Commit | Line | Data |
---|---|---|
c942fddf | 1 | // SPDX-License-Identifier: GPL-2.0-or-later |
1da177e4 LT |
2 | /* |
3 | * TUN - Universal TUN/TAP device driver. | |
4 | * Copyright (C) 1999-2002 Maxim Krasnyansky <maxk@qualcomm.com> | |
5 | * | |
1da177e4 LT |
6 | * $Id: tun.c,v 1.15 2002/03/01 02:44:24 maxk Exp $ |
7 | */ | |
8 | ||
9 | /* | |
10 | * Changes: | |
11 | * | |
ff4cc3ac MK |
12 | * Mike Kershaw <dragorn@kismetwireless.net> 2005/08/14 |
13 | * Add TUNSETLINK ioctl to set the link encapsulation | |
14 | * | |
1da177e4 | 15 | * Mark Smith <markzzzsmith@yahoo.com.au> |
344dc8ed | 16 | * Use eth_random_addr() for tap MAC address. |
1da177e4 LT |
17 | * |
18 | * Harald Roelle <harald.roelle@ifi.lmu.de> 2004/04/20 | |
19 | * Fixes in packet dropping, queue length setting and queue wakeup. | |
20 | * Increased default tx queue length. | |
21 | * Added ethtool API. | |
22 | * Minor cleanups | |
23 | * | |
24 | * Daniel Podlejski <underley@underley.eu.org> | |
25 | * Modifications for 2.3.99-pre5 kernel. | |
26 | */ | |
27 | ||
6b8a66ee JP |
28 | #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt |
29 | ||
1da177e4 LT |
30 | #define DRV_NAME "tun" |
31 | #define DRV_VERSION "1.6" | |
32 | #define DRV_DESCRIPTION "Universal TUN/TAP device driver" | |
33 | #define DRV_COPYRIGHT "(C) 1999-2004 Max Krasnyansky <maxk@qualcomm.com>" | |
34 | ||
1da177e4 LT |
35 | #include <linux/module.h> |
36 | #include <linux/errno.h> | |
37 | #include <linux/kernel.h> | |
174cd4b1 | 38 | #include <linux/sched/signal.h> |
1da177e4 LT |
39 | #include <linux/major.h> |
40 | #include <linux/slab.h> | |
41 | #include <linux/poll.h> | |
42 | #include <linux/fcntl.h> | |
43 | #include <linux/init.h> | |
44 | #include <linux/skbuff.h> | |
45 | #include <linux/netdevice.h> | |
46 | #include <linux/etherdevice.h> | |
47 | #include <linux/miscdevice.h> | |
48 | #include <linux/ethtool.h> | |
49 | #include <linux/rtnetlink.h> | |
50857e2a | 50 | #include <linux/compat.h> |
1da177e4 LT |
51 | #include <linux/if.h> |
52 | #include <linux/if_arp.h> | |
53 | #include <linux/if_ether.h> | |
54 | #include <linux/if_tun.h> | |
6680ec68 | 55 | #include <linux/if_vlan.h> |
1da177e4 | 56 | #include <linux/crc32.h> |
d647a591 | 57 | #include <linux/nsproxy.h> |
f43798c2 | 58 | #include <linux/virtio_net.h> |
99405162 | 59 | #include <linux/rcupdate.h> |
881d966b | 60 | #include <net/net_namespace.h> |
79d17604 | 61 | #include <net/netns/generic.h> |
f019a7a5 | 62 | #include <net/rtnetlink.h> |
33dccbb0 | 63 | #include <net/sock.h> |
735fc405 | 64 | #include <net/xdp.h> |
b9815eb1 | 65 | #include <net/ip_tunnels.h> |
93e14b6d | 66 | #include <linux/seq_file.h> |
e0b46d0e | 67 | #include <linux/uio.h> |
1576d986 | 68 | #include <linux/skb_array.h> |
761876c8 JW |
69 | #include <linux/bpf.h> |
70 | #include <linux/bpf_trace.h> | |
90e33d45 | 71 | #include <linux/mutex.h> |
1da177e4 | 72 | |
7c0f6ba6 | 73 | #include <linux/uaccess.h> |
f2780d6d | 74 | #include <linux/proc_fs.h> |
1da177e4 | 75 | |
4e24f2dd CW |
76 | static void tun_default_link_ksettings(struct net_device *dev, |
77 | struct ethtool_link_ksettings *cmd); | |
78 | ||
7df13219 | 79 | #define TUN_RX_PAD (NET_IP_ALIGN + NET_SKB_PAD) |
66ccbc9c | 80 | |
031f5e03 MT |
81 | /* TUN device flags */ |
82 | ||
83 | /* IFF_ATTACH_QUEUE is never stored in device flags, | |
84 | * overload it to mean fasync when stored there. | |
85 | */ | |
86 | #define TUN_FASYNC IFF_ATTACH_QUEUE | |
1cf8e410 MT |
87 | /* High bits in flags field are unused. */ |
88 | #define TUN_VNET_LE 0x80000000 | |
8b8e658b | 89 | #define TUN_VNET_BE 0x40000000 |
031f5e03 MT |
90 | |
91 | #define TUN_FEATURES (IFF_NO_PI | IFF_ONE_QUEUE | IFF_VNET_HDR | \ | |
90e33d45 PP |
92 | IFF_MULTI_QUEUE | IFF_NAPI | IFF_NAPI_FRAGS) |
93 | ||
0690899b MT |
94 | #define GOODCOPY_LEN 128 |
95 | ||
f271b2cc MK |
96 | #define FLT_EXACT_COUNT 8 |
97 | struct tap_filter { | |
98 | unsigned int count; /* Number of addrs. Zero means disabled */ | |
99 | u32 mask[2]; /* Mask of the hashed addrs */ | |
100 | unsigned char addr[FLT_EXACT_COUNT][ETH_ALEN]; | |
101 | }; | |
102 | ||
baf71c5c PG |
103 | /* MAX_TAP_QUEUES 256 is chosen to allow rx/tx queues to be equal |
104 | * to max number of VCPUs in guest. */ | |
105 | #define MAX_TAP_QUEUES 256 | |
b8732fb7 | 106 | #define MAX_TAP_FLOWS 4096 |
c8d68e6b | 107 | |
96442e42 JW |
108 | #define TUN_FLOW_EXPIRE (3 * HZ) |
109 | ||
54f968d6 | 110 | /* A tun_file connects an open character device to a tuntap netdevice. It |
92d4ea6e | 111 | * also contains all socket related structures (except sock_fprog and tap_filter) |
54f968d6 JW |
112 | * to serve as one transmit queue for tuntap device. The sock_fprog and |
113 | * tap_filter were kept in tun_struct since they were used for filtering for the | |
36fe8c09 | 114 | * netdevice not for a specific queue (at least I didn't see the requirement for |
54f968d6 | 115 | * this). |
6e914fc7 JW |
116 | * |
117 | * RCU usage: | |
36fe8c09 | 118 | * The tun_file and tun_struct are loosely coupled, the pointer from one to the |
6e914fc7 | 119 | * other can only be read while rcu_read_lock or rtnl_lock is held. |
54f968d6 | 120 | */ |
631ab46b | 121 | struct tun_file { |
54f968d6 JW |
122 | struct sock sk; |
123 | struct socket socket; | |
6e914fc7 | 124 | struct tun_struct __rcu *tun; |
54f968d6 JW |
125 | struct fasync_struct *fasync; |
126 | /* only used for fasnyc */ | |
127 | unsigned int flags; | |
fb7589a1 PE |
128 | union { |
129 | u16 queue_index; | |
130 | unsigned int ifindex; | |
131 | }; | |
94317099 | 132 | struct napi_struct napi; |
aec72f33 | 133 | bool napi_enabled; |
af3fb24e | 134 | bool napi_frags_enabled; |
90e33d45 | 135 | struct mutex napi_mutex; /* Protects access to the above napi */ |
4008e97f JW |
136 | struct list_head next; |
137 | struct tun_struct *detached; | |
5990a305 | 138 | struct ptr_ring tx_ring; |
8bf5c4ee | 139 | struct xdp_rxq_info xdp_rxq; |
631ab46b EB |
140 | }; |
141 | ||
f9e06c45 JW |
142 | struct tun_page { |
143 | struct page *page; | |
144 | int count; | |
145 | }; | |
146 | ||
96442e42 JW |
147 | struct tun_flow_entry { |
148 | struct hlist_node hash_link; | |
149 | struct rcu_head rcu; | |
150 | struct tun_struct *tun; | |
151 | ||
152 | u32 rxhash; | |
9bc88939 | 153 | u32 rps_rxhash; |
96442e42 | 154 | int queue_index; |
83b1bc12 | 155 | unsigned long updated ____cacheline_aligned_in_smp; |
96442e42 JW |
156 | }; |
157 | ||
158 | #define TUN_NUM_FLOW_ENTRIES 1024 | |
f13b5468 | 159 | #define TUN_MASK_FLOW_ENTRIES (TUN_NUM_FLOW_ENTRIES - 1) |
96442e42 | 160 | |
cd5681d7 | 161 | struct tun_prog { |
96f84061 JW |
162 | struct rcu_head rcu; |
163 | struct bpf_prog *prog; | |
164 | }; | |
165 | ||
54f968d6 | 166 | /* Since the socket were moved to tun_file, to preserve the behavior of persist |
36fe8c09 | 167 | * device, socket filter, sndbuf and vnet header size were restore when the |
54f968d6 JW |
168 | * file were attached to a persist device. |
169 | */ | |
14daa021 | 170 | struct tun_struct { |
c8d68e6b JW |
171 | struct tun_file __rcu *tfiles[MAX_TAP_QUEUES]; |
172 | unsigned int numqueues; | |
f271b2cc | 173 | unsigned int flags; |
0625c883 EB |
174 | kuid_t owner; |
175 | kgid_t group; | |
14daa021 | 176 | |
14daa021 | 177 | struct net_device *dev; |
c8f44aff | 178 | netdev_features_t set_features; |
88255375 | 179 | #define TUN_USER_FEATURES (NETIF_F_HW_CSUM|NETIF_F_TSO_ECN|NETIF_F_TSO| \ |
d591a1f3 | 180 | NETIF_F_TSO6) |
d9d52b51 | 181 | |
eaea34b2 | 182 | int align; |
d9d52b51 | 183 | int vnet_hdr_sz; |
54f968d6 JW |
184 | int sndbuf; |
185 | struct tap_filter txflt; | |
186 | struct sock_fprog fprog; | |
187 | /* protected by rtnl lock */ | |
188 | bool filter_attached; | |
3424170f | 189 | u32 msg_enable; |
96442e42 | 190 | spinlock_t lock; |
96442e42 JW |
191 | struct hlist_head flows[TUN_NUM_FLOW_ENTRIES]; |
192 | struct timer_list flow_gc_timer; | |
193 | unsigned long ageing_time; | |
4008e97f JW |
194 | unsigned int numdisabled; |
195 | struct list_head disabled; | |
5dbbaf2d | 196 | void *security; |
b8732fb7 | 197 | u32 flow_count; |
5503fcec | 198 | u32 rx_batched; |
497a5757 | 199 | atomic_long_t rx_frame_errors; |
761876c8 | 200 | struct bpf_prog __rcu *xdp_prog; |
cd5681d7 | 201 | struct tun_prog __rcu *steering_prog; |
aff3d70a | 202 | struct tun_prog __rcu *filter_prog; |
4e24f2dd | 203 | struct ethtool_link_ksettings link_ksettings; |
14daa021 | 204 | }; |
1da177e4 | 205 | |
aff3d70a JW |
206 | struct veth { |
207 | __be16 h_vlan_proto; | |
208 | __be16 h_vlan_TCI; | |
14daa021 | 209 | }; |
1da177e4 | 210 | |
94317099 PP |
211 | static int tun_napi_receive(struct napi_struct *napi, int budget) |
212 | { | |
213 | struct tun_file *tfile = container_of(napi, struct tun_file, napi); | |
214 | struct sk_buff_head *queue = &tfile->sk.sk_write_queue; | |
215 | struct sk_buff_head process_queue; | |
216 | struct sk_buff *skb; | |
217 | int received = 0; | |
218 | ||
219 | __skb_queue_head_init(&process_queue); | |
220 | ||
221 | spin_lock(&queue->lock); | |
222 | skb_queue_splice_tail_init(queue, &process_queue); | |
223 | spin_unlock(&queue->lock); | |
224 | ||
225 | while (received < budget && (skb = __skb_dequeue(&process_queue))) { | |
226 | napi_gro_receive(napi, skb); | |
227 | ++received; | |
228 | } | |
229 | ||
230 | if (!skb_queue_empty(&process_queue)) { | |
231 | spin_lock(&queue->lock); | |
232 | skb_queue_splice(&process_queue, queue); | |
233 | spin_unlock(&queue->lock); | |
234 | } | |
235 | ||
236 | return received; | |
237 | } | |
238 | ||
239 | static int tun_napi_poll(struct napi_struct *napi, int budget) | |
240 | { | |
241 | unsigned int received; | |
242 | ||
243 | received = tun_napi_receive(napi, budget); | |
244 | ||
245 | if (received < budget) | |
246 | napi_complete_done(napi, received); | |
247 | ||
248 | return received; | |
249 | } | |
250 | ||
251 | static void tun_napi_init(struct tun_struct *tun, struct tun_file *tfile, | |
af3fb24e | 252 | bool napi_en, bool napi_frags) |
94317099 | 253 | { |
aec72f33 | 254 | tfile->napi_enabled = napi_en; |
af3fb24e | 255 | tfile->napi_frags_enabled = napi_en && napi_frags; |
94317099 | 256 | if (napi_en) { |
c39e342a PP |
257 | netif_tx_napi_add(tun->dev, &tfile->napi, tun_napi_poll, |
258 | NAPI_POLL_WEIGHT); | |
94317099 PP |
259 | napi_enable(&tfile->napi); |
260 | } | |
261 | } | |
262 | ||
06e55add | 263 | static void tun_napi_disable(struct tun_file *tfile) |
94317099 | 264 | { |
aec72f33 | 265 | if (tfile->napi_enabled) |
94317099 PP |
266 | napi_disable(&tfile->napi); |
267 | } | |
268 | ||
06e55add | 269 | static void tun_napi_del(struct tun_file *tfile) |
94317099 | 270 | { |
aec72f33 | 271 | if (tfile->napi_enabled) |
94317099 PP |
272 | netif_napi_del(&tfile->napi); |
273 | } | |
274 | ||
af3fb24e | 275 | static bool tun_napi_frags_enabled(const struct tun_file *tfile) |
90e33d45 | 276 | { |
af3fb24e | 277 | return tfile->napi_frags_enabled; |
90e33d45 PP |
278 | } |
279 | ||
8b8e658b GK |
280 | #ifdef CONFIG_TUN_VNET_CROSS_LE |
281 | static inline bool tun_legacy_is_little_endian(struct tun_struct *tun) | |
282 | { | |
283 | return tun->flags & TUN_VNET_BE ? false : | |
284 | virtio_legacy_is_little_endian(); | |
285 | } | |
286 | ||
287 | static long tun_get_vnet_be(struct tun_struct *tun, int __user *argp) | |
288 | { | |
289 | int be = !!(tun->flags & TUN_VNET_BE); | |
290 | ||
291 | if (put_user(be, argp)) | |
292 | return -EFAULT; | |
293 | ||
294 | return 0; | |
295 | } | |
296 | ||
297 | static long tun_set_vnet_be(struct tun_struct *tun, int __user *argp) | |
298 | { | |
299 | int be; | |
300 | ||
301 | if (get_user(be, argp)) | |
302 | return -EFAULT; | |
303 | ||
304 | if (be) | |
305 | tun->flags |= TUN_VNET_BE; | |
306 | else | |
307 | tun->flags &= ~TUN_VNET_BE; | |
308 | ||
309 | return 0; | |
310 | } | |
311 | #else | |
312 | static inline bool tun_legacy_is_little_endian(struct tun_struct *tun) | |
313 | { | |
314 | return virtio_legacy_is_little_endian(); | |
315 | } | |
316 | ||
317 | static long tun_get_vnet_be(struct tun_struct *tun, int __user *argp) | |
318 | { | |
319 | return -EINVAL; | |
320 | } | |
321 | ||
322 | static long tun_set_vnet_be(struct tun_struct *tun, int __user *argp) | |
323 | { | |
324 | return -EINVAL; | |
325 | } | |
326 | #endif /* CONFIG_TUN_VNET_CROSS_LE */ | |
327 | ||
25bd55bb GK |
328 | static inline bool tun_is_little_endian(struct tun_struct *tun) |
329 | { | |
7d824109 | 330 | return tun->flags & TUN_VNET_LE || |
8b8e658b | 331 | tun_legacy_is_little_endian(tun); |
25bd55bb GK |
332 | } |
333 | ||
56f0dcc5 MT |
334 | static inline u16 tun16_to_cpu(struct tun_struct *tun, __virtio16 val) |
335 | { | |
25bd55bb | 336 | return __virtio16_to_cpu(tun_is_little_endian(tun), val); |
56f0dcc5 MT |
337 | } |
338 | ||
339 | static inline __virtio16 cpu_to_tun16(struct tun_struct *tun, u16 val) | |
340 | { | |
25bd55bb | 341 | return __cpu_to_virtio16(tun_is_little_endian(tun), val); |
56f0dcc5 MT |
342 | } |
343 | ||
96442e42 JW |
344 | static inline u32 tun_hashfn(u32 rxhash) |
345 | { | |
f13b5468 | 346 | return rxhash & TUN_MASK_FLOW_ENTRIES; |
96442e42 JW |
347 | } |
348 | ||
349 | static struct tun_flow_entry *tun_flow_find(struct hlist_head *head, u32 rxhash) | |
350 | { | |
351 | struct tun_flow_entry *e; | |
96442e42 | 352 | |
b67bfe0d | 353 | hlist_for_each_entry_rcu(e, head, hash_link) { |
96442e42 JW |
354 | if (e->rxhash == rxhash) |
355 | return e; | |
356 | } | |
357 | return NULL; | |
358 | } | |
359 | ||
360 | static struct tun_flow_entry *tun_flow_create(struct tun_struct *tun, | |
361 | struct hlist_head *head, | |
362 | u32 rxhash, u16 queue_index) | |
363 | { | |
9fdc6bef ED |
364 | struct tun_flow_entry *e = kmalloc(sizeof(*e), GFP_ATOMIC); |
365 | ||
96442e42 | 366 | if (e) { |
3424170f MK |
367 | netif_info(tun, tx_queued, tun->dev, |
368 | "create flow: hash %u index %u\n", | |
369 | rxhash, queue_index); | |
96442e42 JW |
370 | e->updated = jiffies; |
371 | e->rxhash = rxhash; | |
9bc88939 | 372 | e->rps_rxhash = 0; |
96442e42 JW |
373 | e->queue_index = queue_index; |
374 | e->tun = tun; | |
375 | hlist_add_head_rcu(&e->hash_link, head); | |
b8732fb7 | 376 | ++tun->flow_count; |
96442e42 JW |
377 | } |
378 | return e; | |
379 | } | |
380 | ||
96442e42 JW |
381 | static void tun_flow_delete(struct tun_struct *tun, struct tun_flow_entry *e) |
382 | { | |
3424170f MK |
383 | netif_info(tun, tx_queued, tun->dev, "delete flow: hash %u index %u\n", |
384 | e->rxhash, e->queue_index); | |
96442e42 | 385 | hlist_del_rcu(&e->hash_link); |
9fdc6bef | 386 | kfree_rcu(e, rcu); |
b8732fb7 | 387 | --tun->flow_count; |
96442e42 JW |
388 | } |
389 | ||
390 | static void tun_flow_flush(struct tun_struct *tun) | |
391 | { | |
392 | int i; | |
393 | ||
394 | spin_lock_bh(&tun->lock); | |
395 | for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++) { | |
396 | struct tun_flow_entry *e; | |
b67bfe0d | 397 | struct hlist_node *n; |
96442e42 | 398 | |
b67bfe0d | 399 | hlist_for_each_entry_safe(e, n, &tun->flows[i], hash_link) |
96442e42 JW |
400 | tun_flow_delete(tun, e); |
401 | } | |
402 | spin_unlock_bh(&tun->lock); | |
403 | } | |
404 | ||
405 | static void tun_flow_delete_by_queue(struct tun_struct *tun, u16 queue_index) | |
406 | { | |
407 | int i; | |
408 | ||
409 | spin_lock_bh(&tun->lock); | |
410 | for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++) { | |
411 | struct tun_flow_entry *e; | |
b67bfe0d | 412 | struct hlist_node *n; |
96442e42 | 413 | |
b67bfe0d | 414 | hlist_for_each_entry_safe(e, n, &tun->flows[i], hash_link) { |
96442e42 JW |
415 | if (e->queue_index == queue_index) |
416 | tun_flow_delete(tun, e); | |
417 | } | |
418 | } | |
419 | spin_unlock_bh(&tun->lock); | |
420 | } | |
421 | ||
e99e88a9 | 422 | static void tun_flow_cleanup(struct timer_list *t) |
96442e42 | 423 | { |
e99e88a9 | 424 | struct tun_struct *tun = from_timer(tun, t, flow_gc_timer); |
96442e42 JW |
425 | unsigned long delay = tun->ageing_time; |
426 | unsigned long next_timer = jiffies + delay; | |
427 | unsigned long count = 0; | |
428 | int i; | |
429 | ||
7dbfb4ef | 430 | spin_lock(&tun->lock); |
96442e42 JW |
431 | for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++) { |
432 | struct tun_flow_entry *e; | |
b67bfe0d | 433 | struct hlist_node *n; |
96442e42 | 434 | |
b67bfe0d | 435 | hlist_for_each_entry_safe(e, n, &tun->flows[i], hash_link) { |
96442e42 | 436 | unsigned long this_timer; |
81d98fa4 | 437 | |
96442e42 | 438 | this_timer = e->updated + delay; |
81d98fa4 | 439 | if (time_before_eq(this_timer, jiffies)) { |
96442e42 | 440 | tun_flow_delete(tun, e); |
81d98fa4 ED |
441 | continue; |
442 | } | |
443 | count++; | |
444 | if (time_before(this_timer, next_timer)) | |
96442e42 JW |
445 | next_timer = this_timer; |
446 | } | |
447 | } | |
448 | ||
449 | if (count) | |
