Commit | Line | Data |
---|---|---|
20d29d7a AB |
1 | #include <linux/etherdevice.h> |
2 | #include <linux/if_macvlan.h> | |
3 | #include <linux/interrupt.h> | |
4 | #include <linux/nsproxy.h> | |
5 | #include <linux/compat.h> | |
6 | #include <linux/if_tun.h> | |
7 | #include <linux/module.h> | |
8 | #include <linux/skbuff.h> | |
9 | #include <linux/cache.h> | |
10 | #include <linux/sched.h> | |
11 | #include <linux/types.h> | |
5a0e3ad6 | 12 | #include <linux/slab.h> |
20d29d7a AB |
13 | #include <linux/init.h> |
14 | #include <linux/wait.h> | |
15 | #include <linux/cdev.h> | |
16 | #include <linux/fs.h> | |
17 | ||
18 | #include <net/net_namespace.h> | |
19 | #include <net/rtnetlink.h> | |
20 | #include <net/sock.h> | |
b9fb9ee0 | 21 | #include <linux/virtio_net.h> |
20d29d7a AB |
22 | |
23 | /* | |
24 | * A macvtap queue is the central object of this driver, it connects | |
25 | * an open character device to a macvlan interface. There can be | |
26 | * multiple queues on one interface, which map back to queues | |
27 | * implemented in hardware on the underlying device. | |
28 | * | |
29 | * macvtap_proto is used to allocate queues through the sock allocation | |
30 | * mechanism. | |
31 | * | |
32 | * TODO: multiqueue support is currently not implemented, even though | |
33 | * macvtap is basically prepared for that. We will need to add this | |
34 | * here as well as in virtio-net and qemu to get line rate on 10gbit | |
35 | * adapters from a guest. | |
36 | */ | |
37 | struct macvtap_queue { | |
38 | struct sock sk; | |
39 | struct socket sock; | |
43815482 | 40 | struct socket_wq wq; |
55afbd08 | 41 | int vnet_hdr_sz; |
13707f9e | 42 | struct macvlan_dev __rcu *vlan; |
20d29d7a | 43 | struct file *file; |
b9fb9ee0 | 44 | unsigned int flags; |
20d29d7a AB |
45 | }; |
46 | ||
47 | static struct proto macvtap_proto = { | |
48 | .name = "macvtap", | |
49 | .owner = THIS_MODULE, | |
50 | .obj_size = sizeof (struct macvtap_queue), | |
51 | }; | |
52 | ||
53 | /* | |
54 | * Minor number matches netdev->ifindex, so need a potentially | |
55 | * large value. This also makes it possible to split the | |
56 | * tap functionality out again in the future by offering it | |
57 | * from other drivers besides macvtap. As long as every device | |
58 | * only has one tap, the interface numbers assure that the | |
59 | * device nodes are unique. | |
60 | */ | |
1ebed71a | 61 | static dev_t macvtap_major; |
20d29d7a | 62 | #define MACVTAP_NUM_DEVS 65536 |
97bc3633 | 63 | #define GOODCOPY_LEN 128 |
20d29d7a AB |
64 | static struct class *macvtap_class; |
65 | static struct cdev macvtap_cdev; | |
66 | ||
501c774c AB |
67 | static const struct proto_ops macvtap_socket_ops; |
68 | ||
20d29d7a AB |
69 | /* |
70 | * RCU usage: | |
02df55d2 AB |
71 | * The macvtap_queue and the macvlan_dev are loosely coupled, the |
72 | * pointers from one to the other can only be read while rcu_read_lock | |
73 | * or macvtap_lock is held. | |
20d29d7a | 74 | * |
02df55d2 AB |
75 | * Both the file and the macvlan_dev hold a reference on the macvtap_queue |
76 | * through sock_hold(&q->sk). When the macvlan_dev goes away first, | |
77 | * q->vlan becomes inaccessible. When the files gets closed, | |
78 | * macvtap_get_queue() fails. | |
20d29d7a | 79 | * |
02df55d2 AB |
80 | * There may still be references to the struct sock inside of the |
81 | * queue from outbound SKBs, but these never reference back to the | |
82 | * file or the dev. The data structure is freed through __sk_free | |
83 | * when both our references and any pending SKBs are gone. | |
20d29d7a AB |
84 | */ |
85 | static DEFINE_SPINLOCK(macvtap_lock); | |
86 | ||
87 | /* | |
1565c7c1 KK |
88 | * get_slot: return a [unused/occupied] slot in vlan->taps[]: |
89 | * - if 'q' is NULL, return the first empty slot; | |
90 | * - otherwise, return the slot this pointer occupies. | |
20d29d7a | 91 | */ |
1565c7c1 KK |
92 | static int get_slot(struct macvlan_dev *vlan, struct macvtap_queue *q) |
93 | { | |
94 | int i; | |
95 | ||
96 | for (i = 0; i < MAX_MACVTAP_QUEUES; i++) { | |
97 | if (rcu_dereference(vlan->taps[i]) == q) | |
98 | return i; | |
99 | } | |
100 | ||
101 | /* Should never happen */ | |
102 | BUG_ON(1); | |
103 | } | |
104 | ||
20d29d7a AB |
105 | static int macvtap_set_queue(struct net_device *dev, struct file *file, |
106 | struct macvtap_queue *q) | |
107 | { | |
108 | struct macvlan_dev *vlan = netdev_priv(dev); | |
1565c7c1 | 109 | int index; |
20d29d7a AB |
