blob: 2823a4a6f059035fcbf1892aa714b4569d5a5119 [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
2 * TUN - Universal TUN/TAP device driver.
3 * Copyright (C) 1999-2002 Maxim Krasnyansky <maxk@qualcomm.com>
4 *
5 * This program is free software; you can redistribute it and/or modify
6 * it under the terms of the GNU General Public License as published by
7 * the Free Software Foundation; either version 2 of the License, or
8 * (at your option) any later version.
9 *
10 * This program is distributed in the hope that it will be useful,
11 * but WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
13 * GNU General Public License for more details.
14 *
15 * $Id: tun.c,v 1.15 2002/03/01 02:44:24 maxk Exp $
16 */
17
18/*
19 * Changes:
20 *
Mike Kershawff4cc3a2005-09-01 17:40:05 -070021 * Mike Kershaw <dragorn@kismetwireless.net> 2005/08/14
22 * Add TUNSETLINK ioctl to set the link encapsulation
23 *
Linus Torvalds1da177e2005-04-16 15:20:36 -070024 * Mark Smith <markzzzsmith@yahoo.com.au>
Joe Perches344dc8e2012-07-12 19:33:09 +000025 * Use eth_random_addr() for tap MAC address.
Linus Torvalds1da177e2005-04-16 15:20:36 -070026 *
27 * Harald Roelle <harald.roelle@ifi.lmu.de> 2004/04/20
28 * Fixes in packet dropping, queue length setting and queue wakeup.
29 * Increased default tx queue length.
30 * Added ethtool API.
31 * Minor cleanups
32 *
33 * Daniel Podlejski <underley@underley.eu.org>
34 * Modifications for 2.3.99-pre5 kernel.
35 */
36
Joe Perches6b8a66e2011-03-02 07:18:10 +000037#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
38
Linus Torvalds1da177e2005-04-16 15:20:36 -070039#define DRV_NAME "tun"
40#define DRV_VERSION "1.6"
41#define DRV_DESCRIPTION "Universal TUN/TAP device driver"
42#define DRV_COPYRIGHT "(C) 1999-2004 Max Krasnyansky <maxk@qualcomm.com>"
43
Linus Torvalds1da177e2005-04-16 15:20:36 -070044#include <linux/module.h>
45#include <linux/errno.h>
46#include <linux/kernel.h>
Ingo Molnar174cd4b2017-02-02 19:15:33 +010047#include <linux/sched/signal.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070048#include <linux/major.h>
49#include <linux/slab.h>
50#include <linux/poll.h>
51#include <linux/fcntl.h>
52#include <linux/init.h>
53#include <linux/skbuff.h>
54#include <linux/netdevice.h>
55#include <linux/etherdevice.h>
56#include <linux/miscdevice.h>
57#include <linux/ethtool.h>
58#include <linux/rtnetlink.h>
Arnd Bergmann50857e22009-11-06 22:52:32 -080059#include <linux/compat.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070060#include <linux/if.h>
61#include <linux/if_arp.h>
62#include <linux/if_ether.h>
63#include <linux/if_tun.h>
Jason Wang6680ec62013-07-25 13:00:33 +080064#include <linux/if_vlan.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070065#include <linux/crc32.h>
Pavel Emelyanovd647a592008-04-16 00:41:16 -070066#include <linux/nsproxy.h>
Rusty Russellf43798c2008-07-03 03:48:02 -070067#include <linux/virtio_net.h>
Michael S. Tsirkin99405162010-02-14 01:01:10 +000068#include <linux/rcupdate.h>
Eric W. Biederman881d9662007-09-17 11:56:21 -070069#include <net/net_namespace.h>
Pavel Emelyanov79d17602008-04-16 00:40:46 -070070#include <net/netns/generic.h>
Eric W. Biedermanf019a7a2009-01-21 16:02:16 -080071#include <net/rtnetlink.h>
Herbert Xu33dccbb2009-02-05 21:25:32 -080072#include <net/sock.h>
Masatake YAMATO93e14b62014-01-29 16:43:31 +090073#include <linux/seq_file.h>
Herbert Xue0b46d02014-11-07 21:22:23 +080074#include <linux/uio.h>
Jason Wang1576d982016-06-30 14:45:36 +080075#include <linux/skb_array.h>
Jason Wang761876c2017-08-11 19:41:18 +080076#include <linux/bpf.h>
77#include <linux/bpf_trace.h>
Petar Penkov90e33d42017-09-22 13:49:15 -070078#include <linux/mutex.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070079
Linus Torvalds7c0f6ba2016-12-24 11:46:01 -080080#include <linux/uaccess.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070081
Rusty Russell14daa022008-04-12 18:48:58 -070082/* Uncomment to enable debugging */
83/* #define TUN_DEBUG 1 */
84
Linus Torvalds1da177e2005-04-16 15:20:36 -070085#ifdef TUN_DEBUG
86static int debug;
Rusty Russell14daa022008-04-12 18:48:58 -070087
Joe Perches6b8a66e2011-03-02 07:18:10 +000088#define tun_debug(level, tun, fmt, args...) \
89do { \
90 if (tun->debug) \
91 netdev_printk(level, tun->dev, fmt, ##args); \
92} while (0)
93#define DBG1(level, fmt, args...) \
94do { \
95 if (debug == 2) \
96 printk(level fmt, ##args); \
97} while (0)
Rusty Russell14daa022008-04-12 18:48:58 -070098#else
Joe Perches6b8a66e2011-03-02 07:18:10 +000099#define tun_debug(level, tun, fmt, args...) \
100do { \
101 if (0) \
102 netdev_printk(level, tun->dev, fmt, ##args); \
103} while (0)
104#define DBG1(level, fmt, args...) \
105do { \
106 if (0) \
107 printk(level fmt, ##args); \
108} while (0)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700109#endif
110
Jason Wang761876c2017-08-11 19:41:18 +0800111#define TUN_HEADROOM 256
Jason Wang7df13212017-09-04 11:36:08 +0800112#define TUN_RX_PAD (NET_IP_ALIGN + NET_SKB_PAD)
Jason Wang66ccbc92017-08-11 19:41:16 +0800113
Michael S. Tsirkin031f5e032014-11-19 14:44:40 +0200114/* TUN device flags */
115
116/* IFF_ATTACH_QUEUE is never stored in device flags,
117 * overload it to mean fasync when stored there.
118 */
119#define TUN_FASYNC IFF_ATTACH_QUEUE
Michael S. Tsirkin1cf8e412014-12-16 15:05:06 +0200120/* High bits in flags field are unused. */
121#define TUN_VNET_LE 0x80000000
Greg Kurz8b8e6582015-04-24 14:50:36 +0200122#define TUN_VNET_BE 0x40000000
Michael S. Tsirkin031f5e032014-11-19 14:44:40 +0200123
124#define TUN_FEATURES (IFF_NO_PI | IFF_ONE_QUEUE | IFF_VNET_HDR | \
Petar Penkov90e33d42017-09-22 13:49:15 -0700125 IFF_MULTI_QUEUE | IFF_NAPI | IFF_NAPI_FRAGS)
126
Michael S. Tsirkin06908992012-07-20 09:23:23 +0000127#define GOODCOPY_LEN 128
128
Max Krasnyanskyf271b2cc2008-07-14 22:18:19 -0700129#define FLT_EXACT_COUNT 8
130struct tap_filter {
131 unsigned int count; /* Number of addrs. Zero means disabled */
132 u32 mask[2]; /* Mask of the hashed addrs */
133 unsigned char addr[FLT_EXACT_COUNT][ETH_ALEN];
134};
135
Pankaj Guptabaf71c52015-01-12 11:41:29 +0530136/* MAX_TAP_QUEUES 256 is chosen to allow rx/tx queues to be equal
137 * to max number of VCPUs in guest. */
138#define MAX_TAP_QUEUES 256
Jason Wangb8732fb2013-01-23 03:59:13 +0000139#define MAX_TAP_FLOWS 4096
Jason Wangc8d68e62012-10-31 19:46:00 +0000140
Jason Wang96442e422012-10-31 19:46:02 +0000141#define TUN_FLOW_EXPIRE (3 * HZ)
142
Paolo Abeni608b9972016-04-13 10:52:20 +0200143struct tun_pcpu_stats {
144 u64 rx_packets;
145 u64 rx_bytes;
146 u64 tx_packets;
147 u64 tx_bytes;
148 struct u64_stats_sync syncp;
149 u32 rx_dropped;
150 u32 tx_dropped;
151 u32 rx_frame_errors;
152};
153
Jason Wang54f968d2012-10-31 19:45:57 +0000154/* A tun_file connects an open character device to a tuntap netdevice. It
stephen hemminger92d4ea62013-12-05 20:42:58 -0800155 * also contains all socket related structures (except sock_fprog and tap_filter)
Jason Wang54f968d2012-10-31 19:45:57 +0000156 * to serve as one transmit queue for tuntap device. The sock_fprog and
157 * tap_filter were kept in tun_struct since they were used for filtering for the
Rami Rosen36fe8c092012-11-25 22:07:40 +0000158 * netdevice not for a specific queue (at least I didn't see the requirement for
Jason Wang54f968d2012-10-31 19:45:57 +0000159 * this).
Jason Wang6e914fc2012-10-31 19:45:58 +0000160 *
161 * RCU usage:
Rami Rosen36fe8c092012-11-25 22:07:40 +0000162 * The tun_file and tun_struct are loosely coupled, the pointer from one to the
Jason Wang6e914fc2012-10-31 19:45:58 +0000163 * other can only be read while rcu_read_lock or rtnl_lock is held.
Jason Wang54f968d2012-10-31 19:45:57 +0000164 */
Eric W. Biederman631ab462009-01-20 11:00:40 +0000165struct tun_file {
Jason Wang54f968d2012-10-31 19:45:57 +0000166 struct sock sk;
167 struct socket socket;
168 struct socket_wq wq;
Jason Wang6e914fc2012-10-31 19:45:58 +0000169 struct tun_struct __rcu *tun;
Jason Wang54f968d2012-10-31 19:45:57 +0000170 struct fasync_struct *fasync;
171 /* only used for fasnyc */
172 unsigned int flags;
Pavel Emelyanovfb7589a2013-08-21 14:31:38 +0400173 union {
174 u16 queue_index;
175 unsigned int ifindex;
176 };
Petar Penkov94317092017-09-22 13:49:14 -0700177 struct napi_struct napi;
Eric Dumazetaec72f32017-10-18 12:12:09 -0700178 bool napi_enabled;
Petar Penkov90e33d42017-09-22 13:49:15 -0700179 struct mutex napi_mutex; /* Protects access to the above napi */
Jason Wang4008e972012-12-13 23:53:30 +0000180 struct list_head next;
181 struct tun_struct *detached;
Jason Wang5990a302018-01-04 11:14:27 +0800182 struct ptr_ring tx_ring;
Jesper Dangaard Brouer8bf5c4e2018-01-03 11:25:59 +0100183 struct xdp_rxq_info xdp_rxq;
Eric W. Biederman631ab462009-01-20 11:00:40 +0000184};
185
Jason Wang96442e422012-10-31 19:46:02 +0000186struct tun_flow_entry {
187 struct hlist_node hash_link;
188 struct rcu_head rcu;
189 struct tun_struct *tun;
190
191 u32 rxhash;
Tom Herbert9bc88932013-12-22 18:54:32 +0800192 u32 rps_rxhash;
Jason Wang96442e422012-10-31 19:46:02 +0000193 int queue_index;
194 unsigned long updated;
195};
196
197#define TUN_NUM_FLOW_ENTRIES 1024
198
Jason Wangcd5681d2018-01-16 16:31:01 +0800199struct tun_prog {
Jason Wang96f84062017-12-04 17:31:23 +0800200 struct rcu_head rcu;
201 struct bpf_prog *prog;
202};
203
Jason Wang54f968d2012-10-31 19:45:57 +0000204/* Since the socket were moved to tun_file, to preserve the behavior of persist
Rami Rosen36fe8c092012-11-25 22:07:40 +0000205 * device, socket filter, sndbuf and vnet header size were restore when the
Jason Wang54f968d2012-10-31 19:45:57 +0000206 * file were attached to a persist device.
207 */
Rusty Russell14daa022008-04-12 18:48:58 -0700208struct tun_struct {
Jason Wangc8d68e62012-10-31 19:46:00 +0000209 struct tun_file __rcu *tfiles[MAX_TAP_QUEUES];
210 unsigned int numqueues;
Max Krasnyanskyf271b2cc2008-07-14 22:18:19 -0700211 unsigned int flags;
Eric W. Biederman0625c882012-02-07 16:48:55 -0800212 kuid_t owner;
213 kgid_t group;
Rusty Russell14daa022008-04-12 18:48:58 -0700214
Rusty Russell14daa022008-04-12 18:48:58 -0700215 struct net_device *dev;
Michał Mirosławc8f44af2011-11-15 15:29:55 +0000216 netdev_features_t set_features;
Michał Mirosław88255372011-04-19 06:13:10 +0000217#define TUN_USER_FEATURES (NETIF_F_HW_CSUM|NETIF_F_TSO_ECN|NETIF_F_TSO| \
David S. Millerd591a1f2017-07-03 06:35:32 -0700218 NETIF_F_TSO6)
Michael S. Tsirkind9d52b52010-03-17 17:45:01 +0200219
Paolo Abenieaea34b2016-02-26 10:45:40 +0100220 int align;
Michael S. Tsirkind9d52b52010-03-17 17:45:01 +0200221 int vnet_hdr_sz;
Jason Wang54f968d2012-10-31 19:45:57 +0000222 int sndbuf;
223 struct tap_filter txflt;
224 struct sock_fprog fprog;
225 /* protected by rtnl lock */
226 bool filter_attached;
Rusty Russell14daa022008-04-12 18:48:58 -0700227#ifdef TUN_DEBUG
228 int debug;
229#endif
Jason Wang96442e422012-10-31 19:46:02 +0000230 spinlock_t lock;
Jason Wang96442e422012-10-31 19:46:02 +0000231 struct hlist_head flows[TUN_NUM_FLOW_ENTRIES];
232 struct timer_list flow_gc_timer;
233 unsigned long ageing_time;
Jason Wang4008e972012-12-13 23:53:30 +0000234 unsigned int numdisabled;
235 struct list_head disabled;
Paul Moore5dbbaf22013-01-14 07:12:19 +0000236 void *security;
Jason Wangb8732fb2013-01-23 03:59:13 +0000237 u32 flow_count;
Jason Wang5503fce2017-01-18 15:02:03 +0800238 u32 rx_batched;
Paolo Abeni608b9972016-04-13 10:52:20 +0200239 struct tun_pcpu_stats __percpu *pcpu_stats;
Jason Wang761876c2017-08-11 19:41:18 +0800240 struct bpf_prog __rcu *xdp_prog;
Jason Wangcd5681d2018-01-16 16:31:01 +0800241 struct tun_prog __rcu *steering_prog;
Jason Wangaff3d702018-01-16 16:31:02 +0800242 struct tun_prog __rcu *filter_prog;
Rusty Russell14daa022008-04-12 18:48:58 -0700243};
244
Jason Wangaff3d702018-01-16 16:31:02 +0800245struct veth {
246 __be16 h_vlan_proto;
247 __be16 h_vlan_TCI;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700248};
249
Jason Wangfc72d1d2018-01-04 11:14:28 +0800250bool tun_is_xdp_buff(void *ptr)
251{
252 return (unsigned long)ptr & TUN_XDP_FLAG;
253}
254EXPORT_SYMBOL(tun_is_xdp_buff);
255
256void *tun_xdp_to_ptr(void *ptr)
257{
258 return (void *)((unsigned long)ptr | TUN_XDP_FLAG);
259}
260EXPORT_SYMBOL(tun_xdp_to_ptr);
261
262void *tun_ptr_to_xdp(void *ptr)
263{
264 return (void *)((unsigned long)ptr & ~TUN_XDP_FLAG);
265}
266EXPORT_SYMBOL(tun_ptr_to_xdp);
267
Petar Penkov94317092017-09-22 13:49:14 -0700268static int tun_napi_receive(struct napi_struct *napi, int budget)
269{
270 struct tun_file *tfile = container_of(napi, struct tun_file, napi);
271 struct sk_buff_head *queue = &tfile->sk.sk_write_queue;
272 struct sk_buff_head process_queue;
273 struct sk_buff *skb;
274 int received = 0;
275
276 __skb_queue_head_init(&process_queue);
277
278 spin_lock(&queue->lock);
279 skb_queue_splice_tail_init(queue, &process_queue);
280 spin_unlock(&queue->lock);
281
282 while (received < budget && (skb = __skb_dequeue(&process_queue))) {
283 napi_gro_receive(napi, skb);
284 ++received;
285 }
286
287 if (!skb_queue_empty(&process_queue)) {
288 spin_lock(&queue->lock);
289 skb_queue_splice(&process_queue, queue);
290 spin_unlock(&queue->lock);
291 }
292
293 return received;
294}
295
296static int tun_napi_poll(struct napi_struct *napi, int budget)
297{
298 unsigned int received;
299
300 received = tun_napi_receive(napi, budget);
301
302 if (received < budget)
303 napi_complete_done(napi, received);
304
305 return received;
306}
307
308static void tun_napi_init(struct tun_struct *tun, struct tun_file *tfile,
309 bool napi_en)
310{
Eric Dumazetaec72f32017-10-18 12:12:09 -0700311 tfile->napi_enabled = napi_en;
Petar Penkov94317092017-09-22 13:49:14 -0700312 if (napi_en) {
313 netif_napi_add(tun->dev, &tfile->napi, tun_napi_poll,
314 NAPI_POLL_WEIGHT);
315 napi_enable(&tfile->napi);
Petar Penkov90e33d42017-09-22 13:49:15 -0700316 mutex_init(&tfile->napi_mutex);
Petar Penkov94317092017-09-22 13:49:14 -0700317 }
318}
319
320static void tun_napi_disable(struct tun_struct *tun, struct tun_file *tfile)
321{
Eric Dumazetaec72f32017-10-18 12:12:09 -0700322 if (tfile->napi_enabled)
Petar Penkov94317092017-09-22 13:49:14 -0700323 napi_disable(&tfile->napi);
324}
325
326static void tun_napi_del(struct tun_struct *tun, struct tun_file *tfile)
327{
Eric Dumazetaec72f32017-10-18 12:12:09 -0700328 if (tfile->napi_enabled)
Petar Penkov94317092017-09-22 13:49:14 -0700329 netif_napi_del(&tfile->napi);
330}
331
Petar Penkov90e33d42017-09-22 13:49:15 -0700332static bool tun_napi_frags_enabled(const struct tun_struct *tun)
333{
334 return READ_ONCE(tun->flags) & IFF_NAPI_FRAGS;
335}
336
Greg Kurz8b8e6582015-04-24 14:50:36 +0200337#ifdef CONFIG_TUN_VNET_CROSS_LE
338static inline bool tun_legacy_is_little_endian(struct tun_struct *tun)
339{
340 return tun->flags & TUN_VNET_BE ? false :
341 virtio_legacy_is_little_endian();
342}
343
344static long tun_get_vnet_be(struct tun_struct *tun, int __user *argp)
345{
346 int be = !!(tun->flags & TUN_VNET_BE);
347
348 if (put_user(be, argp))
349 return -EFAULT;
350
351 return 0;
352}
353
354static long tun_set_vnet_be(struct tun_struct *tun, int __user *argp)
355{
356 int be;
357
358 if (get_user(be, argp))
359 return -EFAULT;
360
361 if (be)
362 tun->flags |= TUN_VNET_BE;
363 else
364 tun->flags &= ~TUN_VNET_BE;
365
366 return 0;
367}
368#else
369static inline bool tun_legacy_is_little_endian(struct tun_struct *tun)
370{
371 return virtio_legacy_is_little_endian();
372}
373
374static long tun_get_vnet_be(struct tun_struct *tun, int __user *argp)
375{
376 return -EINVAL;
377}
378
379static long tun_set_vnet_be(struct tun_struct *tun, int __user *argp)
380{
381 return -EINVAL;
382}
383#endif /* CONFIG_TUN_VNET_CROSS_LE */
384
Greg Kurz25bd55bb2015-04-24 14:24:38 +0200385static inline bool tun_is_little_endian(struct tun_struct *tun)
386{
Greg Kurz7d824102015-04-24 14:26:24 +0200387 return tun->flags & TUN_VNET_LE ||
Greg Kurz8b8e6582015-04-24 14:50:36 +0200388 tun_legacy_is_little_endian(tun);
Greg Kurz25bd55bb2015-04-24 14:24:38 +0200389}
390
Michael S. Tsirkin56f0dcc2014-10-23 22:59:31 +0300391static inline u16 tun16_to_cpu(struct tun_struct *tun, __virtio16 val)
392{
Greg Kurz25bd55bb2015-04-24 14:24:38 +0200393 return __virtio16_to_cpu(tun_is_little_endian(tun), val);
Michael S. Tsirkin56f0dcc2014-10-23 22:59:31 +0300394}
395
396static inline __virtio16 cpu_to_tun16(struct tun_struct *tun, u16 val)
397{
Greg Kurz25bd55bb2015-04-24 14:24:38 +0200398 return __cpu_to_virtio16(tun_is_little_endian(tun), val);
Michael S. Tsirkin56f0dcc2014-10-23 22:59:31 +0300399}
400
Jason Wang96442e422012-10-31 19:46:02 +0000401static inline u32 tun_hashfn(u32 rxhash)
402{
403 return rxhash & 0x3ff;
404}
405
406static struct tun_flow_entry *tun_flow_find(struct hlist_head *head, u32 rxhash)
407{
408 struct tun_flow_entry *e;
Jason Wang96442e422012-10-31 19:46:02 +0000409
Sasha Levinb67bfe02013-02-27 17:06:00 -0800410 hlist_for_each_entry_rcu(e, head, hash_link) {
Jason Wang96442e422012-10-31 19:46:02 +0000411 if (e->rxhash == rxhash)
412 return e;
413 }
414 return NULL;
415}
416
417static struct tun_flow_entry *tun_flow_create(struct tun_struct *tun,
418 struct hlist_head *head,
419 u32 rxhash, u16 queue_index)
420{
Eric Dumazet9fdc6be2012-12-21 07:17:21 +0000421 struct tun_flow_entry *e = kmalloc(sizeof(*e), GFP_ATOMIC);
422
Jason Wang96442e422012-10-31 19:46:02 +0000423 if (e) {
424 tun_debug(KERN_INFO, tun, "create flow: hash %u index %u\n",
425 rxhash, queue_index);
426 e->updated = jiffies;
427 e->rxhash = rxhash;
Tom Herbert9bc88932013-12-22 18:54:32 +0800428 e->rps_rxhash = 0;
Jason Wang96442e422012-10-31 19:46:02 +0000429 e->queue_index = queue_index;
430 e->tun = tun;
431 hlist_add_head_rcu(&e->hash_link, head);
Jason Wangb8732fb2013-01-23 03:59:13 +0000432 ++tun->flow_count;
Jason Wang96442e422012-10-31 19:46:02 +0000433 }
434 return e;
435}
436
Jason Wang96442e422012-10-31 19:46:02 +0000437static void tun_flow_delete(struct tun_struct *tun, struct tun_flow_entry *e)
438{
439 tun_debug(KERN_INFO, tun, "delete flow: hash %u index %u\n",
440 e->rxhash, e->queue_index);
441 hlist_del_rcu(&e->hash_link);
Eric Dumazet9fdc6be2012-12-21 07:17:21 +0000442 kfree_rcu(e, rcu);
Jason Wangb8732fb2013-01-23 03:59:13 +0000443 --tun->flow_count;
Jason Wang96442e422012-10-31 19:46:02 +0000444}
445
446static void tun_flow_flush(struct tun_struct *tun)
447{
448 int i;
449
450 spin_lock_bh(&tun->lock);
451 for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++) {
452 struct tun_flow_entry *e;
Sasha Levinb67bfe02013-02-27 17:06:00 -0800453 struct hlist_node *n;
Jason Wang96442e422012-10-31 19:46:02 +0000454
Sasha Levinb67bfe02013-02-27 17:06:00 -0800455 hlist_for_each_entry_safe(e, n, &tun->flows[i], hash_link)
Jason Wang96442e422012-10-31 19:46:02 +0000456 tun_flow_delete(tun, e);
457 }
458 spin_unlock_bh(&tun->lock);
459}
460
461static void tun_flow_delete_by_queue(struct tun_struct *tun, u16 queue_index)
462{
463 int i;
464
465 spin_lock_bh(&tun->lock);
466 for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++) {
467 struct tun_flow_entry *e;
Sasha Levinb67bfe02013-02-27 17:06:00 -0800468 struct hlist_node *n;
Jason Wang96442e422012-10-31 19:46:02 +0000469
Sasha Levinb67bfe02013-02-27 17:06:00 -0800470 hlist_for_each_entry_safe(e, n, &tun->flows[i], hash_link) {
Jason Wang96442e422012-10-31 19:46:02 +0000471 if (e->queue_index == queue_index)
472 tun_flow_delete(tun, e);
473 }
474 }
475 spin_unlock_bh(&tun->lock);
476}
477
Kees Cooke99e88a2017-10-16 14:43:17 -0700478static void tun_flow_cleanup(struct timer_list *t)
Jason Wang96442e422012-10-31 19:46:02 +0000479{
Kees Cooke99e88a2017-10-16 14:43:17 -0700480 struct tun_struct *tun = from_timer(tun, t, flow_gc_timer);
Jason Wang96442e422012-10-31 19:46:02 +0000481 unsigned long delay = tun->ageing_time;
482 unsigned long next_timer = jiffies + delay;
483 unsigned long count = 0;
484 int i;
485
486 tun_debug(KERN_INFO, tun, "tun_flow_cleanup\n");
487
Eric Dumazet7dbfb4e2017-10-20 11:29:55 -0700488 spin_lock(&tun->lock);
Jason Wang96442e422012-10-31 19:46:02 +0000489 for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++) {
490 struct tun_flow_entry *e;
Sasha Levinb67bfe02013-02-27 17:06:00 -0800491 struct hlist_node *n;
Jason Wang96442e422012-10-31 19:46:02 +0000492
Sasha Levinb67bfe02013-02-27 17:06:00 -0800493 hlist_for_each_entry_safe(e, n, &tun->flows[i], hash_link) {
Jason Wang96442e422012-10-31 19:46:02 +0000494 unsigned long this_timer;
Eric Dumazet81d98fa2017-10-20 11:29:56 -0700495
Jason Wang96442e422012-10-31 19:46:02 +0000496 this_timer = e->updated + delay;
Eric Dumazet81d98fa2017-10-20 11:29:56 -0700497 if (time_before_eq(this_timer, jiffies)) {
Jason Wang96442e422012-10-31 19:46:02 +0000498 tun_flow_delete(tun, e);
Eric Dumazet81d98fa2017-10-20 11:29:56 -0700499 continue;
500 }
501 count++;
502 if (time_before(this_timer, next_timer))
Jason Wang96442e422012-10-31 19:46:02 +0000503 next_timer = this_timer;
504 }
505 }
506
507 if (count)
508 mod_timer(&tun->flow_gc_timer, round_jiffies_up(next_timer));
Eric Dumazet7dbfb4e2017-10-20 11:29:55 -0700509 spin_unlock(&tun->lock);
Jason Wang96442e422012-10-31 19:46:02 +0000510}
511
Eric Dumazet49974422012-12-12 19:22:57 +0000512static void tun_flow_update(struct tun_struct *tun, u32 rxhash,
Jason Wang9e857222013-01-28 01:05:19 +0000513 struct tun_file *tfile)
Jason Wang96442e422012-10-31 19:46:02 +0000514{
515 struct hlist_head *head;
516 struct tun_flow_entry *e;
517 unsigned long delay = tun->ageing_time;
Jason Wang9e857222013-01-28 01:05:19 +0000518 u16 queue_index = tfile->queue_index;
Jason Wang96442e422012-10-31 19:46:02 +0000519
520 if (!rxhash)
521 return;
522 else
523 head = &tun->flows[tun_hashfn(rxhash)];
524
525 rcu_read_lock();
526
Jason Wang9e857222013-01-28 01:05:19 +0000527 /* We may get a very small possibility of OOO during switching, not
528 * worth to optimize.*/
529 if (tun->numqueues == 1 || tfile->detached)
Jason Wang96442e422012-10-31 19:46:02 +0000530 goto unlock;
531
532 e = tun_flow_find(head, rxhash);
533 if (likely(e)) {
534 /* TODO: keep queueing to old queue until it's empty? */
535 e->queue_index = queue_index;
536 e->updated = jiffies;
Tom Herbert9bc88932013-12-22 18:54:32 +0800537 sock_rps_record_flow_hash(e->rps_rxhash);
Jason Wang96442e422012-10-31 19:46:02 +0000538 } else {
539 spin_lock_bh(&tun->lock);
Jason Wangb8732fb2013-01-23 03:59:13 +0000540 if (!tun_flow_find(head, rxhash) &&
541 tun->flow_count < MAX_TAP_FLOWS)
Jason Wang96442e422012-10-31 19:46:02 +0000542 tun_flow_create(tun, head, rxhash, queue_index);
543
544 if (!timer_pending(&tun->flow_gc_timer))
545 mod_timer(&tun->flow_gc_timer,
546 round_jiffies_up(jiffies + delay));
547 spin_unlock_bh(&tun->lock);
548 }
549
550unlock:
551 rcu_read_unlock();
552}
553
Tom Herbert9bc88932013-12-22 18:54:32 +0800554/**
555 * Save the hash received in the stack receive path and update the
556 * flow_hash table accordingly.
