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 |
17af2bce MA |
132 | static const void *macvtap_net_namespace(struct device *d) |
133 | { | |
134 | struct net_device *dev = to_net_dev(d->parent); | |
135 | return dev_net(dev); | |
136 | } | |
137 | ||
138 | static struct class macvtap_class = { | |
139 | .name = "macvtap", | |
140 | .owner = THIS_MODULE, | |
141 | .ns_type = &net_ns_type_operations, | |
142 | .namespace = macvtap_net_namespace, | |
143 | }; | |
20d29d7a AB |
144 | static struct cdev macvtap_cdev; |
145 | ||
501c774c AB |
146 | static const struct proto_ops macvtap_socket_ops; |
147 | ||
2be5c767 | 148 | #define TUN_OFFLOADS (NETIF_F_HW_CSUM | NETIF_F_TSO_ECN | NETIF_F_TSO | \ |
e3e3c423 | 149 | NETIF_F_TSO6 | NETIF_F_UFO) |
2be5c767 | 150 | #define RX_OFFLOADS (NETIF_F_GRO | NETIF_F_LRO) |
f23d538b | 151 | #define TAP_FEATURES (NETIF_F_GSO | NETIF_F_SG | NETIF_F_FRAGLIST) |
a567dd62 | 152 | |
6acf54f1 VY |
153 | static struct macvlan_dev *macvtap_get_vlan_rcu(const struct net_device *dev) |
154 | { | |
155 | return rcu_dereference(dev->rx_handler_data); | |
156 | } | |
157 | ||
20d29d7a AB |
158 | /* |
159 | * RCU usage: | |
02df55d2 AB |
160 | * The macvtap_queue and the macvlan_dev are loosely coupled, the |
161 | * pointers from one to the other can only be read while rcu_read_lock | |
441ac0fc | 162 | * or rtnl is held. |
20d29d7a | 163 | * |
02df55d2 AB |
164 | * Both the file and the macvlan_dev hold a reference on the macvtap_queue |
165 | * through sock_hold(&q->sk). When the macvlan_dev goes away first, | |
166 | * q->vlan becomes inaccessible. When the files gets closed, | |
167 | * macvtap_get_queue() fails. | |
20d29d7a | 168 | * |
02df55d2 AB |
169 | * There may still be references to the struct sock inside of the |
170 | * queue from outbound SKBs, but these never reference back to the | |
171 | * file or the dev. The data structure is freed through __sk_free | |
172 | * when both our references and any pending SKBs are gone. | |
20d29d7a | 173 | */ |
20d29d7a | 174 | |
815f236d | 175 | static int macvtap_enable_queue(struct net_device *dev, struct file *file, |
20d29d7a | 176 | struct macvtap_queue *q) |
815f236d JW |
177 | { |
178 | struct macvlan_dev *vlan = netdev_priv(dev); | |
179 | int err = -EINVAL; | |
180 | ||
441ac0fc | 181 | ASSERT_RTNL(); |
815f236d JW |
182 | |
183 | if (q->enabled) | |
184 | goto out; | |
185 | ||
186 | err = 0; | |
187 | rcu_assign_pointer(vlan->taps[vlan->numvtaps], q); | |
188 | q->queue_index = vlan->numvtaps; | |
189 | q->enabled = true; | |
190 | ||
191 | vlan->numvtaps++; | |
192 | out: | |
815f236d JW |
193 | return err; |
194 | } | |
195 | ||
40b8fe45 | 196 | /* Requires RTNL */ |
815f236d JW |
197 | static int macvtap_set_queue(struct net_device *dev, struct file *file, |
198 | struct macvtap_queue *q) | |
20d29d7a AB |
199 | { |
200 | struct macvlan_dev *vlan = netdev_priv(dev); | |
20d29d7a | 201 | |
815f236d | 202 | if (vlan->numqueues == MAX_MACVTAP_QUEUES) |
40b8fe45 | 203 | return -EBUSY; |
20d29d7a | 204 | |
02df55d2 | 205 | rcu_assign_pointer(q->vlan, vlan); |
376b1aab | 206 | rcu_assign_pointer(vlan->taps[vlan->numvtaps], q); |
02df55d2 | 207 | sock_hold(&q->sk); |
20d29d7a AB |
208 | |
209 | q->file = file; | |
376b1aab | 210 | q->queue_index = vlan->numvtaps; |
815f236d | 211 | q->enabled = true; |
02df55d2 | 212 | file->private_data = q; |
815f236d | 213 | list_add_tail(&q->next, &vlan->queue_list); |
20d29d7a | 214 | |
1565c7c1 | 215 | vlan->numvtaps++; |
815f236d | 216 | vlan->numqueues++; |
1565c7c1 | 217 | |
40b8fe45 | 218 | return 0; |
20d29d7a AB |
219 | } |
220 | ||
441ac0fc | 221 | static int macvtap_disable_queue(struct macvtap_queue *q) |
815f236d JW |
222 | { |
223 | struct macvlan_dev *vlan; | |
224 | struct macvtap_queue *nq; | |
225 | ||
441ac0fc | 226 | ASSERT_RTNL(); |
815f236d JW |
227 | if (!q->enabled) |
228 | return -EINVAL; | |
229 | ||
441ac0fc VY |
230 | vlan = rtnl_dereference(q->vlan); |
231 | ||
815f236d JW |
232 | if (vlan) { |
233 | int index = q->queue_index; | |
234 | BUG_ON(index >= vlan->numvtaps); | |
441ac0fc | 235 | nq = rtnl_dereference(vlan->taps[vlan->numvtaps - 1]); |
815f236d JW |
236 | nq->queue_index = index; |
237 | ||
238 | rcu_assign_pointer(vlan->taps[index], nq); | |
239 | RCU_INIT_POINTER(vlan->taps[vlan->numvtaps - 1], NULL); | |
240 | q->enabled = false; | |
241 | ||
242 | vlan->numvtaps--; | |
243 | } | |
244 | ||
245 | return 0; | |
246 | } | |
247 | ||
20d29d7a | 248 | /* |
02df55d2 AB |
249 | * The file owning the queue got closed, give up both |
250 | * the reference that the files holds as well as the | |
251 | * one from the macvlan_dev if that still exists. | |
20d29d7a AB |
252 | * |
253 | * Using the spinlock makes sure that we don't get | |
254 | * to the queue again after destroying it. | |
20d29d7a | 255 | */ |
02df55d2 | 256 | static void macvtap_put_queue(struct macvtap_queue *q) |
20d29d7a | 257 | { |
02df55d2 | 258 | struct macvlan_dev *vlan; |
20d29d7a | 259 | |
441ac0fc VY |
260 | rtnl_lock(); |
261 | vlan = rtnl_dereference(q->vlan); | |
262 | ||
02df55d2 | 263 | if (vlan) { |
815f236d | 264 | if (q->enabled) |
441ac0fc | 265 | BUG_ON(macvtap_disable_queue(q)); |
376b1aab | 266 | |
815f236d | 267 | vlan->numqueues--; |
2cfa5a04 | 268 | RCU_INIT_POINTER(q->vlan, NULL); |
02df55d2 | 269 | sock_put(&q->sk); |
815f236d | 270 | list_del_init(&q->next); |
20d29d7a AB |
271 | } |
272 | ||
441ac0fc | 273 | rtnl_unlock(); |
20d29d7a AB |
274 | |
275 | synchronize_rcu(); | |
276 | sock_put(&q->sk); | |
277 | } | |
278 | ||
279 | /* | |
1565c7c1 KK |
280 | * Select a queue based on the rxq of the device on which this packet |
281 | * arrived. If the incoming device is not mq, calculate a flow hash | |
282 | * to select a queue. If all fails, find the first available queue. | |
283 | * Cache vlan->numvtaps since it can become zero during the execution | |
