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 | |
8b8e658b GK |
51 | #define MACVTAP_VNET_BE 0x40000000 |
52 | ||
53 | #ifdef CONFIG_TUN_VNET_CROSS_LE | |
54 | static inline bool macvtap_legacy_is_little_endian(struct macvtap_queue *q) | |
55 | { | |
56 | return q->flags & MACVTAP_VNET_BE ? false : | |
57 | virtio_legacy_is_little_endian(); | |
58 | } | |
59 | ||
60 | static long macvtap_get_vnet_be(struct macvtap_queue *q, int __user *sp) | |
61 | { | |
62 | int s = !!(q->flags & MACVTAP_VNET_BE); | |
63 | ||
64 | if (put_user(s, sp)) | |
65 | return -EFAULT; | |
66 | ||
67 | return 0; | |
68 | } | |
69 | ||
70 | static long macvtap_set_vnet_be(struct macvtap_queue *q, int __user *sp) | |
71 | { | |
72 | int s; | |
73 | ||
74 | if (get_user(s, sp)) | |
75 | return -EFAULT; | |
76 | ||
77 | if (s) | |
78 | q->flags |= MACVTAP_VNET_BE; | |
79 | else | |
80 | q->flags &= ~MACVTAP_VNET_BE; | |
81 | ||
82 | return 0; | |
83 | } | |
84 | #else | |
85 | static inline bool macvtap_legacy_is_little_endian(struct macvtap_queue *q) | |
86 | { | |
87 | return virtio_legacy_is_little_endian(); | |
88 | } | |
89 | ||
90 | static long macvtap_get_vnet_be(struct macvtap_queue *q, int __user *argp) | |
91 | { | |
92 | return -EINVAL; | |
93 | } | |
94 | ||
95 | static long macvtap_set_vnet_be(struct macvtap_queue *q, int __user *argp) | |
96 | { | |
97 | return -EINVAL; | |
98 | } | |
99 | #endif /* CONFIG_TUN_VNET_CROSS_LE */ | |
6ae7feb3 | 100 | |
5b11e15f GK |
101 | static inline bool macvtap_is_little_endian(struct macvtap_queue *q) |
102 | { | |
7d824109 | 103 | return q->flags & MACVTAP_VNET_LE || |
8b8e658b | 104 | macvtap_legacy_is_little_endian(q); |
5b11e15f | 105 | } |
6ae7feb3 MT |
106 | |
107 | static inline u16 macvtap16_to_cpu(struct macvtap_queue *q, __virtio16 val) | |
108 | { | |
5b11e15f | 109 | return __virtio16_to_cpu(macvtap_is_little_endian(q), val); |
6ae7feb3 MT |
110 | } |
111 | ||
112 | static inline __virtio16 cpu_to_macvtap16(struct macvtap_queue *q, u16 val) | |
113 | { | |
5b11e15f | 114 | return __cpu_to_virtio16(macvtap_is_little_endian(q), val); |
6ae7feb3 MT |
115 | } |
116 | ||
20d29d7a AB |
117 | static struct proto macvtap_proto = { |
118 | .name = "macvtap", | |
119 | .owner = THIS_MODULE, | |
120 | .obj_size = sizeof (struct macvtap_queue), | |
121 | }; | |
122 | ||
123 | /* | |
e09eff7f | 124 | * Variables for dealing with macvtaps device numbers. |
20d29d7a | 125 | */ |
1ebed71a | 126 | static dev_t macvtap_major; |
e09eff7f EB |
127 | #define MACVTAP_NUM_DEVS (1U << MINORBITS) |
128 | static DEFINE_MUTEX(minor_lock); | |
129 | static DEFINE_IDR(minor_idr); | |
130 | ||
97bc3633 | 131 | #define GOODCOPY_LEN 128 |
20d29d7a AB |
132 | static struct class *macvtap_class; |
133 | static struct cdev macvtap_cdev; | |
134 | ||
501c774c AB |
135 | static const struct proto_ops macvtap_socket_ops; |
136 | ||
2be5c767 | 137 | #define TUN_OFFLOADS (NETIF_F_HW_CSUM | NETIF_F_TSO_ECN | NETIF_F_TSO | \ |
e3e3c423 | 138 | NETIF_F_TSO6 | NETIF_F_UFO) |
2be5c767 | 139 | #define RX_OFFLOADS (NETIF_F_GRO | NETIF_F_LRO) |
f23d538b | 140 | #define TAP_FEATURES (NETIF_F_GSO | NETIF_F_SG | NETIF_F_FRAGLIST) |
a567dd62 | 141 | |
6acf54f1 VY |
142 | static struct macvlan_dev *macvtap_get_vlan_rcu(const struct net_device *dev) |
143 | { | |
144 | return rcu_dereference(dev->rx_handler_data); | |
145 | } | |
146 | ||
20d29d7a AB |
147 | /* |
148 | * RCU usage: | |
02df55d2 AB |
149 | * The macvtap_queue and the macvlan_dev are loosely coupled, the |
150 | * pointers from one to the other can only be read while rcu_read_lock | |
441ac0fc | 151 | * or rtnl is held. |
20d29d7a | 152 | * |
02df55d2 AB |
153 | * Both the file and the macvlan_dev hold a reference on the macvtap_queue |
154 | * through sock_hold(&q->sk). When the macvlan_dev goes away first, | |
155 | * q->vlan becomes inaccessible. When the files gets closed, | |
156 | * macvtap_get_queue() fails. | |
20d29d7a | 157 | * |
02df55d2 AB |
158 | * There may still be references to the struct sock inside of the |
159 | * queue from outbound SKBs, but these never reference back to the | |
160 | * file or the dev. The data structure is freed through __sk_free | |
161 | * when both our references and any pending SKBs are gone. | |
20d29d7a | 162 | */ |
20d29d7a | 163 | |
815f236d | 164 | static int macvtap_enable_queue(struct net_device *dev, struct file *file, |
20d29d7a | 165 | struct macvtap_queue *q) |
815f236d JW |
166 | { |
167 | struct macvlan_dev *vlan = netdev_priv(dev); | |
168 | int err = -EINVAL; | |
169 | ||
441ac0fc | 170 | ASSERT_RTNL(); |
815f236d JW |
171 | |
172 | if (q->enabled) | |
173 | goto out; | |
174 | ||
175 | err = 0; | |
176 | rcu_assign_pointer(vlan->taps[vlan->numvtaps], q); | |
177 | q->queue_index = vlan->numvtaps; | |
178 | q->enabled = true; | |
179 | ||
180 | vlan->numvtaps++; | |
181 | out: | |
815f236d JW |
182 | return err; |
183 | } | |
184 | ||
40b8fe45 | 185 | /* Requires RTNL */ |
815f236d JW |
186 | static int macvtap_set_queue(struct net_device *dev, struct file *file, |
187 | struct macvtap_queue *q) | |
20d29d7a AB |
188 | { |
189 | struct macvlan_dev *vlan = netdev_priv(dev); | |
20d29d7a | 190 | |
815f236d | 191 | if (vlan->numqueues == MAX_MACVTAP_QUEUES) |
40b8fe45 | 192 | return -EBUSY; |
20d29d7a | 193 | |
02df55d2 | 194 | rcu_assign_pointer(q->vlan, vlan); |
376b1aab | 195 | rcu_assign_pointer(vlan->taps[vlan->numvtaps], q); |
02df55d2 | 196 | sock_hold(&q->sk); |
20d29d7a AB |
197 | |
198 | q->file = file; | |
376b1aab | 199 | q->queue_index = vlan->numvtaps; |
815f236d | 200 | q->enabled = true; |
02df55d2 | 201 | file->private_data = q; |
815f236d | 202 | list_add_tail(&q->next, &vlan->queue_list); |
20d29d7a | 203 | |
1565c7c1 | 204 | vlan->numvtaps++; |
815f236d | 205 | vlan->numqueues++; |
1565c7c1 | 206 | |
40b8fe45 | 207 | return 0; |
20d29d7a AB |
208 | } |
209 | ||
441ac0fc | 210 | static int macvtap_disable_queue(struct macvtap_queue *q) |
815f236d JW |
211 | { |
212 | struct macvlan_dev *vlan; | |
213 | struct macvtap_queue *nq; | |
214 | ||
