815f236d JW |
159 | if (!q->enabled) |
160 | return -EINVAL; | |
161 | ||
441ac0fc VY |
162 | vlan = rtnl_dereference(q->vlan); |
163 | ||
815f236d JW |
164 | if (vlan) { |
165 | int index = q->queue_index; | |
166 | BUG_ON(index >= vlan->numvtaps); | |
441ac0fc | 167 | nq = rtnl_dereference(vlan->taps[vlan->numvtaps - 1]); |
815f236d JW |
168 | nq->queue_index = index; |
169 | ||
170 | rcu_assign_pointer(vlan->taps[index], nq); | |
171 | RCU_INIT_POINTER(vlan->taps[vlan->numvtaps - 1], NULL); | |
172 | q->enabled = false; | |
173 | ||
174 | vlan->numvtaps--; | |
175 | } | |
176 | ||
177 | return 0; | |
178 | } | |
179 | ||
20d29d7a | 180 | /* |
02df55d2 AB |
181 | * The file owning the queue got closed, give up both |
182 | * the reference that the files holds as well as the | |
183 | * one from the macvlan_dev if that still exists. | |
20d29d7a AB |
184 | * |
185 | * Using the spinlock makes sure that we don't get | |
186 | * to the queue again after destroying it. | |
20d29d7a | 187 | */ |
02df55d2 | 188 | static void macvtap_put_queue(struct macvtap_queue *q) |
20d29d7a | 189 | { |
02df55d2 | 190 | struct macvlan_dev *vlan; |
20d29d7a | 191 | |
441ac0fc VY |
192 | rtnl_lock(); |
193 | vlan = rtnl_dereference(q->vlan); | |
194 | ||
02df55d2 | 195 | if (vlan) { |
815f236d | 196 | if (q->enabled) |
441ac0fc | 197 | BUG_ON(macvtap_disable_queue(q)); |
376b1aab | 198 | |
815f236d | 199 | vlan->numqueues--; |
2cfa5a04 | 200 | RCU_INIT_POINTER(q->vlan, NULL); |
02df55d2 | 201 | sock_put(&q->sk); |
815f236d | 202 | list_del_init(&q->next); |
20d29d7a AB |
203 | } |
204 | ||
441ac0fc | 205 | rtnl_unlock(); |
20d29d7a AB |
206 | |
207 | synchronize_rcu(); | |
208 | sock_put(&q->sk); | |
209 | } | |
210 | ||
211 | /* | |
1565c7c1 KK |
212 | * Select a queue based on the rxq of the device on which this packet |
213 | * arrived. If the incoming device is not mq, calculate a flow hash | |
214 | * to select a queue. If all fails, find the first available queue. | |
215 | * Cache vlan->numvtaps since it can become zero during the execution | |
216 | * of this function. | |
20d29d7a AB |
217 | */ |
218 | static struct macvtap_queue *macvtap_get_queue(struct net_device *dev, | |
219 | struct sk_buff *skb) | |
220 | { | |
221 | struct macvlan_dev *vlan = netdev_priv(dev); | |
1565c7c1 | 222 | struct macvtap_queue *tap = NULL; |
815f236d JW |
223 | /* Access to taps array is protected by rcu, but access to numvtaps |
224 | * isn't. Below we use it to lookup a queue, but treat it as a hint | |
225 | * and validate that the result isn't NULL - in case we are | |
226 | * racing against queue removal. | |
227 | */ | |
ed0483fa | 228 | int numvtaps = ACCESS_ONCE(vlan->numvtaps); |
1565c7c1 KK |
229 | __u32 rxq; |
230 | ||
231 | if (!numvtaps) | |
232 | goto out; | |
233 | ||
ef0002b5 | 234 | /* Check if we can use flow to select a queue */ |
3958afa1 | 235 | rxq = skb_get_hash(skb); |
ef0002b5 KK |
236 | if (rxq) { |
237 | tap = rcu_dereference(vlan->taps[rxq % numvtaps]); | |
376b1aab | 238 | goto out; |
ef0002b5 KK |
239 | } |
240 | ||
1565c7c1 KK |
241 | if (likely(skb_rx_queue_recorded(skb))) { |
242 | rxq = skb_get_rx_queue(skb); | |
20d29d7a | 243 | |
1565c7c1 KK |
244 | while (unlikely(rxq >= numvtaps)) |
245 | rxq -= numvtaps; | |
246 | ||
247 | tap = rcu_dereference(vlan->taps[rxq]); | |
376b1aab | 248 | goto out; |
1565c7c1 KK |
249 | } |
250 | ||
376b1aab | 251 | tap = rcu_dereference(vlan->taps[0]); |
1565c7c1 KK |
252 | out: |
253 | return tap; | |
20d29d7a AB |
254 | } |
255 | ||
02df55d2 AB |
256 | /* |
257 | * The net_device is going away, give up the reference | |
1565c7c1 KK |
258 | * that it holds on all queues and safely set the pointer |
259 | * from the queues to NULL. | |
02df55d2 | 260 | */ |
20d29d7a AB |
261 | static void macvtap_del_queues(struct net_device *dev) |
262 | { | |
263 | struct macvlan_dev *vlan = netdev_priv(dev); | |
815f236d | 264 | struct macvtap_queue *q, *tmp, *qlist[MAX_MACVTAP_QUEUES]; |
1565c7c1 | 265 | int i, j = 0; |
02df55d2 | 266 | |
441ac0fc | 267 | ASSERT_RTNL(); |
815f236d JW |
268 | list_for_each_entry_safe(q, tmp, &vlan->queue_list, next) { |
269 | list_del_init(&q->next); | |
376b1aab | 270 | qlist[j++] = q; |
376b1aab | 271 | RCU_INIT_POINTER(q->vlan, NULL); |
815f236d JW |
272 | if (q->enabled) |
273 | vlan->numvtaps--; | |
274 | vlan->numqueues--; | |
564517e8 | 275 | } |
815f236d JW |
276 | for (i = 0; i < vlan->numvtaps; i++) |
277 | RCU_INIT_POINTER(vlan->taps[i], NULL); | |
278 | BUG_ON(vlan->numvtaps); | |
279 | BUG_ON(vlan->numqueues); | |
99f34b38 EB |
280 | /* guarantee that any future macvtap_set_queue will fail */ |
281 | vlan->numvtaps = MAX_MACVTAP_QUEUES; | |
1565c7c1 KK |
282 | |
283 | for (--j; j >= 0; j--) | |
284 | sock_put(&qlist[j]->sk); | |
20d29d7a AB |
285 | } |
286 | ||
6acf54f1 | 287 | static rx_handler_result_t macvtap_handle_frame(struct sk_buff **pskb) |
20d29d7a | 288 | { |
6acf54f1 VY |
289 | struct sk_buff *skb = *pskb; |
290 | struct net_device *dev = skb->dev; | |
291 | struct macvlan_dev *vlan; | |
