Commit | Line | Data |
---|---|---|
20d29d7a | 1 | #include <linux/etherdevice.h> |
6fe3faf8 | 2 | #include <linux/if_tap.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> | |
c3edc401 | 11 | #include <linux/sched/signal.h> |
20d29d7a | 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> |
362899b8 | 24 | #include <linux/skb_array.h> |
20d29d7a | 25 | |
635b8c8e | 26 | #define TAP_IFFEATURES (IFF_VNET_HDR | IFF_MULTI_QUEUE) |
01b07fb3 | 27 | |
635b8c8e SG |
28 | #define TAP_VNET_LE 0x80000000 |
29 | #define TAP_VNET_BE 0x40000000 | |
8b8e658b GK |
30 | |
31 | #ifdef CONFIG_TUN_VNET_CROSS_LE | |
635b8c8e | 32 | static inline bool tap_legacy_is_little_endian(struct tap_queue *q) |
8b8e658b | 33 | { |
635b8c8e | 34 | return q->flags & TAP_VNET_BE ? false : |
8b8e658b GK |
35 | virtio_legacy_is_little_endian(); |
36 | } | |
37 | ||
635b8c8e | 38 | static long tap_get_vnet_be(struct tap_queue *q, int __user *sp) |
8b8e658b | 39 | { |
635b8c8e | 40 | int s = !!(q->flags & TAP_VNET_BE); |
8b8e658b GK |
41 | |
42 | if (put_user(s, sp)) | |
43 | return -EFAULT; | |
44 | ||
45 | return 0; | |
46 | } | |
47 | ||
635b8c8e | 48 | static long tap_set_vnet_be(struct tap_queue *q, int __user *sp) |
8b8e658b GK |
49 | { |
50 | int s; | |
51 | ||
52 | if (get_user(s, sp)) | |
53 | return -EFAULT; | |
54 | ||
55 | if (s) | |
635b8c8e | 56 | q->flags |= TAP_VNET_BE; |
8b8e658b | 57 | else |
635b8c8e | 58 | q->flags &= ~TAP_VNET_BE; |
8b8e658b GK |
59 | |
60 | return 0; | |
61 | } | |
62 | #else | |
635b8c8e | 63 | static inline bool tap_legacy_is_little_endian(struct tap_queue *q) |
8b8e658b GK |
64 | { |
65 | return virtio_legacy_is_little_endian(); | |
66 | } | |
67 | ||
635b8c8e | 68 | static long tap_get_vnet_be(struct tap_queue *q, int __user *argp) |
8b8e658b GK |
69 | { |
70 | return -EINVAL; | |
71 | } | |
72 | ||
635b8c8e | 73 | static long tap_set_vnet_be(struct tap_queue *q, int __user *argp) |
8b8e658b GK |
74 | { |
75 | return -EINVAL; | |
76 | } | |
77 | #endif /* CONFIG_TUN_VNET_CROSS_LE */ | |
6ae7feb3 | 78 | |
635b8c8e | 79 | static inline bool tap_is_little_endian(struct tap_queue *q) |
5b11e15f | 80 | { |
635b8c8e SG |
81 | return q->flags & TAP_VNET_LE || |
82 | tap_legacy_is_little_endian(q); | |
5b11e15f | 83 | } |
6ae7feb3 | 84 | |
635b8c8e | 85 | static inline u16 tap16_to_cpu(struct tap_queue *q, __virtio16 val) |
6ae7feb3 | 86 | { |
635b8c8e | 87 | return __virtio16_to_cpu(tap_is_little_endian(q), val); |
6ae7feb3 MT |
88 | } |
89 | ||
635b8c8e | 90 | static inline __virtio16 cpu_to_tap16(struct tap_queue *q, u16 val) |
6ae7feb3 | 91 | { |
635b8c8e | 92 | return __cpu_to_virtio16(tap_is_little_endian(q), val); |
6ae7feb3 MT |
93 | } |
94 | ||
635b8c8e SG |
95 | static struct proto tap_proto = { |
96 | .name = "tap", | |
20d29d7a | 97 | .owner = THIS_MODULE, |
635b8c8e | 98 | .obj_size = sizeof(struct tap_queue), |
20d29d7a AB |
99 | }; |
100 | ||
635b8c8e | 101 | #define TAP_NUM_DEVS (1U << MINORBITS) |
d9f1f61c SG |
102 | |
103 | static LIST_HEAD(major_list); | |
104 | ||
ebc05ba7 | 105 | struct major_info { |
d9f1f61c | 106 | struct rcu_head rcu; |
ebc05ba7 SG |
107 | dev_t major; |
108 | struct idr minor_idr; | |
ffa423fb | 109 | spinlock_t minor_lock; |
ebc05ba7 | 110 | const char *device_name; |
d9f1f61c SG |
111 | struct list_head next; |
112 | }; | |
e09eff7f | 113 | |
97bc3633 | 114 | #define GOODCOPY_LEN 128 |
20d29d7a | 115 | |
635b8c8e | 116 | static const struct proto_ops tap_socket_ops; |
501c774c | 117 | |
2be5c767 | 118 | #define RX_OFFLOADS (NETIF_F_GRO | NETIF_F_LRO) |
f23d538b | 119 | #define TAP_FEATURES (NETIF_F_GSO | NETIF_F_SG | NETIF_F_FRAGLIST) |
a567dd62 | 120 | |
6fe3faf8 | 121 | static struct tap_dev *tap_dev_get_rcu(const struct net_device *dev) |
6acf54f1 VY |
122 | { |
123 | return rcu_dereference(dev->rx_handler_data); | |
124 | } | |
125 | ||
20d29d7a AB |
126 | /* |
127 | * RCU usage: | |
635b8c8e | 128 | * The tap_queue and the macvlan_dev are loosely coupled, the |
02df55d2 | 129 | * pointers from one to the other can only be read while rcu_read_lock |
441ac0fc | 130 | * or rtnl is held. |
20d29d7a | 131 | * |
635b8c8e | 132 | * Both the file and the macvlan_dev hold a reference on the tap_queue |
02df55d2 AB |
133 | * through sock_hold(&q->sk). When the macvlan_dev goes away first, |
134 | * q->vlan becomes inaccessible. When the files gets closed, | |
635b8c8e | 135 | * tap_get_queue() fails. |
20d29d7a | 136 | * |
02df55d2 AB |
137 | * There may still be references to the struct sock inside of the |
138 | * queue from outbound SKBs, but these never reference back to the | |
139 | * file or the dev. The data structure is freed through __sk_free | |
140 | * when both our references and any pending SKBs are gone. | |
20d29d7a | 141 | */ |
20d29d7a | 142 | |
6fe3faf8 | 143 | static int tap_enable_queue(struct tap_dev *tap, struct file *file, |
635b8c8e | 144 | struct tap_queue *q) |
815f236d | 145 | { |
815f236d JW |
146 | int err = -EINVAL; |
147 | ||
441ac0fc | 148 | ASSERT_RTNL(); |
815f236d JW |
149 | |
150 | if (q->enabled) | |
151 | goto out; | |
152 | ||
153 | err = 0; | |
6fe3faf8 SG |
154 | rcu_assign_pointer(tap->taps[tap->numvtaps], q); |
155 | q->queue_index = tap->numvtaps; | |
815f236d JW |
156 | q->enabled = true; |
157 | ||
6fe3faf8 | 158 | tap->numvtaps++; |
815f236d | 159 | out: |
815f236d JW |
160 | return err; |
161 | } | |
162 | ||
40b8fe45 | 163 | /* Requires RTNL */ |
6fe3faf8 | 164 | static int tap_set_queue(struct tap_dev *tap, struct file *file, |
635b8c8e | 165 | struct tap_queue *q) |
20d29d7a | 166 | { |
6fe3faf8 | 167 | if (tap->numqueues == MAX_TAP_QUEUES) |
40b8fe45 | 168 | return -EBUSY; |
20d29d7a | 169 | |
6fe3faf8 SG |
170 | rcu_assign_pointer(q->tap, tap); |
171 | rcu_assign_pointer(tap->taps[tap->numvtaps], q); | |
02df55d2 | 172 | sock_hold(&q->sk); |
20d29d7a AB |
173 | |
174 | q->file = file; | |
6fe3faf8 | 175 | q->queue_index = tap->numvtaps; |
815f236d | 176 | q->enabled = true; |