450 | mod_timer(&tun->flow_gc_timer, round_jiffies_up(next_timer)); | |
7dbfb4ef | 451 | spin_unlock(&tun->lock); |
96442e42 JW |
452 | } |
453 | ||
49974420 | 454 | static void tun_flow_update(struct tun_struct *tun, u32 rxhash, |
9e85722d | 455 | struct tun_file *tfile) |
96442e42 JW |
456 | { |
457 | struct hlist_head *head; | |
458 | struct tun_flow_entry *e; | |
459 | unsigned long delay = tun->ageing_time; | |
9e85722d | 460 | u16 queue_index = tfile->queue_index; |
96442e42 | 461 | |
5c327f67 | 462 | head = &tun->flows[tun_hashfn(rxhash)]; |
96442e42 JW |
463 | |
464 | rcu_read_lock(); | |
465 | ||
96442e42 JW |
466 | e = tun_flow_find(head, rxhash); |
467 | if (likely(e)) { | |
468 | /* TODO: keep queueing to old queue until it's empty? */ | |
4ffdd22e ED |
469 | if (READ_ONCE(e->queue_index) != queue_index) |
470 | WRITE_ONCE(e->queue_index, queue_index); | |
83b1bc12 LR |
471 | if (e->updated != jiffies) |
472 | e->updated = jiffies; | |
9bc88939 | 473 | sock_rps_record_flow_hash(e->rps_rxhash); |
96442e42 JW |
474 | } else { |
475 | spin_lock_bh(&tun->lock); | |
b8732fb7 JW |
476 | if (!tun_flow_find(head, rxhash) && |
477 | tun->flow_count < MAX_TAP_FLOWS) | |
96442e42 JW |
478 | tun_flow_create(tun, head, rxhash, queue_index); |
479 | ||
480 | if (!timer_pending(&tun->flow_gc_timer)) | |
481 | mod_timer(&tun->flow_gc_timer, | |
482 | round_jiffies_up(jiffies + delay)); | |
483 | spin_unlock_bh(&tun->lock); | |
484 | } | |
485 | ||
96442e42 JW |
486 | rcu_read_unlock(); |
487 | } | |
488 | ||
516c512b | 489 | /* Save the hash received in the stack receive path and update the |
9bc88939 TH |
490 | * flow_hash table accordingly. |
491 | */ | |
492 | static inline void tun_flow_save_rps_rxhash(struct tun_flow_entry *e, u32 hash) | |
493 | { | |
567e4b79 | 494 | if (unlikely(e->rps_rxhash != hash)) |
9bc88939 | 495 | e->rps_rxhash = hash; |
9bc88939 TH |
496 | } |
497 | ||
4b035271 | 498 | /* We try to identify a flow through its rxhash. The reason that |
92d4ea6e | 499 | * we do not check rxq no. is because some cards(e.g 82599), chooses |
c8d68e6b JW |
500 | * the rxq based on the txq where the last packet of the flow comes. As |
501 | * the userspace application move between processors, we may get a | |
4b035271 | 502 | * different rxq no. here. |
c8d68e6b | 503 | */ |
96f84061 | 504 | static u16 tun_automq_select_queue(struct tun_struct *tun, struct sk_buff *skb) |
c8d68e6b | 505 | { |
96442e42 | 506 | struct tun_flow_entry *e; |
c8d68e6b JW |
507 | u32 txq = 0; |
508 | u32 numqueues = 0; | |
509 | ||
6aa7de05 | 510 | numqueues = READ_ONCE(tun->numqueues); |
c8d68e6b | 511 | |
feec084a | 512 | txq = __skb_get_hash_symmetric(skb); |
4b035271 WL |
513 | e = tun_flow_find(&tun->flows[tun_hashfn(txq)], txq); |
514 | if (e) { | |
515 | tun_flow_save_rps_rxhash(e, txq); | |
516 | txq = e->queue_index; | |
517 | } else { | |
518 | /* use multiply and shift instead of expensive divide */ | |
519 | txq = ((u64)txq * numqueues) >> 32; | |
c8d68e6b JW |
520 | } |
521 | ||
c8d68e6b JW |
522 | return txq; |
523 | } | |
524 | ||
96f84061 JW |
525 | static u16 tun_ebpf_select_queue(struct tun_struct *tun, struct sk_buff *skb) |
526 | { | |
cd5681d7 | 527 | struct tun_prog *prog; |
a35d310f | 528 | u32 numqueues; |
96f84061 JW |
529 | u16 ret = 0; |
530 | ||
a35d310f JW |
531 | numqueues = READ_ONCE(tun->numqueues); |
532 | if (!numqueues) | |
533 | return 0; | |
534 | ||
96f84061 JW |
535 | prog = rcu_dereference(tun->steering_prog); |
536 | if (prog) | |
537 | ret = bpf_prog_run_clear_cb(prog->prog, skb); | |
538 | ||
a35d310f | 539 | return ret % numqueues; |
96f84061 JW |
540 | } |
541 | ||
542 | static u16 tun_select_queue(struct net_device *dev, struct sk_buff *skb, | |
a350ecce | 543 | struct net_device *sb_dev) |
96f84061 JW |
544 | { |
545 | struct tun_struct *tun = netdev_priv(dev); | |
546 | u16 ret; | |
547 | ||
548 | rcu_read_lock(); | |
549 | if (rcu_dereference(tun->steering_prog)) | |
550 | ret = tun_ebpf_select_queue(tun, skb); | |
551 | else | |
552 | ret = tun_automq_select_queue(tun, skb); | |
553 | rcu_read_unlock(); | |
554 | ||
555 | return ret; | |
556 | } | |
557 | ||
cde8b15f JW |
558 | static inline bool tun_not_capable(struct tun_struct *tun) |
559 | { | |
560 | const struct cred *cred = current_cred(); | |
c260b772 | 561 | struct net *net = dev_net(tun->dev); |
cde8b15f JW |
562 | |
563 | return ((uid_valid(tun->owner) && !uid_eq(cred->euid, tun->owner)) || | |
564 | (gid_valid(tun->group) && !in_egroup_p(tun->group))) && | |
c260b772 | 565 | !ns_capable(net->user_ns, CAP_NET_ADMIN); |
cde8b15f JW |
566 | } |
567 | ||
c8d68e6b JW |
568 | static void tun_set_real_num_queues(struct tun_struct *tun) |
569 | { | |
570 | netif_set_real_num_tx_queues(tun->dev, tun->numqueues); | |
571 | netif_set_real_num_rx_queues(tun->dev, tun->numqueues); | |
572 | } | |
573 | ||
4008e97f JW |
574 | static void tun_disable_queue(struct tun_struct *tun, struct tun_file *tfile) |
575 | { | |
576 | tfile->detached = tun; | |
577 | list_add_tail(&tfile->next, &tun->disabled); | |
578 | ++tun->numdisabled; | |
579 | } | |
580 | ||
d32649d1 | 581 | static struct tun_struct *tun_enable_queue(struct tun_file *tfile) |
4008e97f JW |
582 | { |
583 | struct tun_struct *tun = tfile->detached; | |
584 | ||
585 | tfile->detached = NULL; | |
586 | list_del_init(&tfile->next); | |
587 | --tun->numdisabled; | |
588 | return tun; | |
589 | } | |
590 | ||
3a403076 | 591 | void tun_ptr_free(void *ptr) |
fc72d1d5 JW |
592 | { |
593 | if (!ptr) | |
594 | return; | |
1ffcbc85 JDB |
595 | if (tun_is_xdp_frame(ptr)) { |
596 | struct xdp_frame *xdpf = tun_ptr_to_xdp(ptr); | |
fc72d1d5 | 597 | |
03993094 | 598 | xdp_return_frame(xdpf); |
fc72d1d5 JW |
599 | } else { |
600 | __skb_array_destroy_skb(ptr); | |
601 | } | |
602 | } | |
3a403076 | 603 | EXPORT_SYMBOL_GPL(tun_ptr_free); |
fc72d1d5 | 604 | |
4bfb0513 JW |
605 | static void tun_queue_purge(struct tun_file *tfile) |
606 | { | |
fc72d1d5 | 607 | void *ptr; |
1576d986 | 608 | |
fc72d1d5 JW |
609 | while ((ptr = ptr_ring_consume(&tfile->tx_ring)) != NULL) |
610 | tun_ptr_free(ptr); | |
1576d986 | 611 | |
5503fcec | 612 | skb_queue_purge(&tfile->sk.sk_write_queue); |
4bfb0513 JW |
613 | skb_queue_purge(&tfile->sk.sk_error_queue); |
614 | } | |
615 | ||
c8d68e6b JW |
616 | static void __tun_detach(struct tun_file *tfile, bool clean) |
617 | { | |
618 | struct tun_file *ntfile; | |
619 | struct tun_struct *tun; | |
c8d68e6b | 620 | |
b8deabd3 JW |
621 | tun = rtnl_dereference(tfile->tun); |
622 | ||
94317099 | 623 | if (tun && clean) { |
06e55add ED |
624 | tun_napi_disable(tfile); |
625 | tun_napi_del(tfile); | |
94317099 PP |
626 | } |
627 | ||
9e85722d | 628 | if (tun && !tfile->detached) { |
c8d68e6b JW |
629 | u16 index = tfile->queue_index; |
630 | BUG_ON(index >= tun->numqueues); | |
c8d68e6b JW |
631 | |
632 | rcu_assign_pointer(tun->tfiles[index], | |
633 | tun->tfiles[tun->numqueues - 1]); | |
b8deabd3 | 634 | ntfile = rtnl_dereference(tun->tfiles[index]); |
c8d68e6b | 635 | ntfile->queue_index = index; |
9871a9e4 JW |
636 | rcu_assign_pointer(tun->tfiles[tun->numqueues - 1], |
637 | NULL); | |
c8d68e6b JW |
638 | |
639 | --tun->numqueues; | |
9e85722d | 640 | if (clean) { |
c956674b | 641 | RCU_INIT_POINTER(tfile->tun, NULL); |
4008e97f | 642 | sock_put(&tfile->sk); |
9e85722d | 643 | } else |
4008e97f | 644 | tun_disable_queue(tun, tfile); |
c8d68e6b JW |
645 | |
646 | synchronize_net(); | |
96442e42 | 647 | tun_flow_delete_by_queue(tun, tun->numqueues + 1); |
c8d68e6b | 648 | /* Drop read queue */ |
4bfb0513 | 649 | tun_queue_purge(tfile); |
c8d68e6b | 650 | tun_set_real_num_queues(tun); |
dd38bd85 | 651 | } else if (tfile->detached && clean) { |
4008e97f | 652 | tun = tun_enable_queue(tfile); |
dd38bd85 JW |
653 | sock_put(&tfile->sk); |
654 | } | |
c8d68e6b JW |
655 | |
656 | if (clean) { | |
af668b3c MT |
657 | if (tun && tun->numqueues == 0 && tun->numdisabled == 0) { |
658 | netif_carrier_off(tun->dev); | |
659 | ||
40630b82 | 660 | if (!(tun->flags & IFF_PERSIST) && |
af668b3c | 661 | tun->dev->reg_state == NETREG_REGISTERED) |
4008e97f | 662 | unregister_netdevice(tun->dev); |
af668b3c | 663 | } |
b196d88a JW |
664 | if (tun) |
665 | xdp_rxq_info_unreg(&tfile->xdp_rxq); | |
7063efd3 | 666 | ptr_ring_cleanup(&tfile->tx_ring, tun_ptr_free); |
140e807d | 667 | sock_put(&tfile->sk); |
c8d68e6b JW |
668 | } |
669 | } | |
670 | ||
671 | static void tun_detach(struct tun_file *tfile, bool clean) | |
672 | { | |
83c1f36f SD |
673 | struct tun_struct *tun; |
674 | struct net_device *dev; | |
675 | ||
c8d68e6b | 676 | rtnl_lock(); |
83c1f36f SD |
677 | tun = rtnl_dereference(tfile->tun); |
678 | dev = tun ? tun->dev : NULL; | |
c8d68e6b | 679 | __tun_detach(tfile, clean); |
83c1f36f SD |
680 | if (dev) |
681 | netdev_state_change(dev); | |
c8d68e6b JW |
682 | rtnl_unlock(); |
683 | } | |
684 | ||
685 | static void tun_detach_all(struct net_device *dev) | |
686 | { | |
687 | struct tun_struct *tun = netdev_priv(dev); | |
4008e97f | 688 | struct tun_file *tfile, *tmp; |
c8d68e6b JW |
689 | int i, n = tun->numqueues; |
690 | ||
691 | for (i = 0; i < n; i++) { | |
b8deabd3 | 692 | tfile = rtnl_dereference(tun->tfiles[i]); |
c8d68e6b | 693 | BUG_ON(!tfile); |
06e55add | 694 | tun_napi_disable(tfile); |
addf8fc4 | 695 | tfile->socket.sk->sk_shutdown = RCV_SHUTDOWN; |
9e641bdc | 696 | tfile->socket.sk->sk_data_ready(tfile->socket.sk); |
c956674b | 697 | RCU_INIT_POINTER(tfile->tun, NULL); |
c8d68e6b JW |
698 | --tun->numqueues; |
699 | } | |
9e85722d | 700 | list_for_each_entry(tfile, &tun->disabled, next) { |
addf8fc4 | 701 | tfile->socket.sk->sk_shutdown = RCV_SHUTDOWN; |
9e641bdc | 702 | tfile->socket.sk->sk_data_ready(tfile->socket.sk); |
c956674b | 703 | RCU_INIT_POINTER(tfile->tun, NULL); |
9e85722d | 704 | } |
c8d68e6b JW |
705 | BUG_ON(tun->numqueues != 0); |
706 | ||
707 | synchronize_net(); | |
708 | for (i = 0; i < n; i++) { | |
b8deabd3 | 709 | tfile = rtnl_dereference(tun->tfiles[i]); |
06e55add | 710 | tun_napi_del(tfile); |
c8d68e6b | 711 | /* Drop read queue */ |
4bfb0513 | 712 | tun_queue_purge(tfile); |
b196d88a | 713 | xdp_rxq_info_unreg(&tfile->xdp_rxq); |
c8d68e6b JW |
714 | sock_put(&tfile->sk); |
715 | } | |
4008e97f JW |
716 | list_for_each_entry_safe(tfile, tmp, &tun->disabled, next) { |
717 | tun_enable_queue(tfile); | |
4bfb0513 | 718 | tun_queue_purge(tfile); |
b196d88a | 719 | xdp_rxq_info_unreg(&tfile->xdp_rxq); |
4008e97f JW |
720 | sock_put(&tfile->sk); |
721 | } | |
722 | BUG_ON(tun->numdisabled != 0); | |
dd38bd85 | 723 | |
40630b82 | 724 | if (tun->flags & IFF_PERSIST) |
dd38bd85 | 725 | module_put(THIS_MODULE); |
c8d68e6b JW |
726 | } |
727 | ||
94317099 | 728 | static int tun_attach(struct tun_struct *tun, struct file *file, |
77f22f92 YY |
729 | bool skip_filter, bool napi, bool napi_frags, |
730 | bool publish_tun) | |
a7385ba2 | 731 | { |
631ab46b | 732 | struct tun_file *tfile = file->private_data; |
1576d986 | 733 | struct net_device *dev = tun->dev; |
38231b7a | 734 | int err; |
a7385ba2 | 735 | |
5dbbaf2d PM |
736 | err = security_tun_dev_attach(tfile->socket.sk, tun->security); |
737 | if (err < 0) | |
738 | goto out; | |
739 | ||
38231b7a | 740 | err = -EINVAL; |
9e85722d | 741 | if (rtnl_dereference(tfile->tun) && !tfile->detached) |
38231b7a EB |
742 | goto out; |
743 | ||
744 | err = -EBUSY; | |
40630b82 | 745 | if (!(tun->flags & IFF_MULTI_QUEUE) && tun->numqueues == 1) |
c8d68e6b JW |
746 | goto out; |
747 | ||
748 | err = -E2BIG; | |
4008e97f JW |
749 | if (!tfile->detached && |
750 | tun->numqueues + tun->numdisabled == MAX_TAP_QUEUES) | |
38231b7a EB |
751 | goto out; |
752 | ||
753 | err = 0; | |
54f968d6 | 754 | |
92d4ea6e | 755 | /* Re-attach the filter to persist device */ |
849c9b6f | 756 | if (!skip_filter && (tun->filter_attached == true)) { |
8ced425e HFS |
757 | lock_sock(tfile->socket.sk); |
758 | err = sk_attach_filter(&tun->fprog, tfile->socket.sk); | |
759 | release_sock(tfile->socket.sk); | |
54f968d6 JW |
760 | if (!err) |
761 | goto out; | |
762 | } | |
1576d986 JW |
763 | |
764 | if (!tfile->detached && | |
b196d88a JW |
765 | ptr_ring_resize(&tfile->tx_ring, dev->tx_queue_len, |
766 | GFP_KERNEL, tun_ptr_free)) { | |
1576d986 JW |
767 | err = -ENOMEM; |
768 | goto out; | |
769 | } | |
770 | ||
c8d68e6b | 771 | tfile->queue_index = tun->numqueues; |
addf8fc4 | 772 | tfile->socket.sk->sk_shutdown &= ~RCV_SHUTDOWN; |
8bf5c4ee JDB |
773 | |
774 | if (tfile->detached) { | |
775 | /* Re-attach detached tfile, updating XDP queue_index */ | |
776 | WARN_ON(!xdp_rxq_info_is_reg(&tfile->xdp_rxq)); | |
777 | ||
778 | if (tfile->xdp_rxq.queue_index != tfile->queue_index) | |
779 | tfile->xdp_rxq.queue_index = tfile->queue_index; | |
780 | } else { | |
781 | /* Setup XDP RX-queue info, for new tfile getting attached */ | |
782 | err = xdp_rxq_info_reg(&tfile->xdp_rxq, | |
b02e5a0e | 783 | tun->dev, tfile->queue_index, 0); |
8bf5c4ee JDB |
784 | if (err < 0) |
785 | goto out; | |
8d5d8852 JDB |
786 | err = xdp_rxq_info_reg_mem_model(&tfile->xdp_rxq, |
787 | MEM_TYPE_PAGE_SHARED, NULL); | |
788 | if (err < 0) { | |
789 | xdp_rxq_info_unreg(&tfile->xdp_rxq); | |
790 | goto out; | |
791 | } | |
8bf5c4ee JDB |
792 | err = 0; |
793 | } | |
794 | ||
94317099 | 795 | if (tfile->detached) { |
4008e97f | 796 | tun_enable_queue(tfile); |
94317099 | 797 | } else { |
4008e97f | 798 | sock_hold(&tfile->sk); |
af3fb24e | 799 | tun_napi_init(tun, tfile, napi, napi_frags); |
94317099 | 800 | } |
4008e97f | 801 | |
e4a2a304 JW |
802 | if (rtnl_dereference(tun->xdp_prog)) |
803 | sock_set_flag(&tfile->sk, SOCK_XDP); | |
804 | ||
c8d68e6b JW |
805 | /* device is allowed to go away first, so no need to hold extra |
806 | * refcnt. | |
807 | */ | |
808 | ||
0b7959b6 SF |
809 | /* Publish tfile->tun and tun->tfiles only after we've fully |
810 | * initialized tfile; otherwise we risk using half-initialized | |
811 | * object. | |
812 | */ | |
77f22f92 YY |
813 | if (publish_tun) |
814 | rcu_assign_pointer(tfile->tun, tun); | |
0b7959b6 SF |
815 | rcu_assign_pointer(tun->tfiles[tun->numqueues], tfile); |
816 | tun->numqueues++; | |
3a03cb84 | 817 | tun_set_real_num_queues(tun); |
c8d68e6b JW |
818 | out: |
819 | return err; | |
631ab46b EB |
820 | } |
821 | ||
9484dc74 | 822 | static struct tun_struct *tun_get(struct tun_file *tfile) |
631ab46b | 823 | { |
6e914fc7 | 824 | struct tun_struct *tun; |
c70f1829 | 825 | |
6e914fc7 JW |
826 | rcu_read_lock(); |
827 | tun = rcu_dereference(tfile->tun); | |
828 | if (tun) | |
829 | dev_hold(tun->dev); | |
830 | rcu_read_unlock(); | |
c70f1829 EB |
831 | |
832 | return tun; | |
631ab46b EB |
833 | } |
834 | ||
631ab46b EB |
835 | static void tun_put(struct tun_struct *tun) |
836 | { | |
6e914fc7 | 837 | dev_put(tun->dev); |
631ab46b EB |
838 | } |
839 | ||
6b8a66ee | 840 | /* TAP filtering */ |
f271b2cc MK |
841 | static void addr_hash_set(u32 *mask, const u8 *addr) |
842 | { | |
843 | int n = ether_crc(ETH_ALEN, addr) >> 26; | |
844 | mask[n >> 5] |= (1 << (n & 31)); | |
845 | } | |
846 | ||
847 | static unsigned int addr_hash_test(const u32 *mask, const u8 *addr) | |
848 | { | |
849 | int n = ether_crc(ETH_ALEN, addr) >> 26; | |
850 | return mask[n >> 5] & (1 << (n & 31)); | |
851 | } | |
852 | ||
853 | static int update_filter(struct tap_filter *filter, void __user *arg) | |
854 | { | |
855 | struct { u8 u[ETH_ALEN]; } *addr; | |
856 | struct tun_filter uf; | |
857 | int err, alen, n, nexact; | |
858 | ||
859 | if (copy_from_user(&uf, arg, sizeof(uf))) | |
860 | return -EFAULT; | |
861 | ||
862 | if (!uf.count) { | |
863 | /* Disabled */ | |
864 | filter->count = 0; | |
865 | return 0; | |
866 | } | |
867 | ||
868 | alen = ETH_ALEN * uf.count; | |
28e8190d ME |
869 | addr = memdup_user(arg + sizeof(uf), alen); |
870 | if (IS_ERR(addr)) | |
871 | return PTR_ERR(addr); | |
f271b2cc MK |
872 | |
873 | /* The filter is updated without holding any locks. Which is | |
874 | * perfectly safe. We disable it first and in the worst | |
875 | * case we'll accept a few undesired packets. */ | |
876 | filter->count = 0; | |
877 | wmb(); | |
878 | ||
879 | /* Use first set of addresses as an exact filter */ | |
880 | for (n = 0; n < uf.count && n < FLT_EXACT_COUNT; n++) | |
881 | memcpy(filter->addr[n], addr[n].u, ETH_ALEN); | |
882 | ||
883 | nexact = n; | |
884 | ||
cfbf84fc AW |
885 | /* Remaining multicast addresses are hashed, |
886 | * unicast will leave the filter disabled. */ | |
f271b2cc | 887 | memset(filter->mask, 0, sizeof(filter->mask)); |
cfbf84fc AW |
888 | for (; n < uf.count; n++) { |
889 | if (!is_multicast_ether_addr(addr[n].u)) { | |
890 | err = 0; /* no filter */ | |
3b8d2a69 | 891 | goto free_addr; |
cfbf84fc | 892 | } |
f271b2cc | 893 | addr_hash_set(filter->mask, addr[n].u); |
cfbf84fc | 894 | } |
f271b2cc MK |
895 | |
896 | /* For ALLMULTI just set the mask to all ones. | |
