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 | ||
6ae7feb3 MT |
49 | #define MACVTAP_FEATURES (IFF_VNET_HDR | IFF_VNET_LE | IFF_MULTI_QUEUE) |
50 | ||
51 | static inline u16 macvtap16_to_cpu(struct macvtap_queue *q, __virtio16 val) | |
52 | { | |
53 | return __virtio16_to_cpu(q->flags & IFF_VNET_LE, val); | |
54 | } | |
55 | ||
56 | static inline __virtio16 cpu_to_macvtap16(struct macvtap_queue *q, u16 val) | |
57 | { | |
58 | return __cpu_to_virtio16(q->flags & IFF_VNET_LE, val); | |
59 | } | |
60 | ||
20d29d7a AB |
61 | static struct proto macvtap_proto = { |
62 | .name = "macvtap", | |
63 | .owner = THIS_MODULE, | |
64 | .obj_size = sizeof (struct macvtap_queue), | |
65 | }; | |
66 | ||
67 | /* | |
e09eff7f | 68 | * Variables for dealing with macvtaps device numbers. |
20d29d7a | 69 | */ |
1ebed71a | 70 | static dev_t macvtap_major; |
e09eff7f EB |
71 | #define MACVTAP_NUM_DEVS (1U << MINORBITS) |
72 | static DEFINE_MUTEX(minor_lock); | |
73 | static DEFINE_IDR(minor_idr); | |
74 | ||
97bc3633 | 75 | #define GOODCOPY_LEN 128 |
20d29d7a AB |
76 | static struct class *macvtap_class; |
77 | static struct cdev macvtap_cdev; | |
78 | ||
501c774c AB |
79 | static const struct proto_ops macvtap_socket_ops; |
80 | ||
2be5c767 | 81 | #define TUN_OFFLOADS (NETIF_F_HW_CSUM | NETIF_F_TSO_ECN | NETIF_F_TSO | \ |
3d0ad094 | 82 | NETIF_F_TSO6) |
2be5c767 | 83 | #define RX_OFFLOADS (NETIF_F_GRO | NETIF_F_LRO) |
a567dd62 VY |
84 | #define TAP_FEATURES (NETIF_F_GSO | NETIF_F_SG) |
85 | ||
6acf54f1 VY |
86 | static struct macvlan_dev *macvtap_get_vlan_rcu(const struct net_device *dev) |
87 | { | |
88 | return rcu_dereference(dev->rx_handler_data); | |
89 | } | |
90 | ||
20d29d7a AB |
91 | /* |
92 | * RCU usage: | |
02df55d2 AB |
93 | * The macvtap_queue and the macvlan_dev are loosely coupled, the |
94 | * pointers from one to the other can only be read while rcu_read_lock | |
441ac0fc | 95 | * or rtnl is held. |
20d29d7a | 96 | * |
02df55d2 AB |
97 | * Both the file and the macvlan_dev hold a reference on the macvtap_queue |
98 | * through sock_hold(&q->sk). When the macvlan_dev goes away first, | |
99 | * q->vlan becomes inaccessible. When the files gets closed, | |
100 | * macvtap_get_queue() fails. | |
20d29d7a | 101 | * |
02df55d2 AB |
102 | * There may still be references to the struct sock inside of the |
103 | * queue from outbound SKBs, but these never reference back to the | |
104 | * file or the dev. The data structure is freed through __sk_free | |
105 | * when both our references and any pending SKBs are gone. | |
20d29d7a | 106 | */ |
20d29d7a | 107 | |
815f236d | 108 | static int macvtap_enable_queue(struct net_device *dev, struct file *file, |
20d29d7a | 109 | struct macvtap_queue *q) |
815f236d JW |
110 | { |
111 | struct macvlan_dev *vlan = netdev_priv(dev); | |
112 | int err = -EINVAL; | |
113 | ||
441ac0fc | 114 | ASSERT_RTNL(); |
815f236d JW |
115 | |
116 | if (q->enabled) | |
117 | goto out; | |
118 | ||
119 | err = 0; | |
120 | rcu_assign_pointer(vlan->taps[vlan->numvtaps], q); | |
121 | q->queue_index = vlan->numvtaps; | |
122 | q->enabled = true; | |
123 | ||
124 | vlan->numvtaps++; | |
125 | out: | |
815f236d JW |
126 | return err; |
127 | } | |
128 | ||
40b8fe45 | 129 | /* Requires RTNL */ |
815f236d JW |
130 | static int macvtap_set_queue(struct net_device *dev, struct file *file, |
131 | struct macvtap_queue *q) | |
20d29d7a AB |
132 | { |
133 | struct macvlan_dev *vlan = netdev_priv(dev); | |
20d29d7a | 134 | |
815f236d | 135 | if (vlan->numqueues == MAX_MACVTAP_QUEUES) |
40b8fe45 | 136 | return -EBUSY; |
20d29d7a | 137 | |
02df55d2 | 138 | rcu_assign_pointer(q->vlan, vlan); |
376b1aab | 139 | rcu_assign_pointer(vlan->taps[vlan->numvtaps], q); |
02df55d2 | 140 | sock_hold(&q->sk); |
20d29d7a AB |
141 | |
142 | q->file = file; | |
376b1aab | 143 | q->queue_index = vlan->numvtaps; |
815f236d | 144 | q->enabled = true; |
02df55d2 | 145 | file->private_data = q; |
815f236d | 146 | list_add_tail(&q->next, &vlan->queue_list); |
20d29d7a | 147 | |
1565c7c1 | 148 | vlan->numvtaps++; |
815f236d | 149 | vlan->numqueues++; |
1565c7c1 | 150 | |
40b8fe45 | 151 | return 0; |
20d29d7a AB |
152 | } |
153 | ||
441ac0fc | 154 | static int macvtap_disable_queue(struct macvtap_queue *q) |
815f236d JW |
155 | { |
156 | struct macvlan_dev *vlan; | |
157 | struct macvtap_queue *nq; | |
158 | ||
441ac0fc | 159 | ASSERT_RTNL(); |
815f236d JW |
160 | if (!q->enabled) |
161 | return -EINVAL; | |
162 | ||
441ac0fc VY |
163 | vlan = rtnl_dereference(q->vlan); |
164 | ||
815f236d JW |
165 | if (vlan) { |
166 | int index = q->queue_index; | |
167 | BUG_ON(index >= vlan->numvtaps); | |
441ac0fc | 168 | nq = rtnl_dereference(vlan->taps[vlan->numvtaps - 1]); |
815f236d JW |
169 | nq->queue_index = index; |
170 | ||
171 | rcu_assign_pointer(vlan->taps[index], nq); | |
172 | RCU_INIT_POINTER(vlan->taps[vlan->numvtaps - 1], NULL); | |
173 | q->enabled = false; | |
174 | ||
175 | vlan->numvtaps--; | |
176 | } | |
177 | ||
178 | return 0; | |
179 | } | |
180 | ||
20d29d7a | 181 | /* |
02df55d2 AB |
182 | * The file owning the queue got closed, give up both |
183 | * the reference that the files holds as well as the | |
184 | * one from the macvlan_dev if that still exists. | |
20d29d7a AB |
185 | * |
186 | * Using the spinlock makes sure that we don't get | |
187 | * to the queue again after destroying it. | |
20d29d7a | 188 | */ |
02df55d2 | 189 | static void macvtap_put_queue(struct macvtap_queue *q) |
20d29d7a | 190 | { |
02df55d2 | 191 | struct macvlan_dev *vlan; |
20d29d7a | 192 | |
441ac0fc VY |