284 | * of this function. | |
20d29d7a AB |
285 | */ |
286 | static struct macvtap_queue *macvtap_get_queue(struct net_device *dev, | |
287 | struct sk_buff *skb) | |
288 | { | |
289 | struct macvlan_dev *vlan = netdev_priv(dev); | |
1565c7c1 | 290 | struct macvtap_queue *tap = NULL; |
815f236d JW |
291 | /* Access to taps array is protected by rcu, but access to numvtaps |
292 | * isn't. Below we use it to lookup a queue, but treat it as a hint | |
293 | * and validate that the result isn't NULL - in case we are | |
294 | * racing against queue removal. | |
295 | */ | |
ed0483fa | 296 | int numvtaps = ACCESS_ONCE(vlan->numvtaps); |
1565c7c1 KK |
297 | __u32 rxq; |
298 | ||
299 | if (!numvtaps) | |
300 | goto out; | |
301 | ||
ef0002b5 | 302 | /* Check if we can use flow to select a queue */ |
3958afa1 | 303 | rxq = skb_get_hash(skb); |
ef0002b5 KK |
304 | if (rxq) { |
305 | tap = rcu_dereference(vlan->taps[rxq % numvtaps]); | |
376b1aab | 306 | goto out; |
ef0002b5 KK |
307 | } |
308 | ||
1565c7c1 KK |
309 | if (likely(skb_rx_queue_recorded(skb))) { |
310 | rxq = skb_get_rx_queue(skb); | |
20d29d7a | 311 | |
1565c7c1 KK |
312 | while (unlikely(rxq >= numvtaps)) |
313 | rxq -= numvtaps; | |
314 | ||
315 | tap = rcu_dereference(vlan->taps[rxq]); | |
376b1aab | 316 | goto out; |
1565c7c1 KK |
317 | } |
318 | ||
376b1aab | 319 | tap = rcu_dereference(vlan->taps[0]); |
1565c7c1 KK |
320 | out: |
321 | return tap; | |
20d29d7a AB |
322 | } |
323 | ||
02df55d2 AB |
324 | /* |
325 | * The net_device is going away, give up the reference | |
1565c7c1 KK |
326 | * that it holds on all queues and safely set the pointer |
327 | * from the queues to NULL. | |
02df55d2 | 328 | */ |
20d29d7a AB |
329 | static void macvtap_del_queues(struct net_device *dev) |
330 | { | |
331 | struct macvlan_dev *vlan = netdev_priv(dev); | |
dfe816c5 | 332 | struct macvtap_queue *q, *tmp; |
02df55d2 | 333 | |
441ac0fc | 334 | ASSERT_RTNL(); |
815f236d JW |
335 | list_for_each_entry_safe(q, tmp, &vlan->queue_list, next) { |
336 | list_del_init(&q->next); | |
376b1aab | 337 | RCU_INIT_POINTER(q->vlan, NULL); |
815f236d JW |
338 | if (q->enabled) |
339 | vlan->numvtaps--; | |
340 | vlan->numqueues--; | |
dfe816c5 | 341 | sock_put(&q->sk); |
564517e8 | 342 | } |
815f236d JW |
343 | BUG_ON(vlan->numvtaps); |
344 | BUG_ON(vlan->numqueues); | |
99f34b38 EB |
345 | /* guarantee that any future macvtap_set_queue will fail */ |
346 | vlan->numvtaps = MAX_MACVTAP_QUEUES; | |
20d29d7a AB |
347 | } |
348 | ||
6acf54f1 | 349 | static rx_handler_result_t macvtap_handle_frame(struct sk_buff **pskb) |
20d29d7a | 350 | { |
6acf54f1 VY |
351 | struct sk_buff *skb = *pskb; |
352 | struct net_device *dev = skb->dev; | |
353 | struct macvlan_dev *vlan; | |
354 | struct macvtap_queue *q; | |
a567dd62 VY |
355 | netdev_features_t features = TAP_FEATURES; |
356 | ||
6acf54f1 VY |
357 | vlan = macvtap_get_vlan_rcu(dev); |
358 | if (!vlan) | |
359 | return RX_HANDLER_PASS; | |
360 | ||
361 | q = macvtap_get_queue(dev, skb); | |
20d29d7a | 362 | if (!q) |
6acf54f1 | 363 | return RX_HANDLER_PASS; |
8a35747a HX |
364 | |
365 | if (skb_queue_len(&q->sk.sk_receive_queue) >= dev->tx_queue_len) | |
366 | goto drop; | |
20d29d7a | 367 | |
6acf54f1 VY |
368 | skb_push(skb, ETH_HLEN); |
369 | ||
3e4f8b78 | 370 | /* Apply the forward feature mask so that we perform segmentation |
e5733321 VY |
371 | * according to users wishes. This only works if VNET_HDR is |
372 | * enabled. | |
3e4f8b78 | 373 | */ |
e5733321 VY |
374 | if (q->flags & IFF_VNET_HDR) |
375 | features |= vlan->tap_features; | |
8b86a61d | 376 | if (netif_needs_gso(skb, features)) { |
3e4f8b78 VY |
377 | struct sk_buff *segs = __skb_gso_segment(skb, features, false); |
378 | ||
379 | if (IS_ERR(segs)) | |
380 | goto drop; | |
381 | ||
382 | if (!segs) { | |
383 | skb_queue_tail(&q->sk.sk_receive_queue, skb); | |
384 | goto wake_up; | |
385 | } | |
386 | ||
be0bd316 | 387 | consume_skb(skb); |
3e4f8b78 VY |
388 | while (segs) { |
389 | struct sk_buff *nskb = segs->next; | |
390 | ||
391 | segs->next = NULL; | |
392 | skb_queue_tail(&q->sk.sk_receive_queue, segs); | |
393 | segs = nskb; | |
394 | } | |
395 | } else { | |
cbdb0427 VY |
396 | /* If we receive a partial checksum and the tap side |
397 | * doesn't support checksum offload, compute the checksum. | |
398 | * Note: it doesn't matter which checksum feature to | |
399 | * check, we either support them all or none. | |
400 | */ | |
401 | if (skb->ip_summed == CHECKSUM_PARTIAL && | |
a188222b | 402 | !(features & NETIF_F_CSUM_MASK) && |
cbdb0427 VY |
403 | skb_checksum_help(skb)) |
404 | goto drop; | |
3e4f8b78 VY |
405 | skb_queue_tail(&q->sk.sk_receive_queue, skb); |
406 | } | |
407 | ||
408 | wake_up: | |
4a4771a5 | 409 | wake_up_interruptible_poll(sk_sleep(&q->sk), POLLIN | POLLRDNORM | POLLRDBAND); |
6acf54f1 | 410 | return RX_HANDLER_CONSUMED; |
8a35747a HX |
411 | |
412 | drop: | |
6acf54f1 VY |
413 | /* Count errors/drops only here, thus don't care about args. */ |
414 | macvlan_count_rx(vlan, 0, 0, 0); | |
8a35747a | 415 | kfree_skb(skb); |
6acf54f1 | 416 | return RX_HANDLER_CONSUMED; |
20d29d7a AB |
417 | } |
418 | ||
e09eff7f EB |
419 | static int macvtap_get_minor(struct macvlan_dev *vlan) |
420 | { | |
421 | int retval = -ENOMEM; | |
e09eff7f EB |
422 | |
423 | mutex_lock(&minor_lock); | |
ec09ebc1 TH |
424 | retval = idr_alloc(&minor_idr, vlan, 1, MACVTAP_NUM_DEVS, GFP_KERNEL); |
425 | if (retval >= 0) { | |
426 | vlan->minor = retval; | |
427 | } else if (retval == -ENOSPC) { | |
e09eff7f EB |
428 | printk(KERN_ERR "too many macvtap devices\n"); |
429 | retval = -EINVAL; | |
e09eff7f | 430 | } |
e09eff7f | 431 | mutex_unlock(&minor_lock); |
ec09ebc1 | 432 | return retval < 0 ? retval : 0; |
e09eff7f EB |
433 | } |
434 | ||
435 | static void macvtap_free_minor(struct macvlan_dev *vlan) | |