897 | * This overrides the mask populated above. */ | |
898 | if ((uf.flags & TUN_FLT_ALLMULTI)) | |
899 | memset(filter->mask, ~0, sizeof(filter->mask)); | |
900 | ||
901 | /* Now enable the filter */ | |
902 | wmb(); | |
903 | filter->count = nexact; | |
904 | ||
905 | /* Return the number of exact filters */ | |
906 | err = nexact; | |
3b8d2a69 | 907 | free_addr: |
f271b2cc MK |
908 | kfree(addr); |
909 | return err; | |
910 | } | |
911 | ||
912 | /* Returns: 0 - drop, !=0 - accept */ | |
913 | static int run_filter(struct tap_filter *filter, const struct sk_buff *skb) | |
914 | { | |
915 | /* Cannot use eth_hdr(skb) here because skb_mac_hdr() is incorrect | |
916 | * at this point. */ | |
917 | struct ethhdr *eh = (struct ethhdr *) skb->data; | |
918 | int i; | |
919 | ||
920 | /* Exact match */ | |
921 | for (i = 0; i < filter->count; i++) | |
2e42e474 | 922 | if (ether_addr_equal(eh->h_dest, filter->addr[i])) |
f271b2cc MK |
923 | return 1; |
924 | ||
925 | /* Inexact match (multicast only) */ | |
926 | if (is_multicast_ether_addr(eh->h_dest)) | |
927 | return addr_hash_test(filter->mask, eh->h_dest); | |
928 | ||
929 | return 0; | |
930 | } | |
931 | ||
932 | /* | |
933 | * Checks whether the packet is accepted or not. | |
934 | * Returns: 0 - drop, !=0 - accept | |
935 | */ | |
936 | static int check_filter(struct tap_filter *filter, const struct sk_buff *skb) | |
937 | { | |
938 | if (!filter->count) | |
939 | return 1; | |
940 | ||
941 | return run_filter(filter, skb); | |
942 | } | |
943 | ||
1da177e4 LT |
944 | /* Network device part of the driver */ |
945 | ||
7282d491 | 946 | static const struct ethtool_ops tun_ethtool_ops; |
1da177e4 | 947 | |
c70f1829 EB |
948 | /* Net device detach from fd. */ |
949 | static void tun_net_uninit(struct net_device *dev) | |
950 | { | |
c8d68e6b | 951 | tun_detach_all(dev); |
c70f1829 EB |
952 | } |
953 | ||
1da177e4 LT |
954 | /* Net device open. */ |
955 | static int tun_net_open(struct net_device *dev) | |
956 | { | |
c8d68e6b | 957 | netif_tx_start_all_queues(dev); |
b20e2d54 | 958 | |
1da177e4 LT |
959 | return 0; |
960 | } | |
961 | ||
962 | /* Net device close. */ | |
963 | static int tun_net_close(struct net_device *dev) | |
964 | { | |
c8d68e6b | 965 | netif_tx_stop_all_queues(dev); |
1da177e4 LT |
966 | return 0; |
967 | } | |
968 | ||
969 | /* Net device start xmit */ | |
96f84061 | 970 | static void tun_automq_xmit(struct tun_struct *tun, struct sk_buff *skb) |
1da177e4 | 971 | { |
3df97ba8 | 972 | #ifdef CONFIG_RPS |
dc05360f | 973 | if (tun->numqueues == 1 && static_branch_unlikely(&rps_needed)) { |
9bc88939 TH |
974 | /* Select queue was not called for the skbuff, so we extract the |
975 | * RPS hash and save it into the flow_table here. | |
976 | */ | |
4b035271 | 977 | struct tun_flow_entry *e; |
9bc88939 TH |
978 | __u32 rxhash; |
979 | ||
feec084a | 980 | rxhash = __skb_get_hash_symmetric(skb); |
4b035271 WL |
981 | e = tun_flow_find(&tun->flows[tun_hashfn(rxhash)], rxhash); |
982 | if (e) | |
983 | tun_flow_save_rps_rxhash(e, rxhash); | |
9bc88939 | 984 | } |
3df97ba8 | 985 | #endif |
96f84061 JW |
986 | } |
987 | ||
aff3d70a JW |
988 | static unsigned int run_ebpf_filter(struct tun_struct *tun, |
989 | struct sk_buff *skb, | |
990 | int len) | |
991 | { | |
992 | struct tun_prog *prog = rcu_dereference(tun->filter_prog); | |
993 | ||
994 | if (prog) | |
995 | len = bpf_prog_run_clear_cb(prog->prog, skb); | |
996 | ||
997 | return len; | |
998 | } | |
999 | ||
96f84061 JW |
1000 | /* Net device start xmit */ |
1001 | static netdev_tx_t tun_net_xmit(struct sk_buff *skb, struct net_device *dev) | |
1002 | { | |
1003 | struct tun_struct *tun = netdev_priv(dev); | |
1004 | int txq = skb->queue_mapping; | |
1005 | struct tun_file *tfile; | |
aff3d70a | 1006 | int len = skb->len; |
96f84061 JW |
1007 | |
1008 | rcu_read_lock(); | |
1009 | tfile = rcu_dereference(tun->tfiles[txq]); | |
96f84061 JW |
1010 | |
1011 | /* Drop packet if interface is not attached */ | |
9871a9e4 | 1012 | if (!tfile) |
96f84061 JW |
1013 | goto drop; |
1014 | ||
1015 | if (!rcu_dereference(tun->steering_prog)) | |
1016 | tun_automq_xmit(tun, skb); | |
9bc88939 | 1017 | |
3424170f | 1018 | netif_info(tun, tx_queued, tun->dev, "%s %d\n", __func__, skb->len); |
6e914fc7 | 1019 | |
f271b2cc MK |
1020 | /* Drop if the filter does not like it. |
1021 | * This is a noop if the filter is disabled. | |
1022 | * Filter can be enabled only for the TAP devices. */ | |
1023 | if (!check_filter(&tun->txflt, skb)) | |
1024 | goto drop; | |
1025 | ||
54f968d6 JW |
1026 | if (tfile->socket.sk->sk_filter && |
1027 | sk_filter(tfile->socket.sk, skb)) | |
99405162 MT |
1028 | goto drop; |
1029 | ||
aff3d70a | 1030 | len = run_ebpf_filter(tun, skb, len); |
81c89507 | 1031 | if (len == 0 || pskb_trim(skb, len)) |
aff3d70a JW |
1032 | goto drop; |
1033 | ||
1f8b977a | 1034 | if (unlikely(skb_orphan_frags_rx(skb, GFP_ATOMIC))) |
7bf66305 JW |
1035 | goto drop; |
1036 | ||
7b996243 | 1037 | skb_tx_timestamp(skb); |
eda29772 | 1038 | |
0110d6f2 | 1039 | /* Orphan the skb - required as we might hang on to it |
7bf66305 JW |
1040 | * for indefinite time. |
1041 | */ | |
0110d6f2 MT |
1042 | skb_orphan(skb); |
1043 | ||
895b5c9f | 1044 | nf_reset_ct(skb); |
f8af75f3 | 1045 | |
5990a305 | 1046 | if (ptr_ring_produce(&tfile->tx_ring, skb)) |
1576d986 | 1047 | goto drop; |
1da177e4 LT |
1048 | |
1049 | /* Notify and wake up reader process */ | |
54f968d6 JW |
1050 | if (tfile->flags & TUN_FASYNC) |
1051 | kill_fasync(&tfile->fasync, SIGIO, POLL_IN); | |
9e641bdc | 1052 | tfile->socket.sk->sk_data_ready(tfile->socket.sk); |
6e914fc7 JW |
1053 | |
1054 | rcu_read_unlock(); | |
6ed10654 | 1055 | return NETDEV_TX_OK; |
1da177e4 LT |
1056 | |
1057 | drop: | |
497a5757 | 1058 | atomic_long_inc(&dev->tx_dropped); |
149d36f7 | 1059 | skb_tx_error(skb); |
1da177e4 | 1060 | kfree_skb(skb); |
6e914fc7 | 1061 | rcu_read_unlock(); |
baeababb | 1062 | return NET_XMIT_DROP; |
1da177e4 LT |
1063 | } |
1064 | ||
f271b2cc | 1065 | static void tun_net_mclist(struct net_device *dev) |
1da177e4 | 1066 | { |
f271b2cc MK |
1067 | /* |
1068 | * This callback is supposed to deal with mc filter in | |
1069 | * _rx_ path and has nothing to do with the _tx_ path. | |
1070 | * In rx path we always accept everything userspace gives us. | |
1071 | */ | |
1da177e4 LT |
1072 | } |
1073 | ||
c8f44aff MM |
1074 | static netdev_features_t tun_net_fix_features(struct net_device *dev, |
1075 | netdev_features_t features) | |
88255375 MM |
1076 | { |
1077 | struct tun_struct *tun = netdev_priv(dev); | |
1078 | ||
1079 | return (features & tun->set_features) | (features & ~TUN_USER_FEATURES); | |
1080 | } | |
eaea34b2 PA |
1081 | |
1082 | static void tun_set_headroom(struct net_device *dev, int new_hr) | |
1083 | { | |
1084 | struct tun_struct *tun = netdev_priv(dev); | |
1085 | ||
1086 | if (new_hr < NET_SKB_PAD) | |
1087 | new_hr = NET_SKB_PAD; | |
1088 | ||
1089 | tun->align = new_hr; | |
1090 | } | |
1091 | ||
bc1f4470 | 1092 | static void |
608b9977 PA |
1093 | tun_net_get_stats64(struct net_device *dev, struct rtnl_link_stats64 *stats) |
1094 | { | |
608b9977 | 1095 | struct tun_struct *tun = netdev_priv(dev); |
608b9977 | 1096 | |
497a5757 | 1097 | dev_get_tstats64(dev, stats); |
608b9977 | 1098 | |
497a5757 HK |
1099 | stats->rx_frame_errors += |
1100 | (unsigned long)atomic_long_read(&tun->rx_frame_errors); | |
608b9977 PA |
1101 | } |
1102 | ||
761876c8 JW |
1103 | static int tun_xdp_set(struct net_device *dev, struct bpf_prog *prog, |
1104 | struct netlink_ext_ack *extack) | |
1105 | { | |
1106 | struct tun_struct *tun = netdev_priv(dev); | |
e4a2a304 | 1107 | struct tun_file *tfile; |
761876c8 | 1108 | struct bpf_prog *old_prog; |
e4a2a304 | 1109 | int i; |
761876c8 JW |
1110 | |
1111 | old_prog = rtnl_dereference(tun->xdp_prog); | |
1112 | rcu_assign_pointer(tun->xdp_prog, prog); | |
1113 | if (old_prog) | |
1114 | bpf_prog_put(old_prog); | |
1115 | ||
e4a2a304 JW |
1116 | for (i = 0; i < tun->numqueues; i++) { |
1117 | tfile = rtnl_dereference(tun->tfiles[i]); | |
1118 | if (prog) | |
1119 | sock_set_flag(&tfile->sk, SOCK_XDP); | |
1120 | else | |
1121 | sock_reset_flag(&tfile->sk, SOCK_XDP); | |
1122 | } | |
1123 | list_for_each_entry(tfile, &tun->disabled, next) { | |
1124 | if (prog) | |
1125 | sock_set_flag(&tfile->sk, SOCK_XDP); | |
1126 | else | |
1127 | sock_reset_flag(&tfile->sk, SOCK_XDP); | |
1128 | } | |
1129 | ||
761876c8 JW |
1130 | return 0; |
1131 | } | |
1132 | ||
f4e63525 | 1133 | static int tun_xdp(struct net_device *dev, struct netdev_bpf *xdp) |
761876c8 JW |
1134 | { |
1135 | switch (xdp->command) { | |
1136 | case XDP_SETUP_PROG: | |
1137 | return tun_xdp_set(dev, xdp->prog, xdp->extack); | |
761876c8 JW |
1138 | default: |
1139 | return -EINVAL; | |
1140 | } | |
1141 | } | |
1142 | ||
26d31925 ND |
1143 | static int tun_net_change_carrier(struct net_device *dev, bool new_carrier) |
1144 | { | |
1145 | if (new_carrier) { | |
1146 | struct tun_struct *tun = netdev_priv(dev); | |
1147 | ||
1148 | if (!tun->numqueues) | |
1149 | return -EPERM; | |
1150 | ||
1151 | netif_carrier_on(dev); | |
1152 | } else { | |
1153 | netif_carrier_off(dev); | |
1154 | } | |
1155 | return 0; | |
1156 | } | |
1157 | ||
758e43b7 | 1158 | static const struct net_device_ops tun_netdev_ops = { |
c70f1829 | 1159 | .ndo_uninit = tun_net_uninit, |
758e43b7 SH |
1160 | .ndo_open = tun_net_open, |
1161 | .ndo_stop = tun_net_close, | |
00829823 | 1162 | .ndo_start_xmit = tun_net_xmit, |
88255375 | 1163 | .ndo_fix_features = tun_net_fix_features, |
c8d68e6b | 1164 | .ndo_select_queue = tun_select_queue, |
eaea34b2 | 1165 | .ndo_set_rx_headroom = tun_set_headroom, |
608b9977 | 1166 | .ndo_get_stats64 = tun_net_get_stats64, |
26d31925 | 1167 | .ndo_change_carrier = tun_net_change_carrier, |
758e43b7 SH |
1168 | }; |
1169 | ||
0c9d917b JDB |
1170 | static void __tun_xdp_flush_tfile(struct tun_file *tfile) |
1171 | { | |
1172 | /* Notify and wake up reader process */ | |
1173 | if (tfile->flags & TUN_FASYNC) | |
1174 | kill_fasync(&tfile->fasync, SIGIO, POLL_IN); | |
1175 | tfile->socket.sk->sk_data_ready(tfile->socket.sk); | |
1176 | } | |
1177 | ||
42b33468 JDB |
1178 | static int tun_xdp_xmit(struct net_device *dev, int n, |
1179 | struct xdp_frame **frames, u32 flags) | |
fc72d1d5 JW |
1180 | { |
1181 | struct tun_struct *tun = netdev_priv(dev); | |
fc72d1d5 JW |
1182 | struct tun_file *tfile; |
1183 | u32 numqueues; | |
735fc405 JDB |
1184 | int drops = 0; |
1185 | int cnt = n; | |
1186 | int i; | |
fc72d1d5 | 1187 | |
0c9d917b | 1188 | if (unlikely(flags & ~XDP_XMIT_FLAGS_MASK)) |
42b33468 JDB |
1189 | return -EINVAL; |
1190 | ||
fc72d1d5 JW |
1191 | rcu_read_lock(); |
1192 | ||
9871a9e4 | 1193 | resample: |
fc72d1d5 JW |
1194 | numqueues = READ_ONCE(tun->numqueues); |
1195 | if (!numqueues) { | |
735fc405 JDB |
1196 | rcu_read_unlock(); |
1197 | return -ENXIO; /* Caller will free/return all frames */ | |
fc72d1d5 JW |
1198 | } |
1199 | ||
1200 | tfile = rcu_dereference(tun->tfiles[smp_processor_id() % | |
1201 | numqueues]); | |
9871a9e4 JW |
1202 | if (unlikely(!tfile)) |
1203 | goto resample; | |
735fc405 JDB |
1204 | |
1205 | spin_lock(&tfile->tx_ring.producer_lock); | |
1206 | for (i = 0; i < n; i++) { | |
1207 | struct xdp_frame *xdp = frames[i]; | |
1208 | /* Encode the XDP flag into lowest bit for consumer to differ | |
1209 | * XDP buffer from sk_buff. | |
1210 | */ | |
1211 | void *frame = tun_xdp_to_ptr(xdp); | |
1212 | ||
1213 | if (__ptr_ring_produce(&tfile->tx_ring, frame)) { | |
497a5757 | 1214 | atomic_long_inc(&dev->tx_dropped); |
735fc405 JDB |
1215 | xdp_return_frame_rx_napi(xdp); |
1216 | drops++; | |
1217 | } | |
fc72d1d5 | 1218 | } |
735fc405 | 1219 | spin_unlock(&tfile->tx_ring.producer_lock); |
fc72d1d5 | 1220 | |
0c9d917b JDB |
1221 | if (flags & XDP_XMIT_FLUSH) |
1222 | __tun_xdp_flush_tfile(tfile); | |
1223 | ||
fc72d1d5 | 1224 | rcu_read_unlock(); |
735fc405 | 1225 | return cnt - drops; |
fc72d1d5 JW |
1226 | } |
1227 | ||
44fa2dbd JDB |
1228 | static int tun_xdp_tx(struct net_device *dev, struct xdp_buff *xdp) |
1229 | { | |
1b698fa5 | 1230 | struct xdp_frame *frame = xdp_convert_buff_to_frame(xdp); |
44fa2dbd JDB |
1231 | |
1232 | if (unlikely(!frame)) | |
1233 | return -EOVERFLOW; | |
1234 | ||
42421a56 | 1235 | return tun_xdp_xmit(dev, 1, &frame, XDP_XMIT_FLUSH); |
fc72d1d5 JW |
1236 | } |
1237 | ||
758e43b7 | 1238 | static const struct net_device_ops tap_netdev_ops = { |
c70f1829 | 1239 | .ndo_uninit = tun_net_uninit, |
758e43b7 SH |
1240 | .ndo_open = tun_net_open, |
1241 | .ndo_stop = tun_net_close, | |
00829823 | 1242 | .ndo_start_xmit = tun_net_xmit, |
88255375 | 1243 | .ndo_fix_features = tun_net_fix_features, |
afc4b13d | 1244 | .ndo_set_rx_mode = tun_net_mclist, |
758e43b7 SH |
1245 | .ndo_set_mac_address = eth_mac_addr, |
1246 | .ndo_validate_addr = eth_validate_addr, | |
c8d68e6b | 1247 | .ndo_select_queue = tun_select_queue, |
5e52796a | 1248 | .ndo_features_check = passthru_features_check, |
eaea34b2 | 1249 | .ndo_set_rx_headroom = tun_set_headroom, |
497a5757 | 1250 | .ndo_get_stats64 = dev_get_tstats64, |
f4e63525 | 1251 | .ndo_bpf = tun_xdp, |
fc72d1d5 | 1252 | .ndo_xdp_xmit = tun_xdp_xmit, |
26d31925 | 1253 | .ndo_change_carrier = tun_net_change_carrier, |
758e43b7 SH |
1254 | }; |
1255 | ||
944a1376 | 1256 | static void tun_flow_init(struct tun_struct *tun) |
96442e42 JW |
1257 | { |
1258 | int i; | |
1259 | ||
96442e42 JW |
1260 | for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++) |
1261 | INIT_HLIST_HEAD(&tun->flows[i]); | |
1262 | ||
1263 | tun->ageing_time = TUN_FLOW_EXPIRE; | |
e99e88a9 KC |
1264 | timer_setup(&tun->flow_gc_timer, tun_flow_cleanup, 0); |
1265 | mod_timer(&tun->flow_gc_timer, | |
1266 | round_jiffies_up(jiffies + tun->ageing_time)); | |
96442e42 JW |
1267 | } |
1268 | ||
1269 | static void tun_flow_uninit(struct tun_struct *tun) | |
1270 | { | |
1271 | del_timer_sync(&tun->flow_gc_timer); | |
1272 | tun_flow_flush(tun); | |
96442e42 JW |
1273 | } |
1274 | ||
91572088 JW |
1275 | #define MIN_MTU 68 |
1276 | #define MAX_MTU 65535 | |
1277 | ||
1da177e4 LT |
1278 | /* Initialize net device. */ |
1279 | static void tun_net_init(struct net_device *dev) | |
1280 | { | |
1281 | struct tun_struct *tun = netdev_priv(dev); | |
6aa20a22 | 1282 | |
1da177e4 | 1283 | switch (tun->flags & TUN_TYPE_MASK) { |
40630b82 | 1284 | case IFF_TUN: |
758e43b7 | 1285 | dev->netdev_ops = &tun_netdev_ops; |
b9815eb1 | 1286 | dev->header_ops = &ip_tunnel_header_ops; |
758e43b7 | 1287 | |
1da177e4 LT |
1288 | /* Point-to-Point TUN Device */ |
1289 | dev->hard_header_len = 0; | |
1290 | dev->addr_len = 0; | |
1291 | dev->mtu = 1500; | |
1292 | ||
1293 | /* Zero header length */ | |
6aa20a22 | 1294 | dev->type = ARPHRD_NONE; |
1da177e4 | 1295 | dev->flags = IFF_POINTOPOINT | IFF_NOARP | IFF_MULTICAST; |
1da177e4 LT |
1296 | break; |
1297 | ||
40630b82 | 1298 | case IFF_TAP: |
7a0a9608 | 1299 | dev->netdev_ops = &tap_netdev_ops; |
1da177e4 | 1300 | /* Ethernet TAP Device */ |
1da177e4 | 1301 | ether_setup(dev); |
550fd08c | 1302 | dev->priv_flags &= ~IFF_TX_SKB_SHARING; |
a676847b | 1303 | dev->priv_flags |= IFF_LIVE_ADDR_CHANGE; |
36226a8d | 1304 | |
f2cedb63 | 1305 | eth_hw_addr_random(dev); |
36226a8d | 1306 | |
1da177e4 LT |
1307 | break; |
1308 | } | |
91572088 JW |
1309 | |
1310 | dev->min_mtu = MIN_MTU; | |
1311 | dev->max_mtu = MAX_MTU - dev->hard_header_len; | |
1da177e4 LT |
1312 | } |
1313 | ||
2f3ab622 JW |
1314 | static bool tun_sock_writeable(struct tun_struct *tun, struct tun_file *tfile) |
1315 | { | |
1316 | struct sock *sk = tfile->socket.sk; | |
1317 | ||
1318 | return (tun->dev->flags & IFF_UP) && sock_writeable(sk); | |
1319 | } | |
1320 | ||
1da177e4 LT |
1321 | /* Character device part */ |
1322 | ||
1323 | /* Poll */ | |
afc9a42b | 1324 | static __poll_t tun_chr_poll(struct file *file, poll_table *wait) |
6aa20a22 | 1325 | { |
b2430de3 | 1326 | struct tun_file *tfile = file->private_data; |
9484dc74 | 1327 | struct tun_struct *tun = tun_get(tfile); |
3c8a9c63 | 1328 | struct sock *sk; |
afc9a42b | 1329 | __poll_t mask = 0; |
1da177e4 LT |
1330 | |
1331 | if (!tun) | |
a9a08845 | 1332 | return EPOLLERR; |
1da177e4 | 1333 | |
54f968d6 | 1334 | sk = tfile->socket.sk; |
3c8a9c63 | 1335 | |
9e641bdc | 1336 | poll_wait(file, sk_sleep(sk), wait); |
6aa20a22 | 1337 | |
5990a305 | 1338 | if (!ptr_ring_empty(&tfile->tx_ring)) |
a9a08845 | 1339 | mask |= EPOLLIN | EPOLLRDNORM; |
1da177e4 | 1340 | |
2f3ab622 JW |
1341 | /* Make sure SOCKWQ_ASYNC_NOSPACE is set if not writable to |
1342 | * guarantee EPOLLOUT to be raised by either here or | |
1343 | * tun_sock_write_space(). Then process could get notification | |
1344 | * after it writes to a down device and meets -EIO. | |
1345 | */ | |
1346 | if (tun_sock_writeable(tun, tfile) || | |