193 | rtnl_lock(); |
194 | vlan = rtnl_dereference(q->vlan); | |
195 | ||
02df55d2 | 196 | if (vlan) { |
815f236d | 197 | if (q->enabled) |
441ac0fc | 198 | BUG_ON(macvtap_disable_queue(q)); |
376b1aab | 199 | |
815f236d | 200 | vlan->numqueues--; |
2cfa5a04 | 201 | RCU_INIT_POINTER(q->vlan, NULL); |
02df55d2 | 202 | sock_put(&q->sk); |
815f236d | 203 | list_del_init(&q->next); |
20d29d7a AB |
204 | } |
205 | ||
441ac0fc | 206 | rtnl_unlock(); |
20d29d7a AB |
207 | |
208 | synchronize_rcu(); | |
209 | sock_put(&q->sk); | |
210 | } | |
211 | ||
212 | /* | |
1565c7c1 KK |
213 | * Select a queue based on the rxq of the device on which this packet |
214 | * arrived. If the incoming device is not mq, calculate a flow hash | |
215 | * to select a queue. If all fails, find the first available queue. | |
216 | * Cache vlan->numvtaps since it can become zero during the execution | |
217 | * of this function. | |
20d29d7a AB |
218 | */ |
219 | static struct macvtap_queue *macvtap_get_queue(struct net_device *dev, | |
220 | struct sk_buff *skb) | |
221 | { | |
222 | struct macvlan_dev *vlan = netdev_priv(dev); | |
1565c7c1 | 223 | struct macvtap_queue *tap = NULL; |
815f236d JW |
224 | /* Access to taps array is protected by rcu, but access to numvtaps |
225 | * isn't. Below we use it to lookup a queue, but treat it as a hint | |
226 | * and validate that the result isn't NULL - in case we are | |
227 | * racing against queue removal. | |
228 | */ | |
ed0483fa | 229 | int numvtaps = ACCESS_ONCE(vlan->numvtaps); |
1565c7c1 KK |
230 | __u32 rxq; |
231 | ||
232 | if (!numvtaps) | |
233 | goto out; | |
234 | ||
ef0002b5 | 235 | /* Check if we can use flow to select a queue */ |
3958afa1 | 236 | rxq = skb_get_hash(skb); |
ef0002b5 KK |
237 | if (rxq) { |
238 | tap = rcu_dereference(vlan->taps[rxq % numvtaps]); | |
376b1aab | 239 | goto out; |
ef0002b5 KK |
240 | } |
241 | ||
1565c7c1 KK |
242 | if (likely(skb_rx_queue_recorded(skb))) { |
243 | rxq = skb_get_rx_queue(skb); | |
20d29d7a | 244 | |
1565c7c1 KK |
245 | while (unlikely(rxq >= numvtaps)) |
246 | rxq -= numvtaps; | |
247 | ||
248 | tap = rcu_dereference(vlan->taps[rxq]); | |
376b1aab | 249 | goto out; |
1565c7c1 KK |
250 | } |
251 | ||
376b1aab | 252 | tap = rcu_dereference(vlan->taps[0]); |
1565c7c1 KK |
253 | out: |
254 | return tap; | |
20d29d7a AB |
255 | } |
256 | ||
02df55d2 AB |
257 | /* |
258 | * The net_device is going away, give up the reference | |
1565c7c1 KK |
259 | * that it holds on all queues and safely set the pointer |
260 | * from the queues to NULL. | |
02df55d2 | 261 | */ |
20d29d7a AB |
262 | static void macvtap_del_queues(struct net_device *dev) |
263 | { | |
264 | struct macvlan_dev *vlan = netdev_priv(dev); | |
815f236d | 265 | struct macvtap_queue *q, *tmp, *qlist[MAX_MACVTAP_QUEUES]; |
1565c7c1 | 266 | int i, j = 0; |
02df55d2 | 267 | |
441ac0fc | 268 | ASSERT_RTNL(); |
815f236d JW |
269 | list_for_each_entry_safe(q, tmp, &vlan->queue_list, next) { |
270 | list_del_init(&q->next); | |
376b1aab | 271 | qlist[j++] = q; |
376b1aab | 272 | RCU_INIT_POINTER(q->vlan, NULL); |
815f236d JW |
273 | if (q->enabled) |
274 | vlan->numvtaps--; | |
275 | vlan->numqueues--; | |
564517e8 | 276 | } |
815f236d JW |
277 | for (i = 0; i < vlan->numvtaps; i++) |
278 | RCU_INIT_POINTER(vlan->taps[i], NULL); | |
279 | BUG_ON(vlan->numvtaps); | |
280 | BUG_ON(vlan->numqueues); | |
99f34b38 EB |
281 | /* guarantee that any future macvtap_set_queue will fail */ |
282 | vlan->numvtaps = MAX_MACVTAP_QUEUES; | |
1565c7c1 KK |
283 | |
284 | for (--j; j >= 0; j--) | |
285 | sock_put(&qlist[j]->sk); | |
20d29d7a AB |
286 | } |
287 | ||
6acf54f1 | 288 | static rx_handler_result_t macvtap_handle_frame(struct sk_buff **pskb) |
20d29d7a | 289 | { |
6acf54f1 VY |
290 | struct sk_buff *skb = *pskb; |
291 | struct net_device *dev = skb->dev; | |
292 | struct macvlan_dev *vlan; | |
293 | struct macvtap_queue *q; | |
a567dd62 VY |
294 | netdev_features_t features = TAP_FEATURES; |
295 | ||
6acf54f1 VY |
296 | vlan = macvtap_get_vlan_rcu(dev); |
297 | if (!vlan) | |
298 | return RX_HANDLER_PASS; | |
299 | ||
300 | q = macvtap_get_queue(dev, skb); | |
20d29d7a | 301 | if (!q) |
6acf54f1 | 302 | return RX_HANDLER_PASS; |
8a35747a HX |
303 | |
304 | if (skb_queue_len(&q->sk.sk_receive_queue) >= dev->tx_queue_len) | |
305 | goto drop; | |
20d29d7a | 306 | |
6acf54f1 VY |
307 | skb_push(skb, ETH_HLEN); |
308 | ||
3e4f8b78 | 309 | /* Apply the forward feature mask so that we perform segmentation |
e5733321 VY |
310 | * according to users wishes. This only works if VNET_HDR is |
311 | * enabled. | |
3e4f8b78 | 312 | */ |
e5733321 VY |
313 | if (q->flags & IFF_VNET_HDR) |
314 | features |= vlan->tap_features; | |
04ffcb25 | 315 | if (netif_needs_gso(dev, skb, features)) { |
3e4f8b78 VY |
316 | struct sk_buff *segs = __skb_gso_segment(skb, features, false); |
317 | ||
318 | if (IS_ERR(segs)) | |
319 | goto drop; | |
320 | ||
321 | if (!segs) { | |
322 | skb_queue_tail(&q->sk.sk_receive_queue, skb); | |
323 | goto wake_up; | |
324 | } | |
325 | ||
326 | kfree_skb(skb); | |
327 | while (segs) { | |
328 | struct sk_buff *nskb = segs->next; | |
329 | ||
330 | segs->next = NULL; | |
331 | skb_queue_tail(&q->sk.sk_receive_queue, segs); | |
332 | segs = nskb; | |
333 | } | |
334 | } else { | |
cbdb0427 VY |
335 | /* If we receive a partial checksum and the tap side |
336 | * doesn't support checksum offload, compute the checksum. | |
337 | * Note: it doesn't matter which checksum feature to | |
338 | * check, we either support them all or none. | |
339 | */ | |