436 | { | |
437 | mutex_lock(&minor_lock); | |
438 | if (vlan->minor) { | |
439 | idr_remove(&minor_idr, vlan->minor); | |
440 | vlan->minor = 0; | |
441 | } | |
442 | mutex_unlock(&minor_lock); | |
443 | } | |
444 | ||
445 | static struct net_device *dev_get_by_macvtap_minor(int minor) | |
446 | { | |
447 | struct net_device *dev = NULL; | |
448 | struct macvlan_dev *vlan; | |
449 | ||
450 | mutex_lock(&minor_lock); | |
451 | vlan = idr_find(&minor_idr, minor); | |
452 | if (vlan) { | |
453 | dev = vlan->dev; | |
454 | dev_hold(dev); | |
455 | } | |
456 | mutex_unlock(&minor_lock); | |
457 | return dev; | |
458 | } | |
459 | ||
20d29d7a AB |
460 | static int macvtap_newlink(struct net *src_net, |
461 | struct net_device *dev, | |
462 | struct nlattr *tb[], | |
463 | struct nlattr *data[]) | |
464 | { | |
815f236d | 465 | struct macvlan_dev *vlan = netdev_priv(dev); |
6acf54f1 VY |
466 | int err; |
467 | ||
815f236d JW |
468 | INIT_LIST_HEAD(&vlan->queue_list); |
469 | ||
2be5c767 VY |
470 | /* Since macvlan supports all offloads by default, make |
471 | * tap support all offloads also. | |
472 | */ | |
473 | vlan->tap_features = TUN_OFFLOADS; | |
474 | ||
6acf54f1 VY |
475 | err = netdev_rx_handler_register(dev, macvtap_handle_frame, vlan); |
476 | if (err) | |
477 | return err; | |
478 | ||
9bf1907f EB |
479 | /* Don't put anything that may fail after macvlan_common_newlink |
480 | * because we can't undo what it does. | |
481 | */ | |
2f6a1b66 | 482 | return macvlan_common_newlink(src_net, dev, tb, data); |
20d29d7a AB |
483 | } |
484 | ||
485 | static void macvtap_dellink(struct net_device *dev, | |
486 | struct list_head *head) | |
487 | { | |
6acf54f1 | 488 | netdev_rx_handler_unregister(dev); |
20d29d7a AB |
489 | macvtap_del_queues(dev); |
490 | macvlan_dellink(dev, head); | |
491 | } | |
492 | ||
8a35747a HX |
493 | static void macvtap_setup(struct net_device *dev) |
494 | { | |
495 | macvlan_common_setup(dev); | |
496 | dev->tx_queue_len = TUN_READQ_SIZE; | |
497 | } | |
498 | ||
20d29d7a AB |
499 | static struct rtnl_link_ops macvtap_link_ops __read_mostly = { |
500 | .kind = "macvtap", | |
8a35747a | 501 | .setup = macvtap_setup, |
20d29d7a AB |
502 | .newlink = macvtap_newlink, |
503 | .dellink = macvtap_dellink, | |
504 | }; | |
505 | ||
506 | ||
507 | static void macvtap_sock_write_space(struct sock *sk) | |
508 | { | |
43815482 ED |
509 | wait_queue_head_t *wqueue; |
510 | ||
20d29d7a | 511 | if (!sock_writeable(sk) || |
9cd3e072 | 512 | !test_and_clear_bit(SOCKWQ_ASYNC_NOSPACE, &sk->sk_socket->flags)) |
20d29d7a AB |
513 | return; |
514 | ||
43815482 ED |
515 | wqueue = sk_sleep(sk); |
516 | if (wqueue && waitqueue_active(wqueue)) | |
517 | wake_up_interruptible_poll(wqueue, POLLOUT | POLLWRNORM | POLLWRBAND); | |
20d29d7a AB |
518 | } |
519 | ||
2259fef0 EB |
520 | static void macvtap_sock_destruct(struct sock *sk) |
521 | { | |
522 | skb_queue_purge(&sk->sk_receive_queue); | |
523 | } | |
524 | ||
20d29d7a AB |
525 | static int macvtap_open(struct inode *inode, struct file *file) |
526 | { | |
527 | struct net *net = current->nsproxy->net_ns; | |
40b8fe45 | 528 | struct net_device *dev; |
20d29d7a | 529 | struct macvtap_queue *q; |
40b8fe45 | 530 | int err = -ENODEV; |
20d29d7a | 531 | |
40b8fe45 VY |
532 | rtnl_lock(); |
533 | dev = dev_get_by_macvtap_minor(iminor(inode)); | |
20d29d7a AB |
534 | if (!dev) |
535 | goto out; | |
536 | ||
20d29d7a AB |
537 | err = -ENOMEM; |
538 | q = (struct macvtap_queue *)sk_alloc(net, AF_UNSPEC, GFP_KERNEL, | |
11aa9c28 | 539 | &macvtap_proto, 0); |
20d29d7a AB |
540 | if (!q) |
541 | goto out; | |
542 | ||
d9a90a31 | 543 | RCU_INIT_POINTER(q->sock.wq, &q->wq); |
43815482 | 544 | init_waitqueue_head(&q->wq.wait); |
20d29d7a AB |
545 | q->sock.type = SOCK_RAW; |
546 | q->sock.state = SS_CONNECTED; | |
501c774c AB |
547 | q->sock.file = file; |
548 | q->sock.ops = &macvtap_socket_ops; | |
20d29d7a | 549 | sock_init_data(&q->sock, &q->sk); |
20d29d7a | 550 | q->sk.sk_write_space = macvtap_sock_write_space; |
2259fef0 | 551 | q->sk.sk_destruct = macvtap_sock_destruct; |
b9fb9ee0 | 552 | q->flags = IFF_VNET_HDR | IFF_NO_PI | IFF_TAP; |
55afbd08 | 553 | q->vnet_hdr_sz = sizeof(struct virtio_net_hdr); |
20d29d7a | 554 | |
97bc3633 SM |
555 | /* |
556 | * so far only KVM virtio_net uses macvtap, enable zero copy between | |
557 | * guest kernel and host kernel when lower device supports zerocopy | |
047af9cf EB |
558 | * |
559 | * The macvlan supports zerocopy iff the lower device supports zero | |
560 | * copy so we don't have to look at the lower device directly. | |
97bc3633 | 561 | */ |
047af9cf EB |
562 | if ((dev->features & NETIF_F_HIGHDMA) && (dev->features & NETIF_F_SG)) |
563 | sock_set_flag(&q->sk, SOCK_ZEROCOPY); | |
97bc3633 | 564 | |
20d29d7a AB |
565 | err = macvtap_set_queue(dev, file, q); |
566 | if (err) | |
567 | sock_put(&q->sk); | |
568 | ||
569 | out: | |
570 | if (dev) | |
571 | dev_put(dev); | |
572 | ||
40b8fe45 | 573 | rtnl_unlock(); |
20d29d7a AB |
574 | return err; |
575 | } | |
576 | ||
577 | static int macvtap_release(struct inode *inode, struct file *file) | |
578 | { | |
02df55d2 AB |
579 | struct macvtap_queue *q = file->private_data; |
580 | macvtap_put_queue(q); | |
20d29d7a AB |
581 | return 0; |
582 | } | |
583 | ||
584 | static unsigned int macvtap_poll(struct file *file, poll_table * wait) | |
585 | { | |
02df55d2 | 586 | struct macvtap_queue *q = file->private_data; |
20d29d7a AB |
587 | unsigned int mask = POLLERR; |
588 | ||
589 | if (!q) | |
590 | goto out; | |
591 | ||
592 | mask = 0; | |
43815482 | 593 | poll_wait(file, &q->wq.wait, wait); |
20d29d7a AB |
594 | |
595 | if (!skb_queue_empty(&q->sk.sk_receive_queue)) | |
596 | mask |= POLLIN | POLLRDNORM; | |
597 | ||
598 | if (sock_writeable(&q->sk) || | |
9cd3e072 | 599 | (!test_and_set_bit(SOCKWQ_ASYNC_NOSPACE, &q->sock.flags) && |