292 | struct macvtap_queue *q; | |
a567dd62 VY |
293 | netdev_features_t features = TAP_FEATURES; |
294 | ||
6acf54f1 VY |
295 | vlan = macvtap_get_vlan_rcu(dev); |
296 | if (!vlan) | |
297 | return RX_HANDLER_PASS; | |
298 | ||
299 | q = macvtap_get_queue(dev, skb); | |
20d29d7a | 300 | if (!q) |
6acf54f1 | 301 | return RX_HANDLER_PASS; |
8a35747a HX |
302 | |
303 | if (skb_queue_len(&q->sk.sk_receive_queue) >= dev->tx_queue_len) | |
304 | goto drop; | |
20d29d7a | 305 | |
6acf54f1 VY |
306 | skb_push(skb, ETH_HLEN); |
307 | ||
3e4f8b78 | 308 | /* Apply the forward feature mask so that we perform segmentation |
e5733321 VY |
309 | * according to users wishes. This only works if VNET_HDR is |
310 | * enabled. | |
3e4f8b78 | 311 | */ |
e5733321 VY |
312 | if (q->flags & IFF_VNET_HDR) |
313 | features |= vlan->tap_features; | |
04ffcb25 | 314 | if (netif_needs_gso(dev, skb, features)) { |
3e4f8b78 VY |
315 | struct sk_buff *segs = __skb_gso_segment(skb, features, false); |
316 | ||
317 | if (IS_ERR(segs)) | |
318 | goto drop; | |
319 | ||
320 | if (!segs) { | |
321 | skb_queue_tail(&q->sk.sk_receive_queue, skb); | |
322 | goto wake_up; | |
323 | } | |
324 | ||
325 | kfree_skb(skb); | |
326 | while (segs) { | |
327 | struct sk_buff *nskb = segs->next; | |
328 | ||
329 | segs->next = NULL; | |
330 | skb_queue_tail(&q->sk.sk_receive_queue, segs); | |
331 | segs = nskb; | |
332 | } | |
333 | } else { | |
cbdb0427 VY |
334 | /* If we receive a partial checksum and the tap side |
335 | * doesn't support checksum offload, compute the checksum. | |
336 | * Note: it doesn't matter which checksum feature to | |
337 | * check, we either support them all or none. | |
338 | */ | |
339 | if (skb->ip_summed == CHECKSUM_PARTIAL && | |
340 | !(features & NETIF_F_ALL_CSUM) && | |
341 | skb_checksum_help(skb)) | |
342 | goto drop; | |
3e4f8b78 VY |
343 | skb_queue_tail(&q->sk.sk_receive_queue, skb); |
344 | } | |
345 | ||
346 | wake_up: | |
4a4771a5 | 347 | wake_up_interruptible_poll(sk_sleep(&q->sk), POLLIN | POLLRDNORM | POLLRDBAND); |
6acf54f1 | 348 | return RX_HANDLER_CONSUMED; |
8a35747a HX |
349 | |
350 | drop: | |
6acf54f1 VY |
351 | /* Count errors/drops only here, thus don't care about args. */ |
352 | macvlan_count_rx(vlan, 0, 0, 0); | |
8a35747a | 353 | kfree_skb(skb); |
6acf54f1 | 354 | return RX_HANDLER_CONSUMED; |
20d29d7a AB |
355 | } |
356 | ||
e09eff7f EB |
357 | static int macvtap_get_minor(struct macvlan_dev *vlan) |
358 | { | |
359 | int retval = -ENOMEM; | |
e09eff7f EB |
360 | |
361 | mutex_lock(&minor_lock); | |
ec09ebc1 TH |
362 | retval = idr_alloc(&minor_idr, vlan, 1, MACVTAP_NUM_DEVS, GFP_KERNEL); |
363 | if (retval >= 0) { | |
364 | vlan->minor = retval; | |
365 | } else if (retval == -ENOSPC) { | |
e09eff7f EB |
366 | printk(KERN_ERR "too many macvtap devices\n"); |
367 | retval = -EINVAL; | |
e09eff7f | 368 | } |
e09eff7f | 369 | mutex_unlock(&minor_lock); |
ec09ebc1 | 370 | return retval < 0 ? retval : 0; |
e09eff7f EB |
371 | } |
372 | ||
373 | static void macvtap_free_minor(struct macvlan_dev *vlan) | |
374 | { | |
375 | mutex_lock(&minor_lock); | |
376 | if (vlan->minor) { | |
377 | idr_remove(&minor_idr, vlan->minor); | |
378 | vlan->minor = 0; | |
379 | } | |
380 | mutex_unlock(&minor_lock); | |
381 | } | |
382 | ||
383 | static struct net_device *dev_get_by_macvtap_minor(int minor) | |
384 | { | |
385 | struct net_device *dev = NULL; | |
386 | struct macvlan_dev *vlan; | |
387 | ||
388 | mutex_lock(&minor_lock); | |
389 | vlan = idr_find(&minor_idr, minor); | |
390 | if (vlan) { | |
391 | dev = vlan->dev; | |
392 | dev_hold(dev); | |
393 | } | |
394 | mutex_unlock(&minor_lock); | |
395 | return dev; | |
396 | } | |
397 | ||
20d29d7a AB |
398 | static int macvtap_newlink(struct net *src_net, |
399 | struct net_device *dev, | |
400 | struct nlattr *tb[], | |
401 | struct nlattr *data[]) | |
402 | { | |
815f236d | 403 | struct macvlan_dev *vlan = netdev_priv(dev); |
6acf54f1 VY |
404 | int err; |
405 | ||
815f236d JW |
406 | INIT_LIST_HEAD(&vlan->queue_list); |
407 | ||
2be5c767 VY |
408 | /* Since macvlan supports all offloads by default, make |
409 | * tap support all offloads also. | |
410 | */ | |
411 | vlan->tap_features = TUN_OFFLOADS; | |
412 | ||
6acf54f1 VY |
413 | err = netdev_rx_handler_register(dev, macvtap_handle_frame, vlan); |
414 | if (err) | |
415 | return err; | |
416 | ||
9bf1907f EB |
417 | /* Don't put anything that may fail after macvlan_common_newlink |
418 | * because we can't undo what it does. | |
419 | */ | |
2f6a1b66 | 420 | return macvlan_common_newlink(src_net, dev, tb, data); |
20d29d7a AB |
421 | } |
422 | ||
423 | static void macvtap_dellink(struct net_device *dev, | |
424 | struct list_head *head) | |
425 | { | |
6acf54f1 | 426 | netdev_rx_handler_unregister(dev); |
20d29d7a AB |
427 | macvtap_del_queues(dev); |
428 | macvlan_dellink(dev, head); | |
429 | } | |
430 | ||
8a35747a HX |
431 | static void macvtap_setup(struct net_device *dev) |
432 | { | |
433 | macvlan_common_setup(dev); | |
434 | dev->tx_queue_len = TUN_READQ_SIZE; | |