340 | if (skb->ip_summed == CHECKSUM_PARTIAL && | |
341 | !(features & NETIF_F_ALL_CSUM) && | |
342 | skb_checksum_help(skb)) | |
343 | goto drop; | |
3e4f8b78 VY |
344 | skb_queue_tail(&q->sk.sk_receive_queue, skb); |
345 | } | |
346 | ||
347 | wake_up: | |
4a4771a5 | 348 | wake_up_interruptible_poll(sk_sleep(&q->sk), POLLIN | POLLRDNORM | POLLRDBAND); |
6acf54f1 | 349 | return RX_HANDLER_CONSUMED; |
8a35747a HX |
350 | |
351 | drop: | |
6acf54f1 VY |
352 | /* Count errors/drops only here, thus don't care about args. */ |
353 | macvlan_count_rx(vlan, 0, 0, 0); | |
8a35747a | 354 | kfree_skb(skb); |
6acf54f1 | 355 | return RX_HANDLER_CONSUMED; |
20d29d7a AB |
356 | } |
357 | ||
e09eff7f EB |
358 | static int macvtap_get_minor(struct macvlan_dev *vlan) |
359 | { | |
360 | int retval = -ENOMEM; | |
e09eff7f EB |
361 | |
362 | mutex_lock(&minor_lock); | |
ec09ebc1 TH |
363 | retval = idr_alloc(&minor_idr, vlan, 1, MACVTAP_NUM_DEVS, GFP_KERNEL); |
364 | if (retval >= 0) { | |
365 | vlan->minor = retval; | |
366 | } else if (retval == -ENOSPC) { | |
e09eff7f EB |
367 | printk(KERN_ERR "too many macvtap devices\n"); |
368 | retval = -EINVAL; | |
e09eff7f | 369 | } |
e09eff7f | 370 | mutex_unlock(&minor_lock); |
ec09ebc1 | 371 | return retval < 0 ? retval : 0; |
e09eff7f EB |
372 | } |
373 | ||
374 | static void macvtap_free_minor(struct macvlan_dev *vlan) | |
375 | { | |
376 | mutex_lock(&minor_lock); | |
377 | if (vlan->minor) { | |
378 | idr_remove(&minor_idr, vlan->minor); | |
379 | vlan->minor = 0; | |
380 | } | |
381 | mutex_unlock(&minor_lock); | |
382 | } | |
383 | ||
384 | static struct net_device *dev_get_by_macvtap_minor(int minor) | |
385 | { | |
386 | struct net_device *dev = NULL; | |
387 | struct macvlan_dev *vlan; | |
388 | ||
389 | mutex_lock(&minor_lock); | |
390 | vlan = idr_find(&minor_idr, minor); | |
391 | if (vlan) { | |
392 | dev = vlan->dev; | |
393 | dev_hold(dev); | |
394 | } | |
395 | mutex_unlock(&minor_lock); | |
396 | return dev; | |
397 | } | |
398 | ||
20d29d7a AB |
399 | static int macvtap_newlink(struct net *src_net, |
400 | struct net_device *dev, | |
401 | struct nlattr *tb[], | |
402 | struct nlattr *data[]) | |
403 | { | |
815f236d | 404 | struct macvlan_dev *vlan = netdev_priv(dev); |
6acf54f1 VY |
405 | int err; |
406 | ||
815f236d JW |
407 | INIT_LIST_HEAD(&vlan->queue_list); |
408 | ||
2be5c767 VY |
409 | /* Since macvlan supports all offloads by default, make |
410 | * tap support all offloads also. | |
411 | */ | |
412 | vlan->tap_features = TUN_OFFLOADS; | |
413 | ||
6acf54f1 VY |
414 | err = netdev_rx_handler_register(dev, macvtap_handle_frame, vlan); |
415 | if (err) | |
416 | return err; | |
417 | ||
9bf1907f EB |
418 | /* Don't put anything that may fail after macvlan_common_newlink |
419 | * because we can't undo what it does. | |
420 | */ | |
2f6a1b66 | 421 | return macvlan_common_newlink(src_net, dev, tb, data); |
20d29d7a AB |
422 | } |
423 | ||
424 | static void macvtap_dellink(struct net_device *dev, | |
425 | struct list_head *head) | |
426 | { | |
6acf54f1 | 427 | netdev_rx_handler_unregister(dev); |
20d29d7a AB |
428 | macvtap_del_queues(dev); |
429 | macvlan_dellink(dev, head); | |
430 | } | |
431 | ||
8a35747a HX |
432 | static void macvtap_setup(struct net_device *dev) |
433 | { | |
434 | macvlan_common_setup(dev); | |
435 | dev->tx_queue_len = TUN_READQ_SIZE; | |
436 | } | |
437 | ||
20d29d7a AB |
438 | static struct rtnl_link_ops macvtap_link_ops __read_mostly = { |
439 | .kind = "macvtap", | |
8a35747a | 440 | .setup = macvtap_setup, |
20d29d7a AB |
441 | .newlink = macvtap_newlink, |
442 | .dellink = macvtap_dellink, | |
443 | }; | |
444 | ||
445 | ||
446 | static void macvtap_sock_write_space(struct sock *sk) | |
447 | { | |
43815482 ED |
448 | wait_queue_head_t *wqueue; |
449 | ||
20d29d7a AB |
450 | if (!sock_writeable(sk) || |
451 | !test_and_clear_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags)) | |
452 | return; | |
453 | ||
43815482 ED |
454 | wqueue = sk_sleep(sk); |
455 | if (wqueue && waitqueue_active(wqueue)) | |
456 | wake_up_interruptible_poll(wqueue, POLLOUT | POLLWRNORM | POLLWRBAND); | |
20d29d7a AB |
457 | } |
458 | ||
2259fef0 EB |
459 | static void macvtap_sock_destruct(struct sock *sk) |
460 | { | |
461 | skb_queue_purge(&sk->sk_receive_queue); | |
462 | } | |
463 | ||
20d29d7a AB |
464 | static int macvtap_open(struct inode *inode, struct file *file) |
465 | { | |
466 | struct net *net = current->nsproxy->net_ns; | |
40b8fe45 | 467 | struct net_device *dev; |
20d29d7a | 468 | struct macvtap_queue *q; |
40b8fe45 | 469 | int err = -ENODEV; |
20d29d7a | 470 | |
40b8fe45 VY |
471 | rtnl_lock(); |
472 | dev = dev_get_by_macvtap_minor(iminor(inode)); | |
20d29d7a AB |
473 | if (!dev) |
474 | goto out; | |
475 | ||
20d29d7a AB |
476 | err = -ENOMEM; |
477 | q = (struct macvtap_queue *)sk_alloc(net, AF_UNSPEC, GFP_KERNEL, | |
478 | &macvtap_proto); | |
479 | if (!q) | |
480 | goto out; | |
481 | ||
d9a90a31 | 482 | RCU_INIT_POINTER(q->sock.wq, &q->wq); |
43815482 | 483 | init_waitqueue_head(&q->wq.wait); |
20d29d7a AB |
484 | q->sock.type = SOCK_RAW; |
485 | q->sock.state = SS_CONNECTED; | |
501c774c AB |
486 | q->sock.file = file; |
487 | q->sock.ops = &macvtap_socket_ops; | |
20d29d7a | 488 | sock_init_data(&q->sock, &q->sk); |
20d29d7a | 489 | q->sk.sk_write_space = macvtap_sock_write_space; |
2259fef0 | 490 | q->sk.sk_destruct = macvtap_sock_destruct; |
b9fb9ee0 | 491 | q->flags = IFF_VNET_HDR | IFF_NO_PI | IFF_TAP; |
55afbd08 | 492 | q->vnet_hdr_sz = sizeof(struct virtio_net_hdr); |
20d29d7a | 493 | |
