blob: 2ced021f4faf60d31c2c327f0544cb2eed9d21cf [file] [log] [blame]
Thomas Gleixnerc942fdd2019-05-27 08:55:06 +02001// SPDX-License-Identifier: GPL-2.0-or-later
Linus Torvalds1da177e2005-04-16 15:20:36 -07002/*
3 * TUN - Universal TUN/TAP device driver.
4 * Copyright (C) 1999-2002 Maxim Krasnyansky <maxk@qualcomm.com>
5 *
Linus Torvalds1da177e2005-04-16 15:20:36 -07006 * $Id: tun.c,v 1.15 2002/03/01 02:44:24 maxk Exp $
7 */
8
9/*
10 * Changes:
11 *
Mike Kershawff4cc3a2005-09-01 17:40:05 -070012 * Mike Kershaw <dragorn@kismetwireless.net> 2005/08/14
13 * Add TUNSETLINK ioctl to set the link encapsulation
14 *
Linus Torvalds1da177e2005-04-16 15:20:36 -070015 * Mark Smith <markzzzsmith@yahoo.com.au>
Joe Perches344dc8e2012-07-12 19:33:09 +000016 * Use eth_random_addr() for tap MAC address.
Linus Torvalds1da177e2005-04-16 15:20:36 -070017 *
18 * Harald Roelle <harald.roelle@ifi.lmu.de> 2004/04/20
19 * Fixes in packet dropping, queue length setting and queue wakeup.
20 * Increased default tx queue length.
21 * Added ethtool API.
22 * Minor cleanups
23 *
24 * Daniel Podlejski <underley@underley.eu.org>
25 * Modifications for 2.3.99-pre5 kernel.
26 */
27
Joe Perches6b8a66e2011-03-02 07:18:10 +000028#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
29
Linus Torvalds1da177e2005-04-16 15:20:36 -070030#define DRV_NAME "tun"
31#define DRV_VERSION "1.6"
32#define DRV_DESCRIPTION "Universal TUN/TAP device driver"
33#define DRV_COPYRIGHT "(C) 1999-2004 Max Krasnyansky <maxk@qualcomm.com>"
34
Linus Torvalds1da177e2005-04-16 15:20:36 -070035#include <linux/module.h>
36#include <linux/errno.h>
37#include <linux/kernel.h>
Ingo Molnar174cd4b2017-02-02 19:15:33 +010038#include <linux/sched/signal.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070039#include <linux/major.h>
40#include <linux/slab.h>
41#include <linux/poll.h>
42#include <linux/fcntl.h>
43#include <linux/init.h>
44#include <linux/skbuff.h>
45#include <linux/netdevice.h>
46#include <linux/etherdevice.h>
47#include <linux/miscdevice.h>
48#include <linux/ethtool.h>
49#include <linux/rtnetlink.h>
Arnd Bergmann50857e22009-11-06 22:52:32 -080050#include <linux/compat.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070051#include <linux/if.h>
52#include <linux/if_arp.h>
53#include <linux/if_ether.h>
54#include <linux/if_tun.h>
Jason Wang6680ec62013-07-25 13:00:33 +080055#include <linux/if_vlan.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070056#include <linux/crc32.h>
Pavel Emelyanovd647a592008-04-16 00:41:16 -070057#include <linux/nsproxy.h>
Rusty Russellf43798c2008-07-03 03:48:02 -070058#include <linux/virtio_net.h>
Michael S. Tsirkin99405162010-02-14 01:01:10 +000059#include <linux/rcupdate.h>
Eric W. Biederman881d9662007-09-17 11:56:21 -070060#include <net/net_namespace.h>
Pavel Emelyanov79d17602008-04-16 00:40:46 -070061#include <net/netns/generic.h>
Eric W. Biedermanf019a7a2009-01-21 16:02:16 -080062#include <net/rtnetlink.h>
Herbert Xu33dccbb2009-02-05 21:25:32 -080063#include <net/sock.h>
Jesper Dangaard Brouer735fc402018-05-24 16:46:12 +020064#include <net/xdp.h>
Jason A. Donenfeldb9815eb2020-06-29 19:06:22 -060065#include <net/ip_tunnels.h>
Masatake YAMATO93e14b62014-01-29 16:43:31 +090066#include <linux/seq_file.h>
Herbert Xue0b46d02014-11-07 21:22:23 +080067#include <linux/uio.h>
Jason Wang1576d982016-06-30 14:45:36 +080068#include <linux/skb_array.h>
Jason Wang761876c2017-08-11 19:41:18 +080069#include <linux/bpf.h>
70#include <linux/bpf_trace.h>
Petar Penkov90e33d42017-09-22 13:49:15 -070071#include <linux/mutex.h>
Phillip Pottercca8ea32021-04-06 18:45:54 +010072#include <linux/ieee802154.h>
73#include <linux/if_ltalk.h>
74#include <uapi/linux/if_fddi.h>
75#include <uapi/linux/if_hippi.h>
76#include <uapi/linux/if_fc.h>
77#include <net/ax25.h>
78#include <net/rose.h>
79#include <net/6lowpan.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070080
Linus Torvalds7c0f6ba2016-12-24 11:46:01 -080081#include <linux/uaccess.h>
Kirill Tkhaif2780d62018-02-14 16:40:14 +030082#include <linux/proc_fs.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070083
Chas Williams4e24f2d2018-06-02 17:49:53 -040084static void tun_default_link_ksettings(struct net_device *dev,
85 struct ethtool_link_ksettings *cmd);
86
Jason Wang7df13212017-09-04 11:36:08 +080087#define TUN_RX_PAD (NET_IP_ALIGN + NET_SKB_PAD)
Jason Wang66ccbc92017-08-11 19:41:16 +080088
Michael S. Tsirkin031f5e032014-11-19 14:44:40 +020089/* TUN device flags */
90
91/* IFF_ATTACH_QUEUE is never stored in device flags,
92 * overload it to mean fasync when stored there.
93 */
94#define TUN_FASYNC IFF_ATTACH_QUEUE
Michael S. Tsirkin1cf8e412014-12-16 15:05:06 +020095/* High bits in flags field are unused. */
96#define TUN_VNET_LE 0x80000000
Greg Kurz8b8e6582015-04-24 14:50:36 +020097#define TUN_VNET_BE 0x40000000
Michael S. Tsirkin031f5e032014-11-19 14:44:40 +020098
99#define TUN_FEATURES (IFF_NO_PI | IFF_ONE_QUEUE | IFF_VNET_HDR | \
Petar Penkov90e33d42017-09-22 13:49:15 -0700100 IFF_MULTI_QUEUE | IFF_NAPI | IFF_NAPI_FRAGS)
101
Michael S. Tsirkin06908992012-07-20 09:23:23 +0000102#define GOODCOPY_LEN 128
103
Max Krasnyanskyf271b2cc2008-07-14 22:18:19 -0700104#define FLT_EXACT_COUNT 8
105struct tap_filter {
106 unsigned int count; /* Number of addrs. Zero means disabled */
107 u32 mask[2]; /* Mask of the hashed addrs */
108 unsigned char addr[FLT_EXACT_COUNT][ETH_ALEN];
109};
110
Pankaj Guptabaf71c52015-01-12 11:41:29 +0530111/* MAX_TAP_QUEUES 256 is chosen to allow rx/tx queues to be equal
112 * to max number of VCPUs in guest. */
113#define MAX_TAP_QUEUES 256
Jason Wangb8732fb2013-01-23 03:59:13 +0000114#define MAX_TAP_FLOWS 4096
Jason Wangc8d68e62012-10-31 19:46:00 +0000115
Jason Wang96442e422012-10-31 19:46:02 +0000116#define TUN_FLOW_EXPIRE (3 * HZ)
117
Jason Wang54f968d2012-10-31 19:45:57 +0000118/* A tun_file connects an open character device to a tuntap netdevice. It
stephen hemminger92d4ea62013-12-05 20:42:58 -0800119 * also contains all socket related structures (except sock_fprog and tap_filter)
Jason Wang54f968d2012-10-31 19:45:57 +0000120 * to serve as one transmit queue for tuntap device. The sock_fprog and
121 * tap_filter were kept in tun_struct since they were used for filtering for the
Rami Rosen36fe8c092012-11-25 22:07:40 +0000122 * netdevice not for a specific queue (at least I didn't see the requirement for
Jason Wang54f968d2012-10-31 19:45:57 +0000123 * this).
Jason Wang6e914fc2012-10-31 19:45:58 +0000124 *
125 * RCU usage:
Rami Rosen36fe8c092012-11-25 22:07:40 +0000126 * The tun_file and tun_struct are loosely coupled, the pointer from one to the
Jason Wang6e914fc2012-10-31 19:45:58 +0000127 * other can only be read while rcu_read_lock or rtnl_lock is held.
Jason Wang54f968d2012-10-31 19:45:57 +0000128 */
Eric W. Biederman631ab462009-01-20 11:00:40 +0000129struct tun_file {
Jason Wang54f968d2012-10-31 19:45:57 +0000130 struct sock sk;
131 struct socket socket;
Jason Wang6e914fc2012-10-31 19:45:58 +0000132 struct tun_struct __rcu *tun;
Jason Wang54f968d2012-10-31 19:45:57 +0000133 struct fasync_struct *fasync;
134 /* only used for fasnyc */
135 unsigned int flags;
Pavel Emelyanovfb7589a2013-08-21 14:31:38 +0400136 union {
137 u16 queue_index;
138 unsigned int ifindex;
139 };
Petar Penkov94317092017-09-22 13:49:14 -0700140 struct napi_struct napi;
Eric Dumazetaec72f32017-10-18 12:12:09 -0700141 bool napi_enabled;
Eric Dumazetaf3fb242018-09-28 14:51:49 -0700142 bool napi_frags_enabled;
Petar Penkov90e33d42017-09-22 13:49:15 -0700143 struct mutex napi_mutex; /* Protects access to the above napi */
Jason Wang4008e972012-12-13 23:53:30 +0000144 struct list_head next;
145 struct tun_struct *detached;
Jason Wang5990a302018-01-04 11:14:27 +0800146 struct ptr_ring tx_ring;
Jesper Dangaard Brouer8bf5c4e2018-01-03 11:25:59 +0100147 struct xdp_rxq_info xdp_rxq;
Eric W. Biederman631ab462009-01-20 11:00:40 +0000148};
149
Jason Wangf9e06c42018-11-15 17:43:10 +0800150struct tun_page {
151 struct page *page;
152 int count;
153};
154
Jason Wang96442e422012-10-31 19:46:02 +0000155struct tun_flow_entry {
156 struct hlist_node hash_link;
157 struct rcu_head rcu;
158 struct tun_struct *tun;
159
160 u32 rxhash;
Tom Herbert9bc88932013-12-22 18:54:32 +0800161 u32 rps_rxhash;
Jason Wang96442e422012-10-31 19:46:02 +0000162 int queue_index;
Li RongQing83b1bc12018-12-06 16:08:17 +0800163 unsigned long updated ____cacheline_aligned_in_smp;
Jason Wang96442e422012-10-31 19:46:02 +0000164};
165
166#define TUN_NUM_FLOW_ENTRIES 1024
Li RongQingf13b5462018-08-03 15:50:02 +0800167#define TUN_MASK_FLOW_ENTRIES (TUN_NUM_FLOW_ENTRIES - 1)
Jason Wang96442e422012-10-31 19:46:02 +0000168
Jason Wangcd5681d2018-01-16 16:31:01 +0800169struct tun_prog {
Jason Wang96f84062017-12-04 17:31:23 +0800170 struct rcu_head rcu;
171 struct bpf_prog *prog;
172};
173
Jason Wang54f968d2012-10-31 19:45:57 +0000174/* Since the socket were moved to tun_file, to preserve the behavior of persist
Rami Rosen36fe8c092012-11-25 22:07:40 +0000175 * device, socket filter, sndbuf and vnet header size were restore when the
Jason Wang54f968d2012-10-31 19:45:57 +0000176 * file were attached to a persist device.
177 */
Rusty Russell14daa022008-04-12 18:48:58 -0700178struct tun_struct {
Jason Wangc8d68e62012-10-31 19:46:00 +0000179 struct tun_file __rcu *tfiles[MAX_TAP_QUEUES];
180 unsigned int numqueues;
Max Krasnyanskyf271b2cc2008-07-14 22:18:19 -0700181 unsigned int flags;
Eric W. Biederman0625c882012-02-07 16:48:55 -0800182 kuid_t owner;
183 kgid_t group;
Rusty Russell14daa022008-04-12 18:48:58 -0700184
Rusty Russell14daa022008-04-12 18:48:58 -0700185 struct net_device *dev;
Michał Mirosławc8f44af2011-11-15 15:29:55 +0000186 netdev_features_t set_features;
Michał Mirosław88255372011-04-19 06:13:10 +0000187#define TUN_USER_FEATURES (NETIF_F_HW_CSUM|NETIF_F_TSO_ECN|NETIF_F_TSO| \
David S. Millerd591a1f2017-07-03 06:35:32 -0700188 NETIF_F_TSO6)
Michael S. Tsirkind9d52b52010-03-17 17:45:01 +0200189
Paolo Abenieaea34b2016-02-26 10:45:40 +0100190 int align;
Michael S. Tsirkind9d52b52010-03-17 17:45:01 +0200191 int vnet_hdr_sz;
Jason Wang54f968d2012-10-31 19:45:57 +0000192 int sndbuf;
193 struct tap_filter txflt;
194 struct sock_fprog fprog;
195 /* protected by rtnl lock */
196 bool filter_attached;
Michal Kubecek34241702020-03-04 17:24:14 +0100197 u32 msg_enable;
Jason Wang96442e422012-10-31 19:46:02 +0000198 spinlock_t lock;
Jason Wang96442e422012-10-31 19:46:02 +0000199 struct hlist_head flows[TUN_NUM_FLOW_ENTRIES];
200 struct timer_list flow_gc_timer;
201 unsigned long ageing_time;
Jason Wang4008e972012-12-13 23:53:30 +0000202 unsigned int numdisabled;
203 struct list_head disabled;
Paul Moore5dbbaf22013-01-14 07:12:19 +0000204 void *security;
Jason Wangb8732fb2013-01-23 03:59:13 +0000205 u32 flow_count;
Jason Wang5503fce2017-01-18 15:02:03 +0800206 u32 rx_batched;
Heiner Kallweit497a5752020-11-07 21:50:56 +0100207 atomic_long_t rx_frame_errors;
Jason Wang761876c2017-08-11 19:41:18 +0800208 struct bpf_prog __rcu *xdp_prog;
Jason Wangcd5681d2018-01-16 16:31:01 +0800209 struct tun_prog __rcu *steering_prog;
Jason Wangaff3d702018-01-16 16:31:02 +0800210 struct tun_prog __rcu *filter_prog;
Chas Williams4e24f2d2018-06-02 17:49:53 -0400211 struct ethtool_link_ksettings link_ksettings;
Rusty Russell14daa022008-04-12 18:48:58 -0700212};
213
Jason Wangaff3d702018-01-16 16:31:02 +0800214struct veth {
215 __be16 h_vlan_proto;
216 __be16 h_vlan_TCI;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700217};
218
Petar Penkov94317092017-09-22 13:49:14 -0700219static int tun_napi_receive(struct napi_struct *napi, int budget)
220{
221 struct tun_file *tfile = container_of(napi, struct tun_file, napi);
222 struct sk_buff_head *queue = &tfile->sk.sk_write_queue;
223 struct sk_buff_head process_queue;
224 struct sk_buff *skb;
225 int received = 0;
226
227 __skb_queue_head_init(&process_queue);
228
229 spin_lock(&queue->lock);
230 skb_queue_splice_tail_init(queue, &process_queue);
231 spin_unlock(&queue->lock);
232
233 while (received < budget && (skb = __skb_dequeue(&process_queue))) {
234 napi_gro_receive(napi, skb);
235 ++received;
236 }
237
238 if (!skb_queue_empty(&process_queue)) {
239 spin_lock(&queue->lock);
240 skb_queue_splice(&process_queue, queue);
241 spin_unlock(&queue->lock);
242 }
243
244 return received;
245}
246
247static int tun_napi_poll(struct napi_struct *napi, int budget)
248{
249 unsigned int received;
250
251 received = tun_napi_receive(napi, budget);
252
253 if (received < budget)
254 napi_complete_done(napi, received);
255
256 return received;
257}
258
259static void tun_napi_init(struct tun_struct *tun, struct tun_file *tfile,
Eric Dumazetaf3fb242018-09-28 14:51:49 -0700260 bool napi_en, bool napi_frags)
Petar Penkov94317092017-09-22 13:49:14 -0700261{
Eric Dumazetaec72f32017-10-18 12:12:09 -0700262 tfile->napi_enabled = napi_en;
Eric Dumazetaf3fb242018-09-28 14:51:49 -0700263 tfile->napi_frags_enabled = napi_en && napi_frags;
Petar Penkov94317092017-09-22 13:49:14 -0700264 if (napi_en) {
Petar Penkovc39e3422019-11-14 09:52:09 -0800265 netif_tx_napi_add(tun->dev, &tfile->napi, tun_napi_poll,
266 NAPI_POLL_WEIGHT);
Petar Penkov94317092017-09-22 13:49:14 -0700267 napi_enable(&tfile->napi);
268 }
269}
270
Eric Dumazet06e55ad2018-09-28 14:51:47 -0700271static void tun_napi_disable(struct tun_file *tfile)
Petar Penkov94317092017-09-22 13:49:14 -0700272{
Eric Dumazetaec72f32017-10-18 12:12:09 -0700273 if (tfile->napi_enabled)
Petar Penkov94317092017-09-22 13:49:14 -0700274 napi_disable(&tfile->napi);
275}
276
Eric Dumazet06e55ad2018-09-28 14:51:47 -0700277static void tun_napi_del(struct tun_file *tfile)
Petar Penkov94317092017-09-22 13:49:14 -0700278{
Eric Dumazetaec72f32017-10-18 12:12:09 -0700279 if (tfile->napi_enabled)
Petar Penkov94317092017-09-22 13:49:14 -0700280 netif_napi_del(&tfile->napi);
281}
282
Eric Dumazetaf3fb242018-09-28 14:51:49 -0700283static bool tun_napi_frags_enabled(const struct tun_file *tfile)
Petar Penkov90e33d42017-09-22 13:49:15 -0700284{
Eric Dumazetaf3fb242018-09-28 14:51:49 -0700285 return tfile->napi_frags_enabled;
Petar Penkov90e33d42017-09-22 13:49:15 -0700286}
287
Greg Kurz8b8e6582015-04-24 14:50:36 +0200288#ifdef CONFIG_TUN_VNET_CROSS_LE
289static inline bool tun_legacy_is_little_endian(struct tun_struct *tun)
290{
291 return tun->flags & TUN_VNET_BE ? false :
292 virtio_legacy_is_little_endian();
293}
294
295static long tun_get_vnet_be(struct tun_struct *tun, int __user *argp)
296{
297 int be = !!(tun->flags & TUN_VNET_BE);
298
299 if (put_user(be, argp))
300 return -EFAULT;
301
302 return 0;
303}
304
305static long tun_set_vnet_be(struct tun_struct *tun, int __user *argp)
306{
307 int be;
308
309 if (get_user(be, argp))
310 return -EFAULT;
311
312 if (be)
313 tun->flags |= TUN_VNET_BE;
314 else
315 tun->flags &= ~TUN_VNET_BE;
316
317 return 0;
318}
319#else
320static inline bool tun_legacy_is_little_endian(struct tun_struct *tun)
321{
322 return virtio_legacy_is_little_endian();
323}
324
325static long tun_get_vnet_be(struct tun_struct *tun, int __user *argp)