110 | int err = -EBUSY; |
111 | ||
112 | spin_lock(&macvtap_lock); | |
1565c7c1 | 113 | if (vlan->numvtaps == MAX_MACVTAP_QUEUES) |
20d29d7a AB |
114 | goto out; |
115 | ||
116 | err = 0; | |
1565c7c1 | 117 | index = get_slot(vlan, NULL); |
02df55d2 | 118 | rcu_assign_pointer(q->vlan, vlan); |
1565c7c1 | 119 | rcu_assign_pointer(vlan->taps[index], q); |
02df55d2 | 120 | sock_hold(&q->sk); |
20d29d7a AB |
121 | |
122 | q->file = file; | |
02df55d2 | 123 | file->private_data = q; |
20d29d7a | 124 | |
1565c7c1 KK |
125 | vlan->numvtaps++; |
126 | ||
20d29d7a AB |
127 | out: |
128 | spin_unlock(&macvtap_lock); | |
129 | return err; | |
130 | } | |
131 | ||
132 | /* | |
02df55d2 AB |
133 | * The file owning the queue got closed, give up both |
134 | * the reference that the files holds as well as the | |
135 | * one from the macvlan_dev if that still exists. | |
20d29d7a AB |
136 | * |
137 | * Using the spinlock makes sure that we don't get | |
138 | * to the queue again after destroying it. | |
20d29d7a | 139 | */ |
02df55d2 | 140 | static void macvtap_put_queue(struct macvtap_queue *q) |
20d29d7a | 141 | { |
02df55d2 | 142 | struct macvlan_dev *vlan; |
20d29d7a AB |
143 | |
144 | spin_lock(&macvtap_lock); | |
13707f9e ED |
145 | vlan = rcu_dereference_protected(q->vlan, |
146 | lockdep_is_held(&macvtap_lock)); | |
02df55d2 | 147 | if (vlan) { |
1565c7c1 KK |
148 | int index = get_slot(vlan, q); |
149 | ||
150 | rcu_assign_pointer(vlan->taps[index], NULL); | |
02df55d2 AB |
151 | rcu_assign_pointer(q->vlan, NULL); |
152 | sock_put(&q->sk); | |
1565c7c1 | 153 | --vlan->numvtaps; |
20d29d7a AB |
154 | } |
155 | ||
20d29d7a AB |
156 | spin_unlock(&macvtap_lock); |
157 | ||
158 | synchronize_rcu(); | |
159 | sock_put(&q->sk); | |
160 | } | |
161 | ||
162 | /* | |
1565c7c1 KK |
163 | * Select a queue based on the rxq of the device on which this packet |
164 | * arrived. If the incoming device is not mq, calculate a flow hash | |
165 | * to select a queue. If all fails, find the first available queue. | |
166 | * Cache vlan->numvtaps since it can become zero during the execution | |
167 | * of this function. | |
20d29d7a AB |
168 | */ |
169 | static struct macvtap_queue *macvtap_get_queue(struct net_device *dev, | |
170 | struct sk_buff *skb) | |
171 | { | |
172 | struct macvlan_dev *vlan = netdev_priv(dev); | |
1565c7c1 KK |
173 | struct macvtap_queue *tap = NULL; |
174 | int numvtaps = vlan->numvtaps; | |
175 | __u32 rxq; | |
176 | ||
177 | if (!numvtaps) | |
178 | goto out; | |
179 | ||
180 | if (likely(skb_rx_queue_recorded(skb))) { | |
181 | rxq = skb_get_rx_queue(skb); | |
20d29d7a | 182 | |
1565c7c1 KK |
183 | while (unlikely(rxq >= numvtaps)) |
184 | rxq -= numvtaps; | |
185 | ||
186 | tap = rcu_dereference(vlan->taps[rxq]); | |
187 | if (tap) | |
188 | goto out; | |
189 | } | |
190 | ||
191 | /* Check if we can use flow to select a queue */ | |
192 | rxq = skb_get_rxhash(skb); | |
193 | if (rxq) { | |
194 | tap = rcu_dereference(vlan->taps[rxq % numvtaps]); | |
195 | if (tap) | |
196 | goto out; | |
197 | } | |
198 | ||
199 | /* Everything failed - find first available queue */ | |
200 | for (rxq = 0; rxq < MAX_MACVTAP_QUEUES; rxq++) { | |
201 | tap = rcu_dereference(vlan->taps[rxq]); | |
202 | if (tap) | |
203 | break; | |
204 | } | |
205 | ||
206 | out: | |
207 | return tap; | |
20d29d7a AB |
208 | } |
209 | ||
02df55d2 AB |
210 | /* |
211 | * The net_device is going away, give up the reference | |
1565c7c1 KK |
212 | * that it holds on all queues and safely set the pointer |
213 | * from the queues to NULL. | |
02df55d2 | 214 | */ |
20d29d7a AB |
215 | static void macvtap_del_queues(struct net_device *dev) |
216 | { | |
217 | struct macvlan_dev *vlan = netdev_priv(dev); | |
1565c7c1 KK |
218 | struct macvtap_queue *q, *qlist[MAX_MACVTAP_QUEUES]; |
219 | int i, j = 0; | |
02df55d2 | 220 | |
1565c7c1 | 221 | /* macvtap_put_queue can free some slots, so go through all slots */ |
02df55d2 | 222 | spin_lock(&macvtap_lock); |
1565c7c1 | 223 | for (i = 0; i < MAX_MACVTAP_QUEUES && vlan->numvtaps; i++) { |
13707f9e ED |
224 | q = rcu_dereference_protected(vlan->taps[i], |
225 | lockdep_is_held(&macvtap_lock)); | |
1565c7c1 KK |
226 | if (q) { |
227 | qlist[j++] = q; | |
228 | rcu_assign_pointer(vlan->taps[i], NULL); | |
229 | rcu_assign_pointer(q->vlan, NULL); | |
230 | vlan->numvtaps--; | |
231 | } | |
564517e8 | 232 | } |
1565c7c1 | 233 | BUG_ON(vlan->numvtaps != 0); |
02df55d2 AB |