20d29d7a AB |
600 | sock_writeable(&q->sk))) |
601 | mask |= POLLOUT | POLLWRNORM; | |
602 | ||
603 | out: | |
20d29d7a AB |
604 | return mask; |
605 | } | |
606 | ||
b9fb9ee0 AB |
607 | static inline struct sk_buff *macvtap_alloc_skb(struct sock *sk, size_t prepad, |
608 | size_t len, size_t linear, | |
609 | int noblock, int *err) | |
610 | { | |
611 | struct sk_buff *skb; | |
612 | ||
613 | /* Under a page? Don't bother with paged skb. */ | |
614 | if (prepad + len < PAGE_SIZE || !linear) | |
615 | linear = len; | |
616 | ||
617 | skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock, | |
28d64271 | 618 | err, 0); |
b9fb9ee0 AB |
619 | if (!skb) |
620 | return NULL; | |
621 | ||
622 | skb_reserve(skb, prepad); | |
623 | skb_put(skb, linear); | |
624 | skb->data_len = len - linear; | |
625 | skb->len += len - linear; | |
626 | ||
627 | return skb; | |
628 | } | |
629 | ||
630 | /* | |
631 | * macvtap_skb_from_vnet_hdr and macvtap_skb_to_vnet_hdr should | |
632 | * be shared with the tun/tap driver. | |
633 | */ | |
6ae7feb3 MT |
634 | static int macvtap_skb_from_vnet_hdr(struct macvtap_queue *q, |
635 | struct sk_buff *skb, | |
b9fb9ee0 AB |
636 | struct virtio_net_hdr *vnet_hdr) |
637 | { | |
638 | unsigned short gso_type = 0; | |
639 | if (vnet_hdr->gso_type != VIRTIO_NET_HDR_GSO_NONE) { | |
640 | switch (vnet_hdr->gso_type & ~VIRTIO_NET_HDR_GSO_ECN) { | |
641 | case VIRTIO_NET_HDR_GSO_TCPV4: | |
642 | gso_type = SKB_GSO_TCPV4; | |
643 | break; | |
644 | case VIRTIO_NET_HDR_GSO_TCPV6: | |
645 | gso_type = SKB_GSO_TCPV6; | |
646 | break; | |
647 | case VIRTIO_NET_HDR_GSO_UDP: | |
648 | gso_type = SKB_GSO_UDP; | |
649 | break; | |
650 | default: | |
651 | return -EINVAL; | |
652 | } | |
653 | ||
654 | if (vnet_hdr->gso_type & VIRTIO_NET_HDR_GSO_ECN) | |
655 | gso_type |= SKB_GSO_TCP_ECN; | |
656 | ||
657 | if (vnet_hdr->gso_size == 0) | |
658 | return -EINVAL; | |
659 | } | |
660 | ||
661 | if (vnet_hdr->flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) { | |
6ae7feb3 MT |
662 | if (!skb_partial_csum_set(skb, macvtap16_to_cpu(q, vnet_hdr->csum_start), |
663 | macvtap16_to_cpu(q, vnet_hdr->csum_offset))) | |
b9fb9ee0 AB |
664 | return -EINVAL; |
665 | } | |
666 | ||
667 | if (vnet_hdr->gso_type != VIRTIO_NET_HDR_GSO_NONE) { | |
6ae7feb3 | 668 | skb_shinfo(skb)->gso_size = macvtap16_to_cpu(q, vnet_hdr->gso_size); |
c9af6db4 | 669 | skb_shinfo(skb)->gso_type = gso_type; |
b9fb9ee0 AB |
670 | |
671 | /* Header must be checked, and gso_segs computed. */ | |
672 | skb_shinfo(skb)->gso_type |= SKB_GSO_DODGY; | |
673 | skb_shinfo(skb)->gso_segs = 0; | |
674 | } | |
675 | return 0; | |
676 | } | |
677 | ||
6ae7feb3 MT |
678 | static void macvtap_skb_to_vnet_hdr(struct macvtap_queue *q, |
679 | const struct sk_buff *skb, | |
680 | struct virtio_net_hdr *vnet_hdr) | |
b9fb9ee0 AB |
681 | { |
682 | memset(vnet_hdr, 0, sizeof(*vnet_hdr)); | |
683 | ||
684 | if (skb_is_gso(skb)) { | |
685 | struct skb_shared_info *sinfo = skb_shinfo(skb); | |
686 | ||
687 | /* This is a hint as to how much should be linear. */ | |
6ae7feb3 MT |
688 | vnet_hdr->hdr_len = cpu_to_macvtap16(q, skb_headlen(skb)); |
689 | vnet_hdr->gso_size = cpu_to_macvtap16(q, sinfo->gso_size); | |
b9fb9ee0 AB |
690 | if (sinfo->gso_type & SKB_GSO_TCPV4) |
691 | vnet_hdr->gso_type = VIRTIO_NET_HDR_GSO_TCPV4; | |
692 | else if (sinfo->gso_type & SKB_GSO_TCPV6) | |
693 | vnet_hdr->gso_type = VIRTIO_NET_HDR_GSO_TCPV6; | |
e3e3c423 VY |
694 | else if (sinfo->gso_type & SKB_GSO_UDP) |
695 | vnet_hdr->gso_type = VIRTIO_NET_HDR_GSO_UDP; | |
b9fb9ee0 AB |
696 | else |
697 | BUG(); | |
698 | if (sinfo->gso_type & SKB_GSO_TCP_ECN) | |
699 | vnet_hdr->gso_type |= VIRTIO_NET_HDR_GSO_ECN; | |
700 | } else | |
701 | vnet_hdr->gso_type = VIRTIO_NET_HDR_GSO_NONE; | |
702 | ||
703 | if (skb->ip_summed == CHECKSUM_PARTIAL) { | |
704 | vnet_hdr->flags = VIRTIO_NET_HDR_F_NEEDS_CSUM; | |
df8a39de | 705 | if (skb_vlan_tag_present(skb)) |
6ae7feb3 MT |
706 | vnet_hdr->csum_start = cpu_to_macvtap16(q, |
707 | skb_checksum_start_offset(skb) + VLAN_HLEN); | |
708 | else | |
709 | vnet_hdr->csum_start = cpu_to_macvtap16(q, | |
710 | skb_checksum_start_offset(skb)); | |
711 | vnet_hdr->csum_offset = cpu_to_macvtap16(q, skb->csum_offset); | |
10a8d94a JW |
712 | } else if (skb->ip_summed == CHECKSUM_UNNECESSARY) { |
713 | vnet_hdr->flags = VIRTIO_NET_HDR_F_DATA_VALID; | |
b9fb9ee0 | 714 | } /* else everything is zero */ |
b9fb9ee0 AB |
715 | } |
716 | ||
2f1d8b9e ED |
717 | /* Neighbour code has some assumptions on HH_DATA_MOD alignment */ |
718 | #define MACVTAP_RESERVE HH_DATA_OFF(ETH_HLEN) | |
719 | ||
20d29d7a | 720 | /* Get packet from user space buffer */ |
97bc3633 | 721 | static ssize_t macvtap_get_user(struct macvtap_queue *q, struct msghdr *m, |
f5ff53b4 | 722 | struct iov_iter *from, int noblock) |
20d29d7a | 723 | { |
2f1d8b9e | 724 | int good_linear = SKB_MAX_HEAD(MACVTAP_RESERVE); |
20d29d7a | 725 | struct sk_buff *skb; |
02df55d2 | 726 | struct macvlan_dev *vlan; |
f5ff53b4 | 727 | unsigned long total_len = iov_iter_count(from); |
97bc3633 | 728 | unsigned long len = total_len; |
20d29d7a | 729 | int err; |
b9fb9ee0 AB |
730 | struct virtio_net_hdr vnet_hdr = { 0 }; |
731 | int vnet_hdr_len = 0; | |
b92946e2 | 732 | int copylen = 0; |
c5c62f1b | 733 | int depth; |
97bc3633 | 734 | bool zerocopy = false; |
61d46bf9 | 735 | size_t linear; |
f5ff53b4 | 736 | ssize_t n; |
b9fb9ee0 AB |
737 | |
738 | if (q->flags & IFF_VNET_HDR) { | |
55afbd08 | 739 | vnet_hdr_len = q->vnet_hdr_sz; |
b9fb9ee0 AB |
740 | |
741 | err = -EINVAL; | |
ce3c8692 | 742 | if (len < vnet_hdr_len) |
b9fb9ee0 | 743 | goto err; |
ce3c8692 | 744 | len -= vnet_hdr_len; |
b9fb9ee0 | 745 | |
f5ff53b4 AV |
746 | err = -EFAULT; |
747 | n = copy_from_iter(&vnet_hdr, sizeof(vnet_hdr), from); | |
748 | if (n != sizeof(vnet_hdr)) | |