435 | } | |
436 | ||
20d29d7a AB |
437 | static struct rtnl_link_ops macvtap_link_ops __read_mostly = { |
438 | .kind = "macvtap", | |
8a35747a | 439 | .setup = macvtap_setup, |
20d29d7a AB |
440 | .newlink = macvtap_newlink, |
441 | .dellink = macvtap_dellink, | |
442 | }; | |
443 | ||
444 | ||
445 | static void macvtap_sock_write_space(struct sock *sk) | |
446 | { | |
43815482 ED |
447 | wait_queue_head_t *wqueue; |
448 | ||
20d29d7a AB |
449 | if (!sock_writeable(sk) || |
450 | !test_and_clear_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags)) | |
451 | return; | |
452 | ||
43815482 ED |
453 | wqueue = sk_sleep(sk); |
454 | if (wqueue && waitqueue_active(wqueue)) | |
455 | wake_up_interruptible_poll(wqueue, POLLOUT | POLLWRNORM | POLLWRBAND); | |
20d29d7a AB |
456 | } |
457 | ||
2259fef0 EB |
458 | static void macvtap_sock_destruct(struct sock *sk) |
459 | { | |
460 | skb_queue_purge(&sk->sk_receive_queue); | |
461 | } | |
462 | ||
20d29d7a AB |
463 | static int macvtap_open(struct inode *inode, struct file *file) |
464 | { | |
465 | struct net *net = current->nsproxy->net_ns; | |
40b8fe45 | 466 | struct net_device *dev; |
20d29d7a | 467 | struct macvtap_queue *q; |
40b8fe45 | 468 | int err = -ENODEV; |
20d29d7a | 469 | |
40b8fe45 VY |
470 | rtnl_lock(); |
471 | dev = dev_get_by_macvtap_minor(iminor(inode)); | |
20d29d7a AB |
472 | if (!dev) |
473 | goto out; | |
474 | ||
20d29d7a AB |
475 | err = -ENOMEM; |
476 | q = (struct macvtap_queue *)sk_alloc(net, AF_UNSPEC, GFP_KERNEL, | |
477 | &macvtap_proto); | |
478 | if (!q) | |
479 | goto out; | |
480 | ||
d9a90a31 | 481 | RCU_INIT_POINTER(q->sock.wq, &q->wq); |
43815482 | 482 | init_waitqueue_head(&q->wq.wait); |
20d29d7a AB |
483 | q->sock.type = SOCK_RAW; |
484 | q->sock.state = SS_CONNECTED; | |
501c774c AB |
485 | q->sock.file = file; |
486 | q->sock.ops = &macvtap_socket_ops; | |
20d29d7a | 487 | sock_init_data(&q->sock, &q->sk); |
20d29d7a | 488 | q->sk.sk_write_space = macvtap_sock_write_space; |
2259fef0 | 489 | q->sk.sk_destruct = macvtap_sock_destruct; |
b9fb9ee0 | 490 | q->flags = IFF_VNET_HDR | IFF_NO_PI | IFF_TAP; |
55afbd08 | 491 | q->vnet_hdr_sz = sizeof(struct virtio_net_hdr); |
20d29d7a | 492 | |
97bc3633 SM |
493 | /* |
494 | * so far only KVM virtio_net uses macvtap, enable zero copy between | |
495 | * guest kernel and host kernel when lower device supports zerocopy | |
047af9cf EB |
496 | * |
497 | * The macvlan supports zerocopy iff the lower device supports zero | |
498 | * copy so we don't have to look at the lower device directly. | |
97bc3633 | 499 | */ |
047af9cf EB |
500 | if ((dev->features & NETIF_F_HIGHDMA) && (dev->features & NETIF_F_SG)) |
501 | sock_set_flag(&q->sk, SOCK_ZEROCOPY); | |
97bc3633 | 502 | |
20d29d7a AB |
503 | err = macvtap_set_queue(dev, file, q); |
504 | if (err) | |
505 | sock_put(&q->sk); | |
506 | ||
507 | out: | |
508 | if (dev) | |
509 | dev_put(dev); | |
510 | ||
40b8fe45 | 511 | rtnl_unlock(); |
20d29d7a AB |
512 | return err; |
513 | } | |
514 | ||
515 | static int macvtap_release(struct inode *inode, struct file *file) | |
516 | { | |
02df55d2 AB |
517 | struct macvtap_queue *q = file->private_data; |
518 | macvtap_put_queue(q); | |
20d29d7a AB |
519 | return 0; |
520 | } | |
521 | ||
522 | static unsigned int macvtap_poll(struct file *file, poll_table * wait) | |
523 | { | |
02df55d2 | 524 | struct macvtap_queue *q = file->private_data; |
20d29d7a AB |
525 | unsigned int mask = POLLERR; |
526 | ||
527 | if (!q) | |
528 | goto out; | |
529 | ||
530 | mask = 0; | |
43815482 | 531 | poll_wait(file, &q->wq.wait, wait); |
20d29d7a AB |
532 | |
533 | if (!skb_queue_empty(&q->sk.sk_receive_queue)) | |
534 | mask |= POLLIN | POLLRDNORM; | |
535 | ||
536 | if (sock_writeable(&q->sk) || | |
537 | (!test_and_set_bit(SOCK_ASYNC_NOSPACE, &q->sock.flags) && | |
538 | sock_writeable(&q->sk))) | |
539 | mask |= POLLOUT | POLLWRNORM; | |
540 | ||
541 | out: | |
20d29d7a AB |
542 | return mask; |
543 | } | |
544 | ||
b9fb9ee0 AB |
545 | static inline struct sk_buff *macvtap_alloc_skb(struct sock *sk, size_t prepad, |
546 | size_t len, size_t linear, | |
547 | int noblock, int *err) | |
548 | { | |
549 | struct sk_buff *skb; | |
550 | ||
551 | /* Under a page? Don't bother with paged skb. */ | |
552 | if (prepad + len < PAGE_SIZE || !linear) | |
553 | linear = len; | |
554 | ||
555 | skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock, | |
28d64271 | 556 | err, 0); |
b9fb9ee0 AB |
557 | if (!skb) |
558 | return NULL; | |
559 | ||
560 | skb_reserve(skb, prepad); | |
561 | skb_put(skb, linear); | |
562 | skb->data_len = len - linear; | |
563 | skb->len += len - linear; | |
564 | ||
565 | return skb; | |
566 | } | |
567 | ||
568 | /* | |
569 | * macvtap_skb_from_vnet_hdr and macvtap_skb_to_vnet_hdr should | |
570 | * be shared with the tun/tap driver. | |
571 | */ | |
6ae7feb3 MT |
572 | static int macvtap_skb_from_vnet_hdr(struct macvtap_queue *q, |
573 | struct sk_buff *skb, | |
b9fb9ee0 AB |
574 | struct virtio_net_hdr *vnet_hdr) |
575 | { | |
