]>
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 AB |
19 | |
20 | #include <net/net_namespace.h> | |
21 | #include <net/rtnetlink.h> | |
22 | #include <net/sock.h> | |
b9fb9ee0 | 23 | #include <linux/virtio_net.h> |
20d29d7a AB |
24 | |
25 | /* | |
26 | * A macvtap queue is the central object of this driver, it connects | |
27 | * an open character device to a macvlan interface. There can be | |
28 | * multiple queues on one interface, which map back to queues | |
29 | * implemented in hardware on the underlying device. | |
30 | * | |
31 | * macvtap_proto is used to allocate queues through the sock allocation | |
32 | * mechanism. | |
33 | * | |
20d29d7a AB |
34 | */ |
35 | struct macvtap_queue { | |
36 | struct sock sk; | |
37 | struct socket sock; | |
43815482 | 38 | struct socket_wq wq; |
55afbd08 | 39 | int vnet_hdr_sz; |
13707f9e | 40 | struct macvlan_dev __rcu *vlan; |
20d29d7a | 41 | struct file *file; |
b9fb9ee0 | 42 | unsigned int flags; |
376b1aab | 43 | u16 queue_index; |
815f236d JW |
44 | bool enabled; |
45 | struct list_head next; | |
20d29d7a AB |
46 | }; |
47 | ||
01b07fb3 MT |
48 | #define MACVTAP_FEATURES (IFF_VNET_HDR | IFF_MULTI_QUEUE) |
49 | ||
50 | #define MACVTAP_VNET_LE 0x80000000 | |
6ae7feb3 | 51 | |
5b11e15f GK |
52 | static inline bool macvtap_is_little_endian(struct macvtap_queue *q) |
53 | { | |
54 | return q->flags & MACVTAP_VNET_LE; | |
55 | } | |
56 | ||
6ae7feb3 MT |
57 | static inline u16 macvtap16_to_cpu(struct macvtap_queue *q, __virtio16 val) |
58 | { | |
5b11e15f | 59 | return __virtio16_to_cpu(macvtap_is_little_endian(q), val); |
6ae7feb3 MT |
60 | } |
61 | ||
62 | static inline __virtio16 cpu_to_macvtap16(struct macvtap_queue *q, u16 val) | |
63 | { | |
5b11e15f | 64 | return __cpu_to_virtio16(macvtap_is_little_endian(q), val); |
6ae7feb3 MT |
65 | } |
66 | ||
20d29d7a AB |
67 | static struct proto macvtap_proto = { |
68 | .name = "macvtap", | |
69 | .owner = THIS_MODULE, | |
70 | .obj_size = sizeof (struct macvtap_queue), | |
71 | }; | |
72 | ||
73 | /* | |
e09eff7f | 74 | * Variables for dealing with macvtaps device numbers. |
20d29d7a | 75 | */ |
1ebed71a | 76 | static dev_t macvtap_major; |
e09eff7f EB |
77 | #define MACVTAP_NUM_DEVS (1U << MINORBITS) |
78 | static DEFINE_MUTEX(minor_lock); | |
79 | static DEFINE_IDR(minor_idr); | |
80 | ||
97bc3633 | 81 | #define GOODCOPY_LEN 128 |
20d29d7a AB |
82 | static struct class *macvtap_class; |
83 | static struct cdev macvtap_cdev; | |
84 | ||
501c774c AB |
85 | static const struct proto_ops macvtap_socket_ops; |
86 | ||
2be5c767 | 87 | #define TUN_OFFLOADS (NETIF_F_HW_CSUM | NETIF_F_TSO_ECN | NETIF_F_TSO | \ |
e3e3c423 | 88 | NETIF_F_TSO6 | NETIF_F_UFO) |
2be5c767 | 89 | #define RX_OFFLOADS (NETIF_F_GRO | NETIF_F_LRO) |
a567dd62 VY |
90 | #define TAP_FEATURES (NETIF_F_GSO | NETIF_F_SG) |
91 | ||
6acf54f1 VY |
92 | static struct macvlan_dev *macvtap_get_vlan_rcu(const struct net_device *dev) |
93 | { | |
94 | return rcu_dereference(dev->rx_handler_data); | |
95 | } | |
96 | ||
20d29d7a AB |
97 | /* |
98 | * RCU usage: | |
02df55d2 AB |
99 | * The macvtap_queue and the macvlan_dev are loosely coupled, the |
100 | * pointers from one to the other can only be read while rcu_read_lock | |
441ac0fc | 101 | * or rtnl is held. |
20d29d7a | 102 | * |
02df55d2 AB |
103 | * Both the file and the macvlan_dev hold a reference on the macvtap_queue |
104 | * through sock_hold(&q->sk). When the macvlan_dev goes away first, | |
105 | * q->vlan becomes inaccessible. When the files gets closed, | |
106 | * macvtap_get_queue() fails. | |
20d29d7a | 107 | * |
02df55d2 AB |
108 | * There may still be references to the struct sock inside of the |
109 | * queue from outbound SKBs, but these never reference back to the | |
110 | * file or the dev. The data structure is freed through __sk_free | |
111 | * when both our references and any pending SKBs are gone. | |
20d29d7a | 112 | */ |
20d29d7a | 113 | |
815f236d | 114 | static int macvtap_enable_queue(struct net_device *dev, struct file *file, |
20d29d7a | 115 | struct macvtap_queue *q) |
815f236d JW |
116 | { |
117 | struct macvlan_dev *vlan = netdev_priv(dev); | |
118 | int err = -EINVAL; | |
119 | ||
441ac0fc | 120 | ASSERT_RTNL(); |
815f236d JW |
121 | |
122 | if (q->enabled) | |
123 | goto out; | |
124 | ||
125 | err = 0; | |
126 | rcu_assign_pointer(vlan->taps[vlan->numvtaps], q); | |
127 | q->queue_index = vlan->numvtaps; | |
128 | q->enabled = true; | |
129 | ||
130 | vlan->numvtaps++; | |
131 | out: | |
815f236d JW |
132 | return err; |
133 | } | |
134 | ||
40b8fe45 | 135 | /* Requires RTNL */ |
815f236d JW |
136 | static int macvtap_set_queue(struct net_device *dev, struct file *file, |
137 | struct macvtap_queue *q) | |
20d29d7a AB |
138 | { |
139 | struct macvlan_dev *vlan = netdev_priv(dev); | |
20d29d7a | 140 | |
815f236d | 141 | if (vlan->numqueues == MAX_MACVTAP_QUEUES) |
40b8fe45 | 142 | return -EBUSY; |
20d29d7a | 143 | |
02df55d2 | 144 | rcu_assign_pointer(q->vlan, vlan); |
376b1aab | 145 | rcu_assign_pointer(vlan->taps[vlan->numvtaps], q); |
02df55d2 | 146 | sock_hold(&q->sk); |
20d29d7a AB |
147 | |
148 | q->file = file; | |
376b1aab | 149 | q->queue_index = vlan->numvtaps; |
815f236d | 150 | q->enabled = true; |
02df55d2 | 151 | file->private_data = q; |
815f236d | 152 | list_add_tail(&q->next, &vlan->queue_list); |
20d29d7a | 153 | |
1565c7c1 | 154 | vlan->numvtaps++; |
815f236d | 155 | vlan->numqueues++; |
1565c7c1 | 156 | |
40b8fe45 | 157 | return 0; |
20d29d7a AB |
158 | } |
159 | ||
441ac0fc | 160 | static int macvtap_disable_queue(struct macvtap_queue *q) |
815f236d JW |
161 | { |
162 | struct macvlan_dev *vlan; | |
163 | struct macvtap_queue *nq; | |
164 | ||
441ac0fc | 165 | ASSERT_RTNL(); |
815f236d JW |
166 | if (!q->enabled) |
167 | return -EINVAL; | |
168 | ||
441ac0fc VY |