234 | spin_unlock(&macvtap_lock); |
235 | ||
236 | synchronize_rcu(); | |
1565c7c1 KK |
237 | |
238 | for (--j; j >= 0; j--) | |
239 | sock_put(&qlist[j]->sk); | |
20d29d7a AB |
240 | } |
241 | ||
242 | /* | |
243 | * Forward happens for data that gets sent from one macvlan | |
244 | * endpoint to another one in bridge mode. We just take | |
245 | * the skb and put it into the receive queue. | |
246 | */ | |
247 | static int macvtap_forward(struct net_device *dev, struct sk_buff *skb) | |
248 | { | |
249 | struct macvtap_queue *q = macvtap_get_queue(dev, skb); | |
250 | if (!q) | |
8a35747a HX |
251 | goto drop; |
252 | ||
253 | if (skb_queue_len(&q->sk.sk_receive_queue) >= dev->tx_queue_len) | |
254 | goto drop; | |
20d29d7a AB |
255 | |
256 | skb_queue_tail(&q->sk.sk_receive_queue, skb); | |
4a4771a5 | 257 | wake_up_interruptible_poll(sk_sleep(&q->sk), POLLIN | POLLRDNORM | POLLRDBAND); |
8a35747a HX |
258 | return NET_RX_SUCCESS; |
259 | ||
260 | drop: | |
261 | kfree_skb(skb); | |
262 | return NET_RX_DROP; | |
20d29d7a AB |
263 | } |
264 | ||
265 | /* | |
266 | * Receive is for data from the external interface (lowerdev), | |
267 | * in case of macvtap, we can treat that the same way as | |
268 | * forward, which macvlan cannot. | |
269 | */ | |
270 | static int macvtap_receive(struct sk_buff *skb) | |
271 | { | |
272 | skb_push(skb, ETH_HLEN); | |
273 | return macvtap_forward(skb->dev, skb); | |
274 | } | |
275 | ||
276 | static int macvtap_newlink(struct net *src_net, | |
277 | struct net_device *dev, | |
278 | struct nlattr *tb[], | |
279 | struct nlattr *data[]) | |
280 | { | |
281 | struct device *classdev; | |
282 | dev_t devt; | |
283 | int err; | |
284 | ||
285 | err = macvlan_common_newlink(src_net, dev, tb, data, | |
286 | macvtap_receive, macvtap_forward); | |
287 | if (err) | |
288 | goto out; | |
289 | ||
290 | devt = MKDEV(MAJOR(macvtap_major), dev->ifindex); | |
291 | ||
292 | classdev = device_create(macvtap_class, &dev->dev, devt, | |
293 | dev, "tap%d", dev->ifindex); | |
294 | if (IS_ERR(classdev)) { | |
295 | err = PTR_ERR(classdev); | |
296 | macvtap_del_queues(dev); | |
297 | } | |
298 | ||
299 | out: | |
300 | return err; | |
301 | } | |
302 | ||
303 | static void macvtap_dellink(struct net_device *dev, | |
304 | struct list_head *head) | |
305 | { | |
306 | device_destroy(macvtap_class, | |
307 | MKDEV(MAJOR(macvtap_major), dev->ifindex)); | |
308 | ||
309 | macvtap_del_queues(dev); | |
310 | macvlan_dellink(dev, head); | |
311 | } | |
312 | ||
8a35747a HX |
313 | static void macvtap_setup(struct net_device *dev) |
314 | { | |
315 | macvlan_common_setup(dev); | |
316 | dev->tx_queue_len = TUN_READQ_SIZE; | |
317 | } | |
318 | ||
20d29d7a AB |
319 | static struct rtnl_link_ops macvtap_link_ops __read_mostly = { |
320 | .kind = "macvtap", | |
8a35747a | 321 | .setup = macvtap_setup, |
20d29d7a AB |
322 | .newlink = macvtap_newlink, |
323 | .dellink = macvtap_dellink, | |
324 | }; | |
325 | ||
326 | ||
327 | static void macvtap_sock_write_space(struct sock *sk) | |
328 | { | |
43815482 ED |
329 | wait_queue_head_t *wqueue; |
330 | ||
20d29d7a AB |
331 | if (!sock_writeable(sk) || |
332 | !test_and_clear_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags)) | |
333 | return; | |
334 | ||
43815482 ED |
335 | wqueue = sk_sleep(sk); |
336 | if (wqueue && waitqueue_active(wqueue)) | |
337 | wake_up_interruptible_poll(wqueue, POLLOUT | POLLWRNORM | POLLWRBAND); | |
20d29d7a AB |
338 | } |
339 | ||
340 | static int macvtap_open(struct inode *inode, struct file *file) | |
341 | { | |
342 | struct net *net = current->nsproxy->net_ns; | |
343 | struct net_device *dev = dev_get_by_index(net, iminor(inode)); | |
97bc3633 | 344 | struct macvlan_dev *vlan = netdev_priv(dev); |
20d29d7a AB |
345 | struct macvtap_queue *q; |
346 | int err; | |
347 | ||
348 | err = -ENODEV; | |
349 | if (!dev) | |
350 | goto out; | |
351 | ||
352 | /* check if this is a macvtap device */ | |
353 | err = -EINVAL; | |
354 | if (dev->rtnl_link_ops != &macvtap_link_ops) | |
355 | goto out; | |
356 | ||
357 | err = -ENOMEM; | |
358 | q = (struct macvtap_queue *)sk_alloc(net, AF_UNSPEC, GFP_KERNEL, | |
359 | &macvtap_proto); | |
360 | if (!q) | |
361 | goto out; | |
362 | ||
43815482 ED |
363 | q->sock.wq = &q->wq; |
364 | init_waitqueue_head(&q->wq.wait); | |
20d29d7a AB |
365 | q->sock.type = SOCK_RAW; |
366 | q->sock.state = SS_CONNECTED; | |
501c774c AB |
367 | q->sock.file = file; |
368 | q->sock.ops = &macvtap_socket_ops; | |