557 */
558static inline void tun_flow_save_rps_rxhash(struct tun_flow_entry *e, u32 hash)
559{
Eric Dumazet567e4b72015-02-06 12:59:01 -0800560 if (unlikely(e->rps_rxhash != hash))
Tom Herbert9bc88932013-12-22 18:54:32 +0800561 e->rps_rxhash = hash;
Tom Herbert9bc88932013-12-22 18:54:32 +0800562}
563
Jason Wangc8d68e62012-10-31 19:46:00 +0000564/* We try to identify a flow through its rxhash first. The reason that
stephen hemminger92d4ea62013-12-05 20:42:58 -0800565 * we do not check rxq no. is because some cards(e.g 82599), chooses
Jason Wangc8d68e62012-10-31 19:46:00 +0000566 * the rxq based on the txq where the last packet of the flow comes. As
567 * the userspace application move between processors, we may get a
568 * different rxq no. here. If we could not get rxhash, then we would
569 * hope the rxq no. may help here.
570 */
Jason Wang96f84062017-12-04 17:31:23 +0800571static u16 tun_automq_select_queue(struct tun_struct *tun, struct sk_buff *skb)
Jason Wangc8d68e62012-10-31 19:46:00 +0000572{
Jason Wang96442e422012-10-31 19:46:02 +0000573 struct tun_flow_entry *e;
Jason Wangc8d68e62012-10-31 19:46:00 +0000574 u32 txq = 0;
575 u32 numqueues = 0;
576
Mark Rutland6aa7de02017-10-23 14:07:29 -0700577 numqueues = READ_ONCE(tun->numqueues);
Jason Wangc8d68e62012-10-31 19:46:00 +0000578
Jason Wangfeec0842017-06-06 14:09:49 +0800579 txq = __skb_get_hash_symmetric(skb);
Jason Wangc8d68e62012-10-31 19:46:00 +0000580 if (txq) {
Jason Wang96442e422012-10-31 19:46:02 +0000581 e = tun_flow_find(&tun->flows[tun_hashfn(txq)], txq);
Tom Herbert9bc88932013-12-22 18:54:32 +0800582 if (e) {
Tom Herbert9bc88932013-12-22 18:54:32 +0800583 tun_flow_save_rps_rxhash(e, txq);
Zhi Yong Wufbe4d452014-01-02 13:24:28 +0800584 txq = e->queue_index;
Tom Herbert9bc88932013-12-22 18:54:32 +0800585 } else
Jason Wang96442e422012-10-31 19:46:02 +0000586 /* use multiply and shift instead of expensive divide */
587 txq = ((u64)txq * numqueues) >> 32;
Jason Wangc8d68e62012-10-31 19:46:00 +0000588 } else if (likely(skb_rx_queue_recorded(skb))) {
589 txq = skb_get_rx_queue(skb);
590 while (unlikely(txq >= numqueues))
591 txq -= numqueues;
592 }
593
Jason Wangc8d68e62012-10-31 19:46:00 +0000594 return txq;
595}
596
Jason Wang96f84062017-12-04 17:31:23 +0800597static u16 tun_ebpf_select_queue(struct tun_struct *tun, struct sk_buff *skb)
598{
Jason Wangcd5681d2018-01-16 16:31:01 +0800599 struct tun_prog *prog;
Jason Wang96f84062017-12-04 17:31:23 +0800600 u16 ret = 0;
601
602 prog = rcu_dereference(tun->steering_prog);
603 if (prog)
604 ret = bpf_prog_run_clear_cb(prog->prog, skb);
605
606 return ret % tun->numqueues;
607}
608
609static u16 tun_select_queue(struct net_device *dev, struct sk_buff *skb,
610 void *accel_priv, select_queue_fallback_t fallback)
611{
612 struct tun_struct *tun = netdev_priv(dev);
613 u16 ret;
614
615 rcu_read_lock();
616 if (rcu_dereference(tun->steering_prog))
617 ret = tun_ebpf_select_queue(tun, skb);
618 else
619 ret = tun_automq_select_queue(tun, skb);
620 rcu_read_unlock();
621
622 return ret;
623}
624
Jason Wangcde8b152012-10-31 19:46:01 +0000625static inline bool tun_not_capable(struct tun_struct *tun)
626{
627 const struct cred *cred = current_cred();
Eric W. Biedermanc260b772012-11-18 21:34:11 +0000628 struct net *net = dev_net(tun->dev);
Jason Wangcde8b152012-10-31 19:46:01 +0000629
630 return ((uid_valid(tun->owner) && !uid_eq(cred->euid, tun->owner)) ||
631 (gid_valid(tun->group) && !in_egroup_p(tun->group))) &&
Eric W. Biedermanc260b772012-11-18 21:34:11 +0000632 !ns_capable(net->user_ns, CAP_NET_ADMIN);
Jason Wangcde8b152012-10-31 19:46:01 +0000633}
634
Jason Wangc8d68e62012-10-31 19:46:00 +0000635static void tun_set_real_num_queues(struct tun_struct *tun)
636{
637 netif_set_real_num_tx_queues(tun->dev, tun->numqueues);
638 netif_set_real_num_rx_queues(tun->dev, tun->numqueues);
639}
640
Jason Wang4008e972012-12-13 23:53:30 +0000641static void tun_disable_queue(struct tun_struct *tun, struct tun_file *tfile)
642{
643 tfile->detached = tun;
644 list_add_tail(&tfile->next, &tun->disabled);
645 ++tun->numdisabled;
646}
647
Jason Wangd32649d2012-12-18 11:00:27 +0800648static struct tun_struct *tun_enable_queue(struct tun_file *tfile)
Jason Wang4008e972012-12-13 23:53:30 +0000649{
650 struct tun_struct *tun = tfile->detached;
651
652 tfile->detached = NULL;
653 list_del_init(&tfile->next);
654 --tun->numdisabled;
655 return tun;
656}
657
Jason Wangfc72d1d2018-01-04 11:14:28 +0800658static void tun_ptr_free(void *ptr)
659{
660 if (!ptr)
661 return;
662 if (tun_is_xdp_buff(ptr)) {
663 struct xdp_buff *xdp = tun_ptr_to_xdp(ptr);
664
665 put_page(virt_to_head_page(xdp->data));
666 } else {
667 __skb_array_destroy_skb(ptr);
668 }
669}
670
Jason Wang4bfb0512013-09-05 17:53:59 +0800671static void tun_queue_purge(struct tun_file *tfile)
672{
Jason Wangfc72d1d2018-01-04 11:14:28 +0800673 void *ptr;
Jason Wang1576d982016-06-30 14:45:36 +0800674
Jason Wangfc72d1d2018-01-04 11:14:28 +0800675 while ((ptr = ptr_ring_consume(&tfile->tx_ring)) != NULL)
676 tun_ptr_free(ptr);
Jason Wang1576d982016-06-30 14:45:36 +0800677
Jason Wang5503fce2017-01-18 15:02:03 +0800678 skb_queue_purge(&tfile->sk.sk_write_queue);
Jason Wang4bfb0512013-09-05 17:53:59 +0800679 skb_queue_purge(&tfile->sk.sk_error_queue);
680}
681
David S. Miller8565d262018-01-19 22:59:33 -0500682static void tun_cleanup_tx_ring(struct tun_file *tfile)
Cong Wang4df0bfc2018-01-15 11:37:29 -0800683{
David S. Miller8565d262018-01-19 22:59:33 -0500684 if (tfile->tx_ring.queue) {
685 ptr_ring_cleanup(&tfile->tx_ring, tun_ptr_free);
686 xdp_rxq_info_unreg(&tfile->xdp_rxq);
687 memset(&tfile->tx_ring, 0, sizeof(tfile->tx_ring));
Cong Wang4df0bfc2018-01-15 11:37:29 -0800688 }
689}
690
Jason Wangc8d68e62012-10-31 19:46:00 +0000691static void __tun_detach(struct tun_file *tfile, bool clean)
692{
693 struct tun_file *ntfile;
694 struct tun_struct *tun;
Jason Wangc8d68e62012-10-31 19:46:00 +0000695
Jason Wangb8deabd2013-01-11 16:59:32 +0000696 tun = rtnl_dereference(tfile->tun);
697
Petar Penkov94317092017-09-22 13:49:14 -0700698 if (tun && clean) {
699 tun_napi_disable(tun, tfile);
700 tun_napi_del(tun, tfile);
701 }
702
Jason Wang9e857222013-01-28 01:05:19 +0000703 if (tun && !tfile->detached) {
Jason Wangc8d68e62012-10-31 19:46:00 +0000704 u16 index = tfile->queue_index;
705 BUG_ON(index >= tun->numqueues);
Jason Wangc8d68e62012-10-31 19:46:00 +0000706
707 rcu_assign_pointer(tun->tfiles[index],
708 tun->tfiles[tun->numqueues - 1]);
Jason Wangb8deabd2013-01-11 16:59:32 +0000709 ntfile = rtnl_dereference(tun->tfiles[index]);
Jason Wangc8d68e62012-10-31 19:46:00 +0000710 ntfile->queue_index = index;
711
712 --tun->numqueues;
Jason Wang9e857222013-01-28 01:05:19 +0000713 if (clean) {
Monam Agarwalc9566742014-03-24 00:02:32 +0530714 RCU_INIT_POINTER(tfile->tun, NULL);
Jason Wang4008e972012-12-13 23:53:30 +0000715 sock_put(&tfile->sk);
Jason Wang9e857222013-01-28 01:05:19 +0000716 } else
Jason Wang4008e972012-12-13 23:53:30 +0000717 tun_disable_queue(tun, tfile);
Jason Wangc8d68e62012-10-31 19:46:00 +0000718
719 synchronize_net();
Jason Wang96442e422012-10-31 19:46:02 +0000720 tun_flow_delete_by_queue(tun, tun->numqueues + 1);
Jason Wangc8d68e62012-10-31 19:46:00 +0000721 /* Drop read queue */
Jason Wang4bfb0512013-09-05 17:53:59 +0800722 tun_queue_purge(tfile);
Jason Wangc8d68e62012-10-31 19:46:00 +0000723 tun_set_real_num_queues(tun);
Jason Wangdd38bd82013-01-11 16:59:34 +0000724 } else if (tfile->detached && clean) {
Jason Wang4008e972012-12-13 23:53:30 +0000725 tun = tun_enable_queue(tfile);
Jason Wangdd38bd82013-01-11 16:59:34 +0000726 sock_put(&tfile->sk);
727 }
Jason Wangc8d68e62012-10-31 19:46:00 +0000728
729 if (clean) {
Michael S. Tsirkinaf668b32013-01-28 00:38:02 +0000730 if (tun && tun->numqueues == 0 && tun->numdisabled == 0) {
731 netif_carrier_off(tun->dev);
732
Michael S. Tsirkin40630b82014-11-19 15:17:31 +0200733 if (!(tun->flags & IFF_PERSIST) &&
Michael S. Tsirkinaf668b32013-01-28 00:38:02 +0000734 tun->dev->reg_state == NETREG_REGISTERED)
Jason Wang4008e972012-12-13 23:53:30 +0000735 unregister_netdevice(tun->dev);
Michael S. Tsirkinaf668b32013-01-28 00:38:02 +0000736 }
David S. Miller8565d262018-01-19 22:59:33 -0500737 tun_cleanup_tx_ring(tfile);
Eric W. Biederman140e807da2015-05-08 21:07:08 -0500738 sock_put(&tfile->sk);
Jason Wangc8d68e62012-10-31 19:46:00 +0000739 }
740}
741
742static void tun_detach(struct tun_file *tfile, bool clean)
743{
744 rtnl_lock();
745 __tun_detach(tfile, clean);
746 rtnl_unlock();
747}
748
749static void tun_detach_all(struct net_device *dev)
750{
751 struct tun_struct *tun = netdev_priv(dev);
Jason Wang4008e972012-12-13 23:53:30 +0000752 struct tun_file *tfile, *tmp;
Jason Wangc8d68e62012-10-31 19:46:00 +0000753 int i, n = tun->numqueues;
754
755 for (i = 0; i < n; i++) {
Jason Wangb8deabd2013-01-11 16:59:32 +0000756 tfile = rtnl_dereference(tun->tfiles[i]);
Jason Wangc8d68e62012-10-31 19:46:00 +0000757 BUG_ON(!tfile);
Petar Penkov94317092017-09-22 13:49:14 -0700758 tun_napi_disable(tun, tfile);
Jason Wangaddf8fc2016-05-19 13:36:51 +0800759 tfile->socket.sk->sk_shutdown = RCV_SHUTDOWN;
Xi Wang9e641bd2014-05-16 15:11:48 -0700760 tfile->socket.sk->sk_data_ready(tfile->socket.sk);
Monam Agarwalc9566742014-03-24 00:02:32 +0530761 RCU_INIT_POINTER(tfile->tun, NULL);
Jason Wangc8d68e62012-10-31 19:46:00 +0000762 --tun->numqueues;
763 }
Jason Wang9e857222013-01-28 01:05:19 +0000764 list_for_each_entry(tfile, &tun->disabled, next) {
Jason Wangaddf8fc2016-05-19 13:36:51 +0800765 tfile->socket.sk->sk_shutdown = RCV_SHUTDOWN;
Xi Wang9e641bd2014-05-16 15:11:48 -0700766 tfile->socket.sk->sk_data_ready(tfile->socket.sk);
Monam Agarwalc9566742014-03-24 00:02:32 +0530767 RCU_INIT_POINTER(tfile->tun, NULL);
Jason Wang9e857222013-01-28 01:05:19 +0000768 }
Jason Wangc8d68e62012-10-31 19:46:00 +0000769 BUG_ON(tun->numqueues != 0);
770
771 synchronize_net();
772 for (i = 0; i < n; i++) {
Jason Wangb8deabd2013-01-11 16:59:32 +0000773 tfile = rtnl_dereference(tun->tfiles[i]);
Petar Penkov94317092017-09-22 13:49:14 -0700774 tun_napi_del(tun, tfile);
Jason Wangc8d68e62012-10-31 19:46:00 +0000775 /* Drop read queue */
Jason Wang4bfb0512013-09-05 17:53:59 +0800776 tun_queue_purge(tfile);
Jason Wangc8d68e62012-10-31 19:46:00 +0000777 sock_put(&tfile->sk);
David S. Miller8565d262018-01-19 22:59:33 -0500778 tun_cleanup_tx_ring(tfile);
Jason Wangc8d68e62012-10-31 19:46:00 +0000779 }
Jason Wang4008e972012-12-13 23:53:30 +0000780 list_for_each_entry_safe(tfile, tmp, &tun->disabled, next) {
781 tun_enable_queue(tfile);
Jason Wang4bfb0512013-09-05 17:53:59 +0800782 tun_queue_purge(tfile);
Jason Wang4008e972012-12-13 23:53:30 +0000783 sock_put(&tfile->sk);
David S. Miller8565d262018-01-19 22:59:33 -0500784 tun_cleanup_tx_ring(tfile);
Jason Wang4008e972012-12-13 23:53:30 +0000785 }
786 BUG_ON(tun->numdisabled != 0);
Jason Wangdd38bd82013-01-11 16:59:34 +0000787
Michael S. Tsirkin40630b82014-11-19 15:17:31 +0200788 if (tun->flags & IFF_PERSIST)
Jason Wangdd38bd82013-01-11 16:59:34 +0000789 module_put(THIS_MODULE);
Jason Wangc8d68e62012-10-31 19:46:00 +0000790}
791
Petar Penkov94317092017-09-22 13:49:14 -0700792static int tun_attach(struct tun_struct *tun, struct file *file,
793 bool skip_filter, bool napi)
Eric W. Biedermana7385ba2009-01-20 10:57:48 +0000794{
Eric W. Biederman631ab462009-01-20 11:00:40 +0000795 struct tun_file *tfile = file->private_data;
Jason Wang1576d982016-06-30 14:45:36 +0800796 struct net_device *dev = tun->dev;
Eric W. Biederman38231b72009-01-20 11:02:28 +0000797 int err;
Eric W. Biedermana7385ba2009-01-20 10:57:48 +0000798
Paul Moore5dbbaf22013-01-14 07:12:19 +0000799 err = security_tun_dev_attach(tfile->socket.sk, tun->security);
800 if (err < 0)
801 goto out;
802
Eric W. Biederman38231b72009-01-20 11:02:28 +0000803 err = -EINVAL;
Jason Wang9e857222013-01-28 01:05:19 +0000804 if (rtnl_dereference(tfile->tun) && !tfile->detached)
Eric W. Biederman38231b72009-01-20 11:02:28 +0000805 goto out;
806
807 err = -EBUSY;
Michael S. Tsirkin40630b82014-11-19 15:17:31 +0200808 if (!(tun->flags & IFF_MULTI_QUEUE) && tun->numqueues == 1)
Jason Wangc8d68e62012-10-31 19:46:00 +0000809 goto out;
810
811 err = -E2BIG;
Jason Wang4008e972012-12-13 23:53:30 +0000812 if (!tfile->detached &&
813 tun->numqueues + tun->numdisabled == MAX_TAP_QUEUES)
Eric W. Biederman38231b72009-01-20 11:02:28 +0000814 goto out;
815
816 err = 0;
Jason Wang54f968d2012-10-31 19:45:57 +0000817
stephen hemminger92d4ea62013-12-05 20:42:58 -0800818 /* Re-attach the filter to persist device */
Pavel Emelyanov849c9b62013-08-21 14:32:21 +0400819 if (!skip_filter && (tun->filter_attached == true)) {
Hannes Frederic Sowa8ced4252016-04-05 17:10:16 +0200820 lock_sock(tfile->socket.sk);
821 err = sk_attach_filter(&tun->fprog, tfile->socket.sk);
822 release_sock(tfile->socket.sk);
Jason Wang54f968d2012-10-31 19:45:57 +0000823 if (!err)
824 goto out;
825 }
Jason Wang1576d982016-06-30 14:45:36 +0800826
827 if (!tfile->detached &&
Jason Wang5990a302018-01-04 11:14:27 +0800828 ptr_ring_init(&tfile->tx_ring, dev->tx_queue_len, GFP_KERNEL)) {
Jason Wang1576d982016-06-30 14:45:36 +0800829 err = -ENOMEM;
830 goto out;
831 }
832
Jason Wangc8d68e62012-10-31 19:46:00 +0000833 tfile->queue_index = tun->numqueues;
Jason Wangaddf8fc2016-05-19 13:36:51 +0800834 tfile->socket.sk->sk_shutdown &= ~RCV_SHUTDOWN;
Jesper Dangaard Brouer8bf5c4e2018-01-03 11:25:59 +0100835
836 if (tfile->detached) {
837 /* Re-attach detached tfile, updating XDP queue_index */
838 WARN_ON(!xdp_rxq_info_is_reg(&tfile->xdp_rxq));
839
840 if (tfile->xdp_rxq.queue_index != tfile->queue_index)
841 tfile->xdp_rxq.queue_index = tfile->queue_index;
842 } else {
843 /* Setup XDP RX-queue info, for new tfile getting attached */
844 err = xdp_rxq_info_reg(&tfile->xdp_rxq,
845 tun->dev, tfile->queue_index);
846 if (err < 0)
847 goto out;
848 err = 0;
849 }
850
Jason Wang6e914fc2012-10-31 19:45:58 +0000851 rcu_assign_pointer(tfile->tun, tun);
Jason Wangc8d68e62012-10-31 19:46:00 +0000852 rcu_assign_pointer(tun->tfiles[tun->numqueues], tfile);
Jason Wangc8d68e62012-10-31 19:46:00 +0000853 tun->numqueues++;
854
Petar Penkov94317092017-09-22 13:49:14 -0700855 if (tfile->detached) {
Jason Wang4008e972012-12-13 23:53:30 +0000856 tun_enable_queue(tfile);
Petar Penkov94317092017-09-22 13:49:14 -0700857 } else {
Jason Wang4008e972012-12-13 23:53:30 +0000858 sock_hold(&tfile->sk);
Petar Penkov94317092017-09-22 13:49:14 -0700859 tun_napi_init(tun, tfile, napi);
860 }
Jason Wang4008e972012-12-13 23:53:30 +0000861
Jason Wangc8d68e62012-10-31 19:46:00 +0000862 tun_set_real_num_queues(tun);
863
Jason Wangc8d68e62012-10-31 19:46:00 +0000864 /* device is allowed to go away first, so no need to hold extra
865 * refcnt.