b9fb9ee0 | 749 | goto err; |
f5ff53b4 | 750 | iov_iter_advance(from, vnet_hdr_len - sizeof(vnet_hdr)); |
b9fb9ee0 | 751 | if ((vnet_hdr.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) && |
6ae7feb3 MT |
752 | macvtap16_to_cpu(q, vnet_hdr.csum_start) + |
753 | macvtap16_to_cpu(q, vnet_hdr.csum_offset) + 2 > | |
754 | macvtap16_to_cpu(q, vnet_hdr.hdr_len)) | |
755 | vnet_hdr.hdr_len = cpu_to_macvtap16(q, | |
756 | macvtap16_to_cpu(q, vnet_hdr.csum_start) + | |
757 | macvtap16_to_cpu(q, vnet_hdr.csum_offset) + 2); | |
b9fb9ee0 | 758 | err = -EINVAL; |
6ae7feb3 | 759 | if (macvtap16_to_cpu(q, vnet_hdr.hdr_len) > len) |
b9fb9ee0 AB |
760 | goto err; |
761 | } | |
20d29d7a | 762 | |
b9fb9ee0 | 763 | err = -EINVAL; |
20d29d7a | 764 | if (unlikely(len < ETH_HLEN)) |
b9fb9ee0 | 765 | goto err; |
20d29d7a | 766 | |
ece793fc | 767 | if (m && m->msg_control && sock_flag(&q->sk, SOCK_ZEROCOPY)) { |
f5ff53b4 AV |
768 | struct iov_iter i; |
769 | ||
6ae7feb3 MT |
770 | copylen = vnet_hdr.hdr_len ? |
771 | macvtap16_to_cpu(q, vnet_hdr.hdr_len) : GOODCOPY_LEN; | |
16a3fa28 JW |
772 | if (copylen > good_linear) |
773 | copylen = good_linear; | |
8e2ad411 WB |
774 | else if (copylen < ETH_HLEN) |
775 | copylen = ETH_HLEN; | |
61d46bf9 | 776 | linear = copylen; |
f5ff53b4 AV |
777 | i = *from; |
778 | iov_iter_advance(&i, copylen); | |
779 | if (iov_iter_npages(&i, INT_MAX) <= MAX_SKB_FRAGS) | |
ece793fc JW |
780 | zerocopy = true; |
781 | } | |
782 | ||
783 | if (!zerocopy) { | |
97bc3633 | 784 | copylen = len; |
8e2ad411 WB |
785 | linear = macvtap16_to_cpu(q, vnet_hdr.hdr_len); |
786 | if (linear > good_linear) | |
16a3fa28 | 787 | linear = good_linear; |
8e2ad411 WB |
788 | else if (linear < ETH_HLEN) |
789 | linear = ETH_HLEN; | |
61d46bf9 | 790 | } |
97bc3633 | 791 | |
2f1d8b9e | 792 | skb = macvtap_alloc_skb(&q->sk, MACVTAP_RESERVE, copylen, |
61d46bf9 | 793 | linear, noblock, &err); |
02df55d2 AB |
794 | if (!skb) |
795 | goto err; | |
20d29d7a | 796 | |
01d6657b | 797 | if (zerocopy) |
f5ff53b4 | 798 | err = zerocopy_sg_from_iter(skb, from); |
ece793fc | 799 | else { |
f5ff53b4 | 800 | err = skb_copy_datagram_from_iter(skb, 0, from, len); |
ece793fc JW |
801 | if (!err && m && m->msg_control) { |
802 | struct ubuf_info *uarg = m->msg_control; | |
803 | uarg->callback(uarg, false); | |
804 | } | |
805 | } | |
806 | ||
02df55d2 | 807 | if (err) |
b9fb9ee0 | 808 | goto err_kfree; |
20d29d7a AB |
809 | |
810 | skb_set_network_header(skb, ETH_HLEN); | |
b9fb9ee0 AB |
811 | skb_reset_mac_header(skb); |
812 | skb->protocol = eth_hdr(skb)->h_proto; | |
813 | ||
814 | if (vnet_hdr_len) { | |
6ae7feb3 | 815 | err = macvtap_skb_from_vnet_hdr(q, skb, &vnet_hdr); |
b9fb9ee0 AB |
816 | if (err) |
817 | goto err_kfree; | |
818 | } | |
819 | ||
40893fd0 | 820 | skb_probe_transport_header(skb, ETH_HLEN); |
9b4d669b | 821 | |
c5c62f1b IV |
822 | /* Move network header to the right position for VLAN tagged packets */ |
823 | if ((skb->protocol == htons(ETH_P_8021Q) || | |
824 | skb->protocol == htons(ETH_P_8021AD)) && | |
825 | __vlan_get_protocol(skb, skb->protocol, &depth) != 0) | |
826 | skb_set_network_header(skb, depth); | |
827 | ||
ac4e4af1 VY |
828 | rcu_read_lock(); |
829 | vlan = rcu_dereference(q->vlan); | |
97bc3633 | 830 | /* copy skb_ubuf_info for callback when skb has no error */ |
01d6657b | 831 | if (zerocopy) { |
97bc3633 | 832 | skb_shinfo(skb)->destructor_arg = m->msg_control; |
01d6657b | 833 | skb_shinfo(skb)->tx_flags |= SKBTX_DEV_ZEROCOPY; |
c9af6db4 | 834 | skb_shinfo(skb)->tx_flags |= SKBTX_SHARED_FRAG; |
01d6657b | 835 | } |
29d79196 | 836 | if (vlan) { |
6acf54f1 VY |
837 | skb->dev = vlan->dev; |
838 | dev_queue_xmit(skb); | |
29d79196 | 839 | } else { |
02df55d2 | 840 | kfree_skb(skb); |
29d79196 | 841 | } |
ac4e4af1 | 842 | rcu_read_unlock(); |
20d29d7a | 843 | |
97bc3633 | 844 | return total_len; |
02df55d2 | 845 | |
b9fb9ee0 AB |
846 | err_kfree: |
847 | kfree_skb(skb); | |
848 | ||
02df55d2 | 849 | err: |
ac4e4af1 VY |
850 | rcu_read_lock(); |
851 | vlan = rcu_dereference(q->vlan); | |
02df55d2 | 852 | if (vlan) |
cd3e22b7 | 853 | this_cpu_inc(vlan->pcpu_stats->tx_dropped); |
ac4e4af1 | 854 | rcu_read_unlock(); |
02df55d2 | 855 | |
02df55d2 | 856 | return err; |
20d29d7a AB |
857 | } |
858 | ||
f5ff53b4 | 859 | static ssize_t macvtap_write_iter(struct kiocb *iocb, struct iov_iter *from) |
20d29d7a AB |
860 | { |
861 | struct file *file = iocb->ki_filp; | |
02df55d2 | 862 | struct macvtap_queue *q = file->private_data; |
20d29d7a | 863 | |
f5ff53b4 | 864 | return macvtap_get_user(q, NULL, from, file->f_flags & O_NONBLOCK); |
20d29d7a AB |
865 | } |
866 | ||
867 | /* Put packet to the user space buffer */ | |
868 | static ssize_t macvtap_put_user(struct macvtap_queue *q, | |
869 | const struct sk_buff *skb, | |
6c36d2e2 | 870 | struct iov_iter *iter) |
20d29d7a | 871 | { |
20d29d7a | 872 | int ret; |
b9fb9ee0 | 873 | int vnet_hdr_len = 0; |
f09e2249 | 874 | int vlan_offset = 0; |
6c36d2e2 | 875 | int total; |
b9fb9ee0 AB |
876 | |
877 | if (q->flags & IFF_VNET_HDR) { | |
878 | struct virtio_net_hdr vnet_hdr; | |
55afbd08 | 879 | vnet_hdr_len = q->vnet_hdr_sz; |
6c36d2e2 | 880 | if (iov_iter_count(iter) < vnet_hdr_len) |
b9fb9ee0 AB |
881 | return -EINVAL; |
882 | ||
6ae7feb3 | 883 | macvtap_skb_to_vnet_hdr(q, skb, &vnet_hdr); |
b9fb9ee0 | 884 | |
6c36d2e2 HX |
885 | if (copy_to_iter(&vnet_hdr, sizeof(vnet_hdr), iter) != |
886 | sizeof(vnet_hdr)) | |
b9fb9ee0 | 887 | return -EFAULT; |
7cc76f51 JW |
888 | |
889 | iov_iter_advance(iter, vnet_hdr_len - sizeof(vnet_hdr)); | |
b9fb9ee0 | 890 | } |
6c36d2e2 | 891 | total = vnet_hdr_len; |
ce232ce0 | 892 | total += skb->len; |
f09e2249 | 893 | |
df8a39de | 894 | if (skb_vlan_tag_present(skb)) { |
f09e2249 BG |
895 | struct { |
896 | __be16 h_vlan_proto; | |
897 | __be16 h_vlan_TCI; | |
898 | } veth; | |