576 | unsigned short gso_type = 0; | |
577 | if (vnet_hdr->gso_type != VIRTIO_NET_HDR_GSO_NONE) { | |
578 | switch (vnet_hdr->gso_type & ~VIRTIO_NET_HDR_GSO_ECN) { | |
579 | case VIRTIO_NET_HDR_GSO_TCPV4: | |
580 | gso_type = SKB_GSO_TCPV4; | |
581 | break; | |
582 | case VIRTIO_NET_HDR_GSO_TCPV6: | |
583 | gso_type = SKB_GSO_TCPV6; | |
584 | break; | |
585 | case VIRTIO_NET_HDR_GSO_UDP: | |
3d0ad094 BH |
586 | pr_warn_once("macvtap: %s: using disabled UFO feature; please fix this program\n", |
587 | current->comm); | |
b9fb9ee0 | 588 | gso_type = SKB_GSO_UDP; |
5188cd44 BH |
589 | if (skb->protocol == htons(ETH_P_IPV6)) |
590 | ipv6_proxy_select_ident(skb); | |
b9fb9ee0 AB |
591 | break; |
592 | default: | |
593 | return -EINVAL; | |
594 | } | |
595 | ||
596 | if (vnet_hdr->gso_type & VIRTIO_NET_HDR_GSO_ECN) | |
597 | gso_type |= SKB_GSO_TCP_ECN; | |
598 | ||
599 | if (vnet_hdr->gso_size == 0) | |
600 | return -EINVAL; | |
601 | } | |
602 | ||
603 | if (vnet_hdr->flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) { | |
6ae7feb3 MT |
604 | if (!skb_partial_csum_set(skb, macvtap16_to_cpu(q, vnet_hdr->csum_start), |
605 | macvtap16_to_cpu(q, vnet_hdr->csum_offset))) | |
b9fb9ee0 AB |
606 | return -EINVAL; |
607 | } | |
608 | ||
609 | if (vnet_hdr->gso_type != VIRTIO_NET_HDR_GSO_NONE) { | |
6ae7feb3 | 610 | skb_shinfo(skb)->gso_size = macvtap16_to_cpu(q, vnet_hdr->gso_size); |
c9af6db4 | 611 | skb_shinfo(skb)->gso_type = gso_type; |
b9fb9ee0 AB |
612 | |
613 | /* Header must be checked, and gso_segs computed. */ | |
614 | skb_shinfo(skb)->gso_type |= SKB_GSO_DODGY; | |
615 | skb_shinfo(skb)->gso_segs = 0; | |
616 | } | |
617 | return 0; | |
618 | } | |
619 | ||
6ae7feb3 MT |
620 | static void macvtap_skb_to_vnet_hdr(struct macvtap_queue *q, |
621 | const struct sk_buff *skb, | |
622 | struct virtio_net_hdr *vnet_hdr) | |
b9fb9ee0 AB |
623 | { |
624 | memset(vnet_hdr, 0, sizeof(*vnet_hdr)); | |
625 | ||
626 | if (skb_is_gso(skb)) { | |
627 | struct skb_shared_info *sinfo = skb_shinfo(skb); | |
628 | ||
629 | /* This is a hint as to how much should be linear. */ | |
6ae7feb3 MT |
630 | vnet_hdr->hdr_len = cpu_to_macvtap16(q, skb_headlen(skb)); |
631 | vnet_hdr->gso_size = cpu_to_macvtap16(q, sinfo->gso_size); | |
b9fb9ee0 AB |
632 | if (sinfo->gso_type & SKB_GSO_TCPV4) |
633 | vnet_hdr->gso_type = VIRTIO_NET_HDR_GSO_TCPV4; | |
634 | else if (sinfo->gso_type & SKB_GSO_TCPV6) | |
635 | vnet_hdr->gso_type = VIRTIO_NET_HDR_GSO_TCPV6; | |
b9fb9ee0 AB |
636 | else |
637 | BUG(); | |
638 | if (sinfo->gso_type & SKB_GSO_TCP_ECN) | |
639 | vnet_hdr->gso_type |= VIRTIO_NET_HDR_GSO_ECN; | |
640 | } else | |
641 | vnet_hdr->gso_type = VIRTIO_NET_HDR_GSO_NONE; | |
642 | ||
643 | if (skb->ip_summed == CHECKSUM_PARTIAL) { | |
644 | vnet_hdr->flags = VIRTIO_NET_HDR_F_NEEDS_CSUM; | |
3ce9b20f | 645 | if (vlan_tx_tag_present(skb)) |
6ae7feb3 MT |
646 | vnet_hdr->csum_start = cpu_to_macvtap16(q, |
647 | skb_checksum_start_offset(skb) + VLAN_HLEN); | |
648 | else | |
649 | vnet_hdr->csum_start = cpu_to_macvtap16(q, | |
650 | skb_checksum_start_offset(skb)); | |
651 | vnet_hdr->csum_offset = cpu_to_macvtap16(q, skb->csum_offset); | |
10a8d94a JW |
652 | } else if (skb->ip_summed == CHECKSUM_UNNECESSARY) { |
653 | vnet_hdr->flags = VIRTIO_NET_HDR_F_DATA_VALID; | |
b9fb9ee0 | 654 | } /* else everything is zero */ |
b9fb9ee0 AB |
655 | } |
656 | ||
20d29d7a | 657 | /* Get packet from user space buffer */ |
97bc3633 SM |
658 | static ssize_t macvtap_get_user(struct macvtap_queue *q, struct msghdr *m, |
659 | const struct iovec *iv, unsigned long total_len, | |
660 | size_t count, int noblock) | |
20d29d7a | 661 | { |
16a3fa28 | 662 | int good_linear = SKB_MAX_HEAD(NET_IP_ALIGN); |
20d29d7a | 663 | struct sk_buff *skb; |
02df55d2 | 664 | struct macvlan_dev *vlan; |
97bc3633 | 665 | unsigned long len = total_len; |
20d29d7a | 666 | int err; |
b9fb9ee0 AB |
667 | struct virtio_net_hdr vnet_hdr = { 0 }; |
668 | int vnet_hdr_len = 0; | |
b92946e2 | 669 | int copylen = 0; |
97bc3633 | 670 | bool zerocopy = false; |
61d46bf9 | 671 | size_t linear; |
b9fb9ee0 AB |
672 | |
673 | if (q->flags & IFF_VNET_HDR) { | |
55afbd08 | 674 | vnet_hdr_len = q->vnet_hdr_sz; |
b9fb9ee0 AB |
675 | |
676 | err = -EINVAL; | |
ce3c8692 | 677 | if (len < vnet_hdr_len) |
b9fb9ee0 | 678 | goto err; |
ce3c8692 | 679 | len -= vnet_hdr_len; |
b9fb9ee0 AB |
680 | |
681 | err = memcpy_fromiovecend((void *)&vnet_hdr, iv, 0, | |
55afbd08 | 682 | sizeof(vnet_hdr)); |
b9fb9ee0 AB |
683 | if (err < 0) |
684 | goto err; | |
685 | if ((vnet_hdr.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) && | |
6ae7feb3 MT |
686 | macvtap16_to_cpu(q, vnet_hdr.csum_start) + |
687 | macvtap16_to_cpu(q, vnet_hdr.csum_offset) + 2 > | |
688 | macvtap16_to_cpu(q, vnet_hdr.hdr_len)) | |
689 | vnet_hdr.hdr_len = cpu_to_macvtap16(q, | |
690 | macvtap16_to_cpu(q, vnet_hdr.csum_start) + | |
691 | macvtap16_to_cpu(q, vnet_hdr.csum_offset) + 2); | |
b9fb9ee0 | 692 | err = -EINVAL; |
6ae7feb3 | 693 | if (macvtap16_to_cpu(q, vnet_hdr.hdr_len) > len) |
b9fb9ee0 AB |