866 */
Eric W. Biedermana7385ba2009-01-20 10:57:48 +0000867
Eric W. Biederman38231b72009-01-20 11:02:28 +0000868out:
Eric W. Biederman38231b72009-01-20 11:02:28 +0000869 return err;
Eric W. Biedermana7385ba2009-01-20 10:57:48 +0000870}
871
yuan linyu9484dc72017-09-23 22:36:52 +0800872static struct tun_struct *tun_get(struct tun_file *tfile)
Eric W. Biederman631ab462009-01-20 11:00:40 +0000873{
Jason Wang6e914fc2012-10-31 19:45:58 +0000874 struct tun_struct *tun;
Eric W. Biedermanc70f1822009-01-20 11:07:17 +0000875
Jason Wang6e914fc2012-10-31 19:45:58 +0000876 rcu_read_lock();
877 tun = rcu_dereference(tfile->tun);
878 if (tun)
879 dev_hold(tun->dev);
880 rcu_read_unlock();
Eric W. Biedermanc70f1822009-01-20 11:07:17 +0000881
882 return tun;
Eric W. Biederman631ab462009-01-20 11:00:40 +0000883}
884
Eric W. Biederman631ab462009-01-20 11:00:40 +0000885static void tun_put(struct tun_struct *tun)
886{
Jason Wang6e914fc2012-10-31 19:45:58 +0000887 dev_put(tun->dev);
Eric W. Biederman631ab462009-01-20 11:00:40 +0000888}
889
Joe Perches6b8a66e2011-03-02 07:18:10 +0000890/* TAP filtering */
Max Krasnyanskyf271b2cc2008-07-14 22:18:19 -0700891static void addr_hash_set(u32 *mask, const u8 *addr)
892{
893 int n = ether_crc(ETH_ALEN, addr) >> 26;
894 mask[n >> 5] |= (1 << (n & 31));
895}
896
897static unsigned int addr_hash_test(const u32 *mask, const u8 *addr)
898{
899 int n = ether_crc(ETH_ALEN, addr) >> 26;
900 return mask[n >> 5] & (1 << (n & 31));
901}
902
903static int update_filter(struct tap_filter *filter, void __user *arg)
904{
905 struct { u8 u[ETH_ALEN]; } *addr;
906 struct tun_filter uf;
907 int err, alen, n, nexact;
908
909 if (copy_from_user(&uf, arg, sizeof(uf)))
910 return -EFAULT;
911
912 if (!uf.count) {
913 /* Disabled */
914 filter->count = 0;
915 return 0;
916 }
917
918 alen = ETH_ALEN * uf.count;
Markus Elfring28e81902016-08-20 08:54:15 +0200919 addr = memdup_user(arg + sizeof(uf), alen);
920 if (IS_ERR(addr))
921 return PTR_ERR(addr);
Max Krasnyanskyf271b2cc2008-07-14 22:18:19 -0700922
923 /* The filter is updated without holding any locks. Which is
924 * perfectly safe. We disable it first and in the worst
925 * case we'll accept a few undesired packets. */
926 filter->count = 0;
927 wmb();
928
929 /* Use first set of addresses as an exact filter */
930 for (n = 0; n < uf.count && n < FLT_EXACT_COUNT; n++)
931 memcpy(filter->addr[n], addr[n].u, ETH_ALEN);
932
933 nexact = n;
934
Alex Williamsoncfbf84f2009-02-08 17:49:17 -0800935 /* Remaining multicast addresses are hashed,
936 * unicast will leave the filter disabled. */
Max Krasnyanskyf271b2cc2008-07-14 22:18:19 -0700937 memset(filter->mask, 0, sizeof(filter->mask));
Alex Williamsoncfbf84f2009-02-08 17:49:17 -0800938 for (; n < uf.count; n++) {
939 if (!is_multicast_ether_addr(addr[n].u)) {
940 err = 0; /* no filter */
Markus Elfring3b8d2a62016-08-20 09:00:34 +0200941 goto free_addr;
Alex Williamsoncfbf84f2009-02-08 17:49:17 -0800942 }
Max Krasnyanskyf271b2cc2008-07-14 22:18:19 -0700943 addr_hash_set(filter->mask, addr[n].u);
Alex Williamsoncfbf84f2009-02-08 17:49:17 -0800944 }
Max Krasnyanskyf271b2cc2008-07-14 22:18:19 -0700945
946 /* For ALLMULTI just set the mask to all ones.
947 * This overrides the mask populated above. */
948 if ((uf.flags & TUN_FLT_ALLMULTI))
949 memset(filter->mask, ~0, sizeof(filter->mask));
950
951 /* Now enable the filter */
952 wmb();
953 filter->count = nexact;
954
955 /* Return the number of exact filters */
956 err = nexact;
Markus Elfring3b8d2a62016-08-20 09:00:34 +0200957free_addr:
Max Krasnyanskyf271b2cc2008-07-14 22:18:19 -0700958 kfree(addr);
959 return err;
960}
961
962/* Returns: 0 - drop, !=0 - accept */
963static int run_filter(struct tap_filter *filter, const struct sk_buff *skb)
964{
965 /* Cannot use eth_hdr(skb) here because skb_mac_hdr() is incorrect
966 * at this point. */
967 struct ethhdr *eh = (struct ethhdr *) skb->data;
968 int i;
969
970 /* Exact match */
971 for (i = 0; i < filter->count; i++)
Joe Perches2e42e472012-05-09 17:17:46 +0000972 if (ether_addr_equal(eh->h_dest, filter->addr[i]))
Max Krasnyanskyf271b2cc2008-07-14 22:18:19 -0700973 return 1;
974
975 /* Inexact match (multicast only) */
976 if (is_multicast_ether_addr(eh->h_dest))
977 return addr_hash_test(filter->mask, eh->h_dest);
978
979 return 0;
980}
981
982/*
983 * Checks whether the packet is accepted or not.
984 * Returns: 0 - drop, !=0 - accept
985 */
986static int check_filter(struct tap_filter *filter, const struct sk_buff *skb)
987{
988 if (!filter->count)
989 return 1;
990
991 return run_filter(filter, skb);
992}
993
Linus Torvalds1da177e2005-04-16 15:20:36 -0700994/* Network device part of the driver */
995
Linus Torvalds1da177e2005-04-16 15:20:36 -0700996static const struct ethtool_ops tun_ethtool_ops;
997
Eric W. Biedermanc70f1822009-01-20 11:07:17 +0000998/* Net device detach from fd. */
999static void tun_net_uninit(struct net_device *dev)
1000{
Jason Wangc8d68e62012-10-31 19:46:00 +00001001 tun_detach_all(dev);
Eric W. Biedermanc70f1822009-01-20 11:07:17 +00001002}
1003
Linus Torvalds1da177e2005-04-16 15:20:36 -07001004/* Net device open. */
1005static int tun_net_open(struct net_device *dev)
1006{
Hannes Frederic Sowab20e2d52017-03-13 00:00:26 +01001007 struct tun_struct *tun = netdev_priv(dev);
1008 int i;
1009
Jason Wangc8d68e62012-10-31 19:46:00 +00001010 netif_tx_start_all_queues(dev);
Hannes Frederic Sowab20e2d52017-03-13 00:00:26 +01001011
1012 for (i = 0; i < tun->numqueues; i++) {
1013 struct tun_file *tfile;
1014
1015 tfile = rtnl_dereference(tun->tfiles[i]);
1016 tfile->socket.sk->sk_write_space(tfile->socket.sk);
1017 }
1018
Linus Torvalds1da177e2005-04-16 15:20:36 -07001019 return 0;
1020}
1021
1022/* Net device close. */
1023static int tun_net_close(struct net_device *dev)
1024{
Jason Wangc8d68e62012-10-31 19:46:00 +00001025 netif_tx_stop_all_queues(dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001026 return 0;
1027}
1028
1029/* Net device start xmit */
Jason Wang96f84062017-12-04 17:31:23 +08001030static void tun_automq_xmit(struct tun_struct *tun, struct sk_buff *skb)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001031{
Jason Wang3df97ba2016-04-25 23:13:42 -04001032#ifdef CONFIG_RPS
Jason Wang96f84062017-12-04 17:31:23 +08001033 if (tun->numqueues == 1 && static_key_false(&rps_needed)) {
Tom Herbert9bc88932013-12-22 18:54:32 +08001034 /* Select queue was not called for the skbuff, so we extract the
1035 * RPS hash and save it into the flow_table here.
1036 */
1037 __u32 rxhash;
1038
Jason Wangfeec0842017-06-06 14:09:49 +08001039 rxhash = __skb_get_hash_symmetric(skb);
Tom Herbert9bc88932013-12-22 18:54:32 +08001040 if (rxhash) {
1041 struct tun_flow_entry *e;
1042 e = tun_flow_find(&tun->flows[tun_hashfn(rxhash)],
1043 rxhash);
1044 if (e)
1045 tun_flow_save_rps_rxhash(e, rxhash);
1046 }
1047 }
Jason Wang3df97ba2016-04-25 23:13:42 -04001048#endif
Jason Wang96f84062017-12-04 17:31:23 +08001049}
1050
Jason Wangaff3d702018-01-16 16:31:02 +08001051static unsigned int run_ebpf_filter(struct tun_struct *tun,
1052 struct sk_buff *skb,
1053 int len)
1054{
1055 struct tun_prog *prog = rcu_dereference(tun->filter_prog);
1056
1057 if (prog)
1058 len = bpf_prog_run_clear_cb(prog->prog, skb);
1059
1060 return len;
1061}
1062
Jason Wang96f84062017-12-04 17:31:23 +08001063/* Net device start xmit */
1064static netdev_tx_t tun_net_xmit(struct sk_buff *skb, struct net_device *dev)
1065{
1066 struct tun_struct *tun = netdev_priv(dev);
1067 int txq = skb->queue_mapping;
1068 struct tun_file *tfile;
Jason Wangaff3d702018-01-16 16:31:02 +08001069 int len = skb->len;
Jason Wang96f84062017-12-04 17:31:23 +08001070
1071 rcu_read_lock();
1072 tfile = rcu_dereference(tun->tfiles[txq]);
Jason Wang96f84062017-12-04 17:31:23 +08001073
1074 /* Drop packet if interface is not attached */
Willem de Bruijncc166422017-12-05 22:11:17 -05001075 if (txq >= tun->numqueues)
Jason Wang96f84062017-12-04 17:31:23 +08001076 goto drop;
1077
1078 if (!rcu_dereference(tun->steering_prog))
1079 tun_automq_xmit(tun, skb);
Tom Herbert9bc88932013-12-22 18:54:32 +08001080
Jason Wang6e914fc2012-10-31 19:45:58 +00001081 tun_debug(KERN_INFO, tun, "tun_net_xmit %d\n", skb->len);
1082
Jason Wangc8d68e62012-10-31 19:46:00 +00001083 BUG_ON(!tfile);
1084
Max Krasnyanskyf271b2cc2008-07-14 22:18:19 -07001085 /* Drop if the filter does not like it.
1086 * This is a noop if the filter is disabled.
1087 * Filter can be enabled only for the TAP devices. */
1088 if (!check_filter(&tun->txflt, skb))
1089 goto drop;
1090
Jason Wang54f968d2012-10-31 19:45:57 +00001091 if (tfile->socket.sk->sk_filter &&
1092 sk_filter(tfile->socket.sk, skb))
Michael S. Tsirkin99405162010-02-14 01:01:10 +00001093 goto drop;
1094
Jason Wangaff3d702018-01-16 16:31:02 +08001095 len = run_ebpf_filter(tun, skb, len);
1096
1097 /* Trim extra bytes since we may insert vlan proto & TCI
1098 * in tun_put_user().
1099 */
1100 len -= skb_vlan_tag_present(skb) ? sizeof(struct veth) : 0;
1101 if (len <= 0 || pskb_trim(skb, len))
1102 goto drop;
1103
Willem de Bruijn1f8b9772017-08-03 16:29:41 -04001104 if (unlikely(skb_orphan_frags_rx(skb, GFP_ATOMIC)))
Jason Wang7bf66302013-09-05 17:54:00 +08001105 goto drop;
1106
Soheil Hassas Yeganeh7b996242016-08-23 18:22:33 -04001107 skb_tx_timestamp(skb);
Richard Cochraneda29772013-07-19 19:40:10 +02001108
Michael S. Tsirkin0110d6f2010-04-13 04:59:44 +00001109 /* Orphan the skb - required as we might hang on to it
Jason Wang7bf66302013-09-05 17:54:00 +08001110 * for indefinite time.
1111 */
Michael S. Tsirkin0110d6f2010-04-13 04:59:44 +00001112 skb_orphan(skb);
1113
Eric Dumazetf8af75f2013-03-06 11:02:37 +00001114 nf_reset(skb);
1115
Jason Wang5990a302018-01-04 11:14:27 +08001116 if (ptr_ring_produce(&tfile->tx_ring, skb))
Jason Wang1576d982016-06-30 14:45:36 +08001117 goto drop;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001118
1119 /* Notify and wake up reader process */
Jason Wang54f968d2012-10-31 19:45:57 +00001120 if (tfile->flags & TUN_FASYNC)
1121 kill_fasync(&tfile->fasync, SIGIO, POLL_IN);
Xi Wang9e641bd2014-05-16 15:11:48 -07001122 tfile->socket.sk->sk_data_ready(tfile->socket.sk);
Jason Wang6e914fc2012-10-31 19:45:58 +00001123
1124 rcu_read_unlock();
Patrick McHardy6ed10652009-06-23 06:03:08 +00001125 return NETDEV_TX_OK;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001126
1127drop:
Paolo Abeni608b9972016-04-13 10:52:20 +02001128 this_cpu_inc(tun->pcpu_stats->tx_dropped);
Michael S. Tsirkin149d36f2012-11-01 09:16:32 +00001129 skb_tx_error(skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001130 kfree_skb(skb);
Jason Wang6e914fc2012-10-31 19:45:58 +00001131 rcu_read_unlock();
Jason Wangbaeabab2014-11-18 13:20:41 +08001132 return NET_XMIT_DROP;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001133}
1134
Max Krasnyanskyf271b2cc2008-07-14 22:18:19 -07001135static void tun_net_mclist(struct net_device *dev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001136{
Max Krasnyanskyf271b2cc2008-07-14 22:18:19 -07001137 /*
1138 * This callback is supposed to deal with mc filter in
1139 * _rx_ path and has nothing to do with the _tx_ path.
1140 * In rx path we always accept everything userspace gives us.
1141 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001142}
1143
Michał Mirosławc8f44af2011-11-15 15:29:55 +00001144static netdev_features_t tun_net_fix_features(struct net_device *dev,
1145 netdev_features_t features)
Michał Mirosław88255372011-04-19 06:13:10 +00001146{
1147 struct tun_struct *tun = netdev_priv(dev);
1148
1149 return (features & tun->set_features) | (features & ~TUN_USER_FEATURES);
1150}
Neil Hormanbebd0972011-06-15 05:25:01 +00001151#ifdef CONFIG_NET_POLL_CONTROLLER
1152static void tun_poll_controller(struct net_device *dev)
1153{
1154 /*
1155 * Tun only receives frames when:
1156 * 1) the char device endpoint gets data from user space
1157 * 2) the tun socket gets a sendmsg call from user space
Petar Penkov94317092017-09-22 13:49:14 -07001158 * If NAPI is not enabled, since both of those are synchronous
1159 * operations, we are guaranteed never to have pending data when we poll
1160 * for it so there is nothing to do here but return.
Neil Hormanbebd0972011-06-15 05:25:01 +00001161 * We need this though so netpoll recognizes us as an interface that
1162 * supports polling, which enables bridge devices in virt setups to
1163 * still use netconsole
Petar Penkov94317092017-09-22 13:49:14 -07001164 * If NAPI is enabled, however, we need to schedule polling for all
Petar Penkov90e33d42017-09-22 13:49:15 -07001165 * queues unless we are using napi_gro_frags(), which we call in
1166 * process context and not in NAPI context.
Neil Hormanbebd0972011-06-15 05:25:01 +00001167 */
Petar Penkov94317092017-09-22 13:49:14 -07001168 struct tun_struct *tun = netdev_priv(dev);
1169
1170 if (tun->flags & IFF_NAPI) {
1171 struct tun_file *tfile;
1172 int i;
1173
Petar Penkov90e33d42017-09-22 13:49:15 -07001174 if (tun_napi_frags_enabled(tun))
1175 return;
1176
Petar Penkov94317092017-09-22 13:49:14 -07001177 rcu_read_lock();
1178 for (i = 0; i < tun->numqueues; i++) {
1179 tfile = rcu_dereference(tun->tfiles[i]);
Eric Dumazetaec72f32017-10-18 12:12:09 -07001180 if (tfile->napi_enabled)
1181 napi_schedule(&tfile->napi);
Petar Penkov94317092017-09-22 13:49:14 -07001182 }
1183 rcu_read_unlock();
1184 }
Neil Hormanbebd0972011-06-15 05:25:01 +00001185 return;
1186}
1187#endif
Paolo Abenieaea34b2016-02-26 10:45:40 +01001188
1189static void tun_set_headroom(struct net_device *dev, int new_hr)
1190{
1191 struct tun_struct *tun = netdev_priv(dev);
1192
1193 if (new_hr < NET_SKB_PAD)
1194 new_hr = NET_SKB_PAD;
1195
1196 tun->align = new_hr;
1197}
1198
stephen hemmingerbc1f4472017-01-06 19:12:52 -08001199static void
Paolo Abeni608b9972016-04-13 10:52:20 +02001200tun_net_get_stats64(struct net_device *dev, struct rtnl_link_stats64 *stats)
1201{
1202 u32 rx_dropped = 0, tx_dropped = 0, rx_frame_errors = 0;
1203 struct tun_struct *tun = netdev_priv(dev);
1204 struct tun_pcpu_stats *p;
1205 int i;
1206
1207 for_each_possible_cpu(i) {
1208 u64 rxpackets, rxbytes, txpackets, txbytes;
1209 unsigned int start;
1210
1211 p = per_cpu_ptr(tun->pcpu_stats, i);
1212 do {
1213 start = u64_stats_fetch_begin(&p->syncp);
1214 rxpackets = p->rx_packets;
1215 rxbytes = p->rx_bytes;
1216 txpackets = p->tx_packets;
1217 txbytes = p->tx_bytes;
1218 } while (u64_stats_fetch_retry(&p->syncp, start));
1219
1220 stats->rx_packets += rxpackets;
1221 stats->rx_bytes += rxbytes;
1222 stats->tx_packets += txpackets;
1223 stats->tx_bytes += txbytes;
1224
1225 /* u32 counters */
1226 rx_dropped += p->rx_dropped;
1227 rx_frame_errors += p->rx_frame_errors;
1228 tx_dropped += p->tx_dropped;
1229 }
1230 stats->rx_dropped = rx_dropped;
1231 stats->rx_frame_errors = rx_frame_errors;
1232 stats->tx_dropped = tx_dropped;
Paolo Abeni608b9972016-04-13 10:52:20 +02001233}
1234
Jason Wang761876c2017-08-11 19:41:18 +08001235static int tun_xdp_set(struct net_device *dev, struct bpf_prog *prog,
1236 struct netlink_ext_ack *extack)
1237{
1238 struct tun_struct *tun = netdev_priv(dev);
1239 struct bpf_prog *old_prog;
1240
1241 old_prog = rtnl_dereference(tun->xdp_prog);
1242 rcu_assign_pointer(tun->xdp_prog, prog);
1243 if (old_prog)
1244 bpf_prog_put(old_prog);
1245
1246 return 0;
1247}
1248
1249static u32 tun_xdp_query(struct net_device *dev)
1250{
1251 struct tun_struct *tun = netdev_priv(dev);
1252 const struct bpf_prog *xdp_prog;
1253
1254 xdp_prog = rtnl_dereference(tun->xdp_prog);
1255 if (xdp_prog)
1256 return xdp_prog->aux->id;
1257
1258 return 0;
1259}
1260
Jakub Kicinskif4e63522017-11-03 13:56:16 -07001261static int tun_xdp(struct net_device *dev, struct netdev_bpf *xdp)
Jason Wang761876c2017-08-11 19:41:18 +08001262{
1263 switch (xdp->command) {
1264 case XDP_SETUP_PROG:
1265 return tun_xdp_set(dev, xdp->prog, xdp->extack);
1266 case XDP_QUERY_PROG:
1267 xdp->prog_id = tun_xdp_query(dev);
1268 xdp->prog_attached = !!xdp->prog_id;
1269 return 0;
1270 default:
1271 return -EINVAL;
1272 }
1273}
1274
Stephen Hemminger758e43b2008-11-19 22:10:37 -08001275static const struct net_device_ops tun_netdev_ops = {
Eric W. Biedermanc70f1822009-01-20 11:07:17 +00001276 .ndo_uninit = tun_net_uninit,
Stephen Hemminger758e43b2008-11-19 22:10:37 -08001277 .ndo_open = tun_net_open,
1278 .ndo_stop = tun_net_close,
Stephen Hemminger00829822008-11-20 20:14:53 -08001279 .ndo_start_xmit = tun_net_xmit,
Michał Mirosław88255372011-04-19 06:13:10 +00001280 .ndo_fix_features = tun_net_fix_features,
Jason Wangc8d68e62012-10-31 19:46:00 +00001281 .ndo_select_queue = tun_select_queue,
Neil Hormanbebd0972011-06-15 05:25:01 +00001282#ifdef CONFIG_NET_POLL_CONTROLLER
1283 .ndo_poll_controller = tun_poll_controller,
1284#endif
Paolo Abenieaea34b2016-02-26 10:45:40 +01001285 .ndo_set_rx_headroom = tun_set_headroom,
Paolo Abeni608b9972016-04-13 10:52:20 +02001286 .ndo_get_stats64 = tun_net_get_stats64,
Stephen Hemminger758e43b2008-11-19 22:10:37 -08001287};
1288
Jason Wangfc72d1d2018-01-04 11:14:28 +08001289static int tun_xdp_xmit(struct net_device *dev, struct xdp_buff *xdp)
1290{
1291 struct tun_struct *tun = netdev_priv(dev);
1292 struct xdp_buff *buff = xdp->data_hard_start;
1293 int headroom = xdp->data - xdp->data_hard_start;
1294 struct tun_file *tfile;
1295 u32 numqueues;
1296 int ret = 0;
1297
1298 /* Assure headroom is available and buff is properly aligned */
1299 if (unlikely(headroom < sizeof(*xdp) || tun_is_xdp_buff(xdp)))
1300 return -ENOSPC;
1301
1302 *buff = *xdp;
1303
1304 rcu_read_lock();
1305
1306 numqueues = READ_ONCE(tun->numqueues);
1307 if (!numqueues) {
1308 ret = -ENOSPC;
1309 goto out;
1310 }
1311
1312 tfile = rcu_dereference(tun->tfiles[smp_processor_id() %
1313 numqueues]);
1314 /* Encode the XDP flag into lowest bit for consumer to differ
1315 * XDP buffer from sk_buff.