97bc3633 SM |
494 | /* |
495 | * so far only KVM virtio_net uses macvtap, enable zero copy between | |
496 | * guest kernel and host kernel when lower device supports zerocopy | |
047af9cf EB |
497 | * |
498 | * The macvlan supports zerocopy iff the lower device supports zero | |
499 | * copy so we don't have to look at the lower device directly. | |
97bc3633 | 500 | */ |
047af9cf EB |
501 | if ((dev->features & NETIF_F_HIGHDMA) && (dev->features & NETIF_F_SG)) |
502 | sock_set_flag(&q->sk, SOCK_ZEROCOPY); | |
97bc3633 | 503 | |
20d29d7a AB |
504 | err = macvtap_set_queue(dev, file, q); |
505 | if (err) | |
506 | sock_put(&q->sk); | |
507 | ||
508 | out: | |
509 | if (dev) | |
510 | dev_put(dev); | |
511 | ||
40b8fe45 | 512 | rtnl_unlock(); |
20d29d7a AB |
513 | return err; |
514 | } | |
515 | ||
516 | static int macvtap_release(struct inode *inode, struct file *file) | |
517 | { | |
02df55d2 AB |
518 | struct macvtap_queue *q = file->private_data; |
519 | macvtap_put_queue(q); | |
20d29d7a AB |
520 | return 0; |
521 | } | |
522 | ||
523 | static unsigned int macvtap_poll(struct file *file, poll_table * wait) | |
524 | { | |
02df55d2 | 525 | struct macvtap_queue *q = file->private_data; |
20d29d7a AB |
526 | unsigned int mask = POLLERR; |
527 | ||
528 | if (!q) | |
529 | goto out; | |
530 | ||
531 | mask = 0; | |
43815482 | 532 | poll_wait(file, &q->wq.wait, wait); |
20d29d7a AB |
533 | |
534 | if (!skb_queue_empty(&q->sk.sk_receive_queue)) | |
535 | mask |= POLLIN | POLLRDNORM; | |
536 | ||
537 | if (sock_writeable(&q->sk) || | |
538 | (!test_and_set_bit(SOCK_ASYNC_NOSPACE, &q->sock.flags) && | |
539 | sock_writeable(&q->sk))) | |
540 | mask |= POLLOUT | POLLWRNORM; | |
541 | ||
542 | out: | |
20d29d7a AB |
543 | return mask; |
544 | } | |
545 | ||
b9fb9ee0 AB |
546 | static inline struct sk_buff *macvtap_alloc_skb(struct sock *sk, size_t prepad, |
547 | size_t len, size_t linear, | |
548 | int noblock, int *err) | |
549 | { | |
550 | struct sk_buff *skb; | |
551 | ||
552 | /* Under a page? Don't bother with paged skb. */ | |
553 | if (prepad + len < PAGE_SIZE || !linear) | |
554 | linear = len; | |
555 | ||
556 | skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock, | |
28d64271 | 557 | err, 0); |
b9fb9ee0 AB |
558 | if (!skb) |
559 | return NULL; | |
560 | ||
561 | skb_reserve(skb, prepad); | |
562 | skb_put(skb, linear); | |
563 | skb->data_len = len - linear; | |
564 | skb->len += len - linear; | |
565 | ||
566 | return skb; | |
567 | } | |
568 | ||
569 | /* | |
570 | * macvtap_skb_from_vnet_hdr and macvtap_skb_to_vnet_hdr should | |
571 | * be shared with the tun/tap driver. | |
572 | */ | |
6ae7feb3 MT |
573 | static int macvtap_skb_from_vnet_hdr(struct macvtap_queue *q, |
574 | struct sk_buff *skb, | |
b9fb9ee0 AB |
575 | struct virtio_net_hdr *vnet_hdr) |
576 | { | |
577 | unsigned short gso_type = 0; | |
578 | if (vnet_hdr->gso_type != VIRTIO_NET_HDR_GSO_NONE) { | |
579 | switch (vnet_hdr->gso_type & ~VIRTIO_NET_HDR_GSO_ECN) { | |
580 | case VIRTIO_NET_HDR_GSO_TCPV4: | |
581 | gso_type = SKB_GSO_TCPV4; | |
582 | break; | |
583 | case VIRTIO_NET_HDR_GSO_TCPV6: | |
584 | gso_type = SKB_GSO_TCPV6; | |
585 | break; | |
586 | case VIRTIO_NET_HDR_GSO_UDP: | |
3d0ad094 BH |
587 | pr_warn_once("macvtap: %s: using disabled UFO feature; please fix this program\n", |
588 | current->comm); | |
b9fb9ee0 | 589 | gso_type = SKB_GSO_UDP; |
5188cd44 BH |
590 | if (skb->protocol == htons(ETH_P_IPV6)) |
591 | ipv6_proxy_select_ident(skb); | |
b9fb9ee0 AB |
592 | break; |
593 | default: | |
594 | return -EINVAL; | |
595 | } | |
596 | ||
597 | if (vnet_hdr->gso_type & VIRTIO_NET_HDR_GSO_ECN) | |
598 | gso_type |= SKB_GSO_TCP_ECN; | |
599 | ||
600 | if (vnet_hdr->gso_size == 0) | |
601 | return -EINVAL; | |
602 | } | |
603 | ||
604 | if (vnet_hdr->flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) { | |
6ae7feb3 MT |
605 | if (!skb_partial_csum_set(skb, macvtap16_to_cpu(q, vnet_hdr->csum_start), |
606 | macvtap16_to_cpu(q, vnet_hdr->csum_offset))) | |
b9fb9ee0 AB |
607 | return -EINVAL; |
608 | } | |
609 | ||
610 | if (vnet_hdr->gso_type != VIRTIO_NET_HDR_GSO_NONE) { | |
6ae7feb3 | 611 | skb_shinfo(skb)->gso_size = macvtap16_to_cpu(q, vnet_hdr->gso_size); |
c9af6db4 | 612 | skb_shinfo(skb)->gso_type = gso_type; |
b9fb9ee0 AB |
613 | |
614 | /* Header must be checked, and gso_segs computed. */ | |
615 | skb_shinfo(skb)->gso_type |= SKB_GSO_DODGY; | |
616 | skb_shinfo(skb)->gso_segs = 0; | |
617 | } | |
618 | return 0; | |
619 | } | |
620 | ||
6ae7feb3 MT |
621 | static void macvtap_skb_to_vnet_hdr(struct macvtap_queue *q, |
622 | const struct sk_buff *skb, | |
623 | struct virtio_net_hdr *vnet_hdr) | |
b9fb9ee0 AB |
624 | { |
625 | memset(vnet_hdr, 0, sizeof(*vnet_hdr)); | |
626 | ||
627 | if (skb_is_gso(skb)) { | |
628 | struct skb_shared_info *sinfo = skb_shinfo(skb); | |
629 | ||
630 | /* This is a hint as to how much should be linear. */ | |
6ae7feb3 MT |
631 | vnet_hdr->hdr_len = cpu_to_macvtap16(q, skb_headlen(skb)); |
632 | vnet_hdr->gso_size = cpu_to_macvtap16(q, sinfo->gso_size); | |
b9fb9ee0 AB |
633 | if (sinfo->gso_type & SKB_GSO_TCPV4) |
634 | vnet_hdr->gso_type = VIRTIO_NET_HDR_GSO_TCPV4; | |
635 | else if (sinfo->gso_type & SKB_GSO_TCPV6) | |
636 | vnet_hdr->gso_type = VIRTIO_NET_HDR_GSO_TCPV6; | |
b9fb9ee0 AB |
637 | else |
638 | BUG(); | |
639 | if (sinfo->gso_type & SKB_GSO_TCP_ECN) | |
640 | vnet_hdr->gso_type |= VIRTIO_NET_HDR_GSO_ECN; | |
641 | } else | |
642 | vnet_hdr->gso_type = VIRTIO_NET_HDR_GSO_NONE; | |
643 | ||
644 | if (skb->ip_summed == CHECKSUM_PARTIAL) { | |