694 | goto err; |
695 | } | |
20d29d7a | 696 | |
b9fb9ee0 | 697 | err = -EINVAL; |
20d29d7a | 698 | if (unlikely(len < ETH_HLEN)) |
b9fb9ee0 | 699 | goto err; |
20d29d7a | 700 | |
b92946e2 JW |
701 | err = -EMSGSIZE; |
702 | if (unlikely(count > UIO_MAXIOV)) | |
703 | goto err; | |
704 | ||
ece793fc | 705 | if (m && m->msg_control && sock_flag(&q->sk, SOCK_ZEROCOPY)) { |
6ae7feb3 MT |
706 | copylen = vnet_hdr.hdr_len ? |
707 | macvtap16_to_cpu(q, vnet_hdr.hdr_len) : GOODCOPY_LEN; | |
16a3fa28 JW |
708 | if (copylen > good_linear) |
709 | copylen = good_linear; | |
61d46bf9 | 710 | linear = copylen; |
ece793fc JW |
711 | if (iov_pages(iv, vnet_hdr_len + copylen, count) |
712 | <= MAX_SKB_FRAGS) | |
713 | zerocopy = true; | |
714 | } | |
715 | ||
716 | if (!zerocopy) { | |
97bc3633 | 717 | copylen = len; |
6ae7feb3 | 718 | if (macvtap16_to_cpu(q, vnet_hdr.hdr_len) > good_linear) |
16a3fa28 JW |
719 | linear = good_linear; |
720 | else | |
6ae7feb3 | 721 | linear = macvtap16_to_cpu(q, vnet_hdr.hdr_len); |
61d46bf9 | 722 | } |
97bc3633 SM |
723 | |
724 | skb = macvtap_alloc_skb(&q->sk, NET_IP_ALIGN, copylen, | |
61d46bf9 | 725 | linear, noblock, &err); |
02df55d2 AB |
726 | if (!skb) |
727 | goto err; | |
20d29d7a | 728 | |
01d6657b | 729 | if (zerocopy) |
97bc3633 | 730 | err = zerocopy_sg_from_iovec(skb, iv, vnet_hdr_len, count); |
ece793fc | 731 | else { |
97bc3633 SM |
732 | err = skb_copy_datagram_from_iovec(skb, 0, iv, vnet_hdr_len, |
733 | len); | |
ece793fc JW |
734 | if (!err && m && m->msg_control) { |
735 | struct ubuf_info *uarg = m->msg_control; | |
736 | uarg->callback(uarg, false); | |
737 | } | |
738 | } | |
739 | ||
02df55d2 | 740 | if (err) |
b9fb9ee0 | 741 | goto err_kfree; |
20d29d7a AB |
742 | |
743 | skb_set_network_header(skb, ETH_HLEN); | |
b9fb9ee0 AB |
744 | skb_reset_mac_header(skb); |
745 | skb->protocol = eth_hdr(skb)->h_proto; | |
746 | ||
747 | if (vnet_hdr_len) { | |
6ae7feb3 | 748 | err = macvtap_skb_from_vnet_hdr(q, skb, &vnet_hdr); |
b9fb9ee0 AB |
749 | if (err) |
750 | goto err_kfree; | |
751 | } | |
752 | ||
40893fd0 | 753 | skb_probe_transport_header(skb, ETH_HLEN); |
9b4d669b | 754 | |
ac4e4af1 VY |
755 | rcu_read_lock(); |
756 | vlan = rcu_dereference(q->vlan); | |
97bc3633 | 757 | /* copy skb_ubuf_info for callback when skb has no error */ |
01d6657b | 758 | if (zerocopy) { |
97bc3633 | 759 | skb_shinfo(skb)->destructor_arg = m->msg_control; |
01d6657b | 760 | skb_shinfo(skb)->tx_flags |= SKBTX_DEV_ZEROCOPY; |
c9af6db4 | 761 | skb_shinfo(skb)->tx_flags |= SKBTX_SHARED_FRAG; |
01d6657b | 762 | } |
29d79196 | 763 | if (vlan) { |
6acf54f1 VY |
764 | skb->dev = vlan->dev; |
765 | dev_queue_xmit(skb); | |
29d79196 | 766 | } else { |
02df55d2 | 767 | kfree_skb(skb); |
29d79196 | 768 | } |
ac4e4af1 | 769 | rcu_read_unlock(); |
20d29d7a | 770 | |
97bc3633 | 771 | return total_len; |
02df55d2 | 772 | |
b9fb9ee0 AB |
773 | err_kfree: |
774 | kfree_skb(skb); | |
775 | ||
02df55d2 | 776 | err: |
ac4e4af1 VY |
777 | rcu_read_lock(); |
778 | vlan = rcu_dereference(q->vlan); | |
02df55d2 | 779 | if (vlan) |
cd3e22b7 | 780 | this_cpu_inc(vlan->pcpu_stats->tx_dropped); |
ac4e4af1 | 781 | rcu_read_unlock(); |
02df55d2 | 782 | |
02df55d2 | 783 | return err; |
20d29d7a AB |
784 | } |
785 | ||
786 | static ssize_t macvtap_aio_write(struct kiocb *iocb, const struct iovec *iv, | |
787 | unsigned long count, loff_t pos) | |
788 | { | |
789 | struct file *file = iocb->ki_filp; | |
790 | ssize_t result = -ENOLINK; | |
02df55d2 | 791 | struct macvtap_queue *q = file->private_data; |
20d29d7a | 792 | |
97bc3633 SM |
793 | result = macvtap_get_user(q, NULL, iv, iov_length(iv, count), count, |
794 | file->f_flags & O_NONBLOCK); | |
20d29d7a AB |
795 | return result; |
796 | } | |
797 | ||
798 | /* Put packet to the user space buffer */ | |
799 | static ssize_t macvtap_put_user(struct macvtap_queue *q, | |
800 | const struct sk_buff *skb, | |
801 | const struct iovec *iv, int len) | |
802 | { | |
20d29d7a | 803 | int ret; |
b9fb9ee0 | 804 | int vnet_hdr_len = 0; |
f09e2249 | 805 | int vlan_offset = 0; |
ce232ce0 | 806 | int copied, total; |
b9fb9ee0 AB |
807 | |
808 | if (q->flags & IFF_VNET_HDR) { | |
809 | struct virtio_net_hdr vnet_hdr; | |
55afbd08 | 810 | vnet_hdr_len = q->vnet_hdr_sz; |
b9fb9ee0 AB |
811 | if ((len -= vnet_hdr_len) < 0) |
812 | return -EINVAL; | |
813 | ||
6ae7feb3 | 814 | macvtap_skb_to_vnet_hdr(q, skb, &vnet_hdr); |
b9fb9ee0 | 815 | |
55afbd08 | 816 | if (memcpy_toiovecend(iv, (void *)&vnet_hdr, 0, sizeof(vnet_hdr))) |
b9fb9ee0 AB |
817 | return -EFAULT; |
818 | } | |
ce232ce0 JW |
819 | total = copied = vnet_hdr_len; |
820 | total += skb->len; | |
f09e2249 BG |
821 | |
822 | if (!vlan_tx_tag_present(skb)) | |
823 | len = min_t(int, skb->len, len); | |
824 | else { | |
825 | int copy; | |
826 | struct { | |
827 | __be16 h_vlan_proto; | |
828 | __be16 h_vlan_TCI; | |
829 | } veth; | |
0fbe0d47 | 830 | veth.h_vlan_proto = skb->vlan_proto; |
f09e2249 BG |
831 | veth.h_vlan_TCI = htons(vlan_tx_tag_get(skb)); |
832 | ||