20d29d7a | 369 | sock_init_data(&q->sock, &q->sk); |
20d29d7a | 370 | q->sk.sk_write_space = macvtap_sock_write_space; |
b9fb9ee0 | 371 | q->flags = IFF_VNET_HDR | IFF_NO_PI | IFF_TAP; |
55afbd08 | 372 | q->vnet_hdr_sz = sizeof(struct virtio_net_hdr); |
20d29d7a | 373 | |
97bc3633 SM |
374 | /* |
375 | * so far only KVM virtio_net uses macvtap, enable zero copy between | |
376 | * guest kernel and host kernel when lower device supports zerocopy | |
377 | */ | |
378 | if (vlan) { | |
379 | if ((vlan->lowerdev->features & NETIF_F_HIGHDMA) && | |
380 | (vlan->lowerdev->features & NETIF_F_SG)) | |
381 | sock_set_flag(&q->sk, SOCK_ZEROCOPY); | |
382 | } | |
383 | ||
20d29d7a AB |
384 | err = macvtap_set_queue(dev, file, q); |
385 | if (err) | |
386 | sock_put(&q->sk); | |
387 | ||
388 | out: | |
389 | if (dev) | |
390 | dev_put(dev); | |
391 | ||
392 | return err; | |
393 | } | |
394 | ||
395 | static int macvtap_release(struct inode *inode, struct file *file) | |
396 | { | |
02df55d2 AB |
397 | struct macvtap_queue *q = file->private_data; |
398 | macvtap_put_queue(q); | |
20d29d7a AB |
399 | return 0; |
400 | } | |
401 | ||
402 | static unsigned int macvtap_poll(struct file *file, poll_table * wait) | |
403 | { | |
02df55d2 | 404 | struct macvtap_queue *q = file->private_data; |
20d29d7a AB |
405 | unsigned int mask = POLLERR; |
406 | ||
407 | if (!q) | |
408 | goto out; | |
409 | ||
410 | mask = 0; | |
43815482 | 411 | poll_wait(file, &q->wq.wait, wait); |
20d29d7a AB |
412 | |
413 | if (!skb_queue_empty(&q->sk.sk_receive_queue)) | |
414 | mask |= POLLIN | POLLRDNORM; | |
415 | ||
416 | if (sock_writeable(&q->sk) || | |
417 | (!test_and_set_bit(SOCK_ASYNC_NOSPACE, &q->sock.flags) && | |
418 | sock_writeable(&q->sk))) | |
419 | mask |= POLLOUT | POLLWRNORM; | |
420 | ||
421 | out: | |
20d29d7a AB |
422 | return mask; |
423 | } | |
424 | ||
b9fb9ee0 AB |
425 | static inline struct sk_buff *macvtap_alloc_skb(struct sock *sk, size_t prepad, |
426 | size_t len, size_t linear, | |
427 | int noblock, int *err) | |
428 | { | |
429 | struct sk_buff *skb; | |
430 | ||
431 | /* Under a page? Don't bother with paged skb. */ | |
432 | if (prepad + len < PAGE_SIZE || !linear) | |
433 | linear = len; | |
434 | ||
435 | skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock, | |
436 | err); | |
437 | if (!skb) | |
438 | return NULL; | |
439 | ||
440 | skb_reserve(skb, prepad); | |
441 | skb_put(skb, linear); | |
442 | skb->data_len = len - linear; | |
443 | skb->len += len - linear; | |
444 | ||
445 | return skb; | |
446 | } | |
447 | ||
97bc3633 SM |
448 | /* set skb frags from iovec, this can move to core network code for reuse */ |
449 | static int zerocopy_sg_from_iovec(struct sk_buff *skb, const struct iovec *from, | |
450 | int offset, size_t count) | |
451 | { | |
452 | int len = iov_length(from, count) - offset; | |
453 | int copy = skb_headlen(skb); | |
454 | int size, offset1 = 0; | |
455 | int i = 0; | |
456 | skb_frag_t *f; | |
457 | ||
458 | /* Skip over from offset */ | |
459 | while (count && (offset >= from->iov_len)) { | |
460 | offset -= from->iov_len; | |
461 | ++from; | |
462 | --count; | |
463 | } | |
464 | ||
465 | /* copy up to skb headlen */ | |
466 | while (count && (copy > 0)) { | |
467 | size = min_t(unsigned int, copy, from->iov_len - offset); | |
468 | if (copy_from_user(skb->data + offset1, from->iov_base + offset, | |
469 | size)) | |
470 | return -EFAULT; | |
471 | if (copy > size) { | |
472 | ++from; | |
473 | --count; | |
474 | } | |
475 | copy -= size; | |
476 | offset1 += size; | |
477 | offset = 0; | |
478 | } | |
479 | ||
480 | if (len == offset1) | |
481 | return 0; | |
482 | ||
483 | while (count--) { | |
484 | struct page *page[MAX_SKB_FRAGS]; | |
485 | int num_pages; | |
486 | unsigned long base; | |
487 | ||
488 | len = from->iov_len - offset1; | |
489 | if (!len) { | |
490 | offset1 = 0; | |
491 | ++from; | |
492 | continue; | |
493 | } | |
494 | base = (unsigned long)from->iov_base + offset1; | |
495 | size = ((base & ~PAGE_MASK) + len + ~PAGE_MASK) >> PAGE_SHIFT; | |
496 | num_pages = get_user_pages_fast(base, size, 0, &page[i]); | |
497 | if ((num_pages != size) || | |
498 | (num_pages > MAX_SKB_FRAGS - skb_shinfo(skb)->nr_frags)) | |
499 | /* put_page is in skb free */ | |
500 | return -EFAULT; | |
501 | skb->data_len += len; | |
502 | skb->len += len; | |
503 | skb->truesize += len; | |
504 | atomic_add(len, &skb->sk->sk_wmem_alloc); | |
505 | while (len) { | |