0fbe0d47 | 899 | veth.h_vlan_proto = skb->vlan_proto; |
df8a39de | 900 | veth.h_vlan_TCI = htons(skb_vlan_tag_get(skb)); |
f09e2249 BG |
901 | |
902 | vlan_offset = offsetof(struct vlan_ethhdr, h_vlan_proto); | |
ce232ce0 | 903 | total += VLAN_HLEN; |
f09e2249 | 904 | |
6c36d2e2 HX |
905 | ret = skb_copy_datagram_iter(skb, 0, iter, vlan_offset); |
906 | if (ret || !iov_iter_count(iter)) | |
f09e2249 BG |
907 | goto done; |
908 | ||
6c36d2e2 HX |
909 | ret = copy_to_iter(&veth, sizeof(veth), iter); |
910 | if (ret != sizeof(veth) || !iov_iter_count(iter)) | |
f09e2249 BG |
911 | goto done; |
912 | } | |
20d29d7a | 913 | |
6c36d2e2 HX |
914 | ret = skb_copy_datagram_iter(skb, vlan_offset, iter, |
915 | skb->len - vlan_offset); | |
20d29d7a | 916 | |
f09e2249 | 917 | done: |
ce232ce0 | 918 | return ret ? ret : total; |
20d29d7a AB |
919 | } |
920 | ||
55ec8e25 | 921 | static ssize_t macvtap_do_read(struct macvtap_queue *q, |
3af0bfe5 | 922 | struct iov_iter *to, |
501c774c | 923 | int noblock) |
20d29d7a | 924 | { |
ccf7e72b | 925 | DEFINE_WAIT(wait); |
20d29d7a | 926 | struct sk_buff *skb; |
501c774c | 927 | ssize_t ret = 0; |
20d29d7a | 928 | |
3af0bfe5 AV |
929 | if (!iov_iter_count(to)) |
930 | return 0; | |
931 | ||
932 | while (1) { | |
89cee917 JW |
933 | if (!noblock) |
934 | prepare_to_wait(sk_sleep(&q->sk), &wait, | |
935 | TASK_INTERRUPTIBLE); | |
20d29d7a AB |
936 | |
937 | /* Read frames from the queue */ | |
938 | skb = skb_dequeue(&q->sk.sk_receive_queue); | |
3af0bfe5 AV |
939 | if (skb) |
940 | break; | |
941 | if (noblock) { | |
942 | ret = -EAGAIN; | |
943 | break; | |
20d29d7a | 944 | } |
3af0bfe5 AV |
945 | if (signal_pending(current)) { |
946 | ret = -ERESTARTSYS; | |
947 | break; | |
948 | } | |
949 | /* Nothing to read, let's sleep */ | |
950 | schedule(); | |
951 | } | |
a499a2e9 VY |
952 | if (!noblock) |
953 | finish_wait(sk_sleep(&q->sk), &wait); | |
954 | ||
3af0bfe5 AV |
955 | if (skb) { |
956 | ret = macvtap_put_user(q, skb, to); | |
f51a5e82 JW |
957 | if (unlikely(ret < 0)) |
958 | kfree_skb(skb); | |
959 | else | |
960 | consume_skb(skb); | |
20d29d7a | 961 | } |
501c774c AB |
962 | return ret; |
963 | } | |
964 | ||
3af0bfe5 | 965 | static ssize_t macvtap_read_iter(struct kiocb *iocb, struct iov_iter *to) |
501c774c AB |
966 | { |
967 | struct file *file = iocb->ki_filp; | |
968 | struct macvtap_queue *q = file->private_data; | |
3af0bfe5 | 969 | ssize_t len = iov_iter_count(to), ret; |
20d29d7a | 970 | |
3af0bfe5 | 971 | ret = macvtap_do_read(q, to, file->f_flags & O_NONBLOCK); |
ce232ce0 | 972 | ret = min_t(ssize_t, ret, len); |
e6ebc7f1 ZYW |
973 | if (ret > 0) |
974 | iocb->ki_pos = ret; | |
20d29d7a AB |
975 | return ret; |
976 | } | |
977 | ||
8f475a31 JW |
978 | static struct macvlan_dev *macvtap_get_vlan(struct macvtap_queue *q) |
979 | { | |
980 | struct macvlan_dev *vlan; | |
981 | ||
441ac0fc VY |
982 | ASSERT_RTNL(); |
983 | vlan = rtnl_dereference(q->vlan); | |
8f475a31 JW |
984 | if (vlan) |
985 | dev_hold(vlan->dev); | |
8f475a31 JW |
986 | |
987 | return vlan; | |
988 | } | |
989 | ||
990 | static void macvtap_put_vlan(struct macvlan_dev *vlan) | |
991 | { | |
992 | dev_put(vlan->dev); | |
993 | } | |
994 | ||
815f236d JW |
995 | static int macvtap_ioctl_set_queue(struct file *file, unsigned int flags) |
996 | { | |
997 | struct macvtap_queue *q = file->private_data; | |
998 | struct macvlan_dev *vlan; | |
999 | int ret; | |
1000 | ||
1001 | vlan = macvtap_get_vlan(q); | |
1002 | if (!vlan) | |
1003 | return -EINVAL; | |
1004 | ||
1005 | if (flags & IFF_ATTACH_QUEUE) | |
1006 | ret = macvtap_enable_queue(vlan->dev, file, q); | |
1007 | else if (flags & IFF_DETACH_QUEUE) | |
1008 | ret = macvtap_disable_queue(q); | |
f57855a5 JW |
1009 | else |
1010 | ret = -EINVAL; | |
815f236d JW |
1011 | |
1012 | macvtap_put_vlan(vlan); | |
1013 | return ret; | |
1014 | } | |
1015 | ||
2be5c767 VY |
1016 | static int set_offload(struct macvtap_queue *q, unsigned long arg) |
1017 | { | |
1018 | struct macvlan_dev *vlan; | |
1019 | netdev_features_t features; | |
1020 | netdev_features_t feature_mask = 0; | |
1021 | ||
1022 | vlan = rtnl_dereference(q->vlan); | |
1023 | if (!vlan) | |
1024 | return -ENOLINK; | |
1025 | ||
1026 | features = vlan->dev->features; | |
1027 | ||
1028 | if (arg & TUN_F_CSUM) { | |
1029 | feature_mask = NETIF_F_HW_CSUM; | |
1030 | ||
1031 | if (arg & (TUN_F_TSO4 | TUN_F_TSO6)) { | |
1032 | if (arg & TUN_F_TSO_ECN) | |
1033 | feature_mask |= NETIF_F_TSO_ECN; | |
1034 | if (arg & TUN_F_TSO4) | |
1035 | feature_mask |= NETIF_F_TSO; | |
1036 | if (arg & TUN_F_TSO6) | |
1037 | feature_mask |= NETIF_F_TSO6; | |
1038 | } | |
e3e3c423 VY |
1039 | |
1040 | if (arg & TUN_F_UFO) | |
1041 | feature_mask |= NETIF_F_UFO; | |
2be5c767 VY |
1042 | } |
1043 | ||
1044 | /* tun/tap driver inverts the usage for TSO offloads, where | |
1045 | * setting the TSO bit means that the userspace wants to | |
1046 | * accept TSO frames and turning it off means that user space | |
1047 | * does not support TSO. | |
1048 | * For macvtap, we have to invert it to mean the same thing. | |
1049 | * When user space turns off TSO, we turn off GSO/LRO so that | |
1050 | * user-space will not receive TSO frames. | |
1051 | */ | |
e3e3c423 | 1052 | if (feature_mask & (NETIF_F_TSO | NETIF_F_TSO6 | NETIF_F_UFO)) |
2be5c767 VY |
1053 | features |= RX_OFFLOADS; |
1054 | else | |
1055 | features &= ~RX_OFFLOADS; | |
1056 | ||
1057 | /* tap_features are the same as features on tun/tap and | |
1058 | * reflect user expectations. | |
1059 | */ | |
a567dd62 | 1060 | vlan->tap_features = feature_mask; |
2be5c767 VY |
1061 | vlan->set_features = features; |
1062 | netdev_update_features(vlan->dev); | |
1063 | ||
1064 | return 0; | |
1065 | } | |
1066 | ||
20d29d7a AB |
1067 | /* |
1068 | * provide compatibility with generic tun/tap interface | |
1069 | */ | |