833 | vlan_offset = offsetof(struct vlan_ethhdr, h_vlan_proto); | |
834 | len = min_t(int, skb->len + VLAN_HLEN, len); | |
ce232ce0 | 835 | total += VLAN_HLEN; |
f09e2249 BG |
836 | |
837 | copy = min_t(int, vlan_offset, len); | |
838 | ret = skb_copy_datagram_const_iovec(skb, 0, iv, copied, copy); | |
839 | len -= copy; | |
840 | copied += copy; | |
841 | if (ret || !len) | |
842 | goto done; | |
843 | ||
844 | copy = min_t(int, sizeof(veth), len); | |
845 | ret = memcpy_toiovecend(iv, (void *)&veth, copied, copy); | |
846 | len -= copy; | |
847 | copied += copy; | |
848 | if (ret || !len) | |
849 | goto done; | |
850 | } | |
20d29d7a | 851 | |
f09e2249 | 852 | ret = skb_copy_datagram_const_iovec(skb, vlan_offset, iv, copied, len); |
20d29d7a | 853 | |
f09e2249 | 854 | done: |
ce232ce0 | 855 | return ret ? ret : total; |
20d29d7a AB |
856 | } |
857 | ||
55ec8e25 | 858 | static ssize_t macvtap_do_read(struct macvtap_queue *q, |
501c774c AB |
859 | const struct iovec *iv, unsigned long len, |
860 | int noblock) | |
20d29d7a | 861 | { |
ccf7e72b | 862 | DEFINE_WAIT(wait); |
20d29d7a | 863 | struct sk_buff *skb; |
501c774c | 864 | ssize_t ret = 0; |
20d29d7a | 865 | |
20d29d7a | 866 | while (len) { |
89cee917 JW |
867 | if (!noblock) |
868 | prepare_to_wait(sk_sleep(&q->sk), &wait, | |
869 | TASK_INTERRUPTIBLE); | |
20d29d7a AB |
870 | |
871 | /* Read frames from the queue */ | |
872 | skb = skb_dequeue(&q->sk.sk_receive_queue); | |
873 | if (!skb) { | |
501c774c | 874 | if (noblock) { |
20d29d7a AB |
875 | ret = -EAGAIN; |
876 | break; | |
877 | } | |
878 | if (signal_pending(current)) { | |
879 | ret = -ERESTARTSYS; | |
880 | break; | |
881 | } | |
882 | /* Nothing to read, let's sleep */ | |
883 | schedule(); | |
884 | continue; | |
885 | } | |
886 | ret = macvtap_put_user(q, skb, iv, len); | |
887 | kfree_skb(skb); | |
888 | break; | |
889 | } | |
890 | ||
89cee917 JW |
891 | if (!noblock) |
892 | finish_wait(sk_sleep(&q->sk), &wait); | |
501c774c AB |
893 | return ret; |
894 | } | |
895 | ||
896 | static ssize_t macvtap_aio_read(struct kiocb *iocb, const struct iovec *iv, | |
897 | unsigned long count, loff_t pos) | |
898 | { | |
899 | struct file *file = iocb->ki_filp; | |
900 | struct macvtap_queue *q = file->private_data; | |
901 | ssize_t len, ret = 0; | |
20d29d7a | 902 | |
501c774c AB |
903 | len = iov_length(iv, count); |
904 | if (len < 0) { | |
905 | ret = -EINVAL; | |
906 | goto out; | |
907 | } | |
908 | ||
55ec8e25 | 909 | ret = macvtap_do_read(q, iv, len, file->f_flags & O_NONBLOCK); |
ce232ce0 | 910 | ret = min_t(ssize_t, ret, len); |
e6ebc7f1 ZYW |
911 | if (ret > 0) |
912 | iocb->ki_pos = ret; | |
20d29d7a | 913 | out: |
20d29d7a AB |
914 | return ret; |
915 | } | |
916 | ||
8f475a31 JW |
917 | static struct macvlan_dev *macvtap_get_vlan(struct macvtap_queue *q) |
918 | { | |
919 | struct macvlan_dev *vlan; | |
920 | ||
441ac0fc VY |
921 | ASSERT_RTNL(); |
922 | vlan = rtnl_dereference(q->vlan); | |
8f475a31 JW |
923 | if (vlan) |
924 | dev_hold(vlan->dev); | |
8f475a31 JW |
925 | |
926 | return vlan; | |
927 | } | |
928 | ||
929 | static void macvtap_put_vlan(struct macvlan_dev *vlan) | |
930 | { | |
931 | dev_put(vlan->dev); | |
932 | } | |
933 | ||
815f236d JW |
934 | static int macvtap_ioctl_set_queue(struct file *file, unsigned int flags) |
935 | { | |
936 | struct macvtap_queue *q = file->private_data; | |
937 | struct macvlan_dev *vlan; | |
938 | int ret; | |
939 | ||
940 | vlan = macvtap_get_vlan(q); | |
941 | if (!vlan) | |
942 | return -EINVAL; | |
943 | ||
944 | if (flags & IFF_ATTACH_QUEUE) | |
945 | ret = macvtap_enable_queue(vlan->dev, file, q); | |
946 | else if (flags & IFF_DETACH_QUEUE) | |
947 | ret = macvtap_disable_queue(q); | |
f57855a5 JW |
948 | else |
949 | ret = -EINVAL; | |
815f236d JW |
950 | |
951 | macvtap_put_vlan(vlan); | |
952 | return ret; | |
953 | } | |
954 | ||
2be5c767 VY |
955 | static int set_offload(struct macvtap_queue *q, unsigned long arg) |
956 | { | |
957 | struct macvlan_dev *vlan; | |
958 | netdev_features_t features; | |
959 | netdev_features_t feature_mask = 0; | |
960 | ||
961 | vlan = rtnl_dereference(q->vlan); | |
962 | if (!vlan) | |
963 | return -ENOLINK; | |
964 | ||
965 | features = vlan->dev->features; | |
966 | ||
967 | if (arg & TUN_F_CSUM) { | |
968 | feature_mask = NETIF_F_HW_CSUM; | |
969 | ||
970 | if (arg & (TUN_F_TSO4 | TUN_F_TSO6)) { | |
971 | if (arg & TUN_F_TSO_ECN) | |
972 | feature_mask |= NETIF_F_TSO_ECN; | |
973 | if (arg & TUN_F_TSO4) | |
974 | feature_mask |= NETIF_F_TSO; | |
975 | if (arg & TUN_F_TSO6) | |
976 | feature_mask |= NETIF_F_TSO6; | |
977 | } | |
2be5c767 VY |
978 | } |
979 | ||
980 | /* tun/tap driver inverts the usage for TSO offloads, where | |
981 | * setting the TSO bit means that the userspace wants to | |
982 | * accept TSO frames and turning it off means that user space | |
983 | * does not support TSO. | |
984 | * For macvtap, we have to invert it to mean the same thing. | |
985 | * When user space turns off TSO, we turn off GSO/LRO so that | |
986 | * user-space will not receive TSO frames. | |