169 | vlan = rtnl_dereference(q->vlan); |
170 | ||
815f236d JW |
171 | if (vlan) { |
172 | int index = q->queue_index; | |
173 | BUG_ON(index >= vlan->numvtaps); | |
441ac0fc | 174 | nq = rtnl_dereference(vlan->taps[vlan->numvtaps - 1]); |
815f236d JW |
175 | nq->queue_index = index; |
176 | ||
177 | rcu_assign_pointer(vlan->taps[index], nq); | |
178 | RCU_INIT_POINTER(vlan->taps[vlan->numvtaps - 1], NULL); | |
179 | q->enabled = false; | |
180 | ||
181 | vlan->numvtaps--; | |
182 | } | |
183 | ||
184 | return 0; | |
185 | } | |
186 | ||
20d29d7a | 187 | /* |
02df55d2 AB |
188 | * The file owning the queue got closed, give up both |
189 | * the reference that the files holds as well as the | |
190 | * one from the macvlan_dev if that still exists. | |
20d29d7a AB |
191 | * |
192 | * Using the spinlock makes sure that we don't get | |
193 | * to the queue again after destroying it. | |
20d29d7a | 194 | */ |
02df55d2 | 195 | static void macvtap_put_queue(struct macvtap_queue *q) |
20d29d7a | 196 | { |
02df55d2 | 197 | struct macvlan_dev *vlan; |
20d29d7a | 198 | |
441ac0fc VY |
199 | rtnl_lock(); |
200 | vlan = rtnl_dereference(q->vlan); | |
201 | ||
02df55d2 | 202 | if (vlan) { |
815f236d | 203 | if (q->enabled) |
441ac0fc | 204 | BUG_ON(macvtap_disable_queue(q)); |
376b1aab | 205 | |
815f236d | 206 | vlan->numqueues--; |
2cfa5a04 | 207 | RCU_INIT_POINTER(q->vlan, NULL); |
02df55d2 | 208 | sock_put(&q->sk); |
815f236d | 209 | list_del_init(&q->next); |
20d29d7a AB |
210 | } |
211 | ||
441ac0fc | 212 | rtnl_unlock(); |
20d29d7a AB |
213 | |
214 | synchronize_rcu(); | |
215 | sock_put(&q->sk); | |
216 | } | |
217 | ||
218 | /* | |
1565c7c1 KK |
219 | * Select a queue based on the rxq of the device on which this packet |
220 | * arrived. If the incoming device is not mq, calculate a flow hash | |
221 | * to select a queue. If all fails, find the first available queue. | |
222 | * Cache vlan->numvtaps since it can become zero during the execution | |
223 | * of this function. | |
20d29d7a AB |
224 | */ |
225 | static struct macvtap_queue *macvtap_get_queue(struct net_device *dev, | |
226 | struct sk_buff *skb) | |
227 | { | |
228 | struct macvlan_dev *vlan = netdev_priv(dev); | |
1565c7c1 | 229 | struct macvtap_queue *tap = NULL; |
815f236d JW |
230 | /* Access to taps array is protected by rcu, but access to numvtaps |
231 | * isn't. Below we use it to lookup a queue, but treat it as a hint | |
232 | * and validate that the result isn't NULL - in case we are | |
233 | * racing against queue removal. | |
234 | */ | |
ed0483fa | 235 | int numvtaps = ACCESS_ONCE(vlan->numvtaps); |
1565c7c1 KK |
236 | __u32 rxq; |
237 | ||
238 | if (!numvtaps) | |
239 | goto out; | |
240 | ||
ef0002b5 | 241 | /* Check if we can use flow to select a queue */ |
3958afa1 | 242 | rxq = skb_get_hash(skb); |
ef0002b5 KK |
243 | if (rxq) { |
244 | tap = rcu_dereference(vlan->taps[rxq % numvtaps]); | |
376b1aab | 245 | goto out; |
ef0002b5 KK |
246 | } |
247 | ||
1565c7c1 KK |
248 | if (likely(skb_rx_queue_recorded(skb))) { |
249 | rxq = skb_get_rx_queue(skb); | |
20d29d7a | 250 | |
1565c7c1 KK |
251 | while (unlikely(rxq >= numvtaps)) |
252 | rxq -= numvtaps; | |
253 | ||
254 | tap = rcu_dereference(vlan->taps[rxq]); | |
376b1aab | 255 | goto out; |
1565c7c1 KK |
256 | } |
257 | ||
376b1aab | 258 | tap = rcu_dereference(vlan->taps[0]); |
1565c7c1 KK |
259 | out: |
260 | return tap; | |
20d29d7a AB |
261 | } |
262 | ||
02df55d2 AB |
263 | /* |
264 | * The net_device is going away, give up the reference | |
1565c7c1 KK |
265 | * that it holds on all queues and safely set the pointer |
266 | * from the queues to NULL. | |
02df55d2 | 267 | */ |
20d29d7a AB |
268 | static void macvtap_del_queues(struct net_device *dev) |
269 | { | |
270 | struct macvlan_dev *vlan = netdev_priv(dev); | |
815f236d | 271 | struct macvtap_queue *q, *tmp, *qlist[MAX_MACVTAP_QUEUES]; |
1565c7c1 | 272 | int i, j = 0; |
02df55d2 | 273 | |
441ac0fc | 274 | ASSERT_RTNL(); |
815f236d JW |
275 | list_for_each_entry_safe(q, tmp, &vlan->queue_list, next) { |
276 | list_del_init(&q->next); | |
376b1aab | 277 | qlist[j++] = q; |
376b1aab | 278 | RCU_INIT_POINTER(q->vlan, NULL); |
815f236d JW |
279 | if (q->enabled) |
280 | vlan->numvtaps--; | |
281 | vlan->numqueues--; | |
564517e8 | 282 | } |
815f236d JW |
283 | for (i = 0; i < vlan->numvtaps; i++) |
284 | RCU_INIT_POINTER(vlan->taps[i], NULL); | |
285 | BUG_ON(vlan->numvtaps); | |
286 | BUG_ON(vlan->numqueues); | |
99f34b38 EB |
287 | /* guarantee that any future macvtap_set_queue will fail */ |
288 | vlan->numvtaps = MAX_MACVTAP_QUEUES; | |
1565c7c1 KK |
289 | |
290 | for (--j; j >= 0; j--) | |
291 | sock_put(&qlist[j]->sk); | |
20d29d7a AB |
292 | } |
293 | ||
6acf54f1 | 294 | static rx_handler_result_t macvtap_handle_frame(struct sk_buff **pskb) |
20d29d7a | 295 | { |
6acf54f1 VY |
296 | struct sk_buff *skb = *pskb; |
297 | struct net_device *dev = skb->dev; | |
298 | struct macvlan_dev *vlan; | |
299 | struct macvtap_queue *q; | |
a567dd62 VY |
300 | netdev_features_t features = TAP_FEATURES; |
301 | ||
6acf54f1 VY |
302 | vlan = macvtap_get_vlan_rcu(dev); |
303 | if (!vlan) | |
304 | return RX_HANDLER_PASS; | |
305 | ||
306 | q = macvtap_get_queue(dev, skb); | |
20d29d7a | 307 | if (!q) |
6acf54f1 | 308 | return RX_HANDLER_PASS; |
8a35747a HX |
309 | |
310 | if (skb_queue_len(&q->sk.sk_receive_queue) >= dev->tx_queue_len) | |
311 | goto drop; | |
20d29d7a | 312 | |
6acf54f1 VY |
313 | skb_push(skb, ETH_HLEN); |
314 | ||
3e4f8b78 | 315 | /* Apply the forward feature mask so that we perform segmentation |
e5733321 VY |
316 | * according to users wishes. This only works if VNET_HDR is |
317 | * enabled. | |
3e4f8b78 | 318 | */ |
e5733321 VY |
319 | if (q->flags & IFF_VNET_HDR) |
320 | features |= vlan->tap_features; | |