1347 | (!test_and_set_bit(SOCKWQ_ASYNC_NOSPACE, &sk->sk_socket->flags) && | |
1348 | tun_sock_writeable(tun, tfile))) | |
a9a08845 | 1349 | mask |= EPOLLOUT | EPOLLWRNORM; |
33dccbb0 | 1350 | |
c70f1829 | 1351 | if (tun->dev->reg_state != NETREG_REGISTERED) |
a9a08845 | 1352 | mask = EPOLLERR; |
c70f1829 | 1353 | |
631ab46b | 1354 | tun_put(tun); |
1da177e4 LT |
1355 | return mask; |
1356 | } | |
1357 | ||
90e33d45 PP |
1358 | static struct sk_buff *tun_napi_alloc_frags(struct tun_file *tfile, |
1359 | size_t len, | |
1360 | const struct iov_iter *it) | |
1361 | { | |
1362 | struct sk_buff *skb; | |
1363 | size_t linear; | |
1364 | int err; | |
1365 | int i; | |
1366 | ||
1367 | if (it->nr_segs > MAX_SKB_FRAGS + 1) | |
950271d7 | 1368 | return ERR_PTR(-EMSGSIZE); |
90e33d45 PP |
1369 | |
1370 | local_bh_disable(); | |
1371 | skb = napi_get_frags(&tfile->napi); | |
1372 | local_bh_enable(); | |
1373 | if (!skb) | |
1374 | return ERR_PTR(-ENOMEM); | |
1375 | ||
1376 | linear = iov_iter_single_seg_count(it); | |
1377 | err = __skb_grow(skb, linear); | |
1378 | if (err) | |
1379 | goto free; | |
1380 | ||
1381 | skb->len = len; | |
1382 | skb->data_len = len - linear; | |
1383 | skb->truesize += skb->data_len; | |
1384 | ||
1385 | for (i = 1; i < it->nr_segs; i++) { | |
1386 | size_t fragsz = it->iov[i].iov_len; | |
aa6daaca ED |
1387 | struct page *page; |
1388 | void *frag; | |
90e33d45 PP |
1389 | |
1390 | if (fragsz == 0 || fragsz > PAGE_SIZE) { | |
1391 | err = -EINVAL; | |
1392 | goto free; | |
1393 | } | |
aa6daaca ED |
1394 | frag = netdev_alloc_frag(fragsz); |
1395 | if (!frag) { | |
90e33d45 PP |
1396 | err = -ENOMEM; |
1397 | goto free; | |
1398 | } | |
aa6daaca ED |
1399 | page = virt_to_head_page(frag); |
1400 | skb_fill_page_desc(skb, i - 1, page, | |
1401 | frag - page_address(page), fragsz); | |
90e33d45 PP |
1402 | } |
1403 | ||
1404 | return skb; | |
1405 | free: | |
1406 | /* frees skb and all frags allocated with napi_alloc_frag() */ | |
1407 | napi_free_frags(&tfile->napi); | |
1408 | return ERR_PTR(err); | |
1409 | } | |
1410 | ||
f42157cb RR |
1411 | /* prepad is the amount to reserve at front. len is length after that. |
1412 | * linear is a hint as to how much to copy (usually headers). */ | |
54f968d6 | 1413 | static struct sk_buff *tun_alloc_skb(struct tun_file *tfile, |
6f7c156c | 1414 | size_t prepad, size_t len, |
1415 | size_t linear, int noblock) | |
f42157cb | 1416 | { |
54f968d6 | 1417 | struct sock *sk = tfile->socket.sk; |
f42157cb | 1418 | struct sk_buff *skb; |
33dccbb0 | 1419 | int err; |
f42157cb RR |
1420 | |
1421 | /* Under a page? Don't bother with paged skb. */ | |
0eca93bc | 1422 | if (prepad + len < PAGE_SIZE || !linear) |
33dccbb0 | 1423 | linear = len; |
f42157cb | 1424 | |
33dccbb0 | 1425 | skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock, |
28d64271 | 1426 | &err, 0); |
f42157cb | 1427 | if (!skb) |
33dccbb0 | 1428 | return ERR_PTR(err); |
f42157cb RR |
1429 | |
1430 | skb_reserve(skb, prepad); | |
1431 | skb_put(skb, linear); | |
33dccbb0 HX |
1432 | skb->data_len = len - linear; |
1433 | skb->len += len - linear; | |
f42157cb RR |
1434 | |
1435 | return skb; | |
1436 | } | |
1437 | ||
5503fcec JW |
1438 | static void tun_rx_batched(struct tun_struct *tun, struct tun_file *tfile, |
1439 | struct sk_buff *skb, int more) | |
1440 | { | |
1441 | struct sk_buff_head *queue = &tfile->sk.sk_write_queue; | |
1442 | struct sk_buff_head process_queue; | |
1443 | u32 rx_batched = tun->rx_batched; | |
1444 | bool rcv = false; | |
1445 | ||
1446 | if (!rx_batched || (!more && skb_queue_empty(queue))) { | |
1447 | local_bh_disable(); | |
8ebebcba | 1448 | skb_record_rx_queue(skb, tfile->queue_index); |
5503fcec JW |
1449 | netif_receive_skb(skb); |
1450 | local_bh_enable(); | |
1451 | return; | |
1452 | } | |
1453 | ||
1454 | spin_lock(&queue->lock); | |
1455 | if (!more || skb_queue_len(queue) == rx_batched) { | |
1456 | __skb_queue_head_init(&process_queue); | |
1457 | skb_queue_splice_tail_init(queue, &process_queue); | |
1458 | rcv = true; | |
1459 | } else { | |
1460 | __skb_queue_tail(queue, skb); | |
1461 | } | |
1462 | spin_unlock(&queue->lock); | |
1463 | ||
1464 | if (rcv) { | |
1465 | struct sk_buff *nskb; | |
1466 | ||
1467 | local_bh_disable(); | |
8ebebcba MC |
1468 | while ((nskb = __skb_dequeue(&process_queue))) { |
1469 | skb_record_rx_queue(nskb, tfile->queue_index); | |
5503fcec | 1470 | netif_receive_skb(nskb); |
8ebebcba MC |
1471 | } |
1472 | skb_record_rx_queue(skb, tfile->queue_index); | |
5503fcec JW |
1473 | netif_receive_skb(skb); |
1474 | local_bh_enable(); | |
1475 | } | |
1476 | } | |
1477 | ||
66ccbc9c JW |
1478 | static bool tun_can_build_skb(struct tun_struct *tun, struct tun_file *tfile, |
1479 | int len, int noblock, bool zerocopy) | |
1480 | { | |
1481 | if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP) | |
1482 | return false; | |
1483 | ||
1484 | if (tfile->socket.sk->sk_sndbuf != INT_MAX) | |
1485 | return false; | |
1486 | ||
1487 | if (!noblock) | |
1488 | return false; | |
1489 | ||
1490 | if (zerocopy) | |
1491 | return false; | |
1492 | ||
1493 | if (SKB_DATA_ALIGN(len + TUN_RX_PAD) + | |
1494 | SKB_DATA_ALIGN(sizeof(struct skb_shared_info)) > PAGE_SIZE) | |
1495 | return false; | |
1496 | ||
1497 | return true; | |
1498 | } | |
1499 | ||
4b663366 AB |
1500 | static struct sk_buff *__tun_build_skb(struct tun_file *tfile, |
1501 | struct page_frag *alloc_frag, char *buf, | |
8ae1aff0 | 1502 | int buflen, int len, int pad) |
ac1f1f6c JW |
1503 | { |
1504 | struct sk_buff *skb = build_skb(buf, buflen); | |
1505 | ||
1506 | if (!skb) | |
1507 | return ERR_PTR(-ENOMEM); | |
1508 | ||
8ae1aff0 | 1509 | skb_reserve(skb, pad); |
ac1f1f6c | 1510 | skb_put(skb, len); |
4b663366 | 1511 | skb_set_owner_w(skb, tfile->socket.sk); |
ac1f1f6c JW |
1512 | |
1513 | get_page(alloc_frag->page); | |
1514 | alloc_frag->offset += buflen; | |
1515 | ||
1516 | return skb; | |
1517 | } | |
1518 | ||
8ae1aff0 JW |
1519 | static int tun_xdp_act(struct tun_struct *tun, struct bpf_prog *xdp_prog, |
1520 | struct xdp_buff *xdp, u32 act) | |
1521 | { | |
1522 | int err; | |
1523 | ||
1524 | switch (act) { | |
1525 | case XDP_REDIRECT: | |
1526 | err = xdp_do_redirect(tun->dev, xdp, xdp_prog); | |
8ae1aff0 JW |
1527 | if (err) |
1528 | return err; | |
1529 | break; | |
1530 | case XDP_TX: | |
1531 | err = tun_xdp_tx(tun->dev, xdp); | |
1532 | if (err < 0) | |
1533 | return err; | |
1534 | break; | |
1535 | case XDP_PASS: | |
1536 | break; | |
1537 | default: | |
1538 | bpf_warn_invalid_xdp_action(act); | |
df561f66 | 1539 | fallthrough; |
8ae1aff0 JW |
1540 | case XDP_ABORTED: |
1541 | trace_xdp_exception(tun->dev, xdp_prog, act); | |
df561f66 | 1542 | fallthrough; |
8ae1aff0 | 1543 | case XDP_DROP: |
497a5757 | 1544 | atomic_long_inc(&tun->dev->rx_dropped); |
8ae1aff0 JW |
1545 | break; |
1546 | } | |
1547 | ||
1548 | return act; | |
1549 | } | |
1550 | ||
761876c8 JW |
1551 | static struct sk_buff *tun_build_skb(struct tun_struct *tun, |
1552 | struct tun_file *tfile, | |
66ccbc9c | 1553 | struct iov_iter *from, |
761876c8 | 1554 | struct virtio_net_hdr *hdr, |
1cfe6e93 | 1555 | int len, int *skb_xdp) |
66ccbc9c | 1556 | { |
0bbd7dad | 1557 | struct page_frag *alloc_frag = ¤t->task_frag; |
761876c8 | 1558 | struct bpf_prog *xdp_prog; |
7df13219 | 1559 | int buflen = SKB_DATA_ALIGN(sizeof(struct skb_shared_info)); |
66ccbc9c JW |
1560 | char *buf; |
1561 | size_t copied; | |
8ae1aff0 JW |
1562 | int pad = TUN_RX_PAD; |
1563 | int err = 0; | |
7df13219 JW |
1564 | |
1565 | rcu_read_lock(); | |
1566 | xdp_prog = rcu_dereference(tun->xdp_prog); | |
1567 | if (xdp_prog) | |
4f23aff8 | 1568 | pad += XDP_PACKET_HEADROOM; |
7df13219 JW |
1569 | buflen += SKB_DATA_ALIGN(len + pad); |
1570 | rcu_read_unlock(); | |
66ccbc9c | 1571 | |
63b9ab65 | 1572 | alloc_frag->offset = ALIGN((u64)alloc_frag->offset, SMP_CACHE_BYTES); |
66ccbc9c JW |
1573 | if (unlikely(!skb_page_frag_refill(buflen, alloc_frag, GFP_KERNEL))) |
1574 | return ERR_PTR(-ENOMEM); | |
1575 | ||
1576 | buf = (char *)page_address(alloc_frag->page) + alloc_frag->offset; | |
1577 | copied = copy_page_from_iter(alloc_frag->page, | |
7df13219 | 1578 | alloc_frag->offset + pad, |
66ccbc9c JW |
1579 | len, from); |
1580 | if (copied != len) | |
1581 | return ERR_PTR(-EFAULT); | |
1582 | ||
7df13219 JW |
1583 | /* There's a small window that XDP may be set after the check |
1584 | * of xdp_prog above, this should be rare and for simplicity | |
1585 | * we do XDP on skb in case the headroom is not enough. | |
1586 | */ | |
ac1f1f6c | 1587 | if (hdr->gso_type || !xdp_prog) { |
1cfe6e93 | 1588 | *skb_xdp = 1; |
4b663366 AB |
1589 | return __tun_build_skb(tfile, alloc_frag, buf, buflen, len, |
1590 | pad); | |
ac1f1f6c JW |
1591 | } |
1592 | ||
1593 | *skb_xdp = 0; | |
761876c8 | 1594 | |
6547e387 | 1595 | local_bh_disable(); |
761876c8 JW |
1596 | rcu_read_lock(); |
1597 | xdp_prog = rcu_dereference(tun->xdp_prog); | |
8ae1aff0 | 1598 | if (xdp_prog) { |
761876c8 | 1599 | struct xdp_buff xdp; |
761876c8 JW |
1600 | u32 act; |
1601 | ||
43b5169d | 1602 | xdp_init_buff(&xdp, buflen, &tfile->xdp_rxq); |
761876c8 | 1603 | xdp.data_hard_start = buf; |
7df13219 | 1604 | xdp.data = buf + pad; |
de8f3a83 | 1605 | xdp_set_data_meta_invalid(&xdp); |
761876c8 | 1606 | xdp.data_end = xdp.data + len; |
761876c8 | 1607 | |
8ae1aff0 JW |
1608 | act = bpf_prog_run_xdp(xdp_prog, &xdp); |
1609 | if (act == XDP_REDIRECT || act == XDP_TX) { | |
59655a5b JW |
1610 | get_page(alloc_frag->page); |
1611 | alloc_frag->offset += buflen; | |
761876c8 | 1612 | } |
8ae1aff0 | 1613 | err = tun_xdp_act(tun, xdp_prog, &xdp, act); |
bee34890 WD |
1614 | if (err < 0) { |
1615 | if (act == XDP_REDIRECT || act == XDP_TX) | |
1616 | put_page(alloc_frag->page); | |
1617 | goto out; | |
1618 | } | |
1619 | ||
1a097910 | 1620 | if (err == XDP_REDIRECT) |
1d233886 | 1621 | xdp_do_flush(); |
8ae1aff0 JW |
1622 | if (err != XDP_PASS) |
1623 | goto out; | |
1624 | ||
1625 | pad = xdp.data - xdp.data_hard_start; | |
1626 | len = xdp.data_end - xdp.data; | |
761876c8 | 1627 | } |
291aeb2b JW |
1628 | rcu_read_unlock(); |
1629 | local_bh_enable(); | |
761876c8 | 1630 | |
4b663366 | 1631 | return __tun_build_skb(tfile, alloc_frag, buf, buflen, len, pad); |
761876c8 | 1632 | |
f7053b6c | 1633 | out: |
761876c8 | 1634 | rcu_read_unlock(); |
6547e387 | 1635 | local_bh_enable(); |
761876c8 | 1636 | return NULL; |
66ccbc9c JW |
1637 | } |
1638 | ||
1da177e4 | 1639 | /* Get packet from user space buffer */ |
54f968d6 | 1640 | static ssize_t tun_get_user(struct tun_struct *tun, struct tun_file *tfile, |
f5ff53b4 | 1641 | void *msg_control, struct iov_iter *from, |
5503fcec | 1642 | int noblock, bool more) |
1da177e4 | 1643 | { |
09640e63 | 1644 | struct tun_pi pi = { 0, cpu_to_be16(ETH_P_IP) }; |
1da177e4 | 1645 | struct sk_buff *skb; |
f5ff53b4 | 1646 | size_t total_len = iov_iter_count(from); |
eaea34b2 | 1647 | size_t len = total_len, align = tun->align, linear; |
f43798c2 | 1648 | struct virtio_net_hdr gso = { 0 }; |
96f8d9ec | 1649 | int good_linear; |
0690899b MT |
1650 | int copylen; |
1651 | bool zerocopy = false; | |
1652 | int err; | |
96f84061 | 1653 | u32 rxhash = 0; |
1cfe6e93 | 1654 | int skb_xdp = 1; |
af3fb24e | 1655 | bool frags = tun_napi_frags_enabled(tfile); |
1da177e4 | 1656 | |
40630b82 | 1657 | if (!(tun->flags & IFF_NO_PI)) { |
15718ea0 | 1658 | if (len < sizeof(pi)) |
1da177e4 | 1659 | return -EINVAL; |
15718ea0 | 1660 | len -= sizeof(pi); |
1da177e4 | 1661 | |
cbbd26b8 | 1662 | if (!copy_from_iter_full(&pi, sizeof(pi), from)) |
1da177e4 LT |
1663 | return -EFAULT; |
1664 | } | |
1665 | ||
40630b82 | 1666 | if (tun->flags & IFF_VNET_HDR) { |
e1edab87 WB |
1667 | int vnet_hdr_sz = READ_ONCE(tun->vnet_hdr_sz); |
1668 | ||
1669 | if (len < vnet_hdr_sz) | |
f43798c2 | 1670 | return -EINVAL; |
e1edab87 | 1671 | len -= vnet_hdr_sz; |
f43798c2 | 1672 | |
cbbd26b8 | 1673 | if (!copy_from_iter_full(&gso, sizeof(gso), from)) |
f43798c2 RR |
1674 | return -EFAULT; |
1675 | ||
4909122f | 1676 | if ((gso.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) && |
56f0dcc5 MT |
1677 | tun16_to_cpu(tun, gso.csum_start) + tun16_to_cpu(tun, gso.csum_offset) + 2 > tun16_to_cpu(tun, gso.hdr_len)) |
1678 | gso.hdr_len = cpu_to_tun16(tun, tun16_to_cpu(tun, gso.csum_start) + tun16_to_cpu(tun, gso.csum_offset) + 2); | |
4909122f | 1679 | |
56f0dcc5 | 1680 | if (tun16_to_cpu(tun, gso.hdr_len) > len) |
f43798c2 | 1681 | return -EINVAL; |
e1edab87 | 1682 | iov_iter_advance(from, vnet_hdr_sz - sizeof(gso)); |
f43798c2 RR |
1683 | } |
1684 | ||
40630b82 | 1685 | if ((tun->flags & TUN_TYPE_MASK) == IFF_TAP) { |
a504b86e | 1686 | align += NET_IP_ALIGN; |
0eca93bc | 1687 | if (unlikely(len < ETH_HLEN || |
56f0dcc5 | 1688 | (gso.hdr_len && tun16_to_cpu(tun, gso.hdr_len) < ETH_HLEN))) |
e01bf1c8 RR |
1689 | return -EINVAL; |
1690 | } | |
6aa20a22 | 1691 | |
96f8d9ec JW |
1692 | good_linear = SKB_MAX_HEAD(align); |
1693 | ||
88529176 | 1694 | if (msg_control) { |
f5ff53b4 AV |
1695 | struct iov_iter i = *from; |
1696 | ||
88529176 JW |
1697 | /* There are 256 bytes to be copied in skb, so there is |
1698 | * enough room for skb expand head in case it is used. | |
0690899b MT |
1699 | * The rest of the buffer is mapped from userspace. |
1700 | */ | |
56f0dcc5 | 1701 | copylen = gso.hdr_len ? tun16_to_cpu(tun, gso.hdr_len) : GOODCOPY_LEN; |
96f8d9ec JW |
1702 | if (copylen > good_linear) |
1703 | copylen = good_linear; | |
3dd5c330 | 1704 | linear = copylen; |
f5ff53b4 AV |
1705 | iov_iter_advance(&i, copylen); |
1706 | if (iov_iter_npages(&i, INT_MAX) <= MAX_SKB_FRAGS) | |
88529176 JW |
1707 | zerocopy = true; |
1708 | } | |
1709 | ||
90e33d45 | 1710 | if (!frags && tun_can_build_skb(tun, tfile, len, noblock, zerocopy)) { |
1cfe6e93 JW |
1711 | /* For the packet that is not easy to be processed |
1712 | * (e.g gso or jumbo packet), we will do it at after | |
1713 | * skb was created with generic XDP routine. | |
1714 | */ | |
1715 | skb = tun_build_skb(tun, tfile, from, &gso, len, &skb_xdp); | |
66ccbc9c | 1716 | if (IS_ERR(skb)) { |
497a5757 | 1717 | atomic_long_inc(&tun->dev->rx_dropped); |
66ccbc9c JW |
1718 | return PTR_ERR(skb); |
1719 | } | |
761876c8 JW |
1720 | if (!skb) |
1721 | return total_len; | |
66ccbc9c JW |
1722 | } else { |
1723 | if (!zerocopy) { | |
1724 | copylen = len; | |
1725 | if (tun16_to_cpu(tun, gso.hdr_len) > good_linear) | |
1726 | linear = good_linear; | |
1727 | else | |
1728 | linear = tun16_to_cpu(tun, gso.hdr_len); | |
1729 | } | |
1da177e4 | 1730 | |
90e33d45 PP |
1731 | if (frags) { |
1732 | mutex_lock(&tfile->napi_mutex); | |
1733 | skb = tun_napi_alloc_frags(tfile, copylen, from); | |
1734 | /* tun_napi_alloc_frags() enforces a layout for the skb. | |
1735 | * If zerocopy is enabled, then this layout will be | |
1736 | * overwritten by zerocopy_sg_from_iter(). | |
1737 | */ | |
1738 | zerocopy = false; | |
1739 | } else { | |
1740 | skb = tun_alloc_skb(tfile, align, copylen, linear, | |
1741 | noblock); | |
1742 | } | |
1743 | ||
66ccbc9c JW |
1744 | if (IS_ERR(skb)) { |
1745 | if (PTR_ERR(skb) != -EAGAIN) | |
497a5757 | 1746 | atomic_long_inc(&tun->dev->rx_dropped); |
90e33d45 PP |
1747 | if (frags) |
1748 | mutex_unlock(&tfile->napi_mutex); | |
66ccbc9c JW |
1749 | return PTR_ERR(skb); |
1750 | } | |
0690899b | 1751 | |
66ccbc9c JW |
1752 | if (zerocopy) |
1753 | err = zerocopy_sg_from_iter(skb, from); | |
1754 | else | |
1755 | err = skb_copy_datagram_from_iter(skb, 0, from, len); | |
1756 | ||
1757 | if (err) { | |
4477138f ED |
1758 | err = -EFAULT; |
1759 | drop: | |
497a5757 | 1760 | atomic_long_inc(&tun->dev->rx_dropped); |
66ccbc9c | 1761 | kfree_skb(skb); |
90e33d45 PP |
1762 | if (frags) { |
1763 | tfile->napi.skb = NULL; | |
1764 | mutex_unlock(&tfile->napi_mutex); | |
1765 | } | |
1766 | ||
4477138f | 1767 | return err; |
66ccbc9c | 1768 | } |
8f22757e | 1769 | } |
1da177e4 | 1770 | |
3e9e40e7 | 1771 | if (virtio_net_hdr_to_skb(skb, &gso, tun_is_little_endian(tun))) { |
497a5757 | 1772 | atomic_long_inc(&tun->rx_frame_errors); |
df10db98 | 1773 | kfree_skb(skb); |
90e33d45 PP |
1774 | if (frags) { |
1775 | tfile->napi.skb = NULL; | |
1776 | mutex_unlock(&tfile->napi_mutex); | |
1777 | } | |
1778 | ||
df10db98 PA |
1779 | return -EINVAL; |
1780 | } | |
1781 | ||
1da177e4 | 1782 | switch (tun->flags & TUN_TYPE_MASK) { |
40630b82 MT |
1783 | case IFF_TUN: |
1784 | if (tun->flags & IFF_NO_PI) { | |
2580c4c1 AP |
1785 | u8 ip_version = skb->len ? (skb->data[0] >> 4) : 0; |
1786 | ||
1787 | switch (ip_version) { | |
1788 | case 4: | |
f09f7ee2 AWC |
1789 | pi.proto = htons(ETH_P_IP); |
1790 | break; | |
2580c4c1 | 1791 | case 6: |
f09f7ee2 AWC |
1792 | pi.proto = htons(ETH_P_IPV6); |
1793 | break; | |
1794 | default: | |
497a5757 | 1795 | atomic_long_inc(&tun->dev->rx_dropped); |
f09f7ee2 AWC |