326{
327 return -EINVAL;
328}
329
330static long tun_set_vnet_be(struct tun_struct *tun, int __user *argp)
331{
332 return -EINVAL;
333}
334#endif /* CONFIG_TUN_VNET_CROSS_LE */
335
Greg Kurz25bd55bb2015-04-24 14:24:38 +0200336static inline bool tun_is_little_endian(struct tun_struct *tun)
337{
Greg Kurz7d824102015-04-24 14:26:24 +0200338 return tun->flags & TUN_VNET_LE ||
Greg Kurz8b8e6582015-04-24 14:50:36 +0200339 tun_legacy_is_little_endian(tun);
Greg Kurz25bd55bb2015-04-24 14:24:38 +0200340}
341
Michael S. Tsirkin56f0dcc2014-10-23 22:59:31 +0300342static inline u16 tun16_to_cpu(struct tun_struct *tun, __virtio16 val)
343{
Greg Kurz25bd55bb2015-04-24 14:24:38 +0200344 return __virtio16_to_cpu(tun_is_little_endian(tun), val);
Michael S. Tsirkin56f0dcc2014-10-23 22:59:31 +0300345}
346
347static inline __virtio16 cpu_to_tun16(struct tun_struct *tun, u16 val)
348{
Greg Kurz25bd55bb2015-04-24 14:24:38 +0200349 return __cpu_to_virtio16(tun_is_little_endian(tun), val);
Michael S. Tsirkin56f0dcc2014-10-23 22:59:31 +0300350}
351
Jason Wang96442e422012-10-31 19:46:02 +0000352static inline u32 tun_hashfn(u32 rxhash)
353{
Li RongQingf13b5462018-08-03 15:50:02 +0800354 return rxhash & TUN_MASK_FLOW_ENTRIES;
Jason Wang96442e422012-10-31 19:46:02 +0000355}
356
357static struct tun_flow_entry *tun_flow_find(struct hlist_head *head, u32 rxhash)
358{
359 struct tun_flow_entry *e;
Jason Wang96442e422012-10-31 19:46:02 +0000360
Sasha Levinb67bfe02013-02-27 17:06:00 -0800361 hlist_for_each_entry_rcu(e, head, hash_link) {
Jason Wang96442e422012-10-31 19:46:02 +0000362 if (e->rxhash == rxhash)
363 return e;
364 }
365 return NULL;
366}
367
368static struct tun_flow_entry *tun_flow_create(struct tun_struct *tun,
369 struct hlist_head *head,
370 u32 rxhash, u16 queue_index)
371{
Eric Dumazet9fdc6be2012-12-21 07:17:21 +0000372 struct tun_flow_entry *e = kmalloc(sizeof(*e), GFP_ATOMIC);
373
Jason Wang96442e422012-10-31 19:46:02 +0000374 if (e) {
Michal Kubecek34241702020-03-04 17:24:14 +0100375 netif_info(tun, tx_queued, tun->dev,
376 "create flow: hash %u index %u\n",
377 rxhash, queue_index);
Jason Wang96442e422012-10-31 19:46:02 +0000378 e->updated = jiffies;
379 e->rxhash = rxhash;
Tom Herbert9bc88932013-12-22 18:54:32 +0800380 e->rps_rxhash = 0;
Jason Wang96442e422012-10-31 19:46:02 +0000381 e->queue_index = queue_index;
382 e->tun = tun;
383 hlist_add_head_rcu(&e->hash_link, head);
Jason Wangb8732fb2013-01-23 03:59:13 +0000384 ++tun->flow_count;
Jason Wang96442e422012-10-31 19:46:02 +0000385 }
386 return e;
387}
388
Jason Wang96442e422012-10-31 19:46:02 +0000389static void tun_flow_delete(struct tun_struct *tun, struct tun_flow_entry *e)
390{
Michal Kubecek34241702020-03-04 17:24:14 +0100391 netif_info(tun, tx_queued, tun->dev, "delete flow: hash %u index %u\n",
392 e->rxhash, e->queue_index);
Jason Wang96442e422012-10-31 19:46:02 +0000393 hlist_del_rcu(&e->hash_link);
Eric Dumazet9fdc6be2012-12-21 07:17:21 +0000394 kfree_rcu(e, rcu);
Jason Wangb8732fb2013-01-23 03:59:13 +0000395 --tun->flow_count;
Jason Wang96442e422012-10-31 19:46:02 +0000396}
397
398static void tun_flow_flush(struct tun_struct *tun)
399{
400 int i;
401
402 spin_lock_bh(&tun->lock);
403 for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++) {
404 struct tun_flow_entry *e;
Sasha Levinb67bfe02013-02-27 17:06:00 -0800405 struct hlist_node *n;
Jason Wang96442e422012-10-31 19:46:02 +0000406
Sasha Levinb67bfe02013-02-27 17:06:00 -0800407 hlist_for_each_entry_safe(e, n, &tun->flows[i], hash_link)
Jason Wang96442e422012-10-31 19:46:02 +0000408 tun_flow_delete(tun, e);
409 }
410 spin_unlock_bh(&tun->lock);
411}
412
413static void tun_flow_delete_by_queue(struct tun_struct *tun, u16 queue_index)
414{
415 int i;
416
417 spin_lock_bh(&tun->lock);
418 for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++) {
419 struct tun_flow_entry *e;
Sasha Levinb67bfe02013-02-27 17:06:00 -0800420 struct hlist_node *n;
Jason Wang96442e422012-10-31 19:46:02 +0000421
Sasha Levinb67bfe02013-02-27 17:06:00 -0800422 hlist_for_each_entry_safe(e, n, &tun->flows[i], hash_link) {
Jason Wang96442e422012-10-31 19:46:02 +0000423 if (e->queue_index == queue_index)
424 tun_flow_delete(tun, e);
425 }
426 }
427 spin_unlock_bh(&tun->lock);
428}
429
Kees Cooke99e88a2017-10-16 14:43:17 -0700430static void tun_flow_cleanup(struct timer_list *t)
Jason Wang96442e422012-10-31 19:46:02 +0000431{
Kees Cooke99e88a2017-10-16 14:43:17 -0700432 struct tun_struct *tun = from_timer(tun, t, flow_gc_timer);
Jason Wang96442e422012-10-31 19:46:02 +0000433 unsigned long delay = tun->ageing_time;
434 unsigned long next_timer = jiffies + delay;
435 unsigned long count = 0;
436 int i;
437
Eric Dumazet7dbfb4e2017-10-20 11:29:55 -0700438 spin_lock(&tun->lock);
Jason Wang96442e422012-10-31 19:46:02 +0000439 for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++) {
440 struct tun_flow_entry *e;
Sasha Levinb67bfe02013-02-27 17:06:00 -0800441 struct hlist_node *n;
Jason Wang96442e422012-10-31 19:46:02 +0000442
Sasha Levinb67bfe02013-02-27 17:06:00 -0800443 hlist_for_each_entry_safe(e, n, &tun->flows[i], hash_link) {
Jason Wang96442e422012-10-31 19:46:02 +0000444 unsigned long this_timer;
Eric Dumazet81d98fa2017-10-20 11:29:56 -0700445
Jason Wang96442e422012-10-31 19:46:02 +0000446 this_timer = e->updated + delay;
Eric Dumazet81d98fa2017-10-20 11:29:56 -0700447 if (time_before_eq(this_timer, jiffies)) {
Jason Wang96442e422012-10-31 19:46:02 +0000448 tun_flow_delete(tun, e);
Eric Dumazet81d98fa2017-10-20 11:29:56 -0700449 continue;
450 }
451 count++;
452 if (time_before(this_timer, next_timer))
Jason Wang96442e422012-10-31 19:46:02 +0000453 next_timer = this_timer;
454 }
455 }
456
457 if (count)
458 mod_timer(&tun->flow_gc_timer, round_jiffies_up(next_timer));
Eric Dumazet7dbfb4e2017-10-20 11:29:55 -0700459 spin_unlock(&tun->lock);
Jason Wang96442e422012-10-31 19:46:02 +0000460}
461
Eric Dumazet49974422012-12-12 19:22:57 +0000462static void tun_flow_update(struct tun_struct *tun, u32 rxhash,
Jason Wang9e857222013-01-28 01:05:19 +0000463 struct tun_file *tfile)
Jason Wang96442e422012-10-31 19:46:02 +0000464{
465 struct hlist_head *head;
466 struct tun_flow_entry *e;
467 unsigned long delay = tun->ageing_time;
Jason Wang9e857222013-01-28 01:05:19 +0000468 u16 queue_index = tfile->queue_index;
Jason Wang96442e422012-10-31 19:46:02 +0000469
Li RongQing5c327f62018-12-06 16:28:11 +0800470 head = &tun->flows[tun_hashfn(rxhash)];
Jason Wang96442e422012-10-31 19:46:02 +0000471
472 rcu_read_lock();
473
Jason Wang96442e422012-10-31 19:46:02 +0000474 e = tun_flow_find(head, rxhash);
475 if (likely(e)) {
476 /* TODO: keep queueing to old queue until it's empty? */
Eric Dumazet4ffdd222019-10-09 09:20:02 -0700477 if (READ_ONCE(e->queue_index) != queue_index)
478 WRITE_ONCE(e->queue_index, queue_index);
Li RongQing83b1bc12018-12-06 16:08:17 +0800479 if (e->updated != jiffies)
480 e->updated = jiffies;
Tom Herbert9bc88932013-12-22 18:54:32 +0800481 sock_rps_record_flow_hash(e->rps_rxhash);
Jason Wang96442e422012-10-31 19:46:02 +0000482 } else {
483 spin_lock_bh(&tun->lock);
Jason Wangb8732fb2013-01-23 03:59:13 +0000484 if (!tun_flow_find(head, rxhash) &&
485 tun->flow_count < MAX_TAP_FLOWS)
Jason Wang96442e422012-10-31 19:46:02 +0000486 tun_flow_create(tun, head, rxhash, queue_index);
487
488 if (!timer_pending(&tun->flow_gc_timer))
489 mod_timer(&tun->flow_gc_timer,
490 round_jiffies_up(jiffies + delay));
491 spin_unlock_bh(&tun->lock);
492 }
493
Jason Wang96442e422012-10-31 19:46:02 +0000494 rcu_read_unlock();
495}
496
Michal Kubecek516c5122020-03-04 17:23:59 +0100497/* Save the hash received in the stack receive path and update the
Tom Herbert9bc88932013-12-22 18:54:32 +0800498 * flow_hash table accordingly.
499 */
500static inline void tun_flow_save_rps_rxhash(struct tun_flow_entry *e, u32 hash)
501{
Eric Dumazet567e4b72015-02-06 12:59:01 -0800502 if (unlikely(e->rps_rxhash != hash))
Tom Herbert9bc88932013-12-22 18:54:32 +0800503 e->rps_rxhash = hash;
Tom Herbert9bc88932013-12-22 18:54:32 +0800504}
505
Wang Li4b035272018-10-09 10:32:04 +0800506/* We try to identify a flow through its rxhash. The reason that
stephen hemminger92d4ea62013-12-05 20:42:58 -0800507 * we do not check rxq no. is because some cards(e.g 82599), chooses
Jason Wangc8d68e62012-10-31 19:46:00 +0000508 * the rxq based on the txq where the last packet of the flow comes. As
509 * the userspace application move between processors, we may get a
Wang Li4b035272018-10-09 10:32:04 +0800510 * different rxq no. here.
Jason Wangc8d68e62012-10-31 19:46:00 +0000511 */
Jason Wang96f84062017-12-04 17:31:23 +0800512static u16 tun_automq_select_queue(struct tun_struct *tun, struct sk_buff *skb)
Jason Wangc8d68e62012-10-31 19:46:00 +0000513{
Jason Wang96442e422012-10-31 19:46:02 +0000514 struct tun_flow_entry *e;
Jason Wangc8d68e62012-10-31 19:46:00 +0000515 u32 txq = 0;
516 u32 numqueues = 0;
517
Mark Rutland6aa7de02017-10-23 14:07:29 -0700518 numqueues = READ_ONCE(tun->numqueues);
Jason Wangc8d68e62012-10-31 19:46:00 +0000519
Jason Wangfeec0842017-06-06 14:09:49 +0800520 txq = __skb_get_hash_symmetric(skb);
Wang Li4b035272018-10-09 10:32:04 +0800521 e = tun_flow_find(&tun->flows[tun_hashfn(txq)], txq);
522 if (e) {
523 tun_flow_save_rps_rxhash(e, txq);
524 txq = e->queue_index;
525 } else {
526 /* use multiply and shift instead of expensive divide */
527 txq = ((u64)txq * numqueues) >> 32;
Jason Wangc8d68e62012-10-31 19:46:00 +0000528 }
529
Jason Wangc8d68e62012-10-31 19:46:00 +0000530 return txq;
531}
532
Jason Wang96f84062017-12-04 17:31:23 +0800533static u16 tun_ebpf_select_queue(struct tun_struct *tun, struct sk_buff *skb)
534{
Jason Wangcd5681d2018-01-16 16:31:01 +0800535 struct tun_prog *prog;
Jason Wanga35d3102019-05-08 23:20:17 -0400536 u32 numqueues;
Jason Wang96f84062017-12-04 17:31:23 +0800537 u16 ret = 0;
538
Jason Wanga35d3102019-05-08 23:20:17 -0400539 numqueues = READ_ONCE(tun->numqueues);
540 if (!numqueues)
541 return 0;
542
Jason Wang96f84062017-12-04 17:31:23 +0800543 prog = rcu_dereference(tun->steering_prog);
544 if (prog)
545 ret = bpf_prog_run_clear_cb(prog->prog, skb);
546
Jason Wanga35d3102019-05-08 23:20:17 -0400547 return ret % numqueues;
Jason Wang96f84062017-12-04 17:31:23 +0800548}
549
550static u16 tun_select_queue(struct net_device *dev, struct sk_buff *skb,
Paolo Abenia350ecc2019-03-20 11:02:06 +0100551 struct net_device *sb_dev)
Jason Wang96f84062017-12-04 17:31:23 +0800552{
553 struct tun_struct *tun = netdev_priv(dev);
554 u16 ret;
555
556 rcu_read_lock();
557 if (rcu_dereference(tun->steering_prog))
558 ret = tun_ebpf_select_queue(tun, skb);
559 else
560 ret = tun_automq_select_queue(tun, skb);
561 rcu_read_unlock();
562
563 return ret;
564}
565
Jason Wangcde8b152012-10-31 19:46:01 +0000566static inline bool tun_not_capable(struct tun_struct *tun)
567{
568 const struct cred *cred = current_cred();
Eric W. Biedermanc260b772012-11-18 21:34:11 +0000569 struct net *net = dev_net(tun->dev);
Jason Wangcde8b152012-10-31 19:46:01 +0000570
571 return ((uid_valid(tun->owner) && !uid_eq(cred->euid, tun->owner)) ||
572 (gid_valid(tun->group) && !in_egroup_p(tun->group))) &&
Eric W. Biedermanc260b772012-11-18 21:34:11 +0000573 !ns_capable(net->user_ns, CAP_NET_ADMIN);
Jason Wangcde8b152012-10-31 19:46:01 +0000574}
575
Jason Wangc8d68e62012-10-31 19:46:00 +0000576static void tun_set_real_num_queues(struct tun_struct *tun)
577{
578 netif_set_real_num_tx_queues(tun->dev, tun->numqueues);
579 netif_set_real_num_rx_queues(tun->dev, tun->numqueues);
580}
581
Jason Wang4008e972012-12-13 23:53:30 +0000582static void tun_disable_queue(struct tun_struct *tun, struct tun_file *tfile)
583{
584 tfile->detached = tun;
585 list_add_tail(&tfile->next, &tun->disabled);
586 ++tun->numdisabled;
587}
588
Jason Wangd32649d2012-12-18 11:00:27 +0800589static struct tun_struct *tun_enable_queue(struct tun_file *tfile)
Jason Wang4008e972012-12-13 23:53:30 +0000590{
591 struct tun_struct *tun = tfile->detached;
592
593 tfile->detached = NULL;
594 list_del_init(&tfile->next);
595 --tun->numdisabled;
596 return tun;
597}
598
Jason Wang3a403072018-03-09 14:50:34 +0800599void tun_ptr_free(void *ptr)
Jason Wangfc72d1d2018-01-04 11:14:28 +0800600{
601 if (!ptr)
602 return;
Jesper Dangaard Brouer1ffcbc82018-04-17 16:45:47 +0200603 if (tun_is_xdp_frame(ptr)) {
604 struct xdp_frame *xdpf = tun_ptr_to_xdp(ptr);
Jason Wangfc72d1d2018-01-04 11:14:28 +0800605
Jesper Dangaard Brouer03993092018-04-17 16:46:32 +0200606 xdp_return_frame(xdpf);
Jason Wangfc72d1d2018-01-04 11:14:28 +0800607 } else {
608 __skb_array_destroy_skb(ptr);
609 }
610}
Jason Wang3a403072018-03-09 14:50:34 +0800611EXPORT_SYMBOL_GPL(tun_ptr_free);
Jason Wangfc72d1d2018-01-04 11:14:28 +0800612
Jason Wang4bfb0512013-09-05 17:53:59 +0800613static void tun_queue_purge(struct tun_file *tfile)
614{
Jason Wangfc72d1d2018-01-04 11:14:28 +0800615 void *ptr;
Jason Wang1576d982016-06-30 14:45:36 +0800616
Jason Wangfc72d1d2018-01-04 11:14:28 +0800617 while ((ptr = ptr_ring_consume(&tfile->tx_ring)) != NULL)
618 tun_ptr_free(ptr);
Jason Wang1576d982016-06-30 14:45:36 +0800619
Jason Wang5503fce2017-01-18 15:02:03 +0800620 skb_queue_purge(&tfile->sk.sk_write_queue);
Jason Wang4bfb0512013-09-05 17:53:59 +0800621 skb_queue_purge(&tfile->sk.sk_error_queue);
622}
623
Jason Wangc8d68e62012-10-31 19:46:00 +0000624static void __tun_detach(struct tun_file *tfile, bool clean)
625{
626 struct tun_file *ntfile;
627 struct tun_struct *tun;
Jason Wangc8d68e62012-10-31 19:46:00 +0000628
Jason Wangb8deabd2013-01-11 16:59:32 +0000629 tun = rtnl_dereference(tfile->tun);
630
Petar Penkov94317092017-09-22 13:49:14 -0700631 if (tun && clean) {
Eric Dumazet06e55ad2018-09-28 14:51:47 -0700632 tun_napi_disable(tfile);
633 tun_napi_del(tfile);
Petar Penkov94317092017-09-22 13:49:14 -0700634 }
635
Jason Wang9e857222013-01-28 01:05:19 +0000636 if (tun && !tfile->detached) {
Jason Wangc8d68e62012-10-31 19:46:00 +0000637 u16 index = tfile->queue_index;
638 BUG_ON(index >= tun->numqueues);
Jason Wangc8d68e62012-10-31 19:46:00 +0000639
640 rcu_assign_pointer(tun->tfiles[index],
641 tun->tfiles[tun->numqueues - 1]);
Jason Wangb8deabd2013-01-11 16:59:32 +0000642 ntfile = rtnl_dereference(tun->tfiles[index]);
Jason Wangc8d68e62012-10-31 19:46:00 +0000643 ntfile->queue_index = index;
Jason Wang9871a9e2019-05-08 23:20:18 -0400644 rcu_assign_pointer(tun->tfiles[tun->numqueues - 1],
645 NULL);
Jason Wangc8d68e62012-10-31 19:46:00 +0000646
647 --tun->numqueues;
Jason Wang9e857222013-01-28 01:05:19 +0000648 if (clean) {
Monam Agarwalc9566742014-03-24 00:02:32 +0530649 RCU_INIT_POINTER(tfile->tun, NULL);
Jason Wang4008e972012-12-13 23:53:30 +0000650 sock_put(&tfile->sk);
Jason Wang9e857222013-01-28 01:05:19 +0000651 } else
Jason Wang4008e972012-12-13 23:53:30 +0000652 tun_disable_queue(tun, tfile);
Jason Wangc8d68e62012-10-31 19:46:00 +0000653
654 synchronize_net();
Jason Wang96442e422012-10-31 19:46:02 +0000655 tun_flow_delete_by_queue(tun, tun->numqueues + 1);