8b86a61d | 321 | if (netif_needs_gso(skb, features)) { |
3e4f8b78 VY |
322 | struct sk_buff *segs = __skb_gso_segment(skb, features, false); |
323 | ||
324 | if (IS_ERR(segs)) | |
325 | goto drop; | |
326 | ||
327 | if (!segs) { | |
328 | skb_queue_tail(&q->sk.sk_receive_queue, skb); | |
329 | goto wake_up; | |
330 | } | |
331 | ||
332 | kfree_skb(skb); | |
333 | while (segs) { | |
334 | struct sk_buff *nskb = segs->next; | |
335 | ||
336 | segs->next = NULL; | |
337 | skb_queue_tail(&q->sk.sk_receive_queue, segs); | |
338 | segs = nskb; | |
339 | } | |
340 | } else { | |
cbdb0427 VY |
341 | /* If we receive a partial checksum and the tap side |
342 | * doesn't support checksum offload, compute the checksum. | |
343 | * Note: it doesn't matter which checksum feature to | |
344 | * check, we either support them all or none. | |
345 | */ | |
346 | if (skb->ip_summed == CHECKSUM_PARTIAL && | |
347 | !(features & NETIF_F_ALL_CSUM) && | |
348 | skb_checksum_help(skb)) | |
349 | goto drop; | |
3e4f8b78 VY |
350 | skb_queue_tail(&q->sk.sk_receive_queue, skb); |
351 | } | |
352 | ||
353 | wake_up: | |
4a4771a5 | 354 | wake_up_interruptible_poll(sk_sleep(&q->sk), POLLIN | POLLRDNORM | POLLRDBAND); |
6acf54f1 | 355 | return RX_HANDLER_CONSUMED; |
8a35747a HX |
356 | |
357 | drop: | |
6acf54f1 VY |
358 | /* Count errors/drops only here, thus don't care about args. */ |
359 | macvlan_count_rx(vlan, 0, 0, 0); | |
8a35747a | 360 | kfree_skb(skb); |
6acf54f1 | 361 | return RX_HANDLER_CONSUMED; |
20d29d7a AB |
362 | } |
363 | ||
e09eff7f EB |
364 | static int macvtap_get_minor(struct macvlan_dev *vlan) |
365 | { | |
366 | int retval = -ENOMEM; | |
e09eff7f EB |
367 | |
368 | mutex_lock(&minor_lock); | |
ec09ebc1 TH |
369 | retval = idr_alloc(&minor_idr, vlan, 1, MACVTAP_NUM_DEVS, GFP_KERNEL); |
370 | if (retval >= 0) { | |
371 | vlan->minor = retval; | |
372 | } else if (retval == -ENOSPC) { | |
e09eff7f EB |
373 | printk(KERN_ERR "too many macvtap devices\n"); |
374 | retval = -EINVAL; | |
e09eff7f | 375 | } |
e09eff7f | 376 | mutex_unlock(&minor_lock); |
ec09ebc1 | 377 | return retval < 0 ? retval : 0; |
e09eff7f EB |
378 | } |
379 | ||
380 | static void macvtap_free_minor(struct macvlan_dev *vlan) | |
381 | { | |
382 | mutex_lock(&minor_lock); | |
383 | if (vlan->minor) { | |
384 | idr_remove(&minor_idr, vlan->minor); | |
385 | vlan->minor = 0; | |
386 | } | |
387 | mutex_unlock(&minor_lock); | |
388 | } | |
389 | ||
390 | static struct net_device *dev_get_by_macvtap_minor(int minor) | |
391 | { | |
392 | struct net_device *dev = NULL; | |
393 | struct macvlan_dev *vlan; | |
394 | ||
395 | mutex_lock(&minor_lock); | |
396 | vlan = idr_find(&minor_idr, minor); | |
397 | if (vlan) { | |
398 | dev = vlan->dev; | |
399 | dev_hold(dev); | |
400 | } | |
401 | mutex_unlock(&minor_lock); | |
402 | return dev; | |
403 | } | |
404 | ||
20d29d7a AB |
405 | static int macvtap_newlink(struct net *src_net, |
406 | struct net_device *dev, | |
407 | struct nlattr *tb[], | |
408 | struct nlattr *data[]) | |
409 | { | |
815f236d | 410 | struct macvlan_dev *vlan = netdev_priv(dev); |
6acf54f1 VY |
411 | int err; |
412 | ||
815f236d JW |
413 | INIT_LIST_HEAD(&vlan->queue_list); |
414 | ||
2be5c767 VY |
415 | /* Since macvlan supports all offloads by default, make |
416 | * tap support all offloads also. | |
417 | */ | |
418 | vlan->tap_features = TUN_OFFLOADS; | |
419 | ||
6acf54f1 VY |
420 | err = netdev_rx_handler_register(dev, macvtap_handle_frame, vlan); |
421 | if (err) | |
422 | return err; | |
423 | ||
9bf1907f EB |
424 | /* Don't put anything that may fail after macvlan_common_newlink |
425 | * because we can't undo what it does. | |
426 | */ | |
2f6a1b66 | 427 | return macvlan_common_newlink(src_net, dev, tb, data); |
20d29d7a AB |
428 | } |
429 | ||
430 | static void macvtap_dellink(struct net_device *dev, | |
431 | struct list_head *head) | |
432 | { | |
6acf54f1 | 433 | netdev_rx_handler_unregister(dev); |
20d29d7a AB |
434 | macvtap_del_queues(dev); |
435 | macvlan_dellink(dev, head); | |
436 | } | |
437 | ||
8a35747a HX |
438 | static void macvtap_setup(struct net_device *dev) |
439 | { | |
440 | macvlan_common_setup(dev); | |
441 | dev->tx_queue_len = TUN_READQ_SIZE; | |
442 | } | |
443 | ||
20d29d7a AB |
444 | static struct rtnl_link_ops macvtap_link_ops __read_mostly = { |
445 | .kind = "macvtap", | |
8a35747a | 446 | .setup = macvtap_setup, |
20d29d7a AB |
447 | .newlink = macvtap_newlink, |
448 | .dellink = macvtap_dellink, | |
449 | }; | |
450 | ||
451 | ||
452 | static void macvtap_sock_write_space(struct sock *sk) | |
453 | { | |
43815482 ED |
454 | wait_queue_head_t *wqueue; |
455 | ||
20d29d7a AB |
456 | if (!sock_writeable(sk) || |
457 | !test_and_clear_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags)) | |
458 | return; | |
459 | ||
43815482 ED |
460 | wqueue = sk_sleep(sk); |
461 | if (wqueue && waitqueue_active(wqueue)) | |
462 | wake_up_interruptible_poll(wqueue, POLLOUT | POLLWRNORM | POLLWRBAND); | |
20d29d7a AB |
463 | } |
464 | ||
2259fef0 EB |
465 | static void macvtap_sock_destruct(struct sock *sk) |
466 | { | |
467 | skb_queue_purge(&sk->sk_receive_queue); | |
468 | } | |
469 | ||
20d29d7a AB |
470 | static int macvtap_open(struct inode *inode, struct file *file) |
471 | { | |
472 | struct net *net = current->nsproxy->net_ns; | |
40b8fe45 | 473 | struct net_device *dev; |
20d29d7a | 474 | struct macvtap_queue *q; |
40b8fe45 | 475 | int err = -ENODEV; |
20d29d7a | 476 | |
40b8fe45 VY |
477 | rtnl_lock(); |
478 | dev = dev_get_by_macvtap_minor(iminor(inode)); | |
20d29d7a AB |
479 | if (!dev) |
480 | goto out; | |
481 | ||
20d29d7a AB |
482 | err = -ENOMEM; |
483 | q = (struct macvtap_queue *)sk_alloc(net, AF_UNSPEC, GFP_KERNEL, | |
484 | &macvtap_proto); | |
485 | if (!q) | |
486 | goto out; | |
487 | ||
d9a90a31 | 488 | RCU_INIT_POINTER(q->sock.wq, &q->wq); |