02df55d2 | 177 | file->private_data = q; |
6fe3faf8 | 178 | list_add_tail(&q->next, &tap->queue_list); |
20d29d7a | 179 | |
6fe3faf8 SG |
180 | tap->numvtaps++; |
181 | tap->numqueues++; | |
1565c7c1 | 182 | |
40b8fe45 | 183 | return 0; |
20d29d7a AB |
184 | } |
185 | ||
635b8c8e | 186 | static int tap_disable_queue(struct tap_queue *q) |
815f236d | 187 | { |
6fe3faf8 | 188 | struct tap_dev *tap; |
635b8c8e | 189 | struct tap_queue *nq; |
815f236d | 190 | |
441ac0fc | 191 | ASSERT_RTNL(); |
815f236d JW |
192 | if (!q->enabled) |
193 | return -EINVAL; | |
194 | ||
6fe3faf8 | 195 | tap = rtnl_dereference(q->tap); |
441ac0fc | 196 | |
6fe3faf8 | 197 | if (tap) { |
815f236d | 198 | int index = q->queue_index; |
6fe3faf8 SG |
199 | BUG_ON(index >= tap->numvtaps); |
200 | nq = rtnl_dereference(tap->taps[tap->numvtaps - 1]); | |
815f236d JW |
201 | nq->queue_index = index; |
202 | ||
6fe3faf8 SG |
203 | rcu_assign_pointer(tap->taps[index], nq); |
204 | RCU_INIT_POINTER(tap->taps[tap->numvtaps - 1], NULL); | |
815f236d JW |
205 | q->enabled = false; |
206 | ||
6fe3faf8 | 207 | tap->numvtaps--; |
815f236d JW |
208 | } |
209 | ||
210 | return 0; | |
211 | } | |
212 | ||
20d29d7a | 213 | /* |
02df55d2 AB |
214 | * The file owning the queue got closed, give up both |
215 | * the reference that the files holds as well as the | |
216 | * one from the macvlan_dev if that still exists. | |
20d29d7a AB |
217 | * |
218 | * Using the spinlock makes sure that we don't get | |
219 | * to the queue again after destroying it. | |
20d29d7a | 220 | */ |
635b8c8e | 221 | static void tap_put_queue(struct tap_queue *q) |
20d29d7a | 222 | { |
6fe3faf8 | 223 | struct tap_dev *tap; |
20d29d7a | 224 | |
441ac0fc | 225 | rtnl_lock(); |
6fe3faf8 | 226 | tap = rtnl_dereference(q->tap); |
441ac0fc | 227 | |
6fe3faf8 | 228 | if (tap) { |
815f236d | 229 | if (q->enabled) |
635b8c8e | 230 | BUG_ON(tap_disable_queue(q)); |
376b1aab | 231 | |
6fe3faf8 SG |
232 | tap->numqueues--; |
233 | RCU_INIT_POINTER(q->tap, NULL); | |
02df55d2 | 234 | sock_put(&q->sk); |
815f236d | 235 | list_del_init(&q->next); |
20d29d7a AB |
236 | } |
237 | ||
441ac0fc | 238 | rtnl_unlock(); |
20d29d7a AB |
239 | |
240 | synchronize_rcu(); | |
241 | sock_put(&q->sk); | |
242 | } | |
243 | ||
244 | /* | |
1565c7c1 KK |
245 | * Select a queue based on the rxq of the device on which this packet |
246 | * arrived. If the incoming device is not mq, calculate a flow hash | |
247 | * to select a queue. If all fails, find the first available queue. | |
248 | * Cache vlan->numvtaps since it can become zero during the execution | |
249 | * of this function. | |
20d29d7a | 250 | */ |
6fe3faf8 | 251 | static struct tap_queue *tap_get_queue(struct tap_dev *tap, |
635b8c8e | 252 | struct sk_buff *skb) |
20d29d7a | 253 | { |
6fe3faf8 | 254 | struct tap_queue *queue = NULL; |
815f236d JW |
255 | /* Access to taps array is protected by rcu, but access to numvtaps |
256 | * isn't. Below we use it to lookup a queue, but treat it as a hint | |
257 | * and validate that the result isn't NULL - in case we are | |
258 | * racing against queue removal. | |
259 | */ | |
6aa7de05 | 260 | int numvtaps = READ_ONCE(tap->numvtaps); |
1565c7c1 KK |
261 | __u32 rxq; |
262 | ||
263 | if (!numvtaps) | |
264 | goto out; | |
265 | ||
1b16bf42 JW |
266 | if (numvtaps == 1) |
267 | goto single; | |
268 | ||
ef0002b5 | 269 | /* Check if we can use flow to select a queue */ |
3958afa1 | 270 | rxq = skb_get_hash(skb); |
ef0002b5 | 271 | if (rxq) { |
6fe3faf8 | 272 | queue = rcu_dereference(tap->taps[rxq % numvtaps]); |
376b1aab | 273 | goto out; |
ef0002b5 KK |
274 | } |
275 | ||
1565c7c1 KK |
276 | if (likely(skb_rx_queue_recorded(skb))) { |
277 | rxq = skb_get_rx_queue(skb); | |
20d29d7a | 278 | |
1565c7c1 KK |
279 | while (unlikely(rxq >= numvtaps)) |
280 | rxq -= numvtaps; | |
281 | ||
6fe3faf8 | 282 | queue = rcu_dereference(tap->taps[rxq]); |
376b1aab | 283 | goto out; |
1565c7c1 KK |
284 | } |
285 | ||
1b16bf42 | 286 | single: |
6fe3faf8 | 287 | queue = rcu_dereference(tap->taps[0]); |
1565c7c1 | 288 | out: |
6fe3faf8 | 289 | return queue; |
20d29d7a AB |
290 | } |
291 | ||
02df55d2 AB |
292 | /* |
293 | * The net_device is going away, give up the reference | |
1565c7c1 KK |
294 | * that it holds on all queues and safely set the pointer |
295 | * from the queues to NULL. | |
02df55d2 | 296 | */ |
6fe3faf8 | 297 | void tap_del_queues(struct tap_dev *tap) |
20d29d7a | 298 | { |
635b8c8e | 299 | struct tap_queue *q, *tmp; |
02df55d2 | 300 | |
441ac0fc | 301 | ASSERT_RTNL(); |
6fe3faf8 | 302 | list_for_each_entry_safe(q, tmp, &tap->queue_list, next) { |
815f236d | 303 | list_del_init(&q->next); |
6fe3faf8 | 304 | RCU_INIT_POINTER(q->tap, NULL); |
815f236d | 305 | if (q->enabled) |
6fe3faf8 SG |
306 | tap->numvtaps--; |
307 | tap->numqueues--; | |
dfe816c5 | 308 | sock_put(&q->sk); |
564517e8 | 309 | } |
6fe3faf8 SG |
310 | BUG_ON(tap->numvtaps); |
311 | BUG_ON(tap->numqueues); | |
635b8c8e | 312 | /* guarantee that any future tap_set_queue will fail */ |
6fe3faf8 | 313 | tap->numvtaps = MAX_TAP_QUEUES; |
20d29d7a | 314 | } |
9a393b5d | 315 | EXPORT_SYMBOL_GPL(tap_del_queues); |
20d29d7a | 316 | |
635b8c8e | 317 | rx_handler_result_t tap_handle_frame(struct sk_buff **pskb) |
20d29d7a | 318 | { |
6acf54f1 VY |
319 | struct sk_buff *skb = *pskb; |
320 | struct net_device *dev = skb->dev; | |
6fe3faf8 | 321 | struct tap_dev *tap; |
635b8c8e | 322 | struct tap_queue *q; |
a567dd62 VY |
323 | netdev_features_t features = TAP_FEATURES; |
324 | ||
6fe3faf8 SG |
325 | tap = tap_dev_get_rcu(dev); |
326 | if (!tap) | |
6acf54f1 VY |
327 | return RX_HANDLER_PASS; |
328 | ||
6fe3faf8 | 329 | q = tap_get_queue(tap, skb); |
20d29d7a | 330 | if (!q) |
6acf54f1 | 331 | return RX_HANDLER_PASS; |
8a35747a | 332 | |
6acf54f1 VY |
333 | skb_push(skb, ETH_HLEN); |
334 | ||
3e4f8b78 | 335 | /* Apply the forward feature mask so that we perform segmentation |
e5733321 VY |