1070 | static long macvtap_ioctl(struct file *file, unsigned int cmd, | |
1071 | unsigned long arg) | |
1072 | { | |
02df55d2 AB |
1073 | struct macvtap_queue *q = file->private_data; |
1074 | struct macvlan_dev *vlan; | |
20d29d7a AB |
1075 | void __user *argp = (void __user *)arg; |
1076 | struct ifreq __user *ifr = argp; | |
1077 | unsigned int __user *up = argp; | |
39ec7de7 | 1078 | unsigned short u; |
55afbd08 | 1079 | int __user *sp = argp; |
7f460d30 | 1080 | struct sockaddr sa; |
55afbd08 | 1081 | int s; |
02df55d2 | 1082 | int ret; |
20d29d7a AB |
1083 | |
1084 | switch (cmd) { | |
1085 | case TUNSETIFF: | |
1086 | /* ignore the name, just look at flags */ | |
1087 | if (get_user(u, &ifr->ifr_flags)) | |
1088 | return -EFAULT; | |
b9fb9ee0 AB |
1089 | |
1090 | ret = 0; | |
6ae7feb3 | 1091 | if ((u & ~MACVTAP_FEATURES) != (IFF_NO_PI | IFF_TAP)) |
b9fb9ee0 AB |
1092 | ret = -EINVAL; |
1093 | else | |
39ec7de7 | 1094 | q->flags = (q->flags & ~MACVTAP_FEATURES) | u; |
b9fb9ee0 AB |
1095 | |
1096 | return ret; | |
20d29d7a AB |
1097 | |
1098 | case TUNGETIFF: | |
441ac0fc | 1099 | rtnl_lock(); |
8f475a31 | 1100 | vlan = macvtap_get_vlan(q); |
441ac0fc VY |
1101 | if (!vlan) { |
1102 | rtnl_unlock(); | |
20d29d7a | 1103 | return -ENOLINK; |
441ac0fc | 1104 | } |
20d29d7a | 1105 | |
02df55d2 | 1106 | ret = 0; |
39ec7de7 | 1107 | u = q->flags; |
13707f9e | 1108 | if (copy_to_user(&ifr->ifr_name, vlan->dev->name, IFNAMSIZ) || |
39ec7de7 | 1109 | put_user(u, &ifr->ifr_flags)) |
02df55d2 | 1110 | ret = -EFAULT; |
8f475a31 | 1111 | macvtap_put_vlan(vlan); |
441ac0fc | 1112 | rtnl_unlock(); |
02df55d2 | 1113 | return ret; |
20d29d7a | 1114 | |
815f236d JW |
1115 | case TUNSETQUEUE: |
1116 | if (get_user(u, &ifr->ifr_flags)) | |
1117 | return -EFAULT; | |
441ac0fc VY |
1118 | rtnl_lock(); |
1119 | ret = macvtap_ioctl_set_queue(file, u); | |
1120 | rtnl_unlock(); | |
82a19eb8 | 1121 | return ret; |
815f236d | 1122 | |
20d29d7a | 1123 | case TUNGETFEATURES: |
6ae7feb3 | 1124 | if (put_user(IFF_TAP | IFF_NO_PI | MACVTAP_FEATURES, up)) |
20d29d7a AB |
1125 | return -EFAULT; |
1126 | return 0; | |
1127 | ||
1128 | case TUNSETSNDBUF: | |
3ea79249 | 1129 | if (get_user(s, sp)) |
20d29d7a AB |
1130 | return -EFAULT; |
1131 | ||
3ea79249 | 1132 | q->sk.sk_sndbuf = s; |
20d29d7a AB |
1133 | return 0; |
1134 | ||
55afbd08 MT |
1135 | case TUNGETVNETHDRSZ: |
1136 | s = q->vnet_hdr_sz; | |
1137 | if (put_user(s, sp)) | |
1138 | return -EFAULT; | |
1139 | return 0; | |
1140 | ||
1141 | case TUNSETVNETHDRSZ: | |
1142 | if (get_user(s, sp)) | |
1143 | return -EFAULT; | |
1144 | if (s < (int)sizeof(struct virtio_net_hdr)) | |
1145 | return -EINVAL; | |
1146 | ||
1147 | q->vnet_hdr_sz = s; | |
1148 | return 0; | |
1149 | ||
01b07fb3 MT |
1150 | case TUNGETVNETLE: |
1151 | s = !!(q->flags & MACVTAP_VNET_LE); | |
1152 | if (put_user(s, sp)) | |
1153 | return -EFAULT; | |
1154 | return 0; | |
1155 | ||
1156 | case TUNSETVNETLE: | |
1157 | if (get_user(s, sp)) | |
1158 | return -EFAULT; | |
1159 | if (s) | |
1160 | q->flags |= MACVTAP_VNET_LE; | |
1161 | else | |
1162 | q->flags &= ~MACVTAP_VNET_LE; | |
1163 | return 0; | |
1164 | ||
8b8e658b GK |
1165 | case TUNGETVNETBE: |
1166 | return macvtap_get_vnet_be(q, sp); | |
1167 | ||
1168 | case TUNSETVNETBE: | |
1169 | return macvtap_set_vnet_be(q, sp); | |
1170 | ||
20d29d7a AB |
1171 | case TUNSETOFFLOAD: |
1172 | /* let the user check for future flags */ | |
1173 | if (arg & ~(TUN_F_CSUM | TUN_F_TSO4 | TUN_F_TSO6 | | |
e3e3c423 | 1174 | TUN_F_TSO_ECN | TUN_F_UFO)) |
20d29d7a AB |
1175 | return -EINVAL; |
1176 | ||
2be5c767 VY |
1177 | rtnl_lock(); |
1178 | ret = set_offload(q, arg); | |
1179 | rtnl_unlock(); | |
1180 | return ret; | |
20d29d7a | 1181 | |
b5082083 JC |
1182 | case SIOCGIFHWADDR: |
1183 | rtnl_lock(); | |
1184 | vlan = macvtap_get_vlan(q); | |
1185 | if (!vlan) { | |
1186 | rtnl_unlock(); | |
1187 | return -ENOLINK; | |
1188 | } | |
1189 | ret = 0; | |
1190 | u = vlan->dev->type; | |
1191 | if (copy_to_user(&ifr->ifr_name, vlan->dev->name, IFNAMSIZ) || | |
1192 | copy_to_user(&ifr->ifr_hwaddr.sa_data, vlan->dev->dev_addr, ETH_ALEN) || | |
1193 | put_user(u, &ifr->ifr_hwaddr.sa_family)) | |
1194 | ret = -EFAULT; | |
1195 | macvtap_put_vlan(vlan); | |
1196 | rtnl_unlock(); | |
1197 | return ret; | |
1198 | ||
1199 | case SIOCSIFHWADDR: | |
7f460d30 JC |
1200 | if (copy_from_user(&sa, &ifr->ifr_hwaddr, sizeof(sa))) |
1201 | return -EFAULT; | |
b5082083 JC |
1202 | rtnl_lock(); |
1203 | vlan = macvtap_get_vlan(q); | |
1204 | if (!vlan) { | |
1205 | rtnl_unlock(); | |
1206 | return -ENOLINK; | |
1207 | } | |
7f460d30 | 1208 | ret = dev_set_mac_address(vlan->dev, &sa); |
b5082083 JC |
1209 | macvtap_put_vlan(vlan); |
1210 | rtnl_unlock(); | |
1211 | return ret; | |
1212 | ||
20d29d7a AB |
1213 | default: |
1214 | return -EINVAL; | |
1215 | } | |
1216 | } | |
1217 | ||
1218 | #ifdef CONFIG_COMPAT | |
1219 | static long macvtap_compat_ioctl(struct file *file, unsigned int cmd, | |
1220 | unsigned long arg) | |
1221 | { | |
1222 | return macvtap_ioctl(file, cmd, (unsigned long)compat_ptr(arg)); | |
1223 | } | |
1224 | #endif | |
1225 | ||
1226 | static const struct file_operations macvtap_fops = { | |
1227 | .owner = THIS_MODULE, | |
1228 | .open = macvtap_open, | |
1229 | .release = macvtap_release, | |
3af0bfe5 | 1230 | .read_iter = macvtap_read_iter, |
f5ff53b4 | 1231 | .write_iter = macvtap_write_iter, |
20d29d7a AB |
1232 | .poll = macvtap_poll, |
1233 | .llseek = no_llseek, | |
1234 | .unlocked_ioctl = macvtap_ioctl, | |
1235 | #ifdef CONFIG_COMPAT | |
1236 | .compat_ioctl = macvtap_compat_ioctl, | |
1237 | #endif | |
1238 | }; | |
1239 | ||
1b784140 YX |
1240 | static int macvtap_sendmsg(struct socket *sock, struct msghdr *m, |
1241 | size_t total_len) | |
501c774c AB |
1242 | { |
1243 | struct macvtap_queue *q = container_of(sock, struct macvtap_queue, sock); | |