43815482 | 489 | init_waitqueue_head(&q->wq.wait); |
20d29d7a AB |
490 | q->sock.type = SOCK_RAW; |
491 | q->sock.state = SS_CONNECTED; | |
501c774c AB |
492 | q->sock.file = file; |
493 | q->sock.ops = &macvtap_socket_ops; | |
20d29d7a | 494 | sock_init_data(&q->sock, &q->sk); |
20d29d7a | 495 | q->sk.sk_write_space = macvtap_sock_write_space; |
2259fef0 | 496 | q->sk.sk_destruct = macvtap_sock_destruct; |
b9fb9ee0 | 497 | q->flags = IFF_VNET_HDR | IFF_NO_PI | IFF_TAP; |
55afbd08 | 498 | q->vnet_hdr_sz = sizeof(struct virtio_net_hdr); |
20d29d7a | 499 | |
97bc3633 SM |
500 | /* |
501 | * so far only KVM virtio_net uses macvtap, enable zero copy between | |
502 | * guest kernel and host kernel when lower device supports zerocopy | |
047af9cf EB |
503 | * |
504 | * The macvlan supports zerocopy iff the lower device supports zero | |
505 | * copy so we don't have to look at the lower device directly. | |
97bc3633 | 506 | */ |
047af9cf EB |
507 | if ((dev->features & NETIF_F_HIGHDMA) && (dev->features & NETIF_F_SG)) |
508 | sock_set_flag(&q->sk, SOCK_ZEROCOPY); | |
97bc3633 | 509 | |
20d29d7a AB |
510 | err = macvtap_set_queue(dev, file, q); |
511 | if (err) | |
512 | sock_put(&q->sk); | |
513 | ||
514 | out: | |
515 | if (dev) | |
516 | dev_put(dev); | |
517 | ||
40b8fe45 | 518 | rtnl_unlock(); |
20d29d7a AB |
519 | return err; |
520 | } | |
521 | ||
522 | static int macvtap_release(struct inode *inode, struct file *file) | |
523 | { | |
02df55d2 AB |
524 | struct macvtap_queue *q = file->private_data; |
525 | macvtap_put_queue(q); | |
20d29d7a AB |
526 | return 0; |
527 | } | |
528 | ||
529 | static unsigned int macvtap_poll(struct file *file, poll_table * wait) | |
530 | { | |
02df55d2 | 531 | struct macvtap_queue *q = file->private_data; |
20d29d7a AB |
532 | unsigned int mask = POLLERR; |
533 | ||
534 | if (!q) | |
535 | goto out; | |
536 | ||
537 | mask = 0; | |
43815482 | 538 | poll_wait(file, &q->wq.wait, wait); |
20d29d7a AB |
539 | |
540 | if (!skb_queue_empty(&q->sk.sk_receive_queue)) | |
541 | mask |= POLLIN | POLLRDNORM; | |
542 | ||
543 | if (sock_writeable(&q->sk) || | |
544 | (!test_and_set_bit(SOCK_ASYNC_NOSPACE, &q->sock.flags) && | |
545 | sock_writeable(&q->sk))) | |
546 | mask |= POLLOUT | POLLWRNORM; | |
547 | ||
548 | out: | |
20d29d7a AB |
549 | return mask; |
550 | } | |
551 | ||
b9fb9ee0 AB |
552 | static inline struct sk_buff *macvtap_alloc_skb(struct sock *sk, size_t prepad, |
553 | size_t len, size_t linear, | |
554 | int noblock, int *err) | |
555 | { | |
556 | struct sk_buff *skb; | |
557 | ||
558 | /* Under a page? Don't bother with paged skb. */ | |
559 | if (prepad + len < PAGE_SIZE || !linear) | |
560 | linear = len; | |
561 | ||
562 | skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock, | |
28d64271 | 563 | err, 0); |
b9fb9ee0 AB |
564 | if (!skb) |
565 | return NULL; | |
566 | ||
567 | skb_reserve(skb, prepad); | |
568 | skb_put(skb, linear); | |
569 | skb->data_len = len - linear; | |
570 | skb->len += len - linear; | |
571 | ||
572 | return skb; | |
573 | } | |
574 | ||
575 | /* | |
576 | * macvtap_skb_from_vnet_hdr and macvtap_skb_to_vnet_hdr should | |
577 | * be shared with the tun/tap driver. | |
578 | */ | |
6ae7feb3 MT |
579 | static int macvtap_skb_from_vnet_hdr(struct macvtap_queue *q, |
580 | struct sk_buff *skb, | |
b9fb9ee0 AB |
581 | struct virtio_net_hdr *vnet_hdr) |
582 | { | |
583 | unsigned short gso_type = 0; | |
584 | if (vnet_hdr->gso_type != VIRTIO_NET_HDR_GSO_NONE) { | |
585 | switch (vnet_hdr->gso_type & ~VIRTIO_NET_HDR_GSO_ECN) { | |
586 | case VIRTIO_NET_HDR_GSO_TCPV4: | |
587 | gso_type = SKB_GSO_TCPV4; | |
588 | break; | |
589 | case VIRTIO_NET_HDR_GSO_TCPV6: | |
590 | gso_type = SKB_GSO_TCPV6; | |
591 | break; | |
592 | case VIRTIO_NET_HDR_GSO_UDP: | |
593 | gso_type = SKB_GSO_UDP; | |
594 | break; | |
595 | default: | |
596 | return -EINVAL; | |
597 | } | |
598 | ||
599 | if (vnet_hdr->gso_type & VIRTIO_NET_HDR_GSO_ECN) | |
600 | gso_type |= SKB_GSO_TCP_ECN; | |
601 | ||
602 | if (vnet_hdr->gso_size == 0) | |
603 | return -EINVAL; | |
604 | } | |
605 | ||
606 | if (vnet_hdr->flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) { | |
6ae7feb3 MT |
607 | if (!skb_partial_csum_set(skb, macvtap16_to_cpu(q, vnet_hdr->csum_start), |
608 | macvtap16_to_cpu(q, vnet_hdr->csum_offset))) | |
b9fb9ee0 AB |
609 | return -EINVAL; |
610 | } | |
611 | ||
612 | if (vnet_hdr->gso_type != VIRTIO_NET_HDR_GSO_NONE) { | |
6ae7feb3 | 613 | skb_shinfo(skb)->gso_size = macvtap16_to_cpu(q, vnet_hdr->gso_size); |
c9af6db4 | 614 | skb_shinfo(skb)->gso_type = gso_type; |
b9fb9ee0 AB |
615 | |
616 | /* Header must be checked, and gso_segs computed. */ | |
617 | skb_shinfo(skb)->gso_type |= SKB_GSO_DODGY; | |
618 | skb_shinfo(skb)->gso_segs = 0; | |
619 | } | |
620 | return 0; | |
621 | } | |
622 | ||
6ae7feb3 MT |
623 | static void macvtap_skb_to_vnet_hdr(struct macvtap_queue *q, |
624 | const struct sk_buff *skb, | |
625 | struct virtio_net_hdr *vnet_hdr) | |
b9fb9ee0 AB |
626 | { |
627 | memset(vnet_hdr, 0, sizeof(*vnet_hdr)); | |
628 | ||
629 | if (skb_is_gso(skb)) { | |
630 | struct skb_shared_info *sinfo = skb_shinfo(skb); | |
631 | ||
632 | /* This is a hint as to how much should be linear. */ | |
6ae7feb3 MT |
633 | vnet_hdr->hdr_len = cpu_to_macvtap16(q, skb_headlen(skb)); |
634 | vnet_hdr->gso_size = cpu_to_macvtap16(q, sinfo->gso_size); | |
b9fb9ee0 AB |
635 | if (sinfo->gso_type & SKB_GSO_TCPV4) |
636 | vnet_hdr->gso_type = VIRTIO_NET_HDR_GSO_TCPV4; | |
637 | else if (sinfo->gso_type & SKB_GSO_TCPV6) | |
638 | vnet_hdr->gso_type = VIRTIO_NET_HDR_GSO_TCPV6; | |
e3e3c423 VY |
639 | else if (sinfo->gso_type & SKB_GSO_UDP) |
640 | vnet_hdr->gso_type = VIRTIO_NET_HDR_GSO_UDP; | |
b9fb9ee0 AB |
641 | else |
642 | BUG(); | |
643 | if (sinfo->gso_type & SKB_GSO_TCP_ECN) | |