336 | * according to users wishes. This only works if VNET_HDR is |
337 | * enabled. | |
3e4f8b78 | 338 | */ |
e5733321 | 339 | if (q->flags & IFF_VNET_HDR) |
6fe3faf8 | 340 | features |= tap->tap_features; |
8b86a61d | 341 | if (netif_needs_gso(skb, features)) { |
3e4f8b78 VY |
342 | struct sk_buff *segs = __skb_gso_segment(skb, features, false); |
343 | ||
344 | if (IS_ERR(segs)) | |
345 | goto drop; | |
346 | ||
347 | if (!segs) { | |
5990a305 | 348 | if (ptr_ring_produce(&q->ring, skb)) |
362899b8 | 349 | goto drop; |
3e4f8b78 VY |
350 | goto wake_up; |
351 | } | |
352 | ||
be0bd316 | 353 | consume_skb(skb); |
3e4f8b78 VY |
354 | while (segs) { |
355 | struct sk_buff *nskb = segs->next; | |
356 | ||
357 | segs->next = NULL; | |
5990a305 | 358 | if (ptr_ring_produce(&q->ring, segs)) { |
362899b8 JW |
359 | kfree_skb(segs); |
360 | kfree_skb_list(nskb); | |
361 | break; | |
362 | } | |
3e4f8b78 VY |
363 | segs = nskb; |
364 | } | |
365 | } else { | |
cbdb0427 VY |
366 | /* If we receive a partial checksum and the tap side |
367 | * doesn't support checksum offload, compute the checksum. | |
368 | * Note: it doesn't matter which checksum feature to | |
a8e04698 | 369 | * check, we either support them all or none. |
cbdb0427 VY |
370 | */ |
371 | if (skb->ip_summed == CHECKSUM_PARTIAL && | |
a188222b | 372 | !(features & NETIF_F_CSUM_MASK) && |
cbdb0427 VY |
373 | skb_checksum_help(skb)) |
374 | goto drop; | |
5990a305 | 375 | if (ptr_ring_produce(&q->ring, skb)) |
362899b8 | 376 | goto drop; |
3e4f8b78 VY |
377 | } |
378 | ||
379 | wake_up: | |
a9a08845 | 380 | wake_up_interruptible_poll(sk_sleep(&q->sk), EPOLLIN | EPOLLRDNORM | EPOLLRDBAND); |
6acf54f1 | 381 | return RX_HANDLER_CONSUMED; |
8a35747a HX |
382 | |
383 | drop: | |
6acf54f1 | 384 | /* Count errors/drops only here, thus don't care about args. */ |
6fe3faf8 SG |
385 | if (tap->count_rx_dropped) |
386 | tap->count_rx_dropped(tap); | |
8a35747a | 387 | kfree_skb(skb); |
6acf54f1 | 388 | return RX_HANDLER_CONSUMED; |
20d29d7a | 389 | } |
9a393b5d | 390 | EXPORT_SYMBOL_GPL(tap_handle_frame); |
20d29d7a | 391 | |
d9f1f61c SG |
392 | static struct major_info *tap_get_major(int major) |
393 | { | |
394 | struct major_info *tap_major; | |
395 | ||
396 | list_for_each_entry_rcu(tap_major, &major_list, next) { | |
397 | if (tap_major->major == major) | |
398 | return tap_major; | |
399 | } | |
400 | ||
401 | return NULL; | |
402 | } | |
403 | ||
404 | int tap_get_minor(dev_t major, struct tap_dev *tap) | |
e09eff7f EB |
405 | { |
406 | int retval = -ENOMEM; | |
d9f1f61c SG |
407 | struct major_info *tap_major; |
408 | ||
409 | rcu_read_lock(); | |
410 | tap_major = tap_get_major(MAJOR(major)); | |
411 | if (!tap_major) { | |
412 | retval = -EINVAL; | |
413 | goto unlock; | |
414 | } | |
e09eff7f | 415 | |
ffa423fb WC |
416 | spin_lock(&tap_major->minor_lock); |
417 | retval = idr_alloc(&tap_major->minor_idr, tap, 1, TAP_NUM_DEVS, GFP_ATOMIC); | |
ec09ebc1 | 418 | if (retval >= 0) { |
6fe3faf8 | 419 | tap->minor = retval; |
ec09ebc1 | 420 | } else if (retval == -ENOSPC) { |
6fe3faf8 | 421 | netdev_err(tap->dev, "Too many tap devices\n"); |
e09eff7f | 422 | retval = -EINVAL; |
e09eff7f | 423 | } |
ffa423fb | 424 | spin_unlock(&tap_major->minor_lock); |
d9f1f61c SG |
425 | |
426 | unlock: | |
427 | rcu_read_unlock(); | |
ec09ebc1 | 428 | return retval < 0 ? retval : 0; |
e09eff7f | 429 | } |
9a393b5d | 430 | EXPORT_SYMBOL_GPL(tap_get_minor); |
e09eff7f | 431 | |
d9f1f61c | 432 | void tap_free_minor(dev_t major, struct tap_dev *tap) |
e09eff7f | 433 | { |
d9f1f61c SG |
434 | struct major_info *tap_major; |
435 | ||
436 | rcu_read_lock(); | |
437 | tap_major = tap_get_major(MAJOR(major)); | |
438 | if (!tap_major) { | |
439 | goto unlock; | |
440 | } | |
441 | ||
ffa423fb | 442 | spin_lock(&tap_major->minor_lock); |
6fe3faf8 | 443 | if (tap->minor) { |
d9f1f61c | 444 | idr_remove(&tap_major->minor_idr, tap->minor); |
6fe3faf8 | 445 | tap->minor = 0; |
e09eff7f | 446 | } |
ffa423fb | 447 | spin_unlock(&tap_major->minor_lock); |
d9f1f61c SG |
448 | |
449 | unlock: | |
450 | rcu_read_unlock(); | |
e09eff7f | 451 | } |
9a393b5d | 452 | EXPORT_SYMBOL_GPL(tap_free_minor); |
e09eff7f | 453 | |
d9f1f61c | 454 | static struct tap_dev *dev_get_by_tap_file(int major, int minor) |
e09eff7f EB |
455 | { |
456 | struct net_device *dev = NULL; | |
6fe3faf8 | 457 | struct tap_dev *tap; |
d9f1f61c | 458 | struct major_info *tap_major; |
e09eff7f | 459 | |
d9f1f61c SG |
460 | rcu_read_lock(); |
461 | tap_major = tap_get_major(major); | |
462 | if (!tap_major) { | |
463 | tap = NULL; | |
464 | goto unlock; | |
465 | } | |
466 | ||
ffa423fb | 467 | spin_lock(&tap_major->minor_lock); |
d9f1f61c | 468 | tap = idr_find(&tap_major->minor_idr, minor); |
6fe3faf8 SG |
469 | if (tap) { |
470 | dev = tap->dev; | |
e09eff7f EB |
471 | dev_hold(dev); |
472 | } | |
ffa423fb | 473 | spin_unlock(&tap_major->minor_lock); |
d9f1f61c SG |
474 | |
475 | unlock: | |
476 | rcu_read_unlock(); | |
6fe3faf8 | 477 | return tap; |
e09eff7f EB |
478 | } |
479 | ||
635b8c8e | 480 | static void tap_sock_write_space(struct sock *sk) |
20d29d7a | 481 | { |
43815482 ED |
482 | wait_queue_head_t *wqueue; |
483 | ||
20d29d7a | 484 | if (!sock_writeable(sk) || |
9cd3e072 | 485 | !test_and_clear_bit(SOCKWQ_ASYNC_NOSPACE, &sk->sk_socket->flags)) |
20d29d7a AB |
486 | return; |
487 | ||
43815482 ED |
488 | wqueue = sk_sleep(sk); |
489 | if (wqueue && waitqueue_active(wqueue)) | |
a9a08845 | 490 | wake_up_interruptible_poll(wqueue, EPOLLOUT | EPOLLWRNORM | EPOLLWRBAND); |
20d29d7a AB |
491 | } |
492 | ||
635b8c8e | 493 | static void tap_sock_destruct(struct sock *sk) |
2259fef0 | 494 | { |
635b8c8e | 495 | struct tap_queue *q = container_of(sk, struct tap_queue, sk); |
362899b8 | 496 | |
5990a305 | 497 | ptr_ring_cleanup(&q->ring, __skb_array_destroy_skb); |
2259fef0 EB |
498 | } |
499 | ||
635b8c8e | 500 | static int tap_open(struct inode *inode, struct file *file) |
20d29d7a AB |
501 | { |