Jason Wangc8d68e62012-10-31 19:46:00 +0000656 /* Drop read queue */
Jason Wang4bfb0512013-09-05 17:53:59 +0800657 tun_queue_purge(tfile);
Jason Wangc8d68e62012-10-31 19:46:00 +0000658 tun_set_real_num_queues(tun);
Jason Wangdd38bd82013-01-11 16:59:34 +0000659 } else if (tfile->detached && clean) {
Jason Wang4008e972012-12-13 23:53:30 +0000660 tun = tun_enable_queue(tfile);
Jason Wangdd38bd82013-01-11 16:59:34 +0000661 sock_put(&tfile->sk);
662 }
Jason Wangc8d68e62012-10-31 19:46:00 +0000663
664 if (clean) {
Michael S. Tsirkinaf668b32013-01-28 00:38:02 +0000665 if (tun && tun->numqueues == 0 && tun->numdisabled == 0) {
666 netif_carrier_off(tun->dev);
667
Michael S. Tsirkin40630b82014-11-19 15:17:31 +0200668 if (!(tun->flags & IFF_PERSIST) &&
Michael S. Tsirkinaf668b32013-01-28 00:38:02 +0000669 tun->dev->reg_state == NETREG_REGISTERED)
Jason Wang4008e972012-12-13 23:53:30 +0000670 unregister_netdevice(tun->dev);
Michael S. Tsirkinaf668b32013-01-28 00:38:02 +0000671 }
Jason Wangb196d882018-05-11 10:49:25 +0800672 if (tun)
673 xdp_rxq_info_unreg(&tfile->xdp_rxq);
Jason Wang7063efd2018-05-16 20:39:33 +0800674 ptr_ring_cleanup(&tfile->tx_ring, tun_ptr_free);
Eric W. Biederman140e807da2015-05-08 21:07:08 -0500675 sock_put(&tfile->sk);
Jason Wangc8d68e62012-10-31 19:46:00 +0000676 }
677}
678
679static void tun_detach(struct tun_file *tfile, bool clean)
680{
Sabrina Dubroca83c1f362018-04-10 16:28:56 +0200681 struct tun_struct *tun;
682 struct net_device *dev;
683
Jason Wangc8d68e62012-10-31 19:46:00 +0000684 rtnl_lock();
Sabrina Dubroca83c1f362018-04-10 16:28:56 +0200685 tun = rtnl_dereference(tfile->tun);
686 dev = tun ? tun->dev : NULL;
Jason Wangc8d68e62012-10-31 19:46:00 +0000687 __tun_detach(tfile, clean);
Sabrina Dubroca83c1f362018-04-10 16:28:56 +0200688 if (dev)
689 netdev_state_change(dev);
Jason Wangc8d68e62012-10-31 19:46:00 +0000690 rtnl_unlock();
691}
692
693static void tun_detach_all(struct net_device *dev)
694{
695 struct tun_struct *tun = netdev_priv(dev);
Jason Wang4008e972012-12-13 23:53:30 +0000696 struct tun_file *tfile, *tmp;
Jason Wangc8d68e62012-10-31 19:46:00 +0000697 int i, n = tun->numqueues;
698
699 for (i = 0; i < n; i++) {
Jason Wangb8deabd2013-01-11 16:59:32 +0000700 tfile = rtnl_dereference(tun->tfiles[i]);
Jason Wangc8d68e62012-10-31 19:46:00 +0000701 BUG_ON(!tfile);
Eric Dumazet06e55ad2018-09-28 14:51:47 -0700702 tun_napi_disable(tfile);
Jason Wangaddf8fc2016-05-19 13:36:51 +0800703 tfile->socket.sk->sk_shutdown = RCV_SHUTDOWN;
Xi Wang9e641bd2014-05-16 15:11:48 -0700704 tfile->socket.sk->sk_data_ready(tfile->socket.sk);
Monam Agarwalc9566742014-03-24 00:02:32 +0530705 RCU_INIT_POINTER(tfile->tun, NULL);
Jason Wangc8d68e62012-10-31 19:46:00 +0000706 --tun->numqueues;
707 }
Jason Wang9e857222013-01-28 01:05:19 +0000708 list_for_each_entry(tfile, &tun->disabled, next) {
Jason Wangaddf8fc2016-05-19 13:36:51 +0800709 tfile->socket.sk->sk_shutdown = RCV_SHUTDOWN;
Xi Wang9e641bd2014-05-16 15:11:48 -0700710 tfile->socket.sk->sk_data_ready(tfile->socket.sk);
Monam Agarwalc9566742014-03-24 00:02:32 +0530711 RCU_INIT_POINTER(tfile->tun, NULL);
Jason Wang9e857222013-01-28 01:05:19 +0000712 }
Jason Wangc8d68e62012-10-31 19:46:00 +0000713 BUG_ON(tun->numqueues != 0);
714
715 synchronize_net();
716 for (i = 0; i < n; i++) {
Jason Wangb8deabd2013-01-11 16:59:32 +0000717 tfile = rtnl_dereference(tun->tfiles[i]);
Eric Dumazet06e55ad2018-09-28 14:51:47 -0700718 tun_napi_del(tfile);
Jason Wangc8d68e62012-10-31 19:46:00 +0000719 /* Drop read queue */
Jason Wang4bfb0512013-09-05 17:53:59 +0800720 tun_queue_purge(tfile);
Jason Wangb196d882018-05-11 10:49:25 +0800721 xdp_rxq_info_unreg(&tfile->xdp_rxq);
Jason Wangc8d68e62012-10-31 19:46:00 +0000722 sock_put(&tfile->sk);
723 }
Jason Wang4008e972012-12-13 23:53:30 +0000724 list_for_each_entry_safe(tfile, tmp, &tun->disabled, next) {
725 tun_enable_queue(tfile);
Jason Wang4bfb0512013-09-05 17:53:59 +0800726 tun_queue_purge(tfile);
Jason Wangb196d882018-05-11 10:49:25 +0800727 xdp_rxq_info_unreg(&tfile->xdp_rxq);
Jason Wang4008e972012-12-13 23:53:30 +0000728 sock_put(&tfile->sk);
729 }
730 BUG_ON(tun->numdisabled != 0);
Jason Wangdd38bd82013-01-11 16:59:34 +0000731
Michael S. Tsirkin40630b82014-11-19 15:17:31 +0200732 if (tun->flags & IFF_PERSIST)
Jason Wangdd38bd82013-01-11 16:59:34 +0000733 module_put(THIS_MODULE);
Jason Wangc8d68e62012-10-31 19:46:00 +0000734}
735
Petar Penkov94317092017-09-22 13:49:14 -0700736static int tun_attach(struct tun_struct *tun, struct file *file,
Yang Yingliang77f22f92019-09-10 18:56:57 +0800737 bool skip_filter, bool napi, bool napi_frags,
738 bool publish_tun)
Eric W. Biedermana7385ba2009-01-20 10:57:48 +0000739{
Eric W. Biederman631ab462009-01-20 11:00:40 +0000740 struct tun_file *tfile = file->private_data;
Jason Wang1576d982016-06-30 14:45:36 +0800741 struct net_device *dev = tun->dev;
Eric W. Biederman38231b72009-01-20 11:02:28 +0000742 int err;
Eric W. Biedermana7385ba2009-01-20 10:57:48 +0000743
Paul Moore5dbbaf22013-01-14 07:12:19 +0000744 err = security_tun_dev_attach(tfile->socket.sk, tun->security);
745 if (err < 0)
746 goto out;
747
Eric W. Biederman38231b72009-01-20 11:02:28 +0000748 err = -EINVAL;
Jason Wang9e857222013-01-28 01:05:19 +0000749 if (rtnl_dereference(tfile->tun) && !tfile->detached)
Eric W. Biederman38231b72009-01-20 11:02:28 +0000750 goto out;
751
752 err = -EBUSY;
Michael S. Tsirkin40630b82014-11-19 15:17:31 +0200753 if (!(tun->flags & IFF_MULTI_QUEUE) && tun->numqueues == 1)
Jason Wangc8d68e62012-10-31 19:46:00 +0000754 goto out;
755
756 err = -E2BIG;
Jason Wang4008e972012-12-13 23:53:30 +0000757 if (!tfile->detached &&
758 tun->numqueues + tun->numdisabled == MAX_TAP_QUEUES)
Eric W. Biederman38231b72009-01-20 11:02:28 +0000759 goto out;
760
761 err = 0;
Jason Wang54f968d2012-10-31 19:45:57 +0000762
stephen hemminger92d4ea62013-12-05 20:42:58 -0800763 /* Re-attach the filter to persist device */
Pavel Emelyanov849c9b62013-08-21 14:32:21 +0400764 if (!skip_filter && (tun->filter_attached == true)) {
Hannes Frederic Sowa8ced4252016-04-05 17:10:16 +0200765 lock_sock(tfile->socket.sk);
766 err = sk_attach_filter(&tun->fprog, tfile->socket.sk);
767 release_sock(tfile->socket.sk);
Jason Wang54f968d2012-10-31 19:45:57 +0000768 if (!err)
769 goto out;
770 }
Jason Wang1576d982016-06-30 14:45:36 +0800771
772 if (!tfile->detached &&
Jason Wangb196d882018-05-11 10:49:25 +0800773 ptr_ring_resize(&tfile->tx_ring, dev->tx_queue_len,
774 GFP_KERNEL, tun_ptr_free)) {
Jason Wang1576d982016-06-30 14:45:36 +0800775 err = -ENOMEM;
776 goto out;
777 }
778
Jason Wangc8d68e62012-10-31 19:46:00 +0000779 tfile->queue_index = tun->numqueues;
Jason Wangaddf8fc2016-05-19 13:36:51 +0800780 tfile->socket.sk->sk_shutdown &= ~RCV_SHUTDOWN;
Jesper Dangaard Brouer8bf5c4e2018-01-03 11:25:59 +0100781
782 if (tfile->detached) {
783 /* Re-attach detached tfile, updating XDP queue_index */
784 WARN_ON(!xdp_rxq_info_is_reg(&tfile->xdp_rxq));
785
786 if (tfile->xdp_rxq.queue_index != tfile->queue_index)
787 tfile->xdp_rxq.queue_index = tfile->queue_index;
788 } else {
789 /* Setup XDP RX-queue info, for new tfile getting attached */
790 err = xdp_rxq_info_reg(&tfile->xdp_rxq,
Björn Töpelb02e5a02020-11-30 19:52:01 +0100791 tun->dev, tfile->queue_index, 0);
Jesper Dangaard Brouer8bf5c4e2018-01-03 11:25:59 +0100792 if (err < 0)
793 goto out;
Jesper Dangaard Brouer8d5d8852018-04-17 16:46:12 +0200794 err = xdp_rxq_info_reg_mem_model(&tfile->xdp_rxq,
795 MEM_TYPE_PAGE_SHARED, NULL);
796 if (err < 0) {
797 xdp_rxq_info_unreg(&tfile->xdp_rxq);
798 goto out;
799 }
Jesper Dangaard Brouer8bf5c4e2018-01-03 11:25:59 +0100800 err = 0;
801 }
802
Petar Penkov94317092017-09-22 13:49:14 -0700803 if (tfile->detached) {
Jason Wang4008e972012-12-13 23:53:30 +0000804 tun_enable_queue(tfile);
Petar Penkov94317092017-09-22 13:49:14 -0700805 } else {
Jason Wang4008e972012-12-13 23:53:30 +0000806 sock_hold(&tfile->sk);
Eric Dumazetaf3fb242018-09-28 14:51:49 -0700807 tun_napi_init(tun, tfile, napi, napi_frags);
Petar Penkov94317092017-09-22 13:49:14 -0700808 }
Jason Wang4008e972012-12-13 23:53:30 +0000809
Jason Wange4a2a302018-09-12 11:16:59 +0800810 if (rtnl_dereference(tun->xdp_prog))
811 sock_set_flag(&tfile->sk, SOCK_XDP);
812
Jason Wangc8d68e62012-10-31 19:46:00 +0000813 /* device is allowed to go away first, so no need to hold extra
814 * refcnt.
815 */
Eric W. Biedermana7385ba2009-01-20 10:57:48 +0000816
Stanislav Fomichev0b7959b2019-01-07 13:38:38 -0800817 /* Publish tfile->tun and tun->tfiles only after we've fully
818 * initialized tfile; otherwise we risk using half-initialized
819 * object.
820 */
Yang Yingliang77f22f92019-09-10 18:56:57 +0800821 if (publish_tun)
822 rcu_assign_pointer(tfile->tun, tun);
Stanislav Fomichev0b7959b2019-01-07 13:38:38 -0800823 rcu_assign_pointer(tun->tfiles[tun->numqueues], tfile);
824 tun->numqueues++;
George Amanakis3a03cb82019-01-29 22:50:13 -0500825 tun_set_real_num_queues(tun);
Eric W. Biederman38231b72009-01-20 11:02:28 +0000826out:
Eric W. Biederman38231b72009-01-20 11:02:28 +0000827 return err;
Eric W. Biedermana7385ba2009-01-20 10:57:48 +0000828}
829
yuan linyu9484dc72017-09-23 22:36:52 +0800830static struct tun_struct *tun_get(struct tun_file *tfile)
Eric W. Biederman631ab462009-01-20 11:00:40 +0000831{
Jason Wang6e914fc2012-10-31 19:45:58 +0000832 struct tun_struct *tun;
Eric W. Biedermanc70f1822009-01-20 11:07:17 +0000833
Jason Wang6e914fc2012-10-31 19:45:58 +0000834 rcu_read_lock();
835 tun = rcu_dereference(tfile->tun);
836 if (tun)
837 dev_hold(tun->dev);
838 rcu_read_unlock();
Eric W. Biedermanc70f1822009-01-20 11:07:17 +0000839
840 return tun;
Eric W. Biederman631ab462009-01-20 11:00:40 +0000841}
842
Eric W. Biederman631ab462009-01-20 11:00:40 +0000843static void tun_put(struct tun_struct *tun)
844{
Jason Wang6e914fc2012-10-31 19:45:58 +0000845 dev_put(tun->dev);
Eric W. Biederman631ab462009-01-20 11:00:40 +0000846}
847
Joe Perches6b8a66e2011-03-02 07:18:10 +0000848/* TAP filtering */
Max Krasnyanskyf271b2cc2008-07-14 22:18:19 -0700849static void addr_hash_set(u32 *mask, const u8 *addr)
850{
851 int n = ether_crc(ETH_ALEN, addr) >> 26;
852 mask[n >> 5] |= (1 << (n & 31));
853}
854
855static unsigned int addr_hash_test(const u32 *mask, const u8 *addr)
856{
857 int n = ether_crc(ETH_ALEN, addr) >> 26;
858 return mask[n >> 5] & (1 << (n & 31));
859}
860
861static int update_filter(struct tap_filter *filter, void __user *arg)
862{
863 struct { u8 u[ETH_ALEN]; } *addr;
864 struct tun_filter uf;
865 int err, alen, n, nexact;
866
867 if (copy_from_user(&uf, arg, sizeof(uf)))
868 return -EFAULT;
869
870 if (!uf.count) {
871 /* Disabled */
872 filter->count = 0;
873 return 0;
874 }
875
876 alen = ETH_ALEN * uf.count;
Markus Elfring28e81902016-08-20 08:54:15 +0200877 addr = memdup_user(arg + sizeof(uf), alen);
878 if (IS_ERR(addr))
879 return PTR_ERR(addr);
Max Krasnyanskyf271b2cc2008-07-14 22:18:19 -0700880
881 /* The filter is updated without holding any locks. Which is
882 * perfectly safe. We disable it first and in the worst
883 * case we'll accept a few undesired packets. */
884 filter->count = 0;
885 wmb();
886
887 /* Use first set of addresses as an exact filter */
888 for (n = 0; n < uf.count && n < FLT_EXACT_COUNT; n++)
889 memcpy(filter->addr[n], addr[n].u, ETH_ALEN);
890
891 nexact = n;
892
Alex Williamsoncfbf84f2009-02-08 17:49:17 -0800893 /* Remaining multicast addresses are hashed,
894 * unicast will leave the filter disabled. */
Max Krasnyanskyf271b2cc2008-07-14 22:18:19 -0700895 memset(filter->mask, 0, sizeof(filter->mask));
Alex Williamsoncfbf84f2009-02-08 17:49:17 -0800896 for (; n < uf.count; n++) {
897 if (!is_multicast_ether_addr(addr[n].u)) {
898 err = 0; /* no filter */
Markus Elfring3b8d2a62016-08-20 09:00:34 +0200899 goto free_addr;
Alex Williamsoncfbf84f2009-02-08 17:49:17 -0800900 }
Max Krasnyanskyf271b2cc2008-07-14 22:18:19 -0700901 addr_hash_set(filter->mask, addr[n].u);
Alex Williamsoncfbf84f2009-02-08 17:49:17 -0800902 }
Max Krasnyanskyf271b2cc2008-07-14 22:18:19 -0700903
904 /* For ALLMULTI just set the mask to all ones.
905 * This overrides the mask populated above. */
906 if ((uf.flags & TUN_FLT_ALLMULTI))
907 memset(filter->mask, ~0, sizeof(filter->mask));
908
909 /* Now enable the filter */
910 wmb();
911 filter->count = nexact;
912
913 /* Return the number of exact filters */
914 err = nexact;
Markus Elfring3b8d2a62016-08-20 09:00:34 +0200915free_addr:
Max Krasnyanskyf271b2cc2008-07-14 22:18:19 -0700916 kfree(addr);
917 return err;
918}
919
920/* Returns: 0 - drop, !=0 - accept */
921static int run_filter(struct tap_filter *filter, const struct sk_buff *skb)
922{
923 /* Cannot use eth_hdr(skb) here because skb_mac_hdr() is incorrect
924 * at this point. */
925 struct ethhdr *eh = (struct ethhdr *) skb->data;
926 int i;
927
928 /* Exact match */
929 for (i = 0; i < filter->count; i++)
Joe Perches2e42e472012-05-09 17:17:46 +0000930 if (ether_addr_equal(eh->h_dest, filter->addr[i]))
Max Krasnyanskyf271b2cc2008-07-14 22:18:19 -0700931 return 1;
932
933 /* Inexact match (multicast only) */
934 if (is_multicast_ether_addr(eh->h_dest))
935 return addr_hash_test(filter->mask, eh->h_dest);
936
937 return 0;
938}
939
940/*
941 * Checks whether the packet is accepted or not.
942 * Returns: 0 - drop, !=0 - accept
943 */
944static int check_filter(struct tap_filter *filter, const struct sk_buff *skb)
945{
946 if (!filter->count)
947 return 1;
948
949 return run_filter(filter, skb);
950}
951
Linus Torvalds1da177e2005-04-16 15:20:36 -0700952/* Network device part of the driver */
953
Linus Torvalds1da177e2005-04-16 15:20:36 -0700954static const struct ethtool_ops tun_ethtool_ops;
955
Eric W. Biedermanc70f1822009-01-20 11:07:17 +0000956/* Net device detach from fd. */
957static void tun_net_uninit(struct net_device *dev)
958{
Jason Wangc8d68e62012-10-31 19:46:00 +0000959 tun_detach_all(dev);
Eric W. Biedermanc70f1822009-01-20 11:07:17 +0000960}
961
Linus Torvalds1da177e2005-04-16 15:20:36 -0700962/* Net device open. */
963static int tun_net_open(struct net_device *dev)
964{
Jason Wangc8d68e62012-10-31 19:46:00 +0000965 netif_tx_start_all_queues(dev);
Hannes Frederic Sowab20e2d52017-03-13 00:00:26 +0100966
Linus Torvalds1da177e2005-04-16 15:20:36 -0700967 return 0;
968}
969
970/* Net device close. */
971static int tun_net_close(struct net_device *dev)
972{
Jason Wangc8d68e62012-10-31 19:46:00 +0000973 netif_tx_stop_all_queues(dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700974 return 0;
975}
976
977/* Net device start xmit */
Jason Wang96f84062017-12-04 17:31:23 +0800978static void tun_automq_xmit(struct tun_struct *tun, struct sk_buff *skb)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700979{
Jason Wang3df97ba2016-04-25 23:13:42 -0400980#ifdef CONFIG_RPS
Eric Dumazetdc053602019-03-22 08:56:38 -0700981 if (tun->numqueues == 1 && static_branch_unlikely(&rps_needed)) {
Tom Herbert9bc88932013-12-22 18:54:32 +0800982 /* Select queue was not called for the skbuff, so we extract the
983 * RPS hash and save it into the flow_table here.