506 | f = &skb_shinfo(skb)->frags[i]; | |
507 | f->page = page[i]; | |
508 | f->page_offset = base & ~PAGE_MASK; | |
509 | f->size = min_t(int, len, PAGE_SIZE - f->page_offset); | |
510 | skb_shinfo(skb)->nr_frags++; | |
511 | /* increase sk_wmem_alloc */ | |
512 | base += f->size; | |
513 | len -= f->size; | |
514 | i++; | |
515 | } | |
516 | offset1 = 0; | |
517 | ++from; | |
518 | } | |
519 | return 0; | |
520 | } | |
521 | ||
b9fb9ee0 AB |
522 | /* |
523 | * macvtap_skb_from_vnet_hdr and macvtap_skb_to_vnet_hdr should | |
524 | * be shared with the tun/tap driver. | |
525 | */ | |
526 | static int macvtap_skb_from_vnet_hdr(struct sk_buff *skb, | |
527 | struct virtio_net_hdr *vnet_hdr) | |
528 | { | |
529 | unsigned short gso_type = 0; | |
530 | if (vnet_hdr->gso_type != VIRTIO_NET_HDR_GSO_NONE) { | |
531 | switch (vnet_hdr->gso_type & ~VIRTIO_NET_HDR_GSO_ECN) { | |
532 | case VIRTIO_NET_HDR_GSO_TCPV4: | |
533 | gso_type = SKB_GSO_TCPV4; | |
534 | break; | |
535 | case VIRTIO_NET_HDR_GSO_TCPV6: | |
536 | gso_type = SKB_GSO_TCPV6; | |
537 | break; | |
538 | case VIRTIO_NET_HDR_GSO_UDP: | |
539 | gso_type = SKB_GSO_UDP; | |
540 | break; | |
541 | default: | |
542 | return -EINVAL; | |
543 | } | |
544 | ||
545 | if (vnet_hdr->gso_type & VIRTIO_NET_HDR_GSO_ECN) | |
546 | gso_type |= SKB_GSO_TCP_ECN; | |
547 | ||
548 | if (vnet_hdr->gso_size == 0) | |
549 | return -EINVAL; | |
550 | } | |
551 | ||
552 | if (vnet_hdr->flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) { | |
553 | if (!skb_partial_csum_set(skb, vnet_hdr->csum_start, | |
554 | vnet_hdr->csum_offset)) | |
555 | return -EINVAL; | |
556 | } | |
557 | ||
558 | if (vnet_hdr->gso_type != VIRTIO_NET_HDR_GSO_NONE) { | |
559 | skb_shinfo(skb)->gso_size = vnet_hdr->gso_size; | |
560 | skb_shinfo(skb)->gso_type = gso_type; | |
561 | ||
562 | /* Header must be checked, and gso_segs computed. */ | |
563 | skb_shinfo(skb)->gso_type |= SKB_GSO_DODGY; | |
564 | skb_shinfo(skb)->gso_segs = 0; | |
565 | } | |
566 | return 0; | |
567 | } | |
568 | ||
569 | static int macvtap_skb_to_vnet_hdr(const struct sk_buff *skb, | |
570 | struct virtio_net_hdr *vnet_hdr) | |
571 | { | |
572 | memset(vnet_hdr, 0, sizeof(*vnet_hdr)); | |
573 | ||
574 | if (skb_is_gso(skb)) { | |
575 | struct skb_shared_info *sinfo = skb_shinfo(skb); | |
576 | ||
577 | /* This is a hint as to how much should be linear. */ | |
578 | vnet_hdr->hdr_len = skb_headlen(skb); | |
579 | vnet_hdr->gso_size = sinfo->gso_size; | |
580 | if (sinfo->gso_type & SKB_GSO_TCPV4) | |
581 | vnet_hdr->gso_type = VIRTIO_NET_HDR_GSO_TCPV4; | |
582 | else if (sinfo->gso_type & SKB_GSO_TCPV6) | |
583 | vnet_hdr->gso_type = VIRTIO_NET_HDR_GSO_TCPV6; | |
584 | else if (sinfo->gso_type & SKB_GSO_UDP) | |
585 | vnet_hdr->gso_type = VIRTIO_NET_HDR_GSO_UDP; | |
586 | else | |
587 | BUG(); | |
588 | if (sinfo->gso_type & SKB_GSO_TCP_ECN) | |
589 | vnet_hdr->gso_type |= VIRTIO_NET_HDR_GSO_ECN; | |
590 | } else | |
591 | vnet_hdr->gso_type = VIRTIO_NET_HDR_GSO_NONE; | |
592 | ||
593 | if (skb->ip_summed == CHECKSUM_PARTIAL) { | |
594 | vnet_hdr->flags = VIRTIO_NET_HDR_F_NEEDS_CSUM; | |
55508d60 | 595 | vnet_hdr->csum_start = skb_checksum_start_offset(skb); |
b9fb9ee0 | 596 | vnet_hdr->csum_offset = skb->csum_offset; |
10a8d94a JW |
597 | } else if (skb->ip_summed == CHECKSUM_UNNECESSARY) { |
598 | vnet_hdr->flags = VIRTIO_NET_HDR_F_DATA_VALID; | |
b9fb9ee0 AB |
599 | } /* else everything is zero */ |
600 | ||
601 | return 0; | |
602 | } | |
603 | ||
604 | ||
20d29d7a | 605 | /* Get packet from user space buffer */ |
97bc3633 SM |
606 | static ssize_t macvtap_get_user(struct macvtap_queue *q, struct msghdr *m, |
607 | const struct iovec *iv, unsigned long total_len, | |
608 | size_t count, int noblock) | |
20d29d7a AB |
609 | { |
610 | struct sk_buff *skb; | |
02df55d2 | 611 | struct macvlan_dev *vlan; |
97bc3633 | 612 | unsigned long len = total_len; |
20d29d7a | 613 | int err; |
b9fb9ee0 AB |
614 | struct virtio_net_hdr vnet_hdr = { 0 }; |
615 | int vnet_hdr_len = 0; | |
97bc3633 SM |
616 | int copylen; |
617 | bool zerocopy = false; | |
b9fb9ee0 AB |
618 | |
619 | if (q->flags & IFF_VNET_HDR) { | |
55afbd08 | 620 | vnet_hdr_len = q->vnet_hdr_sz; |
b9fb9ee0 AB |
621 | |
622 | err = -EINVAL; | |
ce3c8692 | 623 | if (len < vnet_hdr_len) |
b9fb9ee0 | 624 | goto err; |
ce3c8692 | 625 | len -= vnet_hdr_len; |
b9fb9ee0 AB |
626 | |
627 | err = memcpy_fromiovecend((void *)&vnet_hdr, iv, 0, | |
55afbd08 | 628 | sizeof(vnet_hdr)); |