1316 */
1317 if (ptr_ring_produce(&tfile->tx_ring, tun_xdp_to_ptr(buff))) {
1318 this_cpu_inc(tun->pcpu_stats->tx_dropped);
1319 ret = -ENOSPC;
1320 }
1321
1322out:
1323 rcu_read_unlock();
1324 return ret;
1325}
1326
1327static void tun_xdp_flush(struct net_device *dev)
1328{
1329 struct tun_struct *tun = netdev_priv(dev);
1330 struct tun_file *tfile;
1331 u32 numqueues;
1332
1333 rcu_read_lock();
1334
1335 numqueues = READ_ONCE(tun->numqueues);
1336 if (!numqueues)
1337 goto out;
1338
1339 tfile = rcu_dereference(tun->tfiles[smp_processor_id() %
1340 numqueues]);
1341 /* Notify and wake up reader process */
1342 if (tfile->flags & TUN_FASYNC)
1343 kill_fasync(&tfile->fasync, SIGIO, POLL_IN);
1344 tfile->socket.sk->sk_data_ready(tfile->socket.sk);
1345
1346out:
1347 rcu_read_unlock();
1348}
1349
Stephen Hemminger758e43b2008-11-19 22:10:37 -08001350static const struct net_device_ops tap_netdev_ops = {
Eric W. Biedermanc70f1822009-01-20 11:07:17 +00001351 .ndo_uninit = tun_net_uninit,
Stephen Hemminger758e43b2008-11-19 22:10:37 -08001352 .ndo_open = tun_net_open,
1353 .ndo_stop = tun_net_close,
Stephen Hemminger00829822008-11-20 20:14:53 -08001354 .ndo_start_xmit = tun_net_xmit,
Michał Mirosław88255372011-04-19 06:13:10 +00001355 .ndo_fix_features = tun_net_fix_features,
Jiri Pirkoafc4b132011-08-16 06:29:01 +00001356 .ndo_set_rx_mode = tun_net_mclist,
Stephen Hemminger758e43b2008-11-19 22:10:37 -08001357 .ndo_set_mac_address = eth_mac_addr,
1358 .ndo_validate_addr = eth_validate_addr,
Jason Wangc8d68e62012-10-31 19:46:00 +00001359 .ndo_select_queue = tun_select_queue,
Neil Hormanbebd0972011-06-15 05:25:01 +00001360#ifdef CONFIG_NET_POLL_CONTROLLER
1361 .ndo_poll_controller = tun_poll_controller,
1362#endif
Toshiaki Makita5e527962015-07-31 15:03:27 +09001363 .ndo_features_check = passthru_features_check,
Paolo Abenieaea34b2016-02-26 10:45:40 +01001364 .ndo_set_rx_headroom = tun_set_headroom,
Paolo Abeni608b9972016-04-13 10:52:20 +02001365 .ndo_get_stats64 = tun_net_get_stats64,
Jakub Kicinskif4e63522017-11-03 13:56:16 -07001366 .ndo_bpf = tun_xdp,
Jason Wangfc72d1d2018-01-04 11:14:28 +08001367 .ndo_xdp_xmit = tun_xdp_xmit,
1368 .ndo_xdp_flush = tun_xdp_flush,
Stephen Hemminger758e43b2008-11-19 22:10:37 -08001369};
1370
Pavel Emelyanov944a1372013-06-11 17:01:08 +04001371static void tun_flow_init(struct tun_struct *tun)
Jason Wang96442e422012-10-31 19:46:02 +00001372{
1373 int i;
1374
Jason Wang96442e422012-10-31 19:46:02 +00001375 for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++)
1376 INIT_HLIST_HEAD(&tun->flows[i]);
1377
1378 tun->ageing_time = TUN_FLOW_EXPIRE;
Kees Cooke99e88a2017-10-16 14:43:17 -07001379 timer_setup(&tun->flow_gc_timer, tun_flow_cleanup, 0);
1380 mod_timer(&tun->flow_gc_timer,
1381 round_jiffies_up(jiffies + tun->ageing_time));
Jason Wang96442e422012-10-31 19:46:02 +00001382}
1383
1384static void tun_flow_uninit(struct tun_struct *tun)
1385{
1386 del_timer_sync(&tun->flow_gc_timer);
1387 tun_flow_flush(tun);
Jason Wang96442e422012-10-31 19:46:02 +00001388}
1389
Jarod Wilson91572082016-10-20 13:55:20 -04001390#define MIN_MTU 68
1391#define MAX_MTU 65535
1392
Linus Torvalds1da177e2005-04-16 15:20:36 -07001393/* Initialize net device. */
1394static void tun_net_init(struct net_device *dev)
1395{
1396 struct tun_struct *tun = netdev_priv(dev);
Jeff Garzik6aa20a22006-09-13 13:24:59 -04001397
Linus Torvalds1da177e2005-04-16 15:20:36 -07001398 switch (tun->flags & TUN_TYPE_MASK) {
Michael S. Tsirkin40630b82014-11-19 15:17:31 +02001399 case IFF_TUN:
Stephen Hemminger758e43b2008-11-19 22:10:37 -08001400 dev->netdev_ops = &tun_netdev_ops;
1401
Linus Torvalds1da177e2005-04-16 15:20:36 -07001402 /* Point-to-Point TUN Device */
1403 dev->hard_header_len = 0;
1404 dev->addr_len = 0;
1405 dev->mtu = 1500;
1406
1407 /* Zero header length */
Jeff Garzik6aa20a22006-09-13 13:24:59 -04001408 dev->type = ARPHRD_NONE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001409 dev->flags = IFF_POINTOPOINT | IFF_NOARP | IFF_MULTICAST;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001410 break;
1411
Michael S. Tsirkin40630b82014-11-19 15:17:31 +02001412 case IFF_TAP:
Kusanagi Kouichi7a0a9602008-12-29 18:23:28 -08001413 dev->netdev_ops = &tap_netdev_ops;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001414 /* Ethernet TAP Device */
Max Krasnyanskyf271b2cc2008-07-14 22:18:19 -07001415 ether_setup(dev);
Neil Horman550fd082011-07-26 06:05:38 +00001416 dev->priv_flags &= ~IFF_TX_SKB_SHARING;
stephen hemmingera6768472012-12-10 15:16:00 +00001417 dev->priv_flags |= IFF_LIVE_ADDR_CHANGE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001418
Danny Kukawkaf2cedb62012-02-15 06:45:39 +00001419 eth_hw_addr_random(dev);
Brian Braunstein36226a82007-04-26 01:00:55 -07001420
Linus Torvalds1da177e2005-04-16 15:20:36 -07001421 break;
1422 }
Jarod Wilson91572082016-10-20 13:55:20 -04001423
1424 dev->min_mtu = MIN_MTU;
1425 dev->max_mtu = MAX_MTU - dev->hard_header_len;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001426}
1427
1428/* Character device part */
1429
1430/* Poll */
Al Viroafc9a422017-07-03 06:39:46 -04001431static __poll_t tun_chr_poll(struct file *file, poll_table *wait)
Jeff Garzik6aa20a22006-09-13 13:24:59 -04001432{
Eric W. Biedermanb2430de2009-01-20 11:03:21 +00001433 struct tun_file *tfile = file->private_data;
yuan linyu9484dc72017-09-23 22:36:52 +08001434 struct tun_struct *tun = tun_get(tfile);
Mariusz Kozlowski3c8a9c62009-07-05 19:48:35 +00001435 struct sock *sk;
Al Viroafc9a422017-07-03 06:39:46 -04001436 __poll_t mask = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001437
1438 if (!tun)
Linus Torvaldsa9a08842018-02-11 14:34:03 -08001439 return EPOLLERR;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001440
Jason Wang54f968d2012-10-31 19:45:57 +00001441 sk = tfile->socket.sk;
Mariusz Kozlowski3c8a9c62009-07-05 19:48:35 +00001442
Joe Perches6b8a66e2011-03-02 07:18:10 +00001443 tun_debug(KERN_INFO, tun, "tun_chr_poll\n");
Linus Torvalds1da177e2005-04-16 15:20:36 -07001444
Xi Wang9e641bd2014-05-16 15:11:48 -07001445 poll_wait(file, sk_sleep(sk), wait);
Jeff Garzik6aa20a22006-09-13 13:24:59 -04001446
Jason Wang5990a302018-01-04 11:14:27 +08001447 if (!ptr_ring_empty(&tfile->tx_ring))
Linus Torvaldsa9a08842018-02-11 14:34:03 -08001448 mask |= EPOLLIN | EPOLLRDNORM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001449
Hannes Frederic Sowab20e2d52017-03-13 00:00:26 +01001450 if (tun->dev->flags & IFF_UP &&
1451 (sock_writeable(sk) ||
1452 (!test_and_set_bit(SOCKWQ_ASYNC_NOSPACE, &sk->sk_socket->flags) &&
1453 sock_writeable(sk))))
Linus Torvaldsa9a08842018-02-11 14:34:03 -08001454 mask |= EPOLLOUT | EPOLLWRNORM;
Herbert Xu33dccbb2009-02-05 21:25:32 -08001455
Eric W. Biedermanc70f1822009-01-20 11:07:17 +00001456 if (tun->dev->reg_state != NETREG_REGISTERED)
Linus Torvaldsa9a08842018-02-11 14:34:03 -08001457 mask = EPOLLERR;
Eric W. Biedermanc70f1822009-01-20 11:07:17 +00001458
Eric W. Biederman631ab462009-01-20 11:00:40 +00001459 tun_put(tun);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001460 return mask;
1461}
1462
Petar Penkov90e33d42017-09-22 13:49:15 -07001463static struct sk_buff *tun_napi_alloc_frags(struct tun_file *tfile,
1464 size_t len,
1465 const struct iov_iter *it)
1466{
1467 struct sk_buff *skb;
1468 size_t linear;
1469 int err;
1470 int i;
1471
1472 if (it->nr_segs > MAX_SKB_FRAGS + 1)
1473 return ERR_PTR(-ENOMEM);
1474
1475 local_bh_disable();
1476 skb = napi_get_frags(&tfile->napi);
1477 local_bh_enable();
1478 if (!skb)
1479 return ERR_PTR(-ENOMEM);
1480
1481 linear = iov_iter_single_seg_count(it);
1482 err = __skb_grow(skb, linear);
1483 if (err)
1484 goto free;
1485
1486 skb->len = len;
1487 skb->data_len = len - linear;
1488 skb->truesize += skb->data_len;
1489
1490 for (i = 1; i < it->nr_segs; i++) {
Eric Dumazet43a08e02018-02-15 14:47:15 -08001491 struct page_frag *pfrag = &current->task_frag;
Petar Penkov90e33d42017-09-22 13:49:15 -07001492 size_t fragsz = it->iov[i].iov_len;
Petar Penkov90e33d42017-09-22 13:49:15 -07001493
1494 if (fragsz == 0 || fragsz > PAGE_SIZE) {
1495 err = -EINVAL;
1496 goto free;
1497 }
1498
Eric Dumazet43a08e02018-02-15 14:47:15 -08001499 if (!skb_page_frag_refill(fragsz, pfrag, GFP_KERNEL)) {
Petar Penkov90e33d42017-09-22 13:49:15 -07001500 err = -ENOMEM;
1501 goto free;
1502 }
1503
Eric Dumazet43a08e02018-02-15 14:47:15 -08001504 skb_fill_page_desc(skb, i - 1, pfrag->page,
1505 pfrag->offset, fragsz);
1506 page_ref_inc(pfrag->page);
1507 pfrag->offset += fragsz;
Petar Penkov90e33d42017-09-22 13:49:15 -07001508 }
1509
1510 return skb;
1511free:
1512 /* frees skb and all frags allocated with napi_alloc_frag() */
1513 napi_free_frags(&tfile->napi);
1514 return ERR_PTR(err);
1515}
1516
Rusty Russellf42157c2008-08-15 15:15:10 -07001517/* prepad is the amount to reserve at front. len is length after that.
1518 * linear is a hint as to how much to copy (usually headers). */
Jason Wang54f968d2012-10-31 19:45:57 +00001519static struct sk_buff *tun_alloc_skb(struct tun_file *tfile,
stephen hemminger6f7c1562011-06-08 14:33:08 +00001520 size_t prepad, size_t len,
1521 size_t linear, int noblock)
Rusty Russellf42157c2008-08-15 15:15:10 -07001522{
Jason Wang54f968d2012-10-31 19:45:57 +00001523 struct sock *sk = tfile->socket.sk;
Rusty Russellf42157c2008-08-15 15:15:10 -07001524 struct sk_buff *skb;
Herbert Xu33dccbb2009-02-05 21:25:32 -08001525 int err;
Rusty Russellf42157c2008-08-15 15:15:10 -07001526
1527 /* Under a page? Don't bother with paged skb. */
Herbert Xu0eca93b2009-04-14 02:09:43 -07001528 if (prepad + len < PAGE_SIZE || !linear)
Herbert Xu33dccbb2009-02-05 21:25:32 -08001529 linear = len;
Rusty Russellf42157c2008-08-15 15:15:10 -07001530
Herbert Xu33dccbb2009-02-05 21:25:32 -08001531 skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock,
Eric Dumazet28d64272013-08-08 14:38:47 -07001532 &err, 0);
Rusty Russellf42157c2008-08-15 15:15:10 -07001533 if (!skb)
Herbert Xu33dccbb2009-02-05 21:25:32 -08001534 return ERR_PTR(err);
Rusty Russellf42157c2008-08-15 15:15:10 -07001535
1536 skb_reserve(skb, prepad);
1537 skb_put(skb, linear);
Herbert Xu33dccbb2009-02-05 21:25:32 -08001538 skb->data_len = len - linear;
1539 skb->len += len - linear;
Rusty Russellf42157c2008-08-15 15:15:10 -07001540
1541 return skb;
1542}
1543
Jason Wang5503fce2017-01-18 15:02:03 +08001544static void tun_rx_batched(struct tun_struct *tun, struct tun_file *tfile,
1545 struct sk_buff *skb, int more)
1546{
1547 struct sk_buff_head *queue = &tfile->sk.sk_write_queue;
1548 struct sk_buff_head process_queue;
1549 u32 rx_batched = tun->rx_batched;
1550 bool rcv = false;
1551
1552 if (!rx_batched || (!more && skb_queue_empty(queue))) {
1553 local_bh_disable();
1554 netif_receive_skb(skb);
1555 local_bh_enable();
1556 return;
1557 }
1558
1559 spin_lock(&queue->lock);
1560 if (!more || skb_queue_len(queue) == rx_batched) {
1561 __skb_queue_head_init(&process_queue);
1562 skb_queue_splice_tail_init(queue, &process_queue);
1563 rcv = true;
1564 } else {
1565 __skb_queue_tail(queue, skb);
1566 }
1567 spin_unlock(&queue->lock);
1568
1569 if (rcv) {
1570 struct sk_buff *nskb;
1571
1572 local_bh_disable();
1573 while ((nskb = __skb_dequeue(&process_queue)))
1574 netif_receive_skb(nskb);
1575 netif_receive_skb(skb);
1576 local_bh_enable();
1577 }
1578}
1579
Jason Wang66ccbc92017-08-11 19:41:16 +08001580static bool tun_can_build_skb(struct tun_struct *tun, struct tun_file *tfile,
1581 int len, int noblock, bool zerocopy)
1582{
1583 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
1584 return false;
1585
1586 if (tfile->socket.sk->sk_sndbuf != INT_MAX)
1587 return false;
1588
1589 if (!noblock)
1590 return false;
1591
1592 if (zerocopy)
1593 return false;
1594
1595 if (SKB_DATA_ALIGN(len + TUN_RX_PAD) +
1596 SKB_DATA_ALIGN(sizeof(struct skb_shared_info)) > PAGE_SIZE)
1597 return false;
1598
1599 return true;
1600}
1601
Jason Wang761876c2017-08-11 19:41:18 +08001602static struct sk_buff *tun_build_skb(struct tun_struct *tun,
1603 struct tun_file *tfile,
Jason Wang66ccbc92017-08-11 19:41:16 +08001604 struct iov_iter *from,
Jason Wang761876c2017-08-11 19:41:18 +08001605 struct virtio_net_hdr *hdr,
Jason Wang1cfe6e92017-09-04 11:36:09 +08001606 int len, int *skb_xdp)
Jason Wang66ccbc92017-08-11 19:41:16 +08001607{
Eric Dumazet0bbd7da2017-08-16 22:14:33 +08001608 struct page_frag *alloc_frag = &current->task_frag;
Jason Wang66ccbc92017-08-11 19:41:16 +08001609 struct sk_buff *skb;
Jason Wang761876c2017-08-11 19:41:18 +08001610 struct bpf_prog *xdp_prog;
Jason Wang7df13212017-09-04 11:36:08 +08001611 int buflen = SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
Jason Wang761876c2017-08-11 19:41:18 +08001612 unsigned int delta = 0;
Jason Wang66ccbc92017-08-11 19:41:16 +08001613 char *buf;
1614 size_t copied;
Jason Wang761876c2017-08-11 19:41:18 +08001615 bool xdp_xmit = false;
Jason Wang7df13212017-09-04 11:36:08 +08001616 int err, pad = TUN_RX_PAD;
1617
1618 rcu_read_lock();
1619 xdp_prog = rcu_dereference(tun->xdp_prog);
1620 if (xdp_prog)
1621 pad += TUN_HEADROOM;
1622 buflen += SKB_DATA_ALIGN(len + pad);
1623 rcu_read_unlock();
Jason Wang66ccbc92017-08-11 19:41:16 +08001624
Jason Wang63b9ab62017-10-27 11:05:44 +08001625 alloc_frag->offset = ALIGN((u64)alloc_frag->offset, SMP_CACHE_BYTES);
Jason Wang66ccbc92017-08-11 19:41:16 +08001626 if (unlikely(!skb_page_frag_refill(buflen, alloc_frag, GFP_KERNEL)))
1627 return ERR_PTR(-ENOMEM);
1628
1629 buf = (char *)page_address(alloc_frag->page) + alloc_frag->offset;
1630 copied = copy_page_from_iter(alloc_frag->page,
Jason Wang7df13212017-09-04 11:36:08 +08001631 alloc_frag->offset + pad,
Jason Wang66ccbc92017-08-11 19:41:16 +08001632 len, from);
1633 if (copied != len)
1634 return ERR_PTR(-EFAULT);
1635
Jason Wang7df13212017-09-04 11:36:08 +08001636 /* There's a small window that XDP may be set after the check
1637 * of xdp_prog above, this should be rare and for simplicity
1638 * we do XDP on skb in case the headroom is not enough.
1639 */
1640 if (hdr->gso_type || !xdp_prog)
Jason Wang1cfe6e92017-09-04 11:36:09 +08001641 *skb_xdp = 1;
Jason Wang761876c2017-08-11 19:41:18 +08001642 else
Jason Wang1cfe6e92017-09-04 11:36:09 +08001643 *skb_xdp = 0;
Jason Wang66ccbc92017-08-11 19:41:16 +08001644
Jason Wang761876c2017-08-11 19:41:18 +08001645 rcu_read_lock();
1646 xdp_prog = rcu_dereference(tun->xdp_prog);
Jason Wang1cfe6e92017-09-04 11:36:09 +08001647 if (xdp_prog && !*skb_xdp) {
Jason Wang761876c2017-08-11 19:41:18 +08001648 struct xdp_buff xdp;
1649 void *orig_data;
1650 u32 act;
1651
1652 xdp.data_hard_start = buf;
Jason Wang7df13212017-09-04 11:36:08 +08001653 xdp.data = buf + pad;
Daniel Borkmannde8f3a82017-09-25 02:25:51 +02001654 xdp_set_data_meta_invalid(&xdp);
Jason Wang761876c2017-08-11 19:41:18 +08001655 xdp.data_end = xdp.data + len;
Jesper Dangaard Brouer8bf5c4e2018-01-03 11:25:59 +01001656 xdp.rxq = &tfile->xdp_rxq;
Jason Wang761876c2017-08-11 19:41:18 +08001657 orig_data = xdp.data;
1658 act = bpf_prog_run_xdp(xdp_prog, &xdp);
1659
1660 switch (act) {
1661 case XDP_REDIRECT:
1662 get_page(alloc_frag->page);
1663 alloc_frag->offset += buflen;
1664 err = xdp_do_redirect(tun->dev, &xdp, xdp_prog);
1665 if (err)
1666 goto err_redirect;
Xin Long654d5732017-11-19 19:31:04 +08001667 rcu_read_unlock();
Jason Wang761876c2017-08-11 19:41:18 +08001668 return NULL;
1669 case XDP_TX:
1670 xdp_xmit = true;
1671 /* fall through */
1672 case XDP_PASS:
1673 delta = orig_data - xdp.data;
1674 break;
1675 default:
1676 bpf_warn_invalid_xdp_action(act);
1677 /* fall through */
1678 case XDP_ABORTED:
1679 trace_xdp_exception(tun->dev, xdp_prog, act);
1680 /* fall through */
1681 case XDP_DROP:
1682 goto err_xdp;
1683 }
1684 }
1685
1686 skb = build_skb(buf, buflen);
1687 if (!skb) {
1688 rcu_read_unlock();
1689 return ERR_PTR(-ENOMEM);
1690 }
1691
Jason Wang7df13212017-09-04 11:36:08 +08001692 skb_reserve(skb, pad - delta);
Jason Wang761876c2017-08-11 19:41:18 +08001693 skb_put(skb, len + delta);
Jason Wang66ccbc92017-08-11 19:41:16 +08001694 get_page(alloc_frag->page);
1695 alloc_frag->offset += buflen;
1696
Jason Wang761876c2017-08-11 19:41:18 +08001697 if (xdp_xmit) {
1698 skb->dev = tun->dev;
1699 generic_xdp_tx(skb, xdp_prog);
Xin Long654d5732017-11-19 19:31:04 +08001700 rcu_read_unlock();
Jason Wang761876c2017-08-11 19:41:18 +08001701 return NULL;
1702 }
1703
1704 rcu_read_unlock();
1705
Jason Wang66ccbc92017-08-11 19:41:16 +08001706 return skb;
Jason Wang761876c2017-08-11 19:41:18 +08001707
1708err_redirect:
1709 put_page(alloc_frag->page);
1710err_xdp:
1711 rcu_read_unlock();
1712 this_cpu_inc(tun->pcpu_stats->rx_dropped);
1713 return NULL;
Jason Wang66ccbc92017-08-11 19:41:16 +08001714}
1715
Linus Torvalds1da177e2005-04-16 15:20:36 -07001716/* Get packet from user space buffer */
Jason Wang54f968d2012-10-31 19:45:57 +00001717static ssize_t tun_get_user(struct tun_struct *tun, struct tun_file *tfile,
Al Virof5ff53b2014-06-19 15:36:49 -04001718 void *msg_control, struct iov_iter *from,
Jason Wang5503fce2017-01-18 15:02:03 +08001719 int noblock, bool more)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001720{
Harvey Harrison09640e632009-02-01 00:45:17 -08001721 struct tun_pi pi = { 0, cpu_to_be16(ETH_P_IP) };
Linus Torvalds1da177e2005-04-16 15:20:36 -07001722 struct sk_buff *skb;
Al Virof5ff53b2014-06-19 15:36:49 -04001723 size_t total_len = iov_iter_count(from);
Paolo Abenieaea34b2016-02-26 10:45:40 +01001724 size_t len = total_len, align = tun->align, linear;
Rusty Russellf43798c2008-07-03 03:48:02 -07001725 struct virtio_net_hdr gso = { 0 };
Paolo Abeni608b9972016-04-13 10:52:20 +02001726 struct tun_pcpu_stats *stats;
Jason Wang96f8d9e2013-11-13 14:00:39 +08001727 int good_linear;
Michael S. Tsirkin06908992012-07-20 09:23:23 +00001728 int copylen;
1729 bool zerocopy = false;
1730 int err;
Jason Wang96f84062017-12-04 17:31:23 +08001731 u32 rxhash = 0;
Jason Wang1cfe6e92017-09-04 11:36:09 +08001732 int skb_xdp = 1;
Petar Penkov90e33d42017-09-22 13:49:15 -07001733 bool frags = tun_napi_frags_enabled(tun);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001734
Eric Dumazet1bd4978a2015-12-16 08:57:37 -08001735 if (!(tun->dev->flags & IFF_UP))
1736 return -EIO;
1737
Michael S. Tsirkin40630b82014-11-19 15:17:31 +02001738 if (!(tun->flags & IFF_NO_PI)) {
Dan Carpenter15718ea2013-08-15 15:52:57 +03001739 if (len < sizeof(pi))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001740 return -EINVAL;
Dan Carpenter15718ea2013-08-15 15:52:57 +03001741 len -= sizeof(pi);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001742
Al Virocbbd26b2016-11-01 22:09:04 -04001743 if (!copy_from_iter_full(&pi, sizeof(pi), from))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001744 return -EFAULT;
1745 }
1746
Michael S. Tsirkin40630b82014-11-19 15:17:31 +02001747 if (tun->flags & IFF_VNET_HDR) {
Willem de Bruijne1edab82017-02-03 18:20:48 -05001748 int vnet_hdr_sz = READ_ONCE(tun->vnet_hdr_sz);
1749
1750 if (len < vnet_hdr_sz)
Rusty Russellf43798c2008-07-03 03:48:02 -07001751 return -EINVAL;
Willem de Bruijne1edab82017-02-03 18:20:48 -05001752 len -= vnet_hdr_sz;
Rusty Russellf43798c2008-07-03 03:48:02 -07001753
Al Virocbbd26b2016-11-01 22:09:04 -04001754 if (!copy_from_iter_full(&gso, sizeof(gso), from))
Rusty Russellf43798c2008-07-03 03:48:02 -07001755 return -EFAULT;
1756
Herbert Xu49091222009-06-08 00:20:01 -07001757 if ((gso.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) &&
Michael S. Tsirkin56f0dcc2014-10-23 22:59:31 +03001758 tun16_to_cpu(tun, gso.csum_start) + tun16_to_cpu(tun, gso.csum_offset) + 2 > tun16_to_cpu(tun, gso.hdr_len))
1759 gso.hdr_len = cpu_to_tun16(tun, tun16_to_cpu(tun, gso.csum_start) + tun16_to_cpu(tun, gso.csum_offset) + 2);
Herbert Xu49091222009-06-08 00:20:01 -07001760
Michael S. Tsirkin56f0dcc2014-10-23 22:59:31 +03001761 if (tun16_to_cpu(tun, gso.hdr_len) > len)
Rusty Russellf43798c2008-07-03 03:48:02 -07001762 return -EINVAL;
Willem de Bruijne1edab82017-02-03 18:20:48 -05001763 iov_iter_advance(from, vnet_hdr_sz - sizeof(gso));
Rusty Russellf43798c2008-07-03 03:48:02 -07001764 }
1765
Michael S. Tsirkin40630b82014-11-19 15:17:31 +02001766 if ((tun->flags & TUN_TYPE_MASK) == IFF_TAP) {
stephen hemmingera504b862011-06-08 14:33:07 +00001767 align += NET_IP_ALIGN;
Herbert Xu0eca93b2009-04-14 02:09:43 -07001768 if (unlikely(len < ETH_HLEN ||
Michael S. Tsirkin56f0dcc2014-10-23 22:59:31 +03001769 (gso.hdr_len && tun16_to_cpu(tun, gso.hdr_len) < ETH_HLEN)))
Rusty Russelle01bf1c2008-04-12 18:49:30 -07001770 return -EINVAL;
1771 }
Jeff Garzik6aa20a22006-09-13 13:24:59 -04001772
Jason Wang96f8d9e2013-11-13 14:00:39 +08001773 good_linear = SKB_MAX_HEAD(align);
1774
Jason Wang88529172013-07-18 10:55:15 +08001775 if (msg_control) {
Al Virof5ff53b2014-06-19 15:36:49 -04001776 struct iov_iter i = *from;
1777
Jason Wang88529172013-07-18 10:55:15 +08001778 /* There are 256 bytes to be copied in skb, so there is
1779 * enough room for skb expand head in case it is used.