984 */
Wang Li4b035272018-10-09 10:32:04 +0800985 struct tun_flow_entry *e;
Tom Herbert9bc88932013-12-22 18:54:32 +0800986 __u32 rxhash;
987
Jason Wangfeec0842017-06-06 14:09:49 +0800988 rxhash = __skb_get_hash_symmetric(skb);
Wang Li4b035272018-10-09 10:32:04 +0800989 e = tun_flow_find(&tun->flows[tun_hashfn(rxhash)], rxhash);
990 if (e)
991 tun_flow_save_rps_rxhash(e, rxhash);
Tom Herbert9bc88932013-12-22 18:54:32 +0800992 }
Jason Wang3df97ba2016-04-25 23:13:42 -0400993#endif
Jason Wang96f84062017-12-04 17:31:23 +0800994}
995
Jason Wangaff3d702018-01-16 16:31:02 +0800996static unsigned int run_ebpf_filter(struct tun_struct *tun,
997 struct sk_buff *skb,
998 int len)
999{
1000 struct tun_prog *prog = rcu_dereference(tun->filter_prog);
1001
1002 if (prog)
1003 len = bpf_prog_run_clear_cb(prog->prog, skb);
1004
1005 return len;
1006}
1007
Jason Wang96f84062017-12-04 17:31:23 +08001008/* Net device start xmit */
1009static netdev_tx_t tun_net_xmit(struct sk_buff *skb, struct net_device *dev)
1010{
1011 struct tun_struct *tun = netdev_priv(dev);
1012 int txq = skb->queue_mapping;
1013 struct tun_file *tfile;
Jason Wangaff3d702018-01-16 16:31:02 +08001014 int len = skb->len;
Jason Wang96f84062017-12-04 17:31:23 +08001015
1016 rcu_read_lock();
1017 tfile = rcu_dereference(tun->tfiles[txq]);
Jason Wang96f84062017-12-04 17:31:23 +08001018
1019 /* Drop packet if interface is not attached */
Jason Wang9871a9e2019-05-08 23:20:18 -04001020 if (!tfile)
Jason Wang96f84062017-12-04 17:31:23 +08001021 goto drop;
1022
1023 if (!rcu_dereference(tun->steering_prog))
1024 tun_automq_xmit(tun, skb);
Tom Herbert9bc88932013-12-22 18:54:32 +08001025
Michal Kubecek34241702020-03-04 17:24:14 +01001026 netif_info(tun, tx_queued, tun->dev, "%s %d\n", __func__, skb->len);
Jason Wang6e914fc2012-10-31 19:45:58 +00001027
Max Krasnyanskyf271b2cc2008-07-14 22:18:19 -07001028 /* Drop if the filter does not like it.
1029 * This is a noop if the filter is disabled.
1030 * Filter can be enabled only for the TAP devices. */
1031 if (!check_filter(&tun->txflt, skb))
1032 goto drop;
1033
Jason Wang54f968d2012-10-31 19:45:57 +00001034 if (tfile->socket.sk->sk_filter &&
1035 sk_filter(tfile->socket.sk, skb))
Michael S. Tsirkin99405162010-02-14 01:01:10 +00001036 goto drop;
1037
Jason Wangaff3d702018-01-16 16:31:02 +08001038 len = run_ebpf_filter(tun, skb, len);
Bjørn Mork81c89502018-04-17 22:46:38 +02001039 if (len == 0 || pskb_trim(skb, len))
Jason Wangaff3d702018-01-16 16:31:02 +08001040 goto drop;
1041
Willem de Bruijn1f8b9772017-08-03 16:29:41 -04001042 if (unlikely(skb_orphan_frags_rx(skb, GFP_ATOMIC)))
Jason Wang7bf66302013-09-05 17:54:00 +08001043 goto drop;
1044
Soheil Hassas Yeganeh7b996242016-08-23 18:22:33 -04001045 skb_tx_timestamp(skb);
Richard Cochraneda29772013-07-19 19:40:10 +02001046
Michael S. Tsirkin0110d6f2010-04-13 04:59:44 +00001047 /* Orphan the skb - required as we might hang on to it
Jason Wang7bf66302013-09-05 17:54:00 +08001048 * for indefinite time.
1049 */
Michael S. Tsirkin0110d6f2010-04-13 04:59:44 +00001050 skb_orphan(skb);
1051
Florian Westphal895b5c92019-09-29 20:54:03 +02001052 nf_reset_ct(skb);
Eric Dumazetf8af75f2013-03-06 11:02:37 +00001053
Jason Wang5990a302018-01-04 11:14:27 +08001054 if (ptr_ring_produce(&tfile->tx_ring, skb))
Jason Wang1576d982016-06-30 14:45:36 +08001055 goto drop;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001056
1057 /* Notify and wake up reader process */
Jason Wang54f968d2012-10-31 19:45:57 +00001058 if (tfile->flags & TUN_FASYNC)
1059 kill_fasync(&tfile->fasync, SIGIO, POLL_IN);
Xi Wang9e641bd2014-05-16 15:11:48 -07001060 tfile->socket.sk->sk_data_ready(tfile->socket.sk);
Jason Wang6e914fc2012-10-31 19:45:58 +00001061
1062 rcu_read_unlock();
Patrick McHardy6ed10652009-06-23 06:03:08 +00001063 return NETDEV_TX_OK;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001064
1065drop:
Heiner Kallweit497a5752020-11-07 21:50:56 +01001066 atomic_long_inc(&dev->tx_dropped);
Michael S. Tsirkin149d36f2012-11-01 09:16:32 +00001067 skb_tx_error(skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001068 kfree_skb(skb);
Jason Wang6e914fc2012-10-31 19:45:58 +00001069 rcu_read_unlock();
Jason Wangbaeabab2014-11-18 13:20:41 +08001070 return NET_XMIT_DROP;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001071}
1072
Max Krasnyanskyf271b2cc2008-07-14 22:18:19 -07001073static void tun_net_mclist(struct net_device *dev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001074{
Max Krasnyanskyf271b2cc2008-07-14 22:18:19 -07001075 /*
1076 * This callback is supposed to deal with mc filter in
1077 * _rx_ path and has nothing to do with the _tx_ path.
1078 * In rx path we always accept everything userspace gives us.
1079 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001080}
1081
Michał Mirosławc8f44af2011-11-15 15:29:55 +00001082static netdev_features_t tun_net_fix_features(struct net_device *dev,
1083 netdev_features_t features)
Michał Mirosław88255372011-04-19 06:13:10 +00001084{
1085 struct tun_struct *tun = netdev_priv(dev);
1086
1087 return (features & tun->set_features) | (features & ~TUN_USER_FEATURES);
1088}
Paolo Abenieaea34b2016-02-26 10:45:40 +01001089
1090static void tun_set_headroom(struct net_device *dev, int new_hr)
1091{
1092 struct tun_struct *tun = netdev_priv(dev);
1093
1094 if (new_hr < NET_SKB_PAD)
1095 new_hr = NET_SKB_PAD;
1096
1097 tun->align = new_hr;
1098}
1099
stephen hemmingerbc1f4472017-01-06 19:12:52 -08001100static void
Paolo Abeni608b9972016-04-13 10:52:20 +02001101tun_net_get_stats64(struct net_device *dev, struct rtnl_link_stats64 *stats)
1102{
Paolo Abeni608b9972016-04-13 10:52:20 +02001103 struct tun_struct *tun = netdev_priv(dev);
Paolo Abeni608b9972016-04-13 10:52:20 +02001104
Heiner Kallweit497a5752020-11-07 21:50:56 +01001105 dev_get_tstats64(dev, stats);
Paolo Abeni608b9972016-04-13 10:52:20 +02001106
Heiner Kallweit497a5752020-11-07 21:50:56 +01001107 stats->rx_frame_errors +=
1108 (unsigned long)atomic_long_read(&tun->rx_frame_errors);
Paolo Abeni608b9972016-04-13 10:52:20 +02001109}
1110
Jason Wang761876c2017-08-11 19:41:18 +08001111static int tun_xdp_set(struct net_device *dev, struct bpf_prog *prog,
1112 struct netlink_ext_ack *extack)
1113{
1114 struct tun_struct *tun = netdev_priv(dev);
Jason Wange4a2a302018-09-12 11:16:59 +08001115 struct tun_file *tfile;
Jason Wang761876c2017-08-11 19:41:18 +08001116 struct bpf_prog *old_prog;
Jason Wange4a2a302018-09-12 11:16:59 +08001117 int i;
Jason Wang761876c2017-08-11 19:41:18 +08001118
1119 old_prog = rtnl_dereference(tun->xdp_prog);
1120 rcu_assign_pointer(tun->xdp_prog, prog);
1121 if (old_prog)
1122 bpf_prog_put(old_prog);
1123
Jason Wange4a2a302018-09-12 11:16:59 +08001124 for (i = 0; i < tun->numqueues; i++) {
1125 tfile = rtnl_dereference(tun->tfiles[i]);
1126 if (prog)
1127 sock_set_flag(&tfile->sk, SOCK_XDP);
1128 else
1129 sock_reset_flag(&tfile->sk, SOCK_XDP);
1130 }
1131 list_for_each_entry(tfile, &tun->disabled, next) {
1132 if (prog)
1133 sock_set_flag(&tfile->sk, SOCK_XDP);
1134 else
1135 sock_reset_flag(&tfile->sk, SOCK_XDP);
1136 }
1137
Jason Wang761876c2017-08-11 19:41:18 +08001138 return 0;
1139}
1140
Jakub Kicinskif4e63522017-11-03 13:56:16 -07001141static int tun_xdp(struct net_device *dev, struct netdev_bpf *xdp)
Jason Wang761876c2017-08-11 19:41:18 +08001142{
1143 switch (xdp->command) {
1144 case XDP_SETUP_PROG:
1145 return tun_xdp_set(dev, xdp->prog, xdp->extack);
Jason Wang761876c2017-08-11 19:41:18 +08001146 default:
1147 return -EINVAL;
1148 }
1149}
1150
Nicolas Dichtel26d31922018-11-28 19:12:56 +01001151static int tun_net_change_carrier(struct net_device *dev, bool new_carrier)
1152{
1153 if (new_carrier) {
1154 struct tun_struct *tun = netdev_priv(dev);
1155
1156 if (!tun->numqueues)
1157 return -EPERM;
1158
1159 netif_carrier_on(dev);
1160 } else {
1161 netif_carrier_off(dev);
1162 }
1163 return 0;
1164}
1165
Stephen Hemminger758e43b2008-11-19 22:10:37 -08001166static const struct net_device_ops tun_netdev_ops = {
Eric W. Biedermanc70f1822009-01-20 11:07:17 +00001167 .ndo_uninit = tun_net_uninit,
Stephen Hemminger758e43b2008-11-19 22:10:37 -08001168 .ndo_open = tun_net_open,
1169 .ndo_stop = tun_net_close,
Stephen Hemminger00829822008-11-20 20:14:53 -08001170 .ndo_start_xmit = tun_net_xmit,
Michał Mirosław88255372011-04-19 06:13:10 +00001171 .ndo_fix_features = tun_net_fix_features,
Jason Wangc8d68e62012-10-31 19:46:00 +00001172 .ndo_select_queue = tun_select_queue,
Paolo Abenieaea34b2016-02-26 10:45:40 +01001173 .ndo_set_rx_headroom = tun_set_headroom,
Paolo Abeni608b9972016-04-13 10:52:20 +02001174 .ndo_get_stats64 = tun_net_get_stats64,
Nicolas Dichtel26d31922018-11-28 19:12:56 +01001175 .ndo_change_carrier = tun_net_change_carrier,
Stephen Hemminger758e43b2008-11-19 22:10:37 -08001176};
1177
Jesper Dangaard Brouer0c9d9172018-05-31 11:00:03 +02001178static void __tun_xdp_flush_tfile(struct tun_file *tfile)
1179{
1180 /* Notify and wake up reader process */
1181 if (tfile->flags & TUN_FASYNC)
1182 kill_fasync(&tfile->fasync, SIGIO, POLL_IN);
1183 tfile->socket.sk->sk_data_ready(tfile->socket.sk);
1184}
1185
Jesper Dangaard Brouer42b33462018-05-31 10:59:47 +02001186static int tun_xdp_xmit(struct net_device *dev, int n,
1187 struct xdp_frame **frames, u32 flags)
Jason Wangfc72d1d2018-01-04 11:14:28 +08001188{
1189 struct tun_struct *tun = netdev_priv(dev);
Jason Wangfc72d1d2018-01-04 11:14:28 +08001190 struct tun_file *tfile;
1191 u32 numqueues;
Lorenzo Bianconifdc13972021-03-08 12:06:58 +01001192 int nxmit = 0;
Jesper Dangaard Brouer735fc402018-05-24 16:46:12 +02001193 int i;
Jason Wangfc72d1d2018-01-04 11:14:28 +08001194
Jesper Dangaard Brouer0c9d9172018-05-31 11:00:03 +02001195 if (unlikely(flags & ~XDP_XMIT_FLAGS_MASK))
Jesper Dangaard Brouer42b33462018-05-31 10:59:47 +02001196 return -EINVAL;
1197
Jason Wangfc72d1d2018-01-04 11:14:28 +08001198 rcu_read_lock();
1199
Jason Wang9871a9e2019-05-08 23:20:18 -04001200resample:
Jason Wangfc72d1d2018-01-04 11:14:28 +08001201 numqueues = READ_ONCE(tun->numqueues);
1202 if (!numqueues) {
Jesper Dangaard Brouer735fc402018-05-24 16:46:12 +02001203 rcu_read_unlock();
1204 return -ENXIO; /* Caller will free/return all frames */
Jason Wangfc72d1d2018-01-04 11:14:28 +08001205 }
1206
1207 tfile = rcu_dereference(tun->tfiles[smp_processor_id() %
1208 numqueues]);
Jason Wang9871a9e2019-05-08 23:20:18 -04001209 if (unlikely(!tfile))
1210 goto resample;
Jason Wangfc72d1d2018-01-04 11:14:28 +08001211
Jesper Dangaard Brouer735fc402018-05-24 16:46:12 +02001212 spin_lock(&tfile->tx_ring.producer_lock);
1213 for (i = 0; i < n; i++) {
1214 struct xdp_frame *xdp = frames[i];
1215 /* Encode the XDP flag into lowest bit for consumer to differ
1216 * XDP buffer from sk_buff.
1217 */
1218 void *frame = tun_xdp_to_ptr(xdp);
1219
1220 if (__ptr_ring_produce(&tfile->tx_ring, frame)) {
Heiner Kallweit497a5752020-11-07 21:50:56 +01001221 atomic_long_inc(&dev->tx_dropped);
Lorenzo Bianconifdc13972021-03-08 12:06:58 +01001222 break;
Jesper Dangaard Brouer735fc402018-05-24 16:46:12 +02001223 }
Lorenzo Bianconifdc13972021-03-08 12:06:58 +01001224 nxmit++;
Jesper Dangaard Brouer735fc402018-05-24 16:46:12 +02001225 }
1226 spin_unlock(&tfile->tx_ring.producer_lock);
1227
Jesper Dangaard Brouer0c9d9172018-05-31 11:00:03 +02001228 if (flags & XDP_XMIT_FLUSH)
1229 __tun_xdp_flush_tfile(tfile);
1230
Jason Wangfc72d1d2018-01-04 11:14:28 +08001231 rcu_read_unlock();
Lorenzo Bianconifdc13972021-03-08 12:06:58 +01001232 return nxmit;
Jason Wangfc72d1d2018-01-04 11:14:28 +08001233}
1234
Jesper Dangaard Brouer44fa2db2018-04-17 16:46:37 +02001235static int tun_xdp_tx(struct net_device *dev, struct xdp_buff *xdp)
1236{
Lorenzo Bianconi1b698fa2020-05-28 22:47:29 +02001237 struct xdp_frame *frame = xdp_convert_buff_to_frame(xdp);
Lorenzo Bianconifdc13972021-03-08 12:06:58 +01001238 int nxmit;
Jesper Dangaard Brouer44fa2db2018-04-17 16:46:37 +02001239
1240 if (unlikely(!frame))
1241 return -EOVERFLOW;
1242
Lorenzo Bianconifdc13972021-03-08 12:06:58 +01001243 nxmit = tun_xdp_xmit(dev, 1, &frame, XDP_XMIT_FLUSH);
1244 if (!nxmit)
1245 xdp_return_frame_rx_napi(frame);
1246 return nxmit;
Jason Wangfc72d1d2018-01-04 11:14:28 +08001247}
1248
Stephen Hemminger758e43b2008-11-19 22:10:37 -08001249static const struct net_device_ops tap_netdev_ops = {
Eric W. Biedermanc70f1822009-01-20 11:07:17 +00001250 .ndo_uninit = tun_net_uninit,
Stephen Hemminger758e43b2008-11-19 22:10:37 -08001251 .ndo_open = tun_net_open,
1252 .ndo_stop = tun_net_close,
Stephen Hemminger00829822008-11-20 20:14:53 -08001253 .ndo_start_xmit = tun_net_xmit,
Michał Mirosław88255372011-04-19 06:13:10 +00001254 .ndo_fix_features = tun_net_fix_features,
Jiri Pirkoafc4b132011-08-16 06:29:01 +00001255 .ndo_set_rx_mode = tun_net_mclist,
Stephen Hemminger758e43b2008-11-19 22:10:37 -08001256 .ndo_set_mac_address = eth_mac_addr,
1257 .ndo_validate_addr = eth_validate_addr,
Jason Wangc8d68e62012-10-31 19:46:00 +00001258 .ndo_select_queue = tun_select_queue,
Toshiaki Makita5e527962015-07-31 15:03:27 +09001259 .ndo_features_check = passthru_features_check,
Paolo Abenieaea34b2016-02-26 10:45:40 +01001260 .ndo_set_rx_headroom = tun_set_headroom,
Heiner Kallweit497a5752020-11-07 21:50:56 +01001261 .ndo_get_stats64 = dev_get_tstats64,
Jakub Kicinskif4e63522017-11-03 13:56:16 -07001262 .ndo_bpf = tun_xdp,
Jason Wangfc72d1d2018-01-04 11:14:28 +08001263 .ndo_xdp_xmit = tun_xdp_xmit,
Nicolas Dichtel26d31922018-11-28 19:12:56 +01001264 .ndo_change_carrier = tun_net_change_carrier,
Stephen Hemminger758e43b2008-11-19 22:10:37 -08001265};
1266
Pavel Emelyanov944a1372013-06-11 17:01:08 +04001267static void tun_flow_init(struct tun_struct *tun)
Jason Wang96442e422012-10-31 19:46:02 +00001268{
1269 int i;
1270
Jason Wang96442e422012-10-31 19:46:02 +00001271 for (i = 0; i < TUN_NUM_FLOW_ENTRIES; i++)
1272 INIT_HLIST_HEAD(&tun->flows[i]);
1273
1274 tun->ageing_time = TUN_FLOW_EXPIRE;
Kees Cooke99e88a2017-10-16 14:43:17 -07001275 timer_setup(&tun->flow_gc_timer, tun_flow_cleanup, 0);
1276 mod_timer(&tun->flow_gc_timer,
1277 round_jiffies_up(jiffies + tun->ageing_time));
Jason Wang96442e422012-10-31 19:46:02 +00001278}
1279
1280static void tun_flow_uninit(struct tun_struct *tun)
1281{
1282 del_timer_sync(&tun->flow_gc_timer);
1283 tun_flow_flush(tun);
Jason Wang96442e422012-10-31 19:46:02 +00001284}
1285
Jarod Wilson91572082016-10-20 13:55:20 -04001286#define MIN_MTU 68
1287#define MAX_MTU 65535
1288
Linus Torvalds1da177e2005-04-16 15:20:36 -07001289/* Initialize net device. */
1290static void tun_net_init(struct net_device *dev)
1291{
1292 struct tun_struct *tun = netdev_priv(dev);
Jeff Garzik6aa20a22006-09-13 13:24:59 -04001293
Linus Torvalds1da177e2005-04-16 15:20:36 -07001294 switch (tun->flags & TUN_TYPE_MASK) {
Michael S. Tsirkin40630b82014-11-19 15:17:31 +02001295 case IFF_TUN:
Stephen Hemminger758e43b2008-11-19 22:10:37 -08001296 dev->netdev_ops = &tun_netdev_ops;
Jason A. Donenfeldb9815eb2020-06-29 19:06:22 -06001297 dev->header_ops = &ip_tunnel_header_ops;
Stephen Hemminger758e43b2008-11-19 22:10:37 -08001298
Linus Torvalds1da177e2005-04-16 15:20:36 -07001299 /* Point-to-Point TUN Device */
1300 dev->hard_header_len = 0;
1301 dev->addr_len = 0;
1302 dev->mtu = 1500;
1303
1304 /* Zero header length */
Jeff Garzik6aa20a22006-09-13 13:24:59 -04001305 dev->type = ARPHRD_NONE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001306 dev->flags = IFF_POINTOPOINT | IFF_NOARP | IFF_MULTICAST;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001307 break;
1308
Michael S. Tsirkin40630b82014-11-19 15:17:31 +02001309 case IFF_TAP:
Kusanagi Kouichi7a0a9602008-12-29 18:23:28 -08001310 dev->netdev_ops = &tap_netdev_ops;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001311 /* Ethernet TAP Device */
Max Krasnyanskyf271b2cc2008-07-14 22:18:19 -07001312 ether_setup(dev);
Neil Horman550fd082011-07-26 06:05:38 +00001313 dev->priv_flags &= ~IFF_TX_SKB_SHARING;
stephen hemmingera6768472012-12-10 15:16:00 +00001314 dev->priv_flags |= IFF_LIVE_ADDR_CHANGE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001315
Danny Kukawkaf2cedb62012-02-15 06:45:39 +00001316 eth_hw_addr_random(dev);
Brian Braunstein36226a82007-04-26 01:00:55 -07001317
Linus Torvalds1da177e2005-04-16 15:20:36 -07001318 break;
1319 }
Jarod Wilson91572082016-10-20 13:55:20 -04001320
1321 dev->min_mtu = MIN_MTU;
1322 dev->max_mtu = MAX_MTU - dev->hard_header_len;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001323}
1324
Jason Wang2f3ab622018-05-22 14:21:04 +08001325static bool tun_sock_writeable(struct tun_struct *tun, struct tun_file *tfile)
1326{
1327 struct sock *sk = tfile->socket.sk;
1328
1329 return (tun->dev->flags & IFF_UP) && sock_writeable(sk);
1330}
1331
Linus Torvalds1da177e2005-04-16 15:20:36 -07001332/* Character device part */
1333
1334/* Poll */
Al Viroafc9a422017-07-03 06:39:46 -04001335static __poll_t tun_chr_poll(struct file *file, poll_table *wait)
Jeff Garzik6aa20a22006-09-13 13:24:59 -04001336{
Eric W. Biedermanb2430de2009-01-20 11:03:21 +00001337 struct tun_file *tfile = file->private_data;
yuan linyu9484dc72017-09-23 22:36:52 +08001338 struct tun_struct *tun = tun_get(tfile);
Mariusz Kozlowski3c8a9c62009-07-05 19:48:35 +00001339 struct sock *sk;
Al Viroafc9a422017-07-03 06:39:46 -04001340 __poll_t mask = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001341
1342 if (!tun)
Linus Torvaldsa9a08842018-02-11 14:34:03 -08001343 return EPOLLERR;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001344
Jason Wang54f968d2012-10-31 19:45:57 +00001345 sk = tfile->socket.sk;
Mariusz Kozlowski3c8a9c62009-07-05 19:48:35 +00001346
Xi Wang9e641bd2014-05-16 15:11:48 -07001347 poll_wait(file, sk_sleep(sk), wait);
Jeff Garzik6aa20a22006-09-13 13:24:59 -04001348
Jason Wang5990a302018-01-04 11:14:27 +08001349 if (!ptr_ring_empty(&tfile->tx_ring))
Linus Torvaldsa9a08842018-02-11 14:34:03 -08001350 mask |= EPOLLIN | EPOLLRDNORM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001351
Jason Wang2f3ab622018-05-22 14:21:04 +08001352 /* Make sure SOCKWQ_ASYNC_NOSPACE is set if not writable to
1353 * guarantee EPOLLOUT to be raised by either here or
1354 * tun_sock_write_space(). Then process could get notification
1355 * after it writes to a down device and meets -EIO.