c0371da6 | 1244 | return macvtap_get_user(q, m, &m->msg_iter, m->msg_flags & MSG_DONTWAIT); |
501c774c AB |
1245 | } |
1246 | ||
1b784140 YX |
1247 | static int macvtap_recvmsg(struct socket *sock, struct msghdr *m, |
1248 | size_t total_len, int flags) | |
501c774c AB |
1249 | { |
1250 | struct macvtap_queue *q = container_of(sock, struct macvtap_queue, sock); | |
1251 | int ret; | |
1252 | if (flags & ~(MSG_DONTWAIT|MSG_TRUNC)) | |
1253 | return -EINVAL; | |
c0371da6 | 1254 | ret = macvtap_do_read(q, &m->msg_iter, flags & MSG_DONTWAIT); |
de2aa476 DM |
1255 | if (ret > total_len) { |
1256 | m->msg_flags |= MSG_TRUNC; | |
1257 | ret = flags & MSG_TRUNC ? ret : total_len; | |
1258 | } | |
501c774c AB |
1259 | return ret; |
1260 | } | |
1261 | ||
1262 | /* Ops structure to mimic raw sockets with tun */ | |
1263 | static const struct proto_ops macvtap_socket_ops = { | |
1264 | .sendmsg = macvtap_sendmsg, | |
1265 | .recvmsg = macvtap_recvmsg, | |
1266 | }; | |
1267 | ||
1268 | /* Get an underlying socket object from tun file. Returns error unless file is | |
1269 | * attached to a device. The returned object works like a packet socket, it | |
1270 | * can be used for sock_sendmsg/sock_recvmsg. The caller is responsible for | |
1271 | * holding a reference to the file for as long as the socket is in use. */ | |
1272 | struct socket *macvtap_get_socket(struct file *file) | |
1273 | { | |
1274 | struct macvtap_queue *q; | |
1275 | if (file->f_op != &macvtap_fops) | |
1276 | return ERR_PTR(-EINVAL); | |
1277 | q = file->private_data; | |
1278 | if (!q) | |
1279 | return ERR_PTR(-EBADFD); | |
1280 | return &q->sock; | |
1281 | } | |
1282 | EXPORT_SYMBOL_GPL(macvtap_get_socket); | |
1283 | ||
9bf1907f EB |
1284 | static int macvtap_device_event(struct notifier_block *unused, |
1285 | unsigned long event, void *ptr) | |
1286 | { | |
351638e7 | 1287 | struct net_device *dev = netdev_notifier_info_to_dev(ptr); |
e09eff7f | 1288 | struct macvlan_dev *vlan; |
9bf1907f EB |
1289 | struct device *classdev; |
1290 | dev_t devt; | |
e09eff7f | 1291 | int err; |
17af2bce | 1292 | char tap_name[IFNAMSIZ]; |
9bf1907f EB |
1293 | |
1294 | if (dev->rtnl_link_ops != &macvtap_link_ops) | |
1295 | return NOTIFY_DONE; | |
1296 | ||
17af2bce | 1297 | snprintf(tap_name, IFNAMSIZ, "tap%d", dev->ifindex); |
e09eff7f | 1298 | vlan = netdev_priv(dev); |
9bf1907f EB |
1299 | |
1300 | switch (event) { | |
1301 | case NETDEV_REGISTER: | |
1302 | /* Create the device node here after the network device has | |
1303 | * been registered but before register_netdevice has | |
1304 | * finished running. | |
1305 | */ | |
e09eff7f EB |
1306 | err = macvtap_get_minor(vlan); |
1307 | if (err) | |
1308 | return notifier_from_errno(err); | |
1309 | ||
1310 | devt = MKDEV(MAJOR(macvtap_major), vlan->minor); | |
17af2bce MA |
1311 | classdev = device_create(&macvtap_class, &dev->dev, devt, |
1312 | dev, tap_name); | |
e09eff7f EB |
1313 | if (IS_ERR(classdev)) { |
1314 | macvtap_free_minor(vlan); | |
9bf1907f | 1315 | return notifier_from_errno(PTR_ERR(classdev)); |
e09eff7f | 1316 | } |
17af2bce MA |
1317 | err = sysfs_create_link(&dev->dev.kobj, &classdev->kobj, |
1318 | tap_name); | |
1319 | if (err) | |
1320 | return notifier_from_errno(err); | |
9bf1907f EB |
1321 | break; |
1322 | case NETDEV_UNREGISTER: | |
e96c37f1 FR |
1323 | /* vlan->minor == 0 if NETDEV_REGISTER above failed */ |
1324 | if (vlan->minor == 0) | |
1325 | break; | |
17af2bce | 1326 | sysfs_remove_link(&dev->dev.kobj, tap_name); |
e09eff7f | 1327 | devt = MKDEV(MAJOR(macvtap_major), vlan->minor); |
17af2bce | 1328 | device_destroy(&macvtap_class, devt); |
e09eff7f | 1329 | macvtap_free_minor(vlan); |
9bf1907f EB |
1330 | break; |
1331 | } | |
1332 | ||
1333 | return NOTIFY_DONE; | |
1334 | } | |
1335 | ||
1336 | static struct notifier_block macvtap_notifier_block __read_mostly = { | |
1337 | .notifier_call = macvtap_device_event, | |
1338 | }; | |
1339 | ||
20d29d7a AB |
1340 | static int macvtap_init(void) |
1341 | { | |
1342 | int err; | |
1343 | ||
1344 | err = alloc_chrdev_region(&macvtap_major, 0, | |
1345 | MACVTAP_NUM_DEVS, "macvtap"); | |
1346 | if (err) | |
1347 | goto out1; | |
1348 | ||
1349 | cdev_init(&macvtap_cdev, &macvtap_fops); | |
1350 | err = cdev_add(&macvtap_cdev, macvtap_major, MACVTAP_NUM_DEVS); | |
1351 | if (err) | |
1352 | goto out2; | |
1353 | ||
17af2bce MA |
1354 | err = class_register(&macvtap_class); |
1355 | if (err) | |
20d29d7a | 1356 | goto out3; |
20d29d7a | 1357 | |
9bf1907f | 1358 | err = register_netdevice_notifier(&macvtap_notifier_block); |
20d29d7a AB |
1359 | if (err) |
1360 | goto out4; | |
1361 | ||
9bf1907f EB |
1362 | err = macvlan_link_register(&macvtap_link_ops); |
1363 | if (err) | |
1364 | goto out5; | |
1365 | ||
20d29d7a AB |
1366 | return 0; |
1367 | ||
9bf1907f EB |
1368 | out5: |
1369 | unregister_netdevice_notifier(&macvtap_notifier_block); | |
20d29d7a | 1370 | out4: |
17af2bce | 1371 | class_unregister(&macvtap_class); |
20d29d7a AB |
1372 | out3: |
1373 | cdev_del(&macvtap_cdev); | |
1374 | out2: | |
1375 | unregister_chrdev_region(macvtap_major, MACVTAP_NUM_DEVS); | |
1376 | out1: | |
1377 | return err; | |
1378 | } | |
1379 | module_init(macvtap_init); | |
1380 | ||
1381 | static void macvtap_exit(void) | |
1382 | { | |
1383 | rtnl_link_unregister(&macvtap_link_ops); | |
9bf1907f | 1384 | unregister_netdevice_notifier(&macvtap_notifier_block); |
17af2bce | 1385 | class_unregister(&macvtap_class); |
20d29d7a AB |
1386 | cdev_del(&macvtap_cdev); |
1387 | unregister_chrdev_region(macvtap_major, MACVTAP_NUM_DEVS); | |
d5de1987 | 1388 | idr_destroy(&minor_idr); |
20d29d7a AB |
1389 | } |
1390 | module_exit(macvtap_exit); | |
1391 | ||
1392 | MODULE_ALIAS_RTNL_LINK("macvtap"); | |
1393 | MODULE_AUTHOR("Arnd Bergmann <arnd@arndb.de>"); | |
1394 | MODULE_LICENSE("GPL"); |