Michael S. Tsirkin06908992012-07-20 09:23:23 +00001780 * The rest of the buffer is mapped from userspace.
1781 */
Michael S. Tsirkin56f0dcc2014-10-23 22:59:31 +03001782 copylen = gso.hdr_len ? tun16_to_cpu(tun, gso.hdr_len) : GOODCOPY_LEN;
Jason Wang96f8d9e2013-11-13 14:00:39 +08001783 if (copylen > good_linear)
1784 copylen = good_linear;
Jason Wang3dd5c332013-07-10 13:43:27 +08001785 linear = copylen;
Al Virof5ff53b2014-06-19 15:36:49 -04001786 iov_iter_advance(&i, copylen);
1787 if (iov_iter_npages(&i, INT_MAX) <= MAX_SKB_FRAGS)
Jason Wang88529172013-07-18 10:55:15 +08001788 zerocopy = true;
1789 }
1790
Petar Penkov90e33d42017-09-22 13:49:15 -07001791 if (!frags && tun_can_build_skb(tun, tfile, len, noblock, zerocopy)) {
Jason Wang1cfe6e92017-09-04 11:36:09 +08001792 /* For the packet that is not easy to be processed
1793 * (e.g gso or jumbo packet), we will do it at after
1794 * skb was created with generic XDP routine.
1795 */
1796 skb = tun_build_skb(tun, tfile, from, &gso, len, &skb_xdp);
Jason Wang66ccbc92017-08-11 19:41:16 +08001797 if (IS_ERR(skb)) {
Paolo Abeni608b9972016-04-13 10:52:20 +02001798 this_cpu_inc(tun->pcpu_stats->rx_dropped);
Jason Wang66ccbc92017-08-11 19:41:16 +08001799 return PTR_ERR(skb);
1800 }
Jason Wang761876c2017-08-11 19:41:18 +08001801 if (!skb)
1802 return total_len;
Jason Wang66ccbc92017-08-11 19:41:16 +08001803 } else {
1804 if (!zerocopy) {
1805 copylen = len;
1806 if (tun16_to_cpu(tun, gso.hdr_len) > good_linear)
1807 linear = good_linear;
1808 else
1809 linear = tun16_to_cpu(tun, gso.hdr_len);
1810 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001811
Petar Penkov90e33d42017-09-22 13:49:15 -07001812 if (frags) {
1813 mutex_lock(&tfile->napi_mutex);
1814 skb = tun_napi_alloc_frags(tfile, copylen, from);
1815 /* tun_napi_alloc_frags() enforces a layout for the skb.
1816 * If zerocopy is enabled, then this layout will be
1817 * overwritten by zerocopy_sg_from_iter().
1818 */
1819 zerocopy = false;
1820 } else {
1821 skb = tun_alloc_skb(tfile, align, copylen, linear,
1822 noblock);
1823 }
1824
Jason Wang66ccbc92017-08-11 19:41:16 +08001825 if (IS_ERR(skb)) {
1826 if (PTR_ERR(skb) != -EAGAIN)
1827 this_cpu_inc(tun->pcpu_stats->rx_dropped);
Petar Penkov90e33d42017-09-22 13:49:15 -07001828 if (frags)
1829 mutex_unlock(&tfile->napi_mutex);
Jason Wang66ccbc92017-08-11 19:41:16 +08001830 return PTR_ERR(skb);
1831 }
Michael S. Tsirkin06908992012-07-20 09:23:23 +00001832
Jason Wang66ccbc92017-08-11 19:41:16 +08001833 if (zerocopy)
1834 err = zerocopy_sg_from_iter(skb, from);
1835 else
1836 err = skb_copy_datagram_from_iter(skb, 0, from, len);
1837
1838 if (err) {
1839 this_cpu_inc(tun->pcpu_stats->rx_dropped);
1840 kfree_skb(skb);
Petar Penkov90e33d42017-09-22 13:49:15 -07001841 if (frags) {
1842 tfile->napi.skb = NULL;
1843 mutex_unlock(&tfile->napi_mutex);
1844 }
1845
Jason Wang66ccbc92017-08-11 19:41:16 +08001846 return -EFAULT;
1847 }
Dave Jones8f227572006-03-11 18:49:13 -08001848 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001849
Jarno Rajahalme3e9e40e2016-11-18 15:40:38 -08001850 if (virtio_net_hdr_to_skb(skb, &gso, tun_is_little_endian(tun))) {
Paolo Abenidf10db92016-06-14 00:00:04 +02001851 this_cpu_inc(tun->pcpu_stats->rx_frame_errors);
1852 kfree_skb(skb);
Petar Penkov90e33d42017-09-22 13:49:15 -07001853 if (frags) {
1854 tfile->napi.skb = NULL;
1855 mutex_unlock(&tfile->napi_mutex);
1856 }
1857
Paolo Abenidf10db92016-06-14 00:00:04 +02001858 return -EINVAL;
1859 }
1860
Linus Torvalds1da177e2005-04-16 15:20:36 -07001861 switch (tun->flags & TUN_TYPE_MASK) {
Michael S. Tsirkin40630b82014-11-19 15:17:31 +02001862 case IFF_TUN:
1863 if (tun->flags & IFF_NO_PI) {
Alexander Potapenko2580c4c2017-09-28 11:32:37 +02001864 u8 ip_version = skb->len ? (skb->data[0] >> 4) : 0;
1865
1866 switch (ip_version) {
1867 case 4:
Ang Way Chuangf09f7ee2008-06-17 21:10:33 -07001868 pi.proto = htons(ETH_P_IP);
1869 break;
Alexander Potapenko2580c4c2017-09-28 11:32:37 +02001870 case 6:
Ang Way Chuangf09f7ee2008-06-17 21:10:33 -07001871 pi.proto = htons(ETH_P_IPV6);
1872 break;
1873 default:
Paolo Abeni608b9972016-04-13 10:52:20 +02001874 this_cpu_inc(tun->pcpu_stats->rx_dropped);
Ang Way Chuangf09f7ee2008-06-17 21:10:33 -07001875 kfree_skb(skb);
1876 return -EINVAL;
1877 }
1878 }
1879
Arnaldo Carvalho de Melo459a98e2007-03-19 15:30:44 -07001880 skb_reset_mac_header(skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001881 skb->protocol = pi.proto;
Arnaldo Carvalho de Melo4c13eb62007-04-25 17:40:23 -07001882 skb->dev = tun->dev;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001883 break;
Michael S. Tsirkin40630b82014-11-19 15:17:31 +02001884 case IFF_TAP:
Petar Penkov90e33d42017-09-22 13:49:15 -07001885 if (!frags)
1886 skb->protocol = eth_type_trans(skb, tun->dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001887 break;
Joe Perches6403eab2011-06-03 11:51:20 +00001888 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001889
Michael S. Tsirkin06908992012-07-20 09:23:23 +00001890 /* copy skb_ubuf_info for callback when skb has no error */
1891 if (zerocopy) {
1892 skb_shinfo(skb)->destructor_arg = msg_control;
1893 skb_shinfo(skb)->tx_flags |= SKBTX_DEV_ZEROCOPY;
Pravin B Shelarc9af6db2013-02-11 09:27:41 +00001894 skb_shinfo(skb)->tx_flags |= SKBTX_SHARED_FRAG;
Jason Wangaf1cc7a2016-11-30 13:17:51 +08001895 } else if (msg_control) {
1896 struct ubuf_info *uarg = msg_control;
1897 uarg->callback(uarg, false);
Michael S. Tsirkin06908992012-07-20 09:23:23 +00001898 }
1899
Vlad Yasevich72f65102015-02-03 16:36:16 -05001900 skb_reset_network_header(skb);
Jason Wang40893fd2013-03-26 23:11:22 +00001901 skb_probe_transport_header(skb, 0);
Jason Wang38502af2013-03-25 20:19:56 +00001902
Jason Wang1cfe6e92017-09-04 11:36:09 +08001903 if (skb_xdp) {
Jason Wang761876c2017-08-11 19:41:18 +08001904 struct bpf_prog *xdp_prog;
1905 int ret;
1906
1907 rcu_read_lock();
1908 xdp_prog = rcu_dereference(tun->xdp_prog);
1909 if (xdp_prog) {
1910 ret = do_xdp_generic(xdp_prog, skb);
1911 if (ret != XDP_PASS) {
1912 rcu_read_unlock();
1913 return total_len;
1914 }
1915 }
1916 rcu_read_unlock();
1917 }
1918
Jason Wang96f84062017-12-04 17:31:23 +08001919 rcu_read_lock();
1920 if (!rcu_dereference(tun->steering_prog))
1921 rxhash = __skb_get_hash_symmetric(skb);
1922 rcu_read_unlock();
Petar Penkov94317092017-09-22 13:49:14 -07001923
Petar Penkov90e33d42017-09-22 13:49:15 -07001924 if (frags) {
1925 /* Exercise flow dissector code path. */
1926 u32 headlen = eth_get_headlen(skb->data, skb_headlen(skb));
1927
Eric Dumazet010f2452017-10-17 10:07:44 -07001928 if (unlikely(headlen > skb_headlen(skb))) {
Petar Penkov90e33d42017-09-22 13:49:15 -07001929 this_cpu_inc(tun->pcpu_stats->rx_dropped);
1930 napi_free_frags(&tfile->napi);
1931 mutex_unlock(&tfile->napi_mutex);
1932 WARN_ON(1);
1933 return -ENOMEM;
1934 }
1935
1936 local_bh_disable();
1937 napi_gro_frags(&tfile->napi);
1938 local_bh_enable();
1939 mutex_unlock(&tfile->napi_mutex);
Eric Dumazetaec72f32017-10-18 12:12:09 -07001940 } else if (tfile->napi_enabled) {
Petar Penkov94317092017-09-22 13:49:14 -07001941 struct sk_buff_head *queue = &tfile->sk.sk_write_queue;
1942 int queue_len;
1943
1944 spin_lock_bh(&queue->lock);
1945 __skb_queue_tail(queue, skb);
1946 queue_len = skb_queue_len(queue);
1947 spin_unlock(&queue->lock);
1948
1949 if (!more || queue_len > NAPI_POLL_WEIGHT)
1950 napi_schedule(&tfile->napi);
1951
1952 local_bh_enable();
1953 } else if (!IS_ENABLED(CONFIG_4KSTACKS)) {
1954 tun_rx_batched(tun, tfile, skb, more);
1955 } else {
1956 netif_rx_ni(skb);
1957 }
Jeff Garzik6aa20a22006-09-13 13:24:59 -04001958
Paolo Abeni608b9972016-04-13 10:52:20 +02001959 stats = get_cpu_ptr(tun->pcpu_stats);
1960 u64_stats_update_begin(&stats->syncp);
1961 stats->rx_packets++;
1962 stats->rx_bytes += len;
1963 u64_stats_update_end(&stats->syncp);
1964 put_cpu_ptr(stats);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001965
Jason Wang96f84062017-12-04 17:31:23 +08001966 if (rxhash)
1967 tun_flow_update(tun, rxhash, tfile);
1968
Michael S. Tsirkin06908992012-07-20 09:23:23 +00001969 return total_len;
Jeff Garzik6aa20a22006-09-13 13:24:59 -04001970}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001971
Al Virof5ff53b2014-06-19 15:36:49 -04001972static ssize_t tun_chr_write_iter(struct kiocb *iocb, struct iov_iter *from)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001973{
Herbert Xu33dccbb2009-02-05 21:25:32 -08001974 struct file *file = iocb->ki_filp;
Jason Wang54f968d2012-10-31 19:45:57 +00001975 struct tun_file *tfile = file->private_data;
yuan linyu9484dc72017-09-23 22:36:52 +08001976 struct tun_struct *tun = tun_get(tfile);
Eric W. Biederman631ab462009-01-20 11:00:40 +00001977 ssize_t result;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001978
1979 if (!tun)
1980 return -EBADFD;
1981
Jason Wang5503fce2017-01-18 15:02:03 +08001982 result = tun_get_user(tun, tfile, NULL, from,
1983 file->f_flags & O_NONBLOCK, false);
Eric W. Biederman631ab462009-01-20 11:00:40 +00001984
1985 tun_put(tun);
1986 return result;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001987}
1988
Jason Wangfc72d1d2018-01-04 11:14:28 +08001989static ssize_t tun_put_user_xdp(struct tun_struct *tun,
1990 struct tun_file *tfile,
1991 struct xdp_buff *xdp,
1992 struct iov_iter *iter)
1993{
1994 int vnet_hdr_sz = 0;
1995 size_t size = xdp->data_end - xdp->data;
1996 struct tun_pcpu_stats *stats;
1997 size_t ret;
1998
1999 if (tun->flags & IFF_VNET_HDR) {
2000 struct virtio_net_hdr gso = { 0 };
2001
2002 vnet_hdr_sz = READ_ONCE(tun->vnet_hdr_sz);
2003 if (unlikely(iov_iter_count(iter) < vnet_hdr_sz))
2004 return -EINVAL;
2005 if (unlikely(copy_to_iter(&gso, sizeof(gso), iter) !=
2006 sizeof(gso)))
2007 return -EFAULT;
2008 iov_iter_advance(iter, vnet_hdr_sz - sizeof(gso));
2009 }
2010
2011 ret = copy_to_iter(xdp->data, size, iter) + vnet_hdr_sz;
2012
2013 stats = get_cpu_ptr(tun->pcpu_stats);
2014 u64_stats_update_begin(&stats->syncp);
2015 stats->tx_packets++;
2016 stats->tx_bytes += ret;
2017 u64_stats_update_end(&stats->syncp);
2018 put_cpu_ptr(tun->pcpu_stats);
2019
2020 return ret;
2021}
2022
Linus Torvalds1da177e2005-04-16 15:20:36 -07002023/* Put packet to the user space buffer */
stephen hemminger6f7c1562011-06-08 14:33:08 +00002024static ssize_t tun_put_user(struct tun_struct *tun,
Jason Wang54f968d2012-10-31 19:45:57 +00002025 struct tun_file *tfile,
stephen hemminger6f7c1562011-06-08 14:33:08 +00002026 struct sk_buff *skb,
Herbert Xue0b46d02014-11-07 21:22:23 +08002027 struct iov_iter *iter)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002028{
2029 struct tun_pi pi = { 0, skb->protocol };
Paolo Abeni608b9972016-04-13 10:52:20 +02002030 struct tun_pcpu_stats *stats;
Herbert Xue0b46d02014-11-07 21:22:23 +08002031 ssize_t total;
Jason Wang8c847d22014-11-13 16:54:14 +08002032 int vlan_offset = 0;
Herbert Xua8f9bfd2014-11-03 04:30:13 +08002033 int vlan_hlen = 0;
Herbert Xu2eb783c2014-11-03 04:30:14 +08002034 int vnet_hdr_sz = 0;
Herbert Xua8f9bfd2014-11-03 04:30:13 +08002035
Jiri Pirkodf8a39d2015-01-13 17:13:44 +01002036 if (skb_vlan_tag_present(skb))
Herbert Xua8f9bfd2014-11-03 04:30:13 +08002037 vlan_hlen = VLAN_HLEN;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002038
Michael S. Tsirkin40630b82014-11-19 15:17:31 +02002039 if (tun->flags & IFF_VNET_HDR)
Willem de Bruijne1edab82017-02-03 18:20:48 -05002040 vnet_hdr_sz = READ_ONCE(tun->vnet_hdr_sz);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002041
Herbert Xue0b46d02014-11-07 21:22:23 +08002042 total = skb->len + vlan_hlen + vnet_hdr_sz;
2043
Michael S. Tsirkin40630b82014-11-19 15:17:31 +02002044 if (!(tun->flags & IFF_NO_PI)) {
Herbert Xue0b46d02014-11-07 21:22:23 +08002045 if (iov_iter_count(iter) < sizeof(pi))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002046 return -EINVAL;
2047
Herbert Xue0b46d02014-11-07 21:22:23 +08002048 total += sizeof(pi);
2049 if (iov_iter_count(iter) < total) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002050 /* Packet will be striped */
2051 pi.flags |= TUN_PKT_STRIP;
2052 }
Jeff Garzik6aa20a22006-09-13 13:24:59 -04002053
Herbert Xue0b46d02014-11-07 21:22:23 +08002054 if (copy_to_iter(&pi, sizeof(pi), iter) != sizeof(pi))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002055 return -EFAULT;
Jeff Garzik6aa20a22006-09-13 13:24:59 -04002056 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002057
Herbert Xu2eb783c2014-11-03 04:30:14 +08002058 if (vnet_hdr_sz) {
Jarno Rajahalme9403cd72016-11-18 15:40:40 -08002059 struct virtio_net_hdr gso;
Mike Rapoport34166092016-06-08 16:09:20 +03002060
Herbert Xue0b46d02014-11-07 21:22:23 +08002061 if (iov_iter_count(iter) < vnet_hdr_sz)
Rusty Russellf43798c2008-07-03 03:48:02 -07002062 return -EINVAL;
2063
Jarno Rajahalme3e9e40e2016-11-18 15:40:38 -08002064 if (virtio_net_hdr_from_skb(skb, &gso,
Jason Wang6391a442017-01-20 14:32:42 +08002065 tun_is_little_endian(tun), true)) {
Rusty Russellf43798c2008-07-03 03:48:02 -07002066 struct skb_shared_info *sinfo = skb_shinfo(skb);
Mike Rapoport34166092016-06-08 16:09:20 +03002067 pr_err("unexpected GSO type: "
2068 "0x%x, gso_size %d, hdr_len %d\n",
2069 sinfo->gso_type, tun16_to_cpu(tun, gso.gso_size),
2070 tun16_to_cpu(tun, gso.hdr_len));
2071 print_hex_dump(KERN_ERR, "tun: ",
2072 DUMP_PREFIX_NONE,
2073 16, 1, skb->head,
2074 min((int)tun16_to_cpu(tun, gso.hdr_len), 64), true);
2075 WARN_ON_ONCE(1);
2076 return -EINVAL;
2077 }
Rusty Russellf43798c2008-07-03 03:48:02 -07002078
Herbert Xue0b46d02014-11-07 21:22:23 +08002079 if (copy_to_iter(&gso, sizeof(gso), iter) != sizeof(gso))
Rusty Russellf43798c2008-07-03 03:48:02 -07002080 return -EFAULT;
Jason Wang8c847d22014-11-13 16:54:14 +08002081
2082 iov_iter_advance(iter, vnet_hdr_sz - sizeof(gso));
Rusty Russellf43798c2008-07-03 03:48:02 -07002083 }
2084
Herbert Xua8f9bfd2014-11-03 04:30:13 +08002085 if (vlan_hlen) {
Herbert Xue0b46d02014-11-07 21:22:23 +08002086 int ret;
Jason Wangaff3d702018-01-16 16:31:02 +08002087 struct veth veth;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002088
Jason Wang6680ec62013-07-25 13:00:33 +08002089 veth.h_vlan_proto = skb->vlan_proto;
Jiri Pirkodf8a39d2015-01-13 17:13:44 +01002090 veth.h_vlan_TCI = htons(skb_vlan_tag_get(skb));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002091
Jason Wang6680ec62013-07-25 13:00:33 +08002092 vlan_offset = offsetof(struct vlan_ethhdr, h_vlan_proto);
Jason Wang6680ec62013-07-25 13:00:33 +08002093
Herbert Xue0b46d02014-11-07 21:22:23 +08002094 ret = skb_copy_datagram_iter(skb, 0, iter, vlan_offset);
2095 if (ret || !iov_iter_count(iter))
Jason Wang6680ec62013-07-25 13:00:33 +08002096 goto done;
2097
Herbert Xue0b46d02014-11-07 21:22:23 +08002098 ret = copy_to_iter(&veth, sizeof(veth), iter);
2099 if (ret != sizeof(veth) || !iov_iter_count(iter))
Jason Wang6680ec62013-07-25 13:00:33 +08002100 goto done;
2101 }
2102
Herbert Xue0b46d02014-11-07 21:22:23 +08002103 skb_copy_datagram_iter(skb, vlan_offset, iter, skb->len - vlan_offset);
Jason Wang6680ec62013-07-25 13:00:33 +08002104
2105done:
Paolo Abeni608b9972016-04-13 10:52:20 +02002106 /* caller is in process context, */
2107 stats = get_cpu_ptr(tun->pcpu_stats);
2108 u64_stats_update_begin(&stats->syncp);
2109 stats->tx_packets++;
2110 stats->tx_bytes += skb->len + vlan_hlen;
2111 u64_stats_update_end(&stats->syncp);
2112 put_cpu_ptr(tun->pcpu_stats);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002113
2114 return total;
2115}
2116
Jason Wangfc72d1d2018-01-04 11:14:28 +08002117static void *tun_ring_recv(struct tun_file *tfile, int noblock, int *err)
Jason Wang1576d982016-06-30 14:45:36 +08002118{
2119 DECLARE_WAITQUEUE(wait, current);
Jason Wangfc72d1d2018-01-04 11:14:28 +08002120 void *ptr = NULL;
Jason Wangf48cc6b2016-07-04 13:53:38 +08002121 int error = 0;
Jason Wang1576d982016-06-30 14:45:36 +08002122
Jason Wangfc72d1d2018-01-04 11:14:28 +08002123 ptr = ptr_ring_consume(&tfile->tx_ring);
2124 if (ptr)
Jason Wang1576d982016-06-30 14:45:36 +08002125 goto out;
2126 if (noblock) {
Jason Wangf48cc6b2016-07-04 13:53:38 +08002127 error = -EAGAIN;
Jason Wang1576d982016-06-30 14:45:36 +08002128 goto out;
2129 }
2130
2131 add_wait_queue(&tfile->wq.wait, &wait);
2132 current->state = TASK_INTERRUPTIBLE;
2133
2134 while (1) {
Jason Wangfc72d1d2018-01-04 11:14:28 +08002135 ptr = ptr_ring_consume(&tfile->tx_ring);
2136 if (ptr)
Jason Wang1576d982016-06-30 14:45:36 +08002137 break;
2138 if (signal_pending(current)) {
Jason Wangf48cc6b2016-07-04 13:53:38 +08002139 error = -ERESTARTSYS;
Jason Wang1576d982016-06-30 14:45:36 +08002140 break;
2141 }
2142 if (tfile->socket.sk->sk_shutdown & RCV_SHUTDOWN) {
Jason Wangf48cc6b2016-07-04 13:53:38 +08002143 error = -EFAULT;
Jason Wang1576d982016-06-30 14:45:36 +08002144 break;
2145 }
2146
2147 schedule();
2148 }
2149
2150 current->state = TASK_RUNNING;
2151 remove_wait_queue(&tfile->wq.wait, &wait);
2152
2153out:
Jason Wangf48cc6b2016-07-04 13:53:38 +08002154 *err = error;
Jason Wangfc72d1d2018-01-04 11:14:28 +08002155 return ptr;
Jason Wang1576d982016-06-30 14:45:36 +08002156}
2157
Jason Wang54f968d2012-10-31 19:45:57 +00002158static ssize_t tun_do_read(struct tun_struct *tun, struct tun_file *tfile,
Al Viro9b067032014-11-07 13:52:07 -05002159 struct iov_iter *to,
Jason Wangfc72d1d2018-01-04 11:14:28 +08002160 int noblock, void *ptr)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002161{
Al Viro9b067032014-11-07 13:52:07 -05002162 ssize_t ret;
Jason Wang1576d982016-06-30 14:45:36 +08002163 int err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002164
Rami Rosen3872baf2012-11-25 22:07:41 +00002165 tun_debug(KERN_INFO, tun, "tun_do_read\n");
Linus Torvalds1da177e2005-04-16 15:20:36 -07002166
Wei Xuc33ee152017-12-01 05:10:37 -05002167 if (!iov_iter_count(to)) {
Jason Wangfc72d1d2018-01-04 11:14:28 +08002168 tun_ptr_free(ptr);
Al Viro9b067032014-11-07 13:52:07 -05002169 return 0;
Wei Xuc33ee152017-12-01 05:10:37 -05002170 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002171
Jason Wangfc72d1d2018-01-04 11:14:28 +08002172 if (!ptr) {
Jason Wangac77cfd2017-05-17 12:14:43 +08002173 /* Read frames from ring */
Jason Wangfc72d1d2018-01-04 11:14:28 +08002174 ptr = tun_ring_recv(tfile, noblock, &err);
2175 if (!ptr)
Jason Wangac77cfd2017-05-17 12:14:43 +08002176 return err;
2177 }
Herbert Xue0b46d02014-11-07 21:22:23 +08002178
Jason Wangfc72d1d2018-01-04 11:14:28 +08002179 if (tun_is_xdp_buff(ptr)) {
2180 struct xdp_buff *xdp = tun_ptr_to_xdp(ptr);
2181
2182 ret = tun_put_user_xdp(tun, tfile, xdp, to);
2183 put_page(virt_to_head_page(xdp->data));
2184 } else {
2185 struct sk_buff *skb = ptr;
2186
2187 ret = tun_put_user(tun, tfile, skb, to);
2188 if (unlikely(ret < 0))
2189 kfree_skb(skb);
2190 else
2191 consume_skb(skb);
2192 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002193
Michael S. Tsirkin05c28282010-01-14 06:17:09 +00002194 return ret;
2195}
2196
Al Viro9b067032014-11-07 13:52:07 -05002197static ssize_t tun_chr_read_iter(struct kiocb *iocb, struct iov_iter *to)
Michael S. Tsirkin05c28282010-01-14 06:17:09 +00002198{
2199 struct file *file = iocb->ki_filp;
2200 struct tun_file *tfile = file->private_data;
yuan linyu9484dc72017-09-23 22:36:52 +08002201 struct tun_struct *tun = tun_get(tfile);
Al Viro9b067032014-11-07 13:52:07 -05002202 ssize_t len = iov_iter_count(to), ret;
Michael S. Tsirkin05c28282010-01-14 06:17:09 +00002203
2204 if (!tun)
2205 return -EBADFD;
Jason Wangac77cfd2017-05-17 12:14:43 +08002206 ret = tun_do_read(tun, tfile, to, file->f_flags & O_NONBLOCK, NULL);
David S. Miller42404c02013-12-10 22:05:45 -05002207 ret = min_t(ssize_t, ret, len);
Zhi Yong Wud0b7da8a2013-12-06 14:16:51 +08002208 if (ret > 0)
2209 iocb->ki_pos = ret;
Eric W. Biederman631ab462009-01-20 11:00:40 +00002210 tun_put(tun);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002211 return ret;
2212}
2213
Jason Wangcd5681d2018-01-16 16:31:01 +08002214static void tun_prog_free(struct rcu_head *rcu)
Jason Wang96f84062017-12-04 17:31:23 +08002215{
Jason Wangcd5681d2018-01-16 16:31:01 +08002216 struct tun_prog *prog = container_of(rcu, struct tun_prog, rcu);
Jason Wang96f84062017-12-04 17:31:23 +08002217
2218 bpf_prog_destroy(prog->prog);
2219 kfree(prog);
2220}
2221
Jason Wang9d6474e2018-01-22 10:55:38 +08002222static int __tun_set_ebpf(struct tun_struct *tun,
2223 struct tun_prog __rcu **prog_p,
Jason Wangcd5681d2018-01-16 16:31:01 +08002224 struct bpf_prog *prog)
Jason Wang96f84062017-12-04 17:31:23 +08002225{
Jason Wangcd5681d2018-01-16 16:31:01 +08002226 struct tun_prog *old, *new = NULL;
Jason Wang96f84062017-12-04 17:31:23 +08002227
2228 if (prog) {
2229 new = kmalloc(sizeof(*new), GFP_KERNEL);
2230 if (!new)
2231 return -ENOMEM;
2232 new->prog = prog;
2233 }
2234
Jason Wang124da8f2017-12-08 12:02:30 +08002235 spin_lock_bh(&tun->lock);
Jason Wangcd5681d2018-01-16 16:31:01 +08002236 old = rcu_dereference_protected(*prog_p,
Jason Wang124da8f2017-12-08 12:02:30 +08002237 lockdep_is_held(&tun->lock));
Jason Wangcd5681d2018-01-16 16:31:01 +08002238 rcu_assign_pointer(*prog_p, new);
Jason Wang124da8f2017-12-08 12:02:30 +08002239 spin_unlock_bh(&tun->lock);
Jason Wang96f84062017-12-04 17:31:23 +08002240
2241 if (old)
Jason Wangcd5681d2018-01-16 16:31:01 +08002242 call_rcu(&old->rcu, tun_prog_free);
Jason Wang96f84062017-12-04 17:31:23 +08002243
2244 return 0;
2245}
2246
Jason Wang96442e422012-10-31 19:46:02 +00002247static void tun_free_netdev(struct net_device *dev)
2248{
2249 struct tun_struct *tun = netdev_priv(dev);
2250
Jason Wang4008e972012-12-13 23:53:30 +00002251 BUG_ON(!(list_empty(&tun->disabled)));
Paolo Abeni608b9972016-04-13 10:52:20 +02002252 free_percpu(tun->pcpu_stats);
Jason Wang96442e422012-10-31 19:46:02 +00002253 tun_flow_uninit(tun);
Paul Moore5dbbaf22013-01-14 07:12:19 +00002254 security_tun_dev_free_security(tun->security);
Jason Wangcd5681d2018-01-16 16:31:01 +08002255 __tun_set_ebpf(tun, &tun->steering_prog, NULL);
Jason Wangaff3d702018-01-16 16:31:02 +08002256 __tun_set_ebpf(tun, &tun->filter_prog, NULL);
Jason Wang96442e422012-10-31 19:46:02 +00002257}
2258
Linus Torvalds1da177e2005-04-16 15:20:36 -07002259static void tun_setup(struct net_device *dev)
2260{
2261 struct tun_struct *tun = netdev_priv(dev);
2262
Eric W. Biederman0625c882012-02-07 16:48:55 -08002263 tun->owner = INVALID_UID;
2264 tun->group = INVALID_GID;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002265
Linus Torvalds1da177e2005-04-16 15:20:36 -07002266 dev->ethtool_ops = &tun_ethtool_ops;
David S. Millercf124db2017-05-08 12:52:56 -04002267 dev->needs_free_netdev = true;
2268 dev->priv_destructor = tun_free_netdev;
Jason Wang016adb72016-04-08 13:26:48 +08002269 /* We prefer our own queue length */
2270 dev->tx_queue_len = TUN_READQ_SIZE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002271}
2272
Eric W. Biedermanf019a7a2009-01-21 16:02:16 -08002273/* Trivial set of netlink ops to allow deleting tun or tap
2274 * device with netlink.