644 | vnet_hdr->gso_type |= VIRTIO_NET_HDR_GSO_ECN; | |
645 | } else | |
646 | vnet_hdr->gso_type = VIRTIO_NET_HDR_GSO_NONE; | |
647 | ||
648 | if (skb->ip_summed == CHECKSUM_PARTIAL) { | |
649 | vnet_hdr->flags = VIRTIO_NET_HDR_F_NEEDS_CSUM; | |
df8a39de | 650 | if (skb_vlan_tag_present(skb)) |
6ae7feb3 MT |
651 | vnet_hdr->csum_start = cpu_to_macvtap16(q, |
652 | skb_checksum_start_offset(skb) + VLAN_HLEN); | |
653 | else | |
654 | vnet_hdr->csum_start = cpu_to_macvtap16(q, | |
655 | skb_checksum_start_offset(skb)); | |
656 | vnet_hdr->csum_offset = cpu_to_macvtap16(q, skb->csum_offset); | |
10a8d94a JW |
657 | } else if (skb->ip_summed == CHECKSUM_UNNECESSARY) { |
658 | vnet_hdr->flags = VIRTIO_NET_HDR_F_DATA_VALID; | |
b9fb9ee0 | 659 | } /* else everything is zero */ |
b9fb9ee0 AB |
660 | } |
661 | ||
2f1d8b9e ED |
662 | /* Neighbour code has some assumptions on HH_DATA_MOD alignment */ |
663 | #define MACVTAP_RESERVE HH_DATA_OFF(ETH_HLEN) | |
664 | ||
20d29d7a | 665 | /* Get packet from user space buffer */ |
97bc3633 | 666 | static ssize_t macvtap_get_user(struct macvtap_queue *q, struct msghdr *m, |
f5ff53b4 | 667 | struct iov_iter *from, int noblock) |
20d29d7a | 668 | { |
2f1d8b9e | 669 | int good_linear = SKB_MAX_HEAD(MACVTAP_RESERVE); |
20d29d7a | 670 | struct sk_buff *skb; |
02df55d2 | 671 | struct macvlan_dev *vlan; |
f5ff53b4 | 672 | unsigned long total_len = iov_iter_count(from); |
97bc3633 | 673 | unsigned long len = total_len; |
20d29d7a | 674 | int err; |
b9fb9ee0 AB |
675 | struct virtio_net_hdr vnet_hdr = { 0 }; |
676 | int vnet_hdr_len = 0; | |
b92946e2 | 677 | int copylen = 0; |
97bc3633 | 678 | bool zerocopy = false; |
61d46bf9 | 679 | size_t linear; |
f5ff53b4 | 680 | ssize_t n; |
b9fb9ee0 AB |
681 | |
682 | if (q->flags & IFF_VNET_HDR) { | |
55afbd08 | 683 | vnet_hdr_len = q->vnet_hdr_sz; |
b9fb9ee0 AB |
684 | |
685 | err = -EINVAL; | |
ce3c8692 | 686 | if (len < vnet_hdr_len) |
b9fb9ee0 | 687 | goto err; |
ce3c8692 | 688 | len -= vnet_hdr_len; |
b9fb9ee0 | 689 | |
f5ff53b4 AV |
690 | err = -EFAULT; |
691 | n = copy_from_iter(&vnet_hdr, sizeof(vnet_hdr), from); | |
692 | if (n != sizeof(vnet_hdr)) | |
b9fb9ee0 | 693 | goto err; |
f5ff53b4 | 694 | iov_iter_advance(from, vnet_hdr_len - sizeof(vnet_hdr)); |
b9fb9ee0 | 695 | if ((vnet_hdr.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) && |
6ae7feb3 MT |
696 | macvtap16_to_cpu(q, vnet_hdr.csum_start) + |
697 | macvtap16_to_cpu(q, vnet_hdr.csum_offset) + 2 > | |
698 | macvtap16_to_cpu(q, vnet_hdr.hdr_len)) | |
699 | vnet_hdr.hdr_len = cpu_to_macvtap16(q, | |
700 | macvtap16_to_cpu(q, vnet_hdr.csum_start) + | |
701 | macvtap16_to_cpu(q, vnet_hdr.csum_offset) + 2); | |
b9fb9ee0 | 702 | err = -EINVAL; |
6ae7feb3 | 703 | if (macvtap16_to_cpu(q, vnet_hdr.hdr_len) > len) |
b9fb9ee0 AB |
704 | goto err; |
705 | } | |
20d29d7a | 706 | |
b9fb9ee0 | 707 | err = -EINVAL; |
20d29d7a | 708 | if (unlikely(len < ETH_HLEN)) |
b9fb9ee0 | 709 | goto err; |
20d29d7a | 710 | |
ece793fc | 711 | if (m && m->msg_control && sock_flag(&q->sk, SOCK_ZEROCOPY)) { |
f5ff53b4 AV |
712 | struct iov_iter i; |
713 | ||
6ae7feb3 MT |
714 | copylen = vnet_hdr.hdr_len ? |
715 | macvtap16_to_cpu(q, vnet_hdr.hdr_len) : GOODCOPY_LEN; | |
16a3fa28 JW |
716 | if (copylen > good_linear) |
717 | copylen = good_linear; | |
61d46bf9 | 718 | linear = copylen; |
f5ff53b4 AV |
719 | i = *from; |
720 | iov_iter_advance(&i, copylen); | |
721 | if (iov_iter_npages(&i, INT_MAX) <= MAX_SKB_FRAGS) | |
ece793fc JW |
722 | zerocopy = true; |
723 | } | |
724 | ||
725 | if (!zerocopy) { | |
97bc3633 | 726 | copylen = len; |
6ae7feb3 | 727 | if (macvtap16_to_cpu(q, vnet_hdr.hdr_len) > good_linear) |
16a3fa28 JW |
728 | linear = good_linear; |
729 | else | |
6ae7feb3 | 730 | linear = macvtap16_to_cpu(q, vnet_hdr.hdr_len); |
61d46bf9 | 731 | } |
97bc3633 | 732 | |
2f1d8b9e | 733 | skb = macvtap_alloc_skb(&q->sk, MACVTAP_RESERVE, copylen, |
61d46bf9 | 734 | linear, noblock, &err); |
02df55d2 AB |
735 | if (!skb) |
736 | goto err; | |
20d29d7a | 737 | |
01d6657b | 738 | if (zerocopy) |
f5ff53b4 | 739 | err = zerocopy_sg_from_iter(skb, from); |
ece793fc | 740 | else { |
f5ff53b4 | 741 | err = skb_copy_datagram_from_iter(skb, 0, from, len); |
ece793fc JW |
742 | if (!err && m && m->msg_control) { |
743 | struct ubuf_info *uarg = m->msg_control; | |
744 | uarg->callback(uarg, false); | |
745 | } | |
746 | } | |
747 | ||
02df55d2 | 748 | if (err) |
b9fb9ee0 | 749 | goto err_kfree; |
20d29d7a AB |
750 | |
751 | skb_set_network_header(skb, ETH_HLEN); | |
b9fb9ee0 AB |
752 | skb_reset_mac_header(skb); |
753 | skb->protocol = eth_hdr(skb)->h_proto; | |
754 | ||
755 | if (vnet_hdr_len) { | |
6ae7feb3 | 756 | err = macvtap_skb_from_vnet_hdr(q, skb, &vnet_hdr); |
b9fb9ee0 AB |
757 | if (err) |
758 | goto err_kfree; | |
759 | } | |
760 | ||
40893fd0 | 761 | skb_probe_transport_header(skb, ETH_HLEN); |
9b4d669b | 762 | |
ac4e4af1 VY |
763 | rcu_read_lock(); |
764 | vlan = rcu_dereference(q->vlan); | |
97bc3633 | 765 | /* copy skb_ubuf_info for callback when skb has no error */ |
01d6657b | 766 | if (zerocopy) { |
97bc3633 | 767 | skb_shinfo(skb)->destructor_arg = m->msg_control; |
01d6657b | 768 | skb_shinfo(skb)->tx_flags |= SKBTX_DEV_ZEROCOPY; |
c9af6db4 | 769 | skb_shinfo(skb)->tx_flags |= SKBTX_SHARED_FRAG; |
01d6657b | 770 | } |
29d79196 | 771 | if (vlan) { |
6acf54f1 VY |
772 | skb->dev = vlan->dev; |
773 | dev_queue_xmit(skb); | |
29d79196 | 774 | } else { |
02df55d2 | 775 | kfree_skb(skb); |
29d79196 | 776 | } |
ac4e4af1 | 777 | rcu_read_unlock(); |
20d29d7a | 778 | |
97bc3633 | 779 | return total_len; |
02df55d2 | 780 | |
b9fb9ee0 AB |
781 | err_kfree: |
782 | kfree_skb(skb); | |
783 | ||
02df55d2 | 784 | err: |
ac4e4af1 VY |
785 | rcu_read_lock(); |
786 | vlan = rcu_dereference(q->vlan); | |
02df55d2 | 787 | if (vlan) |