645 | vnet_hdr->flags = VIRTIO_NET_HDR_F_NEEDS_CSUM; | |
3ce9b20f | 646 | if (vlan_tx_tag_present(skb)) |
6ae7feb3 MT |
647 | vnet_hdr->csum_start = cpu_to_macvtap16(q, |
648 | skb_checksum_start_offset(skb) + VLAN_HLEN); | |
649 | else | |
650 | vnet_hdr->csum_start = cpu_to_macvtap16(q, | |
651 | skb_checksum_start_offset(skb)); | |
652 | vnet_hdr->csum_offset = cpu_to_macvtap16(q, skb->csum_offset); | |
10a8d94a JW |
653 | } else if (skb->ip_summed == CHECKSUM_UNNECESSARY) { |
654 | vnet_hdr->flags = VIRTIO_NET_HDR_F_DATA_VALID; | |
b9fb9ee0 | 655 | } /* else everything is zero */ |
b9fb9ee0 AB |
656 | } |
657 | ||
20d29d7a | 658 | /* Get packet from user space buffer */ |
97bc3633 | 659 | static ssize_t macvtap_get_user(struct macvtap_queue *q, struct msghdr *m, |
f5ff53b4 | 660 | struct iov_iter *from, 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; |
f5ff53b4 | 665 | unsigned long total_len = iov_iter_count(from); |
97bc3633 | 666 | unsigned long len = total_len; |
20d29d7a | 667 | int err; |
b9fb9ee0 AB |
668 | struct virtio_net_hdr vnet_hdr = { 0 }; |
669 | int vnet_hdr_len = 0; | |
b92946e2 | 670 | int copylen = 0; |
97bc3633 | 671 | bool zerocopy = false; |
61d46bf9 | 672 | size_t linear; |
f5ff53b4 | 673 | ssize_t n; |
b9fb9ee0 AB |
674 | |
675 | if (q->flags & IFF_VNET_HDR) { | |
55afbd08 | 676 | vnet_hdr_len = q->vnet_hdr_sz; |
b9fb9ee0 AB |
677 | |
678 | err = -EINVAL; | |
ce3c8692 | 679 | if (len < vnet_hdr_len) |
b9fb9ee0 | 680 | goto err; |
ce3c8692 | 681 | len -= vnet_hdr_len; |
b9fb9ee0 | 682 | |
f5ff53b4 AV |
683 | err = -EFAULT; |
684 | n = copy_from_iter(&vnet_hdr, sizeof(vnet_hdr), from); | |
685 | if (n != sizeof(vnet_hdr)) | |
b9fb9ee0 | 686 | goto err; |
f5ff53b4 | 687 | iov_iter_advance(from, vnet_hdr_len - sizeof(vnet_hdr)); |
b9fb9ee0 | 688 | if ((vnet_hdr.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) && |
6ae7feb3 MT |
689 | macvtap16_to_cpu(q, vnet_hdr.csum_start) + |
690 | macvtap16_to_cpu(q, vnet_hdr.csum_offset) + 2 > | |
691 | macvtap16_to_cpu(q, vnet_hdr.hdr_len)) | |
692 | vnet_hdr.hdr_len = cpu_to_macvtap16(q, | |
693 | macvtap16_to_cpu(q, vnet_hdr.csum_start) + | |
694 | macvtap16_to_cpu(q, vnet_hdr.csum_offset) + 2); | |
b9fb9ee0 | 695 | err = -EINVAL; |
6ae7feb3 | 696 | if (macvtap16_to_cpu(q, vnet_hdr.hdr_len) > len) |
b9fb9ee0 AB |
697 | goto err; |
698 | } | |
20d29d7a | 699 | |
b9fb9ee0 | 700 | err = -EINVAL; |
20d29d7a | 701 | if (unlikely(len < ETH_HLEN)) |
b9fb9ee0 | 702 | goto err; |
20d29d7a | 703 | |
ece793fc | 704 | if (m && m->msg_control && sock_flag(&q->sk, SOCK_ZEROCOPY)) { |
f5ff53b4 AV |
705 | struct iov_iter i; |
706 | ||
6ae7feb3 MT |
707 | copylen = vnet_hdr.hdr_len ? |
708 | macvtap16_to_cpu(q, vnet_hdr.hdr_len) : GOODCOPY_LEN; | |
16a3fa28 JW |
709 | if (copylen > good_linear) |
710 | copylen = good_linear; | |
61d46bf9 | 711 | linear = copylen; |
f5ff53b4 AV |
712 | i = *from; |
713 | iov_iter_advance(&i, copylen); | |
714 | if (iov_iter_npages(&i, INT_MAX) <= MAX_SKB_FRAGS) | |
ece793fc JW |
715 | zerocopy = true; |
716 | } | |
717 | ||
718 | if (!zerocopy) { | |
97bc3633 | 719 | copylen = len; |
6ae7feb3 | 720 | if (macvtap16_to_cpu(q, vnet_hdr.hdr_len) > good_linear) |
16a3fa28 JW |
721 | linear = good_linear; |
722 | else | |
6ae7feb3 | 723 | linear = macvtap16_to_cpu(q, vnet_hdr.hdr_len); |
61d46bf9 | 724 | } |
97bc3633 SM |
725 | |
726 | skb = macvtap_alloc_skb(&q->sk, NET_IP_ALIGN, copylen, | |
61d46bf9 | 727 | linear, noblock, &err); |
02df55d2 AB |
728 | if (!skb) |
729 | goto err; | |
20d29d7a | 730 | |
01d6657b | 731 | if (zerocopy) |
f5ff53b4 | 732 | err = zerocopy_sg_from_iter(skb, from); |
ece793fc | 733 | else { |
f5ff53b4 | 734 | err = skb_copy_datagram_from_iter(skb, 0, from, len); |
ece793fc JW |
735 | if (!err && m && m->msg_control) { |
736 | struct ubuf_info *uarg = m->msg_control; | |
737 | uarg->callback(uarg, false); | |
738 | } | |
739 | } | |
740 | ||
02df55d2 | 741 | if (err) |
b9fb9ee0 | 742 | goto err_kfree; |
20d29d7a AB |
743 | |
744 | skb_set_network_header(skb, ETH_HLEN); | |
b9fb9ee0 AB |
745 | skb_reset_mac_header(skb); |
746 | skb->protocol = eth_hdr(skb)->h_proto; | |
747 | ||
748 | if (vnet_hdr_len) { | |
6ae7feb3 | 749 | err = macvtap_skb_from_vnet_hdr(q, skb, &vnet_hdr); |
b9fb9ee0 AB |
750 | if (err) |
751 | goto err_kfree; | |
752 | } | |
753 | ||
40893fd0 | 754 | skb_probe_transport_header(skb, ETH_HLEN); |
9b4d669b | 755 | |
ac4e4af1 VY |
756 | rcu_read_lock(); |
757 | vlan = rcu_dereference(q->vlan); | |
97bc3633 | 758 | /* copy skb_ubuf_info for callback when skb has no error */ |
01d6657b | 759 | if (zerocopy) { |
97bc3633 | 760 | skb_shinfo(skb)->destructor_arg = m->msg_control; |
01d6657b | 761 | skb_shinfo(skb)->tx_flags |= SKBTX_DEV_ZEROCOPY; |
c9af6db4 | 762 | skb_shinfo(skb)->tx_flags |= SKBTX_SHARED_FRAG; |
01d6657b | 763 | } |
29d79196 | 764 | if (vlan) { |
6acf54f1 VY |
765 | skb->dev = vlan->dev; |
766 | dev_queue_xmit(skb); | |
29d79196 | 767 | } else { |
02df55d2 | 768 | kfree_skb(skb); |
29d79196 | 769 | } |
ac4e4af1 | 770 | rcu_read_unlock(); |
20d29d7a | 771 | |
97bc3633 | 772 | return total_len; |
02df55d2 | 773 | |
b9fb9ee0 AB |
774 | err_kfree: |
775 | kfree_skb(skb); | |
776 | ||
02df55d2 | 777 | err: |
ac4e4af1 VY |
778 | rcu_read_lock(); |
779 | vlan = rcu_dereference(q->vlan); | |
02df55d2 | 780 | if (vlan) |
cd3e22b7 | 781 | this_cpu_inc(vlan->pcpu_stats->tx_dropped); |
ac4e4af1 | 782 | rcu_read_unlock(); |
02df55d2 | 783 | |
02df55d2 | 784 | return err; |
20d29d7a AB |
785 | } |
786 | ||