441ac0fc | 215 | ASSERT_RTNL(); |
815f236d JW |
216 | if (!q->enabled) |
217 | return -EINVAL; | |
218 | ||
441ac0fc VY |
219 | vlan = rtnl_dereference(q->vlan); |
220 | ||
815f236d JW |
221 | if (vlan) { |
222 | int index = q->queue_index; | |
223 | BUG_ON(index >= vlan->numvtaps); | |
441ac0fc | 224 | nq = rtnl_dereference(vlan->taps[vlan->numvtaps - 1]); |
815f236d JW |
225 | nq->queue_index = index; |
226 | ||
227 | rcu_assign_pointer(vlan->taps[index], nq); | |
228 | RCU_INIT_POINTER(vlan->taps[vlan->numvtaps - 1], NULL); | |
229 | q->enabled = false; | |
230 | ||
231 | vlan->numvtaps--; | |
232 | } | |
233 | ||
234 | return 0; | |
235 | } | |
236 | ||
20d29d7a | 237 | /* |
02df55d2 AB |
238 | * The file owning the queue got closed, give up both |
239 | * the reference that the files holds as well as the | |
240 | * one from the macvlan_dev if that still exists. | |
20d29d7a AB |
241 | * |
242 | * Using the spinlock makes sure that we don't get | |
243 | * to the queue again after destroying it. | |
20d29d7a | 244 | */ |
02df55d2 | 245 | static void macvtap_put_queue(struct macvtap_queue *q) |
20d29d7a | 246 | { |
02df55d2 | 247 | struct macvlan_dev *vlan; |
20d29d7a | 248 | |
441ac0fc VY |
249 | rtnl_lock(); |
250 | vlan = rtnl_dereference(q->vlan); | |
251 | ||
02df55d2 | 252 | if (vlan) { |
815f236d | 253 | if (q->enabled) |
441ac0fc | 254 | BUG_ON(macvtap_disable_queue(q)); |
376b1aab | 255 | |
815f236d | 256 | vlan->numqueues--; |
2cfa5a04 | 257 | RCU_INIT_POINTER(q->vlan, NULL); |
02df55d2 | 258 | sock_put(&q->sk); |
815f236d | 259 | list_del_init(&q->next); |
20d29d7a AB |
260 | } |
261 | ||
441ac0fc | 262 | rtnl_unlock(); |
20d29d7a AB |
263 | |
264 | synchronize_rcu(); | |
265 | sock_put(&q->sk); | |
266 | } | |
267 | ||
268 | /* | |
1565c7c1 KK |
269 | * Select a queue based on the rxq of the device on which this packet |
270 | * arrived. If the incoming device is not mq, calculate a flow hash | |
271 | * to select a queue. If all fails, find the first available queue. | |
272 | * Cache vlan->numvtaps since it can become zero during the execution | |
273 | * of this function. | |
20d29d7a AB |
274 | */ |
275 | static struct macvtap_queue *macvtap_get_queue(struct net_device *dev, | |
276 | struct sk_buff *skb) | |
277 | { | |
278 | struct macvlan_dev *vlan = netdev_priv(dev); | |
1565c7c1 | 279 | struct macvtap_queue *tap = NULL; |
815f236d JW |
280 | /* Access to taps array is protected by rcu, but access to numvtaps |
281 | * isn't. Below we use it to lookup a queue, but treat it as a hint | |
282 | * and validate that the result isn't NULL - in case we are | |
283 | * racing against queue removal. | |
284 | */ | |
ed0483fa | 285 | int numvtaps = ACCESS_ONCE(vlan->numvtaps); |
1565c7c1 KK |
286 | __u32 rxq; |
287 | ||
288 | if (!numvtaps) | |
289 | goto out; | |
290 | ||
ef0002b5 | 291 | /* Check if we can use flow to select a queue */ |
3958afa1 | 292 | rxq = skb_get_hash(skb); |
ef0002b5 KK |
293 | if (rxq) { |
294 | tap = rcu_dereference(vlan->taps[rxq % numvtaps]); | |
376b1aab | 295 | goto out; |
ef0002b5 KK |
296 | } |
297 | ||
1565c7c1 KK |
298 | if (likely(skb_rx_queue_recorded(skb))) { |
299 | rxq = skb_get_rx_queue(skb); | |
20d29d7a | 300 | |
1565c7c1 KK |
301 | while (unlikely(rxq >= numvtaps)) |
302 | rxq -= numvtaps; | |
303 | ||
304 | tap = rcu_dereference(vlan->taps[rxq]); | |
376b1aab | 305 | goto out; |
1565c7c1 KK |
306 | } |
307 | ||
376b1aab | 308 | tap = rcu_dereference(vlan->taps[0]); |
1565c7c1 KK |
309 | out: |
310 | return tap; | |
20d29d7a AB |
311 | } |
312 | ||
02df55d2 AB |
313 | /* |
314 | * The net_device is going away, give up the reference | |
1565c7c1 KK |
315 | * that it holds on all queues and safely set the pointer |
316 | * from the queues to NULL. | |
02df55d2 | 317 | */ |
20d29d7a AB |
318 | static void macvtap_del_queues(struct net_device *dev) |
319 | { | |
320 | struct macvlan_dev *vlan = netdev_priv(dev); | |
dfe816c5 | 321 | struct macvtap_queue *q, *tmp; |
02df55d2 | 322 | |
441ac0fc | 323 | ASSERT_RTNL(); |
815f236d JW |
324 | list_for_each_entry_safe(q, tmp, &vlan->queue_list, next) { |
325 | list_del_init(&q->next); | |
376b1aab | 326 | RCU_INIT_POINTER(q->vlan, NULL); |
815f236d JW |
327 | if (q->enabled) |
328 | vlan->numvtaps--; | |
329 | vlan->numqueues--; | |
dfe816c5 | 330 | sock_put(&q->sk); |
564517e8 | 331 | } |
815f236d JW |
332 | BUG_ON(vlan->numvtaps); |
333 | BUG_ON(vlan->numqueues); | |
99f34b38 EB |
334 | /* guarantee that any future macvtap_set_queue will fail */ |
335 | vlan->numvtaps = MAX_MACVTAP_QUEUES; | |
20d29d7a AB |
336 | } |
337 | ||
6acf54f1 | 338 | static rx_handler_result_t macvtap_handle_frame(struct sk_buff **pskb) |
20d29d7a | 339 | { |
6acf54f1 VY |
340 | struct sk_buff *skb = *pskb; |
341 | struct net_device *dev = skb->dev; | |
342 | struct macvlan_dev *vlan; | |
343 | struct macvtap_queue *q; | |
a567dd62 VY |
344 | netdev_features_t features = TAP_FEATURES; |
345 | ||
6acf54f1 VY |
346 | vlan = macvtap_get_vlan_rcu(dev); |
347 | if (!vlan) | |
348 | return RX_HANDLER_PASS; | |
349 | ||
350 | q = macvtap_get_queue(dev, skb); | |
20d29d7a | 351 | if (!q) |
6acf54f1 | 352 | return RX_HANDLER_PASS; |
8a35747a HX |
353 | |
354 | if (skb_queue_len(&q->sk.sk_receive_queue) >= dev->tx_queue_len) | |
355 | goto drop; | |
20d29d7a | 356 | |
6acf54f1 VY |
357 | skb_push(skb, ETH_HLEN); |
358 | ||
3e4f8b78 | 359 | /* Apply the forward feature mask so that we perform segmentation |
e5733321 VY |
360 | * according to users wishes. This only works if VNET_HDR is |
361 | * enabled. | |
3e4f8b78 | 362 | */ |
e5733321 VY |
363 | if (q->flags & IFF_VNET_HDR) |
364 | features |= vlan->tap_features; | |
8b86a61d | 365 | if (netif_needs_gso(skb, features)) { |
3e4f8b78 VY |
366 | struct sk_buff *segs = __skb_gso_segment(skb, features, false); |
367 | ||
368 | if (IS_ERR(segs)) | |
369 | goto drop; | |
370 | ||
371 | if (!segs) { | |
372 | skb_queue_tail(&q->sk.sk_receive_queue, skb); | |
373 | goto wake_up; | |
374 | } | |
375 | ||
376 | kfree_skb(skb); | |
377 | while (segs) { | |