502 | struct net *net = current->nsproxy->net_ns; | |
6fe3faf8 | 503 | struct tap_dev *tap; |
635b8c8e | 504 | struct tap_queue *q; |
40b8fe45 | 505 | int err = -ENODEV; |
20d29d7a | 506 | |
40b8fe45 | 507 | rtnl_lock(); |
d9f1f61c | 508 | tap = dev_get_by_tap_file(imajor(inode), iminor(inode)); |
6fe3faf8 | 509 | if (!tap) |
362899b8 | 510 | goto err; |
20d29d7a | 511 | |
20d29d7a | 512 | err = -ENOMEM; |
635b8c8e SG |
513 | q = (struct tap_queue *)sk_alloc(net, AF_UNSPEC, GFP_KERNEL, |
514 | &tap_proto, 0); | |
20d29d7a | 515 | if (!q) |
362899b8 | 516 | goto err; |
5990a305 | 517 | if (ptr_ring_init(&q->ring, tap->dev->tx_queue_len, GFP_KERNEL)) { |
78e0ea67 GM |
518 | sk_free(&q->sk); |
519 | goto err; | |
520 | } | |
20d29d7a | 521 | |
d9a90a31 | 522 | RCU_INIT_POINTER(q->sock.wq, &q->wq); |
43815482 | 523 | init_waitqueue_head(&q->wq.wait); |
20d29d7a AB |
524 | q->sock.type = SOCK_RAW; |
525 | q->sock.state = SS_CONNECTED; | |
501c774c | 526 | q->sock.file = file; |
635b8c8e | 527 | q->sock.ops = &tap_socket_ops; |
20d29d7a | 528 | sock_init_data(&q->sock, &q->sk); |
635b8c8e SG |
529 | q->sk.sk_write_space = tap_sock_write_space; |
530 | q->sk.sk_destruct = tap_sock_destruct; | |
b9fb9ee0 | 531 | q->flags = IFF_VNET_HDR | IFF_NO_PI | IFF_TAP; |
55afbd08 | 532 | q->vnet_hdr_sz = sizeof(struct virtio_net_hdr); |
20d29d7a | 533 | |
97bc3633 | 534 | /* |
635b8c8e | 535 | * so far only KVM virtio_net uses tap, enable zero copy between |
97bc3633 | 536 | * guest kernel and host kernel when lower device supports zerocopy |
047af9cf EB |
537 | * |
538 | * The macvlan supports zerocopy iff the lower device supports zero | |
539 | * copy so we don't have to look at the lower device directly. | |
97bc3633 | 540 | */ |
6fe3faf8 | 541 | if ((tap->dev->features & NETIF_F_HIGHDMA) && (tap->dev->features & NETIF_F_SG)) |
047af9cf | 542 | sock_set_flag(&q->sk, SOCK_ZEROCOPY); |
97bc3633 | 543 | |
6fe3faf8 | 544 | err = tap_set_queue(tap, file, q); |
78e0ea67 | 545 | if (err) { |
5990a305 | 546 | /* tap_sock_destruct() will take care of freeing ptr_ring */ |
78e0ea67 GM |
547 | goto err_put; |
548 | } | |
20d29d7a | 549 | |
6fe3faf8 | 550 | dev_put(tap->dev); |
362899b8 JW |
551 | |
552 | rtnl_unlock(); | |
553 | return err; | |
554 | ||
78e0ea67 | 555 | err_put: |
362899b8 JW |
556 | sock_put(&q->sk); |
557 | err: | |
6fe3faf8 SG |
558 | if (tap) |
559 | dev_put(tap->dev); | |
20d29d7a | 560 | |
40b8fe45 | 561 | rtnl_unlock(); |
20d29d7a AB |
562 | return err; |
563 | } | |
564 | ||
635b8c8e | 565 | static int tap_release(struct inode *inode, struct file *file) |
20d29d7a | 566 | { |
635b8c8e SG |
567 | struct tap_queue *q = file->private_data; |
568 | tap_put_queue(q); | |
20d29d7a AB |
569 | return 0; |
570 | } | |
571 | ||
afc9a42b | 572 | static __poll_t tap_poll(struct file *file, poll_table *wait) |
20d29d7a | 573 | { |
635b8c8e | 574 | struct tap_queue *q = file->private_data; |
a9a08845 | 575 | __poll_t mask = EPOLLERR; |
20d29d7a AB |
576 | |
577 | if (!q) | |
578 | goto out; | |
579 | ||
580 | mask = 0; | |
43815482 | 581 | poll_wait(file, &q->wq.wait, wait); |
20d29d7a | 582 | |
5990a305 | 583 | if (!ptr_ring_empty(&q->ring)) |
a9a08845 | 584 | mask |= EPOLLIN | EPOLLRDNORM; |
20d29d7a AB |
585 | |
586 | if (sock_writeable(&q->sk) || | |
9cd3e072 | 587 | (!test_and_set_bit(SOCKWQ_ASYNC_NOSPACE, &q->sock.flags) && |
20d29d7a | 588 | sock_writeable(&q->sk))) |
a9a08845 | 589 | mask |= EPOLLOUT | EPOLLWRNORM; |
20d29d7a AB |
590 | |
591 | out: | |
20d29d7a AB |
592 | return mask; |
593 | } | |
594 | ||
635b8c8e SG |
595 | static inline struct sk_buff *tap_alloc_skb(struct sock *sk, size_t prepad, |
596 | size_t len, size_t linear, | |
b9fb9ee0 AB |
597 | int noblock, int *err) |
598 | { | |
599 | struct sk_buff *skb; | |
600 | ||
601 | /* Under a page? Don't bother with paged skb. */ | |
602 | if (prepad + len < PAGE_SIZE || !linear) | |
603 | linear = len; | |
604 | ||
605 | skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock, | |
28d64271 | 606 | err, 0); |
b9fb9ee0 AB |
607 | if (!skb) |
608 | return NULL; | |
609 | ||
610 | skb_reserve(skb, prepad); | |
611 | skb_put(skb, linear); | |
612 | skb->data_len = len - linear; | |
613 | skb->len += len - linear; | |
614 | ||
615 | return skb; | |
616 | } | |
617 | ||
2f1d8b9e | 618 | /* Neighbour code has some assumptions on HH_DATA_MOD alignment */ |
635b8c8e | 619 | #define TAP_RESERVE HH_DATA_OFF(ETH_HLEN) |
2f1d8b9e | 620 | |
20d29d7a | 621 | /* Get packet from user space buffer */ |
635b8c8e SG |
622 | static ssize_t tap_get_user(struct tap_queue *q, struct msghdr *m, |
623 | struct iov_iter *from, int noblock) | |
20d29d7a | 624 | { |
635b8c8e | 625 | int good_linear = SKB_MAX_HEAD(TAP_RESERVE); |
20d29d7a | 626 | struct sk_buff *skb; |
6fe3faf8 | 627 | struct tap_dev *tap; |
f5ff53b4 | 628 | unsigned long total_len = iov_iter_count(from); |
97bc3633 | 629 | unsigned long len = total_len; |
20d29d7a | 630 | int err; |
b9fb9ee0 AB |
631 | struct virtio_net_hdr vnet_hdr = { 0 }; |
632 | int vnet_hdr_len = 0; | |
b92946e2 | 633 | int copylen = 0; |
c5c62f1b | 634 | int depth; |
97bc3633 | 635 | bool zerocopy = false; |
61d46bf9 | 636 | size_t linear; |
b9fb9ee0 AB |
637 | |
638 | if (q->flags & IFF_VNET_HDR) { | |
837585a5 | 639 | vnet_hdr_len = READ_ONCE(q->vnet_hdr_sz); |
b9fb9ee0 AB |
640 | |
641 | err = -EINVAL; | |
ce3c8692 | 642 | if (len < vnet_hdr_len) |
b9fb9ee0 | 643 | goto err; |
ce3c8692 | 644 | len -= vnet_hdr_len; |
b9fb9ee0 | 645 | |
f5ff53b4 | 646 | err = -EFAULT; |
cbbd26b8 | 647 | if (!copy_from_iter_full(&vnet_hdr, sizeof(vnet_hdr), from)) |
b9fb9ee0 | 648 | goto err; |
f5ff53b4 | 649 | iov_iter_advance(from, vnet_hdr_len - sizeof(vnet_hdr)); |
b9fb9ee0 | 650 | if ((vnet_hdr.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) && |
635b8c8e SG |
651 | tap16_to_cpu(q, vnet_hdr.csum_start) + |
652 | tap16_to_cpu(q, vnet_hdr.csum_offset) + 2 > | |
653 | tap16_to_cpu(q, vnet_hdr.hdr_len)) | |
654 | vnet_hdr.hdr_len = cpu_to_tap16(q, | |