f5ff53b4 | 787 | static ssize_t macvtap_write_iter(struct kiocb *iocb, struct iov_iter *from) |
20d29d7a AB |
788 | { |
789 | struct file *file = iocb->ki_filp; | |
02df55d2 | 790 | struct macvtap_queue *q = file->private_data; |
20d29d7a | 791 | |
f5ff53b4 | 792 | return macvtap_get_user(q, NULL, from, file->f_flags & O_NONBLOCK); |
20d29d7a AB |
793 | } |
794 | ||
795 | /* Put packet to the user space buffer */ | |
796 | static ssize_t macvtap_put_user(struct macvtap_queue *q, | |
797 | const struct sk_buff *skb, | |
6c36d2e2 | 798 | struct iov_iter *iter) |
20d29d7a | 799 | { |
20d29d7a | 800 | int ret; |
b9fb9ee0 | 801 | int vnet_hdr_len = 0; |
f09e2249 | 802 | int vlan_offset = 0; |
6c36d2e2 | 803 | int total; |
b9fb9ee0 AB |
804 | |
805 | if (q->flags & IFF_VNET_HDR) { | |
806 | struct virtio_net_hdr vnet_hdr; | |
55afbd08 | 807 | vnet_hdr_len = q->vnet_hdr_sz; |
6c36d2e2 | 808 | if (iov_iter_count(iter) < vnet_hdr_len) |
b9fb9ee0 AB |
809 | return -EINVAL; |
810 | ||
6ae7feb3 | 811 | macvtap_skb_to_vnet_hdr(q, skb, &vnet_hdr); |
b9fb9ee0 | 812 | |
6c36d2e2 HX |
813 | if (copy_to_iter(&vnet_hdr, sizeof(vnet_hdr), iter) != |
814 | sizeof(vnet_hdr)) | |
b9fb9ee0 | 815 | return -EFAULT; |
7cc76f51 JW |
816 | |
817 | iov_iter_advance(iter, vnet_hdr_len - sizeof(vnet_hdr)); | |
b9fb9ee0 | 818 | } |
6c36d2e2 | 819 | total = vnet_hdr_len; |
ce232ce0 | 820 | total += skb->len; |
f09e2249 | 821 | |
6c36d2e2 | 822 | if (vlan_tx_tag_present(skb)) { |
f09e2249 BG |
823 | struct { |
824 | __be16 h_vlan_proto; | |
825 | __be16 h_vlan_TCI; | |
826 | } veth; | |
0fbe0d47 | 827 | veth.h_vlan_proto = skb->vlan_proto; |
f09e2249 BG |
828 | veth.h_vlan_TCI = htons(vlan_tx_tag_get(skb)); |
829 | ||
830 | vlan_offset = offsetof(struct vlan_ethhdr, h_vlan_proto); | |
ce232ce0 | 831 | total += VLAN_HLEN; |
f09e2249 | 832 | |
6c36d2e2 HX |
833 | ret = skb_copy_datagram_iter(skb, 0, iter, vlan_offset); |
834 | if (ret || !iov_iter_count(iter)) | |
f09e2249 BG |
835 | goto done; |
836 | ||
6c36d2e2 HX |
837 | ret = copy_to_iter(&veth, sizeof(veth), iter); |
838 | if (ret != sizeof(veth) || !iov_iter_count(iter)) | |
f09e2249 BG |
839 | goto done; |
840 | } | |
20d29d7a | 841 | |
6c36d2e2 HX |
842 | ret = skb_copy_datagram_iter(skb, vlan_offset, iter, |
843 | skb->len - vlan_offset); | |
20d29d7a | 844 | |
f09e2249 | 845 | done: |
ce232ce0 | 846 | return ret ? ret : total; |
20d29d7a AB |
847 | } |
848 | ||
55ec8e25 | 849 | static ssize_t macvtap_do_read(struct macvtap_queue *q, |
3af0bfe5 | 850 | struct iov_iter *to, |
501c774c | 851 | int noblock) |
20d29d7a | 852 | { |
ccf7e72b | 853 | DEFINE_WAIT(wait); |
20d29d7a | 854 | struct sk_buff *skb; |
501c774c | 855 | ssize_t ret = 0; |
20d29d7a | 856 | |
3af0bfe5 AV |
857 | if (!iov_iter_count(to)) |
858 | return 0; | |
859 | ||
860 | while (1) { | |
89cee917 JW |
861 | if (!noblock) |
862 | prepare_to_wait(sk_sleep(&q->sk), &wait, | |
863 | TASK_INTERRUPTIBLE); | |
20d29d7a AB |
864 | |
865 | /* Read frames from the queue */ | |
866 | skb = skb_dequeue(&q->sk.sk_receive_queue); | |
3af0bfe5 AV |
867 | if (skb) |
868 | break; | |
869 | if (noblock) { | |
870 | ret = -EAGAIN; | |
871 | break; | |
20d29d7a | 872 | } |
3af0bfe5 AV |
873 | if (signal_pending(current)) { |
874 | ret = -ERESTARTSYS; | |
875 | break; | |
876 | } | |
877 | /* Nothing to read, let's sleep */ | |
878 | schedule(); | |
879 | } | |
880 | if (skb) { | |
881 | ret = macvtap_put_user(q, skb, to); | |
f51a5e82 JW |
882 | if (unlikely(ret < 0)) |
883 | kfree_skb(skb); | |
884 | else | |
885 | consume_skb(skb); | |
20d29d7a | 886 | } |
89cee917 JW |
887 | if (!noblock) |
888 | finish_wait(sk_sleep(&q->sk), &wait); | |
501c774c AB |
889 | return ret; |
890 | } | |
891 | ||
3af0bfe5 | 892 | static ssize_t macvtap_read_iter(struct kiocb *iocb, struct iov_iter *to) |
501c774c AB |
893 | { |
894 | struct file *file = iocb->ki_filp; | |
895 | struct macvtap_queue *q = file->private_data; | |
3af0bfe5 | 896 | ssize_t len = iov_iter_count(to), ret; |
20d29d7a | 897 | |
3af0bfe5 | 898 | ret = macvtap_do_read(q, to, file->f_flags & O_NONBLOCK); |
ce232ce0 | 899 | ret = min_t(ssize_t, ret, len); |
e6ebc7f1 ZYW |
900 | if (ret > 0) |
901 | iocb->ki_pos = ret; | |
20d29d7a AB |
902 | return ret; |
903 | } | |
904 | ||
8f475a31 JW |
905 | static struct macvlan_dev *macvtap_get_vlan(struct macvtap_queue *q) |
906 | { | |
907 | struct macvlan_dev *vlan; | |
908 | ||
441ac0fc VY |
909 | ASSERT_RTNL(); |
910 | vlan = rtnl_dereference(q->vlan); | |
8f475a31 JW |
911 | if (vlan) |
912 | dev_hold(vlan->dev); | |
8f475a31 JW |
913 | |
914 | return vlan; | |
915 | } | |
916 | ||
917 | static void macvtap_put_vlan(struct macvlan_dev *vlan) | |
918 | { | |
919 | dev_put(vlan->dev); | |
920 | } | |
921 | ||
815f236d JW |
922 | static int macvtap_ioctl_set_queue(struct file *file, unsigned int flags) |
923 | { | |
924 | struct macvtap_queue *q = file->private_data; | |
925 | struct macvlan_dev *vlan; | |
926 | int ret; | |
927 | ||
928 | vlan = macvtap_get_vlan(q); | |
929 | if (!vlan) | |
930 | return -EINVAL; | |
931 | ||
932 | if (flags & IFF_ATTACH_QUEUE) | |
933 | ret = macvtap_enable_queue(vlan->dev, file, q); | |
934 | else if (flags & IFF_DETACH_QUEUE) | |
935 | ret = macvtap_disable_queue(q); | |
f57855a5 JW |
936 | else |
937 | ret = -EINVAL; | |
815f236d JW |
938 | |
939 | macvtap_put_vlan(vlan); | |
940 | return ret; | |
941 | } | |
942 | ||
2be5c767 VY |
943 | static int set_offload(struct macvtap_queue *q, unsigned long arg) |
944 | { | |
945 | struct macvlan_dev *vlan; | |
946 | netdev_features_t features; | |
947 | netdev_features_t feature_mask = 0; | |