378 | struct sk_buff *nskb = segs->next; | |
379 | ||
380 | segs->next = NULL; | |
381 | skb_queue_tail(&q->sk.sk_receive_queue, segs); | |
382 | segs = nskb; | |
383 | } | |
384 | } else { | |
cbdb0427 VY |
385 | /* If we receive a partial checksum and the tap side |
386 | * doesn't support checksum offload, compute the checksum. | |
387 | * Note: it doesn't matter which checksum feature to | |
388 | * check, we either support them all or none. | |
389 | */ | |
390 | if (skb->ip_summed == CHECKSUM_PARTIAL && | |
a188222b | 391 | !(features & NETIF_F_CSUM_MASK) && |
cbdb0427 VY |
392 | skb_checksum_help(skb)) |
393 | goto drop; | |
3e4f8b78 VY |
394 | skb_queue_tail(&q->sk.sk_receive_queue, skb); |
395 | } | |
396 | ||
397 | wake_up: | |
4a4771a5 | 398 | wake_up_interruptible_poll(sk_sleep(&q->sk), POLLIN | POLLRDNORM | POLLRDBAND); |
6acf54f1 | 399 | return RX_HANDLER_CONSUMED; |
8a35747a HX |
400 | |
401 | drop: | |
6acf54f1 VY |
402 | /* Count errors/drops only here, thus don't care about args. */ |
403 | macvlan_count_rx(vlan, 0, 0, 0); | |
8a35747a | 404 | kfree_skb(skb); |
6acf54f1 | 405 | return RX_HANDLER_CONSUMED; |
20d29d7a AB |
406 | } |
407 | ||
e09eff7f EB |
408 | static int macvtap_get_minor(struct macvlan_dev *vlan) |
409 | { | |
410 | int retval = -ENOMEM; | |
e09eff7f EB |
411 | |
412 | mutex_lock(&minor_lock); | |
ec09ebc1 TH |
413 | retval = idr_alloc(&minor_idr, vlan, 1, MACVTAP_NUM_DEVS, GFP_KERNEL); |
414 | if (retval >= 0) { | |
415 | vlan->minor = retval; | |
416 | } else if (retval == -ENOSPC) { | |
e09eff7f EB |
417 | printk(KERN_ERR "too many macvtap devices\n"); |
418 | retval = -EINVAL; | |
e09eff7f | 419 | } |
e09eff7f | 420 | mutex_unlock(&minor_lock); |
ec09ebc1 | 421 | return retval < 0 ? retval : 0; |
e09eff7f EB |
422 | } |
423 | ||
424 | static void macvtap_free_minor(struct macvlan_dev *vlan) | |
425 | { | |
426 | mutex_lock(&minor_lock); | |
427 | if (vlan->minor) { | |
428 | idr_remove(&minor_idr, vlan->minor); | |
429 | vlan->minor = 0; | |
430 | } | |
431 | mutex_unlock(&minor_lock); | |
432 | } | |
433 | ||
434 | static struct net_device *dev_get_by_macvtap_minor(int minor) | |
435 | { | |
436 | struct net_device *dev = NULL; | |
437 | struct macvlan_dev *vlan; | |
438 | ||
439 | mutex_lock(&minor_lock); | |
440 | vlan = idr_find(&minor_idr, minor); | |
441 | if (vlan) { | |
442 | dev = vlan->dev; | |
443 | dev_hold(dev); | |
444 | } | |
445 | mutex_unlock(&minor_lock); | |
446 | return dev; | |
447 | } | |
448 | ||
20d29d7a AB |
449 | static int macvtap_newlink(struct net *src_net, |
450 | struct net_device *dev, | |
451 | struct nlattr *tb[], | |
452 | struct nlattr *data[]) | |
453 | { | |
815f236d | 454 | struct macvlan_dev *vlan = netdev_priv(dev); |
6acf54f1 VY |
455 | int err; |
456 | ||
815f236d JW |
457 | INIT_LIST_HEAD(&vlan->queue_list); |
458 | ||
2be5c767 VY |
459 | /* Since macvlan supports all offloads by default, make |
460 | * tap support all offloads also. | |
461 | */ | |
462 | vlan->tap_features = TUN_OFFLOADS; | |
463 | ||
6acf54f1 VY |
464 | err = netdev_rx_handler_register(dev, macvtap_handle_frame, vlan); |
465 | if (err) | |
466 | return err; | |
467 | ||
9bf1907f EB |
468 | /* Don't put anything that may fail after macvlan_common_newlink |
469 | * because we can't undo what it does. | |
470 | */ | |
2f6a1b66 | 471 | return macvlan_common_newlink(src_net, dev, tb, data); |
20d29d7a AB |
472 | } |
473 | ||
474 | static void macvtap_dellink(struct net_device *dev, | |
475 | struct list_head *head) | |
476 | { | |
6acf54f1 | 477 | netdev_rx_handler_unregister(dev); |
20d29d7a AB |
478 | macvtap_del_queues(dev); |
479 | macvlan_dellink(dev, head); | |
480 | } | |
481 | ||
8a35747a HX |
482 | static void macvtap_setup(struct net_device *dev) |
483 | { | |
484 | macvlan_common_setup(dev); | |
485 | dev->tx_queue_len = TUN_READQ_SIZE; | |
486 | } | |
487 | ||
20d29d7a AB |
488 | static struct rtnl_link_ops macvtap_link_ops __read_mostly = { |
489 | .kind = "macvtap", | |
8a35747a | 490 | .setup = macvtap_setup, |
20d29d7a AB |
491 | .newlink = macvtap_newlink, |
492 | .dellink = macvtap_dellink, | |
493 | }; | |
494 | ||
495 | ||
496 | static void macvtap_sock_write_space(struct sock *sk) | |
497 | { | |
43815482 ED |
498 | wait_queue_head_t *wqueue; |
499 | ||
20d29d7a | 500 | if (!sock_writeable(sk) || |
9cd3e072 | 501 | !test_and_clear_bit(SOCKWQ_ASYNC_NOSPACE, &sk->sk_socket->flags)) |
20d29d7a AB |
502 | return; |
503 | ||
43815482 ED |
504 | wqueue = sk_sleep(sk); |
505 | if (wqueue && waitqueue_active(wqueue)) | |
506 | wake_up_interruptible_poll(wqueue, POLLOUT | POLLWRNORM | POLLWRBAND); | |
20d29d7a AB |
507 | } |
508 | ||
2259fef0 EB |
509 | static void macvtap_sock_destruct(struct sock *sk) |
510 | { | |
511 | skb_queue_purge(&sk->sk_receive_queue); | |
512 | } | |
513 | ||
20d29d7a AB |
514 | static int macvtap_open(struct inode *inode, struct file *file) |
515 | { | |
516 | struct net *net = current->nsproxy->net_ns; | |
40b8fe45 | 517 | struct net_device *dev; |
20d29d7a | 518 | struct macvtap_queue *q; |
40b8fe45 | 519 | int err = -ENODEV; |
20d29d7a | 520 | |
40b8fe45 VY |
521 | rtnl_lock(); |
522 | dev = dev_get_by_macvtap_minor(iminor(inode)); | |
20d29d7a AB |
523 | if (!dev) |
524 | goto out; | |
525 | ||
20d29d7a AB |
526 | err = -ENOMEM; |
527 | q = (struct macvtap_queue *)sk_alloc(net, AF_UNSPEC, GFP_KERNEL, | |
11aa9c28 | 528 | &macvtap_proto, 0); |
20d29d7a AB |
529 | if (!q) |
530 | goto out; | |
531 | ||
d9a90a31 | 532 | RCU_INIT_POINTER(q->sock.wq, &q->wq); |
43815482 | 533 | init_waitqueue_head(&q->wq.wait); |
20d29d7a AB |
534 | q->sock.type = SOCK_RAW; |
535 | q->sock.state = SS_CONNECTED; | |
501c774c AB |
536 | q->sock.file = file; |
537 | q->sock.ops = &macvtap_socket_ops; | |
20d29d7a | 538 | sock_init_data(&q->sock, &q->sk); |
20d29d7a | 539 | q->sk.sk_write_space = macvtap_sock_write_space; |
2259fef0 | 540 | q->sk.sk_destruct = macvtap_sock_destruct; |
b9fb9ee0 | 541 | q->flags = IFF_VNET_HDR | IFF_NO_PI | IFF_TAP; |
55afbd08 | 542 | q->vnet_hdr_sz = sizeof(struct virtio_net_hdr); |
20d29d7a | 543 | |