987 | */ | |
3d0ad094 | 988 | if (feature_mask & (NETIF_F_TSO | NETIF_F_TSO6)) |
2be5c767 VY |
989 | features |= RX_OFFLOADS; |
990 | else | |
991 | features &= ~RX_OFFLOADS; | |
992 | ||
993 | /* tap_features are the same as features on tun/tap and | |
994 | * reflect user expectations. | |
995 | */ | |
a567dd62 | 996 | vlan->tap_features = feature_mask; |
2be5c767 VY |
997 | vlan->set_features = features; |
998 | netdev_update_features(vlan->dev); | |
999 | ||
1000 | return 0; | |
1001 | } | |
1002 | ||
20d29d7a AB |
1003 | /* |
1004 | * provide compatibility with generic tun/tap interface | |
1005 | */ | |
1006 | static long macvtap_ioctl(struct file *file, unsigned int cmd, | |
1007 | unsigned long arg) | |
1008 | { | |
02df55d2 AB |
1009 | struct macvtap_queue *q = file->private_data; |
1010 | struct macvlan_dev *vlan; | |
20d29d7a AB |
1011 | void __user *argp = (void __user *)arg; |
1012 | struct ifreq __user *ifr = argp; | |
1013 | unsigned int __user *up = argp; | |
1014 | unsigned int u; | |
55afbd08 MT |
1015 | int __user *sp = argp; |
1016 | int s; | |
02df55d2 | 1017 | int ret; |
20d29d7a AB |
1018 | |
1019 | switch (cmd) { | |
1020 | case TUNSETIFF: | |
1021 | /* ignore the name, just look at flags */ | |
1022 | if (get_user(u, &ifr->ifr_flags)) | |
1023 | return -EFAULT; | |
b9fb9ee0 AB |
1024 | |
1025 | ret = 0; | |
6ae7feb3 | 1026 | if ((u & ~MACVTAP_FEATURES) != (IFF_NO_PI | IFF_TAP)) |
b9fb9ee0 AB |
1027 | ret = -EINVAL; |
1028 | else | |
1029 | q->flags = u; | |
1030 | ||
1031 | return ret; | |
20d29d7a AB |
1032 | |
1033 | case TUNGETIFF: | |
441ac0fc | 1034 | rtnl_lock(); |
8f475a31 | 1035 | vlan = macvtap_get_vlan(q); |
441ac0fc VY |
1036 | if (!vlan) { |
1037 | rtnl_unlock(); | |
20d29d7a | 1038 | return -ENOLINK; |
441ac0fc | 1039 | } |
20d29d7a | 1040 | |
02df55d2 | 1041 | ret = 0; |
13707f9e | 1042 | if (copy_to_user(&ifr->ifr_name, vlan->dev->name, IFNAMSIZ) || |
b9fb9ee0 | 1043 | put_user(q->flags, &ifr->ifr_flags)) |
02df55d2 | 1044 | ret = -EFAULT; |
8f475a31 | 1045 | macvtap_put_vlan(vlan); |
441ac0fc | 1046 | rtnl_unlock(); |
02df55d2 | 1047 | return ret; |
20d29d7a | 1048 | |
815f236d JW |
1049 | case TUNSETQUEUE: |
1050 | if (get_user(u, &ifr->ifr_flags)) | |
1051 | return -EFAULT; | |
441ac0fc VY |
1052 | rtnl_lock(); |
1053 | ret = macvtap_ioctl_set_queue(file, u); | |
1054 | rtnl_unlock(); | |
82a19eb8 | 1055 | return ret; |
815f236d | 1056 | |
20d29d7a | 1057 | case TUNGETFEATURES: |
6ae7feb3 | 1058 | if (put_user(IFF_TAP | IFF_NO_PI | MACVTAP_FEATURES, up)) |
20d29d7a AB |
1059 | return -EFAULT; |
1060 | return 0; | |
1061 | ||
1062 | case TUNSETSNDBUF: | |
1063 | if (get_user(u, up)) | |
1064 | return -EFAULT; | |
1065 | ||
20d29d7a | 1066 | q->sk.sk_sndbuf = u; |
20d29d7a AB |
1067 | return 0; |
1068 | ||
55afbd08 MT |
1069 | case TUNGETVNETHDRSZ: |
1070 | s = q->vnet_hdr_sz; | |
1071 | if (put_user(s, sp)) | |
1072 | return -EFAULT; | |
1073 | return 0; | |
1074 | ||
1075 | case TUNSETVNETHDRSZ: | |
1076 | if (get_user(s, sp)) | |
1077 | return -EFAULT; | |
1078 | if (s < (int)sizeof(struct virtio_net_hdr)) | |
1079 | return -EINVAL; | |
1080 | ||
1081 | q->vnet_hdr_sz = s; | |
1082 | return 0; | |
1083 | ||
20d29d7a AB |
1084 | case TUNSETOFFLOAD: |
1085 | /* let the user check for future flags */ | |
1086 | if (arg & ~(TUN_F_CSUM | TUN_F_TSO4 | TUN_F_TSO6 | | |
3d0ad094 | 1087 | TUN_F_TSO_ECN)) |
20d29d7a AB |
1088 | return -EINVAL; |
1089 | ||
2be5c767 VY |
1090 | rtnl_lock(); |
1091 | ret = set_offload(q, arg); | |
1092 | rtnl_unlock(); | |
1093 | return ret; | |
20d29d7a AB |
1094 | |
1095 | default: | |
1096 | return -EINVAL; | |
1097 | } | |
1098 | } | |
1099 | ||
1100 | #ifdef CONFIG_COMPAT | |
1101 | static long macvtap_compat_ioctl(struct file *file, unsigned int cmd, | |
1102 | unsigned long arg) | |
1103 | { | |
1104 | return macvtap_ioctl(file, cmd, (unsigned long)compat_ptr(arg)); | |
1105 | } | |
1106 | #endif | |
1107 | ||
1108 | static const struct file_operations macvtap_fops = { | |
1109 | .owner = THIS_MODULE, | |
1110 | .open = macvtap_open, | |
1111 | .release = macvtap_release, | |
1112 | .aio_read = macvtap_aio_read, | |
1113 | .aio_write = macvtap_aio_write, | |
1114 | .poll = macvtap_poll, | |
1115 | .llseek = no_llseek, | |
1116 | .unlocked_ioctl = macvtap_ioctl, | |
1117 | #ifdef CONFIG_COMPAT | |
1118 | .compat_ioctl = macvtap_compat_ioctl, | |
1119 | #endif | |
1120 | }; | |
1121 | ||
501c774c AB |
1122 | static int macvtap_sendmsg(struct kiocb *iocb, struct socket *sock, |
1123 | struct msghdr *m, size_t total_len) | |
1124 | { | |
1125 | struct macvtap_queue *q = container_of(sock, struct macvtap_queue, sock); | |
97bc3633 | 1126 | return macvtap_get_user(q, m, m->msg_iov, total_len, m->msg_iovlen, |
501c774c AB |
1127 | m->msg_flags & MSG_DONTWAIT); |
1128 | } | |
1129 | ||
1130 | static int macvtap_recvmsg(struct kiocb *iocb, struct socket *sock, | |
1131 | struct msghdr *m, size_t total_len, | |
1132 | int flags) | |
1133 | { | |
1134 | struct macvtap_queue *q = container_of(sock, struct macvtap_queue, sock); | |