655 | tap16_to_cpu(q, vnet_hdr.csum_start) + | |
656 | tap16_to_cpu(q, vnet_hdr.csum_offset) + 2); | |
b9fb9ee0 | 657 | err = -EINVAL; |
635b8c8e | 658 | if (tap16_to_cpu(q, vnet_hdr.hdr_len) > len) |
b9fb9ee0 AB |
659 | goto err; |
660 | } | |
20d29d7a | 661 | |
b9fb9ee0 | 662 | err = -EINVAL; |
20d29d7a | 663 | if (unlikely(len < ETH_HLEN)) |
b9fb9ee0 | 664 | goto err; |
20d29d7a | 665 | |
ece793fc | 666 | if (m && m->msg_control && sock_flag(&q->sk, SOCK_ZEROCOPY)) { |
f5ff53b4 AV |
667 | struct iov_iter i; |
668 | ||
6ae7feb3 | 669 | copylen = vnet_hdr.hdr_len ? |
635b8c8e | 670 | tap16_to_cpu(q, vnet_hdr.hdr_len) : GOODCOPY_LEN; |
16a3fa28 JW |
671 | if (copylen > good_linear) |
672 | copylen = good_linear; | |
8e2ad411 WB |
673 | else if (copylen < ETH_HLEN) |
674 | copylen = ETH_HLEN; | |
61d46bf9 | 675 | linear = copylen; |
f5ff53b4 AV |
676 | i = *from; |
677 | iov_iter_advance(&i, copylen); | |
678 | if (iov_iter_npages(&i, INT_MAX) <= MAX_SKB_FRAGS) | |
ece793fc JW |
679 | zerocopy = true; |
680 | } | |
681 | ||
682 | if (!zerocopy) { | |
97bc3633 | 683 | copylen = len; |
635b8c8e | 684 | linear = tap16_to_cpu(q, vnet_hdr.hdr_len); |
8e2ad411 | 685 | if (linear > good_linear) |
16a3fa28 | 686 | linear = good_linear; |
8e2ad411 WB |
687 | else if (linear < ETH_HLEN) |
688 | linear = ETH_HLEN; | |
61d46bf9 | 689 | } |
97bc3633 | 690 | |
635b8c8e SG |
691 | skb = tap_alloc_skb(&q->sk, TAP_RESERVE, copylen, |
692 | linear, noblock, &err); | |
02df55d2 AB |
693 | if (!skb) |
694 | goto err; | |
20d29d7a | 695 | |
01d6657b | 696 | if (zerocopy) |
f5ff53b4 | 697 | err = zerocopy_sg_from_iter(skb, from); |
aa196eed | 698 | else |
f5ff53b4 | 699 | err = skb_copy_datagram_from_iter(skb, 0, from, len); |
ece793fc | 700 | |
02df55d2 | 701 | if (err) |
b9fb9ee0 | 702 | goto err_kfree; |
20d29d7a AB |
703 | |
704 | skb_set_network_header(skb, ETH_HLEN); | |
b9fb9ee0 AB |
705 | skb_reset_mac_header(skb); |
706 | skb->protocol = eth_hdr(skb)->h_proto; | |
707 | ||
708 | if (vnet_hdr_len) { | |
fd88d68b | 709 | err = virtio_net_hdr_to_skb(skb, &vnet_hdr, |
635b8c8e | 710 | tap_is_little_endian(q)); |
b9fb9ee0 AB |
711 | if (err) |
712 | goto err_kfree; | |
713 | } | |
714 | ||
40893fd0 | 715 | skb_probe_transport_header(skb, ETH_HLEN); |
9b4d669b | 716 | |
c5c62f1b IV |
717 | /* Move network header to the right position for VLAN tagged packets */ |
718 | if ((skb->protocol == htons(ETH_P_8021Q) || | |
719 | skb->protocol == htons(ETH_P_8021AD)) && | |
720 | __vlan_get_protocol(skb, skb->protocol, &depth) != 0) | |
721 | skb_set_network_header(skb, depth); | |
722 | ||
ac4e4af1 | 723 | rcu_read_lock(); |
6fe3faf8 | 724 | tap = rcu_dereference(q->tap); |
97bc3633 | 725 | /* copy skb_ubuf_info for callback when skb has no error */ |
01d6657b | 726 | if (zerocopy) { |
97bc3633 | 727 | skb_shinfo(skb)->destructor_arg = m->msg_control; |
01d6657b | 728 | skb_shinfo(skb)->tx_flags |= SKBTX_DEV_ZEROCOPY; |
c9af6db4 | 729 | skb_shinfo(skb)->tx_flags |= SKBTX_SHARED_FRAG; |
aa196eed JW |
730 | } else if (m && m->msg_control) { |
731 | struct ubuf_info *uarg = m->msg_control; | |
732 | uarg->callback(uarg, false); | |
01d6657b | 733 | } |
aa196eed | 734 | |
6fe3faf8 SG |
735 | if (tap) { |
736 | skb->dev = tap->dev; | |
6acf54f1 | 737 | dev_queue_xmit(skb); |
29d79196 | 738 | } else { |
02df55d2 | 739 | kfree_skb(skb); |
29d79196 | 740 | } |
ac4e4af1 | 741 | rcu_read_unlock(); |
20d29d7a | 742 | |
97bc3633 | 743 | return total_len; |
02df55d2 | 744 | |
b9fb9ee0 AB |
745 | err_kfree: |
746 | kfree_skb(skb); | |
747 | ||
02df55d2 | 748 | err: |
ac4e4af1 | 749 | rcu_read_lock(); |
6fe3faf8 SG |
750 | tap = rcu_dereference(q->tap); |
751 | if (tap && tap->count_tx_dropped) | |
752 | tap->count_tx_dropped(tap); | |
ac4e4af1 | 753 | rcu_read_unlock(); |
02df55d2 | 754 | |
02df55d2 | 755 | return err; |
20d29d7a AB |
756 | } |
757 | ||
635b8c8e | 758 | static ssize_t tap_write_iter(struct kiocb *iocb, struct iov_iter *from) |
20d29d7a AB |
759 | { |
760 | struct file *file = iocb->ki_filp; | |
635b8c8e | 761 | struct tap_queue *q = file->private_data; |
20d29d7a | 762 | |
635b8c8e | 763 | return tap_get_user(q, NULL, from, file->f_flags & O_NONBLOCK); |
20d29d7a AB |
764 | } |
765 | ||
766 | /* Put packet to the user space buffer */ | |
635b8c8e SG |
767 | static ssize_t tap_put_user(struct tap_queue *q, |
768 | const struct sk_buff *skb, | |
769 | struct iov_iter *iter) | |
20d29d7a | 770 | { |
20d29d7a | 771 | int ret; |
b9fb9ee0 | 772 | int vnet_hdr_len = 0; |
f09e2249 | 773 | int vlan_offset = 0; |
6c36d2e2 | 774 | int total; |
b9fb9ee0 AB |
775 | |
776 | if (q->flags & IFF_VNET_HDR) { | |
fd3a8862 | 777 | int vlan_hlen = skb_vlan_tag_present(skb) ? VLAN_HLEN : 0; |
b9fb9ee0 | 778 | struct virtio_net_hdr vnet_hdr; |
fd3a8862 | 779 | |
837585a5 | 780 | vnet_hdr_len = READ_ONCE(q->vnet_hdr_sz); |
6c36d2e2 | 781 | if (iov_iter_count(iter) < vnet_hdr_len) |
b9fb9ee0 AB |
782 | return -EINVAL; |
783 | ||
3e9e40e7 | 784 | if (virtio_net_hdr_from_skb(skb, &vnet_hdr, |
fd3a8862 WB |
785 | tap_is_little_endian(q), true, |
786 | vlan_hlen)) | |
fd88d68b | 787 | BUG(); |
b9fb9ee0 | 788 | |
6c36d2e2 HX |
789 | if (copy_to_iter(&vnet_hdr, sizeof(vnet_hdr), iter) != |
790 | sizeof(vnet_hdr)) | |
b9fb9ee0 | 791 | return -EFAULT; |
7cc76f51 JW |
792 | |
793 | iov_iter_advance(iter, vnet_hdr_len - sizeof(vnet_hdr)); | |
b9fb9ee0 | 794 | } |
6c36d2e2 | 795 | total = vnet_hdr_len; |
ce232ce0 | 796 | total += skb->len; |
f09e2249 | 797 | |
df8a39de | 798 | if (skb_vlan_tag_present(skb)) { |
f09e2249 BG |
799 | struct { |
800 | __be16 h_vlan_proto; | |
801 | __be16 h_vlan_TCI; | |
802 | } veth; | |
0fbe0d47 | 803 | veth.h_vlan_proto = skb->vlan_proto; |
df8a39de | 804 | veth.h_vlan_TCI = htons(skb_vlan_tag_get(skb)); |
f09e2249 BG |
805 | |
806 | vlan_offset = offsetof(struct vlan_ethhdr, h_vlan_proto); | |
ce232ce0 | 807 | total += VLAN_HLEN; |
f09e2249 | 808 | |
6c36d2e2 HX |