b9fb9ee0 AB |
629 | if (err < 0) |
630 | goto err; | |
631 | if ((vnet_hdr.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) && | |
632 | vnet_hdr.csum_start + vnet_hdr.csum_offset + 2 > | |
633 | vnet_hdr.hdr_len) | |
634 | vnet_hdr.hdr_len = vnet_hdr.csum_start + | |
635 | vnet_hdr.csum_offset + 2; | |
636 | err = -EINVAL; | |
637 | if (vnet_hdr.hdr_len > len) | |
638 | goto err; | |
639 | } | |
20d29d7a | 640 | |
b9fb9ee0 | 641 | err = -EINVAL; |
20d29d7a | 642 | if (unlikely(len < ETH_HLEN)) |
b9fb9ee0 | 643 | goto err; |
20d29d7a | 644 | |
97bc3633 SM |
645 | if (m && m->msg_control && sock_flag(&q->sk, SOCK_ZEROCOPY)) |
646 | zerocopy = true; | |
647 | ||
648 | if (zerocopy) { | |
649 | /* There are 256 bytes to be copied in skb, so there is enough | |
650 | * room for skb expand head in case it is used. | |
651 | * The rest buffer is mapped from userspace. | |
652 | */ | |
653 | copylen = vnet_hdr.hdr_len; | |
654 | if (!copylen) | |
655 | copylen = GOODCOPY_LEN; | |
656 | } else | |
657 | copylen = len; | |
658 | ||
659 | skb = macvtap_alloc_skb(&q->sk, NET_IP_ALIGN, copylen, | |
660 | vnet_hdr.hdr_len, noblock, &err); | |
02df55d2 AB |
661 | if (!skb) |
662 | goto err; | |
20d29d7a | 663 | |
97bc3633 SM |
664 | if (zerocopy) { |
665 | err = zerocopy_sg_from_iovec(skb, iv, vnet_hdr_len, count); | |
666 | skb_shinfo(skb)->tx_flags |= SKBTX_DEV_ZEROCOPY; | |
667 | } else | |
668 | err = skb_copy_datagram_from_iovec(skb, 0, iv, vnet_hdr_len, | |
669 | len); | |
02df55d2 | 670 | if (err) |
b9fb9ee0 | 671 | goto err_kfree; |
20d29d7a AB |
672 | |
673 | skb_set_network_header(skb, ETH_HLEN); | |
b9fb9ee0 AB |
674 | skb_reset_mac_header(skb); |
675 | skb->protocol = eth_hdr(skb)->h_proto; | |
676 | ||
677 | if (vnet_hdr_len) { | |
678 | err = macvtap_skb_from_vnet_hdr(skb, &vnet_hdr); | |
679 | if (err) | |
680 | goto err_kfree; | |
681 | } | |
682 | ||
02df55d2 | 683 | rcu_read_lock_bh(); |
13707f9e | 684 | vlan = rcu_dereference_bh(q->vlan); |
97bc3633 SM |
685 | /* copy skb_ubuf_info for callback when skb has no error */ |
686 | if (zerocopy) | |
687 | skb_shinfo(skb)->destructor_arg = m->msg_control; | |
02df55d2 AB |
688 | if (vlan) |
689 | macvlan_start_xmit(skb, vlan->dev); | |
690 | else | |
691 | kfree_skb(skb); | |
692 | rcu_read_unlock_bh(); | |
20d29d7a | 693 | |
97bc3633 | 694 | return total_len; |
02df55d2 | 695 | |
b9fb9ee0 AB |
696 | err_kfree: |
697 | kfree_skb(skb); | |
698 | ||
02df55d2 AB |
699 | err: |
700 | rcu_read_lock_bh(); | |
13707f9e | 701 | vlan = rcu_dereference_bh(q->vlan); |
02df55d2 | 702 | if (vlan) |
1ac9ad13 | 703 | vlan->dev->stats.tx_dropped++; |
02df55d2 AB |
704 | rcu_read_unlock_bh(); |
705 | ||
02df55d2 | 706 | return err; |
20d29d7a AB |
707 | } |
708 | ||
709 | static ssize_t macvtap_aio_write(struct kiocb *iocb, const struct iovec *iv, | |
710 | unsigned long count, loff_t pos) | |
711 | { | |
712 | struct file *file = iocb->ki_filp; | |
713 | ssize_t result = -ENOLINK; | |
02df55d2 | 714 | struct macvtap_queue *q = file->private_data; |
20d29d7a | 715 | |
97bc3633 SM |
716 | result = macvtap_get_user(q, NULL, iv, iov_length(iv, count), count, |
717 | file->f_flags & O_NONBLOCK); | |
20d29d7a AB |
718 | return result; |
719 | } | |
720 | ||
721 | /* Put packet to the user space buffer */ | |
722 | static ssize_t macvtap_put_user(struct macvtap_queue *q, | |
723 | const struct sk_buff *skb, | |
724 | const struct iovec *iv, int len) | |
725 | { | |
02df55d2 | 726 | struct macvlan_dev *vlan; |
20d29d7a | 727 | int ret; |
b9fb9ee0 AB |
728 | int vnet_hdr_len = 0; |
729 | ||
730 | if (q->flags & IFF_VNET_HDR) { | |
731 | struct virtio_net_hdr vnet_hdr; | |
55afbd08 | 732 | vnet_hdr_len = q->vnet_hdr_sz; |
b9fb9ee0 AB |
733 | if ((len -= vnet_hdr_len) < 0) |
734 | return -EINVAL; | |
735 | ||
736 | ret = macvtap_skb_to_vnet_hdr(skb, &vnet_hdr); | |
737 | if (ret) | |
738 | return ret; | |
739 | ||
55afbd08 | 740 | if (memcpy_toiovecend(iv, (void *)&vnet_hdr, 0, sizeof(vnet_hdr))) |
b9fb9ee0 AB |
741 | return -EFAULT; |
742 | } | |
20d29d7a AB |
743 | |
744 | len = min_t(int, skb->len, len); | |
745 | ||
b9fb9ee0 | 746 | ret = skb_copy_datagram_const_iovec(skb, 0, iv, vnet_hdr_len, len); |
20d29d7a | 747 | |
02df55d2 | 748 | rcu_read_lock_bh(); |
13707f9e | 749 | vlan = rcu_dereference_bh(q->vlan); |
501c774c AB |
750 | if (vlan) |
751 | macvlan_count_rx(vlan, len, ret == 0, 0); | |
02df55d2 | 752 | rcu_read_unlock_bh(); |
20d29d7a | 753 | |