97bc3633 SM |
544 | /* |
545 | * so far only KVM virtio_net uses macvtap, enable zero copy between | |
546 | * guest kernel and host kernel when lower device supports zerocopy | |
047af9cf EB |
547 | * |
548 | * The macvlan supports zerocopy iff the lower device supports zero | |
549 | * copy so we don't have to look at the lower device directly. | |
97bc3633 | 550 | */ |
047af9cf EB |
551 | if ((dev->features & NETIF_F_HIGHDMA) && (dev->features & NETIF_F_SG)) |
552 | sock_set_flag(&q->sk, SOCK_ZEROCOPY); | |
97bc3633 | 553 | |
20d29d7a AB |
554 | err = macvtap_set_queue(dev, file, q); |
555 | if (err) | |
556 | sock_put(&q->sk); | |
557 | ||
558 | out: | |
559 | if (dev) | |
560 | dev_put(dev); | |
561 | ||
40b8fe45 | 562 | rtnl_unlock(); |
20d29d7a AB |
563 | return err; |
564 | } | |
565 | ||
566 | static int macvtap_release(struct inode *inode, struct file *file) | |
567 | { | |
02df55d2 AB |
568 | struct macvtap_queue *q = file->private_data; |
569 | macvtap_put_queue(q); | |
20d29d7a AB |
570 | return 0; |
571 | } | |
572 | ||
573 | static unsigned int macvtap_poll(struct file *file, poll_table * wait) | |
574 | { | |
02df55d2 | 575 | struct macvtap_queue *q = file->private_data; |
20d29d7a AB |
576 | unsigned int mask = POLLERR; |
577 | ||
578 | if (!q) | |
579 | goto out; | |
580 | ||
581 | mask = 0; | |
43815482 | 582 | poll_wait(file, &q->wq.wait, wait); |
20d29d7a AB |
583 | |
584 | if (!skb_queue_empty(&q->sk.sk_receive_queue)) | |
585 | mask |= POLLIN | POLLRDNORM; | |
586 | ||
587 | if (sock_writeable(&q->sk) || | |
9cd3e072 | 588 | (!test_and_set_bit(SOCKWQ_ASYNC_NOSPACE, &q->sock.flags) && |
20d29d7a AB |
589 | sock_writeable(&q->sk))) |
590 | mask |= POLLOUT | POLLWRNORM; | |
591 | ||
592 | out: | |
20d29d7a AB |
593 | return mask; |
594 | } | |
595 | ||
b9fb9ee0 AB |
596 | static inline struct sk_buff *macvtap_alloc_skb(struct sock *sk, size_t prepad, |
597 | size_t len, size_t linear, | |
598 | int noblock, int *err) | |
599 | { | |
600 | struct sk_buff *skb; | |
601 | ||
602 | /* Under a page? Don't bother with paged skb. */ | |
603 | if (prepad + len < PAGE_SIZE || !linear) | |
604 | linear = len; | |
605 | ||
606 | skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock, | |
28d64271 | 607 | err, 0); |
b9fb9ee0 AB |
608 | if (!skb) |
609 | return NULL; | |
610 | ||
611 | skb_reserve(skb, prepad); | |
612 | skb_put(skb, linear); | |
613 | skb->data_len = len - linear; | |
614 | skb->len += len - linear; | |
615 | ||
616 | return skb; | |
617 | } | |
618 | ||
619 | /* | |
620 | * macvtap_skb_from_vnet_hdr and macvtap_skb_to_vnet_hdr should | |
621 | * be shared with the tun/tap driver. | |
622 | */ | |
6ae7feb3 MT |
623 | static int macvtap_skb_from_vnet_hdr(struct macvtap_queue *q, |
624 | struct sk_buff *skb, | |
b9fb9ee0 AB |
625 | struct virtio_net_hdr *vnet_hdr) |
626 | { | |
627 | unsigned short gso_type = 0; | |
628 | if (vnet_hdr->gso_type != VIRTIO_NET_HDR_GSO_NONE) { | |
629 | switch (vnet_hdr->gso_type & ~VIRTIO_NET_HDR_GSO_ECN) { | |
630 | case VIRTIO_NET_HDR_GSO_TCPV4: | |
631 | gso_type = SKB_GSO_TCPV4; | |
632 | break; | |
633 | case VIRTIO_NET_HDR_GSO_TCPV6: | |
634 | gso_type = SKB_GSO_TCPV6; | |
635 | break; | |
636 | case VIRTIO_NET_HDR_GSO_UDP: | |
637 | gso_type = SKB_GSO_UDP; | |
638 | break; | |
639 | default: | |
640 | return -EINVAL; | |
641 | } | |
642 | ||
643 | if (vnet_hdr->gso_type & VIRTIO_NET_HDR_GSO_ECN) | |
644 | gso_type |= SKB_GSO_TCP_ECN; | |
645 | ||
646 | if (vnet_hdr->gso_size == 0) | |
647 | return -EINVAL; | |
648 | } | |
649 | ||
650 | if (vnet_hdr->flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) { | |
6ae7feb3 MT |
651 | if (!skb_partial_csum_set(skb, macvtap16_to_cpu(q, vnet_hdr->csum_start), |
652 | macvtap16_to_cpu(q, vnet_hdr->csum_offset))) | |
b9fb9ee0 AB |
653 | return -EINVAL; |
654 | } | |
655 | ||
656 | if (vnet_hdr->gso_type != VIRTIO_NET_HDR_GSO_NONE) { | |
6ae7feb3 | 657 | skb_shinfo(skb)->gso_size = macvtap16_to_cpu(q, vnet_hdr->gso_size); |
c9af6db4 | 658 | skb_shinfo(skb)->gso_type = gso_type; |
b9fb9ee0 AB |
659 | |
660 | /* Header must be checked, and gso_segs computed. */ | |
661 | skb_shinfo(skb)->gso_type |= SKB_GSO_DODGY; | |
662 | skb_shinfo(skb)->gso_segs = 0; | |
663 | } | |
664 | return 0; | |
665 | } | |
666 | ||
6ae7feb3 MT |
667 | static void macvtap_skb_to_vnet_hdr(struct macvtap_queue *q, |
668 | const struct sk_buff *skb, | |
669 | struct virtio_net_hdr *vnet_hdr) | |
b9fb9ee0 AB |
670 | { |
671 | memset(vnet_hdr, 0, sizeof(*vnet_hdr)); | |
672 | ||
673 | if (skb_is_gso(skb)) { | |
674 | struct skb_shared_info *sinfo = skb_shinfo(skb); | |
675 | ||
676 | /* This is a hint as to how much should be linear. */ | |
6ae7feb3 MT |
677 | vnet_hdr->hdr_len = cpu_to_macvtap16(q, skb_headlen(skb)); |
678 | vnet_hdr->gso_size = cpu_to_macvtap16(q, sinfo->gso_size); | |
b9fb9ee0 AB |
679 | if (sinfo->gso_type & SKB_GSO_TCPV4) |
680 | vnet_hdr->gso_type = VIRTIO_NET_HDR_GSO_TCPV4; | |
681 | else if (sinfo->gso_type & SKB_GSO_TCPV6) | |
682 | vnet_hdr->gso_type = VIRTIO_NET_HDR_GSO_TCPV6; | |
e3e3c423 VY |
683 | else if (sinfo->gso_type & SKB_GSO_UDP) |
684 | vnet_hdr->gso_type = VIRTIO_NET_HDR_GSO_UDP; | |
b9fb9ee0 AB |
685 | else |
686 | BUG(); | |
687 | if (sinfo->gso_type & SKB_GSO_TCP_ECN) | |
688 | vnet_hdr->gso_type |= VIRTIO_NET_HDR_GSO_ECN; | |
689 | } else | |
690 | vnet_hdr->gso_type = VIRTIO_NET_HDR_GSO_NONE; | |
691 | ||
692 | if (skb->ip_summed == CHECKSUM_PARTIAL) { | |
693 | vnet_hdr->flags = VIRTIO_NET_HDR_F_NEEDS_CSUM; | |
df8a39de | 694 | if (skb_vlan_tag_present(skb)) |
6ae7feb3 MT |
695 | vnet_hdr->csum_start = cpu_to_macvtap16(q, |
696 | skb_checksum_start_offset(skb) + VLAN_HLEN); | |
697 | else | |
698 | vnet_hdr->csum_start = cpu_to_macvtap16(q, | |
699 | skb_checksum_start_offset(skb)); | |
700 | vnet_hdr->csum_offset = cpu_to_macvtap16(q, skb->csum_offset); | |
10a8d94a JW |
701 | } else if (skb->ip_summed == CHECKSUM_UNNECESSARY) { |
702 | vnet_hdr->flags = VIRTIO_NET_HDR_F_DATA_VALID; | |
b9fb9ee0 | 703 | } /* else everything is zero */ |
b9fb9ee0 AB |