809 | ret = skb_copy_datagram_iter(skb, 0, iter, vlan_offset); |
810 | if (ret || !iov_iter_count(iter)) | |
f09e2249 BG |
811 | goto done; |
812 | ||
6c36d2e2 HX |
813 | ret = copy_to_iter(&veth, sizeof(veth), iter); |
814 | if (ret != sizeof(veth) || !iov_iter_count(iter)) | |
f09e2249 BG |
815 | goto done; |
816 | } | |
20d29d7a | 817 | |
6c36d2e2 HX |
818 | ret = skb_copy_datagram_iter(skb, vlan_offset, iter, |
819 | skb->len - vlan_offset); | |
20d29d7a | 820 | |
f09e2249 | 821 | done: |
ce232ce0 | 822 | return ret ? ret : total; |
20d29d7a AB |
823 | } |
824 | ||
635b8c8e SG |
825 | static ssize_t tap_do_read(struct tap_queue *q, |
826 | struct iov_iter *to, | |
3b4ba04a | 827 | int noblock, struct sk_buff *skb) |
20d29d7a | 828 | { |
ccf7e72b | 829 | DEFINE_WAIT(wait); |
501c774c | 830 | ssize_t ret = 0; |
20d29d7a | 831 | |
61d78537 WX |
832 | if (!iov_iter_count(to)) { |
833 | if (skb) | |
834 | kfree_skb(skb); | |
3af0bfe5 | 835 | return 0; |
61d78537 | 836 | } |
3af0bfe5 | 837 | |
3b4ba04a JW |
838 | if (skb) |
839 | goto put; | |
840 | ||
3af0bfe5 | 841 | while (1) { |
89cee917 JW |
842 | if (!noblock) |
843 | prepare_to_wait(sk_sleep(&q->sk), &wait, | |
844 | TASK_INTERRUPTIBLE); | |
20d29d7a AB |
845 | |
846 | /* Read frames from the queue */ | |
5990a305 | 847 | skb = ptr_ring_consume(&q->ring); |
3af0bfe5 AV |
848 | if (skb) |
849 | break; | |
850 | if (noblock) { | |
851 | ret = -EAGAIN; | |
852 | break; | |
20d29d7a | 853 | } |
3af0bfe5 AV |
854 | if (signal_pending(current)) { |
855 | ret = -ERESTARTSYS; | |
856 | break; | |
857 | } | |
858 | /* Nothing to read, let's sleep */ | |
859 | schedule(); | |
860 | } | |
a499a2e9 VY |
861 | if (!noblock) |
862 | finish_wait(sk_sleep(&q->sk), &wait); | |
863 | ||
3b4ba04a | 864 | put: |
3af0bfe5 | 865 | if (skb) { |
635b8c8e | 866 | ret = tap_put_user(q, skb, to); |
f51a5e82 JW |
867 | if (unlikely(ret < 0)) |
868 | kfree_skb(skb); | |
869 | else | |
870 | consume_skb(skb); | |
20d29d7a | 871 | } |
501c774c AB |
872 | return ret; |
873 | } | |
874 | ||
635b8c8e | 875 | static ssize_t tap_read_iter(struct kiocb *iocb, struct iov_iter *to) |
501c774c AB |
876 | { |
877 | struct file *file = iocb->ki_filp; | |
635b8c8e | 878 | struct tap_queue *q = file->private_data; |
3af0bfe5 | 879 | ssize_t len = iov_iter_count(to), ret; |
20d29d7a | 880 | |
3b4ba04a | 881 | ret = tap_do_read(q, to, file->f_flags & O_NONBLOCK, NULL); |
ce232ce0 | 882 | ret = min_t(ssize_t, ret, len); |
e6ebc7f1 ZYW |
883 | if (ret > 0) |
884 | iocb->ki_pos = ret; | |
20d29d7a AB |
885 | return ret; |
886 | } | |
887 | ||
6fe3faf8 | 888 | static struct tap_dev *tap_get_tap_dev(struct tap_queue *q) |
8f475a31 | 889 | { |
6fe3faf8 | 890 | struct tap_dev *tap; |
8f475a31 | 891 | |
441ac0fc | 892 | ASSERT_RTNL(); |
6fe3faf8 SG |
893 | tap = rtnl_dereference(q->tap); |
894 | if (tap) | |
895 | dev_hold(tap->dev); | |
8f475a31 | 896 | |
6fe3faf8 | 897 | return tap; |
8f475a31 JW |
898 | } |
899 | ||
6fe3faf8 | 900 | static void tap_put_tap_dev(struct tap_dev *tap) |
8f475a31 | 901 | { |
6fe3faf8 | 902 | dev_put(tap->dev); |
8f475a31 JW |
903 | } |
904 | ||
635b8c8e | 905 | static int tap_ioctl_set_queue(struct file *file, unsigned int flags) |
815f236d | 906 | { |
635b8c8e | 907 | struct tap_queue *q = file->private_data; |
6fe3faf8 | 908 | struct tap_dev *tap; |
815f236d JW |
909 | int ret; |
910 | ||
6fe3faf8 SG |
911 | tap = tap_get_tap_dev(q); |
912 | if (!tap) | |
815f236d JW |
913 | return -EINVAL; |
914 | ||
915 | if (flags & IFF_ATTACH_QUEUE) | |
6fe3faf8 | 916 | ret = tap_enable_queue(tap, file, q); |
815f236d | 917 | else if (flags & IFF_DETACH_QUEUE) |
635b8c8e | 918 | ret = tap_disable_queue(q); |
f57855a5 JW |
919 | else |
920 | ret = -EINVAL; | |
815f236d | 921 | |
6fe3faf8 | 922 | tap_put_tap_dev(tap); |
815f236d JW |
923 | return ret; |
924 | } | |
925 | ||
635b8c8e | 926 | static int set_offload(struct tap_queue *q, unsigned long arg) |
2be5c767 | 927 | { |
6fe3faf8 | 928 | struct tap_dev *tap; |
2be5c767 VY |
929 | netdev_features_t features; |
930 | netdev_features_t feature_mask = 0; | |
931 | ||
6fe3faf8 SG |
932 | tap = rtnl_dereference(q->tap); |
933 | if (!tap) | |
2be5c767 VY |
934 | return -ENOLINK; |
935 | ||
6fe3faf8 | 936 | features = tap->dev->features; |
2be5c767 VY |
937 | |
938 | if (arg & TUN_F_CSUM) { | |
939 | feature_mask = NETIF_F_HW_CSUM; | |
940 | ||
941 | if (arg & (TUN_F_TSO4 | TUN_F_TSO6)) { | |
942 | if (arg & TUN_F_TSO_ECN) | |
943 | feature_mask |= NETIF_F_TSO_ECN; | |
944 | if (arg & TUN_F_TSO4) | |
945 | feature_mask |= NETIF_F_TSO; | |
946 | if (arg & TUN_F_TSO6) | |
947 | feature_mask |= NETIF_F_TSO6; | |
948 | } | |
2be5c767 VY |
949 | } |
950 | ||
951 | /* tun/tap driver inverts the usage for TSO offloads, where | |
952 | * setting the TSO bit means that the userspace wants to | |
953 | * accept TSO frames and turning it off means that user space | |
954 | * does not support TSO. | |
635b8c8e | 955 | * For tap, we have to invert it to mean the same thing. |
2be5c767 VY |
956 | * When user space turns off TSO, we turn off GSO/LRO so that |
957 | * user-space will not receive TSO frames. | |
958 | */ | |
d591a1f3 | 959 | if (feature_mask & (NETIF_F_TSO | NETIF_F_TSO6)) |
2be5c767 VY |
960 | features |= RX_OFFLOADS; |
961 | else | |
962 | features &= ~RX_OFFLOADS; | |
963 | ||
964 | /* tap_features are the same as features on tun/tap and | |
965 | * reflect user expectations. | |
966 | */ | |
6fe3faf8 SG |
967 | tap->tap_features = feature_mask; |
968 | if (tap->update_features) | |
969 | tap->update_features(tap, features); | |
2be5c767 VY |
970 | |
971 | return 0; | |
972 | } | |
973 | ||
20d29d7a AB |
974 | /* |
975 | * provide compatibility with generic tun/tap interface | |
976 | */ | |
635b8c8e SG |
977 | static long tap_ioctl(struct file *file, unsigned int cmd, |
978 | unsigned long arg) | |
20d29d7a | 979 | { |
635b8c8e | 980 | struct tap_queue *q = file->private_data; |
6fe3faf8 | 981 | struct tap_dev *tap; |