cd3e22b7 | 788 | this_cpu_inc(vlan->pcpu_stats->tx_dropped); |
ac4e4af1 | 789 | rcu_read_unlock(); |
02df55d2 | 790 | |
02df55d2 | 791 | return err; |
20d29d7a AB |
792 | } |
793 | ||
f5ff53b4 | 794 | static ssize_t macvtap_write_iter(struct kiocb *iocb, struct iov_iter *from) |
20d29d7a AB |
795 | { |
796 | struct file *file = iocb->ki_filp; | |
02df55d2 | 797 | struct macvtap_queue *q = file->private_data; |
20d29d7a | 798 | |
f5ff53b4 | 799 | return macvtap_get_user(q, NULL, from, file->f_flags & O_NONBLOCK); |
20d29d7a AB |
800 | } |
801 | ||
802 | /* Put packet to the user space buffer */ | |
803 | static ssize_t macvtap_put_user(struct macvtap_queue *q, | |
804 | const struct sk_buff *skb, | |
6c36d2e2 | 805 | struct iov_iter *iter) |
20d29d7a | 806 | { |
20d29d7a | 807 | int ret; |
b9fb9ee0 | 808 | int vnet_hdr_len = 0; |
f09e2249 | 809 | int vlan_offset = 0; |
6c36d2e2 | 810 | int total; |
b9fb9ee0 AB |
811 | |
812 | if (q->flags & IFF_VNET_HDR) { | |
813 | struct virtio_net_hdr vnet_hdr; | |
55afbd08 | 814 | vnet_hdr_len = q->vnet_hdr_sz; |
6c36d2e2 | 815 | if (iov_iter_count(iter) < vnet_hdr_len) |
b9fb9ee0 AB |
816 | return -EINVAL; |
817 | ||
6ae7feb3 | 818 | macvtap_skb_to_vnet_hdr(q, skb, &vnet_hdr); |
b9fb9ee0 | 819 | |
6c36d2e2 HX |
820 | if (copy_to_iter(&vnet_hdr, sizeof(vnet_hdr), iter) != |
821 | sizeof(vnet_hdr)) | |
b9fb9ee0 | 822 | return -EFAULT; |
7cc76f51 JW |
823 | |
824 | iov_iter_advance(iter, vnet_hdr_len - sizeof(vnet_hdr)); | |
b9fb9ee0 | 825 | } |
6c36d2e2 | 826 | total = vnet_hdr_len; |
ce232ce0 | 827 | total += skb->len; |
f09e2249 | 828 | |
df8a39de | 829 | if (skb_vlan_tag_present(skb)) { |
f09e2249 BG |
830 | struct { |
831 | __be16 h_vlan_proto; | |
832 | __be16 h_vlan_TCI; | |
833 | } veth; | |
0fbe0d47 | 834 | veth.h_vlan_proto = skb->vlan_proto; |
df8a39de | 835 | veth.h_vlan_TCI = htons(skb_vlan_tag_get(skb)); |
f09e2249 BG |
836 | |
837 | vlan_offset = offsetof(struct vlan_ethhdr, h_vlan_proto); | |
ce232ce0 | 838 | total += VLAN_HLEN; |
f09e2249 | 839 | |
6c36d2e2 HX |
840 | ret = skb_copy_datagram_iter(skb, 0, iter, vlan_offset); |
841 | if (ret || !iov_iter_count(iter)) | |
f09e2249 BG |
842 | goto done; |
843 | ||
6c36d2e2 HX |
844 | ret = copy_to_iter(&veth, sizeof(veth), iter); |
845 | if (ret != sizeof(veth) || !iov_iter_count(iter)) | |
f09e2249 BG |
846 | goto done; |
847 | } | |
20d29d7a | 848 | |
6c36d2e2 HX |
849 | ret = skb_copy_datagram_iter(skb, vlan_offset, iter, |
850 | skb->len - vlan_offset); | |
20d29d7a | 851 | |
f09e2249 | 852 | done: |
ce232ce0 | 853 | return ret ? ret : total; |
20d29d7a AB |
854 | } |
855 | ||
55ec8e25 | 856 | static ssize_t macvtap_do_read(struct macvtap_queue *q, |
3af0bfe5 | 857 | struct iov_iter *to, |
501c774c | 858 | int noblock) |
20d29d7a | 859 | { |
ccf7e72b | 860 | DEFINE_WAIT(wait); |
20d29d7a | 861 | struct sk_buff *skb; |
501c774c | 862 | ssize_t ret = 0; |
20d29d7a | 863 | |
3af0bfe5 AV |
864 | if (!iov_iter_count(to)) |
865 | return 0; | |
866 | ||
867 | while (1) { | |
89cee917 JW |
868 | if (!noblock) |
869 | prepare_to_wait(sk_sleep(&q->sk), &wait, | |
870 | TASK_INTERRUPTIBLE); | |
20d29d7a AB |
871 | |
872 | /* Read frames from the queue */ | |
873 | skb = skb_dequeue(&q->sk.sk_receive_queue); | |
3af0bfe5 AV |
874 | if (skb) |
875 | break; | |
876 | if (noblock) { | |
877 | ret = -EAGAIN; | |
878 | break; | |
20d29d7a | 879 | } |
3af0bfe5 AV |
880 | if (signal_pending(current)) { |
881 | ret = -ERESTARTSYS; | |
882 | break; | |
883 | } | |
884 | /* Nothing to read, let's sleep */ | |
885 | schedule(); | |
886 | } | |
887 | if (skb) { | |
888 | ret = macvtap_put_user(q, skb, to); | |
f51a5e82 JW |
889 | if (unlikely(ret < 0)) |
890 | kfree_skb(skb); | |
891 | else | |
892 | consume_skb(skb); | |
20d29d7a | 893 | } |
89cee917 JW |
894 | if (!noblock) |
895 | finish_wait(sk_sleep(&q->sk), &wait); | |
501c774c AB |
896 | return ret; |
897 | } | |
898 | ||
3af0bfe5 | 899 | static ssize_t macvtap_read_iter(struct kiocb *iocb, struct iov_iter *to) |
501c774c AB |
900 | { |
901 | struct file *file = iocb->ki_filp; | |
902 | struct macvtap_queue *q = file->private_data; | |
3af0bfe5 | 903 | ssize_t len = iov_iter_count(to), ret; |
20d29d7a | 904 | |
3af0bfe5 | 905 | ret = macvtap_do_read(q, to, file->f_flags & O_NONBLOCK); |
ce232ce0 | 906 | ret = min_t(ssize_t, ret, len); |
e6ebc7f1 ZYW |
907 | if (ret > 0) |
908 | iocb->ki_pos = ret; | |
20d29d7a AB |
909 | return ret; |
910 | } | |
911 | ||
8f475a31 JW |
912 | static struct macvlan_dev *macvtap_get_vlan(struct macvtap_queue *q) |
913 | { | |
914 | struct macvlan_dev *vlan; | |
915 | ||
441ac0fc VY |
916 | ASSERT_RTNL(); |
917 | vlan = rtnl_dereference(q->vlan); | |
8f475a31 JW |
918 | if (vlan) |
919 | dev_hold(vlan->dev); | |
8f475a31 JW |
920 | |
921 | return vlan; | |
922 | } | |
923 | ||
924 | static void macvtap_put_vlan(struct macvlan_dev *vlan) | |
925 | { | |
926 | dev_put(vlan->dev); | |
927 | } | |
928 | ||
815f236d JW |
929 | static int macvtap_ioctl_set_queue(struct file *file, unsigned int flags) |
930 | { | |
931 | struct macvtap_queue *q = file->private_data; | |
932 | struct macvlan_dev *vlan; | |
933 | int ret; | |
934 | ||
935 | vlan = macvtap_get_vlan(q); | |
936 | if (!vlan) | |
937 | return -EINVAL; | |
938 | ||
939 | if (flags & IFF_ATTACH_QUEUE) | |
940 | ret = macvtap_enable_queue(vlan->dev, file, q); | |
941 | else if (flags & IFF_DETACH_QUEUE) | |
942 | ret = macvtap_disable_queue(q); | |
f57855a5 JW |
943 | else |
944 | ret = -EINVAL; | |
815f236d JW |
945 | |
946 | macvtap_put_vlan(vlan); | |
947 | return ret; | |
948 | } | |
949 | ||
2be5c767 VY |
950 | static int set_offload(struct macvtap_queue *q, unsigned long arg) |
951 | { | |
952 | struct macvlan_dev *vlan; | |
953 | netdev_features_t features; | |
954 | netdev_features_t feature_mask = 0; | |