2275 */
Matthias Schiffera8b8a8892017-06-25 23:56:01 +02002276static int tun_validate(struct nlattr *tb[], struct nlattr *data[],
2277 struct netlink_ext_ack *extack)
Eric W. Biedermanf019a7a2009-01-21 16:02:16 -08002278{
2279 return -EINVAL;
2280}
2281
2282static struct rtnl_link_ops tun_link_ops __read_mostly = {
2283 .kind = DRV_NAME,
2284 .priv_size = sizeof(struct tun_struct),
2285 .setup = tun_setup,
2286 .validate = tun_validate,
2287};
2288
Herbert Xu33dccbb2009-02-05 21:25:32 -08002289static void tun_sock_write_space(struct sock *sk)
2290{
Jason Wang54f968d2012-10-31 19:45:57 +00002291 struct tun_file *tfile;
Eric Dumazet43815482010-04-29 11:01:49 +00002292 wait_queue_head_t *wqueue;
Herbert Xu33dccbb2009-02-05 21:25:32 -08002293
2294 if (!sock_writeable(sk))
2295 return;
2296
Eric Dumazet9cd3e072015-11-29 20:03:10 -08002297 if (!test_and_clear_bit(SOCKWQ_ASYNC_NOSPACE, &sk->sk_socket->flags))
Herbert Xu33dccbb2009-02-05 21:25:32 -08002298 return;
2299
Eric Dumazet43815482010-04-29 11:01:49 +00002300 wqueue = sk_sleep(sk);
2301 if (wqueue && waitqueue_active(wqueue))
Linus Torvaldsa9a08842018-02-11 14:34:03 -08002302 wake_up_interruptible_sync_poll(wqueue, EPOLLOUT |
2303 EPOLLWRNORM | EPOLLWRBAND);
Herbert Xuc722c622009-06-03 21:45:55 -07002304
Jason Wang54f968d2012-10-31 19:45:57 +00002305 tfile = container_of(sk, struct tun_file, sk);
2306 kill_fasync(&tfile->fasync, SIGIO, POLL_OUT);
Herbert Xu33dccbb2009-02-05 21:25:32 -08002307}
2308
Ying Xue1b784142015-03-02 15:37:48 +08002309static int tun_sendmsg(struct socket *sock, struct msghdr *m, size_t total_len)
Michael S. Tsirkin05c28282010-01-14 06:17:09 +00002310{
Jason Wang54f968d2012-10-31 19:45:57 +00002311 int ret;
2312 struct tun_file *tfile = container_of(sock, struct tun_file, socket);
yuan linyu9484dc72017-09-23 22:36:52 +08002313 struct tun_struct *tun = tun_get(tfile);
Jason Wang54f968d2012-10-31 19:45:57 +00002314
2315 if (!tun)
2316 return -EBADFD;
Al Virof5ff53b2014-06-19 15:36:49 -04002317
Al Viroc0371da2014-11-24 10:42:55 -05002318 ret = tun_get_user(tun, tfile, m->msg_control, &m->msg_iter,
Jason Wang5503fce2017-01-18 15:02:03 +08002319 m->msg_flags & MSG_DONTWAIT,
2320 m->msg_flags & MSG_MORE);
Jason Wang54f968d2012-10-31 19:45:57 +00002321 tun_put(tun);
2322 return ret;
Michael S. Tsirkin05c28282010-01-14 06:17:09 +00002323}
2324
Ying Xue1b784142015-03-02 15:37:48 +08002325static int tun_recvmsg(struct socket *sock, struct msghdr *m, size_t total_len,
Michael S. Tsirkin05c28282010-01-14 06:17:09 +00002326 int flags)
2327{
Jason Wang54f968d2012-10-31 19:45:57 +00002328 struct tun_file *tfile = container_of(sock, struct tun_file, socket);
yuan linyu9484dc72017-09-23 22:36:52 +08002329 struct tun_struct *tun = tun_get(tfile);
Jason Wangfc72d1d2018-01-04 11:14:28 +08002330 void *ptr = m->msg_control;
Michael S. Tsirkin05c28282010-01-14 06:17:09 +00002331 int ret;
Jason Wang54f968d2012-10-31 19:45:57 +00002332
Wei Xuc33ee152017-12-01 05:10:37 -05002333 if (!tun) {
2334 ret = -EBADFD;
Jason Wangfc72d1d2018-01-04 11:14:28 +08002335 goto out_free;
Wei Xuc33ee152017-12-01 05:10:37 -05002336 }
Jason Wang54f968d2012-10-31 19:45:57 +00002337
Richard Cochraneda29772013-07-19 19:40:10 +02002338 if (flags & ~(MSG_DONTWAIT|MSG_TRUNC|MSG_ERRQUEUE)) {
Gao feng3811ae72013-04-24 21:59:23 +00002339 ret = -EINVAL;
Wei Xuc33ee152017-12-01 05:10:37 -05002340 goto out_put_tun;
Gao feng3811ae72013-04-24 21:59:23 +00002341 }
Richard Cochraneda29772013-07-19 19:40:10 +02002342 if (flags & MSG_ERRQUEUE) {
2343 ret = sock_recv_errqueue(sock->sk, m, total_len,
2344 SOL_PACKET, TUN_TX_TIMESTAMP);
2345 goto out;
2346 }
Jason Wangfc72d1d2018-01-04 11:14:28 +08002347 ret = tun_do_read(tun, tfile, &m->msg_iter, flags & MSG_DONTWAIT, ptr);
Alex Gartrell87897932014-12-25 23:05:03 -08002348 if (ret > (ssize_t)total_len) {
David S. Miller42404c02013-12-10 22:05:45 -05002349 m->msg_flags |= MSG_TRUNC;
2350 ret = flags & MSG_TRUNC ? ret : total_len;
2351 }
Gao feng3811ae72013-04-24 21:59:23 +00002352out:
Jason Wang54f968d2012-10-31 19:45:57 +00002353 tun_put(tun);
Michael S. Tsirkin05c28282010-01-14 06:17:09 +00002354 return ret;
Wei Xuc33ee152017-12-01 05:10:37 -05002355
2356out_put_tun:
2357 tun_put(tun);
Jason Wangfc72d1d2018-01-04 11:14:28 +08002358out_free:
2359 tun_ptr_free(ptr);
Wei Xuc33ee152017-12-01 05:10:37 -05002360 return ret;
Michael S. Tsirkin05c28282010-01-14 06:17:09 +00002361}
2362
Jason Wangfc72d1d2018-01-04 11:14:28 +08002363static int tun_ptr_peek_len(void *ptr)
2364{
2365 if (likely(ptr)) {
2366 if (tun_is_xdp_buff(ptr)) {
2367 struct xdp_buff *xdp = tun_ptr_to_xdp(ptr);
2368
2369 return xdp->data_end - xdp->data;
2370 }
2371 return __skb_array_len_with_tag(ptr);
2372 } else {
2373 return 0;
2374 }
2375}
2376
Jason Wang1576d982016-06-30 14:45:36 +08002377static int tun_peek_len(struct socket *sock)
2378{
2379 struct tun_file *tfile = container_of(sock, struct tun_file, socket);
2380 struct tun_struct *tun;
2381 int ret = 0;
2382
yuan linyu9484dc72017-09-23 22:36:52 +08002383 tun = tun_get(tfile);
Jason Wang1576d982016-06-30 14:45:36 +08002384 if (!tun)
2385 return 0;
2386
Jason Wangfc72d1d2018-01-04 11:14:28 +08002387 ret = PTR_RING_PEEK_CALL(&tfile->tx_ring, tun_ptr_peek_len);
Jason Wang1576d982016-06-30 14:45:36 +08002388 tun_put(tun);
2389
2390 return ret;
2391}
2392
Michael S. Tsirkin05c28282010-01-14 06:17:09 +00002393/* Ops structure to mimic raw sockets with tun */
2394static const struct proto_ops tun_socket_ops = {
Jason Wang1576d982016-06-30 14:45:36 +08002395 .peek_len = tun_peek_len,
Michael S. Tsirkin05c28282010-01-14 06:17:09 +00002396 .sendmsg = tun_sendmsg,
2397 .recvmsg = tun_recvmsg,
2398};
2399
Herbert Xu33dccbb2009-02-05 21:25:32 -08002400static struct proto tun_proto = {
2401 .name = "tun",
2402 .owner = THIS_MODULE,
Jason Wang54f968d2012-10-31 19:45:57 +00002403 .obj_size = sizeof(struct tun_file),
Herbert Xu33dccbb2009-02-05 21:25:32 -08002404};
Eric W. Biedermanf019a7a2009-01-21 16:02:16 -08002405
David Woodhouse980c9e82009-05-09 22:54:21 -07002406static int tun_flags(struct tun_struct *tun)
2407{
Michael S. Tsirkin031f5e032014-11-19 14:44:40 +02002408 return tun->flags & (TUN_FEATURES | IFF_PERSIST | IFF_TUN | IFF_TAP);
David Woodhouse980c9e82009-05-09 22:54:21 -07002409}
2410
2411static ssize_t tun_show_flags(struct device *dev, struct device_attribute *attr,
2412 char *buf)
2413{
2414 struct tun_struct *tun = netdev_priv(to_net_dev(dev));
2415 return sprintf(buf, "0x%x\n", tun_flags(tun));
2416}
2417
2418static ssize_t tun_show_owner(struct device *dev, struct device_attribute *attr,
2419 char *buf)
2420{
2421 struct tun_struct *tun = netdev_priv(to_net_dev(dev));
Eric W. Biederman0625c882012-02-07 16:48:55 -08002422 return uid_valid(tun->owner)?
2423 sprintf(buf, "%u\n",
2424 from_kuid_munged(current_user_ns(), tun->owner)):
2425 sprintf(buf, "-1\n");
David Woodhouse980c9e82009-05-09 22:54:21 -07002426}
2427
2428static ssize_t tun_show_group(struct device *dev, struct device_attribute *attr,
2429 char *buf)
2430{
2431 struct tun_struct *tun = netdev_priv(to_net_dev(dev));
Eric W. Biederman0625c882012-02-07 16:48:55 -08002432 return gid_valid(tun->group) ?
2433 sprintf(buf, "%u\n",
2434 from_kgid_munged(current_user_ns(), tun->group)):
2435 sprintf(buf, "-1\n");
David Woodhouse980c9e82009-05-09 22:54:21 -07002436}
2437
2438static DEVICE_ATTR(tun_flags, 0444, tun_show_flags, NULL);
2439static DEVICE_ATTR(owner, 0444, tun_show_owner, NULL);
2440static DEVICE_ATTR(group, 0444, tun_show_group, NULL);
2441
Takashi Iwaic4d33e22015-02-04 14:37:34 +01002442static struct attribute *tun_dev_attrs[] = {
2443 &dev_attr_tun_flags.attr,
2444 &dev_attr_owner.attr,
2445 &dev_attr_group.attr,
2446 NULL
2447};
2448
2449static const struct attribute_group tun_attr_group = {
2450 .attrs = tun_dev_attrs
2451};
2452
Pavel Emelyanovd647a592008-04-16 00:41:16 -07002453static int tun_set_iff(struct net *net, struct file *file, struct ifreq *ifr)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002454{
2455 struct tun_struct *tun;
Jason Wang54f968d2012-10-31 19:45:57 +00002456 struct tun_file *tfile = file->private_data;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002457 struct net_device *dev;
2458 int err;
2459
Jason Wang7c0c3b12013-01-11 16:59:33 +00002460 if (tfile->detached)
2461 return -EINVAL;
2462
Petar Penkov90e33d42017-09-22 13:49:15 -07002463 if ((ifr->ifr_flags & IFF_NAPI_FRAGS)) {
2464 if (!capable(CAP_NET_ADMIN))
2465 return -EPERM;
2466
2467 if (!(ifr->ifr_flags & IFF_NAPI) ||
2468 (ifr->ifr_flags & TUN_TYPE_MASK) != IFF_TAP)
2469 return -EINVAL;
2470 }
2471
Eric W. Biederman74a3e5a2009-01-20 10:56:20 +00002472 dev = __dev_get_by_name(net, ifr->ifr_name);
2473 if (dev) {
David Woodhousef85ba782009-04-27 03:23:54 -07002474 if (ifr->ifr_flags & IFF_TUN_EXCL)
2475 return -EBUSY;
Eric W. Biederman74a3e5a2009-01-20 10:56:20 +00002476 if ((ifr->ifr_flags & IFF_TUN) && dev->netdev_ops == &tun_netdev_ops)
2477 tun = netdev_priv(dev);
2478 else if ((ifr->ifr_flags & IFF_TAP) && dev->netdev_ops == &tap_netdev_ops)
2479 tun = netdev_priv(dev);
2480 else
2481 return -EINVAL;
2482
Jason Wang8e6d91a2013-05-28 18:32:11 +00002483 if (!!(ifr->ifr_flags & IFF_MULTI_QUEUE) !=
Michael S. Tsirkin40630b82014-11-19 15:17:31 +02002484 !!(tun->flags & IFF_MULTI_QUEUE))
Jason Wang8e6d91a2013-05-28 18:32:11 +00002485 return -EINVAL;
2486
Jason Wangcde8b152012-10-31 19:46:01 +00002487 if (tun_not_capable(tun))
Paul Moore2b980db2009-08-28 18:12:43 -04002488 return -EPERM;
Paul Moore5dbbaf22013-01-14 07:12:19 +00002489 err = security_tun_dev_open(tun->security);
Paul Moore2b980db2009-08-28 18:12:43 -04002490 if (err < 0)
2491 return err;
2492
Petar Penkov94317092017-09-22 13:49:14 -07002493 err = tun_attach(tun, file, ifr->ifr_flags & IFF_NOFILTER,
2494 ifr->ifr_flags & IFF_NAPI);
Eric W. Biedermana7385ba2009-01-20 10:57:48 +00002495 if (err < 0)
2496 return err;
Jason Wang4008e972012-12-13 23:53:30 +00002497
Michael S. Tsirkin40630b82014-11-19 15:17:31 +02002498 if (tun->flags & IFF_MULTI_QUEUE &&
Jason Wange8dbad62013-04-22 20:40:39 +00002499 (tun->numqueues + tun->numdisabled > 1)) {
2500 /* One or more queue has already been attached, no need
2501 * to initialize the device again.
2502 */
2503 return 0;
2504 }
Jeff Garzik6aa20a22006-09-13 13:24:59 -04002505 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002506 else {
2507 char *name;
2508 unsigned long flags = 0;
Jason Wangedfb6a12013-01-23 03:59:12 +00002509 int queues = ifr->ifr_flags & IFF_MULTI_QUEUE ?
2510 MAX_TAP_QUEUES : 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002511
Eric W. Biedermanc260b772012-11-18 21:34:11 +00002512 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
David Woodhouseca6bb5d2006-06-22 16:07:52 -07002513 return -EPERM;
Paul Moore2b980db2009-08-28 18:12:43 -04002514 err = security_tun_dev_create();
2515 if (err < 0)
2516 return err;
David Woodhouseca6bb5d2006-06-22 16:07:52 -07002517
Linus Torvalds1da177e2005-04-16 15:20:36 -07002518 /* Set dev type */
2519 if (ifr->ifr_flags & IFF_TUN) {
2520 /* TUN device */
Michael S. Tsirkin40630b82014-11-19 15:17:31 +02002521 flags |= IFF_TUN;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002522 name = "tun%d";
2523 } else if (ifr->ifr_flags & IFF_TAP) {
2524 /* TAP device */
Michael S. Tsirkin40630b82014-11-19 15:17:31 +02002525 flags |= IFF_TAP;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002526 name = "tap%d";
Jeff Garzik6aa20a22006-09-13 13:24:59 -04002527 } else
Kusanagi Kouichi36989b92009-09-16 21:36:13 +00002528 return -EINVAL;
Jeff Garzik6aa20a22006-09-13 13:24:59 -04002529
Linus Torvalds1da177e2005-04-16 15:20:36 -07002530 if (*ifr->ifr_name)
2531 name = ifr->ifr_name;
2532
Jason Wangc8d68e62012-10-31 19:46:00 +00002533 dev = alloc_netdev_mqs(sizeof(struct tun_struct), name,
Tom Gundersenc835a672014-07-14 16:37:24 +02002534 NET_NAME_UNKNOWN, tun_setup, queues,
2535 queues);
Jason Wangedfb6a12013-01-23 03:59:12 +00002536
Linus Torvalds1da177e2005-04-16 15:20:36 -07002537 if (!dev)
2538 return -ENOMEM;
Cong Wang0ad646c2017-10-13 11:58:53 -07002539 err = dev_get_valid_name(net, dev, name);
Julien Gomes5c25f652017-10-25 11:50:50 -07002540 if (err < 0)
Cong Wang0ad646c2017-10-13 11:58:53 -07002541 goto err_free_dev;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002542
Pavel Emelyanovfc54c652008-04-16 00:41:53 -07002543 dev_net_set(dev, net);
Eric W. Biedermanf019a7a2009-01-21 16:02:16 -08002544 dev->rtnl_link_ops = &tun_link_ops;
Pavel Emelyanovfb7589a2013-08-21 14:31:38 +04002545 dev->ifindex = tfile->ifindex;
Takashi Iwaic4d33e22015-02-04 14:37:34 +01002546 dev->sysfs_groups[0] = &tun_attr_group;
Stephen Hemminger758e43b2008-11-19 22:10:37 -08002547
Linus Torvalds1da177e2005-04-16 15:20:36 -07002548 tun = netdev_priv(dev);
2549 tun->dev = dev;
2550 tun->flags = flags;
Max Krasnyanskyf271b2cc2008-07-14 22:18:19 -07002551 tun->txflt.count = 0;
Michael S. Tsirkind9d52b52010-03-17 17:45:01 +02002552 tun->vnet_hdr_sz = sizeof(struct virtio_net_hdr);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002553
Paolo Abenieaea34b2016-02-26 10:45:40 +01002554 tun->align = NET_SKB_PAD;
Jason Wang54f968d2012-10-31 19:45:57 +00002555 tun->filter_attached = false;
2556 tun->sndbuf = tfile->socket.sk->sk_sndbuf;
Jason Wang5503fce2017-01-18 15:02:03 +08002557 tun->rx_batched = 0;
Jason Wang96f84062017-12-04 17:31:23 +08002558 RCU_INIT_POINTER(tun->steering_prog, NULL);
Herbert Xu33dccbb2009-02-05 21:25:32 -08002559
Paolo Abeni608b9972016-04-13 10:52:20 +02002560 tun->pcpu_stats = netdev_alloc_pcpu_stats(struct tun_pcpu_stats);
2561 if (!tun->pcpu_stats) {
2562 err = -ENOMEM;
2563 goto err_free_dev;
2564 }
2565
Jason Wang96442e422012-10-31 19:46:02 +00002566 spin_lock_init(&tun->lock);
2567
Paul Moore5dbbaf22013-01-14 07:12:19 +00002568 err = security_tun_dev_alloc_security(&tun->security);
2569 if (err < 0)
Paolo Abeni608b9972016-04-13 10:52:20 +02002570 goto err_free_stat;
Paul Moore2b980db2009-08-28 18:12:43 -04002571
Linus Torvalds1da177e2005-04-16 15:20:36 -07002572 tun_net_init(dev);
Pavel Emelyanov944a1372013-06-11 17:01:08 +04002573 tun_flow_init(tun);
Jason Wang96442e422012-10-31 19:46:02 +00002574
Michał Mirosław88255372011-04-19 06:13:10 +00002575 dev->hw_features = NETIF_F_SG | NETIF_F_FRAGLIST |
Jason Wang6680ec62013-07-25 13:00:33 +08002576 TUN_USER_FEATURES | NETIF_F_HW_VLAN_CTAG_TX |
2577 NETIF_F_HW_VLAN_STAG_TX;
Paolo Abeni2a2bbf12016-04-14 18:39:39 +02002578 dev->features = dev->hw_features | NETIF_F_LLTX;
Fernando Luis Vazquez Cao6671b222014-02-18 21:20:09 +09002579 dev->vlan_features = dev->features &
2580 ~(NETIF_F_HW_VLAN_CTAG_TX |
2581 NETIF_F_HW_VLAN_STAG_TX);
Michał Mirosław88255372011-04-19 06:13:10 +00002582
Jason Wang4008e972012-12-13 23:53:30 +00002583 INIT_LIST_HEAD(&tun->disabled);
Petar Penkov94317092017-09-22 13:49:14 -07002584 err = tun_attach(tun, file, false, ifr->ifr_flags & IFF_NAPI);
Jason Wangeb0fb362012-12-02 17:19:45 +00002585 if (err < 0)
Jason Wang662ca432013-09-11 18:09:48 +08002586 goto err_free_flow;
Jason Wangeb0fb362012-12-02 17:19:45 +00002587
Linus Torvalds1da177e2005-04-16 15:20:36 -07002588 err = register_netdevice(tun->dev);
2589 if (err < 0)
Jason Wang662ca432013-09-11 18:09:48 +08002590 goto err_detach;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002591 }
2592
Michael S. Tsirkinaf668b32013-01-28 00:38:02 +00002593 netif_carrier_on(tun->dev);
2594
Joe Perches6b8a66e2011-03-02 07:18:10 +00002595 tun_debug(KERN_INFO, tun, "tun_set_iff\n");
Linus Torvalds1da177e2005-04-16 15:20:36 -07002596
Michael S. Tsirkin031f5e032014-11-19 14:44:40 +02002597 tun->flags = (tun->flags & ~TUN_FEATURES) |
2598 (ifr->ifr_flags & TUN_FEATURES);
Jason Wangc8d68e62012-10-31 19:46:00 +00002599
Max Krasnyanskye35259a2008-07-10 16:59:11 -07002600 /* Make sure persistent devices do not get stuck in
2601 * xoff state.