1356 */
1357 if (tun_sock_writeable(tun, tfile) ||
1358 (!test_and_set_bit(SOCKWQ_ASYNC_NOSPACE, &sk->sk_socket->flags) &&
1359 tun_sock_writeable(tun, tfile)))
Linus Torvaldsa9a08842018-02-11 14:34:03 -08001360 mask |= EPOLLOUT | EPOLLWRNORM;
Herbert Xu33dccbb2009-02-05 21:25:32 -08001361
Eric W. Biedermanc70f1822009-01-20 11:07:17 +00001362 if (tun->dev->reg_state != NETREG_REGISTERED)
Linus Torvaldsa9a08842018-02-11 14:34:03 -08001363 mask = EPOLLERR;
Eric W. Biedermanc70f1822009-01-20 11:07:17 +00001364
Eric W. Biederman631ab462009-01-20 11:00:40 +00001365 tun_put(tun);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001366 return mask;
1367}
1368
Petar Penkov90e33d42017-09-22 13:49:15 -07001369static struct sk_buff *tun_napi_alloc_frags(struct tun_file *tfile,
1370 size_t len,
1371 const struct iov_iter *it)
1372{
1373 struct sk_buff *skb;
1374 size_t linear;
1375 int err;
1376 int i;
1377
1378 if (it->nr_segs > MAX_SKB_FRAGS + 1)
Yunjian Wang950271d2020-12-25 10:52:16 +08001379 return ERR_PTR(-EMSGSIZE);
Petar Penkov90e33d42017-09-22 13:49:15 -07001380
1381 local_bh_disable();
1382 skb = napi_get_frags(&tfile->napi);
1383 local_bh_enable();
1384 if (!skb)
1385 return ERR_PTR(-ENOMEM);
1386
1387 linear = iov_iter_single_seg_count(it);
1388 err = __skb_grow(skb, linear);
1389 if (err)
1390 goto free;
1391
1392 skb->len = len;
1393 skb->data_len = len - linear;
1394 skb->truesize += skb->data_len;
1395
1396 for (i = 1; i < it->nr_segs; i++) {
1397 size_t fragsz = it->iov[i].iov_len;
Eric Dumazetaa6daac2018-11-18 07:37:33 -08001398 struct page *page;
1399 void *frag;
Petar Penkov90e33d42017-09-22 13:49:15 -07001400
1401 if (fragsz == 0 || fragsz > PAGE_SIZE) {
1402 err = -EINVAL;
1403 goto free;
1404 }
Eric Dumazetaa6daac2018-11-18 07:37:33 -08001405 frag = netdev_alloc_frag(fragsz);
1406 if (!frag) {
Petar Penkov90e33d42017-09-22 13:49:15 -07001407 err = -ENOMEM;
1408 goto free;
1409 }
Eric Dumazetaa6daac2018-11-18 07:37:33 -08001410 page = virt_to_head_page(frag);
1411 skb_fill_page_desc(skb, i - 1, page,
1412 frag - page_address(page), fragsz);
Petar Penkov90e33d42017-09-22 13:49:15 -07001413 }
1414
1415 return skb;
1416free:
1417 /* frees skb and all frags allocated with napi_alloc_frag() */
1418 napi_free_frags(&tfile->napi);
1419 return ERR_PTR(err);
1420}
1421
Rusty Russellf42157c2008-08-15 15:15:10 -07001422/* prepad is the amount to reserve at front. len is length after that.
1423 * linear is a hint as to how much to copy (usually headers). */
Jason Wang54f968d2012-10-31 19:45:57 +00001424static struct sk_buff *tun_alloc_skb(struct tun_file *tfile,
stephen hemminger6f7c1562011-06-08 14:33:08 +00001425 size_t prepad, size_t len,
1426 size_t linear, int noblock)
Rusty Russellf42157c2008-08-15 15:15:10 -07001427{
Jason Wang54f968d2012-10-31 19:45:57 +00001428 struct sock *sk = tfile->socket.sk;
Rusty Russellf42157c2008-08-15 15:15:10 -07001429 struct sk_buff *skb;
Herbert Xu33dccbb2009-02-05 21:25:32 -08001430 int err;
Rusty Russellf42157c2008-08-15 15:15:10 -07001431
1432 /* Under a page? Don't bother with paged skb. */
Herbert Xu0eca93b2009-04-14 02:09:43 -07001433 if (prepad + len < PAGE_SIZE || !linear)
Herbert Xu33dccbb2009-02-05 21:25:32 -08001434 linear = len;
Rusty Russellf42157c2008-08-15 15:15:10 -07001435
Herbert Xu33dccbb2009-02-05 21:25:32 -08001436 skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock,
Eric Dumazet28d64272013-08-08 14:38:47 -07001437 &err, 0);
Rusty Russellf42157c2008-08-15 15:15:10 -07001438 if (!skb)
Herbert Xu33dccbb2009-02-05 21:25:32 -08001439 return ERR_PTR(err);
Rusty Russellf42157c2008-08-15 15:15:10 -07001440
1441 skb_reserve(skb, prepad);
1442 skb_put(skb, linear);
Herbert Xu33dccbb2009-02-05 21:25:32 -08001443 skb->data_len = len - linear;
1444 skb->len += len - linear;
Rusty Russellf42157c2008-08-15 15:15:10 -07001445
1446 return skb;
1447}
1448
Jason Wang5503fce2017-01-18 15:02:03 +08001449static void tun_rx_batched(struct tun_struct *tun, struct tun_file *tfile,
1450 struct sk_buff *skb, int more)
1451{
1452 struct sk_buff_head *queue = &tfile->sk.sk_write_queue;
1453 struct sk_buff_head process_queue;
1454 u32 rx_batched = tun->rx_batched;
1455 bool rcv = false;
1456
1457 if (!rx_batched || (!more && skb_queue_empty(queue))) {
1458 local_bh_disable();
Matthew Cover8ebebcb2018-11-18 00:46:00 -07001459 skb_record_rx_queue(skb, tfile->queue_index);
Jason Wang5503fce2017-01-18 15:02:03 +08001460 netif_receive_skb(skb);
1461 local_bh_enable();
1462 return;
1463 }
1464
1465 spin_lock(&queue->lock);
1466 if (!more || skb_queue_len(queue) == rx_batched) {
1467 __skb_queue_head_init(&process_queue);
1468 skb_queue_splice_tail_init(queue, &process_queue);
1469 rcv = true;
1470 } else {
1471 __skb_queue_tail(queue, skb);
1472 }
1473 spin_unlock(&queue->lock);
1474
1475 if (rcv) {
1476 struct sk_buff *nskb;
1477
1478 local_bh_disable();
Matthew Cover8ebebcb2018-11-18 00:46:00 -07001479 while ((nskb = __skb_dequeue(&process_queue))) {
1480 skb_record_rx_queue(nskb, tfile->queue_index);
Jason Wang5503fce2017-01-18 15:02:03 +08001481 netif_receive_skb(nskb);
Matthew Cover8ebebcb2018-11-18 00:46:00 -07001482 }
1483 skb_record_rx_queue(skb, tfile->queue_index);
Jason Wang5503fce2017-01-18 15:02:03 +08001484 netif_receive_skb(skb);
1485 local_bh_enable();
1486 }
1487}
1488
Jason Wang66ccbc92017-08-11 19:41:16 +08001489static bool tun_can_build_skb(struct tun_struct *tun, struct tun_file *tfile,
1490 int len, int noblock, bool zerocopy)
1491{
1492 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
1493 return false;
1494
1495 if (tfile->socket.sk->sk_sndbuf != INT_MAX)
1496 return false;
1497
1498 if (!noblock)
1499 return false;
1500
1501 if (zerocopy)
1502 return false;
1503
1504 if (SKB_DATA_ALIGN(len + TUN_RX_PAD) +
1505 SKB_DATA_ALIGN(sizeof(struct skb_shared_info)) > PAGE_SIZE)
1506 return false;
1507
1508 return true;
1509}
1510
Alexis Bauvin4b663362019-07-23 16:23:01 +02001511static struct sk_buff *__tun_build_skb(struct tun_file *tfile,
1512 struct page_frag *alloc_frag, char *buf,
Jason Wang8ae1aff2018-09-12 11:17:04 +08001513 int buflen, int len, int pad)
Jason Wangac1f1f62018-09-12 11:17:03 +08001514{
1515 struct sk_buff *skb = build_skb(buf, buflen);
1516
1517 if (!skb)
1518 return ERR_PTR(-ENOMEM);
1519
Jason Wang8ae1aff2018-09-12 11:17:04 +08001520 skb_reserve(skb, pad);
Jason Wangac1f1f62018-09-12 11:17:03 +08001521 skb_put(skb, len);
Alexis Bauvin4b663362019-07-23 16:23:01 +02001522 skb_set_owner_w(skb, tfile->socket.sk);
Jason Wangac1f1f62018-09-12 11:17:03 +08001523
1524 get_page(alloc_frag->page);
1525 alloc_frag->offset += buflen;
1526
1527 return skb;
1528}
1529
Jason Wang8ae1aff2018-09-12 11:17:04 +08001530static int tun_xdp_act(struct tun_struct *tun, struct bpf_prog *xdp_prog,
1531 struct xdp_buff *xdp, u32 act)
1532{
1533 int err;
1534
1535 switch (act) {
1536 case XDP_REDIRECT:
1537 err = xdp_do_redirect(tun->dev, xdp, xdp_prog);
Jason Wang8ae1aff2018-09-12 11:17:04 +08001538 if (err)
1539 return err;
1540 break;
1541 case XDP_TX:
1542 err = tun_xdp_tx(tun->dev, xdp);
1543 if (err < 0)
1544 return err;
1545 break;
1546 case XDP_PASS:
1547 break;
1548 default:
1549 bpf_warn_invalid_xdp_action(act);
Gustavo A. R. Silvadf561f662020-08-23 17:36:59 -05001550 fallthrough;
Jason Wang8ae1aff2018-09-12 11:17:04 +08001551 case XDP_ABORTED:
1552 trace_xdp_exception(tun->dev, xdp_prog, act);
Gustavo A. R. Silvadf561f662020-08-23 17:36:59 -05001553 fallthrough;
Jason Wang8ae1aff2018-09-12 11:17:04 +08001554 case XDP_DROP:
Heiner Kallweit497a5752020-11-07 21:50:56 +01001555 atomic_long_inc(&tun->dev->rx_dropped);
Jason Wang8ae1aff2018-09-12 11:17:04 +08001556 break;
1557 }
1558
1559 return act;
1560}
1561
Jason Wang761876c2017-08-11 19:41:18 +08001562static struct sk_buff *tun_build_skb(struct tun_struct *tun,
1563 struct tun_file *tfile,
Jason Wang66ccbc92017-08-11 19:41:16 +08001564 struct iov_iter *from,
Jason Wang761876c2017-08-11 19:41:18 +08001565 struct virtio_net_hdr *hdr,
Jason Wang1cfe6e92017-09-04 11:36:09 +08001566 int len, int *skb_xdp)
Jason Wang66ccbc92017-08-11 19:41:16 +08001567{
Eric Dumazet0bbd7da2017-08-16 22:14:33 +08001568 struct page_frag *alloc_frag = &current->task_frag;
Jason Wang761876c2017-08-11 19:41:18 +08001569 struct bpf_prog *xdp_prog;
Jason Wang7df13212017-09-04 11:36:08 +08001570 int buflen = SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
Jason Wang66ccbc92017-08-11 19:41:16 +08001571 char *buf;
1572 size_t copied;
Jason Wang8ae1aff2018-09-12 11:17:04 +08001573 int pad = TUN_RX_PAD;
1574 int err = 0;
Jason Wang7df13212017-09-04 11:36:08 +08001575
1576 rcu_read_lock();
1577 xdp_prog = rcu_dereference(tun->xdp_prog);
1578 if (xdp_prog)
Jason Wang4f23aff2018-09-12 11:17:00 +08001579 pad += XDP_PACKET_HEADROOM;
Jason Wang7df13212017-09-04 11:36:08 +08001580 buflen += SKB_DATA_ALIGN(len + pad);
1581 rcu_read_unlock();
Jason Wang66ccbc92017-08-11 19:41:16 +08001582
Jason Wang63b9ab62017-10-27 11:05:44 +08001583 alloc_frag->offset = ALIGN((u64)alloc_frag->offset, SMP_CACHE_BYTES);
Jason Wang66ccbc92017-08-11 19:41:16 +08001584 if (unlikely(!skb_page_frag_refill(buflen, alloc_frag, GFP_KERNEL)))
1585 return ERR_PTR(-ENOMEM);
1586
1587 buf = (char *)page_address(alloc_frag->page) + alloc_frag->offset;
1588 copied = copy_page_from_iter(alloc_frag->page,
Jason Wang7df13212017-09-04 11:36:08 +08001589 alloc_frag->offset + pad,
Jason Wang66ccbc92017-08-11 19:41:16 +08001590 len, from);
1591 if (copied != len)
1592 return ERR_PTR(-EFAULT);
1593
Jason Wang7df13212017-09-04 11:36:08 +08001594 /* There's a small window that XDP may be set after the check
1595 * of xdp_prog above, this should be rare and for simplicity
1596 * we do XDP on skb in case the headroom is not enough.