1796 | kfree_skb(skb); |
1797 | return -EINVAL; | |
1798 | } | |
1799 | } | |
1800 | ||
459a98ed | 1801 | skb_reset_mac_header(skb); |
1da177e4 | 1802 | skb->protocol = pi.proto; |
4c13eb66 | 1803 | skb->dev = tun->dev; |
1da177e4 | 1804 | break; |
40630b82 | 1805 | case IFF_TAP: |
96aa1b22 WB |
1806 | if (frags && !pskb_may_pull(skb, ETH_HLEN)) { |
1807 | err = -ENOMEM; | |
1808 | goto drop; | |
1809 | } | |
1810 | skb->protocol = eth_type_trans(skb, tun->dev); | |
1da177e4 | 1811 | break; |
6403eab1 | 1812 | } |
1da177e4 | 1813 | |
0690899b MT |
1814 | /* copy skb_ubuf_info for callback when skb has no error */ |
1815 | if (zerocopy) { | |
9ee5e5ad | 1816 | skb_zcopy_init(skb, msg_control); |
af1cc7a2 JW |
1817 | } else if (msg_control) { |
1818 | struct ubuf_info *uarg = msg_control; | |
36177832 | 1819 | uarg->callback(NULL, uarg, false); |
0690899b MT |
1820 | } |
1821 | ||
72f65107 | 1822 | skb_reset_network_header(skb); |
d2aa125d | 1823 | skb_probe_transport_header(skb); |
3fe260e0 | 1824 | skb_record_rx_queue(skb, tfile->queue_index); |
38502af7 | 1825 | |
1cfe6e93 | 1826 | if (skb_xdp) { |
761876c8 JW |
1827 | struct bpf_prog *xdp_prog; |
1828 | int ret; | |
1829 | ||
6547e387 | 1830 | local_bh_disable(); |
761876c8 JW |
1831 | rcu_read_lock(); |
1832 | xdp_prog = rcu_dereference(tun->xdp_prog); | |
1833 | if (xdp_prog) { | |
1834 | ret = do_xdp_generic(xdp_prog, skb); | |
1835 | if (ret != XDP_PASS) { | |
1836 | rcu_read_unlock(); | |
6547e387 | 1837 | local_bh_enable(); |
1efba987 ED |
1838 | if (frags) { |
1839 | tfile->napi.skb = NULL; | |
1840 | mutex_unlock(&tfile->napi_mutex); | |
1841 | } | |
761876c8 JW |
1842 | return total_len; |
1843 | } | |
1844 | } | |
1845 | rcu_read_unlock(); | |
6547e387 | 1846 | local_bh_enable(); |
761876c8 JW |
1847 | } |
1848 | ||
cf1a1e07 PA |
1849 | /* Compute the costly rx hash only if needed for flow updates. |
1850 | * We may get a very small possibility of OOO during switching, not | |
1851 | * worth to optimize. | |
1852 | */ | |
1853 | if (!rcu_access_pointer(tun->steering_prog) && tun->numqueues > 1 && | |
1854 | !tfile->detached) | |
96f84061 | 1855 | rxhash = __skb_get_hash_symmetric(skb); |
94317099 | 1856 | |
4477138f ED |
1857 | rcu_read_lock(); |
1858 | if (unlikely(!(tun->dev->flags & IFF_UP))) { | |
1859 | err = -EIO; | |
9180bb4f | 1860 | rcu_read_unlock(); |
4477138f ED |
1861 | goto drop; |
1862 | } | |
1863 | ||
90e33d45 | 1864 | if (frags) { |
96aa1b22 WB |
1865 | u32 headlen; |
1866 | ||
90e33d45 | 1867 | /* Exercise flow dissector code path. */ |
96aa1b22 WB |
1868 | skb_push(skb, ETH_HLEN); |
1869 | headlen = eth_get_headlen(tun->dev, skb->data, | |
1870 | skb_headlen(skb)); | |
90e33d45 | 1871 | |
010f245b | 1872 | if (unlikely(headlen > skb_headlen(skb))) { |
497a5757 | 1873 | atomic_long_inc(&tun->dev->rx_dropped); |
90e33d45 | 1874 | napi_free_frags(&tfile->napi); |
4477138f | 1875 | rcu_read_unlock(); |
90e33d45 PP |
1876 | mutex_unlock(&tfile->napi_mutex); |
1877 | WARN_ON(1); | |
1878 | return -ENOMEM; | |
1879 | } | |
1880 | ||
1881 | local_bh_disable(); | |
1882 | napi_gro_frags(&tfile->napi); | |
1883 | local_bh_enable(); | |
1884 | mutex_unlock(&tfile->napi_mutex); | |
aec72f33 | 1885 | } else if (tfile->napi_enabled) { |
94317099 PP |
1886 | struct sk_buff_head *queue = &tfile->sk.sk_write_queue; |
1887 | int queue_len; | |
1888 | ||
1889 | spin_lock_bh(&queue->lock); | |
1890 | __skb_queue_tail(queue, skb); | |
1891 | queue_len = skb_queue_len(queue); | |
1892 | spin_unlock(&queue->lock); | |
1893 | ||
1894 | if (!more || queue_len > NAPI_POLL_WEIGHT) | |
1895 | napi_schedule(&tfile->napi); | |
1896 | ||
1897 | local_bh_enable(); | |
1898 | } else if (!IS_ENABLED(CONFIG_4KSTACKS)) { | |
1899 | tun_rx_batched(tun, tfile, skb, more); | |
1900 | } else { | |
1901 | netif_rx_ni(skb); | |
1902 | } | |
4477138f | 1903 | rcu_read_unlock(); |
6aa20a22 | 1904 | |
497a5757 HK |
1905 | preempt_disable(); |
1906 | dev_sw_netstats_rx_add(tun->dev, len); | |
1907 | preempt_enable(); | |
1da177e4 | 1908 | |
96f84061 JW |
1909 | if (rxhash) |
1910 | tun_flow_update(tun, rxhash, tfile); | |
1911 | ||
0690899b | 1912 | return total_len; |
6aa20a22 | 1913 | } |
1da177e4 | 1914 | |
f5ff53b4 | 1915 | static ssize_t tun_chr_write_iter(struct kiocb *iocb, struct iov_iter *from) |
1da177e4 | 1916 | { |
33dccbb0 | 1917 | struct file *file = iocb->ki_filp; |
54f968d6 | 1918 | struct tun_file *tfile = file->private_data; |
9484dc74 | 1919 | struct tun_struct *tun = tun_get(tfile); |
631ab46b | 1920 | ssize_t result; |
5aac0390 | 1921 | int noblock = 0; |
1da177e4 LT |
1922 | |
1923 | if (!tun) | |
1924 | return -EBADFD; | |
1925 | ||
5aac0390 JA |
1926 | if ((file->f_flags & O_NONBLOCK) || (iocb->ki_flags & IOCB_NOWAIT)) |
1927 | noblock = 1; | |
1928 | ||
1929 | result = tun_get_user(tun, tfile, NULL, from, noblock, false); | |
631ab46b EB |
1930 | |
1931 | tun_put(tun); | |
1932 | return result; | |
1da177e4 LT |
1933 | } |
1934 | ||
fc72d1d5 JW |
1935 | static ssize_t tun_put_user_xdp(struct tun_struct *tun, |
1936 | struct tun_file *tfile, | |
1ffcbc85 | 1937 | struct xdp_frame *xdp_frame, |
fc72d1d5 JW |
1938 | struct iov_iter *iter) |
1939 | { | |
1940 | int vnet_hdr_sz = 0; | |
1ffcbc85 | 1941 | size_t size = xdp_frame->len; |
fc72d1d5 JW |
1942 | size_t ret; |
1943 | ||
1944 | if (tun->flags & IFF_VNET_HDR) { | |
1945 | struct virtio_net_hdr gso = { 0 }; | |
1946 | ||
1947 | vnet_hdr_sz = READ_ONCE(tun->vnet_hdr_sz); | |
1948 | if (unlikely(iov_iter_count(iter) < vnet_hdr_sz)) | |
1949 | return -EINVAL; | |
1950 | if (unlikely(copy_to_iter(&gso, sizeof(gso), iter) != | |
1951 | sizeof(gso))) | |
1952 | return -EFAULT; | |
1953 | iov_iter_advance(iter, vnet_hdr_sz - sizeof(gso)); | |
1954 | } | |
1955 | ||
1ffcbc85 | 1956 | ret = copy_to_iter(xdp_frame->data, size, iter) + vnet_hdr_sz; |
fc72d1d5 | 1957 | |
497a5757 HK |
1958 | preempt_disable(); |
1959 | dev_sw_netstats_tx_add(tun->dev, 1, ret); | |
1960 | preempt_enable(); | |
fc72d1d5 JW |
1961 | |
1962 | return ret; | |
1963 | } | |
1964 | ||
1da177e4 | 1965 | /* Put packet to the user space buffer */ |
6f7c156c | 1966 | static ssize_t tun_put_user(struct tun_struct *tun, |
54f968d6 | 1967 | struct tun_file *tfile, |
6f7c156c | 1968 | struct sk_buff *skb, |
e0b46d0e | 1969 | struct iov_iter *iter) |
1da177e4 LT |
1970 | { |
1971 | struct tun_pi pi = { 0, skb->protocol }; | |
e0b46d0e | 1972 | ssize_t total; |
8c847d25 | 1973 | int vlan_offset = 0; |
a8f9bfdf | 1974 | int vlan_hlen = 0; |
2eb783c4 | 1975 | int vnet_hdr_sz = 0; |
a8f9bfdf | 1976 | |
df8a39de | 1977 | if (skb_vlan_tag_present(skb)) |
a8f9bfdf | 1978 | vlan_hlen = VLAN_HLEN; |
1da177e4 | 1979 | |
40630b82 | 1980 | if (tun->flags & IFF_VNET_HDR) |
e1edab87 | 1981 | vnet_hdr_sz = READ_ONCE(tun->vnet_hdr_sz); |
1da177e4 | 1982 | |
e0b46d0e HX |
1983 | total = skb->len + vlan_hlen + vnet_hdr_sz; |
1984 | ||
40630b82 | 1985 | if (!(tun->flags & IFF_NO_PI)) { |
e0b46d0e | 1986 | if (iov_iter_count(iter) < sizeof(pi)) |
1da177e4 LT |
1987 | return -EINVAL; |
1988 | ||
e0b46d0e HX |
1989 | total += sizeof(pi); |
1990 | if (iov_iter_count(iter) < total) { | |
1da177e4 LT |
1991 | /* Packet will be striped */ |
1992 | pi.flags |= TUN_PKT_STRIP; | |
1993 | } | |
6aa20a22 | 1994 | |
e0b46d0e | 1995 | if (copy_to_iter(&pi, sizeof(pi), iter) != sizeof(pi)) |
1da177e4 | 1996 | return -EFAULT; |
6aa20a22 | 1997 | } |
1da177e4 | 1998 | |
2eb783c4 | 1999 | if (vnet_hdr_sz) { |
9403cd7c | 2000 | struct virtio_net_hdr gso; |
34166093 | 2001 | |
e0b46d0e | 2002 | if (iov_iter_count(iter) < vnet_hdr_sz) |
f43798c2 RR |
2003 | return -EINVAL; |
2004 | ||
3e9e40e7 | 2005 | if (virtio_net_hdr_from_skb(skb, &gso, |
fd3a8862 WB |
2006 | tun_is_little_endian(tun), true, |
2007 | vlan_hlen)) { | |
f43798c2 | 2008 | struct skb_shared_info *sinfo = skb_shinfo(skb); |
34166093 MR |
2009 | pr_err("unexpected GSO type: " |
2010 | "0x%x, gso_size %d, hdr_len %d\n", | |
2011 | sinfo->gso_type, tun16_to_cpu(tun, gso.gso_size), | |
2012 | tun16_to_cpu(tun, gso.hdr_len)); | |
2013 | print_hex_dump(KERN_ERR, "tun: ", | |
2014 | DUMP_PREFIX_NONE, | |
2015 | 16, 1, skb->head, | |
2016 | min((int)tun16_to_cpu(tun, gso.hdr_len), 64), true); | |
2017 | WARN_ON_ONCE(1); | |
2018 | return -EINVAL; | |
2019 | } | |
f43798c2 | 2020 | |
e0b46d0e | 2021 | if (copy_to_iter(&gso, sizeof(gso), iter) != sizeof(gso)) |
f43798c2 | 2022 | return -EFAULT; |
8c847d25 JW |
2023 | |
2024 | iov_iter_advance(iter, vnet_hdr_sz - sizeof(gso)); | |
f43798c2 RR |
2025 | } |
2026 | ||
a8f9bfdf | 2027 | if (vlan_hlen) { |
e0b46d0e | 2028 | int ret; |
aff3d70a | 2029 | struct veth veth; |
6680ec68 JW |
2030 | |
2031 | veth.h_vlan_proto = skb->vlan_proto; | |
df8a39de | 2032 | veth.h_vlan_TCI = htons(skb_vlan_tag_get(skb)); |
6680ec68 JW |
2033 | |
2034 | vlan_offset = offsetof(struct vlan_ethhdr, h_vlan_proto); | |
6680ec68 | 2035 | |
e0b46d0e HX |
2036 | ret = skb_copy_datagram_iter(skb, 0, iter, vlan_offset); |
2037 | if (ret || !iov_iter_count(iter)) | |
6680ec68 JW |
2038 | goto done; |
2039 | ||
e0b46d0e HX |
2040 | ret = copy_to_iter(&veth, sizeof(veth), iter); |
2041 | if (ret != sizeof(veth) || !iov_iter_count(iter)) | |
6680ec68 JW |
2042 | goto done; |
2043 | } | |
1da177e4 | 2044 | |
e0b46d0e | 2045 | skb_copy_datagram_iter(skb, vlan_offset, iter, skb->len - vlan_offset); |
1da177e4 | 2046 | |
6680ec68 | 2047 | done: |
608b9977 | 2048 | /* caller is in process context, */ |
497a5757 HK |
2049 | preempt_disable(); |
2050 | dev_sw_netstats_tx_add(tun->dev, 1, skb->len + vlan_hlen); | |
2051 | preempt_enable(); | |
1da177e4 LT |
2052 | |
2053 | return total; | |
2054 | } | |
2055 | ||
fc72d1d5 | 2056 | static void *tun_ring_recv(struct tun_file *tfile, int noblock, int *err) |
1576d986 JW |
2057 | { |
2058 | DECLARE_WAITQUEUE(wait, current); | |
fc72d1d5 | 2059 | void *ptr = NULL; |
f48cc6b2 | 2060 | int error = 0; |
1576d986 | 2061 | |
fc72d1d5 JW |
2062 | ptr = ptr_ring_consume(&tfile->tx_ring); |
2063 | if (ptr) | |
1576d986 JW |
2064 | goto out; |
2065 | if (noblock) { | |
f48cc6b2 | 2066 | error = -EAGAIN; |
1576d986 JW |
2067 | goto out; |
2068 | } | |
2069 | ||
333f7909 | 2070 | add_wait_queue(&tfile->socket.wq.wait, &wait); |
1576d986 JW |
2071 | |
2072 | while (1) { | |
71828b22 | 2073 | set_current_state(TASK_INTERRUPTIBLE); |
fc72d1d5 JW |
2074 | ptr = ptr_ring_consume(&tfile->tx_ring); |
2075 | if (ptr) | |
1576d986 JW |
2076 | break; |
2077 | if (signal_pending(current)) { | |
f48cc6b2 | 2078 | error = -ERESTARTSYS; |
1576d986 JW |
2079 | break; |
2080 | } | |
2081 | if (tfile->socket.sk->sk_shutdown & RCV_SHUTDOWN) { | |
f48cc6b2 | 2082 | error = -EFAULT; |
1576d986 JW |
2083 | break; |
2084 | } | |
2085 | ||
2086 | schedule(); | |
2087 | } | |
2088 | ||
ecef67cb | 2089 | __set_current_state(TASK_RUNNING); |
333f7909 | 2090 | remove_wait_queue(&tfile->socket.wq.wait, &wait); |
1576d986 JW |
2091 | |
2092 | out: | |
f48cc6b2 | 2093 | *err = error; |
fc72d1d5 | 2094 | return ptr; |
1576d986 JW |
2095 | } |
2096 | ||
54f968d6 | 2097 | static ssize_t tun_do_read(struct tun_struct *tun, struct tun_file *tfile, |
9b067034 | 2098 | struct iov_iter *to, |
fc72d1d5 | 2099 | int noblock, void *ptr) |
1da177e4 | 2100 | { |
9b067034 | 2101 | ssize_t ret; |
1576d986 | 2102 | int err; |
1da177e4 | 2103 | |
c33ee15b | 2104 | if (!iov_iter_count(to)) { |
fc72d1d5 | 2105 | tun_ptr_free(ptr); |
9b067034 | 2106 | return 0; |
c33ee15b | 2107 | } |
1da177e4 | 2108 | |
fc72d1d5 | 2109 | if (!ptr) { |
ac77cfd4 | 2110 | /* Read frames from ring */ |
fc72d1d5 JW |
2111 | ptr = tun_ring_recv(tfile, noblock, &err); |
2112 | if (!ptr) | |
ac77cfd4 JW |
2113 | return err; |
2114 | } | |
e0b46d0e | 2115 | |
1ffcbc85 JDB |
2116 | if (tun_is_xdp_frame(ptr)) { |
2117 | struct xdp_frame *xdpf = tun_ptr_to_xdp(ptr); | |
fc72d1d5 | 2118 | |
1ffcbc85 | 2119 | ret = tun_put_user_xdp(tun, tfile, xdpf, to); |
03993094 | 2120 | xdp_return_frame(xdpf); |
fc72d1d5 JW |
2121 | } else { |
2122 | struct sk_buff *skb = ptr; | |
2123 | ||
2124 | ret = tun_put_user(tun, tfile, skb, to); | |
2125 | if (unlikely(ret < 0)) | |
2126 | kfree_skb(skb); | |
2127 | else | |
2128 | consume_skb(skb); | |
2129 | } | |
1da177e4 | 2130 | |
05c2828c MT |
2131 | return ret; |
2132 | } | |
2133 | ||
9b067034 | 2134 | static ssize_t tun_chr_read_iter(struct kiocb *iocb, struct iov_iter *to) |
05c2828c MT |
2135 | { |
2136 | struct file *file = iocb->ki_filp; | |
2137 | struct tun_file *tfile = file->private_data; | |
9484dc74 | 2138 | struct tun_struct *tun = tun_get(tfile); |
9b067034 | 2139 | ssize_t len = iov_iter_count(to), ret; |
5aac0390 | 2140 | int noblock = 0; |
05c2828c MT |
2141 | |
2142 | if (!tun) | |
2143 | return -EBADFD; | |
5aac0390 JA |
2144 | |
2145 | if ((file->f_flags & O_NONBLOCK) || (iocb->ki_flags & IOCB_NOWAIT)) | |
2146 | noblock = 1; | |
2147 | ||
2148 | ret = tun_do_read(tun, tfile, to, noblock, NULL); | |
42404c09 | 2149 | ret = min_t(ssize_t, ret, len); |
d0b7da8a ZYW |
2150 | if (ret > 0) |
2151 | iocb->ki_pos = ret; | |
631ab46b | 2152 | tun_put(tun); |
1da177e4 LT |
2153 | return ret; |
2154 | } | |
2155 | ||
cd5681d7 | 2156 | static void tun_prog_free(struct rcu_head *rcu) |
96f84061 | 2157 | { |
cd5681d7 | 2158 | struct tun_prog *prog = container_of(rcu, struct tun_prog, rcu); |
96f84061 JW |
2159 | |
2160 | bpf_prog_destroy(prog->prog); | |
2161 | kfree(prog); | |
2162 | } | |
2163 | ||
9d6474e4 JW |
2164 | static int __tun_set_ebpf(struct tun_struct *tun, |
2165 | struct tun_prog __rcu **prog_p, | |
cd5681d7 | 2166 | struct bpf_prog *prog) |
96f84061 | 2167 | { |
cd5681d7 | 2168 | struct tun_prog *old, *new = NULL; |
96f84061 JW |
2169 | |
2170 | if (prog) { | |
2171 | new = kmalloc(sizeof(*new), GFP_KERNEL); | |
2172 | if (!new) | |
2173 | return -ENOMEM; | |
2174 | new->prog = prog; | |
2175 | } | |
2176 | ||
124da8f6 | 2177 | spin_lock_bh(&tun->lock); |
cd5681d7 | 2178 | old = rcu_dereference_protected(*prog_p, |
124da8f6 | 2179 | lockdep_is_held(&tun->lock)); |
cd5681d7 | 2180 | rcu_assign_pointer(*prog_p, new); |
124da8f6 | 2181 | spin_unlock_bh(&tun->lock); |
96f84061 JW |
2182 | |
2183 | if (old) | |
cd5681d7 | 2184 | call_rcu(&old->rcu, tun_prog_free); |
96f84061 JW |
2185 | |
2186 | return 0; | |
2187 | } | |
2188 | ||
96442e42 JW |
2189 | static void tun_free_netdev(struct net_device *dev) |
2190 | { | |
2191 | struct tun_struct *tun = netdev_priv(dev); | |
2192 | ||
4008e97f | 2193 | BUG_ON(!(list_empty(&tun->disabled))); |
11fc7d5a | 2194 | |
497a5757 HK |
2195 | free_percpu(dev->tstats); |
2196 | /* We clear tstats so that tun_set_iff() can tell if | |
11fc7d5a ED |
2197 | * tun_free_netdev() has been called from register_netdevice(). |
2198 | */ | |
497a5757 | 2199 | dev->tstats = NULL; |
11fc7d5a | 2200 | |
96442e42 | 2201 | tun_flow_uninit(tun); |
5dbbaf2d | 2202 | security_tun_dev_free_security(tun->security); |
cd5681d7 | 2203 | __tun_set_ebpf(tun, &tun->steering_prog, NULL); |
aff3d70a | 2204 | __tun_set_ebpf(tun, &tun->filter_prog, NULL); |
96442e42 JW |
2205 | } |
2206 | ||
1da177e4 LT |
2207 | static void tun_setup(struct net_device *dev) |
2208 | { | |
2209 | struct tun_struct *tun = netdev_priv(dev); | |
2210 | ||
0625c883 EB |
2211 | tun->owner = INVALID_UID; |
2212 | tun->group = INVALID_GID; | |
4e24f2dd | 2213 | tun_default_link_ksettings(dev, &tun->link_ksettings); |
1da177e4 | 2214 | |
1da177e4 | 2215 | dev->ethtool_ops = &tun_ethtool_ops; |
cf124db5 DM |
2216 | dev->needs_free_netdev = true; |
2217 | dev->priv_destructor = tun_free_netdev; | |
016adb72 JW |
2218 | /* We prefer our own queue length */ |
2219 | dev->tx_queue_len = TUN_READQ_SIZE; | |
1da177e4 LT |
2220 | } |
2221 | ||
f019a7a5 EB |
2222 | /* Trivial set of netlink ops to allow deleting tun or tap |
2223 | * device with netlink. | |
2224 | */ | |
a8b8a889 MS |
2225 | static int tun_validate(struct nlattr *tb[], struct nlattr *data[], |
2226 | struct netlink_ext_ack *extack) | |
f019a7a5 | 2227 | { |
35b827b6 ND |
2228 | NL_SET_ERR_MSG(extack, |
2229 | "tun/tap creation via rtnetlink is not supported."); | |
2230 | return -EOPNOTSUPP; | |
f019a7a5 EB |
2231 | } |
2232 | ||
1ec010e7 SD |
2233 | static size_t tun_get_size(const struct net_device *dev) |
2234 | { | |
2235 | BUILD_BUG_ON(sizeof(u32) != sizeof(uid_t)); | |
2236 | BUILD_BUG_ON(sizeof(u32) != sizeof(gid_t)); | |
2237 | ||
2238 | return nla_total_size(sizeof(uid_t)) + /* OWNER */ | |
2239 | nla_total_size(sizeof(gid_t)) + /* GROUP */ | |
2240 | nla_total_size(sizeof(u8)) + /* TYPE */ | |
2241 | nla_total_size(sizeof(u8)) + /* PI */ | |
2242 | nla_total_size(sizeof(u8)) + /* VNET_HDR */ | |
2243 | nla_total_size(sizeof(u8)) + /* PERSIST */ | |
2244 | nla_total_size(sizeof(u8)) + /* MULTI_QUEUE */ | |
2245 | nla_total_size(sizeof(u32)) + /* NUM_QUEUES */ | |
2246 | nla_total_size(sizeof(u32)) + /* NUM_DISABLED_QUEUES */ | |