20d29d7a AB |
982 | void __user *argp = (void __user *)arg; |
983 | struct ifreq __user *ifr = argp; | |
984 | unsigned int __user *up = argp; | |
39ec7de7 | 985 | unsigned short u; |
55afbd08 | 986 | int __user *sp = argp; |
7f460d30 | 987 | struct sockaddr sa; |
55afbd08 | 988 | int s; |
02df55d2 | 989 | int ret; |
20d29d7a AB |
990 | |
991 | switch (cmd) { | |
992 | case TUNSETIFF: | |
993 | /* ignore the name, just look at flags */ | |
994 | if (get_user(u, &ifr->ifr_flags)) | |
995 | return -EFAULT; | |
b9fb9ee0 AB |
996 | |
997 | ret = 0; | |
635b8c8e | 998 | if ((u & ~TAP_IFFEATURES) != (IFF_NO_PI | IFF_TAP)) |
b9fb9ee0 AB |
999 | ret = -EINVAL; |
1000 | else | |
635b8c8e | 1001 | q->flags = (q->flags & ~TAP_IFFEATURES) | u; |
b9fb9ee0 AB |
1002 | |
1003 | return ret; | |
20d29d7a AB |
1004 | |
1005 | case TUNGETIFF: | |
441ac0fc | 1006 | rtnl_lock(); |
6fe3faf8 SG |
1007 | tap = tap_get_tap_dev(q); |
1008 | if (!tap) { | |
441ac0fc | 1009 | rtnl_unlock(); |
20d29d7a | 1010 | return -ENOLINK; |
441ac0fc | 1011 | } |
20d29d7a | 1012 | |
02df55d2 | 1013 | ret = 0; |
39ec7de7 | 1014 | u = q->flags; |
6fe3faf8 | 1015 | if (copy_to_user(&ifr->ifr_name, tap->dev->name, IFNAMSIZ) || |
39ec7de7 | 1016 | put_user(u, &ifr->ifr_flags)) |
02df55d2 | 1017 | ret = -EFAULT; |
6fe3faf8 | 1018 | tap_put_tap_dev(tap); |
441ac0fc | 1019 | rtnl_unlock(); |
02df55d2 | 1020 | return ret; |
20d29d7a | 1021 | |
815f236d JW |
1022 | case TUNSETQUEUE: |
1023 | if (get_user(u, &ifr->ifr_flags)) | |
1024 | return -EFAULT; | |
441ac0fc | 1025 | rtnl_lock(); |
635b8c8e | 1026 | ret = tap_ioctl_set_queue(file, u); |
441ac0fc | 1027 | rtnl_unlock(); |
82a19eb8 | 1028 | return ret; |
815f236d | 1029 | |
20d29d7a | 1030 | case TUNGETFEATURES: |
635b8c8e | 1031 | if (put_user(IFF_TAP | IFF_NO_PI | TAP_IFFEATURES, up)) |
20d29d7a AB |
1032 | return -EFAULT; |
1033 | return 0; | |
1034 | ||
1035 | case TUNSETSNDBUF: | |
3ea79249 | 1036 | if (get_user(s, sp)) |
20d29d7a | 1037 | return -EFAULT; |
93161922 CG |
1038 | if (s <= 0) |
1039 | return -EINVAL; | |
20d29d7a | 1040 | |
3ea79249 | 1041 | q->sk.sk_sndbuf = s; |
20d29d7a AB |
1042 | return 0; |
1043 | ||
55afbd08 MT |
1044 | case TUNGETVNETHDRSZ: |
1045 | s = q->vnet_hdr_sz; | |
1046 | if (put_user(s, sp)) | |
1047 | return -EFAULT; | |
1048 | return 0; | |
1049 | ||
1050 | case TUNSETVNETHDRSZ: | |
1051 | if (get_user(s, sp)) | |
1052 | return -EFAULT; | |
1053 | if (s < (int)sizeof(struct virtio_net_hdr)) | |
1054 | return -EINVAL; | |
1055 | ||
1056 | q->vnet_hdr_sz = s; | |
1057 | return 0; | |
1058 | ||
01b07fb3 | 1059 | case TUNGETVNETLE: |
635b8c8e | 1060 | s = !!(q->flags & TAP_VNET_LE); |
01b07fb3 MT |
1061 | if (put_user(s, sp)) |
1062 | return -EFAULT; | |
1063 | return 0; | |
1064 | ||
1065 | case TUNSETVNETLE: | |
1066 | if (get_user(s, sp)) | |
1067 | return -EFAULT; | |
1068 | if (s) | |
635b8c8e | 1069 | q->flags |= TAP_VNET_LE; |
01b07fb3 | 1070 | else |
635b8c8e | 1071 | q->flags &= ~TAP_VNET_LE; |
01b07fb3 MT |
1072 | return 0; |
1073 | ||
8b8e658b | 1074 | case TUNGETVNETBE: |
635b8c8e | 1075 | return tap_get_vnet_be(q, sp); |
8b8e658b GK |
1076 | |
1077 | case TUNSETVNETBE: | |
635b8c8e | 1078 | return tap_set_vnet_be(q, sp); |
8b8e658b | 1079 | |
20d29d7a AB |
1080 | case TUNSETOFFLOAD: |
1081 | /* let the user check for future flags */ | |
1082 | if (arg & ~(TUN_F_CSUM | TUN_F_TSO4 | TUN_F_TSO6 | | |
0c19f846 | 1083 | TUN_F_TSO_ECN | TUN_F_UFO)) |
20d29d7a AB |
1084 | return -EINVAL; |
1085 | ||
2be5c767 VY |
1086 | rtnl_lock(); |
1087 | ret = set_offload(q, arg); | |
1088 | rtnl_unlock(); | |
1089 | return ret; | |
20d29d7a | 1090 | |
b5082083 JC |
1091 | case SIOCGIFHWADDR: |
1092 | rtnl_lock(); | |
6fe3faf8 SG |
1093 | tap = tap_get_tap_dev(q); |
1094 | if (!tap) { | |
b5082083 JC |
1095 | rtnl_unlock(); |
1096 | return -ENOLINK; | |
1097 | } | |
1098 | ret = 0; | |
6fe3faf8 SG |
1099 | u = tap->dev->type; |
1100 | if (copy_to_user(&ifr->ifr_name, tap->dev->name, IFNAMSIZ) || | |
1101 | copy_to_user(&ifr->ifr_hwaddr.sa_data, tap->dev->dev_addr, ETH_ALEN) || | |
b5082083 JC |
1102 | put_user(u, &ifr->ifr_hwaddr.sa_family)) |
1103 | ret = -EFAULT; | |
6fe3faf8 | 1104 | tap_put_tap_dev(tap); |
b5082083 JC |
1105 | rtnl_unlock(); |
1106 | return ret; | |
1107 | ||
1108 | case SIOCSIFHWADDR: | |
7f460d30 JC |
1109 | if (copy_from_user(&sa, &ifr->ifr_hwaddr, sizeof(sa))) |
1110 | return -EFAULT; | |
b5082083 | 1111 | rtnl_lock(); |
6fe3faf8 SG |
1112 | tap = tap_get_tap_dev(q); |
1113 | if (!tap) { | |
b5082083 JC |
1114 | rtnl_unlock(); |
1115 | return -ENOLINK; | |
1116 | } | |
6fe3faf8 SG |
1117 | ret = dev_set_mac_address(tap->dev, &sa); |
1118 | tap_put_tap_dev(tap); | |
b5082083 JC |
1119 | rtnl_unlock(); |
1120 | return ret; | |
1121 | ||
20d29d7a AB |
1122 | default: |
1123 | return -EINVAL; | |
1124 | } | |
1125 | } | |
1126 | ||
1127 | #ifdef CONFIG_COMPAT | |
635b8c8e SG |
1128 | static long tap_compat_ioctl(struct file *file, unsigned int cmd, |
1129 | unsigned long arg) | |
20d29d7a | 1130 | { |
635b8c8e | 1131 | return tap_ioctl(file, cmd, (unsigned long)compat_ptr(arg)); |
20d29d7a AB |
1132 | } |
1133 | #endif | |
1134 | ||
d17eb73b | 1135 | static const struct file_operations tap_fops = { |
20d29d7a | 1136 | .owner = THIS_MODULE, |
635b8c8e SG |
1137 | .open = tap_open, |
1138 | .release = tap_release, | |
1139 | .read_iter = tap_read_iter, | |
1140 | .write_iter = tap_write_iter, | |
1141 | .poll = tap_poll, | |
20d29d7a | 1142 | .llseek = no_llseek, |
635b8c8e | 1143 | .unlocked_ioctl = tap_ioctl, |
20d29d7a | 1144 | #ifdef CONFIG_COMPAT |
635b8c8e | 1145 | .compat_ioctl = tap_compat_ioctl, |
20d29d7a AB |
1146 | #endif |
1147 | }; | |
1148 | ||
635b8c8e SG |
1149 | static int tap_sendmsg(struct socket *sock, struct msghdr *m, |
1150 | size_t total_len) | |
501c774c | 1151 | { |
635b8c8e SG |
1152 | struct tap_queue *q = container_of(sock, struct tap_queue, sock); |
1153 | return tap_get_user(q, m, &m->msg_iter, m->msg_flags & MSG_DONTWAIT); | |