948 | ||
949 | vlan = rtnl_dereference(q->vlan); | |
950 | if (!vlan) | |
951 | return -ENOLINK; | |
952 | ||
953 | features = vlan->dev->features; | |
954 | ||
955 | if (arg & TUN_F_CSUM) { | |
956 | feature_mask = NETIF_F_HW_CSUM; | |
957 | ||
958 | if (arg & (TUN_F_TSO4 | TUN_F_TSO6)) { | |
959 | if (arg & TUN_F_TSO_ECN) | |
960 | feature_mask |= NETIF_F_TSO_ECN; | |
961 | if (arg & TUN_F_TSO4) | |
962 | feature_mask |= NETIF_F_TSO; | |
963 | if (arg & TUN_F_TSO6) | |
964 | feature_mask |= NETIF_F_TSO6; | |
965 | } | |
2be5c767 VY |
966 | } |
967 | ||
968 | /* tun/tap driver inverts the usage for TSO offloads, where | |
969 | * setting the TSO bit means that the userspace wants to | |
970 | * accept TSO frames and turning it off means that user space | |
971 | * does not support TSO. | |
972 | * For macvtap, we have to invert it to mean the same thing. | |
973 | * When user space turns off TSO, we turn off GSO/LRO so that | |
974 | * user-space will not receive TSO frames. | |
975 | */ | |
3d0ad094 | 976 | if (feature_mask & (NETIF_F_TSO | NETIF_F_TSO6)) |
2be5c767 VY |
977 | features |= RX_OFFLOADS; |
978 | else | |
979 | features &= ~RX_OFFLOADS; | |
980 | ||
981 | /* tap_features are the same as features on tun/tap and | |
982 | * reflect user expectations. | |
983 | */ | |
a567dd62 | 984 | vlan->tap_features = feature_mask; |
2be5c767 VY |
985 | vlan->set_features = features; |
986 | netdev_update_features(vlan->dev); | |
987 | ||
988 | return 0; | |
989 | } | |
990 | ||
20d29d7a AB |
991 | /* |
992 | * provide compatibility with generic tun/tap interface | |
993 | */ | |
994 | static long macvtap_ioctl(struct file *file, unsigned int cmd, | |
995 | unsigned long arg) | |
996 | { | |
02df55d2 AB |
997 | struct macvtap_queue *q = file->private_data; |
998 | struct macvlan_dev *vlan; | |
20d29d7a AB |
999 | void __user *argp = (void __user *)arg; |
1000 | struct ifreq __user *ifr = argp; | |
1001 | unsigned int __user *up = argp; | |
1002 | unsigned int u; | |
55afbd08 MT |
1003 | int __user *sp = argp; |
1004 | int s; | |
02df55d2 | 1005 | int ret; |
20d29d7a AB |
1006 | |
1007 | switch (cmd) { | |
1008 | case TUNSETIFF: | |
1009 | /* ignore the name, just look at flags */ | |
1010 | if (get_user(u, &ifr->ifr_flags)) | |
1011 | return -EFAULT; | |
b9fb9ee0 AB |
1012 | |
1013 | ret = 0; | |
6ae7feb3 | 1014 | if ((u & ~MACVTAP_FEATURES) != (IFF_NO_PI | IFF_TAP)) |
b9fb9ee0 AB |
1015 | ret = -EINVAL; |
1016 | else | |
1017 | q->flags = u; | |
1018 | ||
1019 | return ret; | |
20d29d7a AB |
1020 | |
1021 | case TUNGETIFF: | |
441ac0fc | 1022 | rtnl_lock(); |
8f475a31 | 1023 | vlan = macvtap_get_vlan(q); |
441ac0fc VY |
1024 | if (!vlan) { |
1025 | rtnl_unlock(); | |
20d29d7a | 1026 | return -ENOLINK; |
441ac0fc | 1027 | } |
20d29d7a | 1028 | |
02df55d2 | 1029 | ret = 0; |
13707f9e | 1030 | if (copy_to_user(&ifr->ifr_name, vlan->dev->name, IFNAMSIZ) || |
b9fb9ee0 | 1031 | put_user(q->flags, &ifr->ifr_flags)) |
02df55d2 | 1032 | ret = -EFAULT; |
8f475a31 | 1033 | macvtap_put_vlan(vlan); |
441ac0fc | 1034 | rtnl_unlock(); |
02df55d2 | 1035 | return ret; |
20d29d7a | 1036 | |
815f236d JW |
1037 | case TUNSETQUEUE: |
1038 | if (get_user(u, &ifr->ifr_flags)) | |
1039 | return -EFAULT; | |
441ac0fc VY |
1040 | rtnl_lock(); |
1041 | ret = macvtap_ioctl_set_queue(file, u); | |
1042 | rtnl_unlock(); | |
82a19eb8 | 1043 | return ret; |
815f236d | 1044 | |
20d29d7a | 1045 | case TUNGETFEATURES: |
6ae7feb3 | 1046 | if (put_user(IFF_TAP | IFF_NO_PI | MACVTAP_FEATURES, up)) |
20d29d7a AB |
1047 | return -EFAULT; |
1048 | return 0; | |
1049 | ||
1050 | case TUNSETSNDBUF: | |
1051 | if (get_user(u, up)) | |
1052 | return -EFAULT; | |
1053 | ||
20d29d7a | 1054 | q->sk.sk_sndbuf = u; |
20d29d7a AB |
1055 | return 0; |
1056 | ||
55afbd08 MT |
1057 | case TUNGETVNETHDRSZ: |
1058 | s = q->vnet_hdr_sz; | |
1059 | if (put_user(s, sp)) | |
1060 | return -EFAULT; | |
1061 | return 0; | |
1062 | ||
1063 | case TUNSETVNETHDRSZ: | |
1064 | if (get_user(s, sp)) | |
1065 | return -EFAULT; | |
1066 | if (s < (int)sizeof(struct virtio_net_hdr)) | |
1067 | return -EINVAL; | |
1068 | ||
1069 | q->vnet_hdr_sz = s; | |
1070 | return 0; | |
1071 | ||
20d29d7a AB |
1072 | case TUNSETOFFLOAD: |
1073 | /* let the user check for future flags */ | |
1074 | if (arg & ~(TUN_F_CSUM | TUN_F_TSO4 | TUN_F_TSO6 | | |
3d0ad094 | 1075 | TUN_F_TSO_ECN)) |
20d29d7a AB |
1076 | return -EINVAL; |
1077 | ||
2be5c767 VY |
1078 | rtnl_lock(); |
1079 | ret = set_offload(q, arg); | |
1080 | rtnl_unlock(); | |
1081 | return ret; | |
20d29d7a AB |
1082 | |
1083 | default: | |
1084 | return -EINVAL; | |
1085 | } | |
1086 | } | |
1087 | ||
1088 | #ifdef CONFIG_COMPAT | |
1089 | static long macvtap_compat_ioctl(struct file *file, unsigned int cmd, | |
1090 | unsigned long arg) | |
1091 | { | |
1092 | return macvtap_ioctl(file, cmd, (unsigned long)compat_ptr(arg)); | |
1093 | } | |
1094 | #endif | |
1095 | ||
1096 | static const struct file_operations macvtap_fops = { | |
1097 | .owner = THIS_MODULE, | |
1098 | .open = macvtap_open, | |
1099 | .release = macvtap_release, | |
3af0bfe5 | 1100 | .read = new_sync_read, |
f5ff53b4 | 1101 | .write = new_sync_write, |
3af0bfe5 | 1102 | .read_iter = macvtap_read_iter, |
f5ff53b4 | 1103 | .write_iter = macvtap_write_iter, |
20d29d7a AB |
1104 | .poll = macvtap_poll, |
1105 | .llseek = no_llseek, | |
1106 | .unlocked_ioctl = macvtap_ioctl, | |
1107 | #ifdef CONFIG_COMPAT | |
1108 | .compat_ioctl = macvtap_compat_ioctl, | |
1109 | #endif | |
1110 | }; | |
1111 | ||
501c774c AB |