b9fb9ee0 | 754 | return ret ? ret : (len + vnet_hdr_len); |
20d29d7a AB |
755 | } |
756 | ||
501c774c AB |
757 | static ssize_t macvtap_do_read(struct macvtap_queue *q, struct kiocb *iocb, |
758 | const struct iovec *iv, unsigned long len, | |
759 | int noblock) | |
20d29d7a | 760 | { |
20d29d7a AB |
761 | DECLARE_WAITQUEUE(wait, current); |
762 | struct sk_buff *skb; | |
501c774c | 763 | ssize_t ret = 0; |
20d29d7a | 764 | |
4a4771a5 | 765 | add_wait_queue(sk_sleep(&q->sk), &wait); |
20d29d7a AB |
766 | while (len) { |
767 | current->state = TASK_INTERRUPTIBLE; | |
768 | ||
769 | /* Read frames from the queue */ | |
770 | skb = skb_dequeue(&q->sk.sk_receive_queue); | |
771 | if (!skb) { | |
501c774c | 772 | if (noblock) { |
20d29d7a AB |
773 | ret = -EAGAIN; |
774 | break; | |
775 | } | |
776 | if (signal_pending(current)) { | |
777 | ret = -ERESTARTSYS; | |
778 | break; | |
779 | } | |
780 | /* Nothing to read, let's sleep */ | |
781 | schedule(); | |
782 | continue; | |
783 | } | |
784 | ret = macvtap_put_user(q, skb, iv, len); | |
785 | kfree_skb(skb); | |
786 | break; | |
787 | } | |
788 | ||
789 | current->state = TASK_RUNNING; | |
4a4771a5 | 790 | remove_wait_queue(sk_sleep(&q->sk), &wait); |
501c774c AB |
791 | return ret; |
792 | } | |
793 | ||
794 | static ssize_t macvtap_aio_read(struct kiocb *iocb, const struct iovec *iv, | |
795 | unsigned long count, loff_t pos) | |
796 | { | |
797 | struct file *file = iocb->ki_filp; | |
798 | struct macvtap_queue *q = file->private_data; | |
799 | ssize_t len, ret = 0; | |
20d29d7a | 800 | |
501c774c AB |
801 | len = iov_length(iv, count); |
802 | if (len < 0) { | |
803 | ret = -EINVAL; | |
804 | goto out; | |
805 | } | |
806 | ||
807 | ret = macvtap_do_read(q, iocb, iv, len, file->f_flags & O_NONBLOCK); | |
808 | ret = min_t(ssize_t, ret, len); /* XXX copied from tun.c. Why? */ | |
20d29d7a | 809 | out: |
20d29d7a AB |
810 | return ret; |
811 | } | |
812 | ||
813 | /* | |
814 | * provide compatibility with generic tun/tap interface | |
815 | */ | |
816 | static long macvtap_ioctl(struct file *file, unsigned int cmd, | |
817 | unsigned long arg) | |
818 | { | |
02df55d2 AB |
819 | struct macvtap_queue *q = file->private_data; |
820 | struct macvlan_dev *vlan; | |
20d29d7a AB |
821 | void __user *argp = (void __user *)arg; |
822 | struct ifreq __user *ifr = argp; | |
823 | unsigned int __user *up = argp; | |
824 | unsigned int u; | |
55afbd08 MT |
825 | int __user *sp = argp; |
826 | int s; | |
02df55d2 | 827 | int ret; |
20d29d7a AB |
828 | |
829 | switch (cmd) { | |
830 | case TUNSETIFF: | |
831 | /* ignore the name, just look at flags */ | |
832 | if (get_user(u, &ifr->ifr_flags)) | |
833 | return -EFAULT; | |
b9fb9ee0 AB |
834 | |
835 | ret = 0; | |
836 | if ((u & ~IFF_VNET_HDR) != (IFF_NO_PI | IFF_TAP)) | |
837 | ret = -EINVAL; | |
838 | else | |
839 | q->flags = u; | |
840 | ||
841 | return ret; | |
20d29d7a AB |
842 | |
843 | case TUNGETIFF: | |
02df55d2 | 844 | rcu_read_lock_bh(); |
13707f9e | 845 | vlan = rcu_dereference_bh(q->vlan); |
02df55d2 AB |
846 | if (vlan) |
847 | dev_hold(vlan->dev); | |
848 | rcu_read_unlock_bh(); | |
849 | ||
850 | if (!vlan) | |
20d29d7a | 851 | return -ENOLINK; |
20d29d7a | 852 | |
02df55d2 | 853 | ret = 0; |
13707f9e | 854 | if (copy_to_user(&ifr->ifr_name, vlan->dev->name, IFNAMSIZ) || |
b9fb9ee0 | 855 | put_user(q->flags, &ifr->ifr_flags)) |
02df55d2 AB |
856 | ret = -EFAULT; |
857 | dev_put(vlan->dev); | |
858 | return ret; | |
20d29d7a AB |
859 | |
860 | case TUNGETFEATURES: | |
b9fb9ee0 | 861 | if (put_user(IFF_TAP | IFF_NO_PI | IFF_VNET_HDR, up)) |
20d29d7a AB |
862 | return -EFAULT; |
863 | return 0; | |
864 | ||
865 | case TUNSETSNDBUF: | |
866 | if (get_user(u, up)) | |
867 | return -EFAULT; | |
868 | ||
20d29d7a | 869 | q->sk.sk_sndbuf = u; |
20d29d7a AB |
870 | return 0; |
871 | ||
55afbd08 MT |
872 | case TUNGETVNETHDRSZ: |
873 | s = q->vnet_hdr_sz; | |
874 | if (put_user(s, sp)) | |
875 | return -EFAULT; | |
876 | return 0; | |
877 | ||
878 | case TUNSETVNETHDRSZ: | |
879 | if (get_user(s, sp)) | |
880 | return -EFAULT; | |
881 | if (s < (int)sizeof(struct virtio_net_hdr)) | |
882 | return -EINVAL; | |
883 | ||
884 | q->vnet_hdr_sz = s; | |
885 | return 0; | |
886 | ||
20d29d7a AB |
887 | case TUNSETOFFLOAD: |
888 | /* let the user check for future flags */ | |
889 | if (arg & ~(TUN_F_CSUM | TUN_F_TSO4 | TUN_F_TSO6 | | |
b9fb9ee0 | 890 | TUN_F_TSO_ECN | TUN_F_UFO)) |
20d29d7a AB |
891 | return -EINVAL; |