704 | } |
705 | ||
2f1d8b9e ED |
706 | /* Neighbour code has some assumptions on HH_DATA_MOD alignment */ |
707 | #define MACVTAP_RESERVE HH_DATA_OFF(ETH_HLEN) | |
708 | ||
20d29d7a | 709 | /* Get packet from user space buffer */ |
97bc3633 | 710 | static ssize_t macvtap_get_user(struct macvtap_queue *q, struct msghdr *m, |
f5ff53b4 | 711 | struct iov_iter *from, int noblock) |
20d29d7a | 712 | { |
2f1d8b9e | 713 | int good_linear = SKB_MAX_HEAD(MACVTAP_RESERVE); |
20d29d7a | 714 | struct sk_buff *skb; |
02df55d2 | 715 | struct macvlan_dev *vlan; |
f5ff53b4 | 716 | unsigned long total_len = iov_iter_count(from); |
97bc3633 | 717 | unsigned long len = total_len; |
20d29d7a | 718 | int err; |
b9fb9ee0 AB |
719 | struct virtio_net_hdr vnet_hdr = { 0 }; |
720 | int vnet_hdr_len = 0; | |
b92946e2 | 721 | int copylen = 0; |
c5c62f1b | 722 | int depth; |
97bc3633 | 723 | bool zerocopy = false; |
61d46bf9 | 724 | size_t linear; |
f5ff53b4 | 725 | ssize_t n; |
b9fb9ee0 AB |
726 | |
727 | if (q->flags & IFF_VNET_HDR) { | |
55afbd08 | 728 | vnet_hdr_len = q->vnet_hdr_sz; |
b9fb9ee0 AB |
729 | |
730 | err = -EINVAL; | |
ce3c8692 | 731 | if (len < vnet_hdr_len) |
b9fb9ee0 | 732 | goto err; |
ce3c8692 | 733 | len -= vnet_hdr_len; |
b9fb9ee0 | 734 | |
f5ff53b4 AV |
735 | err = -EFAULT; |
736 | n = copy_from_iter(&vnet_hdr, sizeof(vnet_hdr), from); | |
737 | if (n != sizeof(vnet_hdr)) | |
b9fb9ee0 | 738 | goto err; |
f5ff53b4 | 739 | iov_iter_advance(from, vnet_hdr_len - sizeof(vnet_hdr)); |
b9fb9ee0 | 740 | if ((vnet_hdr.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) && |
6ae7feb3 MT |
741 | macvtap16_to_cpu(q, vnet_hdr.csum_start) + |
742 | macvtap16_to_cpu(q, vnet_hdr.csum_offset) + 2 > | |
743 | macvtap16_to_cpu(q, vnet_hdr.hdr_len)) | |
744 | vnet_hdr.hdr_len = cpu_to_macvtap16(q, | |
745 | macvtap16_to_cpu(q, vnet_hdr.csum_start) + | |
746 | macvtap16_to_cpu(q, vnet_hdr.csum_offset) + 2); | |
b9fb9ee0 | 747 | err = -EINVAL; |
6ae7feb3 | 748 | if (macvtap16_to_cpu(q, vnet_hdr.hdr_len) > len) |
b9fb9ee0 AB |
749 | goto err; |
750 | } | |
20d29d7a | 751 | |
b9fb9ee0 | 752 | err = -EINVAL; |
20d29d7a | 753 | if (unlikely(len < ETH_HLEN)) |
b9fb9ee0 | 754 | goto err; |
20d29d7a | 755 | |
ece793fc | 756 | if (m && m->msg_control && sock_flag(&q->sk, SOCK_ZEROCOPY)) { |
f5ff53b4 AV |
757 | struct iov_iter i; |
758 | ||
6ae7feb3 MT |
759 | copylen = vnet_hdr.hdr_len ? |
760 | macvtap16_to_cpu(q, vnet_hdr.hdr_len) : GOODCOPY_LEN; | |
16a3fa28 JW |
761 | if (copylen > good_linear) |
762 | copylen = good_linear; | |
61d46bf9 | 763 | linear = copylen; |
f5ff53b4 AV |
764 | i = *from; |
765 | iov_iter_advance(&i, copylen); | |
766 | if (iov_iter_npages(&i, INT_MAX) <= MAX_SKB_FRAGS) | |
ece793fc JW |
767 | zerocopy = true; |
768 | } | |
769 | ||
770 | if (!zerocopy) { | |
97bc3633 | 771 | copylen = len; |
6ae7feb3 | 772 | if (macvtap16_to_cpu(q, vnet_hdr.hdr_len) > good_linear) |
16a3fa28 JW |
773 | linear = good_linear; |
774 | else | |
6ae7feb3 | 775 | linear = macvtap16_to_cpu(q, vnet_hdr.hdr_len); |
61d46bf9 | 776 | } |
97bc3633 | 777 | |
2f1d8b9e | 778 | skb = macvtap_alloc_skb(&q->sk, MACVTAP_RESERVE, copylen, |
61d46bf9 | 779 | linear, noblock, &err); |
02df55d2 AB |
780 | if (!skb) |
781 | goto err; | |
20d29d7a | 782 | |
01d6657b | 783 | if (zerocopy) |
f5ff53b4 | 784 | err = zerocopy_sg_from_iter(skb, from); |
ece793fc | 785 | else { |
f5ff53b4 | 786 | err = skb_copy_datagram_from_iter(skb, 0, from, len); |
ece793fc JW |
787 | if (!err && m && m->msg_control) { |
788 | struct ubuf_info *uarg = m->msg_control; | |
789 | uarg->callback(uarg, false); | |
790 | } | |
791 | } | |
792 | ||
02df55d2 | 793 | if (err) |
b9fb9ee0 | 794 | goto err_kfree; |
20d29d7a AB |
795 | |
796 | skb_set_network_header(skb, ETH_HLEN); | |
b9fb9ee0 AB |
797 | skb_reset_mac_header(skb); |
798 | skb->protocol = eth_hdr(skb)->h_proto; | |
799 | ||
800 | if (vnet_hdr_len) { | |
6ae7feb3 | 801 | err = macvtap_skb_from_vnet_hdr(q, skb, &vnet_hdr); |
b9fb9ee0 AB |
802 | if (err) |
803 | goto err_kfree; | |
804 | } | |
805 | ||
40893fd0 | 806 | skb_probe_transport_header(skb, ETH_HLEN); |
9b4d669b | 807 | |
c5c62f1b IV |
808 | /* Move network header to the right position for VLAN tagged packets */ |
809 | if ((skb->protocol == htons(ETH_P_8021Q) || | |
810 | skb->protocol == htons(ETH_P_8021AD)) && | |
811 | __vlan_get_protocol(skb, skb->protocol, &depth) != 0) | |
812 | skb_set_network_header(skb, depth); | |
813 | ||
ac4e4af1 VY |
814 | rcu_read_lock(); |
815 | vlan = rcu_dereference(q->vlan); | |
97bc3633 | 816 | /* copy skb_ubuf_info for callback when skb has no error */ |
01d6657b | 817 | if (zerocopy) { |
97bc3633 | 818 | skb_shinfo(skb)->destructor_arg = m->msg_control; |
01d6657b | 819 | skb_shinfo(skb)->tx_flags |= SKBTX_DEV_ZEROCOPY; |
c9af6db4 | 820 | skb_shinfo(skb)->tx_flags |= SKBTX_SHARED_FRAG; |
01d6657b | 821 | } |
29d79196 | 822 | if (vlan) { |
6acf54f1 VY |
823 | skb->dev = vlan->dev; |
824 | dev_queue_xmit(skb); | |
29d79196 | 825 | } else { |
02df55d2 | 826 | kfree_skb(skb); |
29d79196 | 827 | } |
ac4e4af1 | 828 | rcu_read_unlock(); |
20d29d7a | 829 | |
97bc3633 | 830 | return total_len; |
02df55d2 | 831 | |
b9fb9ee0 AB |
832 | err_kfree: |
833 | kfree_skb(skb); | |
834 | ||
02df55d2 | 835 | err: |
ac4e4af1 VY |
836 | rcu_read_lock(); |
837 | vlan = rcu_dereference(q->vlan); | |
02df55d2 | 838 | if (vlan) |
cd3e22b7 | 839 | this_cpu_inc(vlan->pcpu_stats->tx_dropped); |
ac4e4af1 | 840 | rcu_read_unlock(); |
02df55d2 | 841 | |
02df55d2 | 842 | return err; |
20d29d7a AB |
843 | } |
844 | ||
f5ff53b4 | 845 | static ssize_t macvtap_write_iter(struct kiocb *iocb, struct iov_iter *from) |
20d29d7a AB |
846 | { |
847 | struct file *file = iocb->ki_filp; | |
02df55d2 | 848 | struct macvtap_queue *q = file->private_data; |
20d29d7a | 849 | |
f5ff53b4 | 850 | return macvtap_get_user(q, NULL, from, file->f_flags & O_NONBLOCK); |
20d29d7a AB |
851 | } |
852 | ||
853 | /* Put packet to the user space buffer */ | |
854 | static ssize_t macvtap_put_user(struct macvtap_queue *q, | |
855 | const struct sk_buff *skb, | |