2602 */
2603 if (netif_running(tun->dev))
Jason Wangc8d68e62012-10-31 19:46:00 +00002604 netif_tx_wake_all_queues(tun->dev);
Max Krasnyanskye35259a2008-07-10 16:59:11 -07002605
Linus Torvalds1da177e2005-04-16 15:20:36 -07002606 strcpy(ifr->ifr_name, tun->dev->name);
2607 return 0;
2608
Jason Wang662ca432013-09-11 18:09:48 +08002609err_detach:
2610 tun_detach_all(dev);
Eric Dumazetff244c62017-08-18 13:39:56 -07002611 /* register_netdevice() already called tun_free_netdev() */
2612 goto err_free_dev;
2613
Jason Wang662ca432013-09-11 18:09:48 +08002614err_free_flow:
2615 tun_flow_uninit(tun);
2616 security_tun_dev_free_security(tun->security);
Paolo Abeni608b9972016-04-13 10:52:20 +02002617err_free_stat:
2618 free_percpu(tun->pcpu_stats);
Jason Wang662ca432013-09-11 18:09:48 +08002619err_free_dev:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002620 free_netdev(dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002621 return err;
2622}
2623
Rami Rosen9ce99cf2012-11-23 03:58:10 +00002624static void tun_get_iff(struct net *net, struct tun_struct *tun,
Herbert Xu876bfd42009-08-06 14:22:44 +00002625 struct ifreq *ifr)
Mark McLoughline3b99552008-08-15 15:09:56 -07002626{
Joe Perches6b8a66e2011-03-02 07:18:10 +00002627 tun_debug(KERN_INFO, tun, "tun_get_iff\n");
Mark McLoughline3b99552008-08-15 15:09:56 -07002628
2629 strcpy(ifr->ifr_name, tun->dev->name);
2630
David Woodhouse980c9e82009-05-09 22:54:21 -07002631 ifr->ifr_flags = tun_flags(tun);
Mark McLoughline3b99552008-08-15 15:09:56 -07002632
Mark McLoughline3b99552008-08-15 15:09:56 -07002633}
2634
Rusty Russell5228ddc2008-07-03 03:46:16 -07002635/* This is like a cut-down ethtool ops, except done via tun fd so no
2636 * privs required. */
Michał Mirosław88255372011-04-19 06:13:10 +00002637static int set_offload(struct tun_struct *tun, unsigned long arg)
Rusty Russell5228ddc2008-07-03 03:46:16 -07002638{
Michał Mirosławc8f44af2011-11-15 15:29:55 +00002639 netdev_features_t features = 0;
Rusty Russell5228ddc2008-07-03 03:46:16 -07002640
2641 if (arg & TUN_F_CSUM) {
Michał Mirosław88255372011-04-19 06:13:10 +00002642 features |= NETIF_F_HW_CSUM;
Rusty Russell5228ddc2008-07-03 03:46:16 -07002643 arg &= ~TUN_F_CSUM;
2644
2645 if (arg & (TUN_F_TSO4|TUN_F_TSO6)) {
2646 if (arg & TUN_F_TSO_ECN) {
2647 features |= NETIF_F_TSO_ECN;
2648 arg &= ~TUN_F_TSO_ECN;
2649 }
2650 if (arg & TUN_F_TSO4)
2651 features |= NETIF_F_TSO;
2652 if (arg & TUN_F_TSO6)
2653 features |= NETIF_F_TSO6;
2654 arg &= ~(TUN_F_TSO4|TUN_F_TSO6);
2655 }
Willem de Bruijn0c19f8462017-11-21 10:22:25 -05002656
2657 arg &= ~TUN_F_UFO;
Rusty Russell5228ddc2008-07-03 03:46:16 -07002658 }
2659
2660 /* This gives the user a way to test for new features in future by
2661 * trying to set them. */
2662 if (arg)
2663 return -EINVAL;
2664
Michał Mirosław88255372011-04-19 06:13:10 +00002665 tun->set_features = features;
Yaroslav Isakov09050952017-03-16 22:44:10 +03002666 tun->dev->wanted_features &= ~TUN_USER_FEATURES;
2667 tun->dev->wanted_features |= features;
Michał Mirosław88255372011-04-19 06:13:10 +00002668 netdev_update_features(tun->dev);
Rusty Russell5228ddc2008-07-03 03:46:16 -07002669
2670 return 0;
2671}
2672
Jason Wangc8d68e62012-10-31 19:46:00 +00002673static void tun_detach_filter(struct tun_struct *tun, int n)
2674{
2675 int i;
2676 struct tun_file *tfile;
2677
2678 for (i = 0; i < n; i++) {
Jason Wangb8deabd2013-01-11 16:59:32 +00002679 tfile = rtnl_dereference(tun->tfiles[i]);
Hannes Frederic Sowa8ced4252016-04-05 17:10:16 +02002680 lock_sock(tfile->socket.sk);
2681 sk_detach_filter(tfile->socket.sk);
2682 release_sock(tfile->socket.sk);
Jason Wangc8d68e62012-10-31 19:46:00 +00002683 }
2684
2685 tun->filter_attached = false;
2686}
2687
2688static int tun_attach_filter(struct tun_struct *tun)
2689{
2690 int i, ret = 0;
2691 struct tun_file *tfile;
2692
2693 for (i = 0; i < tun->numqueues; i++) {
Jason Wangb8deabd2013-01-11 16:59:32 +00002694 tfile = rtnl_dereference(tun->tfiles[i]);
Hannes Frederic Sowa8ced4252016-04-05 17:10:16 +02002695 lock_sock(tfile->socket.sk);
2696 ret = sk_attach_filter(&tun->fprog, tfile->socket.sk);
2697 release_sock(tfile->socket.sk);
Jason Wangc8d68e62012-10-31 19:46:00 +00002698 if (ret) {
2699 tun_detach_filter(tun, i);
2700 return ret;
2701 }
2702 }
2703
2704 tun->filter_attached = true;
2705 return ret;
2706}
2707
2708static void tun_set_sndbuf(struct tun_struct *tun)
2709{
2710 struct tun_file *tfile;
2711 int i;
2712
2713 for (i = 0; i < tun->numqueues; i++) {
Jason Wangb8deabd2013-01-11 16:59:32 +00002714 tfile = rtnl_dereference(tun->tfiles[i]);
Jason Wangc8d68e62012-10-31 19:46:00 +00002715 tfile->socket.sk->sk_sndbuf = tun->sndbuf;
2716 }
2717}
2718
Jason Wangcde8b152012-10-31 19:46:01 +00002719static int tun_set_queue(struct file *file, struct ifreq *ifr)
2720{
2721 struct tun_file *tfile = file->private_data;
2722 struct tun_struct *tun;
Jason Wangcde8b152012-10-31 19:46:01 +00002723 int ret = 0;
2724
2725 rtnl_lock();
2726
2727 if (ifr->ifr_flags & IFF_ATTACH_QUEUE) {
Jason Wang4008e972012-12-13 23:53:30 +00002728 tun = tfile->detached;
Paul Moore5dbbaf22013-01-14 07:12:19 +00002729 if (!tun) {
Jason Wangcde8b152012-10-31 19:46:01 +00002730 ret = -EINVAL;
Paul Moore5dbbaf22013-01-14 07:12:19 +00002731 goto unlock;
2732 }
2733 ret = security_tun_dev_attach_queue(tun->security);
2734 if (ret < 0)
2735 goto unlock;
Petar Penkov94317092017-09-22 13:49:14 -07002736 ret = tun_attach(tun, file, false, tun->flags & IFF_NAPI);
Jason Wang4008e972012-12-13 23:53:30 +00002737 } else if (ifr->ifr_flags & IFF_DETACH_QUEUE) {
Jason Wangb8deabd2013-01-11 16:59:32 +00002738 tun = rtnl_dereference(tfile->tun);
Michael S. Tsirkin40630b82014-11-19 15:17:31 +02002739 if (!tun || !(tun->flags & IFF_MULTI_QUEUE) || tfile->detached)
Jason Wang4008e972012-12-13 23:53:30 +00002740 ret = -EINVAL;
2741 else
2742 __tun_detach(tfile, false);
2743 } else
Jason Wangcde8b152012-10-31 19:46:01 +00002744 ret = -EINVAL;
2745
Paul Moore5dbbaf22013-01-14 07:12:19 +00002746unlock:
Jason Wangcde8b152012-10-31 19:46:01 +00002747 rtnl_unlock();
2748 return ret;
2749}
2750
Jason Wangcd5681d2018-01-16 16:31:01 +08002751static int tun_set_ebpf(struct tun_struct *tun, struct tun_prog **prog_p,
2752 void __user *data)
Jason Wang96f84062017-12-04 17:31:23 +08002753{
2754 struct bpf_prog *prog;
2755 int fd;
2756
2757 if (copy_from_user(&fd, data, sizeof(fd)))
2758 return -EFAULT;
2759
2760 if (fd == -1) {
2761 prog = NULL;
2762 } else {
2763 prog = bpf_prog_get_type(fd, BPF_PROG_TYPE_SOCKET_FILTER);
2764 if (IS_ERR(prog))
2765 return PTR_ERR(prog);
2766 }
2767
Jason Wangcd5681d2018-01-16 16:31:01 +08002768 return __tun_set_ebpf(tun, prog_p, prog);
Jason Wang96f84062017-12-04 17:31:23 +08002769}
2770
Arnd Bergmann50857e22009-11-06 22:52:32 -08002771static long __tun_chr_ioctl(struct file *file, unsigned int cmd,
2772 unsigned long arg, int ifreq_len)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002773{
Eric W. Biederman36b50ba2009-01-20 11:01:48 +00002774 struct tun_file *tfile = file->private_data;
Eric W. Biederman631ab462009-01-20 11:00:40 +00002775 struct tun_struct *tun;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002776 void __user* argp = (void __user*)arg;
2777 struct ifreq ifr;
Eric W. Biederman0625c882012-02-07 16:48:55 -08002778 kuid_t owner;
2779 kgid_t group;
Herbert Xu33dccbb2009-02-05 21:25:32 -08002780 int sndbuf;
Michael S. Tsirkind9d52b52010-03-17 17:45:01 +02002781 int vnet_hdr_sz;
Pavel Emelyanovfb7589a2013-08-21 14:31:38 +04002782 unsigned int ifindex;
Michael S. Tsirkin1cf8e412014-12-16 15:05:06 +02002783 int le;
Max Krasnyanskyf271b2cc2008-07-14 22:18:19 -07002784 int ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002785
Gao Feng20861f22016-10-27 09:05:22 +08002786 if (cmd == TUNSETIFF || cmd == TUNSETQUEUE || _IOC_TYPE(cmd) == SOCK_IOC_TYPE) {
Arnd Bergmann50857e22009-11-06 22:52:32 -08002787 if (copy_from_user(&ifr, argp, ifreq_len))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002788 return -EFAULT;
David S. Miller8bbb1812012-07-30 14:52:48 -07002789 } else {
Mathias Krausea117dac2012-07-29 19:45:14 +00002790 memset(&ifr, 0, sizeof(ifr));
David S. Miller8bbb1812012-07-30 14:52:48 -07002791 }
Eric W. Biederman631ab462009-01-20 11:00:40 +00002792 if (cmd == TUNGETFEATURES) {
2793 /* Currently this just means: "what IFF flags are valid?".
2794 * This is needed because we never checked for invalid flags on
Michael S. Tsirkin031f5e032014-11-19 14:44:40 +02002795 * TUNSETIFF.
2796 */
2797 return put_user(IFF_TUN | IFF_TAP | TUN_FEATURES,
Eric W. Biederman631ab462009-01-20 11:00:40 +00002798 (unsigned int __user*)argp);
Jason Wangcde8b152012-10-31 19:46:01 +00002799 } else if (cmd == TUNSETQUEUE)
2800 return tun_set_queue(file, &ifr);
Eric W. Biederman631ab462009-01-20 11:00:40 +00002801
Jason Wangc8d68e62012-10-31 19:46:00 +00002802 ret = 0;
Herbert Xu876bfd42009-08-06 14:22:44 +00002803 rtnl_lock();
2804
yuan linyu9484dc72017-09-23 22:36:52 +08002805 tun = tun_get(tfile);
Gao Feng0f16bc12016-10-25 22:26:09 +08002806 if (cmd == TUNSETIFF) {
2807 ret = -EEXIST;
2808 if (tun)
2809 goto unlock;
2810
Linus Torvalds1da177e2005-04-16 15:20:36 -07002811 ifr.ifr_name[IFNAMSIZ-1] = '\0';
2812
Eric W. Biederman140e807da2015-05-08 21:07:08 -05002813 ret = tun_set_iff(sock_net(&tfile->sk), file, &ifr);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002814
Herbert Xu876bfd42009-08-06 14:22:44 +00002815 if (ret)
2816 goto unlock;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002817
Arnd Bergmann50857e22009-11-06 22:52:32 -08002818 if (copy_to_user(argp, &ifr, ifreq_len))
Herbert Xu876bfd42009-08-06 14:22:44 +00002819 ret = -EFAULT;
2820 goto unlock;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002821 }
Pavel Emelyanovfb7589a2013-08-21 14:31:38 +04002822 if (cmd == TUNSETIFINDEX) {
2823 ret = -EPERM;
2824 if (tun)
2825 goto unlock;
2826
2827 ret = -EFAULT;
2828 if (copy_from_user(&ifindex, argp, sizeof(ifindex)))
2829 goto unlock;
2830
2831 ret = 0;
2832 tfile->ifindex = ifindex;
2833 goto unlock;
2834 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002835
Herbert Xu876bfd42009-08-06 14:22:44 +00002836 ret = -EBADFD;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002837 if (!tun)
Herbert Xu876bfd42009-08-06 14:22:44 +00002838 goto unlock;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002839
Jason Wang1e588332012-10-31 19:45:56 +00002840 tun_debug(KERN_INFO, tun, "tun_chr_ioctl cmd %u\n", cmd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002841
Eric W. Biederman631ab462009-01-20 11:00:40 +00002842 ret = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002843 switch (cmd) {
Mark McLoughline3b99552008-08-15 15:09:56 -07002844 case TUNGETIFF:
Rami Rosen9ce99cf2012-11-23 03:58:10 +00002845 tun_get_iff(current->nsproxy->net_ns, tun, &ifr);
Mark McLoughline3b99552008-08-15 15:09:56 -07002846
Pavel Emelyanov3d407a82013-08-21 14:32:00 +04002847 if (tfile->detached)
2848 ifr.ifr_flags |= IFF_DETACH_QUEUE;
Pavel Emelyanov849c9b62013-08-21 14:32:21 +04002849 if (!tfile->socket.sk->sk_filter)
2850 ifr.ifr_flags |= IFF_NOFILTER;
Pavel Emelyanov3d407a82013-08-21 14:32:00 +04002851
Arnd Bergmann50857e22009-11-06 22:52:32 -08002852 if (copy_to_user(argp, &ifr, ifreq_len))
Eric W. Biederman631ab462009-01-20 11:00:40 +00002853 ret = -EFAULT;
Mark McLoughline3b99552008-08-15 15:09:56 -07002854 break;
2855
Linus Torvalds1da177e2005-04-16 15:20:36 -07002856 case TUNSETNOCSUM:
2857 /* Disable/Enable checksum */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002858
Michał Mirosław88255372011-04-19 06:13:10 +00002859 /* [unimplemented] */
2860 tun_debug(KERN_INFO, tun, "ignored: set checksum %s\n",
Joe Perches6b8a66e2011-03-02 07:18:10 +00002861 arg ? "disabled" : "enabled");
Linus Torvalds1da177e2005-04-16 15:20:36 -07002862 break;
2863
2864 case TUNSETPERSIST:
Jason Wang54f968d2012-10-31 19:45:57 +00002865 /* Disable/Enable persist mode. Keep an extra reference to the
2866 * module to prevent the module being unprobed.
2867 */
Michael S. Tsirkin40630b82014-11-19 15:17:31 +02002868 if (arg && !(tun->flags & IFF_PERSIST)) {
2869 tun->flags |= IFF_PERSIST;
Jason Wang54f968d2012-10-31 19:45:57 +00002870 __module_get(THIS_MODULE);
Jason Wangdd38bd82013-01-11 16:59:34 +00002871 }
Michael S. Tsirkin40630b82014-11-19 15:17:31 +02002872 if (!arg && (tun->flags & IFF_PERSIST)) {
2873 tun->flags &= ~IFF_PERSIST;
Jason Wang54f968d2012-10-31 19:45:57 +00002874 module_put(THIS_MODULE);
2875 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002876
Joe Perches6b8a66e2011-03-02 07:18:10 +00002877 tun_debug(KERN_INFO, tun, "persist %s\n",
2878 arg ? "enabled" : "disabled");
Linus Torvalds1da177e2005-04-16 15:20:36 -07002879 break;
2880
2881 case TUNSETOWNER:
2882 /* Set owner of the device */
Eric W. Biederman0625c882012-02-07 16:48:55 -08002883 owner = make_kuid(current_user_ns(), arg);
2884 if (!uid_valid(owner)) {
2885 ret = -EINVAL;
2886 break;
2887 }
2888 tun->owner = owner;
Jason Wang1e588332012-10-31 19:45:56 +00002889 tun_debug(KERN_INFO, tun, "owner set to %u\n",
Eric W. Biederman0625c882012-02-07 16:48:55 -08002890 from_kuid(&init_user_ns, tun->owner));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002891 break;
2892
Guido Guenther8c644622007-07-02 22:50:25 -07002893 case TUNSETGROUP:
2894 /* Set group of the device */
Eric W. Biederman0625c882012-02-07 16:48:55 -08002895 group = make_kgid(current_user_ns(), arg);
2896 if (!gid_valid(group)) {
2897 ret = -EINVAL;
2898 break;
2899 }
2900 tun->group = group;
Jason Wang1e588332012-10-31 19:45:56 +00002901 tun_debug(KERN_INFO, tun, "group set to %u\n",
Eric W. Biederman0625c882012-02-07 16:48:55 -08002902 from_kgid(&init_user_ns, tun->group));
Guido Guenther8c644622007-07-02 22:50:25 -07002903 break;
2904
Mike Kershawff4cc3a2005-09-01 17:40:05 -07002905 case TUNSETLINK:
2906 /* Only allow setting the type when the interface is down */
2907 if (tun->dev->flags & IFF_UP) {
Joe Perches6b8a66e2011-03-02 07:18:10 +00002908 tun_debug(KERN_INFO, tun,
2909 "Linktype set failed because interface is up\n");
David S. Miller48abfe02008-04-23 19:37:58 -07002910 ret = -EBUSY;
Mike Kershawff4cc3a2005-09-01 17:40:05 -07002911 } else {
2912 tun->dev->type = (int) arg;
Joe Perches6b8a66e2011-03-02 07:18:10 +00002913 tun_debug(KERN_INFO, tun, "linktype set to %d\n",
2914 tun->dev->type);
David S. Miller48abfe02008-04-23 19:37:58 -07002915 ret = 0;
Mike Kershawff4cc3a2005-09-01 17:40:05 -07002916 }
Eric W. Biederman631ab462009-01-20 11:00:40 +00002917 break;
Mike Kershawff4cc3a2005-09-01 17:40:05 -07002918
Linus Torvalds1da177e2005-04-16 15:20:36 -07002919#ifdef TUN_DEBUG
2920 case TUNSETDEBUG:
2921 tun->debug = arg;
2922 break;
2923#endif
Rusty Russell5228ddc2008-07-03 03:46:16 -07002924 case TUNSETOFFLOAD:
Michał Mirosław88255372011-04-19 06:13:10 +00002925 ret = set_offload(tun, arg);
Eric W. Biederman631ab462009-01-20 11:00:40 +00002926 break;
Rusty Russell5228ddc2008-07-03 03:46:16 -07002927
Max Krasnyanskyf271b2cc2008-07-14 22:18:19 -07002928 case TUNSETTXFILTER:
2929 /* Can be set only for TAPs */
Eric W. Biederman631ab462009-01-20 11:00:40 +00002930 ret = -EINVAL;
Michael S. Tsirkin40630b82014-11-19 15:17:31 +02002931 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
Eric W. Biederman631ab462009-01-20 11:00:40 +00002932 break;
Harvey Harrisonc0e5a8c2008-07-16 12:45:34 -07002933 ret = update_filter(&tun->txflt, (void __user *)arg);
Eric W. Biederman631ab462009-01-20 11:00:40 +00002934 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002935
2936 case SIOCGIFHWADDR:
Uwe Kleine-Königb5950762010-11-01 15:38:34 -04002937 /* Get hw address */
Max Krasnyanskyf271b2cc2008-07-14 22:18:19 -07002938 memcpy(ifr.ifr_hwaddr.sa_data, tun->dev->dev_addr, ETH_ALEN);
2939 ifr.ifr_hwaddr.sa_family = tun->dev->type;
Arnd Bergmann50857e22009-11-06 22:52:32 -08002940 if (copy_to_user(argp, &ifr, ifreq_len))
Eric W. Biederman631ab462009-01-20 11:00:40 +00002941 ret = -EFAULT;
2942 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002943
2944 case SIOCSIFHWADDR:
Max Krasnyanskyf271b2cc2008-07-14 22:18:19 -07002945 /* Set hw address */
Joe Perches6b8a66e2011-03-02 07:18:10 +00002946 tun_debug(KERN_DEBUG, tun, "set hw address: %pM\n",
2947 ifr.ifr_hwaddr.sa_data);
Kim B. Heino40102372008-02-29 12:26:21 -08002948
Kim B. Heino40102372008-02-29 12:26:21 -08002949 ret = dev_set_mac_address(tun->dev, &ifr.ifr_hwaddr);
Eric W. Biederman631ab462009-01-20 11:00:40 +00002950 break;
Herbert Xu33dccbb2009-02-05 21:25:32 -08002951
2952 case TUNGETSNDBUF:
Jason Wang54f968d2012-10-31 19:45:57 +00002953 sndbuf = tfile->socket.sk->sk_sndbuf;
Herbert Xu33dccbb2009-02-05 21:25:32 -08002954 if (copy_to_user(argp, &sndbuf, sizeof(sndbuf)))
2955 ret = -EFAULT;
2956 break;
2957
2958 case TUNSETSNDBUF:
2959 if (copy_from_user(&sndbuf, argp, sizeof(sndbuf))) {
2960 ret = -EFAULT;
2961 break;
2962 }
Craig Gallek931619222017-10-30 18:50:11 -04002963 if (sndbuf <= 0) {
2964 ret = -EINVAL;
2965 break;
2966 }
Herbert Xu33dccbb2009-02-05 21:25:32 -08002967
Jason Wangc8d68e62012-10-31 19:46:00 +00002968 tun->sndbuf = sndbuf;
2969 tun_set_sndbuf(tun);
Herbert Xu33dccbb2009-02-05 21:25:32 -08002970 break;
2971
Michael S. Tsirkind9d52b52010-03-17 17:45:01 +02002972 case TUNGETVNETHDRSZ:
2973 vnet_hdr_sz = tun->vnet_hdr_sz;
2974 if (copy_to_user(argp, &vnet_hdr_sz, sizeof(vnet_hdr_sz)))
2975 ret = -EFAULT;
2976 break;
2977
2978 case TUNSETVNETHDRSZ:
2979 if (copy_from_user(&vnet_hdr_sz, argp, sizeof(vnet_hdr_sz))) {
2980 ret = -EFAULT;
2981 break;
2982 }
2983 if (vnet_hdr_sz < (int)sizeof(struct virtio_net_hdr)) {
2984 ret = -EINVAL;
2985 break;
2986 }
2987
2988 tun->vnet_hdr_sz = vnet_hdr_sz;
2989 break;
2990
Michael S. Tsirkin1cf8e412014-12-16 15:05:06 +02002991 case TUNGETVNETLE:
2992 le = !!(tun->flags & TUN_VNET_LE);
2993 if (put_user(le, (int __user *)argp))
2994 ret = -EFAULT;
2995 break;
2996
2997 case TUNSETVNETLE:
2998 if (get_user(le, (int __user *)argp)) {
2999 ret = -EFAULT;
3000 break;
3001 }
3002 if (le)
3003 tun->flags |= TUN_VNET_LE;
3004 else
3005 tun->flags &= ~TUN_VNET_LE;
3006 break;
3007
Greg Kurz8b8e6582015-04-24 14:50:36 +02003008 case TUNGETVNETBE:
3009 ret = tun_get_vnet_be(tun, argp);
3010 break;
3011
3012 case TUNSETVNETBE:
3013 ret = tun_set_vnet_be(tun, argp);
3014 break;
3015
Michael S. Tsirkin99405162010-02-14 01:01:10 +00003016 case TUNATTACHFILTER:
3017 /* Can be set only for TAPs */
3018 ret = -EINVAL;
Michael S. Tsirkin40630b82014-11-19 15:17:31 +02003019 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
Michael S. Tsirkin99405162010-02-14 01:01:10 +00003020 break;
3021 ret = -EFAULT;
Jason Wang54f968d2012-10-31 19:45:57 +00003022 if (copy_from_user(&tun->fprog, argp, sizeof(tun->fprog)))
Michael S. Tsirkin99405162010-02-14 01:01:10 +00003023 break;
3024
Jason Wangc8d68e62012-10-31 19:46:00 +00003025 ret = tun_attach_filter(tun);
Michael S. Tsirkin99405162010-02-14 01:01:10 +00003026 break;
3027
3028 case TUNDETACHFILTER:
3029 /* Can be set only for TAPs */
3030 ret = -EINVAL;
Michael S. Tsirkin40630b82014-11-19 15:17:31 +02003031 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
Michael S. Tsirkin99405162010-02-14 01:01:10 +00003032 break;
Jason Wangc8d68e62012-10-31 19:46:00 +00003033 ret = 0;
3034 tun_detach_filter(tun, tun->numqueues);
Michael S. Tsirkin99405162010-02-14 01:01:10 +00003035 break;
3036
Pavel Emelyanov76975e92013-08-21 14:32:39 +04003037 case TUNGETFILTER:
3038 ret = -EINVAL;
Michael S. Tsirkin40630b82014-11-19 15:17:31 +02003039 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
Pavel Emelyanov76975e92013-08-21 14:32:39 +04003040 break;
3041 ret = -EFAULT;
3042 if (copy_to_user(argp, &tun->fprog, sizeof(tun->fprog)))
3043 break;
3044 ret = 0;
3045 break;
3046
Jason Wang96f84062017-12-04 17:31:23 +08003047 case TUNSETSTEERINGEBPF:
Jason Wangcd5681d2018-01-16 16:31:01 +08003048 ret = tun_set_ebpf(tun, &tun->steering_prog, argp);
Jason Wang96f84062017-12-04 17:31:23 +08003049 break;
3050
Jason Wangaff3d702018-01-16 16:31:02 +08003051 case TUNSETFILTEREBPF:
3052 ret = tun_set_ebpf(tun, &tun->filter_prog, argp);
3053 break;
3054
Linus Torvalds1da177e2005-04-16 15:20:36 -07003055 default:
Eric W. Biederman631ab462009-01-20 11:00:40 +00003056 ret = -EINVAL;
3057 break;
Joe Perchesee289b62010-05-17 22:47:34 -07003058 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003059
Herbert Xu876bfd42009-08-06 14:22:44 +00003060unlock:
3061 rtnl_unlock();
3062 if (tun)
3063 tun_put(tun);
Eric W. Biederman631ab462009-01-20 11:00:40 +00003064 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003065}
3066
Arnd Bergmann50857e22009-11-06 22:52:32 -08003067static long tun_chr_ioctl(struct file *file,
3068 unsigned int cmd, unsigned long arg)
3069{
3070 return __tun_chr_ioctl(file, cmd, arg, sizeof (struct ifreq));
3071}
3072
3073#ifdef CONFIG_COMPAT
3074static long tun_chr_compat_ioctl(struct file *file,
3075 unsigned int cmd, unsigned long arg)
3076{
3077 switch (cmd) {
3078 case TUNSETIFF:
3079 case TUNGETIFF:
3080 case TUNSETTXFILTER:
3081 case TUNGETSNDBUF:
3082 case TUNSETSNDBUF:
3083 case SIOCGIFHWADDR:
3084 case SIOCSIFHWADDR:
3085 arg = (unsigned long)compat_ptr(arg);
3086 break;
3087 default:
3088 arg = (compat_ulong_t)arg;
3089 break;
3090 }
3091
3092 /*
3093 * compat_ifreq is shorter than ifreq, so we must not access beyond
3094 * the end of that structure. All fields that are used in this
3095 * driver are compatible though, we don't need to convert the
3096 * contents.