2247 | 0; | |
2248 | } | |
2249 | ||
2250 | static int tun_fill_info(struct sk_buff *skb, const struct net_device *dev) | |
2251 | { | |
2252 | struct tun_struct *tun = netdev_priv(dev); | |
2253 | ||
2254 | if (nla_put_u8(skb, IFLA_TUN_TYPE, tun->flags & TUN_TYPE_MASK)) | |
2255 | goto nla_put_failure; | |
2256 | if (uid_valid(tun->owner) && | |
2257 | nla_put_u32(skb, IFLA_TUN_OWNER, | |
2258 | from_kuid_munged(current_user_ns(), tun->owner))) | |
2259 | goto nla_put_failure; | |
2260 | if (gid_valid(tun->group) && | |
2261 | nla_put_u32(skb, IFLA_TUN_GROUP, | |
2262 | from_kgid_munged(current_user_ns(), tun->group))) | |
2263 | goto nla_put_failure; | |
2264 | if (nla_put_u8(skb, IFLA_TUN_PI, !(tun->flags & IFF_NO_PI))) | |
2265 | goto nla_put_failure; | |
2266 | if (nla_put_u8(skb, IFLA_TUN_VNET_HDR, !!(tun->flags & IFF_VNET_HDR))) | |
2267 | goto nla_put_failure; | |
2268 | if (nla_put_u8(skb, IFLA_TUN_PERSIST, !!(tun->flags & IFF_PERSIST))) | |
2269 | goto nla_put_failure; | |
2270 | if (nla_put_u8(skb, IFLA_TUN_MULTI_QUEUE, | |
2271 | !!(tun->flags & IFF_MULTI_QUEUE))) | |
2272 | goto nla_put_failure; | |
2273 | if (tun->flags & IFF_MULTI_QUEUE) { | |
2274 | if (nla_put_u32(skb, IFLA_TUN_NUM_QUEUES, tun->numqueues)) | |
2275 | goto nla_put_failure; | |
2276 | if (nla_put_u32(skb, IFLA_TUN_NUM_DISABLED_QUEUES, | |
2277 | tun->numdisabled)) | |
2278 | goto nla_put_failure; | |
2279 | } | |
2280 | ||
2281 | return 0; | |
2282 | ||
2283 | nla_put_failure: | |
2284 | return -EMSGSIZE; | |
2285 | } | |
2286 | ||
f019a7a5 EB |
2287 | static struct rtnl_link_ops tun_link_ops __read_mostly = { |
2288 | .kind = DRV_NAME, | |
2289 | .priv_size = sizeof(struct tun_struct), | |
2290 | .setup = tun_setup, | |
2291 | .validate = tun_validate, | |
1ec010e7 SD |
2292 | .get_size = tun_get_size, |
2293 | .fill_info = tun_fill_info, | |
f019a7a5 EB |
2294 | }; |
2295 | ||
33dccbb0 HX |
2296 | static void tun_sock_write_space(struct sock *sk) |
2297 | { | |
54f968d6 | 2298 | struct tun_file *tfile; |
43815482 | 2299 | wait_queue_head_t *wqueue; |
33dccbb0 HX |
2300 | |
2301 | if (!sock_writeable(sk)) | |
2302 | return; | |
2303 | ||
9cd3e072 | 2304 | if (!test_and_clear_bit(SOCKWQ_ASYNC_NOSPACE, &sk->sk_socket->flags)) |
33dccbb0 HX |
2305 | return; |
2306 | ||
43815482 ED |
2307 | wqueue = sk_sleep(sk); |
2308 | if (wqueue && waitqueue_active(wqueue)) | |
a9a08845 LT |
2309 | wake_up_interruptible_sync_poll(wqueue, EPOLLOUT | |
2310 | EPOLLWRNORM | EPOLLWRBAND); | |
c722c625 | 2311 | |
54f968d6 JW |
2312 | tfile = container_of(sk, struct tun_file, sk); |
2313 | kill_fasync(&tfile->fasync, SIGIO, POLL_OUT); | |
33dccbb0 HX |
2314 | } |
2315 | ||
f9e06c45 JW |
2316 | static void tun_put_page(struct tun_page *tpage) |
2317 | { | |
2318 | if (tpage->page) | |
2319 | __page_frag_cache_drain(tpage->page, tpage->count); | |
2320 | } | |
2321 | ||
043d222f JW |
2322 | static int tun_xdp_one(struct tun_struct *tun, |
2323 | struct tun_file *tfile, | |
f9e06c45 JW |
2324 | struct xdp_buff *xdp, int *flush, |
2325 | struct tun_page *tpage) | |
043d222f | 2326 | { |
4e4b08e5 | 2327 | unsigned int datasize = xdp->data_end - xdp->data; |
043d222f JW |
2328 | struct tun_xdp_hdr *hdr = xdp->data_hard_start; |
2329 | struct virtio_net_hdr *gso = &hdr->gso; | |
043d222f JW |
2330 | struct bpf_prog *xdp_prog; |
2331 | struct sk_buff *skb = NULL; | |
2332 | u32 rxhash = 0, act; | |
2333 | int buflen = hdr->buflen; | |
2334 | int err = 0; | |
2335 | bool skb_xdp = false; | |
f9e06c45 | 2336 | struct page *page; |
043d222f JW |
2337 | |
2338 | xdp_prog = rcu_dereference(tun->xdp_prog); | |
2339 | if (xdp_prog) { | |
2340 | if (gso->gso_type) { | |
2341 | skb_xdp = true; | |
2342 | goto build; | |
2343 | } | |
43b5169d LB |
2344 | |
2345 | xdp_init_buff(xdp, buflen, &tfile->xdp_rxq); | |
043d222f | 2346 | xdp_set_data_meta_invalid(xdp); |
043d222f JW |
2347 | |
2348 | act = bpf_prog_run_xdp(xdp_prog, xdp); | |
2349 | err = tun_xdp_act(tun, xdp_prog, xdp, act); | |
2350 | if (err < 0) { | |
2351 | put_page(virt_to_head_page(xdp->data)); | |
2352 | return err; | |
2353 | } | |
2354 | ||
2355 | switch (err) { | |
2356 | case XDP_REDIRECT: | |
2357 | *flush = true; | |
df561f66 | 2358 | fallthrough; |
043d222f JW |
2359 | case XDP_TX: |
2360 | return 0; | |
2361 | case XDP_PASS: | |
2362 | break; | |
2363 | default: | |
f9e06c45 JW |
2364 | page = virt_to_head_page(xdp->data); |
2365 | if (tpage->page == page) { | |
2366 | ++tpage->count; | |
2367 | } else { | |
2368 | tun_put_page(tpage); | |
2369 | tpage->page = page; | |
2370 | tpage->count = 1; | |
2371 | } | |
043d222f JW |
2372 | return 0; |
2373 | } | |
2374 | } | |
2375 | ||
2376 | build: | |
2377 | skb = build_skb(xdp->data_hard_start, buflen); | |
2378 | if (!skb) { | |
2379 | err = -ENOMEM; | |
2380 | goto out; | |
2381 | } | |
2382 | ||
2383 | skb_reserve(skb, xdp->data - xdp->data_hard_start); | |
2384 | skb_put(skb, xdp->data_end - xdp->data); | |
2385 | ||
2386 | if (virtio_net_hdr_to_skb(skb, gso, tun_is_little_endian(tun))) { | |
497a5757 | 2387 | atomic_long_inc(&tun->rx_frame_errors); |
043d222f JW |
2388 | kfree_skb(skb); |
2389 | err = -EINVAL; | |
2390 | goto out; | |
2391 | } | |
2392 | ||
2393 | skb->protocol = eth_type_trans(skb, tun->dev); | |
2394 | skb_reset_network_header(skb); | |
d2aa125d | 2395 | skb_probe_transport_header(skb); |
3fe260e0 | 2396 | skb_record_rx_queue(skb, tfile->queue_index); |
043d222f JW |
2397 | |
2398 | if (skb_xdp) { | |
2399 | err = do_xdp_generic(xdp_prog, skb); | |
2400 | if (err != XDP_PASS) | |
2401 | goto out; | |
2402 | } | |
2403 | ||
f29eb2a9 PA |
2404 | if (!rcu_dereference(tun->steering_prog) && tun->numqueues > 1 && |
2405 | !tfile->detached) | |
043d222f JW |
2406 | rxhash = __skb_get_hash_symmetric(skb); |
2407 | ||
2408 | netif_receive_skb(skb); | |
2409 | ||
497a5757 | 2410 | /* No need to disable preemption here since this function is |
6342ca64 PB |
2411 | * always called with bh disabled |
2412 | */ | |
497a5757 | 2413 | dev_sw_netstats_rx_add(tun->dev, datasize); |
043d222f JW |
2414 | |
2415 | if (rxhash) | |
2416 | tun_flow_update(tun, rxhash, tfile); | |
2417 | ||
2418 | out: | |
2419 | return err; | |
2420 | } | |
2421 | ||
1b784140 | 2422 | static int tun_sendmsg(struct socket *sock, struct msghdr *m, size_t total_len) |
05c2828c | 2423 | { |
043d222f | 2424 | int ret, i; |
54f968d6 | 2425 | struct tun_file *tfile = container_of(sock, struct tun_file, socket); |
9484dc74 | 2426 | struct tun_struct *tun = tun_get(tfile); |
fe8dd45b | 2427 | struct tun_msg_ctl *ctl = m->msg_control; |
043d222f | 2428 | struct xdp_buff *xdp; |
54f968d6 JW |
2429 | |
2430 | if (!tun) | |
2431 | return -EBADFD; | |
f5ff53b4 | 2432 | |
043d222f | 2433 | if (ctl && (ctl->type == TUN_MSG_PTR)) { |
6f0271d9 | 2434 | struct tun_page tpage; |
043d222f JW |
2435 | int n = ctl->num; |
2436 | int flush = 0; | |
2437 | ||
6f0271d9 DM |
2438 | memset(&tpage, 0, sizeof(tpage)); |
2439 | ||
043d222f JW |
2440 | local_bh_disable(); |
2441 | rcu_read_lock(); | |
2442 | ||
2443 | for (i = 0; i < n; i++) { | |
2444 | xdp = &((struct xdp_buff *)ctl->ptr)[i]; | |
f9e06c45 | 2445 | tun_xdp_one(tun, tfile, xdp, &flush, &tpage); |
043d222f JW |
2446 | } |
2447 | ||
2448 | if (flush) | |
1d233886 | 2449 | xdp_do_flush(); |
043d222f JW |
2450 | |
2451 | rcu_read_unlock(); | |
2452 | local_bh_enable(); | |
2453 | ||
f9e06c45 JW |
2454 | tun_put_page(&tpage); |
2455 | ||
043d222f JW |
2456 | ret = total_len; |
2457 | goto out; | |
2458 | } | |
fe8dd45b JW |
2459 | |
2460 | ret = tun_get_user(tun, tfile, ctl ? ctl->ptr : NULL, &m->msg_iter, | |
5503fcec JW |
2461 | m->msg_flags & MSG_DONTWAIT, |
2462 | m->msg_flags & MSG_MORE); | |
043d222f | 2463 | out: |
54f968d6 JW |
2464 | tun_put(tun); |
2465 | return ret; | |
05c2828c MT |
2466 | } |
2467 | ||
1b784140 | 2468 | static int tun_recvmsg(struct socket *sock, struct msghdr *m, size_t total_len, |
05c2828c MT |
2469 | int flags) |
2470 | { | |
54f968d6 | 2471 | struct tun_file *tfile = container_of(sock, struct tun_file, socket); |
9484dc74 | 2472 | struct tun_struct *tun = tun_get(tfile); |
fc72d1d5 | 2473 | void *ptr = m->msg_control; |
05c2828c | 2474 | int ret; |
54f968d6 | 2475 | |
c33ee15b WX |
2476 | if (!tun) { |
2477 | ret = -EBADFD; | |
fc72d1d5 | 2478 | goto out_free; |
c33ee15b | 2479 | } |
54f968d6 | 2480 | |
eda29772 | 2481 | if (flags & ~(MSG_DONTWAIT|MSG_TRUNC|MSG_ERRQUEUE)) { |
3811ae76 | 2482 | ret = -EINVAL; |
c33ee15b | 2483 | goto out_put_tun; |
3811ae76 | 2484 | } |
eda29772 RC |
2485 | if (flags & MSG_ERRQUEUE) { |
2486 | ret = sock_recv_errqueue(sock->sk, m, total_len, | |
2487 | SOL_PACKET, TUN_TX_TIMESTAMP); | |
2488 | goto out; | |
2489 | } | |
fc72d1d5 | 2490 | ret = tun_do_read(tun, tfile, &m->msg_iter, flags & MSG_DONTWAIT, ptr); |
87897931 | 2491 | if (ret > (ssize_t)total_len) { |
42404c09 DM |
2492 | m->msg_flags |= MSG_TRUNC; |
2493 | ret = flags & MSG_TRUNC ? ret : total_len; | |
2494 | } | |
3811ae76 | 2495 | out: |
54f968d6 | 2496 | tun_put(tun); |
05c2828c | 2497 | return ret; |
c33ee15b WX |
2498 | |
2499 | out_put_tun: | |
2500 | tun_put(tun); | |
fc72d1d5 JW |
2501 | out_free: |
2502 | tun_ptr_free(ptr); | |
c33ee15b | 2503 | return ret; |
05c2828c MT |
2504 | } |
2505 | ||
fc72d1d5 JW |
2506 | static int tun_ptr_peek_len(void *ptr) |
2507 | { | |
2508 | if (likely(ptr)) { | |
1ffcbc85 JDB |
2509 | if (tun_is_xdp_frame(ptr)) { |
2510 | struct xdp_frame *xdpf = tun_ptr_to_xdp(ptr); | |
fc72d1d5 | 2511 | |
1ffcbc85 | 2512 | return xdpf->len; |
fc72d1d5 JW |
2513 | } |
2514 | return __skb_array_len_with_tag(ptr); | |
2515 | } else { | |
2516 | return 0; | |
2517 | } | |
2518 | } | |
2519 | ||
1576d986 JW |
2520 | static int tun_peek_len(struct socket *sock) |
2521 | { | |
2522 | struct tun_file *tfile = container_of(sock, struct tun_file, socket); | |
2523 | struct tun_struct *tun; | |
2524 | int ret = 0; | |
2525 | ||
9484dc74 | 2526 | tun = tun_get(tfile); |
1576d986 JW |
2527 | if (!tun) |
2528 | return 0; | |
2529 | ||
fc72d1d5 | 2530 | ret = PTR_RING_PEEK_CALL(&tfile->tx_ring, tun_ptr_peek_len); |
1576d986 JW |
2531 | tun_put(tun); |
2532 | ||
2533 | return ret; | |
2534 | } | |
2535 | ||
05c2828c MT |
2536 | /* Ops structure to mimic raw sockets with tun */ |
2537 | static const struct proto_ops tun_socket_ops = { | |
1576d986 | 2538 | .peek_len = tun_peek_len, |
05c2828c MT |
2539 | .sendmsg = tun_sendmsg, |
2540 | .recvmsg = tun_recvmsg, | |
2541 | }; | |
2542 | ||
33dccbb0 HX |
2543 | static struct proto tun_proto = { |
2544 | .name = "tun", | |
2545 | .owner = THIS_MODULE, | |
54f968d6 | 2546 | .obj_size = sizeof(struct tun_file), |
33dccbb0 | 2547 | }; |
f019a7a5 | 2548 | |
980c9e8c DW |
2549 | static int tun_flags(struct tun_struct *tun) |
2550 | { | |
031f5e03 | 2551 | return tun->flags & (TUN_FEATURES | IFF_PERSIST | IFF_TUN | IFF_TAP); |
980c9e8c DW |
2552 | } |
2553 | ||
2554 | static ssize_t tun_show_flags(struct device *dev, struct device_attribute *attr, | |
2555 | char *buf) | |
2556 | { | |
2557 | struct tun_struct *tun = netdev_priv(to_net_dev(dev)); | |
2558 | return sprintf(buf, "0x%x\n", tun_flags(tun)); | |
2559 | } | |
2560 | ||
2561 | static ssize_t tun_show_owner(struct device *dev, struct device_attribute *attr, | |
2562 | char *buf) | |
2563 | { | |
2564 | struct tun_struct *tun = netdev_priv(to_net_dev(dev)); | |
0625c883 EB |
2565 | return uid_valid(tun->owner)? |
2566 | sprintf(buf, "%u\n", | |
2567 | from_kuid_munged(current_user_ns(), tun->owner)): | |
2568 | sprintf(buf, "-1\n"); | |
980c9e8c DW |
2569 | } |
2570 | ||
2571 | static ssize_t tun_show_group(struct device *dev, struct device_attribute *attr, | |
2572 | char *buf) | |
2573 | { | |
2574 | struct tun_struct *tun = netdev_priv(to_net_dev(dev)); | |
0625c883 EB |
2575 | return gid_valid(tun->group) ? |
2576 | sprintf(buf, "%u\n", | |
2577 | from_kgid_munged(current_user_ns(), tun->group)): | |
2578 | sprintf(buf, "-1\n"); | |
980c9e8c DW |
2579 | } |
2580 | ||
2581 | static DEVICE_ATTR(tun_flags, 0444, tun_show_flags, NULL); | |
2582 | static DEVICE_ATTR(owner, 0444, tun_show_owner, NULL); | |
2583 | static DEVICE_ATTR(group, 0444, tun_show_group, NULL); | |
2584 | ||
c4d33e24 TI |
2585 | static struct attribute *tun_dev_attrs[] = { |
2586 | &dev_attr_tun_flags.attr, | |
2587 | &dev_attr_owner.attr, | |
2588 | &dev_attr_group.attr, | |
2589 | NULL | |
2590 | }; | |
2591 | ||
2592 | static const struct attribute_group tun_attr_group = { | |
2593 | .attrs = tun_dev_attrs | |
2594 | }; | |
2595 | ||
d647a591 | 2596 | static int tun_set_iff(struct net *net, struct file *file, struct ifreq *ifr) |
1da177e4 LT |
2597 | { |
2598 | struct tun_struct *tun; | |
54f968d6 | 2599 | struct tun_file *tfile = file->private_data; |
1da177e4 LT |
2600 | struct net_device *dev; |
2601 | int err; | |
2602 | ||
7c0c3b1a JW |
2603 | if (tfile->detached) |
2604 | return -EINVAL; | |
2605 | ||
90e33d45 PP |
2606 | if ((ifr->ifr_flags & IFF_NAPI_FRAGS)) { |
2607 | if (!capable(CAP_NET_ADMIN)) | |
2608 | return -EPERM; | |
2609 | ||
2610 | if (!(ifr->ifr_flags & IFF_NAPI) || | |
2611 | (ifr->ifr_flags & TUN_TYPE_MASK) != IFF_TAP) | |
2612 | return -EINVAL; | |
2613 | } | |
2614 | ||
74a3e5a7 EB |
2615 | dev = __dev_get_by_name(net, ifr->ifr_name); |
2616 | if (dev) { | |
f85ba780 DW |
2617 | if (ifr->ifr_flags & IFF_TUN_EXCL) |
2618 | return -EBUSY; | |
74a3e5a7 EB |
2619 | if ((ifr->ifr_flags & IFF_TUN) && dev->netdev_ops == &tun_netdev_ops) |
2620 | tun = netdev_priv(dev); | |
2621 | else if ((ifr->ifr_flags & IFF_TAP) && dev->netdev_ops == &tap_netdev_ops) | |
2622 | tun = netdev_priv(dev); | |
2623 | else | |
2624 | return -EINVAL; | |
2625 | ||
8e6d91ae | 2626 | if (!!(ifr->ifr_flags & IFF_MULTI_QUEUE) != |
40630b82 | 2627 | !!(tun->flags & IFF_MULTI_QUEUE)) |
8e6d91ae JW |
2628 | return -EINVAL; |
2629 | ||
cde8b15f | 2630 | if (tun_not_capable(tun)) |
2b980dbd | 2631 | return -EPERM; |
5dbbaf2d | 2632 | err = security_tun_dev_open(tun->security); |
2b980dbd PM |
2633 | if (err < 0) |
2634 | return err; | |
2635 | ||
94317099 | 2636 | err = tun_attach(tun, file, ifr->ifr_flags & IFF_NOFILTER, |
af3fb24e | 2637 | ifr->ifr_flags & IFF_NAPI, |
77f22f92 | 2638 | ifr->ifr_flags & IFF_NAPI_FRAGS, true); |
a7385ba2 EB |
2639 | if (err < 0) |
2640 | return err; | |
4008e97f | 2641 | |
40630b82 | 2642 | if (tun->flags & IFF_MULTI_QUEUE && |
e8dbad66 JW |
2643 | (tun->numqueues + tun->numdisabled > 1)) { |
2644 | /* One or more queue has already been attached, no need | |
2645 | * to initialize the device again. | |
2646 | */ | |
83c1f36f | 2647 | netdev_state_change(dev); |
e8dbad66 JW |
2648 | return 0; |
2649 | } | |
9fffc5c6 SD |
2650 | |
2651 | tun->flags = (tun->flags & ~TUN_FEATURES) | | |
2652 | (ifr->ifr_flags & TUN_FEATURES); | |
83c1f36f SD |
2653 | |
2654 | netdev_state_change(dev); | |
2655 | } else { | |
1da177e4 LT |
2656 | char *name; |
2657 | unsigned long flags = 0; | |
edfb6a14 JW |
2658 | int queues = ifr->ifr_flags & IFF_MULTI_QUEUE ? |
2659 | MAX_TAP_QUEUES : 1; | |
1da177e4 | 2660 | |
c260b772 | 2661 | if (!ns_capable(net->user_ns, CAP_NET_ADMIN)) |
ca6bb5d7 | 2662 | return -EPERM; |
2b980dbd PM |
2663 | err = security_tun_dev_create(); |
2664 | if (err < 0) | |
2665 | return err; | |
ca6bb5d7 | 2666 | |
1da177e4 LT |
2667 | /* Set dev type */ |
2668 | if (ifr->ifr_flags & IFF_TUN) { | |
2669 | /* TUN device */ | |
40630b82 | 2670 | flags |= IFF_TUN; |
1da177e4 LT |
2671 | name = "tun%d"; |
2672 | } else if (ifr->ifr_flags & IFF_TAP) { | |
2673 | /* TAP device */ | |
40630b82 | 2674 | flags |= IFF_TAP; |
1da177e4 | 2675 | name = "tap%d"; |
6aa20a22 | 2676 | } else |
36989b90 | 2677 | return -EINVAL; |
6aa20a22 | 2678 | |
1da177e4 LT |
2679 | if (*ifr->ifr_name) |
2680 | name = ifr->ifr_name; | |
2681 | ||
c8d68e6b | 2682 | dev = alloc_netdev_mqs(sizeof(struct tun_struct), name, |
c835a677 TG |
2683 | NET_NAME_UNKNOWN, tun_setup, queues, |
2684 | queues); | |
edfb6a14 | 2685 | |
1da177e4 LT |
2686 | if (!dev) |
2687 | return -ENOMEM; | |
2688 | ||
fc54c658 | 2689 | dev_net_set(dev, net); |
f019a7a5 | 2690 | dev->rtnl_link_ops = &tun_link_ops; |
fb7589a1 | 2691 | dev->ifindex = tfile->ifindex; |
c4d33e24 | 2692 | dev->sysfs_groups[0] = &tun_attr_group; |
758e43b7 | 2693 | |
1da177e4 LT |
2694 | tun = netdev_priv(dev); |
2695 | tun->dev = dev; | |
2696 | tun->flags = flags; | |
f271b2cc | 2697 | tun->txflt.count = 0; |
d9d52b51 | 2698 | tun->vnet_hdr_sz = sizeof(struct virtio_net_hdr); |
33dccbb0 | 2699 | |
eaea34b2 | 2700 | tun->align = NET_SKB_PAD; |
54f968d6 JW |
2701 | tun->filter_attached = false; |
2702 | tun->sndbuf = tfile->socket.sk->sk_sndbuf; | |
5503fcec | 2703 | tun->rx_batched = 0; |
96f84061 | 2704 | RCU_INIT_POINTER(tun->steering_prog, NULL); |
33dccbb0 | 2705 | |
497a5757 HK |
2706 | dev->tstats = netdev_alloc_pcpu_stats(struct pcpu_sw_netstats); |