1135 | int ret; | |
1136 | if (flags & ~(MSG_DONTWAIT|MSG_TRUNC)) | |
1137 | return -EINVAL; | |
55ec8e25 | 1138 | ret = macvtap_do_read(q, m->msg_iov, total_len, |
501c774c | 1139 | flags & MSG_DONTWAIT); |
de2aa476 DM |
1140 | if (ret > total_len) { |
1141 | m->msg_flags |= MSG_TRUNC; | |
1142 | ret = flags & MSG_TRUNC ? ret : total_len; | |
1143 | } | |
501c774c AB |
1144 | return ret; |
1145 | } | |
1146 | ||
1147 | /* Ops structure to mimic raw sockets with tun */ | |
1148 | static const struct proto_ops macvtap_socket_ops = { | |
1149 | .sendmsg = macvtap_sendmsg, | |
1150 | .recvmsg = macvtap_recvmsg, | |
1151 | }; | |
1152 | ||
1153 | /* Get an underlying socket object from tun file. Returns error unless file is | |
1154 | * attached to a device. The returned object works like a packet socket, it | |
1155 | * can be used for sock_sendmsg/sock_recvmsg. The caller is responsible for | |
1156 | * holding a reference to the file for as long as the socket is in use. */ | |
1157 | struct socket *macvtap_get_socket(struct file *file) | |
1158 | { | |
1159 | struct macvtap_queue *q; | |
1160 | if (file->f_op != &macvtap_fops) | |
1161 | return ERR_PTR(-EINVAL); | |
1162 | q = file->private_data; | |
1163 | if (!q) | |
1164 | return ERR_PTR(-EBADFD); | |
1165 | return &q->sock; | |
1166 | } | |
1167 | EXPORT_SYMBOL_GPL(macvtap_get_socket); | |
1168 | ||
9bf1907f EB |
1169 | static int macvtap_device_event(struct notifier_block *unused, |
1170 | unsigned long event, void *ptr) | |
1171 | { | |
351638e7 | 1172 | struct net_device *dev = netdev_notifier_info_to_dev(ptr); |
e09eff7f | 1173 | struct macvlan_dev *vlan; |
9bf1907f EB |
1174 | struct device *classdev; |
1175 | dev_t devt; | |
e09eff7f | 1176 | int err; |
9bf1907f EB |
1177 | |
1178 | if (dev->rtnl_link_ops != &macvtap_link_ops) | |
1179 | return NOTIFY_DONE; | |
1180 | ||
e09eff7f | 1181 | vlan = netdev_priv(dev); |
9bf1907f EB |
1182 | |
1183 | switch (event) { | |
1184 | case NETDEV_REGISTER: | |
1185 | /* Create the device node here after the network device has | |
1186 | * been registered but before register_netdevice has | |
1187 | * finished running. | |
1188 | */ | |
e09eff7f EB |
1189 | err = macvtap_get_minor(vlan); |
1190 | if (err) | |
1191 | return notifier_from_errno(err); | |
1192 | ||
1193 | devt = MKDEV(MAJOR(macvtap_major), vlan->minor); | |
9bf1907f EB |
1194 | classdev = device_create(macvtap_class, &dev->dev, devt, |
1195 | dev, "tap%d", dev->ifindex); | |
e09eff7f EB |
1196 | if (IS_ERR(classdev)) { |
1197 | macvtap_free_minor(vlan); | |
9bf1907f | 1198 | return notifier_from_errno(PTR_ERR(classdev)); |
e09eff7f | 1199 | } |
9bf1907f EB |
1200 | break; |
1201 | case NETDEV_UNREGISTER: | |
e09eff7f | 1202 | devt = MKDEV(MAJOR(macvtap_major), vlan->minor); |
9bf1907f | 1203 | device_destroy(macvtap_class, devt); |
e09eff7f | 1204 | macvtap_free_minor(vlan); |
9bf1907f EB |
1205 | break; |
1206 | } | |
1207 | ||
1208 | return NOTIFY_DONE; | |
1209 | } | |
1210 | ||
1211 | static struct notifier_block macvtap_notifier_block __read_mostly = { | |
1212 | .notifier_call = macvtap_device_event, | |
1213 | }; | |
1214 | ||
20d29d7a AB |
1215 | static int macvtap_init(void) |
1216 | { | |
1217 | int err; | |
1218 | ||
1219 | err = alloc_chrdev_region(&macvtap_major, 0, | |
1220 | MACVTAP_NUM_DEVS, "macvtap"); | |
1221 | if (err) | |
1222 | goto out1; | |
1223 | ||
1224 | cdev_init(&macvtap_cdev, &macvtap_fops); | |
1225 | err = cdev_add(&macvtap_cdev, macvtap_major, MACVTAP_NUM_DEVS); | |
1226 | if (err) | |
1227 | goto out2; | |
1228 | ||
1229 | macvtap_class = class_create(THIS_MODULE, "macvtap"); | |
1230 | if (IS_ERR(macvtap_class)) { | |
1231 | err = PTR_ERR(macvtap_class); | |
1232 | goto out3; | |
1233 | } | |
1234 | ||
9bf1907f | 1235 | err = register_netdevice_notifier(&macvtap_notifier_block); |
20d29d7a AB |
1236 | if (err) |
1237 | goto out4; | |
1238 | ||
9bf1907f EB |
1239 | err = macvlan_link_register(&macvtap_link_ops); |
1240 | if (err) | |
1241 | goto out5; | |
1242 | ||
20d29d7a AB |
1243 | return 0; |
1244 | ||
9bf1907f EB |
1245 | out5: |
1246 | unregister_netdevice_notifier(&macvtap_notifier_block); | |
20d29d7a AB |
1247 | out4: |
1248 | class_unregister(macvtap_class); | |
1249 | out3: | |
1250 | cdev_del(&macvtap_cdev); | |
1251 | out2: | |
1252 | unregister_chrdev_region(macvtap_major, MACVTAP_NUM_DEVS); | |
1253 | out1: | |
1254 | return err; | |
1255 | } | |
1256 | module_init(macvtap_init); | |
1257 | ||
1258 | static void macvtap_exit(void) | |
1259 | { | |
1260 | rtnl_link_unregister(&macvtap_link_ops); | |
9bf1907f | 1261 | unregister_netdevice_notifier(&macvtap_notifier_block); |
20d29d7a AB |
1262 | class_unregister(macvtap_class); |
1263 | cdev_del(&macvtap_cdev); | |
1264 | unregister_chrdev_region(macvtap_major, MACVTAP_NUM_DEVS); | |
1265 | } | |
1266 | module_exit(macvtap_exit); | |
1267 | ||
1268 | MODULE_ALIAS_RTNL_LINK("macvtap"); | |
1269 | MODULE_AUTHOR("Arnd Bergmann <arnd@arndb.de>"); | |
1270 | MODULE_LICENSE("GPL"); |