3097 */
3098 return __tun_chr_ioctl(file, cmd, arg, sizeof(struct compat_ifreq));
3099}
3100#endif /* CONFIG_COMPAT */
3101
Linus Torvalds1da177e2005-04-16 15:20:36 -07003102static int tun_chr_fasync(int fd, struct file *file, int on)
3103{
Jason Wang54f968d2012-10-31 19:45:57 +00003104 struct tun_file *tfile = file->private_data;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003105 int ret;
3106
Jason Wang54f968d2012-10-31 19:45:57 +00003107 if ((ret = fasync_helper(fd, file, on, &tfile->fasync)) < 0)
Jonathan Corbet9d319522008-06-19 15:50:37 -06003108 goto out;
Jeff Garzik6aa20a22006-09-13 13:24:59 -04003109
Linus Torvalds1da177e2005-04-16 15:20:36 -07003110 if (on) {
Jeff Laytone0b93ed2014-08-22 11:27:32 -04003111 __f_setown(file, task_pid(current), PIDTYPE_PID, 0);
Jason Wang54f968d2012-10-31 19:45:57 +00003112 tfile->flags |= TUN_FASYNC;
Jeff Garzik6aa20a22006-09-13 13:24:59 -04003113 } else
Jason Wang54f968d2012-10-31 19:45:57 +00003114 tfile->flags &= ~TUN_FASYNC;
Jonathan Corbet9d319522008-06-19 15:50:37 -06003115 ret = 0;
3116out:
Jonathan Corbet9d319522008-06-19 15:50:37 -06003117 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003118}
3119
3120static int tun_chr_open(struct inode *inode, struct file * file)
3121{
Eric W. Biederman140e807da2015-05-08 21:07:08 -05003122 struct net *net = current->nsproxy->net_ns;
Eric W. Biederman631ab462009-01-20 11:00:40 +00003123 struct tun_file *tfile;
Thomas Gleixnerdeed49f2009-10-14 01:19:46 -07003124
Joe Perches6b8a66e2011-03-02 07:18:10 +00003125 DBG1(KERN_INFO, "tunX: tun_chr_open\n");
Eric W. Biederman631ab462009-01-20 11:00:40 +00003126
Eric W. Biederman140e807da2015-05-08 21:07:08 -05003127 tfile = (struct tun_file *)sk_alloc(net, AF_UNSPEC, GFP_KERNEL,
Eric W. Biederman11aa9c22015-05-08 21:09:13 -05003128 &tun_proto, 0);
Eric W. Biederman631ab462009-01-20 11:00:40 +00003129 if (!tfile)
3130 return -ENOMEM;
Monam Agarwalc9566742014-03-24 00:02:32 +05303131 RCU_INIT_POINTER(tfile->tun, NULL);
Jason Wang54f968d2012-10-31 19:45:57 +00003132 tfile->flags = 0;
Pavel Emelyanovfb7589a2013-08-21 14:31:38 +04003133 tfile->ifindex = 0;
Jason Wang54f968d2012-10-31 19:45:57 +00003134
Jason Wang54f968d2012-10-31 19:45:57 +00003135 init_waitqueue_head(&tfile->wq.wait);
Xi Wang9e641bd2014-05-16 15:11:48 -07003136 RCU_INIT_POINTER(tfile->socket.wq, &tfile->wq);
Jason Wang54f968d2012-10-31 19:45:57 +00003137
3138 tfile->socket.file = file;
3139 tfile->socket.ops = &tun_socket_ops;
3140
3141 sock_init_data(&tfile->socket, &tfile->sk);
Jason Wang54f968d2012-10-31 19:45:57 +00003142
3143 tfile->sk.sk_write_space = tun_sock_write_space;
3144 tfile->sk.sk_sndbuf = INT_MAX;
3145
Eric W. Biederman631ab462009-01-20 11:00:40 +00003146 file->private_data = tfile;
Jason Wang4008e972012-12-13 23:53:30 +00003147 INIT_LIST_HEAD(&tfile->next);
Jason Wang54f968d2012-10-31 19:45:57 +00003148
Jason Wang19a6afb2013-06-08 14:17:41 +08003149 sock_set_flag(&tfile->sk, SOCK_ZEROCOPY);
3150
David S. Miller8565d262018-01-19 22:59:33 -05003151 memset(&tfile->tx_ring, 0, sizeof(tfile->tx_ring));
Cong Wang4df0bfc2018-01-15 11:37:29 -08003152
Linus Torvalds1da177e2005-04-16 15:20:36 -07003153 return 0;
3154}
3155
3156static int tun_chr_close(struct inode *inode, struct file *file)
3157{
Eric W. Biederman631ab462009-01-20 11:00:40 +00003158 struct tun_file *tfile = file->private_data;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003159
Jason Wangc8d68e62012-10-31 19:46:00 +00003160 tun_detach(tfile, true);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003161
3162 return 0;
3163}
3164
Masatake YAMATO93e14b62014-01-29 16:43:31 +09003165#ifdef CONFIG_PROC_FS
yuan linyu9484dc72017-09-23 22:36:52 +08003166static void tun_chr_show_fdinfo(struct seq_file *m, struct file *file)
Masatake YAMATO93e14b62014-01-29 16:43:31 +09003167{
yuan linyu9484dc72017-09-23 22:36:52 +08003168 struct tun_file *tfile = file->private_data;
Masatake YAMATO93e14b62014-01-29 16:43:31 +09003169 struct tun_struct *tun;
3170 struct ifreq ifr;
3171
3172 memset(&ifr, 0, sizeof(ifr));
3173
3174 rtnl_lock();
yuan linyu9484dc72017-09-23 22:36:52 +08003175 tun = tun_get(tfile);
Masatake YAMATO93e14b62014-01-29 16:43:31 +09003176 if (tun)
3177 tun_get_iff(current->nsproxy->net_ns, tun, &ifr);
3178 rtnl_unlock();
3179
3180 if (tun)
3181 tun_put(tun);
3182
Joe Perchesa3816ab2014-09-29 16:08:25 -07003183 seq_printf(m, "iff:\t%s\n", ifr.ifr_name);
Masatake YAMATO93e14b62014-01-29 16:43:31 +09003184}
3185#endif
3186
Arjan van de Vend54b1fd2007-02-12 00:55:34 -08003187static const struct file_operations tun_fops = {
Jeff Garzik6aa20a22006-09-13 13:24:59 -04003188 .owner = THIS_MODULE,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003189 .llseek = no_llseek,
Al Viro9b067032014-11-07 13:52:07 -05003190 .read_iter = tun_chr_read_iter,
Al Virof5ff53b2014-06-19 15:36:49 -04003191 .write_iter = tun_chr_write_iter,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003192 .poll = tun_chr_poll,
Arnd Bergmann50857e22009-11-06 22:52:32 -08003193 .unlocked_ioctl = tun_chr_ioctl,
3194#ifdef CONFIG_COMPAT
3195 .compat_ioctl = tun_chr_compat_ioctl,
3196#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07003197 .open = tun_chr_open,
3198 .release = tun_chr_close,
Masatake YAMATO93e14b62014-01-29 16:43:31 +09003199 .fasync = tun_chr_fasync,
3200#ifdef CONFIG_PROC_FS
3201 .show_fdinfo = tun_chr_show_fdinfo,
3202#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07003203};
3204
3205static struct miscdevice tun_miscdev = {
3206 .minor = TUN_MINOR,
3207 .name = "tun",
Kay Sieverse454cea2009-09-18 23:01:12 +02003208 .nodename = "net/tun",
Linus Torvalds1da177e2005-04-16 15:20:36 -07003209 .fops = &tun_fops,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003210};
3211
3212/* ethtool interface */
3213
Philippe Reynes29ccc492017-03-11 22:03:50 +01003214static int tun_get_link_ksettings(struct net_device *dev,
3215 struct ethtool_link_ksettings *cmd)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003216{
Philippe Reynes29ccc492017-03-11 22:03:50 +01003217 ethtool_link_ksettings_zero_link_mode(cmd, supported);
3218 ethtool_link_ksettings_zero_link_mode(cmd, advertising);
3219 cmd->base.speed = SPEED_10;
3220 cmd->base.duplex = DUPLEX_FULL;
3221 cmd->base.port = PORT_TP;
3222 cmd->base.phy_address = 0;
3223 cmd->base.autoneg = AUTONEG_DISABLE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003224 return 0;
3225}
3226
3227static void tun_get_drvinfo(struct net_device *dev, struct ethtool_drvinfo *info)
3228{
3229 struct tun_struct *tun = netdev_priv(dev);
3230
Rick Jones33a5ba12011-11-15 14:59:53 +00003231 strlcpy(info->driver, DRV_NAME, sizeof(info->driver));
3232 strlcpy(info->version, DRV_VERSION, sizeof(info->version));
Linus Torvalds1da177e2005-04-16 15:20:36 -07003233
3234 switch (tun->flags & TUN_TYPE_MASK) {
Michael S. Tsirkin40630b82014-11-19 15:17:31 +02003235 case IFF_TUN:
Rick Jones33a5ba12011-11-15 14:59:53 +00003236 strlcpy(info->bus_info, "tun", sizeof(info->bus_info));
Linus Torvalds1da177e2005-04-16 15:20:36 -07003237 break;
Michael S. Tsirkin40630b82014-11-19 15:17:31 +02003238 case IFF_TAP:
Rick Jones33a5ba12011-11-15 14:59:53 +00003239 strlcpy(info->bus_info, "tap", sizeof(info->bus_info));
Linus Torvalds1da177e2005-04-16 15:20:36 -07003240 break;
3241 }
3242}
3243
3244static u32 tun_get_msglevel(struct net_device *dev)
3245{
3246#ifdef TUN_DEBUG
3247 struct tun_struct *tun = netdev_priv(dev);
3248 return tun->debug;
3249#else
3250 return -EOPNOTSUPP;
3251#endif
3252}
3253
3254static void tun_set_msglevel(struct net_device *dev, u32 value)
3255{
3256#ifdef TUN_DEBUG
3257 struct tun_struct *tun = netdev_priv(dev);
3258 tun->debug = value;
3259#endif
3260}
3261
Jason Wang5503fce2017-01-18 15:02:03 +08003262static int tun_get_coalesce(struct net_device *dev,
3263 struct ethtool_coalesce *ec)
3264{
3265 struct tun_struct *tun = netdev_priv(dev);
3266
3267 ec->rx_max_coalesced_frames = tun->rx_batched;
3268
3269 return 0;
3270}
3271
3272static int tun_set_coalesce(struct net_device *dev,
3273 struct ethtool_coalesce *ec)
3274{
3275 struct tun_struct *tun = netdev_priv(dev);
3276
3277 if (ec->rx_max_coalesced_frames > NAPI_POLL_WEIGHT)
3278 tun->rx_batched = NAPI_POLL_WEIGHT;
3279 else
3280 tun->rx_batched = ec->rx_max_coalesced_frames;
3281
3282 return 0;
3283}
3284
Jeff Garzik7282d492006-09-13 14:30:00 -04003285static const struct ethtool_ops tun_ethtool_ops = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003286 .get_drvinfo = tun_get_drvinfo,
3287 .get_msglevel = tun_get_msglevel,
3288 .set_msglevel = tun_set_msglevel,
Nolan Leakebee31362010-07-27 13:53:43 +00003289 .get_link = ethtool_op_get_link,
Richard Cochraneda29772013-07-19 19:40:10 +02003290 .get_ts_info = ethtool_op_get_ts_info,
Jason Wang5503fce2017-01-18 15:02:03 +08003291 .get_coalesce = tun_get_coalesce,
3292 .set_coalesce = tun_set_coalesce,
Philippe Reynes29ccc492017-03-11 22:03:50 +01003293 .get_link_ksettings = tun_get_link_ksettings,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003294};
3295
Jason Wang1576d982016-06-30 14:45:36 +08003296static int tun_queue_resize(struct tun_struct *tun)
3297{
3298 struct net_device *dev = tun->dev;
3299 struct tun_file *tfile;
Jason Wang5990a302018-01-04 11:14:27 +08003300 struct ptr_ring **rings;
Jason Wang1576d982016-06-30 14:45:36 +08003301 int n = tun->numqueues + tun->numdisabled;
3302 int ret, i;
3303
Jason Wang5990a302018-01-04 11:14:27 +08003304 rings = kmalloc_array(n, sizeof(*rings), GFP_KERNEL);
3305 if (!rings)
Jason Wang1576d982016-06-30 14:45:36 +08003306 return -ENOMEM;
3307
3308 for (i = 0; i < tun->numqueues; i++) {
3309 tfile = rtnl_dereference(tun->tfiles[i]);
Jason Wang5990a302018-01-04 11:14:27 +08003310 rings[i] = &tfile->tx_ring;
Jason Wang1576d982016-06-30 14:45:36 +08003311 }
3312 list_for_each_entry(tfile, &tun->disabled, next)
Jason Wang5990a302018-01-04 11:14:27 +08003313 rings[i++] = &tfile->tx_ring;
Jason Wang1576d982016-06-30 14:45:36 +08003314
Jason Wang5990a302018-01-04 11:14:27 +08003315 ret = ptr_ring_resize_multiple(rings, n,
3316 dev->tx_queue_len, GFP_KERNEL,
Jason Wangfc72d1d2018-01-04 11:14:28 +08003317 tun_ptr_free);
Jason Wang1576d982016-06-30 14:45:36 +08003318
Jason Wang5990a302018-01-04 11:14:27 +08003319 kfree(rings);
Jason Wang1576d982016-06-30 14:45:36 +08003320 return ret;
3321}
3322
3323static int tun_device_event(struct notifier_block *unused,
3324 unsigned long event, void *ptr)
3325{
3326 struct net_device *dev = netdev_notifier_info_to_dev(ptr);
3327 struct tun_struct *tun = netdev_priv(dev);
3328
Craig Gallek86dfb4ac2016-07-06 18:44:20 -04003329 if (dev->rtnl_link_ops != &tun_link_ops)
3330 return NOTIFY_DONE;
3331
Jason Wang1576d982016-06-30 14:45:36 +08003332 switch (event) {
3333 case NETDEV_CHANGE_TX_QUEUE_LEN:
3334 if (tun_queue_resize(tun))
3335 return NOTIFY_BAD;
3336 break;
3337 default:
3338 break;
3339 }
3340
3341 return NOTIFY_DONE;
3342}
3343
3344static struct notifier_block tun_notifier_block __read_mostly = {
3345 .notifier_call = tun_device_event,
3346};
Pavel Emelyanov79d17602008-04-16 00:40:46 -07003347
Linus Torvalds1da177e2005-04-16 15:20:36 -07003348static int __init tun_init(void)
3349{
3350 int ret = 0;
3351
Joe Perches6b8a66e2011-03-02 07:18:10 +00003352 pr_info("%s, %s\n", DRV_DESCRIPTION, DRV_VERSION);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003353
Eric W. Biedermanf019a7a2009-01-21 16:02:16 -08003354 ret = rtnl_link_register(&tun_link_ops);
Pavel Emelyanov79d17602008-04-16 00:40:46 -07003355 if (ret) {
Joe Perches6b8a66e2011-03-02 07:18:10 +00003356 pr_err("Can't register link_ops\n");
Eric W. Biedermanf019a7a2009-01-21 16:02:16 -08003357 goto err_linkops;
Pavel Emelyanov79d17602008-04-16 00:40:46 -07003358 }
3359
Linus Torvalds1da177e2005-04-16 15:20:36 -07003360 ret = misc_register(&tun_miscdev);
Pavel Emelyanov79d17602008-04-16 00:40:46 -07003361 if (ret) {
Joe Perches6b8a66e2011-03-02 07:18:10 +00003362 pr_err("Can't register misc device %d\n", TUN_MINOR);
Pavel Emelyanov79d17602008-04-16 00:40:46 -07003363 goto err_misc;
3364 }
Jason Wang1576d982016-06-30 14:45:36 +08003365
Tonghao Zhang5edfbd32017-07-20 02:41:34 -07003366 ret = register_netdevice_notifier(&tun_notifier_block);
3367 if (ret) {
3368 pr_err("Can't register netdevice notifier\n");
3369 goto err_notifier;
3370 }
3371
Eric W. Biedermanf019a7a2009-01-21 16:02:16 -08003372 return 0;
Tonghao Zhang5edfbd32017-07-20 02:41:34 -07003373
3374err_notifier:
3375 misc_deregister(&tun_miscdev);
Pavel Emelyanov79d17602008-04-16 00:40:46 -07003376err_misc:
Eric W. Biedermanf019a7a2009-01-21 16:02:16 -08003377 rtnl_link_unregister(&tun_link_ops);
3378err_linkops:
Linus Torvalds1da177e2005-04-16 15:20:36 -07003379 return ret;
3380}
3381
3382static void tun_cleanup(void)
3383{
Jeff Garzik6aa20a22006-09-13 13:24:59 -04003384 misc_deregister(&tun_miscdev);
Eric W. Biedermanf019a7a2009-01-21 16:02:16 -08003385 rtnl_link_unregister(&tun_link_ops);
Jason Wang1576d982016-06-30 14:45:36 +08003386 unregister_netdevice_notifier(&tun_notifier_block);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003387}
3388
Michael S. Tsirkin05c28282010-01-14 06:17:09 +00003389/* Get an underlying socket object from tun file. Returns error unless file is
3390 * attached to a device. The returned object works like a packet socket, it
3391 * can be used for sock_sendmsg/sock_recvmsg. The caller is responsible for
3392 * holding a reference to the file for as long as the socket is in use. */
3393struct socket *tun_get_socket(struct file *file)
3394{
Jason Wang6e914fc2012-10-31 19:45:58 +00003395 struct tun_file *tfile;
Michael S. Tsirkin05c28282010-01-14 06:17:09 +00003396 if (file->f_op != &tun_fops)
3397 return ERR_PTR(-EINVAL);
Jason Wang6e914fc2012-10-31 19:45:58 +00003398 tfile = file->private_data;
3399 if (!tfile)
Michael S. Tsirkin05c28282010-01-14 06:17:09 +00003400 return ERR_PTR(-EBADFD);
Jason Wang54f968d2012-10-31 19:45:57 +00003401 return &tfile->socket;
Michael S. Tsirkin05c28282010-01-14 06:17:09 +00003402}
3403EXPORT_SYMBOL_GPL(tun_get_socket);
3404
Jason Wang5990a302018-01-04 11:14:27 +08003405struct ptr_ring *tun_get_tx_ring(struct file *file)
Jason Wang83339c62017-05-17 12:14:41 +08003406{
3407 struct tun_file *tfile;
3408
3409 if (file->f_op != &tun_fops)
3410 return ERR_PTR(-EINVAL);
3411 tfile = file->private_data;
3412 if (!tfile)
3413 return ERR_PTR(-EBADFD);
Jason Wang5990a302018-01-04 11:14:27 +08003414 return &tfile->tx_ring;
Jason Wang83339c62017-05-17 12:14:41 +08003415}
Jason Wang5990a302018-01-04 11:14:27 +08003416EXPORT_SYMBOL_GPL(tun_get_tx_ring);
Jason Wang83339c62017-05-17 12:14:41 +08003417
Linus Torvalds1da177e2005-04-16 15:20:36 -07003418module_init(tun_init);
3419module_exit(tun_cleanup);
3420MODULE_DESCRIPTION(DRV_DESCRIPTION);
3421MODULE_AUTHOR(DRV_COPYRIGHT);
3422MODULE_LICENSE("GPL");
3423MODULE_ALIAS_MISCDEV(TUN_MINOR);
Kay Sievers578454f2010-05-20 18:07:20 +02003424MODULE_ALIAS("devname:net/tun");