6c36d2e2 | 856 | struct iov_iter *iter) |
20d29d7a | 857 | { |
20d29d7a | 858 | int ret; |
b9fb9ee0 | 859 | int vnet_hdr_len = 0; |
f09e2249 | 860 | int vlan_offset = 0; |
6c36d2e2 | 861 | int total; |
b9fb9ee0 AB |
862 | |
863 | if (q->flags & IFF_VNET_HDR) { | |
864 | struct virtio_net_hdr vnet_hdr; | |
55afbd08 | 865 | vnet_hdr_len = q->vnet_hdr_sz; |
6c36d2e2 | 866 | if (iov_iter_count(iter) < vnet_hdr_len) |
b9fb9ee0 AB |
867 | return -EINVAL; |
868 | ||
6ae7feb3 | 869 | macvtap_skb_to_vnet_hdr(q, skb, &vnet_hdr); |
b9fb9ee0 | 870 | |
6c36d2e2 HX |
871 | if (copy_to_iter(&vnet_hdr, sizeof(vnet_hdr), iter) != |
872 | sizeof(vnet_hdr)) | |
b9fb9ee0 | 873 | return -EFAULT; |
7cc76f51 JW |
874 | |
875 | iov_iter_advance(iter, vnet_hdr_len - sizeof(vnet_hdr)); | |
b9fb9ee0 | 876 | } |
6c36d2e2 | 877 | total = vnet_hdr_len; |
ce232ce0 | 878 | total += skb->len; |
f09e2249 | 879 | |
df8a39de | 880 | if (skb_vlan_tag_present(skb)) { |
f09e2249 BG |
881 | struct { |
882 | __be16 h_vlan_proto; | |
883 | __be16 h_vlan_TCI; | |
884 | } veth; | |
0fbe0d47 | 885 | veth.h_vlan_proto = skb->vlan_proto; |
df8a39de | 886 | veth.h_vlan_TCI = htons(skb_vlan_tag_get(skb)); |
f09e2249 BG |
887 | |
888 | vlan_offset = offsetof(struct vlan_ethhdr, h_vlan_proto); | |
ce232ce0 | 889 | total += VLAN_HLEN; |
f09e2249 | 890 | |
6c36d2e2 HX |
891 | ret = skb_copy_datagram_iter(skb, 0, iter, vlan_offset); |
892 | if (ret || !iov_iter_count(iter)) | |
f09e2249 BG |
893 | goto done; |
894 | ||
6c36d2e2 HX |
895 | ret = copy_to_iter(&veth, sizeof(veth), iter); |
896 | if (ret != sizeof(veth) || !iov_iter_count(iter)) | |
f09e2249 BG |
897 | goto done; |
898 | } | |
20d29d7a | 899 | |
6c36d2e2 HX |
900 | ret = skb_copy_datagram_iter(skb, vlan_offset, iter, |
901 | skb->len - vlan_offset); | |
20d29d7a | 902 | |
f09e2249 | 903 | done: |
ce232ce0 | 904 | return ret ? ret : total; |
20d29d7a AB |
905 | } |
906 | ||
55ec8e25 | 907 | static ssize_t macvtap_do_read(struct macvtap_queue *q, |
3af0bfe5 | 908 | struct iov_iter *to, |
501c774c | 909 | int noblock) |
20d29d7a | 910 | { |
ccf7e72b | 911 | DEFINE_WAIT(wait); |
20d29d7a | 912 | struct sk_buff *skb; |
501c774c | 913 | ssize_t ret = 0; |
20d29d7a | 914 | |
3af0bfe5 AV |
915 | if (!iov_iter_count(to)) |
916 | return 0; | |
917 | ||
918 | while (1) { | |
89cee917 JW |
919 | if (!noblock) |
920 | prepare_to_wait(sk_sleep(&q->sk), &wait, | |
921 | TASK_INTERRUPTIBLE); | |
20d29d7a AB |
922 | |
923 | /* Read frames from the queue */ | |
924 | skb = skb_dequeue(&q->sk.sk_receive_queue); | |
3af0bfe5 AV |
925 | if (skb) |
926 | break; | |
927 | if (noblock) { | |
928 | ret = -EAGAIN; | |
929 | break; | |
20d29d7a | 930 | } |
3af0bfe5 AV |
931 | if (signal_pending(current)) { |
932 | ret = -ERESTARTSYS; | |
933 | break; | |
934 | } | |
935 | /* Nothing to read, let's sleep */ | |
936 | schedule(); | |
937 | } | |
a499a2e9 VY |
938 | if (!noblock) |
939 | finish_wait(sk_sleep(&q->sk), &wait); | |
940 | ||
3af0bfe5 AV |
941 | if (skb) { |
942 | ret = macvtap_put_user(q, skb, to); | |
f51a5e82 JW |
943 | if (unlikely(ret < 0)) |
944 | kfree_skb(skb); | |
945 | else | |
946 | consume_skb(skb); | |
20d29d7a | 947 | } |
501c774c AB |
948 | return ret; |
949 | } | |
950 | ||
3af0bfe5 | 951 | static ssize_t macvtap_read_iter(struct kiocb *iocb, struct iov_iter *to) |
501c774c AB |
952 | { |
953 | struct file *file = iocb->ki_filp; | |
954 | struct macvtap_queue *q = file->private_data; | |
3af0bfe5 | 955 | ssize_t len = iov_iter_count(to), ret; |
20d29d7a | 956 | |
3af0bfe5 | 957 | ret = macvtap_do_read(q, to, file->f_flags & O_NONBLOCK); |
ce232ce0 | 958 | ret = min_t(ssize_t, ret, len); |
e6ebc7f1 ZYW |
959 | if (ret > 0) |
960 | iocb->ki_pos = ret; | |
20d29d7a AB |
961 | return ret; |
962 | } | |
963 | ||
8f475a31 JW |
964 | static struct macvlan_dev *macvtap_get_vlan(struct macvtap_queue *q) |
965 | { | |
966 | struct macvlan_dev *vlan; | |
967 | ||
441ac0fc VY |
968 | ASSERT_RTNL(); |
969 | vlan = rtnl_dereference(q->vlan); | |
8f475a31 JW |
970 | if (vlan) |
971 | dev_hold(vlan->dev); | |
8f475a31 JW |
972 | |
973 | return vlan; | |
974 | } | |
975 | ||
976 | static void macvtap_put_vlan(struct macvlan_dev *vlan) | |
977 | { | |
978 | dev_put(vlan->dev); | |
979 | } | |
980 | ||
815f236d JW |
981 | static int macvtap_ioctl_set_queue(struct file *file, unsigned int flags) |
982 | { | |
983 | struct macvtap_queue *q = file->private_data; | |
984 | struct macvlan_dev *vlan; | |
985 | int ret; | |
986 | ||
987 | vlan = macvtap_get_vlan(q); | |
988 | if (!vlan) | |
989 | return -EINVAL; | |
990 | ||
991 | if (flags & IFF_ATTACH_QUEUE) | |
992 | ret = macvtap_enable_queue(vlan->dev, file, q); | |
993 | else if (flags & IFF_DETACH_QUEUE) | |
994 | ret = macvtap_disable_queue(q); | |
f57855a5 JW |
995 | else |
996 | ret = -EINVAL; | |
815f236d JW |
997 | |
998 | macvtap_put_vlan(vlan); | |
999 | return ret; | |
1000 | } | |
1001 | ||
2be5c767 VY |
1002 | static int set_offload(struct macvtap_queue *q, unsigned long arg) |
1003 | { | |
1004 | struct macvlan_dev *vlan; | |
1005 | netdev_features_t features; | |
1006 | netdev_features_t feature_mask = 0; | |
1007 | ||
1008 | vlan = rtnl_dereference(q->vlan); | |
1009 | if (!vlan) | |
1010 | return -ENOLINK; | |
1011 | ||
1012 | features = vlan->dev->features; | |
1013 | ||
1014 | if (arg & TUN_F_CSUM) { | |
1015 | feature_mask = NETIF_F_HW_CSUM; | |
1016 | ||
1017 | if (arg & (TUN_F_TSO4 | TUN_F_TSO6)) { | |
1018 | if (arg & TUN_F_TSO_ECN) | |
1019 | feature_mask |= NETIF_F_TSO_ECN; | |
1020 | if (arg & TUN_F_TSO4) | |
1021 | feature_mask |= NETIF_F_TSO; | |
1022 | if (arg & TUN_F_TSO6) | |
1023 | feature_mask |= NETIF_F_TSO6; | |
1024 | } | |
e3e3c423 VY |
1025 | |
1026 | if (arg & TUN_F_UFO) | |
1027 | feature_mask |= NETIF_F_UFO; | |
2be5c767 VY |
1028 | } |
1029 | ||
1030 | /* tun/tap driver inverts the usage for TSO offloads, where | |
1031 | * setting the TSO bit means that the userspace wants to | |
1032 | * accept TSO frames and turning it off means that user space | |
1033 | * does not support TSO. | |