2707 | if (!dev->tstats) { | |
608b9977 PA |
2708 | err = -ENOMEM; |
2709 | goto err_free_dev; | |
2710 | } | |
2711 | ||
96442e42 JW |
2712 | spin_lock_init(&tun->lock); |
2713 | ||
5dbbaf2d PM |
2714 | err = security_tun_dev_alloc_security(&tun->security); |
2715 | if (err < 0) | |
608b9977 | 2716 | goto err_free_stat; |
2b980dbd | 2717 | |
1da177e4 | 2718 | tun_net_init(dev); |
944a1376 | 2719 | tun_flow_init(tun); |
96442e42 | 2720 | |
88255375 | 2721 | dev->hw_features = NETIF_F_SG | NETIF_F_FRAGLIST | |
6680ec68 JW |
2722 | TUN_USER_FEATURES | NETIF_F_HW_VLAN_CTAG_TX | |
2723 | NETIF_F_HW_VLAN_STAG_TX; | |
2a2bbf17 | 2724 | dev->features = dev->hw_features | NETIF_F_LLTX; |
6671b224 FLVC |
2725 | dev->vlan_features = dev->features & |
2726 | ~(NETIF_F_HW_VLAN_CTAG_TX | | |
2727 | NETIF_F_HW_VLAN_STAG_TX); | |
88255375 | 2728 | |
9fffc5c6 SD |
2729 | tun->flags = (tun->flags & ~TUN_FEATURES) | |
2730 | (ifr->ifr_flags & TUN_FEATURES); | |
2731 | ||
4008e97f | 2732 | INIT_LIST_HEAD(&tun->disabled); |
af3fb24e | 2733 | err = tun_attach(tun, file, false, ifr->ifr_flags & IFF_NAPI, |
77f22f92 | 2734 | ifr->ifr_flags & IFF_NAPI_FRAGS, false); |
eb0fb363 | 2735 | if (err < 0) |
662ca437 | 2736 | goto err_free_flow; |
eb0fb363 | 2737 | |
1da177e4 LT |
2738 | err = register_netdevice(tun->dev); |
2739 | if (err < 0) | |
662ca437 | 2740 | goto err_detach; |
77f22f92 YY |
2741 | /* free_netdev() won't check refcnt, to aovid race |
2742 | * with dev_put() we need publish tun after registration. | |
2743 | */ | |
2744 | rcu_assign_pointer(tfile->tun, tun); | |
1da177e4 LT |
2745 | } |
2746 | ||
af668b3c MT |
2747 | netif_carrier_on(tun->dev); |
2748 | ||
e35259a9 MK |
2749 | /* Make sure persistent devices do not get stuck in |
2750 | * xoff state. | |
2751 | */ | |
2752 | if (netif_running(tun->dev)) | |
c8d68e6b | 2753 | netif_tx_wake_all_queues(tun->dev); |
e35259a9 | 2754 | |
1da177e4 LT |
2755 | strcpy(ifr->ifr_name, tun->dev->name); |
2756 | return 0; | |
2757 | ||
662ca437 JW |
2758 | err_detach: |
2759 | tun_detach_all(dev); | |
11fc7d5a ED |
2760 | /* We are here because register_netdevice() has failed. |
2761 | * If register_netdevice() already called tun_free_netdev() | |
497a5757 | 2762 | * while dealing with the error, dev->stats has been cleared. |
11fc7d5a | 2763 | */ |
497a5757 | 2764 | if (!dev->tstats) |
11fc7d5a | 2765 | goto err_free_dev; |
ff244c6b | 2766 | |
662ca437 JW |
2767 | err_free_flow: |
2768 | tun_flow_uninit(tun); | |
2769 | security_tun_dev_free_security(tun->security); | |
608b9977 | 2770 | err_free_stat: |
497a5757 | 2771 | free_percpu(dev->tstats); |
662ca437 | 2772 | err_free_dev: |
1da177e4 | 2773 | free_netdev(dev); |
1da177e4 LT |
2774 | return err; |
2775 | } | |
2776 | ||
12132768 | 2777 | static void tun_get_iff(struct tun_struct *tun, struct ifreq *ifr) |
e3b99556 | 2778 | { |
e3b99556 MM |
2779 | strcpy(ifr->ifr_name, tun->dev->name); |
2780 | ||
980c9e8c | 2781 | ifr->ifr_flags = tun_flags(tun); |
e3b99556 | 2782 | |
e3b99556 MM |
2783 | } |
2784 | ||
5228ddc9 RR |
2785 | /* This is like a cut-down ethtool ops, except done via tun fd so no |
2786 | * privs required. */ | |
88255375 | 2787 | static int set_offload(struct tun_struct *tun, unsigned long arg) |
5228ddc9 | 2788 | { |
c8f44aff | 2789 | netdev_features_t features = 0; |
5228ddc9 RR |
2790 | |
2791 | if (arg & TUN_F_CSUM) { | |
88255375 | 2792 | features |= NETIF_F_HW_CSUM; |
5228ddc9 RR |
2793 | arg &= ~TUN_F_CSUM; |
2794 | ||
2795 | if (arg & (TUN_F_TSO4|TUN_F_TSO6)) { | |
2796 | if (arg & TUN_F_TSO_ECN) { | |
2797 | features |= NETIF_F_TSO_ECN; | |
2798 | arg &= ~TUN_F_TSO_ECN; | |
2799 | } | |
2800 | if (arg & TUN_F_TSO4) | |
2801 | features |= NETIF_F_TSO; | |
2802 | if (arg & TUN_F_TSO6) | |
2803 | features |= NETIF_F_TSO6; | |
2804 | arg &= ~(TUN_F_TSO4|TUN_F_TSO6); | |
2805 | } | |
0c19f846 WB |
2806 | |
2807 | arg &= ~TUN_F_UFO; | |
5228ddc9 RR |
2808 | } |
2809 | ||
2810 | /* This gives the user a way to test for new features in future by | |
2811 | * trying to set them. */ | |
2812 | if (arg) | |
2813 | return -EINVAL; | |
2814 | ||
88255375 | 2815 | tun->set_features = features; |
09050957 YI |
2816 | tun->dev->wanted_features &= ~TUN_USER_FEATURES; |
2817 | tun->dev->wanted_features |= features; | |
88255375 | 2818 | netdev_update_features(tun->dev); |
5228ddc9 RR |
2819 | |
2820 | return 0; | |
2821 | } | |
2822 | ||
c8d68e6b JW |
2823 | static void tun_detach_filter(struct tun_struct *tun, int n) |
2824 | { | |
2825 | int i; | |
2826 | struct tun_file *tfile; | |
2827 | ||
2828 | for (i = 0; i < n; i++) { | |
b8deabd3 | 2829 | tfile = rtnl_dereference(tun->tfiles[i]); |
8ced425e HFS |
2830 | lock_sock(tfile->socket.sk); |
2831 | sk_detach_filter(tfile->socket.sk); | |
2832 | release_sock(tfile->socket.sk); | |
c8d68e6b JW |
2833 | } |
2834 | ||
2835 | tun->filter_attached = false; | |
2836 | } | |
2837 | ||
2838 | static int tun_attach_filter(struct tun_struct *tun) | |
2839 | { | |
2840 | int i, ret = 0; | |
2841 | struct tun_file *tfile; | |
2842 | ||
2843 | for (i = 0; i < tun->numqueues; i++) { | |
b8deabd3 | 2844 | tfile = rtnl_dereference(tun->tfiles[i]); |
8ced425e HFS |
2845 | lock_sock(tfile->socket.sk); |
2846 | ret = sk_attach_filter(&tun->fprog, tfile->socket.sk); | |
2847 | release_sock(tfile->socket.sk); | |
c8d68e6b JW |
2848 | if (ret) { |
2849 | tun_detach_filter(tun, i); | |
2850 | return ret; | |
2851 | } | |
2852 | } | |
2853 | ||
2854 | tun->filter_attached = true; | |
2855 | return ret; | |
2856 | } | |
2857 | ||
2858 | static void tun_set_sndbuf(struct tun_struct *tun) | |
2859 | { | |
2860 | struct tun_file *tfile; | |
2861 | int i; | |
2862 | ||
2863 | for (i = 0; i < tun->numqueues; i++) { | |
b8deabd3 | 2864 | tfile = rtnl_dereference(tun->tfiles[i]); |
c8d68e6b JW |
2865 | tfile->socket.sk->sk_sndbuf = tun->sndbuf; |
2866 | } | |
2867 | } | |
2868 | ||
cde8b15f JW |
2869 | static int tun_set_queue(struct file *file, struct ifreq *ifr) |
2870 | { | |
2871 | struct tun_file *tfile = file->private_data; | |
2872 | struct tun_struct *tun; | |
cde8b15f JW |
2873 | int ret = 0; |
2874 | ||
2875 | rtnl_lock(); | |
2876 | ||
2877 | if (ifr->ifr_flags & IFF_ATTACH_QUEUE) { | |
4008e97f | 2878 | tun = tfile->detached; |
5dbbaf2d | 2879 | if (!tun) { |
cde8b15f | 2880 | ret = -EINVAL; |
5dbbaf2d PM |
2881 | goto unlock; |
2882 | } | |
2883 | ret = security_tun_dev_attach_queue(tun->security); | |
2884 | if (ret < 0) | |
2885 | goto unlock; | |
af3fb24e | 2886 | ret = tun_attach(tun, file, false, tun->flags & IFF_NAPI, |
77f22f92 | 2887 | tun->flags & IFF_NAPI_FRAGS, true); |
4008e97f | 2888 | } else if (ifr->ifr_flags & IFF_DETACH_QUEUE) { |
b8deabd3 | 2889 | tun = rtnl_dereference(tfile->tun); |
40630b82 | 2890 | if (!tun || !(tun->flags & IFF_MULTI_QUEUE) || tfile->detached) |
4008e97f JW |
2891 | ret = -EINVAL; |
2892 | else | |
2893 | __tun_detach(tfile, false); | |
2894 | } else | |
cde8b15f JW |
2895 | ret = -EINVAL; |
2896 | ||
83c1f36f SD |
2897 | if (ret >= 0) |
2898 | netdev_state_change(tun->dev); | |
2899 | ||
5dbbaf2d | 2900 | unlock: |
cde8b15f JW |
2901 | rtnl_unlock(); |
2902 | return ret; | |
2903 | } | |
2904 | ||
8f3f330d | 2905 | static int tun_set_ebpf(struct tun_struct *tun, struct tun_prog __rcu **prog_p, |
cd5681d7 | 2906 | void __user *data) |
96f84061 JW |
2907 | { |
2908 | struct bpf_prog *prog; | |
2909 | int fd; | |
2910 | ||
2911 | if (copy_from_user(&fd, data, sizeof(fd))) | |
2912 | return -EFAULT; | |
2913 | ||
2914 | if (fd == -1) { | |
2915 | prog = NULL; | |
2916 | } else { | |
2917 | prog = bpf_prog_get_type(fd, BPF_PROG_TYPE_SOCKET_FILTER); | |
2918 | if (IS_ERR(prog)) | |
2919 | return PTR_ERR(prog); | |
2920 | } | |
2921 | ||
cd5681d7 | 2922 | return __tun_set_ebpf(tun, prog_p, prog); |
96f84061 JW |
2923 | } |
2924 | ||
50857e2a AB |
2925 | static long __tun_chr_ioctl(struct file *file, unsigned int cmd, |
2926 | unsigned long arg, int ifreq_len) | |
1da177e4 | 2927 | { |
36b50bab | 2928 | struct tun_file *tfile = file->private_data; |
f663706a | 2929 | struct net *net = sock_net(&tfile->sk); |
631ab46b | 2930 | struct tun_struct *tun; |
1da177e4 | 2931 | void __user* argp = (void __user*)arg; |
26d31925 | 2932 | unsigned int ifindex, carrier; |
1da177e4 | 2933 | struct ifreq ifr; |
0625c883 EB |
2934 | kuid_t owner; |
2935 | kgid_t group; | |
33dccbb0 | 2936 | int sndbuf; |
d9d52b51 | 2937 | int vnet_hdr_sz; |
1cf8e410 | 2938 | int le; |
f271b2cc | 2939 | int ret; |
83c1f36f | 2940 | bool do_notify = false; |
1da177e4 | 2941 | |
f2780d6d KT |
2942 | if (cmd == TUNSETIFF || cmd == TUNSETQUEUE || |
2943 | (_IOC_TYPE(cmd) == SOCK_IOC_TYPE && cmd != SIOCGSKNS)) { | |
50857e2a | 2944 | if (copy_from_user(&ifr, argp, ifreq_len)) |
1da177e4 | 2945 | return -EFAULT; |
8bbb1813 | 2946 | } else { |
a117dacd | 2947 | memset(&ifr, 0, sizeof(ifr)); |
8bbb1813 | 2948 | } |
631ab46b EB |
2949 | if (cmd == TUNGETFEATURES) { |
2950 | /* Currently this just means: "what IFF flags are valid?". | |
2951 | * This is needed because we never checked for invalid flags on | |
031f5e03 MT |
2952 | * TUNSETIFF. |
2953 | */ | |
2954 | return put_user(IFF_TUN | IFF_TAP | TUN_FEATURES, | |
631ab46b | 2955 | (unsigned int __user*)argp); |
f663706a | 2956 | } else if (cmd == TUNSETQUEUE) { |
cde8b15f | 2957 | return tun_set_queue(file, &ifr); |
f663706a KT |
2958 | } else if (cmd == SIOCGSKNS) { |
2959 | if (!ns_capable(net->user_ns, CAP_NET_ADMIN)) | |
2960 | return -EPERM; | |
2961 | return open_related_ns(&net->ns, get_net_ns); | |
2962 | } | |
631ab46b | 2963 | |
c8d68e6b | 2964 | ret = 0; |
876bfd4d HX |
2965 | rtnl_lock(); |
2966 | ||
9484dc74 | 2967 | tun = tun_get(tfile); |
0f16bc13 GF |
2968 | if (cmd == TUNSETIFF) { |
2969 | ret = -EEXIST; | |
2970 | if (tun) | |
2971 | goto unlock; | |
2972 | ||
1da177e4 LT |
2973 | ifr.ifr_name[IFNAMSIZ-1] = '\0'; |
2974 | ||
f2780d6d | 2975 | ret = tun_set_iff(net, file, &ifr); |
1da177e4 | 2976 | |
876bfd4d HX |
2977 | if (ret) |
2978 | goto unlock; | |
1da177e4 | 2979 | |
50857e2a | 2980 | if (copy_to_user(argp, &ifr, ifreq_len)) |
876bfd4d HX |
2981 | ret = -EFAULT; |
2982 | goto unlock; | |
1da177e4 | 2983 | } |
fb7589a1 PE |
2984 | if (cmd == TUNSETIFINDEX) { |
2985 | ret = -EPERM; | |
2986 | if (tun) | |
2987 | goto unlock; | |
2988 | ||
2989 | ret = -EFAULT; | |
2990 | if (copy_from_user(&ifindex, argp, sizeof(ifindex))) | |
2991 | goto unlock; | |
2992 | ||
2993 | ret = 0; | |
2994 | tfile->ifindex = ifindex; | |
2995 | goto unlock; | |
2996 | } | |
1da177e4 | 2997 | |
876bfd4d | 2998 | ret = -EBADFD; |
1da177e4 | 2999 | if (!tun) |
876bfd4d | 3000 | goto unlock; |
1da177e4 | 3001 | |
3424170f | 3002 | netif_info(tun, drv, tun->dev, "tun_chr_ioctl cmd %u\n", cmd); |
1da177e4 | 3003 | |
0c3e0e3b | 3004 | net = dev_net(tun->dev); |
631ab46b | 3005 | ret = 0; |
1da177e4 | 3006 | switch (cmd) { |
e3b99556 | 3007 | case TUNGETIFF: |
12132768 | 3008 | tun_get_iff(tun, &ifr); |
e3b99556 | 3009 | |
3d407a80 PE |
3010 | if (tfile->detached) |
3011 | ifr.ifr_flags |= IFF_DETACH_QUEUE; | |
849c9b6f PE |
3012 | if (!tfile->socket.sk->sk_filter) |
3013 | ifr.ifr_flags |= IFF_NOFILTER; | |
3d407a80 | 3014 | |
50857e2a | 3015 | if (copy_to_user(argp, &ifr, ifreq_len)) |
631ab46b | 3016 | ret = -EFAULT; |
e3b99556 MM |
3017 | break; |
3018 | ||
1da177e4 LT |
3019 | case TUNSETNOCSUM: |
3020 | /* Disable/Enable checksum */ | |
1da177e4 | 3021 | |
88255375 | 3022 | /* [unimplemented] */ |
3424170f MK |
3023 | netif_info(tun, drv, tun->dev, "ignored: set checksum %s\n", |
3024 | arg ? "disabled" : "enabled"); | |
1da177e4 LT |
3025 | break; |
3026 | ||
3027 | case TUNSETPERSIST: | |
54f968d6 JW |
3028 | /* Disable/Enable persist mode. Keep an extra reference to the |
3029 | * module to prevent the module being unprobed. | |
3030 | */ | |
40630b82 MT |
3031 | if (arg && !(tun->flags & IFF_PERSIST)) { |
3032 | tun->flags |= IFF_PERSIST; | |
54f968d6 | 3033 | __module_get(THIS_MODULE); |
83c1f36f | 3034 | do_notify = true; |
dd38bd85 | 3035 | } |
40630b82 MT |
3036 | if (!arg && (tun->flags & IFF_PERSIST)) { |
3037 | tun->flags &= ~IFF_PERSIST; | |
54f968d6 | 3038 | module_put(THIS_MODULE); |
83c1f36f | 3039 | do_notify = true; |
54f968d6 | 3040 | } |
1da177e4 | 3041 | |
3424170f MK |
3042 | netif_info(tun, drv, tun->dev, "persist %s\n", |
3043 | arg ? "enabled" : "disabled"); | |
1da177e4 LT |
3044 | break; |
3045 | ||
3046 | case TUNSETOWNER: | |
3047 | /* Set owner of the device */ | |
0625c883 EB |
3048 | owner = make_kuid(current_user_ns(), arg); |
3049 | if (!uid_valid(owner)) { | |
3050 | ret = -EINVAL; | |
3051 | break; | |
3052 | } | |
3053 | tun->owner = owner; | |
83c1f36f | 3054 | do_notify = true; |
3424170f MK |
3055 | netif_info(tun, drv, tun->dev, "owner set to %u\n", |
3056 | from_kuid(&init_user_ns, tun->owner)); | |
1da177e4 LT |
3057 | break; |
3058 | ||
8c644623 GG |
3059 | case TUNSETGROUP: |
3060 | /* Set group of the device */ | |
0625c883 EB |
3061 | group = make_kgid(current_user_ns(), arg); |
3062 | if (!gid_valid(group)) { | |
3063 | ret = -EINVAL; | |
3064 | break; | |
3065 | } | |
3066 | tun->group = group; | |
83c1f36f | 3067 | do_notify = true; |
3424170f MK |
3068 | netif_info(tun, drv, tun->dev, "group set to %u\n", |
3069 | from_kgid(&init_user_ns, tun->group)); | |
8c644623 GG |
3070 | break; |
3071 | ||
ff4cc3ac MK |
3072 | case TUNSETLINK: |
3073 | /* Only allow setting the type when the interface is down */ | |
3074 | if (tun->dev->flags & IFF_UP) { | |
3424170f MK |
3075 | netif_info(tun, drv, tun->dev, |
3076 | "Linktype set failed because interface is up\n"); | |
48abfe05 | 3077 | ret = -EBUSY; |
ff4cc3ac | 3078 | } else { |
8e1e33ff MS |
3079 | ret = call_netdevice_notifiers(NETDEV_PRE_TYPE_CHANGE, |
3080 | tun->dev); | |
3081 | ret = notifier_to_errno(ret); | |
3082 | if (ret) { | |
3083 | netif_info(tun, drv, tun->dev, | |
3084 | "Refused to change device type\n"); | |
3085 | break; | |
3086 | } | |
ff4cc3ac | 3087 | tun->dev->type = (int) arg; |
3424170f MK |
3088 | netif_info(tun, drv, tun->dev, "linktype set to %d\n", |
3089 | tun->dev->type); | |
8e1e33ff MS |
3090 | call_netdevice_notifiers(NETDEV_POST_TYPE_CHANGE, |
3091 | tun->dev); | |
ff4cc3ac | 3092 | } |
631ab46b | 3093 | break; |
ff4cc3ac | 3094 | |
1da177e4 | 3095 | case TUNSETDEBUG: |
3424170f | 3096 | tun->msg_enable = (u32)arg; |
1da177e4 | 3097 | break; |
3424170f | 3098 | |
5228ddc9 | 3099 | case TUNSETOFFLOAD: |
88255375 | 3100 | ret = set_offload(tun, arg); |
631ab46b | 3101 | break; |
5228ddc9 | 3102 | |
f271b2cc MK |
3103 | case TUNSETTXFILTER: |
3104 | /* Can be set only for TAPs */ | |
631ab46b | 3105 | ret = -EINVAL; |
40630b82 | 3106 | if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP) |
631ab46b | 3107 | break; |
c0e5a8c2 | 3108 | ret = update_filter(&tun->txflt, (void __user *)arg); |
631ab46b | 3109 | break; |
1da177e4 LT |
3110 | |
3111 | case SIOCGIFHWADDR: | |
b595076a | 3112 | /* Get hw address */ |
f271b2cc MK |
3113 | memcpy(ifr.ifr_hwaddr.sa_data, tun->dev->dev_addr, ETH_ALEN); |
3114 | ifr.ifr_hwaddr.sa_family = tun->dev->type; | |
50857e2a | 3115 | if (copy_to_user(argp, &ifr, ifreq_len)) |
631ab46b EB |
3116 | ret = -EFAULT; |
3117 | break; | |
1da177e4 LT |
3118 | |
3119 | case SIOCSIFHWADDR: | |
f271b2cc | 3120 | /* Set hw address */ |
3a37a963 | 3121 | ret = dev_set_mac_address(tun->dev, &ifr.ifr_hwaddr, NULL); |
631ab46b | 3122 | break; |
33dccbb0 HX |
3123 | |
3124 | case TUNGETSNDBUF: | |
54f968d6 | 3125 | sndbuf = tfile->socket.sk->sk_sndbuf; |
33dccbb0 HX |
3126 | if (copy_to_user(argp, &sndbuf, sizeof(sndbuf))) |
3127 | ret = -EFAULT; | |
3128 | break; | |
3129 | ||
3130 | case TUNSETSNDBUF: | |
3131 | if (copy_from_user(&sndbuf, argp, sizeof(sndbuf))) { | |
3132 | ret = -EFAULT; | |
3133 | break; | |
3134 | } | |
93161922 CG |
3135 | if (sndbuf <= 0) { |
3136 | ret = -EINVAL; | |
3137 | break; | |
3138 | } | |
33dccbb0 | 3139 | |
c8d68e6b JW |
3140 | tun->sndbuf = sndbuf; |
3141 | tun_set_sndbuf(tun); | |
33dccbb0 HX |
3142 | break; |
3143 | ||
d9d52b51 MT |
3144 | case TUNGETVNETHDRSZ: |
3145 | vnet_hdr_sz = tun->vnet_hdr_sz; | |
3146 | if (copy_to_user(argp, &vnet_hdr_sz, sizeof(vnet_hdr_sz))) | |
3147 | ret = -EFAULT; | |
3148 | break; | |
3149 | ||
3150 | case TUNSETVNETHDRSZ: | |
3151 | if (copy_from_user(&vnet_hdr_sz, argp, sizeof(vnet_hdr_sz))) { | |
3152 | ret = -EFAULT; | |
3153 | break; | |
3154 | } | |
3155 | if (vnet_hdr_sz < (int)sizeof(struct virtio_net_hdr)) { | |
3156 | ret = -EINVAL; | |
3157 | break; | |
3158 | } | |
3159 | ||
3160 | tun->vnet_hdr_sz = vnet_hdr_sz; | |
3161 | break; | |
3162 | ||
1cf8e410 MT |
3163 | case TUNGETVNETLE: |
3164 | le = !!(tun->flags & TUN_VNET_LE); | |
3165 | if (put_user(le, (int __user *)argp)) | |
3166 | ret = -EFAULT; | |
3167 | break; | |
3168 | ||
3169 | case TUNSETVNETLE: | |