1597 */
Jason Wangac1f1f62018-09-12 11:17:03 +08001598 if (hdr->gso_type || !xdp_prog) {
Jason Wang1cfe6e92017-09-04 11:36:09 +08001599 *skb_xdp = 1;
Alexis Bauvin4b663362019-07-23 16:23:01 +02001600 return __tun_build_skb(tfile, alloc_frag, buf, buflen, len,
1601 pad);
Jason Wangac1f1f62018-09-12 11:17:03 +08001602 }
1603
1604 *skb_xdp = 0;
Jason Wang66ccbc92017-08-11 19:41:16 +08001605
Toshiaki Makita6547e382018-05-28 19:37:49 +09001606 local_bh_disable();
Jason Wang761876c2017-08-11 19:41:18 +08001607 rcu_read_lock();
1608 xdp_prog = rcu_dereference(tun->xdp_prog);
Jason Wang8ae1aff2018-09-12 11:17:04 +08001609 if (xdp_prog) {
Jason Wang761876c2017-08-11 19:41:18 +08001610 struct xdp_buff xdp;
Jason Wang761876c2017-08-11 19:41:18 +08001611 u32 act;
1612
Lorenzo Bianconi43b51692020-12-22 22:09:28 +01001613 xdp_init_buff(&xdp, buflen, &tfile->xdp_rxq);
Lorenzo Bianconibe9df4a2020-12-22 22:09:29 +01001614 xdp_prepare_buff(&xdp, buf, pad, len, false);
Jason Wang761876c2017-08-11 19:41:18 +08001615
Jason Wang8ae1aff2018-09-12 11:17:04 +08001616 act = bpf_prog_run_xdp(xdp_prog, &xdp);
1617 if (act == XDP_REDIRECT || act == XDP_TX) {
Jason Wang761876c2017-08-11 19:41:18 +08001618 get_page(alloc_frag->page);
1619 alloc_frag->offset += buflen;
Jason Wang761876c2017-08-11 19:41:18 +08001620 }
Jason Wang8ae1aff2018-09-12 11:17:04 +08001621 err = tun_xdp_act(tun, xdp_prog, &xdp, act);
Will Deaconbee34892020-04-03 16:13:21 +01001622 if (err < 0) {
1623 if (act == XDP_REDIRECT || act == XDP_TX)
1624 put_page(alloc_frag->page);
1625 goto out;
1626 }
1627
Jason Wang1a097912018-09-12 11:17:05 +08001628 if (err == XDP_REDIRECT)
Toke Høiland-Jørgensen1d233882020-01-16 16:14:45 +01001629 xdp_do_flush();
Jason Wang8ae1aff2018-09-12 11:17:04 +08001630 if (err != XDP_PASS)
1631 goto out;
1632
1633 pad = xdp.data - xdp.data_hard_start;
1634 len = xdp.data_end - xdp.data;
Jason Wang761876c2017-08-11 19:41:18 +08001635 }
Jason Wang291aeb22018-09-12 11:17:01 +08001636 rcu_read_unlock();
1637 local_bh_enable();
Jason Wang761876c2017-08-11 19:41:18 +08001638
Alexis Bauvin4b663362019-07-23 16:23:01 +02001639 return __tun_build_skb(tfile, alloc_frag, buf, buflen, len, pad);
Jason Wang761876c2017-08-11 19:41:18 +08001640
Jason Wangf7053b62018-09-12 11:17:02 +08001641out:
Jason Wang761876c2017-08-11 19:41:18 +08001642 rcu_read_unlock();
Toshiaki Makita6547e382018-05-28 19:37:49 +09001643 local_bh_enable();
Jason Wang761876c2017-08-11 19:41:18 +08001644 return NULL;
Jason Wang66ccbc92017-08-11 19:41:16 +08001645}
1646
Linus Torvalds1da177e2005-04-16 15:20:36 -07001647/* Get packet from user space buffer */
Jason Wang54f968d2012-10-31 19:45:57 +00001648static ssize_t tun_get_user(struct tun_struct *tun, struct tun_file *tfile,
Al Virof5ff53b2014-06-19 15:36:49 -04001649 void *msg_control, struct iov_iter *from,
Jason Wang5503fce2017-01-18 15:02:03 +08001650 int noblock, bool more)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001651{
Harvey Harrison09640e632009-02-01 00:45:17 -08001652 struct tun_pi pi = { 0, cpu_to_be16(ETH_P_IP) };
Linus Torvalds1da177e2005-04-16 15:20:36 -07001653 struct sk_buff *skb;
Al Virof5ff53b2014-06-19 15:36:49 -04001654 size_t total_len = iov_iter_count(from);
Paolo Abenieaea34b2016-02-26 10:45:40 +01001655 size_t len = total_len, align = tun->align, linear;
Rusty Russellf43798c2008-07-03 03:48:02 -07001656 struct virtio_net_hdr gso = { 0 };
Jason Wang96f8d9e2013-11-13 14:00:39 +08001657 int good_linear;
Michael S. Tsirkin06908992012-07-20 09:23:23 +00001658 int copylen;
1659 bool zerocopy = false;
1660 int err;
Jason Wang96f84062017-12-04 17:31:23 +08001661 u32 rxhash = 0;
Jason Wang1cfe6e92017-09-04 11:36:09 +08001662 int skb_xdp = 1;
Eric Dumazetaf3fb242018-09-28 14:51:49 -07001663 bool frags = tun_napi_frags_enabled(tfile);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001664
Michael S. Tsirkin40630b82014-11-19 15:17:31 +02001665 if (!(tun->flags & IFF_NO_PI)) {
Dan Carpenter15718ea2013-08-15 15:52:57 +03001666 if (len < sizeof(pi))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001667 return -EINVAL;
Dan Carpenter15718ea2013-08-15 15:52:57 +03001668 len -= sizeof(pi);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001669
Al Virocbbd26b2016-11-01 22:09:04 -04001670 if (!copy_from_iter_full(&pi, sizeof(pi), from))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001671 return -EFAULT;
1672 }
1673
Michael S. Tsirkin40630b82014-11-19 15:17:31 +02001674 if (tun->flags & IFF_VNET_HDR) {
Willem de Bruijne1edab82017-02-03 18:20:48 -05001675 int vnet_hdr_sz = READ_ONCE(tun->vnet_hdr_sz);
1676
1677 if (len < vnet_hdr_sz)
Rusty Russellf43798c2008-07-03 03:48:02 -07001678 return -EINVAL;
Willem de Bruijne1edab82017-02-03 18:20:48 -05001679 len -= vnet_hdr_sz;
Rusty Russellf43798c2008-07-03 03:48:02 -07001680
Al Virocbbd26b2016-11-01 22:09:04 -04001681 if (!copy_from_iter_full(&gso, sizeof(gso), from))
Rusty Russellf43798c2008-07-03 03:48:02 -07001682 return -EFAULT;
1683
Herbert Xu49091222009-06-08 00:20:01 -07001684 if ((gso.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) &&
Michael S. Tsirkin56f0dcc2014-10-23 22:59:31 +03001685 tun16_to_cpu(tun, gso.csum_start) + tun16_to_cpu(tun, gso.csum_offset) + 2 > tun16_to_cpu(tun, gso.hdr_len))
1686 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 -07001687
Michael S. Tsirkin56f0dcc2014-10-23 22:59:31 +03001688 if (tun16_to_cpu(tun, gso.hdr_len) > len)
Rusty Russellf43798c2008-07-03 03:48:02 -07001689 return -EINVAL;
Willem de Bruijne1edab82017-02-03 18:20:48 -05001690 iov_iter_advance(from, vnet_hdr_sz - sizeof(gso));
Rusty Russellf43798c2008-07-03 03:48:02 -07001691 }
1692
Michael S. Tsirkin40630b82014-11-19 15:17:31 +02001693 if ((tun->flags & TUN_TYPE_MASK) == IFF_TAP) {
stephen hemmingera504b862011-06-08 14:33:07 +00001694 align += NET_IP_ALIGN;
Herbert Xu0eca93b2009-04-14 02:09:43 -07001695 if (unlikely(len < ETH_HLEN ||
Michael S. Tsirkin56f0dcc2014-10-23 22:59:31 +03001696 (gso.hdr_len && tun16_to_cpu(tun, gso.hdr_len) < ETH_HLEN)))
Rusty Russelle01bf1c2008-04-12 18:49:30 -07001697 return -EINVAL;
1698 }
Jeff Garzik6aa20a22006-09-13 13:24:59 -04001699
Jason Wang96f8d9e2013-11-13 14:00:39 +08001700 good_linear = SKB_MAX_HEAD(align);
1701
Jason Wang88529172013-07-18 10:55:15 +08001702 if (msg_control) {
Al Virof5ff53b2014-06-19 15:36:49 -04001703 struct iov_iter i = *from;
1704
Jason Wang88529172013-07-18 10:55:15 +08001705 /* There are 256 bytes to be copied in skb, so there is
1706 * enough room for skb expand head in case it is used.
Michael S. Tsirkin06908992012-07-20 09:23:23 +00001707 * The rest of the buffer is mapped from userspace.
1708 */
Michael S. Tsirkin56f0dcc2014-10-23 22:59:31 +03001709 copylen = gso.hdr_len ? tun16_to_cpu(tun, gso.hdr_len) : GOODCOPY_LEN;
Jason Wang96f8d9e2013-11-13 14:00:39 +08001710 if (copylen > good_linear)
1711 copylen = good_linear;
Jason Wang3dd5c332013-07-10 13:43:27 +08001712 linear = copylen;
Al Virof5ff53b2014-06-19 15:36:49 -04001713 iov_iter_advance(&i, copylen);
1714 if (iov_iter_npages(&i, INT_MAX) <= MAX_SKB_FRAGS)
Jason Wang88529172013-07-18 10:55:15 +08001715 zerocopy = true;
1716 }
1717
Petar Penkov90e33d42017-09-22 13:49:15 -07001718 if (!frags && tun_can_build_skb(tun, tfile, len, noblock, zerocopy)) {
Jason Wang1cfe6e92017-09-04 11:36:09 +08001719 /* For the packet that is not easy to be processed
1720 * (e.g gso or jumbo packet), we will do it at after
1721 * skb was created with generic XDP routine.
1722 */
1723 skb = tun_build_skb(tun, tfile, from, &gso, len, &skb_xdp);
Jason Wang66ccbc92017-08-11 19:41:16 +08001724 if (IS_ERR(skb)) {
Heiner Kallweit497a5752020-11-07 21:50:56 +01001725 atomic_long_inc(&tun->dev->rx_dropped);
Jason Wang66ccbc92017-08-11 19:41:16 +08001726 return PTR_ERR(skb);
1727 }
Jason Wang761876c2017-08-11 19:41:18 +08001728 if (!skb)
1729 return total_len;
Jason Wang66ccbc92017-08-11 19:41:16 +08001730 } else {
1731 if (!zerocopy) {
1732 copylen = len;
1733 if (tun16_to_cpu(tun, gso.hdr_len) > good_linear)
1734 linear = good_linear;
1735 else
1736 linear = tun16_to_cpu(tun, gso.hdr_len);
1737 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001738
Petar Penkov90e33d42017-09-22 13:49:15 -07001739 if (frags) {
1740 mutex_lock(&tfile->napi_mutex);
1741 skb = tun_napi_alloc_frags(tfile, copylen, from);
1742 /* tun_napi_alloc_frags() enforces a layout for the skb.
1743 * If zerocopy is enabled, then this layout will be
1744 * overwritten by zerocopy_sg_from_iter().
1745 */
1746 zerocopy = false;
1747 } else {
1748 skb = tun_alloc_skb(tfile, align, copylen, linear,
1749 noblock);
1750 }
1751
Jason Wang66ccbc92017-08-11 19:41:16 +08001752 if (IS_ERR(skb)) {
1753 if (PTR_ERR(skb) != -EAGAIN)
Heiner Kallweit497a5752020-11-07 21:50:56 +01001754 atomic_long_inc(&tun->dev->rx_dropped);
Petar Penkov90e33d42017-09-22 13:49:15 -07001755 if (frags)
1756 mutex_unlock(&tfile->napi_mutex);
Jason Wang66ccbc92017-08-11 19:41:16 +08001757 return PTR_ERR(skb);
1758 }
Michael S. Tsirkin06908992012-07-20 09:23:23 +00001759
Jason Wang66ccbc92017-08-11 19:41:16 +08001760 if (zerocopy)
1761 err = zerocopy_sg_from_iter(skb, from);
1762 else
1763 err = skb_copy_datagram_from_iter(skb, 0, from, len);
1764
1765 if (err) {
Eric Dumazet44771382019-03-14 20:19:47 -07001766 err = -EFAULT;
1767drop:
Heiner Kallweit497a5752020-11-07 21:50:56 +01001768 atomic_long_inc(&tun->dev->rx_dropped);
Jason Wang66ccbc92017-08-11 19:41:16 +08001769 kfree_skb(skb);
Petar Penkov90e33d42017-09-22 13:49:15 -07001770 if (frags) {
1771 tfile->napi.skb = NULL;
1772 mutex_unlock(&tfile->napi_mutex);
1773 }
1774
Eric Dumazet44771382019-03-14 20:19:47 -07001775 return err;
Jason Wang66ccbc92017-08-11 19:41:16 +08001776 }
Dave Jones8f227572006-03-11 18:49:13 -08001777 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001778
Jarno Rajahalme3e9e40e2016-11-18 15:40:38 -08001779 if (virtio_net_hdr_to_skb(skb, &gso, tun_is_little_endian(tun))) {
Heiner Kallweit497a5752020-11-07 21:50:56 +01001780 atomic_long_inc(&tun->rx_frame_errors);
Paolo Abenidf10db92016-06-14 00:00:04 +02001781 kfree_skb(skb);
Petar Penkov90e33d42017-09-22 13:49:15 -07001782 if (frags) {
1783 tfile->napi.skb = NULL;
1784 mutex_unlock(&tfile->napi_mutex);
1785 }
1786
Paolo Abenidf10db92016-06-14 00:00:04 +02001787 return -EINVAL;
1788 }
1789
Linus Torvalds1da177e2005-04-16 15:20:36 -07001790 switch (tun->flags & TUN_TYPE_MASK) {
Michael S. Tsirkin40630b82014-11-19 15:17:31 +02001791 case IFF_TUN:
1792 if (tun->flags & IFF_NO_PI) {
Alexander Potapenko2580c4c2017-09-28 11:32:37 +02001793 u8 ip_version = skb->len ? (skb->data[0] >> 4) : 0;
1794
1795 switch (ip_version) {
1796 case 4:
Ang Way Chuangf09f7ee2008-06-17 21:10:33 -07001797 pi.proto = htons(ETH_P_IP);
1798 break;
Alexander Potapenko2580c4c2017-09-28 11:32:37 +02001799 case 6:
Ang Way Chuangf09f7ee2008-06-17 21:10:33 -07001800 pi.proto = htons(ETH_P_IPV6);
1801 break;
1802 default:
Heiner Kallweit497a5752020-11-07 21:50:56 +01001803 atomic_long_inc(&tun->dev->rx_dropped);
Ang Way Chuangf09f7ee2008-06-17 21:10:33 -07001804 kfree_skb(skb);
1805 return -EINVAL;
1806 }
1807 }
1808
Arnaldo Carvalho de Melo459a98e2007-03-19 15:30:44 -07001809 skb_reset_mac_header(skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001810 skb->protocol = pi.proto;
Arnaldo Carvalho de Melo4c13eb62007-04-25 17:40:23 -07001811 skb->dev = tun->dev;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001812 break;
Michael S. Tsirkin40630b82014-11-19 15:17:31 +02001813 case IFF_TAP:
Willem de Bruijn96aa1b22020-05-30 15:41:31 -04001814 if (frags && !pskb_may_pull(skb, ETH_HLEN)) {
1815 err = -ENOMEM;
1816 goto drop;
1817 }
1818 skb->protocol = eth_type_trans(skb, tun->dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001819 break;
Joe Perches6403eab2011-06-03 11:51:20 +00001820 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001821
Michael S. Tsirkin06908992012-07-20 09:23:23 +00001822 /* copy skb_ubuf_info for callback when skb has no error */
1823 if (zerocopy) {
Jonathan Lemon9ee5e5a2021-01-06 14:18:40 -08001824 skb_zcopy_init(skb, msg_control);
Jason Wangaf1cc7a2016-11-30 13:17:51 +08001825 } else if (msg_control) {
1826 struct ubuf_info *uarg = msg_control;
Jonathan Lemon36177832021-01-06 14:18:34 -08001827 uarg->callback(NULL, uarg, false);
Michael S. Tsirkin06908992012-07-20 09:23:23 +00001828 }
1829
Vlad Yasevich72f65102015-02-03 16:36:16 -05001830 skb_reset_network_header(skb);
Maxim Mikityanskiyd2aa1252019-02-21 12:39:57 +00001831 skb_probe_transport_header(skb);
Gilberto Bertin3fe260e2020-04-10 18:20:59 +02001832 skb_record_rx_queue(skb, tfile->queue_index);
Jason Wang38502af2013-03-25 20:19:56 +00001833
Jason Wang1cfe6e92017-09-04 11:36:09 +08001834 if (skb_xdp) {
Jason Wang761876c2017-08-11 19:41:18 +08001835 struct bpf_prog *xdp_prog;
1836 int ret;
1837
Toshiaki Makita6547e382018-05-28 19:37:49 +09001838 local_bh_disable();
Jason Wang761876c2017-08-11 19:41:18 +08001839 rcu_read_lock();
1840 xdp_prog = rcu_dereference(tun->xdp_prog);
1841 if (xdp_prog) {
1842 ret = do_xdp_generic(xdp_prog, skb);
1843 if (ret != XDP_PASS) {
1844 rcu_read_unlock();
Toshiaki Makita6547e382018-05-28 19:37:49 +09001845 local_bh_enable();
Eric Dumazet1efba982020-01-22 09:07:35 -08001846 if (frags) {
1847 tfile->napi.skb = NULL;
1848 mutex_unlock(&tfile->napi_mutex);
1849 }
Jason Wang761876c2017-08-11 19:41:18 +08001850 return total_len;
1851 }
1852 }
1853 rcu_read_unlock();
Toshiaki Makita6547e382018-05-28 19:37:49 +09001854 local_bh_enable();
Jason Wang761876c2017-08-11 19:41:18 +08001855 }
1856
Paolo Abenicf1a1e02018-04-20 13:18:16 +02001857 /* Compute the costly rx hash only if needed for flow updates.
1858 * We may get a very small possibility of OOO during switching, not
1859 * worth to optimize.