892 | ||
b9fb9ee0 AB |
893 | /* TODO: only accept frames with the features that |
894 | got enabled for forwarded frames */ | |
895 | if (!(q->flags & IFF_VNET_HDR)) | |
896 | return -EINVAL; | |
20d29d7a AB |
897 | return 0; |
898 | ||
899 | default: | |
900 | return -EINVAL; | |
901 | } | |
902 | } | |
903 | ||
904 | #ifdef CONFIG_COMPAT | |
905 | static long macvtap_compat_ioctl(struct file *file, unsigned int cmd, | |
906 | unsigned long arg) | |
907 | { | |
908 | return macvtap_ioctl(file, cmd, (unsigned long)compat_ptr(arg)); | |
909 | } | |
910 | #endif | |
911 | ||
912 | static const struct file_operations macvtap_fops = { | |
913 | .owner = THIS_MODULE, | |
914 | .open = macvtap_open, | |
915 | .release = macvtap_release, | |
916 | .aio_read = macvtap_aio_read, | |
917 | .aio_write = macvtap_aio_write, | |
918 | .poll = macvtap_poll, | |
919 | .llseek = no_llseek, | |
920 | .unlocked_ioctl = macvtap_ioctl, | |
921 | #ifdef CONFIG_COMPAT | |
922 | .compat_ioctl = macvtap_compat_ioctl, | |
923 | #endif | |
924 | }; | |
925 | ||
501c774c AB |
926 | static int macvtap_sendmsg(struct kiocb *iocb, struct socket *sock, |
927 | struct msghdr *m, size_t total_len) | |
928 | { | |
929 | struct macvtap_queue *q = container_of(sock, struct macvtap_queue, sock); | |
97bc3633 | 930 | return macvtap_get_user(q, m, m->msg_iov, total_len, m->msg_iovlen, |
501c774c AB |
931 | m->msg_flags & MSG_DONTWAIT); |
932 | } | |
933 | ||
934 | static int macvtap_recvmsg(struct kiocb *iocb, struct socket *sock, | |
935 | struct msghdr *m, size_t total_len, | |
936 | int flags) | |
937 | { | |
938 | struct macvtap_queue *q = container_of(sock, struct macvtap_queue, sock); | |
939 | int ret; | |
940 | if (flags & ~(MSG_DONTWAIT|MSG_TRUNC)) | |
941 | return -EINVAL; | |
942 | ret = macvtap_do_read(q, iocb, m->msg_iov, total_len, | |
943 | flags & MSG_DONTWAIT); | |
944 | if (ret > total_len) { | |
945 | m->msg_flags |= MSG_TRUNC; | |
946 | ret = flags & MSG_TRUNC ? ret : total_len; | |
947 | } | |
948 | return ret; | |
949 | } | |
950 | ||
951 | /* Ops structure to mimic raw sockets with tun */ | |
952 | static const struct proto_ops macvtap_socket_ops = { | |
953 | .sendmsg = macvtap_sendmsg, | |
954 | .recvmsg = macvtap_recvmsg, | |
955 | }; | |
956 | ||
957 | /* Get an underlying socket object from tun file. Returns error unless file is | |
958 | * attached to a device. The returned object works like a packet socket, it | |
959 | * can be used for sock_sendmsg/sock_recvmsg. The caller is responsible for | |
960 | * holding a reference to the file for as long as the socket is in use. */ | |
961 | struct socket *macvtap_get_socket(struct file *file) | |
962 | { | |
963 | struct macvtap_queue *q; | |
964 | if (file->f_op != &macvtap_fops) | |
965 | return ERR_PTR(-EINVAL); | |
966 | q = file->private_data; | |
967 | if (!q) | |
968 | return ERR_PTR(-EBADFD); | |
969 | return &q->sock; | |
970 | } | |
971 | EXPORT_SYMBOL_GPL(macvtap_get_socket); | |
972 | ||
20d29d7a AB |
973 | static int macvtap_init(void) |
974 | { | |
975 | int err; | |
976 | ||
977 | err = alloc_chrdev_region(&macvtap_major, 0, | |
978 | MACVTAP_NUM_DEVS, "macvtap"); | |
979 | if (err) | |
980 | goto out1; | |
981 | ||
982 | cdev_init(&macvtap_cdev, &macvtap_fops); | |
983 | err = cdev_add(&macvtap_cdev, macvtap_major, MACVTAP_NUM_DEVS); | |
984 | if (err) | |
985 | goto out2; | |
986 | ||
987 | macvtap_class = class_create(THIS_MODULE, "macvtap"); | |
988 | if (IS_ERR(macvtap_class)) { | |
989 | err = PTR_ERR(macvtap_class); | |
990 | goto out3; | |
991 | } | |
992 | ||
993 | err = macvlan_link_register(&macvtap_link_ops); | |
994 | if (err) | |
995 | goto out4; | |
996 | ||
997 | return 0; | |
998 | ||
999 | out4: | |
1000 | class_unregister(macvtap_class); | |
1001 | out3: | |
1002 | cdev_del(&macvtap_cdev); | |
1003 | out2: | |
1004 | unregister_chrdev_region(macvtap_major, MACVTAP_NUM_DEVS); | |
1005 | out1: | |
1006 | return err; | |
1007 | } | |
1008 | module_init(macvtap_init); | |
1009 | ||
1010 | static void macvtap_exit(void) | |
1011 | { | |
1012 | rtnl_link_unregister(&macvtap_link_ops); | |
1013 | class_unregister(macvtap_class); | |
1014 | cdev_del(&macvtap_cdev); | |
1015 | unregister_chrdev_region(macvtap_major, MACVTAP_NUM_DEVS); | |
1016 | } | |
1017 | module_exit(macvtap_exit); | |
1018 | ||
1019 | MODULE_ALIAS_RTNL_LINK("macvtap"); | |
1020 | MODULE_AUTHOR("Arnd Bergmann <arnd@arndb.de>"); | |
1021 | MODULE_LICENSE("GPL"); |