3170 | if (get_user(le, (int __user *)argp)) { | |
3171 | ret = -EFAULT; | |
3172 | break; | |
3173 | } | |
3174 | if (le) | |
3175 | tun->flags |= TUN_VNET_LE; | |
3176 | else | |
3177 | tun->flags &= ~TUN_VNET_LE; | |
3178 | break; | |
3179 | ||
8b8e658b GK |
3180 | case TUNGETVNETBE: |
3181 | ret = tun_get_vnet_be(tun, argp); | |
3182 | break; | |
3183 | ||
3184 | case TUNSETVNETBE: | |
3185 | ret = tun_set_vnet_be(tun, argp); | |
3186 | break; | |
3187 | ||
99405162 MT |
3188 | case TUNATTACHFILTER: |
3189 | /* Can be set only for TAPs */ | |
3190 | ret = -EINVAL; | |
40630b82 | 3191 | if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP) |
99405162 MT |
3192 | break; |
3193 | ret = -EFAULT; | |
54f968d6 | 3194 | if (copy_from_user(&tun->fprog, argp, sizeof(tun->fprog))) |
99405162 MT |
3195 | break; |
3196 | ||
c8d68e6b | 3197 | ret = tun_attach_filter(tun); |
99405162 MT |
3198 | break; |
3199 | ||
3200 | case TUNDETACHFILTER: | |
3201 | /* Can be set only for TAPs */ | |
3202 | ret = -EINVAL; | |
40630b82 | 3203 | if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP) |
99405162 | 3204 | break; |
c8d68e6b JW |
3205 | ret = 0; |
3206 | tun_detach_filter(tun, tun->numqueues); | |
99405162 MT |
3207 | break; |
3208 | ||
76975e9c PE |
3209 | case TUNGETFILTER: |
3210 | ret = -EINVAL; | |
40630b82 | 3211 | if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP) |
76975e9c PE |
3212 | break; |
3213 | ret = -EFAULT; | |
3214 | if (copy_to_user(argp, &tun->fprog, sizeof(tun->fprog))) | |
3215 | break; | |
3216 | ret = 0; | |
3217 | break; | |
3218 | ||
96f84061 | 3219 | case TUNSETSTEERINGEBPF: |
cd5681d7 | 3220 | ret = tun_set_ebpf(tun, &tun->steering_prog, argp); |
96f84061 JW |
3221 | break; |
3222 | ||
aff3d70a JW |
3223 | case TUNSETFILTEREBPF: |
3224 | ret = tun_set_ebpf(tun, &tun->filter_prog, argp); | |
3225 | break; | |
3226 | ||
26d31925 ND |
3227 | case TUNSETCARRIER: |
3228 | ret = -EFAULT; | |
3229 | if (copy_from_user(&carrier, argp, sizeof(carrier))) | |
3230 | goto unlock; | |
3231 | ||
3232 | ret = tun_net_change_carrier(tun->dev, (bool)carrier); | |
3233 | break; | |
3234 | ||
0c3e0e3b KT |
3235 | case TUNGETDEVNETNS: |
3236 | ret = -EPERM; | |
3237 | if (!ns_capable(net->user_ns, CAP_NET_ADMIN)) | |
3238 | goto unlock; | |
3239 | ret = open_related_ns(&net->ns, get_net_ns); | |
3240 | break; | |
3241 | ||
1da177e4 | 3242 | default: |
631ab46b EB |
3243 | ret = -EINVAL; |
3244 | break; | |
ee289b64 | 3245 | } |
1da177e4 | 3246 | |
83c1f36f SD |
3247 | if (do_notify) |
3248 | netdev_state_change(tun->dev); | |
3249 | ||
876bfd4d HX |
3250 | unlock: |
3251 | rtnl_unlock(); | |
3252 | if (tun) | |
3253 | tun_put(tun); | |
631ab46b | 3254 | return ret; |
1da177e4 LT |
3255 | } |
3256 | ||
50857e2a AB |
3257 | static long tun_chr_ioctl(struct file *file, |
3258 | unsigned int cmd, unsigned long arg) | |
3259 | { | |
3260 | return __tun_chr_ioctl(file, cmd, arg, sizeof (struct ifreq)); | |
3261 | } | |
3262 | ||
3263 | #ifdef CONFIG_COMPAT | |
3264 | static long tun_chr_compat_ioctl(struct file *file, | |
3265 | unsigned int cmd, unsigned long arg) | |
3266 | { | |
3267 | switch (cmd) { | |
3268 | case TUNSETIFF: | |
3269 | case TUNGETIFF: | |
3270 | case TUNSETTXFILTER: | |
3271 | case TUNGETSNDBUF: | |
3272 | case TUNSETSNDBUF: | |
3273 | case SIOCGIFHWADDR: | |
3274 | case SIOCSIFHWADDR: | |
3275 | arg = (unsigned long)compat_ptr(arg); | |
3276 | break; | |
3277 | default: | |
3278 | arg = (compat_ulong_t)arg; | |
3279 | break; | |
3280 | } | |
3281 | ||
3282 | /* | |
3283 | * compat_ifreq is shorter than ifreq, so we must not access beyond | |
3284 | * the end of that structure. All fields that are used in this | |
3285 | * driver are compatible though, we don't need to convert the | |
3286 | * contents. | |
3287 | */ | |
3288 | return __tun_chr_ioctl(file, cmd, arg, sizeof(struct compat_ifreq)); | |
3289 | } | |
3290 | #endif /* CONFIG_COMPAT */ | |
3291 | ||
1da177e4 LT |
3292 | static int tun_chr_fasync(int fd, struct file *file, int on) |
3293 | { | |
54f968d6 | 3294 | struct tun_file *tfile = file->private_data; |
1da177e4 LT |
3295 | int ret; |
3296 | ||
54f968d6 | 3297 | if ((ret = fasync_helper(fd, file, on, &tfile->fasync)) < 0) |
9d319522 | 3298 | goto out; |
6aa20a22 | 3299 | |
1da177e4 | 3300 | if (on) { |
01919134 | 3301 | __f_setown(file, task_pid(current), PIDTYPE_TGID, 0); |
54f968d6 | 3302 | tfile->flags |= TUN_FASYNC; |
6aa20a22 | 3303 | } else |
54f968d6 | 3304 | tfile->flags &= ~TUN_FASYNC; |
9d319522 JC |
3305 | ret = 0; |
3306 | out: | |
9d319522 | 3307 | return ret; |
1da177e4 LT |
3308 | } |
3309 | ||
3310 | static int tun_chr_open(struct inode *inode, struct file * file) | |
3311 | { | |
140e807d | 3312 | struct net *net = current->nsproxy->net_ns; |
631ab46b | 3313 | struct tun_file *tfile; |
deed49fb | 3314 | |
140e807d | 3315 | tfile = (struct tun_file *)sk_alloc(net, AF_UNSPEC, GFP_KERNEL, |
11aa9c28 | 3316 | &tun_proto, 0); |
631ab46b EB |
3317 | if (!tfile) |
3318 | return -ENOMEM; | |
b196d88a JW |
3319 | if (ptr_ring_init(&tfile->tx_ring, 0, GFP_KERNEL)) { |
3320 | sk_free(&tfile->sk); | |
3321 | return -ENOMEM; | |
3322 | } | |
3323 | ||
c7256f57 | 3324 | mutex_init(&tfile->napi_mutex); |
c956674b | 3325 | RCU_INIT_POINTER(tfile->tun, NULL); |
54f968d6 | 3326 | tfile->flags = 0; |
fb7589a1 | 3327 | tfile->ifindex = 0; |
54f968d6 | 3328 | |
333f7909 | 3329 | init_waitqueue_head(&tfile->socket.wq.wait); |
54f968d6 JW |
3330 | |
3331 | tfile->socket.file = file; | |
3332 | tfile->socket.ops = &tun_socket_ops; | |
3333 | ||
3334 | sock_init_data(&tfile->socket, &tfile->sk); | |
54f968d6 JW |
3335 | |
3336 | tfile->sk.sk_write_space = tun_sock_write_space; | |
3337 | tfile->sk.sk_sndbuf = INT_MAX; | |
3338 | ||
631ab46b | 3339 | file->private_data = tfile; |
4008e97f | 3340 | INIT_LIST_HEAD(&tfile->next); |
54f968d6 | 3341 | |
19a6afb2 JW |
3342 | sock_set_flag(&tfile->sk, SOCK_ZEROCOPY); |
3343 | ||
1da177e4 LT |
3344 | return 0; |
3345 | } | |
3346 | ||
3347 | static int tun_chr_close(struct inode *inode, struct file *file) | |
3348 | { | |
631ab46b | 3349 | struct tun_file *tfile = file->private_data; |
1da177e4 | 3350 | |
c8d68e6b | 3351 | tun_detach(tfile, true); |
1da177e4 LT |
3352 | |
3353 | return 0; | |
3354 | } | |
3355 | ||
93e14b6d | 3356 | #ifdef CONFIG_PROC_FS |
9484dc74 | 3357 | static void tun_chr_show_fdinfo(struct seq_file *m, struct file *file) |
93e14b6d | 3358 | { |
9484dc74 | 3359 | struct tun_file *tfile = file->private_data; |
93e14b6d MY |
3360 | struct tun_struct *tun; |
3361 | struct ifreq ifr; | |
3362 | ||
3363 | memset(&ifr, 0, sizeof(ifr)); | |
3364 | ||
3365 | rtnl_lock(); | |
9484dc74 | 3366 | tun = tun_get(tfile); |
93e14b6d | 3367 | if (tun) |
12132768 | 3368 | tun_get_iff(tun, &ifr); |
93e14b6d MY |
3369 | rtnl_unlock(); |
3370 | ||
3371 | if (tun) | |
3372 | tun_put(tun); | |
3373 | ||
a3816ab0 | 3374 | seq_printf(m, "iff:\t%s\n", ifr.ifr_name); |
93e14b6d MY |
3375 | } |
3376 | #endif | |
3377 | ||
d54b1fdb | 3378 | static const struct file_operations tun_fops = { |
6aa20a22 | 3379 | .owner = THIS_MODULE, |
1da177e4 | 3380 | .llseek = no_llseek, |
9b067034 | 3381 | .read_iter = tun_chr_read_iter, |
f5ff53b4 | 3382 | .write_iter = tun_chr_write_iter, |
1da177e4 | 3383 | .poll = tun_chr_poll, |
50857e2a AB |
3384 | .unlocked_ioctl = tun_chr_ioctl, |
3385 | #ifdef CONFIG_COMPAT | |
3386 | .compat_ioctl = tun_chr_compat_ioctl, | |
3387 | #endif | |
1da177e4 LT |
3388 | .open = tun_chr_open, |
3389 | .release = tun_chr_close, | |
93e14b6d MY |
3390 | .fasync = tun_chr_fasync, |
3391 | #ifdef CONFIG_PROC_FS | |
3392 | .show_fdinfo = tun_chr_show_fdinfo, | |
3393 | #endif | |
1da177e4 LT |
3394 | }; |
3395 | ||
3396 | static struct miscdevice tun_miscdev = { | |
3397 | .minor = TUN_MINOR, | |
3398 | .name = "tun", | |
e454cea2 | 3399 | .nodename = "net/tun", |
1da177e4 | 3400 | .fops = &tun_fops, |
1da177e4 LT |
3401 | }; |
3402 | ||
3403 | /* ethtool interface */ | |
3404 | ||
4e24f2dd CW |
3405 | static void tun_default_link_ksettings(struct net_device *dev, |
3406 | struct ethtool_link_ksettings *cmd) | |
29ccc49d PR |
3407 | { |
3408 | ethtool_link_ksettings_zero_link_mode(cmd, supported); | |
3409 | ethtool_link_ksettings_zero_link_mode(cmd, advertising); | |
3410 | cmd->base.speed = SPEED_10; | |
3411 | cmd->base.duplex = DUPLEX_FULL; | |
3412 | cmd->base.port = PORT_TP; | |
3413 | cmd->base.phy_address = 0; | |
3414 | cmd->base.autoneg = AUTONEG_DISABLE; | |
4e24f2dd CW |
3415 | } |
3416 | ||
3417 | static int tun_get_link_ksettings(struct net_device *dev, | |
3418 | struct ethtool_link_ksettings *cmd) | |
3419 | { | |
3420 | struct tun_struct *tun = netdev_priv(dev); | |
3421 | ||
3422 | memcpy(cmd, &tun->link_ksettings, sizeof(*cmd)); | |
3423 | return 0; | |
3424 | } | |
3425 | ||
3426 | static int tun_set_link_ksettings(struct net_device *dev, | |
3427 | const struct ethtool_link_ksettings *cmd) | |
3428 | { | |
3429 | struct tun_struct *tun = netdev_priv(dev); | |
3430 | ||
3431 | memcpy(&tun->link_ksettings, cmd, sizeof(*cmd)); | |
1da177e4 LT |
3432 | return 0; |
3433 | } | |
3434 | ||
3435 | static void tun_get_drvinfo(struct net_device *dev, struct ethtool_drvinfo *info) | |
3436 | { | |
3437 | struct tun_struct *tun = netdev_priv(dev); | |
3438 | ||
33a5ba14 RJ |
3439 | strlcpy(info->driver, DRV_NAME, sizeof(info->driver)); |
3440 | strlcpy(info->version, DRV_VERSION, sizeof(info->version)); | |
1da177e4 LT |
3441 | |
3442 | switch (tun->flags & TUN_TYPE_MASK) { | |
40630b82 | 3443 | case IFF_TUN: |
33a5ba14 | 3444 | strlcpy(info->bus_info, "tun", sizeof(info->bus_info)); |
1da177e4 | 3445 | break; |
40630b82 | 3446 | case IFF_TAP: |
33a5ba14 | 3447 | strlcpy(info->bus_info, "tap", sizeof(info->bus_info)); |
1da177e4 LT |
3448 | break; |
3449 | } | |
3450 | } | |
3451 | ||
3452 | static u32 tun_get_msglevel(struct net_device *dev) | |
3453 | { | |
1da177e4 | 3454 | struct tun_struct *tun = netdev_priv(dev); |
3424170f MK |
3455 | |
3456 | return tun->msg_enable; | |
1da177e4 LT |
3457 | } |
3458 | ||
3459 | static void tun_set_msglevel(struct net_device *dev, u32 value) | |
3460 | { | |
1da177e4 | 3461 | struct tun_struct *tun = netdev_priv(dev); |
3424170f MK |
3462 | |
3463 | tun->msg_enable = value; | |
1da177e4 LT |
3464 | } |
3465 | ||
5503fcec JW |
3466 | static int tun_get_coalesce(struct net_device *dev, |
3467 | struct ethtool_coalesce *ec) | |
3468 | { | |
3469 | struct tun_struct *tun = netdev_priv(dev); | |
3470 | ||
3471 | ec->rx_max_coalesced_frames = tun->rx_batched; | |
3472 | ||
3473 | return 0; | |
3474 | } | |
3475 | ||
3476 | static int tun_set_coalesce(struct net_device *dev, | |
3477 | struct ethtool_coalesce *ec) | |
3478 | { | |
3479 | struct tun_struct *tun = netdev_priv(dev); | |
3480 | ||
3481 | if (ec->rx_max_coalesced_frames > NAPI_POLL_WEIGHT) | |
3482 | tun->rx_batched = NAPI_POLL_WEIGHT; | |
3483 | else | |
3484 | tun->rx_batched = ec->rx_max_coalesced_frames; | |
3485 | ||
3486 | return 0; | |
3487 | } | |
3488 | ||
7282d491 | 3489 | static const struct ethtool_ops tun_ethtool_ops = { |
e5ad00b3 | 3490 | .supported_coalesce_params = ETHTOOL_COALESCE_RX_MAX_FRAMES, |
1da177e4 LT |
3491 | .get_drvinfo = tun_get_drvinfo, |
3492 | .get_msglevel = tun_get_msglevel, | |
3493 | .set_msglevel = tun_set_msglevel, | |
bee31369 | 3494 | .get_link = ethtool_op_get_link, |
eda29772 | 3495 | .get_ts_info = ethtool_op_get_ts_info, |
5503fcec JW |
3496 | .get_coalesce = tun_get_coalesce, |
3497 | .set_coalesce = tun_set_coalesce, | |
29ccc49d | 3498 | .get_link_ksettings = tun_get_link_ksettings, |
4e24f2dd | 3499 | .set_link_ksettings = tun_set_link_ksettings, |
1da177e4 LT |
3500 | }; |
3501 | ||
1576d986 JW |
3502 | static int tun_queue_resize(struct tun_struct *tun) |
3503 | { | |
3504 | struct net_device *dev = tun->dev; | |
3505 | struct tun_file *tfile; | |
5990a305 | 3506 | struct ptr_ring **rings; |
1576d986 JW |
3507 | int n = tun->numqueues + tun->numdisabled; |
3508 | int ret, i; | |
3509 | ||
5990a305 JW |
3510 | rings = kmalloc_array(n, sizeof(*rings), GFP_KERNEL); |
3511 | if (!rings) | |
1576d986 JW |
3512 | return -ENOMEM; |
3513 | ||
3514 | for (i = 0; i < tun->numqueues; i++) { | |
3515 | tfile = rtnl_dereference(tun->tfiles[i]); | |
5990a305 | 3516 | rings[i] = &tfile->tx_ring; |
1576d986 JW |
3517 | } |
3518 | list_for_each_entry(tfile, &tun->disabled, next) | |
5990a305 | 3519 | rings[i++] = &tfile->tx_ring; |
1576d986 | 3520 | |
5990a305 JW |
3521 | ret = ptr_ring_resize_multiple(rings, n, |
3522 | dev->tx_queue_len, GFP_KERNEL, | |
fc72d1d5 | 3523 | tun_ptr_free); |
1576d986 | 3524 | |
5990a305 | 3525 | kfree(rings); |
1576d986 JW |
3526 | return ret; |
3527 | } | |
3528 | ||
3529 | static int tun_device_event(struct notifier_block *unused, | |
3530 | unsigned long event, void *ptr) | |
3531 | { | |
3532 | struct net_device *dev = netdev_notifier_info_to_dev(ptr); | |
3533 | struct tun_struct *tun = netdev_priv(dev); | |
72b319dc | 3534 | int i; |
1576d986 | 3535 | |
86dfb4ac CG |
3536 | if (dev->rtnl_link_ops != &tun_link_ops) |
3537 | return NOTIFY_DONE; | |
3538 | ||
1576d986 JW |
3539 | switch (event) { |
3540 | case NETDEV_CHANGE_TX_QUEUE_LEN: | |
3541 | if (tun_queue_resize(tun)) | |
3542 | return NOTIFY_BAD; | |
3543 | break; | |
72b319dc FL |
3544 | case NETDEV_UP: |
3545 | for (i = 0; i < tun->numqueues; i++) { | |
3546 | struct tun_file *tfile; | |
3547 | ||
3548 | tfile = rtnl_dereference(tun->tfiles[i]); | |
3549 | tfile->socket.sk->sk_write_space(tfile->socket.sk); | |
3550 | } | |
3551 | break; | |
1576d986 JW |
3552 | default: |
3553 | break; | |
3554 | } | |
3555 | ||
3556 | return NOTIFY_DONE; | |
3557 | } | |
3558 | ||
3559 | static struct notifier_block tun_notifier_block __read_mostly = { | |
3560 | .notifier_call = tun_device_event, | |
3561 | }; | |
79d17604 | 3562 | |
1da177e4 LT |
3563 | static int __init tun_init(void) |
3564 | { | |
3565 | int ret = 0; | |
3566 | ||
6b8a66ee | 3567 | pr_info("%s, %s\n", DRV_DESCRIPTION, DRV_VERSION); |
1da177e4 | 3568 | |
f019a7a5 | 3569 | ret = rtnl_link_register(&tun_link_ops); |
79d17604 | 3570 | if (ret) { |
6b8a66ee | 3571 | pr_err("Can't register link_ops\n"); |
f019a7a5 | 3572 | goto err_linkops; |
79d17604 PE |
3573 | } |
3574 | ||
1da177e4 | 3575 | ret = misc_register(&tun_miscdev); |
79d17604 | 3576 | if (ret) { |
6b8a66ee | 3577 | pr_err("Can't register misc device %d\n", TUN_MINOR); |
79d17604 PE |
3578 | goto err_misc; |
3579 | } | |
1576d986 | 3580 | |
5edfbd3c TZ |
3581 | ret = register_netdevice_notifier(&tun_notifier_block); |
3582 | if (ret) { | |
3583 | pr_err("Can't register netdevice notifier\n"); | |
3584 | goto err_notifier; | |
3585 | } | |
3586 | ||
f019a7a5 | 3587 | return 0; |
5edfbd3c TZ |
3588 | |
3589 | err_notifier: | |
3590 | misc_deregister(&tun_miscdev); | |
79d17604 | 3591 | err_misc: |
f019a7a5 EB |
3592 | rtnl_link_unregister(&tun_link_ops); |
3593 | err_linkops: | |
1da177e4 LT |
3594 | return ret; |
3595 | } | |
3596 | ||
3597 | static void tun_cleanup(void) | |
3598 | { | |
6aa20a22 | 3599 | misc_deregister(&tun_miscdev); |
f019a7a5 | 3600 | rtnl_link_unregister(&tun_link_ops); |
1576d986 | 3601 | unregister_netdevice_notifier(&tun_notifier_block); |
1da177e4 LT |
3602 | } |
3603 | ||
05c2828c MT |
3604 | /* Get an underlying socket object from tun file. Returns error unless file is |
3605 | * attached to a device. The returned object works like a packet socket, it | |
3606 | * can be used for sock_sendmsg/sock_recvmsg. The caller is responsible for | |
3607 | * holding a reference to the file for as long as the socket is in use. */ | |
3608 | struct socket *tun_get_socket(struct file *file) | |
3609 | { | |
6e914fc7 | 3610 | struct tun_file *tfile; |
05c2828c MT |
3611 | if (file->f_op != &tun_fops) |
3612 | return ERR_PTR(-EINVAL); | |
6e914fc7 JW |
3613 | tfile = file->private_data; |
3614 | if (!tfile) | |
05c2828c | 3615 | return ERR_PTR(-EBADFD); |
54f968d6 | 3616 | return &tfile->socket; |
05c2828c MT |
3617 | } |
3618 | EXPORT_SYMBOL_GPL(tun_get_socket); | |
3619 | ||
5990a305 | 3620 | struct ptr_ring *tun_get_tx_ring(struct file *file) |
83339c6b JW |
3621 | { |
3622 | struct tun_file *tfile; | |
3623 | ||
3624 | if (file->f_op != &tun_fops) | |
3625 | return ERR_PTR(-EINVAL); | |
3626 | tfile = file->private_data; | |
3627 | if (!tfile) | |
3628 | return ERR_PTR(-EBADFD); | |
5990a305 | 3629 | return &tfile->tx_ring; |
83339c6b | 3630 | } |
5990a305 | 3631 | EXPORT_SYMBOL_GPL(tun_get_tx_ring); |
83339c6b | 3632 | |
1da177e4 LT |
3633 | module_init(tun_init); |
3634 | module_exit(tun_cleanup); | |
3635 | MODULE_DESCRIPTION(DRV_DESCRIPTION); | |
3636 | MODULE_AUTHOR(DRV_COPYRIGHT); | |
3637 | MODULE_LICENSE("GPL"); | |
3638 | MODULE_ALIAS_MISCDEV(TUN_MINOR); | |
578454ff | 3639 | MODULE_ALIAS("devname:net/tun"); |