1860 */
1861 if (!rcu_access_pointer(tun->steering_prog) && tun->numqueues > 1 &&
1862 !tfile->detached)
Jason Wang96f84062017-12-04 17:31:23 +08001863 rxhash = __skb_get_hash_symmetric(skb);
Petar Penkov94317092017-09-22 13:49:14 -07001864
Eric Dumazet44771382019-03-14 20:19:47 -07001865 rcu_read_lock();
1866 if (unlikely(!(tun->dev->flags & IFF_UP))) {
1867 err = -EIO;
Eric Dumazet9180bb42019-03-16 13:09:53 -07001868 rcu_read_unlock();
Eric Dumazet44771382019-03-14 20:19:47 -07001869 goto drop;
1870 }
1871
Petar Penkov90e33d42017-09-22 13:49:15 -07001872 if (frags) {
Willem de Bruijn96aa1b22020-05-30 15:41:31 -04001873 u32 headlen;
1874
Petar Penkov90e33d42017-09-22 13:49:15 -07001875 /* Exercise flow dissector code path. */
Willem de Bruijn96aa1b22020-05-30 15:41:31 -04001876 skb_push(skb, ETH_HLEN);
1877 headlen = eth_get_headlen(tun->dev, skb->data,
1878 skb_headlen(skb));
Petar Penkov90e33d42017-09-22 13:49:15 -07001879
Eric Dumazet010f2452017-10-17 10:07:44 -07001880 if (unlikely(headlen > skb_headlen(skb))) {
Heiner Kallweit497a5752020-11-07 21:50:56 +01001881 atomic_long_inc(&tun->dev->rx_dropped);
Petar Penkov90e33d42017-09-22 13:49:15 -07001882 napi_free_frags(&tfile->napi);
Eric Dumazet44771382019-03-14 20:19:47 -07001883 rcu_read_unlock();
Petar Penkov90e33d42017-09-22 13:49:15 -07001884 mutex_unlock(&tfile->napi_mutex);
1885 WARN_ON(1);
1886 return -ENOMEM;
1887 }
1888
1889 local_bh_disable();
1890 napi_gro_frags(&tfile->napi);
1891 local_bh_enable();
1892 mutex_unlock(&tfile->napi_mutex);
Eric Dumazetaec72f32017-10-18 12:12:09 -07001893 } else if (tfile->napi_enabled) {
Petar Penkov94317092017-09-22 13:49:14 -07001894 struct sk_buff_head *queue = &tfile->sk.sk_write_queue;
1895 int queue_len;
1896
1897 spin_lock_bh(&queue->lock);
1898 __skb_queue_tail(queue, skb);
1899 queue_len = skb_queue_len(queue);
1900 spin_unlock(&queue->lock);
1901
1902 if (!more || queue_len > NAPI_POLL_WEIGHT)
1903 napi_schedule(&tfile->napi);
1904
1905 local_bh_enable();
1906 } else if (!IS_ENABLED(CONFIG_4KSTACKS)) {
1907 tun_rx_batched(tun, tfile, skb, more);
1908 } else {
1909 netif_rx_ni(skb);
1910 }
Eric Dumazet44771382019-03-14 20:19:47 -07001911 rcu_read_unlock();
Jeff Garzik6aa20a22006-09-13 13:24:59 -04001912
Heiner Kallweit497a5752020-11-07 21:50:56 +01001913 preempt_disable();
1914 dev_sw_netstats_rx_add(tun->dev, len);
1915 preempt_enable();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001916
Jason Wang96f84062017-12-04 17:31:23 +08001917 if (rxhash)
1918 tun_flow_update(tun, rxhash, tfile);
1919
Michael S. Tsirkin06908992012-07-20 09:23:23 +00001920 return total_len;
Jeff Garzik6aa20a22006-09-13 13:24:59 -04001921}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001922
Al Virof5ff53b2014-06-19 15:36:49 -04001923static ssize_t tun_chr_write_iter(struct kiocb *iocb, struct iov_iter *from)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001924{
Herbert Xu33dccbb2009-02-05 21:25:32 -08001925 struct file *file = iocb->ki_filp;
Jason Wang54f968d2012-10-31 19:45:57 +00001926 struct tun_file *tfile = file->private_data;
yuan linyu9484dc72017-09-23 22:36:52 +08001927 struct tun_struct *tun = tun_get(tfile);
Eric W. Biederman631ab462009-01-20 11:00:40 +00001928 ssize_t result;
Jens Axboe5aac0392020-11-20 07:59:54 -07001929 int noblock = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001930
1931 if (!tun)
1932 return -EBADFD;
1933
Jens Axboe5aac0392020-11-20 07:59:54 -07001934 if ((file->f_flags & O_NONBLOCK) || (iocb->ki_flags & IOCB_NOWAIT))
1935 noblock = 1;
1936
1937 result = tun_get_user(tun, tfile, NULL, from, noblock, false);
Eric W. Biederman631ab462009-01-20 11:00:40 +00001938
1939 tun_put(tun);
1940 return result;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001941}
1942
Jason Wangfc72d1d2018-01-04 11:14:28 +08001943static ssize_t tun_put_user_xdp(struct tun_struct *tun,
1944 struct tun_file *tfile,
Jesper Dangaard Brouer1ffcbc82018-04-17 16:45:47 +02001945 struct xdp_frame *xdp_frame,
Jason Wangfc72d1d2018-01-04 11:14:28 +08001946 struct iov_iter *iter)
1947{
1948 int vnet_hdr_sz = 0;
Jesper Dangaard Brouer1ffcbc82018-04-17 16:45:47 +02001949 size_t size = xdp_frame->len;
Jason Wangfc72d1d2018-01-04 11:14:28 +08001950 size_t ret;
1951
1952 if (tun->flags & IFF_VNET_HDR) {
1953 struct virtio_net_hdr gso = { 0 };
1954
1955 vnet_hdr_sz = READ_ONCE(tun->vnet_hdr_sz);
1956 if (unlikely(iov_iter_count(iter) < vnet_hdr_sz))
1957 return -EINVAL;
1958 if (unlikely(copy_to_iter(&gso, sizeof(gso), iter) !=
1959 sizeof(gso)))
1960 return -EFAULT;
1961 iov_iter_advance(iter, vnet_hdr_sz - sizeof(gso));
1962 }
1963
Jesper Dangaard Brouer1ffcbc82018-04-17 16:45:47 +02001964 ret = copy_to_iter(xdp_frame->data, size, iter) + vnet_hdr_sz;
Jason Wangfc72d1d2018-01-04 11:14:28 +08001965
Heiner Kallweit497a5752020-11-07 21:50:56 +01001966 preempt_disable();
1967 dev_sw_netstats_tx_add(tun->dev, 1, ret);
1968 preempt_enable();
Jason Wangfc72d1d2018-01-04 11:14:28 +08001969
1970 return ret;
1971}
1972
Linus Torvalds1da177e2005-04-16 15:20:36 -07001973/* Put packet to the user space buffer */
stephen hemminger6f7c1562011-06-08 14:33:08 +00001974static ssize_t tun_put_user(struct tun_struct *tun,
Jason Wang54f968d2012-10-31 19:45:57 +00001975 struct tun_file *tfile,
stephen hemminger6f7c1562011-06-08 14:33:08 +00001976 struct sk_buff *skb,
Herbert Xue0b46d02014-11-07 21:22:23 +08001977 struct iov_iter *iter)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001978{
1979 struct tun_pi pi = { 0, skb->protocol };
Herbert Xue0b46d02014-11-07 21:22:23 +08001980 ssize_t total;
Jason Wang8c847d22014-11-13 16:54:14 +08001981 int vlan_offset = 0;
Herbert Xua8f9bfd2014-11-03 04:30:13 +08001982 int vlan_hlen = 0;
Herbert Xu2eb783c2014-11-03 04:30:14 +08001983 int vnet_hdr_sz = 0;
Herbert Xua8f9bfd2014-11-03 04:30:13 +08001984
Jiri Pirkodf8a39d2015-01-13 17:13:44 +01001985 if (skb_vlan_tag_present(skb))
Herbert Xua8f9bfd2014-11-03 04:30:13 +08001986 vlan_hlen = VLAN_HLEN;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001987
Michael S. Tsirkin40630b82014-11-19 15:17:31 +02001988 if (tun->flags & IFF_VNET_HDR)
Willem de Bruijne1edab82017-02-03 18:20:48 -05001989 vnet_hdr_sz = READ_ONCE(tun->vnet_hdr_sz);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001990
Herbert Xue0b46d02014-11-07 21:22:23 +08001991 total = skb->len + vlan_hlen + vnet_hdr_sz;
1992
Michael S. Tsirkin40630b82014-11-19 15:17:31 +02001993 if (!(tun->flags & IFF_NO_PI)) {
Herbert Xue0b46d02014-11-07 21:22:23 +08001994 if (iov_iter_count(iter) < sizeof(pi))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001995 return -EINVAL;
1996
Herbert Xue0b46d02014-11-07 21:22:23 +08001997 total += sizeof(pi);
1998 if (iov_iter_count(iter) < total) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001999 /* Packet will be striped */
2000 pi.flags |= TUN_PKT_STRIP;
2001 }
Jeff Garzik6aa20a22006-09-13 13:24:59 -04002002
Herbert Xue0b46d02014-11-07 21:22:23 +08002003 if (copy_to_iter(&pi, sizeof(pi), iter) != sizeof(pi))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002004 return -EFAULT;
Jeff Garzik6aa20a22006-09-13 13:24:59 -04002005 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002006
Herbert Xu2eb783c2014-11-03 04:30:14 +08002007 if (vnet_hdr_sz) {
Jarno Rajahalme9403cd72016-11-18 15:40:40 -08002008 struct virtio_net_hdr gso;
Mike Rapoport34166092016-06-08 16:09:20 +03002009
Herbert Xue0b46d02014-11-07 21:22:23 +08002010 if (iov_iter_count(iter) < vnet_hdr_sz)
Rusty Russellf43798c2008-07-03 03:48:02 -07002011 return -EINVAL;
2012
Jarno Rajahalme3e9e40e2016-11-18 15:40:38 -08002013 if (virtio_net_hdr_from_skb(skb, &gso,
Willem de Bruijnfd3a8862018-06-06 11:23:01 -04002014 tun_is_little_endian(tun), true,
2015 vlan_hlen)) {
Rusty Russellf43798c2008-07-03 03:48:02 -07002016 struct skb_shared_info *sinfo = skb_shinfo(skb);
Mike Rapoport34166092016-06-08 16:09:20 +03002017 pr_err("unexpected GSO type: "
2018 "0x%x, gso_size %d, hdr_len %d\n",
2019 sinfo->gso_type, tun16_to_cpu(tun, gso.gso_size),
2020 tun16_to_cpu(tun, gso.hdr_len));
2021 print_hex_dump(KERN_ERR, "tun: ",
2022 DUMP_PREFIX_NONE,
2023 16, 1, skb->head,
2024 min((int)tun16_to_cpu(tun, gso.hdr_len), 64), true);
2025 WARN_ON_ONCE(1);
2026 return -EINVAL;
2027 }
Rusty Russellf43798c2008-07-03 03:48:02 -07002028
Herbert Xue0b46d02014-11-07 21:22:23 +08002029 if (copy_to_iter(&gso, sizeof(gso), iter) != sizeof(gso))
Rusty Russellf43798c2008-07-03 03:48:02 -07002030 return -EFAULT;
Jason Wang8c847d22014-11-13 16:54:14 +08002031
2032 iov_iter_advance(iter, vnet_hdr_sz - sizeof(gso));
Rusty Russellf43798c2008-07-03 03:48:02 -07002033 }
2034
Herbert Xua8f9bfd2014-11-03 04:30:13 +08002035 if (vlan_hlen) {
Herbert Xue0b46d02014-11-07 21:22:23 +08002036 int ret;
Jason Wangaff3d702018-01-16 16:31:02 +08002037 struct veth veth;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002038
Jason Wang6680ec62013-07-25 13:00:33 +08002039 veth.h_vlan_proto = skb->vlan_proto;
Jiri Pirkodf8a39d2015-01-13 17:13:44 +01002040 veth.h_vlan_TCI = htons(skb_vlan_tag_get(skb));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002041
Jason Wang6680ec62013-07-25 13:00:33 +08002042 vlan_offset = offsetof(struct vlan_ethhdr, h_vlan_proto);
Jason Wang6680ec62013-07-25 13:00:33 +08002043
Herbert Xue0b46d02014-11-07 21:22:23 +08002044 ret = skb_copy_datagram_iter(skb, 0, iter, vlan_offset);
2045 if (ret || !iov_iter_count(iter))
Jason Wang6680ec62013-07-25 13:00:33 +08002046 goto done;
2047
Herbert Xue0b46d02014-11-07 21:22:23 +08002048 ret = copy_to_iter(&veth, sizeof(veth), iter);
2049 if (ret != sizeof(veth) || !iov_iter_count(iter))
Jason Wang6680ec62013-07-25 13:00:33 +08002050 goto done;
2051 }
2052
Herbert Xue0b46d02014-11-07 21:22:23 +08002053 skb_copy_datagram_iter(skb, vlan_offset, iter, skb->len - vlan_offset);
Jason Wang6680ec62013-07-25 13:00:33 +08002054
2055done:
Paolo Abeni608b9972016-04-13 10:52:20 +02002056 /* caller is in process context, */
Heiner Kallweit497a5752020-11-07 21:50:56 +01002057 preempt_disable();
2058 dev_sw_netstats_tx_add(tun->dev, 1, skb->len + vlan_hlen);
2059 preempt_enable();
Linus Torvalds1da177e2005-04-16 15:20:36 -07002060
2061 return total;
2062}
2063
Jason Wangfc72d1d2018-01-04 11:14:28 +08002064static void *tun_ring_recv(struct tun_file *tfile, int noblock, int *err)
Jason Wang1576d982016-06-30 14:45:36 +08002065{
2066 DECLARE_WAITQUEUE(wait, current);
Jason Wangfc72d1d2018-01-04 11:14:28 +08002067 void *ptr = NULL;
Jason Wangf48cc6b2016-07-04 13:53:38 +08002068 int error = 0;
Jason Wang1576d982016-06-30 14:45:36 +08002069
Jason Wangfc72d1d2018-01-04 11:14:28 +08002070 ptr = ptr_ring_consume(&tfile->tx_ring);
2071 if (ptr)
Jason Wang1576d982016-06-30 14:45:36 +08002072 goto out;
2073 if (noblock) {
Jason Wangf48cc6b2016-07-04 13:53:38 +08002074 error = -EAGAIN;
Jason Wang1576d982016-06-30 14:45:36 +08002075 goto out;
2076 }
2077
Al Viro333f7902019-07-05 20:14:16 +01002078 add_wait_queue(&tfile->socket.wq.wait, &wait);
Jason Wang1576d982016-06-30 14:45:36 +08002079
2080 while (1) {
Timur Celik71828b22019-02-23 12:53:13 +01002081 set_current_state(TASK_INTERRUPTIBLE);
Jason Wangfc72d1d2018-01-04 11:14:28 +08002082 ptr = ptr_ring_consume(&tfile->tx_ring);
2083 if (ptr)
Jason Wang1576d982016-06-30 14:45:36 +08002084 break;
2085 if (signal_pending(current)) {
Jason Wangf48cc6b2016-07-04 13:53:38 +08002086 error = -ERESTARTSYS;
Jason Wang1576d982016-06-30 14:45:36 +08002087 break;
2088 }
2089 if (tfile->socket.sk->sk_shutdown & RCV_SHUTDOWN) {
Jason Wangf48cc6b2016-07-04 13:53:38 +08002090 error = -EFAULT;
Jason Wang1576d982016-06-30 14:45:36 +08002091 break;
2092 }
2093
2094 schedule();
2095 }
2096
Timur Celikecef67c2019-02-25 21:13:13 +01002097 __set_current_state(TASK_RUNNING);
Al Viro333f7902019-07-05 20:14:16 +01002098 remove_wait_queue(&tfile->socket.wq.wait, &wait);
Jason Wang1576d982016-06-30 14:45:36 +08002099
2100out:
Jason Wangf48cc6b2016-07-04 13:53:38 +08002101 *err = error;
Jason Wangfc72d1d2018-01-04 11:14:28 +08002102 return ptr;
Jason Wang1576d982016-06-30 14:45:36 +08002103}
2104
Jason Wang54f968d2012-10-31 19:45:57 +00002105static ssize_t tun_do_read(struct tun_struct *tun, struct tun_file *tfile,
Al Viro9b067032014-11-07 13:52:07 -05002106 struct iov_iter *to,
Jason Wangfc72d1d2018-01-04 11:14:28 +08002107 int noblock, void *ptr)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002108{
Al Viro9b067032014-11-07 13:52:07 -05002109 ssize_t ret;
Jason Wang1576d982016-06-30 14:45:36 +08002110 int err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002111
Wei Xuc33ee152017-12-01 05:10:37 -05002112 if (!iov_iter_count(to)) {
Jason Wangfc72d1d2018-01-04 11:14:28 +08002113 tun_ptr_free(ptr);
Al Viro9b067032014-11-07 13:52:07 -05002114 return 0;
Wei Xuc33ee152017-12-01 05:10:37 -05002115 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002116
Jason Wangfc72d1d2018-01-04 11:14:28 +08002117 if (!ptr) {
Jason Wangac77cfd2017-05-17 12:14:43 +08002118 /* Read frames from ring */
Jason Wangfc72d1d2018-01-04 11:14:28 +08002119 ptr = tun_ring_recv(tfile, noblock, &err);
2120 if (!ptr)
Jason Wangac77cfd2017-05-17 12:14:43 +08002121 return err;
2122 }
Herbert Xue0b46d02014-11-07 21:22:23 +08002123
Jesper Dangaard Brouer1ffcbc82018-04-17 16:45:47 +02002124 if (tun_is_xdp_frame(ptr)) {
2125 struct xdp_frame *xdpf = tun_ptr_to_xdp(ptr);
Jason Wangfc72d1d2018-01-04 11:14:28 +08002126
Jesper Dangaard Brouer1ffcbc82018-04-17 16:45:47 +02002127 ret = tun_put_user_xdp(tun, tfile, xdpf, to);
Jesper Dangaard Brouer03993092018-04-17 16:46:32 +02002128 xdp_return_frame(xdpf);
Jason Wangfc72d1d2018-01-04 11:14:28 +08002129 } else {
2130 struct sk_buff *skb = ptr;
2131
2132 ret = tun_put_user(tun, tfile, skb, to);
2133 if (unlikely(ret < 0))
2134 kfree_skb(skb);
2135 else
2136 consume_skb(skb);
2137 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002138
Michael S. Tsirkin05c28282010-01-14 06:17:09 +00002139 return ret;
2140}
2141
Al Viro9b067032014-11-07 13:52:07 -05002142static ssize_t tun_chr_read_iter(struct kiocb *iocb, struct iov_iter *to)
Michael S. Tsirkin05c28282010-01-14 06:17:09 +00002143{
2144 struct file *file = iocb->ki_filp;
2145 struct tun_file *tfile = file->private_data;
yuan linyu9484dc72017-09-23 22:36:52 +08002146 struct tun_struct *tun = tun_get(tfile);
Al Viro9b067032014-11-07 13:52:07 -05002147 ssize_t len = iov_iter_count(to), ret;
Jens Axboe5aac0392020-11-20 07:59:54 -07002148 int noblock = 0;
Michael S. Tsirkin05c28282010-01-14 06:17:09 +00002149
2150 if (!tun)
2151 return -EBADFD;
Jens Axboe5aac0392020-11-20 07:59:54 -07002152
2153 if ((file->f_flags & O_NONBLOCK) || (iocb->ki_flags & IOCB_NOWAIT))
2154 noblock = 1;
2155
2156 ret = tun_do_read(tun, tfile, to, noblock, NULL);
David S. Miller42404c02013-12-10 22:05:45 -05002157 ret = min_t(ssize_t, ret, len);
Zhi Yong Wud0b7da8a2013-12-06 14:16:51 +08002158 if (ret > 0)
2159 iocb->ki_pos = ret;
Eric W. Biederman631ab462009-01-20 11:00:40 +00002160 tun_put(tun);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002161 return ret;
2162}
2163
Jason Wangcd5681d2018-01-16 16:31:01 +08002164static void tun_prog_free(struct rcu_head *rcu)
Jason Wang96f84062017-12-04 17:31:23 +08002165{
Jason Wangcd5681d2018-01-16 16:31:01 +08002166 struct tun_prog *prog = container_of(rcu, struct tun_prog, rcu);
Jason Wang96f84062017-12-04 17:31:23 +08002167
2168 bpf_prog_destroy(prog->prog);
2169 kfree(prog);
2170}
2171
Jason Wang9d6474e2018-01-22 10:55:38 +08002172static int __tun_set_ebpf(struct tun_struct *tun,
2173 struct tun_prog __rcu **prog_p,
Jason Wangcd5681d2018-01-16 16:31:01 +08002174 struct bpf_prog *prog)
Jason Wang96f84062017-12-04 17:31:23 +08002175{
Jason Wangcd5681d2018-01-16 16:31:01 +08002176 struct tun_prog *old, *new = NULL;
Jason Wang96f84062017-12-04 17:31:23 +08002177
2178 if (prog) {
2179 new = kmalloc(sizeof(*new), GFP_KERNEL);
2180 if (!new)
2181 return -ENOMEM;
2182 new->prog = prog;
2183 }
2184
Jason Wang124da8f2017-12-08 12:02:30 +08002185 spin_lock_bh(&tun->lock);
Jason Wangcd5681d2018-01-16 16:31:01 +08002186 old = rcu_dereference_protected(*prog_p,
Jason Wang124da8f2017-12-08 12:02:30 +08002187 lockdep_is_held(&tun->lock));
Jason Wangcd5681d2018-01-16 16:31:01 +08002188 rcu_assign_pointer(*prog_p, new);
Jason Wang124da8f2017-12-08 12:02:30 +08002189 spin_unlock_bh(&tun->lock);
Jason Wang96f84062017-12-04 17:31:23 +08002190
2191 if (old)
Jason Wangcd5681d2018-01-16 16:31:01 +08002192 call_rcu(&old->rcu, tun_prog_free);
Jason Wang96f84062017-12-04 17:31:23 +08002193
2194 return 0;
2195}
2196
Jason Wang96442e422012-10-31 19:46:02 +00002197static void tun_free_netdev(struct net_device *dev)
2198{
2199 struct tun_struct *tun = netdev_priv(dev);
2200
Jason Wang4008e972012-12-13 23:53:30 +00002201 BUG_ON(!(list_empty(&tun->disabled)));
Eric Dumazet11fc7d52019-10-07 12:21:05 -07002202
Heiner Kallweit497a5752020-11-07 21:50:56 +01002203 free_percpu(dev->tstats);
2204 /* We clear tstats so that tun_set_iff() can tell if
Eric Dumazet11fc7d52019-10-07 12:21:05 -07002205 * tun_free_netdev() has been called from register_netdevice().
2206 */
Heiner Kallweit497a5752020-11-07 21:50:56 +01002207 dev->tstats = NULL;
Eric Dumazet11fc7d52019-10-07 12:21:05 -07002208
Jason Wang96442e422012-10-31 19:46:02 +00002209 tun_flow_uninit(tun);
Paul Moore5dbbaf22013-01-14 07:12:19 +00002210 security_tun_dev_free_security(tun->security);
Jason Wangcd5681d2018-01-16 16:31:01 +08002211 __tun_set_ebpf(tun, &tun->steering_prog, NULL);
Jason Wangaff3d702018-01-16 16:31:02 +08002212 __tun_set_ebpf(tun, &tun->filter_prog, NULL);
Jason Wang96442e422012-10-31 19:46:02 +00002213}
2214
Linus Torvalds1da177e2005-04-16 15:20:36 -07002215static void tun_setup(struct net_device *dev)
2216{
2217 struct tun_struct *tun = netdev_priv(dev);
2218
Eric W. Biederman0625c882012-02-07 16:48:55 -08002219 tun->owner = INVALID_UID;
2220 tun->group = INVALID_GID;
Chas Williams4e24f2d2018-06-02 17:49:53 -04002221 tun_default_link_ksettings(dev, &tun->link_ksettings);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002222
Linus Torvalds1da177e2005-04-16 15:20:36 -07002223 dev->ethtool_ops = &tun_ethtool_ops;
David S. Millercf124db2017-05-08 12:52:56 -04002224 dev->needs_free_netdev = true;
2225 dev->priv_destructor = tun_free_netdev;
Jason Wang016adb72016-04-08 13:26:48 +08002226 /* We prefer our own queue length */
2227 dev->tx_queue_len = TUN_READQ_SIZE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002228}
2229
Eric W. Biedermanf019a7a2009-01-21 16:02:16 -08002230/* Trivial set of netlink ops to allow deleting tun or tap
2231 * device with netlink.