1112 | static int macvtap_sendmsg(struct kiocb *iocb, struct socket *sock, |
1113 | struct msghdr *m, size_t total_len) | |
1114 | { | |
1115 | struct macvtap_queue *q = container_of(sock, struct macvtap_queue, sock); | |
c0371da6 | 1116 | return macvtap_get_user(q, m, &m->msg_iter, m->msg_flags & MSG_DONTWAIT); |
501c774c AB |
1117 | } |
1118 | ||
1119 | static int macvtap_recvmsg(struct kiocb *iocb, struct socket *sock, | |
1120 | struct msghdr *m, size_t total_len, | |
1121 | int flags) | |
1122 | { | |
1123 | struct macvtap_queue *q = container_of(sock, struct macvtap_queue, sock); | |
1124 | int ret; | |
1125 | if (flags & ~(MSG_DONTWAIT|MSG_TRUNC)) | |
1126 | return -EINVAL; | |
c0371da6 | 1127 | ret = macvtap_do_read(q, &m->msg_iter, flags & MSG_DONTWAIT); |
de2aa476 DM |
1128 | if (ret > total_len) { |
1129 | m->msg_flags |= MSG_TRUNC; | |
1130 | ret = flags & MSG_TRUNC ? ret : total_len; | |
1131 | } | |
501c774c AB |
1132 | return ret; |
1133 | } | |
1134 | ||
1135 | /* Ops structure to mimic raw sockets with tun */ | |
1136 | static const struct proto_ops macvtap_socket_ops = { | |
1137 | .sendmsg = macvtap_sendmsg, | |
1138 | .recvmsg = macvtap_recvmsg, | |
1139 | }; | |
1140 | ||
1141 | /* Get an underlying socket object from tun file. Returns error unless file is | |
1142 | * attached to a device. The returned object works like a packet socket, it | |
1143 | * can be used for sock_sendmsg/sock_recvmsg. The caller is responsible for | |
1144 | * holding a reference to the file for as long as the socket is in use. */ | |
1145 | struct socket *macvtap_get_socket(struct file *file) | |
1146 | { | |
1147 | struct macvtap_queue *q; | |
1148 | if (file->f_op != &macvtap_fops) | |
1149 | return ERR_PTR(-EINVAL); | |
1150 | q = file->private_data; | |
1151 | if (!q) | |
1152 | return ERR_PTR(-EBADFD); | |
1153 | return &q->sock; | |
1154 | } | |
1155 | EXPORT_SYMBOL_GPL(macvtap_get_socket); | |
1156 | ||
9bf1907f EB |
1157 | static int macvtap_device_event(struct notifier_block *unused, |
1158 | unsigned long event, void *ptr) | |
1159 | { | |
351638e7 | 1160 | struct net_device *dev = netdev_notifier_info_to_dev(ptr); |
e09eff7f | 1161 | struct macvlan_dev *vlan; |
9bf1907f EB |
1162 | struct device *classdev; |
1163 | dev_t devt; | |
e09eff7f | 1164 | int err; |
9bf1907f EB |
1165 | |
1166 | if (dev->rtnl_link_ops != &macvtap_link_ops) | |
1167 | return NOTIFY_DONE; | |
1168 | ||
e09eff7f | 1169 | vlan = netdev_priv(dev); |
9bf1907f EB |
1170 | |
1171 | switch (event) { | |
1172 | case NETDEV_REGISTER: | |
1173 | /* Create the device node here after the network device has | |
1174 | * been registered but before register_netdevice has | |
1175 | * finished running. | |
1176 | */ | |
e09eff7f EB |
1177 | err = macvtap_get_minor(vlan); |
1178 | if (err) | |
1179 | return notifier_from_errno(err); | |
1180 | ||
1181 | devt = MKDEV(MAJOR(macvtap_major), vlan->minor); | |
9bf1907f EB |
1182 | classdev = device_create(macvtap_class, &dev->dev, devt, |
1183 | dev, "tap%d", dev->ifindex); | |
e09eff7f EB |
1184 | if (IS_ERR(classdev)) { |
1185 | macvtap_free_minor(vlan); | |
9bf1907f | 1186 | return notifier_from_errno(PTR_ERR(classdev)); |
e09eff7f | 1187 | } |
9bf1907f EB |
1188 | break; |
1189 | case NETDEV_UNREGISTER: | |
e09eff7f | 1190 | devt = MKDEV(MAJOR(macvtap_major), vlan->minor); |
9bf1907f | 1191 | device_destroy(macvtap_class, devt); |
e09eff7f | 1192 | macvtap_free_minor(vlan); |
9bf1907f EB |
1193 | break; |
1194 | } | |
1195 | ||
1196 | return NOTIFY_DONE; | |
1197 | } | |
1198 | ||
1199 | static struct notifier_block macvtap_notifier_block __read_mostly = { | |
1200 | .notifier_call = macvtap_device_event, | |
1201 | }; | |
1202 | ||
20d29d7a AB |
1203 | static int macvtap_init(void) |
1204 | { | |
1205 | int err; | |
1206 | ||
1207 | err = alloc_chrdev_region(&macvtap_major, 0, | |
1208 | MACVTAP_NUM_DEVS, "macvtap"); | |
1209 | if (err) | |
1210 | goto out1; | |
1211 | ||
1212 | cdev_init(&macvtap_cdev, &macvtap_fops); | |
1213 | err = cdev_add(&macvtap_cdev, macvtap_major, MACVTAP_NUM_DEVS); | |
1214 | if (err) | |
1215 | goto out2; | |
1216 | ||
1217 | macvtap_class = class_create(THIS_MODULE, "macvtap"); | |
1218 | if (IS_ERR(macvtap_class)) { | |
1219 | err = PTR_ERR(macvtap_class); | |
1220 | goto out3; | |
1221 | } | |
1222 | ||
9bf1907f | 1223 | err = register_netdevice_notifier(&macvtap_notifier_block); |
20d29d7a AB |
1224 | if (err) |
1225 | goto out4; | |
1226 | ||
9bf1907f EB |
1227 | err = macvlan_link_register(&macvtap_link_ops); |
1228 | if (err) | |
1229 | goto out5; | |
1230 | ||
20d29d7a AB |
1231 | return 0; |
1232 | ||
9bf1907f EB |
1233 | out5: |
1234 | unregister_netdevice_notifier(&macvtap_notifier_block); | |
20d29d7a AB |
1235 | out4: |
1236 | class_unregister(macvtap_class); | |
1237 | out3: | |
1238 | cdev_del(&macvtap_cdev); | |
1239 | out2: | |
1240 | unregister_chrdev_region(macvtap_major, MACVTAP_NUM_DEVS); | |
1241 | out1: | |
1242 | return err; | |
1243 | } | |
1244 | module_init(macvtap_init); | |
1245 | ||
1246 | static void macvtap_exit(void) | |
1247 | { | |
1248 | rtnl_link_unregister(&macvtap_link_ops); | |
9bf1907f | 1249 | unregister_netdevice_notifier(&macvtap_notifier_block); |
20d29d7a AB |
1250 | class_unregister(macvtap_class); |
1251 | cdev_del(&macvtap_cdev); | |
1252 | unregister_chrdev_region(macvtap_major, MACVTAP_NUM_DEVS); | |
1253 | } | |
1254 | module_exit(macvtap_exit); | |
1255 | ||
1256 | MODULE_ALIAS_RTNL_LINK("macvtap"); | |
1257 | MODULE_AUTHOR("Arnd Bergmann <arnd@arndb.de>"); | |
1258 | MODULE_LICENSE("GPL"); |