955 | ||
956 | vlan = rtnl_dereference(q->vlan); | |
957 | if (!vlan) | |
958 | return -ENOLINK; | |
959 | ||
960 | features = vlan->dev->features; | |
961 | ||
962 | if (arg & TUN_F_CSUM) { | |
963 | feature_mask = NETIF_F_HW_CSUM; | |
964 | ||
965 | if (arg & (TUN_F_TSO4 | TUN_F_TSO6)) { | |
966 | if (arg & TUN_F_TSO_ECN) | |
967 | feature_mask |= NETIF_F_TSO_ECN; | |
968 | if (arg & TUN_F_TSO4) | |
969 | feature_mask |= NETIF_F_TSO; | |
970 | if (arg & TUN_F_TSO6) | |
971 | feature_mask |= NETIF_F_TSO6; | |
972 | } | |
e3e3c423 VY |
973 | |
974 | if (arg & TUN_F_UFO) | |
975 | feature_mask |= NETIF_F_UFO; | |
2be5c767 VY |
976 | } |
977 | ||
978 | /* tun/tap driver inverts the usage for TSO offloads, where | |
979 | * setting the TSO bit means that the userspace wants to | |
980 | * accept TSO frames and turning it off means that user space | |
981 | * does not support TSO. | |
982 | * For macvtap, we have to invert it to mean the same thing. | |
983 | * When user space turns off TSO, we turn off GSO/LRO so that | |
984 | * user-space will not receive TSO frames. | |
985 | */ | |
e3e3c423 | 986 | if (feature_mask & (NETIF_F_TSO | NETIF_F_TSO6 | NETIF_F_UFO)) |
2be5c767 VY |
987 | features |= RX_OFFLOADS; |
988 | else | |
989 | features &= ~RX_OFFLOADS; | |
990 | ||
991 | /* tap_features are the same as features on tun/tap and | |
992 | * reflect user expectations. | |
993 | */ | |
a567dd62 | 994 | vlan->tap_features = feature_mask; |
2be5c767 VY |
995 | vlan->set_features = features; |
996 | netdev_update_features(vlan->dev); | |
997 | ||
998 | return 0; | |
999 | } | |
1000 | ||
20d29d7a AB |
1001 | /* |
1002 | * provide compatibility with generic tun/tap interface | |
1003 | */ | |
1004 | static long macvtap_ioctl(struct file *file, unsigned int cmd, | |
1005 | unsigned long arg) | |
1006 | { | |
02df55d2 AB |
1007 | struct macvtap_queue *q = file->private_data; |
1008 | struct macvlan_dev *vlan; | |
20d29d7a AB |
1009 | void __user *argp = (void __user *)arg; |
1010 | struct ifreq __user *ifr = argp; | |
1011 | unsigned int __user *up = argp; | |
39ec7de7 | 1012 | unsigned short u; |
55afbd08 MT |
1013 | int __user *sp = argp; |
1014 | int s; | |
02df55d2 | 1015 | int ret; |
20d29d7a AB |
1016 | |
1017 | switch (cmd) { | |
1018 | case TUNSETIFF: | |
1019 | /* ignore the name, just look at flags */ | |
1020 | if (get_user(u, &ifr->ifr_flags)) | |
1021 | return -EFAULT; | |
b9fb9ee0 AB |
1022 | |
1023 | ret = 0; | |
6ae7feb3 | 1024 | if ((u & ~MACVTAP_FEATURES) != (IFF_NO_PI | IFF_TAP)) |
b9fb9ee0 AB |
1025 | ret = -EINVAL; |
1026 | else | |
39ec7de7 | 1027 | q->flags = (q->flags & ~MACVTAP_FEATURES) | u; |
b9fb9ee0 AB |
1028 | |
1029 | return ret; | |
20d29d7a AB |
1030 | |
1031 | case TUNGETIFF: | |
441ac0fc | 1032 | rtnl_lock(); |
8f475a31 | 1033 | vlan = macvtap_get_vlan(q); |
441ac0fc VY |
1034 | if (!vlan) { |
1035 | rtnl_unlock(); | |
20d29d7a | 1036 | return -ENOLINK; |
441ac0fc | 1037 | } |
20d29d7a | 1038 | |
02df55d2 | 1039 | ret = 0; |
39ec7de7 | 1040 | u = q->flags; |
13707f9e | 1041 | if (copy_to_user(&ifr->ifr_name, vlan->dev->name, IFNAMSIZ) || |
39ec7de7 | 1042 | put_user(u, &ifr->ifr_flags)) |
02df55d2 | 1043 | ret = -EFAULT; |
8f475a31 | 1044 | macvtap_put_vlan(vlan); |
441ac0fc | 1045 | rtnl_unlock(); |
02df55d2 | 1046 | return ret; |
20d29d7a | 1047 | |
815f236d JW |
1048 | case TUNSETQUEUE: |
1049 | if (get_user(u, &ifr->ifr_flags)) | |
1050 | return -EFAULT; | |
441ac0fc VY |
1051 | rtnl_lock(); |
1052 | ret = macvtap_ioctl_set_queue(file, u); | |
1053 | rtnl_unlock(); | |
82a19eb8 | 1054 | return ret; |
815f236d | 1055 | |
20d29d7a | 1056 | case TUNGETFEATURES: |
6ae7feb3 | 1057 | if (put_user(IFF_TAP | IFF_NO_PI | MACVTAP_FEATURES, up)) |
20d29d7a AB |
1058 | return -EFAULT; |
1059 | return 0; | |
1060 | ||
1061 | case TUNSETSNDBUF: | |
1062 | if (get_user(u, up)) | |
1063 | return -EFAULT; | |
1064 | ||
20d29d7a | 1065 | q->sk.sk_sndbuf = u; |
20d29d7a AB |
1066 | return 0; |
1067 | ||
55afbd08 MT |
1068 | case TUNGETVNETHDRSZ: |
1069 | s = q->vnet_hdr_sz; | |
1070 | if (put_user(s, sp)) | |
1071 | return -EFAULT; | |
1072 | return 0; | |
1073 | ||
1074 | case TUNSETVNETHDRSZ: | |
1075 | if (get_user(s, sp)) | |
1076 | return -EFAULT; | |
1077 | if (s < (int)sizeof(struct virtio_net_hdr)) | |
1078 | return -EINVAL; | |
1079 | ||
1080 | q->vnet_hdr_sz = s; | |
1081 | return 0; | |
1082 | ||
01b07fb3 MT |
1083 | case TUNGETVNETLE: |
1084 | s = !!(q->flags & MACVTAP_VNET_LE); | |
1085 | if (put_user(s, sp)) | |
1086 | return -EFAULT; | |
1087 | return 0; | |
1088 | ||
1089 | case TUNSETVNETLE: | |
1090 | if (get_user(s, sp)) | |
1091 | return -EFAULT; | |
1092 | if (s) | |
1093 | q->flags |= MACVTAP_VNET_LE; | |
1094 | else | |
1095 | q->flags &= ~MACVTAP_VNET_LE; | |
1096 | return 0; | |
1097 | ||
20d29d7a AB |
1098 | case TUNSETOFFLOAD: |
1099 | /* let the user check for future flags */ | |
1100 | if (arg & ~(TUN_F_CSUM | TUN_F_TSO4 | TUN_F_TSO6 | | |
e3e3c423 | 1101 | TUN_F_TSO_ECN | TUN_F_UFO)) |
20d29d7a AB |
1102 | return -EINVAL; |
1103 | ||
2be5c767 VY |
1104 | rtnl_lock(); |
1105 | ret = set_offload(q, arg); | |
1106 | rtnl_unlock(); | |
1107 | return ret; | |
20d29d7a AB |
1108 | |
1109 | default: | |
1110 | return -EINVAL; | |
1111 | } | |
1112 | } | |
1113 | ||
1114 | #ifdef CONFIG_COMPAT | |
1115 | static long macvtap_compat_ioctl(struct file *file, unsigned int cmd, | |
1116 | unsigned long arg) | |
1117 | { | |
1118 | return macvtap_ioctl(file, cmd, (unsigned long)compat_ptr(arg)); | |
1119 | } | |
1120 | #endif | |
1121 | ||
1122 | static const struct file_operations macvtap_fops = { | |
1123 | .owner = THIS_MODULE, | |
1124 | .open = macvtap_open, | |
1125 | .release = macvtap_release, | |
3af0bfe5 | 1126 | .read_iter = macvtap_read_iter, |
f5ff53b4 | 1127 | .write_iter = macvtap_write_iter, |