1034 | * For macvtap, we have to invert it to mean the same thing. | |
1035 | * When user space turns off TSO, we turn off GSO/LRO so that | |
1036 | * user-space will not receive TSO frames. | |
1037 | */ | |
e3e3c423 | 1038 | if (feature_mask & (NETIF_F_TSO | NETIF_F_TSO6 | NETIF_F_UFO)) |
2be5c767 VY |
1039 | features |= RX_OFFLOADS; |
1040 | else | |
1041 | features &= ~RX_OFFLOADS; | |
1042 | ||
1043 | /* tap_features are the same as features on tun/tap and | |
1044 | * reflect user expectations. | |
1045 | */ | |
a567dd62 | 1046 | vlan->tap_features = feature_mask; |
2be5c767 VY |
1047 | vlan->set_features = features; |
1048 | netdev_update_features(vlan->dev); | |
1049 | ||
1050 | return 0; | |
1051 | } | |
1052 | ||
20d29d7a AB |
1053 | /* |
1054 | * provide compatibility with generic tun/tap interface | |
1055 | */ | |
1056 | static long macvtap_ioctl(struct file *file, unsigned int cmd, | |
1057 | unsigned long arg) | |
1058 | { | |
02df55d2 AB |
1059 | struct macvtap_queue *q = file->private_data; |
1060 | struct macvlan_dev *vlan; | |
20d29d7a AB |
1061 | void __user *argp = (void __user *)arg; |
1062 | struct ifreq __user *ifr = argp; | |
1063 | unsigned int __user *up = argp; | |
39ec7de7 | 1064 | unsigned short u; |
55afbd08 | 1065 | int __user *sp = argp; |
7f460d30 | 1066 | struct sockaddr sa; |
55afbd08 | 1067 | int s; |
02df55d2 | 1068 | int ret; |
20d29d7a AB |
1069 | |
1070 | switch (cmd) { | |
1071 | case TUNSETIFF: | |
1072 | /* ignore the name, just look at flags */ | |
1073 | if (get_user(u, &ifr->ifr_flags)) | |
1074 | return -EFAULT; | |
b9fb9ee0 AB |
1075 | |
1076 | ret = 0; | |
6ae7feb3 | 1077 | if ((u & ~MACVTAP_FEATURES) != (IFF_NO_PI | IFF_TAP)) |
b9fb9ee0 AB |
1078 | ret = -EINVAL; |
1079 | else | |
39ec7de7 | 1080 | q->flags = (q->flags & ~MACVTAP_FEATURES) | u; |
b9fb9ee0 AB |
1081 | |
1082 | return ret; | |
20d29d7a AB |
1083 | |
1084 | case TUNGETIFF: | |
441ac0fc | 1085 | rtnl_lock(); |
8f475a31 | 1086 | vlan = macvtap_get_vlan(q); |
441ac0fc VY |
1087 | if (!vlan) { |
1088 | rtnl_unlock(); | |
20d29d7a | 1089 | return -ENOLINK; |
441ac0fc | 1090 | } |
20d29d7a | 1091 | |
02df55d2 | 1092 | ret = 0; |
39ec7de7 | 1093 | u = q->flags; |
13707f9e | 1094 | if (copy_to_user(&ifr->ifr_name, vlan->dev->name, IFNAMSIZ) || |
39ec7de7 | 1095 | put_user(u, &ifr->ifr_flags)) |
02df55d2 | 1096 | ret = -EFAULT; |
8f475a31 | 1097 | macvtap_put_vlan(vlan); |
441ac0fc | 1098 | rtnl_unlock(); |
02df55d2 | 1099 | return ret; |
20d29d7a | 1100 | |
815f236d JW |
1101 | case TUNSETQUEUE: |
1102 | if (get_user(u, &ifr->ifr_flags)) | |
1103 | return -EFAULT; | |
441ac0fc VY |
1104 | rtnl_lock(); |
1105 | ret = macvtap_ioctl_set_queue(file, u); | |
1106 | rtnl_unlock(); | |
82a19eb8 | 1107 | return ret; |
815f236d | 1108 | |
20d29d7a | 1109 | case TUNGETFEATURES: |
6ae7feb3 | 1110 | if (put_user(IFF_TAP | IFF_NO_PI | MACVTAP_FEATURES, up)) |
20d29d7a AB |
1111 | return -EFAULT; |
1112 | return 0; | |
1113 | ||
1114 | case TUNSETSNDBUF: | |
3ea79249 | 1115 | if (get_user(s, sp)) |
20d29d7a AB |
1116 | return -EFAULT; |
1117 | ||
3ea79249 | 1118 | q->sk.sk_sndbuf = s; |
20d29d7a AB |
1119 | return 0; |
1120 | ||
55afbd08 MT |
1121 | case TUNGETVNETHDRSZ: |
1122 | s = q->vnet_hdr_sz; | |
1123 | if (put_user(s, sp)) | |
1124 | return -EFAULT; | |
1125 | return 0; | |
1126 | ||
1127 | case TUNSETVNETHDRSZ: | |
1128 | if (get_user(s, sp)) | |
1129 | return -EFAULT; | |
1130 | if (s < (int)sizeof(struct virtio_net_hdr)) | |
1131 | return -EINVAL; | |
1132 | ||
1133 | q->vnet_hdr_sz = s; | |
1134 | return 0; | |
1135 | ||
01b07fb3 MT |
1136 | case TUNGETVNETLE: |
1137 | s = !!(q->flags & MACVTAP_VNET_LE); | |
1138 | if (put_user(s, sp)) | |
1139 | return -EFAULT; | |
1140 | return 0; | |
1141 | ||
1142 | case TUNSETVNETLE: | |
1143 | if (get_user(s, sp)) | |
1144 | return -EFAULT; | |
1145 | if (s) | |
1146 | q->flags |= MACVTAP_VNET_LE; | |
1147 | else | |
1148 | q->flags &= ~MACVTAP_VNET_LE; | |
1149 | return 0; | |
1150 | ||
8b8e658b GK |
1151 | case TUNGETVNETBE: |
1152 | return macvtap_get_vnet_be(q, sp); | |
1153 | ||
1154 | case TUNSETVNETBE: | |
1155 | return macvtap_set_vnet_be(q, sp); | |
1156 | ||
20d29d7a AB |
1157 | case TUNSETOFFLOAD: |
1158 | /* let the user check for future flags */ | |
1159 | if (arg & ~(TUN_F_CSUM | TUN_F_TSO4 | TUN_F_TSO6 | | |
e3e3c423 | 1160 | TUN_F_TSO_ECN | TUN_F_UFO)) |
20d29d7a AB |
1161 | return -EINVAL; |
1162 | ||
2be5c767 VY |
1163 | rtnl_lock(); |
1164 | ret = set_offload(q, arg); | |
1165 | rtnl_unlock(); | |
1166 | return ret; | |
20d29d7a | 1167 | |
b5082083 JC |
1168 | case SIOCGIFHWADDR: |
1169 | rtnl_lock(); | |
1170 | vlan = macvtap_get_vlan(q); | |
1171 | if (!vlan) { | |
1172 | rtnl_unlock(); | |
1173 | return -ENOLINK; | |
1174 | } | |
1175 | ret = 0; | |
1176 | u = vlan->dev->type; | |
1177 | if (copy_to_user(&ifr->ifr_name, vlan->dev->name, IFNAMSIZ) || | |
1178 | copy_to_user(&ifr->ifr_hwaddr.sa_data, vlan->dev->dev_addr, ETH_ALEN) || | |
1179 | put_user(u, &ifr->ifr_hwaddr.sa_family)) | |
1180 | ret = -EFAULT; | |
1181 | macvtap_put_vlan(vlan); | |
1182 | rtnl_unlock(); | |
1183 | return ret; | |
1184 | ||
1185 | case SIOCSIFHWADDR: | |
7f460d30 JC |
1186 | if (copy_from_user(&sa, &ifr->ifr_hwaddr, sizeof(sa))) |
1187 | return -EFAULT; | |
b5082083 JC |
1188 | rtnl_lock(); |
1189 | vlan = macvtap_get_vlan(q); | |
1190 | if (!vlan) { | |
1191 | rtnl_unlock(); | |
1192 | return -ENOLINK; | |
1193 | } | |
7f460d30 | 1194 | ret = dev_set_mac_address(vlan->dev, &sa); |
b5082083 JC |
1195 | macvtap_put_vlan(vlan); |
1196 | rtnl_unlock(); | |
1197 | return ret; | |
1198 | ||
20d29d7a AB |
1199 | default: |
1200 | return -EINVAL; | |
1201 | } | |
1202 | } | |
1203 | ||
1204 | #ifdef CONFIG_COMPAT | |
1205 | static long macvtap_compat_ioctl(struct file *file, unsigned int cmd, | |
1206 | unsigned long arg) | |
1207 | { | |
1208 | return macvtap_ioctl(file, cmd, (unsigned long)compat_ptr(arg)); | |
1209 | } | |
1210 | #endif | |
1211 | ||
1212 | static const struct file_operations macvtap_fops = { | |
1213 | .owner = THIS_MODULE, | |