501c774c AB |
1154 | } |
1155 | ||
635b8c8e SG |
1156 | static int tap_recvmsg(struct socket *sock, struct msghdr *m, |
1157 | size_t total_len, int flags) | |
501c774c | 1158 | { |
635b8c8e | 1159 | struct tap_queue *q = container_of(sock, struct tap_queue, sock); |
61d78537 | 1160 | struct sk_buff *skb = m->msg_control; |
501c774c | 1161 | int ret; |
61d78537 WX |
1162 | if (flags & ~(MSG_DONTWAIT|MSG_TRUNC)) { |
1163 | if (skb) | |
1164 | kfree_skb(skb); | |
501c774c | 1165 | return -EINVAL; |
61d78537 WX |
1166 | } |
1167 | ret = tap_do_read(q, &m->msg_iter, flags & MSG_DONTWAIT, skb); | |
de2aa476 DM |
1168 | if (ret > total_len) { |
1169 | m->msg_flags |= MSG_TRUNC; | |
1170 | ret = flags & MSG_TRUNC ? ret : total_len; | |
1171 | } | |
501c774c AB |
1172 | return ret; |
1173 | } | |
1174 | ||
635b8c8e | 1175 | static int tap_peek_len(struct socket *sock) |
362899b8 | 1176 | { |
635b8c8e | 1177 | struct tap_queue *q = container_of(sock, struct tap_queue, |
362899b8 | 1178 | sock); |
5990a305 | 1179 | return PTR_RING_PEEK_CALL(&q->ring, __skb_array_len_with_tag); |
362899b8 JW |
1180 | } |
1181 | ||
501c774c | 1182 | /* Ops structure to mimic raw sockets with tun */ |
635b8c8e SG |
1183 | static const struct proto_ops tap_socket_ops = { |
1184 | .sendmsg = tap_sendmsg, | |
1185 | .recvmsg = tap_recvmsg, | |
1186 | .peek_len = tap_peek_len, | |
501c774c AB |
1187 | }; |
1188 | ||
1189 | /* Get an underlying socket object from tun file. Returns error unless file is | |
1190 | * attached to a device. The returned object works like a packet socket, it | |
1191 | * can be used for sock_sendmsg/sock_recvmsg. The caller is responsible for | |
1192 | * holding a reference to the file for as long as the socket is in use. */ | |
635b8c8e | 1193 | struct socket *tap_get_socket(struct file *file) |
501c774c | 1194 | { |
635b8c8e SG |
1195 | struct tap_queue *q; |
1196 | if (file->f_op != &tap_fops) | |
501c774c AB |
1197 | return ERR_PTR(-EINVAL); |
1198 | q = file->private_data; | |
1199 | if (!q) | |
1200 | return ERR_PTR(-EBADFD); | |
1201 | return &q->sock; | |
1202 | } | |
635b8c8e | 1203 | EXPORT_SYMBOL_GPL(tap_get_socket); |
501c774c | 1204 | |
5990a305 | 1205 | struct ptr_ring *tap_get_ptr_ring(struct file *file) |
49f96fd0 JW |
1206 | { |
1207 | struct tap_queue *q; | |
1208 | ||
1209 | if (file->f_op != &tap_fops) | |
1210 | return ERR_PTR(-EINVAL); | |
1211 | q = file->private_data; | |
1212 | if (!q) | |
1213 | return ERR_PTR(-EBADFD); | |
5990a305 | 1214 | return &q->ring; |
49f96fd0 | 1215 | } |
5990a305 | 1216 | EXPORT_SYMBOL_GPL(tap_get_ptr_ring); |
49f96fd0 | 1217 | |
6fe3faf8 | 1218 | int tap_queue_resize(struct tap_dev *tap) |
362899b8 | 1219 | { |
6fe3faf8 | 1220 | struct net_device *dev = tap->dev; |
635b8c8e | 1221 | struct tap_queue *q; |
5990a305 | 1222 | struct ptr_ring **rings; |
6fe3faf8 | 1223 | int n = tap->numqueues; |
362899b8 JW |
1224 | int ret, i = 0; |
1225 | ||
5990a305 JW |
1226 | rings = kmalloc_array(n, sizeof(*rings), GFP_KERNEL); |
1227 | if (!rings) | |
362899b8 JW |
1228 | return -ENOMEM; |
1229 | ||
6fe3faf8 | 1230 | list_for_each_entry(q, &tap->queue_list, next) |
5990a305 | 1231 | rings[i++] = &q->ring; |
362899b8 | 1232 | |
5990a305 JW |
1233 | ret = ptr_ring_resize_multiple(rings, n, |
1234 | dev->tx_queue_len, GFP_KERNEL, | |
1235 | __skb_array_destroy_skb); | |
362899b8 | 1236 | |
5990a305 | 1237 | kfree(rings); |
362899b8 JW |
1238 | return ret; |
1239 | } | |
9a393b5d | 1240 | EXPORT_SYMBOL_GPL(tap_queue_resize); |
ebc05ba7 | 1241 | |
d9f1f61c SG |
1242 | static int tap_list_add(dev_t major, const char *device_name) |
1243 | { | |
1244 | struct major_info *tap_major; | |
1245 | ||
1246 | tap_major = kzalloc(sizeof(*tap_major), GFP_ATOMIC); | |
1247 | if (!tap_major) | |
1248 | return -ENOMEM; | |
1249 | ||
1250 | tap_major->major = MAJOR(major); | |
1251 | ||
1252 | idr_init(&tap_major->minor_idr); | |
ffa423fb | 1253 | spin_lock_init(&tap_major->minor_lock); |
d9f1f61c SG |
1254 | |
1255 | tap_major->device_name = device_name; | |
1256 | ||
1257 | list_add_tail_rcu(&tap_major->next, &major_list); | |
1258 | return 0; | |
1259 | } | |
1260 | ||
dea6e19f GM |
1261 | int tap_create_cdev(struct cdev *tap_cdev, dev_t *tap_major, |
1262 | const char *device_name, struct module *module) | |
ebc05ba7 SG |
1263 | { |
1264 | int err; | |
1265 | ||
1266 | err = alloc_chrdev_region(tap_major, 0, TAP_NUM_DEVS, device_name); | |
1267 | if (err) | |
1268 | goto out1; | |
1269 | ||
1270 | cdev_init(tap_cdev, &tap_fops); | |
dea6e19f | 1271 | tap_cdev->owner = module; |
ebc05ba7 SG |
1272 | err = cdev_add(tap_cdev, *tap_major, TAP_NUM_DEVS); |
1273 | if (err) | |
1274 | goto out2; | |
1275 | ||
d9f1f61c SG |
1276 | err = tap_list_add(*tap_major, device_name); |
1277 | if (err) | |
1278 | goto out3; | |
ebc05ba7 SG |
1279 | |
1280 | return 0; | |
1281 | ||
d9f1f61c SG |
1282 | out3: |
1283 | cdev_del(tap_cdev); | |
ebc05ba7 SG |
1284 | out2: |
1285 | unregister_chrdev_region(*tap_major, TAP_NUM_DEVS); | |
1286 | out1: | |
1287 | return err; | |
1288 | } | |
9a393b5d | 1289 | EXPORT_SYMBOL_GPL(tap_create_cdev); |
ebc05ba7 SG |
1290 | |
1291 | void tap_destroy_cdev(dev_t major, struct cdev *tap_cdev) | |
1292 | { | |
d9f1f61c SG |
1293 | struct major_info *tap_major, *tmp; |
1294 | ||
ebc05ba7 SG |
1295 | cdev_del(tap_cdev); |
1296 | unregister_chrdev_region(major, TAP_NUM_DEVS); | |
d9f1f61c SG |
1297 | list_for_each_entry_safe(tap_major, tmp, &major_list, next) { |
1298 | if (tap_major->major == MAJOR(major)) { | |
1299 | idr_destroy(&tap_major->minor_idr); | |
1300 | list_del_rcu(&tap_major->next); | |
1301 | kfree_rcu(tap_major, rcu); | |
1302 | } | |
1303 | } | |
ebc05ba7 | 1304 | } |
9a393b5d SG |
1305 | EXPORT_SYMBOL_GPL(tap_destroy_cdev); |
1306 | ||
1307 | MODULE_AUTHOR("Arnd Bergmann <arnd@arndb.de>"); | |
1308 | MODULE_AUTHOR("Sainath Grandhi <sainath.grandhi@intel.com>"); | |
1309 | MODULE_LICENSE("GPL"); |