20d29d7a AB |
1128 | .poll = macvtap_poll, |
1129 | .llseek = no_llseek, | |
1130 | .unlocked_ioctl = macvtap_ioctl, | |
1131 | #ifdef CONFIG_COMPAT | |
1132 | .compat_ioctl = macvtap_compat_ioctl, | |
1133 | #endif | |
1134 | }; | |
1135 | ||
1b784140 YX |
1136 | static int macvtap_sendmsg(struct socket *sock, struct msghdr *m, |
1137 | size_t total_len) | |
501c774c AB |
1138 | { |
1139 | struct macvtap_queue *q = container_of(sock, struct macvtap_queue, sock); | |
c0371da6 | 1140 | return macvtap_get_user(q, m, &m->msg_iter, m->msg_flags & MSG_DONTWAIT); |
501c774c AB |
1141 | } |
1142 | ||
1b784140 YX |
1143 | static int macvtap_recvmsg(struct socket *sock, struct msghdr *m, |
1144 | size_t total_len, int flags) | |
501c774c AB |
1145 | { |
1146 | struct macvtap_queue *q = container_of(sock, struct macvtap_queue, sock); | |
1147 | int ret; | |
1148 | if (flags & ~(MSG_DONTWAIT|MSG_TRUNC)) | |
1149 | return -EINVAL; | |
c0371da6 | 1150 | ret = macvtap_do_read(q, &m->msg_iter, flags & MSG_DONTWAIT); |
de2aa476 DM |
1151 | if (ret > total_len) { |
1152 | m->msg_flags |= MSG_TRUNC; | |
1153 | ret = flags & MSG_TRUNC ? ret : total_len; | |
1154 | } | |
501c774c AB |
1155 | return ret; |
1156 | } | |
1157 | ||
1158 | /* Ops structure to mimic raw sockets with tun */ | |
1159 | static const struct proto_ops macvtap_socket_ops = { | |
1160 | .sendmsg = macvtap_sendmsg, | |
1161 | .recvmsg = macvtap_recvmsg, | |
1162 | }; | |
1163 | ||
1164 | /* Get an underlying socket object from tun file. Returns error unless file is | |
1165 | * attached to a device. The returned object works like a packet socket, it | |
1166 | * can be used for sock_sendmsg/sock_recvmsg. The caller is responsible for | |
1167 | * holding a reference to the file for as long as the socket is in use. */ | |
1168 | struct socket *macvtap_get_socket(struct file *file) | |
1169 | { | |
1170 | struct macvtap_queue *q; | |
1171 | if (file->f_op != &macvtap_fops) | |
1172 | return ERR_PTR(-EINVAL); | |
1173 | q = file->private_data; | |
1174 | if (!q) | |
1175 | return ERR_PTR(-EBADFD); | |
1176 | return &q->sock; | |
1177 | } | |
1178 | EXPORT_SYMBOL_GPL(macvtap_get_socket); | |
1179 | ||
9bf1907f EB |
1180 | static int macvtap_device_event(struct notifier_block *unused, |
1181 | unsigned long event, void *ptr) | |
1182 | { | |
351638e7 | 1183 | struct net_device *dev = netdev_notifier_info_to_dev(ptr); |
e09eff7f | 1184 | struct macvlan_dev *vlan; |
9bf1907f EB |
1185 | struct device *classdev; |
1186 | dev_t devt; | |
e09eff7f | 1187 | int err; |
9bf1907f EB |
1188 | |
1189 | if (dev->rtnl_link_ops != &macvtap_link_ops) | |
1190 | return NOTIFY_DONE; | |
1191 | ||
e09eff7f | 1192 | vlan = netdev_priv(dev); |
9bf1907f EB |
1193 | |
1194 | switch (event) { | |
1195 | case NETDEV_REGISTER: | |
1196 | /* Create the device node here after the network device has | |
1197 | * been registered but before register_netdevice has | |
1198 | * finished running. | |
1199 | */ | |
e09eff7f EB |
1200 | err = macvtap_get_minor(vlan); |
1201 | if (err) | |
1202 | return notifier_from_errno(err); | |
1203 | ||
1204 | devt = MKDEV(MAJOR(macvtap_major), vlan->minor); | |
9bf1907f EB |
1205 | classdev = device_create(macvtap_class, &dev->dev, devt, |
1206 | dev, "tap%d", dev->ifindex); | |
e09eff7f EB |
1207 | if (IS_ERR(classdev)) { |
1208 | macvtap_free_minor(vlan); | |
9bf1907f | 1209 | return notifier_from_errno(PTR_ERR(classdev)); |
e09eff7f | 1210 | } |
9bf1907f EB |
1211 | break; |
1212 | case NETDEV_UNREGISTER: | |
e09eff7f | 1213 | devt = MKDEV(MAJOR(macvtap_major), vlan->minor); |
9bf1907f | 1214 | device_destroy(macvtap_class, devt); |
e09eff7f | 1215 | macvtap_free_minor(vlan); |
9bf1907f EB |
1216 | break; |
1217 | } | |
1218 | ||
1219 | return NOTIFY_DONE; | |
1220 | } | |
1221 | ||
1222 | static struct notifier_block macvtap_notifier_block __read_mostly = { | |
1223 | .notifier_call = macvtap_device_event, | |
1224 | }; | |
1225 | ||
20d29d7a AB |
1226 | static int macvtap_init(void) |
1227 | { | |
1228 | int err; | |
1229 | ||
1230 | err = alloc_chrdev_region(&macvtap_major, 0, | |
1231 | MACVTAP_NUM_DEVS, "macvtap"); | |
1232 | if (err) | |
1233 | goto out1; | |
1234 | ||
1235 | cdev_init(&macvtap_cdev, &macvtap_fops); | |
1236 | err = cdev_add(&macvtap_cdev, macvtap_major, MACVTAP_NUM_DEVS); | |
1237 | if (err) | |
1238 | goto out2; | |
1239 | ||
1240 | macvtap_class = class_create(THIS_MODULE, "macvtap"); | |
1241 | if (IS_ERR(macvtap_class)) { | |
1242 | err = PTR_ERR(macvtap_class); | |
1243 | goto out3; | |
1244 | } | |
1245 | ||
9bf1907f | 1246 | err = register_netdevice_notifier(&macvtap_notifier_block); |
20d29d7a AB |
1247 | if (err) |
1248 | goto out4; | |
1249 | ||
9bf1907f EB |
1250 | err = macvlan_link_register(&macvtap_link_ops); |
1251 | if (err) | |
1252 | goto out5; | |
1253 | ||
20d29d7a AB |
1254 | return 0; |
1255 | ||
9bf1907f EB |
1256 | out5: |
1257 | unregister_netdevice_notifier(&macvtap_notifier_block); | |
20d29d7a AB |
1258 | out4: |
1259 | class_unregister(macvtap_class); | |
1260 | out3: | |
1261 | cdev_del(&macvtap_cdev); | |
1262 | out2: | |
1263 | unregister_chrdev_region(macvtap_major, MACVTAP_NUM_DEVS); | |
1264 | out1: | |
1265 | return err; | |
1266 | } | |
1267 | module_init(macvtap_init); | |
1268 | ||
1269 | static void macvtap_exit(void) | |
1270 | { | |
1271 | rtnl_link_unregister(&macvtap_link_ops); | |
9bf1907f | 1272 | unregister_netdevice_notifier(&macvtap_notifier_block); |
20d29d7a AB |
1273 | class_unregister(macvtap_class); |
1274 | cdev_del(&macvtap_cdev); | |
1275 | unregister_chrdev_region(macvtap_major, MACVTAP_NUM_DEVS); | |
1276 | } | |
1277 | module_exit(macvtap_exit); | |
1278 | ||
1279 | MODULE_ALIAS_RTNL_LINK("macvtap"); | |
1280 | MODULE_AUTHOR("Arnd Bergmann <arnd@arndb.de>"); | |
1281 | MODULE_LICENSE("GPL"); |