1214 | .open = macvtap_open, | |
1215 | .release = macvtap_release, | |
3af0bfe5 | 1216 | .read_iter = macvtap_read_iter, |
f5ff53b4 | 1217 | .write_iter = macvtap_write_iter, |
20d29d7a AB |
1218 | .poll = macvtap_poll, |
1219 | .llseek = no_llseek, | |
1220 | .unlocked_ioctl = macvtap_ioctl, | |
1221 | #ifdef CONFIG_COMPAT | |
1222 | .compat_ioctl = macvtap_compat_ioctl, | |
1223 | #endif | |
1224 | }; | |
1225 | ||
1b784140 YX |
1226 | static int macvtap_sendmsg(struct socket *sock, struct msghdr *m, |
1227 | size_t total_len) | |
501c774c AB |
1228 | { |
1229 | struct macvtap_queue *q = container_of(sock, struct macvtap_queue, sock); | |
c0371da6 | 1230 | return macvtap_get_user(q, m, &m->msg_iter, m->msg_flags & MSG_DONTWAIT); |
501c774c AB |
1231 | } |
1232 | ||
1b784140 YX |
1233 | static int macvtap_recvmsg(struct socket *sock, struct msghdr *m, |
1234 | size_t total_len, int flags) | |
501c774c AB |
1235 | { |
1236 | struct macvtap_queue *q = container_of(sock, struct macvtap_queue, sock); | |
1237 | int ret; | |
1238 | if (flags & ~(MSG_DONTWAIT|MSG_TRUNC)) | |
1239 | return -EINVAL; | |
c0371da6 | 1240 | ret = macvtap_do_read(q, &m->msg_iter, flags & MSG_DONTWAIT); |
de2aa476 DM |
1241 | if (ret > total_len) { |
1242 | m->msg_flags |= MSG_TRUNC; | |
1243 | ret = flags & MSG_TRUNC ? ret : total_len; | |
1244 | } | |
501c774c AB |
1245 | return ret; |
1246 | } | |
1247 | ||
1248 | /* Ops structure to mimic raw sockets with tun */ | |
1249 | static const struct proto_ops macvtap_socket_ops = { | |
1250 | .sendmsg = macvtap_sendmsg, | |
1251 | .recvmsg = macvtap_recvmsg, | |
1252 | }; | |
1253 | ||
1254 | /* Get an underlying socket object from tun file. Returns error unless file is | |
1255 | * attached to a device. The returned object works like a packet socket, it | |
1256 | * can be used for sock_sendmsg/sock_recvmsg. The caller is responsible for | |
1257 | * holding a reference to the file for as long as the socket is in use. */ | |
1258 | struct socket *macvtap_get_socket(struct file *file) | |
1259 | { | |
1260 | struct macvtap_queue *q; | |
1261 | if (file->f_op != &macvtap_fops) | |
1262 | return ERR_PTR(-EINVAL); | |
1263 | q = file->private_data; | |
1264 | if (!q) | |
1265 | return ERR_PTR(-EBADFD); | |
1266 | return &q->sock; | |
1267 | } | |
1268 | EXPORT_SYMBOL_GPL(macvtap_get_socket); | |
1269 | ||
9bf1907f EB |
1270 | static int macvtap_device_event(struct notifier_block *unused, |
1271 | unsigned long event, void *ptr) | |
1272 | { | |
351638e7 | 1273 | struct net_device *dev = netdev_notifier_info_to_dev(ptr); |
e09eff7f | 1274 | struct macvlan_dev *vlan; |
9bf1907f EB |
1275 | struct device *classdev; |
1276 | dev_t devt; | |
e09eff7f | 1277 | int err; |
9bf1907f EB |
1278 | |
1279 | if (dev->rtnl_link_ops != &macvtap_link_ops) | |
1280 | return NOTIFY_DONE; | |
1281 | ||
e09eff7f | 1282 | vlan = netdev_priv(dev); |
9bf1907f EB |
1283 | |
1284 | switch (event) { | |
1285 | case NETDEV_REGISTER: | |
1286 | /* Create the device node here after the network device has | |
1287 | * been registered but before register_netdevice has | |
1288 | * finished running. | |
1289 | */ | |
e09eff7f EB |
1290 | err = macvtap_get_minor(vlan); |
1291 | if (err) | |
1292 | return notifier_from_errno(err); | |
1293 | ||
1294 | devt = MKDEV(MAJOR(macvtap_major), vlan->minor); | |
9bf1907f EB |
1295 | classdev = device_create(macvtap_class, &dev->dev, devt, |
1296 | dev, "tap%d", dev->ifindex); | |
e09eff7f EB |
1297 | if (IS_ERR(classdev)) { |
1298 | macvtap_free_minor(vlan); | |
9bf1907f | 1299 | return notifier_from_errno(PTR_ERR(classdev)); |
e09eff7f | 1300 | } |
9bf1907f EB |
1301 | break; |
1302 | case NETDEV_UNREGISTER: | |
e09eff7f | 1303 | devt = MKDEV(MAJOR(macvtap_major), vlan->minor); |
9bf1907f | 1304 | device_destroy(macvtap_class, devt); |
e09eff7f | 1305 | macvtap_free_minor(vlan); |
9bf1907f EB |
1306 | break; |
1307 | } | |
1308 | ||
1309 | return NOTIFY_DONE; | |
1310 | } | |
1311 | ||
1312 | static struct notifier_block macvtap_notifier_block __read_mostly = { | |
1313 | .notifier_call = macvtap_device_event, | |
1314 | }; | |
1315 | ||
20d29d7a AB |
1316 | static int macvtap_init(void) |
1317 | { | |
1318 | int err; | |
1319 | ||
1320 | err = alloc_chrdev_region(&macvtap_major, 0, | |
1321 | MACVTAP_NUM_DEVS, "macvtap"); | |
1322 | if (err) | |
1323 | goto out1; | |
1324 | ||
1325 | cdev_init(&macvtap_cdev, &macvtap_fops); | |
1326 | err = cdev_add(&macvtap_cdev, macvtap_major, MACVTAP_NUM_DEVS); | |
1327 | if (err) | |
1328 | goto out2; | |
1329 | ||
1330 | macvtap_class = class_create(THIS_MODULE, "macvtap"); | |
1331 | if (IS_ERR(macvtap_class)) { | |
1332 | err = PTR_ERR(macvtap_class); | |
1333 | goto out3; | |
1334 | } | |
1335 | ||
9bf1907f | 1336 | err = register_netdevice_notifier(&macvtap_notifier_block); |
20d29d7a AB |
1337 | if (err) |
1338 | goto out4; | |
1339 | ||
9bf1907f EB |
1340 | err = macvlan_link_register(&macvtap_link_ops); |
1341 | if (err) | |
1342 | goto out5; | |
1343 | ||
20d29d7a AB |
1344 | return 0; |
1345 | ||
9bf1907f EB |
1346 | out5: |
1347 | unregister_netdevice_notifier(&macvtap_notifier_block); | |
20d29d7a AB |
1348 | out4: |
1349 | class_unregister(macvtap_class); | |
1350 | out3: | |
1351 | cdev_del(&macvtap_cdev); | |
1352 | out2: | |
1353 | unregister_chrdev_region(macvtap_major, MACVTAP_NUM_DEVS); | |
1354 | out1: | |
1355 | return err; | |
1356 | } | |
1357 | module_init(macvtap_init); | |
1358 | ||
1359 | static void macvtap_exit(void) | |
1360 | { | |
1361 | rtnl_link_unregister(&macvtap_link_ops); | |
9bf1907f | 1362 | unregister_netdevice_notifier(&macvtap_notifier_block); |
20d29d7a AB |
1363 | class_unregister(macvtap_class); |
1364 | cdev_del(&macvtap_cdev); | |
1365 | unregister_chrdev_region(macvtap_major, MACVTAP_NUM_DEVS); | |
d5de1987 | 1366 | idr_destroy(&minor_idr); |
20d29d7a AB |
1367 | } |
1368 | module_exit(macvtap_exit); | |
1369 | ||
1370 | MODULE_ALIAS_RTNL_LINK("macvtap"); | |
1371 | MODULE_AUTHOR("Arnd Bergmann <arnd@arndb.de>"); | |
1372 | MODULE_LICENSE("GPL"); |