2232 */
Matthias Schiffera8b8a8892017-06-25 23:56:01 +02002233static int tun_validate(struct nlattr *tb[], struct nlattr *data[],
2234 struct netlink_ext_ack *extack)
Eric W. Biedermanf019a7a2009-01-21 16:02:16 -08002235{
Nicolas Dichtel35b827b2018-11-29 14:45:39 +01002236 NL_SET_ERR_MSG(extack,
2237 "tun/tap creation via rtnetlink is not supported.");
2238 return -EOPNOTSUPP;
Eric W. Biedermanf019a7a2009-01-21 16:02:16 -08002239}
2240
Sabrina Dubroca1ec010e2018-02-16 11:03:07 +01002241static size_t tun_get_size(const struct net_device *dev)
2242{
2243 BUILD_BUG_ON(sizeof(u32) != sizeof(uid_t));
2244 BUILD_BUG_ON(sizeof(u32) != sizeof(gid_t));
2245
2246 return nla_total_size(sizeof(uid_t)) + /* OWNER */
2247 nla_total_size(sizeof(gid_t)) + /* GROUP */
2248 nla_total_size(sizeof(u8)) + /* TYPE */
2249 nla_total_size(sizeof(u8)) + /* PI */
2250 nla_total_size(sizeof(u8)) + /* VNET_HDR */
2251 nla_total_size(sizeof(u8)) + /* PERSIST */
2252 nla_total_size(sizeof(u8)) + /* MULTI_QUEUE */
2253 nla_total_size(sizeof(u32)) + /* NUM_QUEUES */
2254 nla_total_size(sizeof(u32)) + /* NUM_DISABLED_QUEUES */
2255 0;
2256}
2257
2258static int tun_fill_info(struct sk_buff *skb, const struct net_device *dev)
2259{
2260 struct tun_struct *tun = netdev_priv(dev);
2261
2262 if (nla_put_u8(skb, IFLA_TUN_TYPE, tun->flags & TUN_TYPE_MASK))
2263 goto nla_put_failure;
2264 if (uid_valid(tun->owner) &&
2265 nla_put_u32(skb, IFLA_TUN_OWNER,
2266 from_kuid_munged(current_user_ns(), tun->owner)))
2267 goto nla_put_failure;
2268 if (gid_valid(tun->group) &&
2269 nla_put_u32(skb, IFLA_TUN_GROUP,
2270 from_kgid_munged(current_user_ns(), tun->group)))
2271 goto nla_put_failure;
2272 if (nla_put_u8(skb, IFLA_TUN_PI, !(tun->flags & IFF_NO_PI)))
2273 goto nla_put_failure;
2274 if (nla_put_u8(skb, IFLA_TUN_VNET_HDR, !!(tun->flags & IFF_VNET_HDR)))
2275 goto nla_put_failure;
2276 if (nla_put_u8(skb, IFLA_TUN_PERSIST, !!(tun->flags & IFF_PERSIST)))
2277 goto nla_put_failure;
2278 if (nla_put_u8(skb, IFLA_TUN_MULTI_QUEUE,
2279 !!(tun->flags & IFF_MULTI_QUEUE)))
2280 goto nla_put_failure;
2281 if (tun->flags & IFF_MULTI_QUEUE) {
2282 if (nla_put_u32(skb, IFLA_TUN_NUM_QUEUES, tun->numqueues))
2283 goto nla_put_failure;
2284 if (nla_put_u32(skb, IFLA_TUN_NUM_DISABLED_QUEUES,
2285 tun->numdisabled))
2286 goto nla_put_failure;
2287 }
2288
2289 return 0;
2290
2291nla_put_failure:
2292 return -EMSGSIZE;
2293}
2294
Eric W. Biedermanf019a7a2009-01-21 16:02:16 -08002295static struct rtnl_link_ops tun_link_ops __read_mostly = {
2296 .kind = DRV_NAME,
2297 .priv_size = sizeof(struct tun_struct),
2298 .setup = tun_setup,
2299 .validate = tun_validate,
Sabrina Dubroca1ec010e2018-02-16 11:03:07 +01002300 .get_size = tun_get_size,
2301 .fill_info = tun_fill_info,
Eric W. Biedermanf019a7a2009-01-21 16:02:16 -08002302};
2303
Herbert Xu33dccbb2009-02-05 21:25:32 -08002304static void tun_sock_write_space(struct sock *sk)
2305{
Jason Wang54f968d2012-10-31 19:45:57 +00002306 struct tun_file *tfile;
Eric Dumazet43815482010-04-29 11:01:49 +00002307 wait_queue_head_t *wqueue;
Herbert Xu33dccbb2009-02-05 21:25:32 -08002308
2309 if (!sock_writeable(sk))
2310 return;
2311
Eric Dumazet9cd3e072015-11-29 20:03:10 -08002312 if (!test_and_clear_bit(SOCKWQ_ASYNC_NOSPACE, &sk->sk_socket->flags))
Herbert Xu33dccbb2009-02-05 21:25:32 -08002313 return;
2314
Eric Dumazet43815482010-04-29 11:01:49 +00002315 wqueue = sk_sleep(sk);
2316 if (wqueue && waitqueue_active(wqueue))
Linus Torvaldsa9a08842018-02-11 14:34:03 -08002317 wake_up_interruptible_sync_poll(wqueue, EPOLLOUT |
2318 EPOLLWRNORM | EPOLLWRBAND);
Herbert Xuc722c622009-06-03 21:45:55 -07002319
Jason Wang54f968d2012-10-31 19:45:57 +00002320 tfile = container_of(sk, struct tun_file, sk);
2321 kill_fasync(&tfile->fasync, SIGIO, POLL_OUT);
Herbert Xu33dccbb2009-02-05 21:25:32 -08002322}
2323
Jason Wangf9e06c42018-11-15 17:43:10 +08002324static void tun_put_page(struct tun_page *tpage)
2325{
2326 if (tpage->page)
2327 __page_frag_cache_drain(tpage->page, tpage->count);
2328}
2329
Jason Wang043d2222018-09-12 11:17:07 +08002330static int tun_xdp_one(struct tun_struct *tun,
2331 struct tun_file *tfile,
Jason Wangf9e06c42018-11-15 17:43:10 +08002332 struct xdp_buff *xdp, int *flush,
2333 struct tun_page *tpage)
Jason Wang043d2222018-09-12 11:17:07 +08002334{
Prashant Bhole4e4b08e2018-12-03 18:09:24 +09002335 unsigned int datasize = xdp->data_end - xdp->data;
Jason Wang043d2222018-09-12 11:17:07 +08002336 struct tun_xdp_hdr *hdr = xdp->data_hard_start;
2337 struct virtio_net_hdr *gso = &hdr->gso;
Jason Wang043d2222018-09-12 11:17:07 +08002338 struct bpf_prog *xdp_prog;
2339 struct sk_buff *skb = NULL;
2340 u32 rxhash = 0, act;
2341 int buflen = hdr->buflen;
2342 int err = 0;
2343 bool skb_xdp = false;
Jason Wangf9e06c42018-11-15 17:43:10 +08002344 struct page *page;
Jason Wang043d2222018-09-12 11:17:07 +08002345
2346 xdp_prog = rcu_dereference(tun->xdp_prog);
2347 if (xdp_prog) {
2348 if (gso->gso_type) {
2349 skb_xdp = true;
2350 goto build;
2351 }
Lorenzo Bianconi43b51692020-12-22 22:09:28 +01002352
2353 xdp_init_buff(xdp, buflen, &tfile->xdp_rxq);
Jason Wang043d2222018-09-12 11:17:07 +08002354 xdp_set_data_meta_invalid(xdp);
Jason Wang043d2222018-09-12 11:17:07 +08002355
2356 act = bpf_prog_run_xdp(xdp_prog, xdp);
2357 err = tun_xdp_act(tun, xdp_prog, xdp, act);
2358 if (err < 0) {
2359 put_page(virt_to_head_page(xdp->data));
2360 return err;
2361 }
2362
2363 switch (err) {
2364 case XDP_REDIRECT:
2365 *flush = true;
Gustavo A. R. Silvadf561f662020-08-23 17:36:59 -05002366 fallthrough;
Jason Wang043d2222018-09-12 11:17:07 +08002367 case XDP_TX:
2368 return 0;
2369 case XDP_PASS:
2370 break;
2371 default:
Jason Wangf9e06c42018-11-15 17:43:10 +08002372 page = virt_to_head_page(xdp->data);
2373 if (tpage->page == page) {
2374 ++tpage->count;
2375 } else {
2376 tun_put_page(tpage);
2377 tpage->page = page;
2378 tpage->count = 1;
2379 }
Jason Wang043d2222018-09-12 11:17:07 +08002380 return 0;
2381 }
2382 }
2383
2384build:
2385 skb = build_skb(xdp->data_hard_start, buflen);
2386 if (!skb) {
2387 err = -ENOMEM;
2388 goto out;
2389 }
2390
2391 skb_reserve(skb, xdp->data - xdp->data_hard_start);
2392 skb_put(skb, xdp->data_end - xdp->data);
2393
2394 if (virtio_net_hdr_to_skb(skb, gso, tun_is_little_endian(tun))) {
Heiner Kallweit497a5752020-11-07 21:50:56 +01002395 atomic_long_inc(&tun->rx_frame_errors);
Jason Wang043d2222018-09-12 11:17:07 +08002396 kfree_skb(skb);
2397 err = -EINVAL;
2398 goto out;
2399 }
2400
2401 skb->protocol = eth_type_trans(skb, tun->dev);
2402 skb_reset_network_header(skb);
Maxim Mikityanskiyd2aa1252019-02-21 12:39:57 +00002403 skb_probe_transport_header(skb);
Gilberto Bertin3fe260e2020-04-10 18:20:59 +02002404 skb_record_rx_queue(skb, tfile->queue_index);
Jason Wang043d2222018-09-12 11:17:07 +08002405
2406 if (skb_xdp) {
2407 err = do_xdp_generic(xdp_prog, skb);
2408 if (err != XDP_PASS)
2409 goto out;
2410 }
2411
Paolo Abenif29eb2a2018-11-07 10:34:36 +01002412 if (!rcu_dereference(tun->steering_prog) && tun->numqueues > 1 &&
2413 !tfile->detached)
Jason Wang043d2222018-09-12 11:17:07 +08002414 rxhash = __skb_get_hash_symmetric(skb);
2415
2416 netif_receive_skb(skb);
2417
Heiner Kallweit497a5752020-11-07 21:50:56 +01002418 /* No need to disable preemption here since this function is
Prashant Bhole6342ca62018-12-11 11:43:07 +09002419 * always called with bh disabled
2420 */
Heiner Kallweit497a5752020-11-07 21:50:56 +01002421 dev_sw_netstats_rx_add(tun->dev, datasize);
Jason Wang043d2222018-09-12 11:17:07 +08002422
2423 if (rxhash)
2424 tun_flow_update(tun, rxhash, tfile);
2425
2426out:
2427 return err;
2428}
2429
Ying Xue1b784142015-03-02 15:37:48 +08002430static int tun_sendmsg(struct socket *sock, struct msghdr *m, size_t total_len)
Michael S. Tsirkin05c28282010-01-14 06:17:09 +00002431{
Jason Wang043d2222018-09-12 11:17:07 +08002432 int ret, i;
Jason Wang54f968d2012-10-31 19:45:57 +00002433 struct tun_file *tfile = container_of(sock, struct tun_file, socket);
yuan linyu9484dc72017-09-23 22:36:52 +08002434 struct tun_struct *tun = tun_get(tfile);
Jason Wangfe8dd452018-09-12 11:17:06 +08002435 struct tun_msg_ctl *ctl = m->msg_control;
Jason Wang043d2222018-09-12 11:17:07 +08002436 struct xdp_buff *xdp;
Jason Wang54f968d2012-10-31 19:45:57 +00002437
2438 if (!tun)
2439 return -EBADFD;
Al Virof5ff53b2014-06-19 15:36:49 -04002440
Jason Wang043d2222018-09-12 11:17:07 +08002441 if (ctl && (ctl->type == TUN_MSG_PTR)) {
David S. Miller6f0271d2018-11-17 16:53:46 -08002442 struct tun_page tpage;
Jason Wang043d2222018-09-12 11:17:07 +08002443 int n = ctl->num;
2444 int flush = 0;
2445
David S. Miller6f0271d2018-11-17 16:53:46 -08002446 memset(&tpage, 0, sizeof(tpage));
2447
Jason Wang043d2222018-09-12 11:17:07 +08002448 local_bh_disable();
2449 rcu_read_lock();
2450
2451 for (i = 0; i < n; i++) {
2452 xdp = &((struct xdp_buff *)ctl->ptr)[i];
Jason Wangf9e06c42018-11-15 17:43:10 +08002453 tun_xdp_one(tun, tfile, xdp, &flush, &tpage);
Jason Wang043d2222018-09-12 11:17:07 +08002454 }
2455
2456 if (flush)
Toke Høiland-Jørgensen1d233882020-01-16 16:14:45 +01002457 xdp_do_flush();
Jason Wang043d2222018-09-12 11:17:07 +08002458
2459 rcu_read_unlock();
2460 local_bh_enable();
2461
Jason Wangf9e06c42018-11-15 17:43:10 +08002462 tun_put_page(&tpage);
2463
Jason Wang043d2222018-09-12 11:17:07 +08002464 ret = total_len;
2465 goto out;
2466 }
Jason Wangfe8dd452018-09-12 11:17:06 +08002467
2468 ret = tun_get_user(tun, tfile, ctl ? ctl->ptr : NULL, &m->msg_iter,
Jason Wang5503fce2017-01-18 15:02:03 +08002469 m->msg_flags & MSG_DONTWAIT,
2470 m->msg_flags & MSG_MORE);
Jason Wang043d2222018-09-12 11:17:07 +08002471out:
Jason Wang54f968d2012-10-31 19:45:57 +00002472 tun_put(tun);
2473 return ret;
Michael S. Tsirkin05c28282010-01-14 06:17:09 +00002474}
2475
Ying Xue1b784142015-03-02 15:37:48 +08002476static int tun_recvmsg(struct socket *sock, struct msghdr *m, size_t total_len,
Michael S. Tsirkin05c28282010-01-14 06:17:09 +00002477 int flags)
2478{
Jason Wang54f968d2012-10-31 19:45:57 +00002479 struct tun_file *tfile = container_of(sock, struct tun_file, socket);
yuan linyu9484dc72017-09-23 22:36:52 +08002480 struct tun_struct *tun = tun_get(tfile);
Jason Wangfc72d1d2018-01-04 11:14:28 +08002481 void *ptr = m->msg_control;
Michael S. Tsirkin05c28282010-01-14 06:17:09 +00002482 int ret;
Jason Wang54f968d2012-10-31 19:45:57 +00002483
Wei Xuc33ee152017-12-01 05:10:37 -05002484 if (!tun) {
2485 ret = -EBADFD;
Jason Wangfc72d1d2018-01-04 11:14:28 +08002486 goto out_free;
Wei Xuc33ee152017-12-01 05:10:37 -05002487 }
Jason Wang54f968d2012-10-31 19:45:57 +00002488
Richard Cochraneda29772013-07-19 19:40:10 +02002489 if (flags & ~(MSG_DONTWAIT|MSG_TRUNC|MSG_ERRQUEUE)) {
Gao feng3811ae72013-04-24 21:59:23 +00002490 ret = -EINVAL;
Wei Xuc33ee152017-12-01 05:10:37 -05002491 goto out_put_tun;
Gao feng3811ae72013-04-24 21:59:23 +00002492 }
Richard Cochraneda29772013-07-19 19:40:10 +02002493 if (flags & MSG_ERRQUEUE) {
2494 ret = sock_recv_errqueue(sock->sk, m, total_len,
2495 SOL_PACKET, TUN_TX_TIMESTAMP);
2496 goto out;
2497 }
Jason Wangfc72d1d2018-01-04 11:14:28 +08002498 ret = tun_do_read(tun, tfile, &m->msg_iter, flags & MSG_DONTWAIT, ptr);
Alex Gartrell87897932014-12-25 23:05:03 -08002499 if (ret > (ssize_t)total_len) {
David S. Miller42404c02013-12-10 22:05:45 -05002500 m->msg_flags |= MSG_TRUNC;
2501 ret = flags & MSG_TRUNC ? ret : total_len;
2502 }
Gao feng3811ae72013-04-24 21:59:23 +00002503out:
Jason Wang54f968d2012-10-31 19:45:57 +00002504 tun_put(tun);
Michael S. Tsirkin05c28282010-01-14 06:17:09 +00002505 return ret;
Wei Xuc33ee152017-12-01 05:10:37 -05002506
2507out_put_tun:
2508 tun_put(tun);
Jason Wangfc72d1d2018-01-04 11:14:28 +08002509out_free:
2510 tun_ptr_free(ptr);
Wei Xuc33ee152017-12-01 05:10:37 -05002511 return ret;
Michael S. Tsirkin05c28282010-01-14 06:17:09 +00002512}
2513
Jason Wangfc72d1d2018-01-04 11:14:28 +08002514static int tun_ptr_peek_len(void *ptr)
2515{
2516 if (likely(ptr)) {
Jesper Dangaard Brouer1ffcbc82018-04-17 16:45:47 +02002517 if (tun_is_xdp_frame(ptr)) {
2518 struct xdp_frame *xdpf = tun_ptr_to_xdp(ptr);
Jason Wangfc72d1d2018-01-04 11:14:28 +08002519
Jesper Dangaard Brouer1ffcbc82018-04-17 16:45:47 +02002520 return xdpf->len;
Jason Wangfc72d1d2018-01-04 11:14:28 +08002521 }
2522 return __skb_array_len_with_tag(ptr);
2523 } else {
2524 return 0;
2525 }
2526}
2527
Jason Wang1576d982016-06-30 14:45:36 +08002528static int tun_peek_len(struct socket *sock)
2529{
2530 struct tun_file *tfile = container_of(sock, struct tun_file, socket);
2531 struct tun_struct *tun;
2532 int ret = 0;
2533
yuan linyu9484dc72017-09-23 22:36:52 +08002534 tun = tun_get(tfile);
Jason Wang1576d982016-06-30 14:45:36 +08002535 if (!tun)
2536 return 0;
2537
Jason Wangfc72d1d2018-01-04 11:14:28 +08002538 ret = PTR_RING_PEEK_CALL(&tfile->tx_ring, tun_ptr_peek_len);
Jason Wang1576d982016-06-30 14:45:36 +08002539 tun_put(tun);
2540
2541 return ret;
2542}
2543
Michael S. Tsirkin05c28282010-01-14 06:17:09 +00002544/* Ops structure to mimic raw sockets with tun */
2545static const struct proto_ops tun_socket_ops = {
Jason Wang1576d982016-06-30 14:45:36 +08002546 .peek_len = tun_peek_len,
Michael S. Tsirkin05c28282010-01-14 06:17:09 +00002547 .sendmsg = tun_sendmsg,
2548 .recvmsg = tun_recvmsg,
2549};
2550
Herbert Xu33dccbb2009-02-05 21:25:32 -08002551static struct proto tun_proto = {
2552 .name = "tun",
2553 .owner = THIS_MODULE,
Jason Wang54f968d2012-10-31 19:45:57 +00002554 .obj_size = sizeof(struct tun_file),
Herbert Xu33dccbb2009-02-05 21:25:32 -08002555};
Eric W. Biedermanf019a7a2009-01-21 16:02:16 -08002556
David Woodhouse980c9e82009-05-09 22:54:21 -07002557static int tun_flags(struct tun_struct *tun)
2558{
Michael S. Tsirkin031f5e032014-11-19 14:44:40 +02002559 return tun->flags & (TUN_FEATURES | IFF_PERSIST | IFF_TUN | IFF_TAP);
David Woodhouse980c9e82009-05-09 22:54:21 -07002560}
2561
YueHaibingbc6d0762021-05-19 10:38:50 +08002562static ssize_t tun_flags_show(struct device *dev, struct device_attribute *attr,
David Woodhouse980c9e82009-05-09 22:54:21 -07002563 char *buf)
2564{
2565 struct tun_struct *tun = netdev_priv(to_net_dev(dev));
2566 return sprintf(buf, "0x%x\n", tun_flags(tun));
2567}
2568
YueHaibingbc6d0762021-05-19 10:38:50 +08002569static ssize_t owner_show(struct device *dev, struct device_attribute *attr,
2570 char *buf)
David Woodhouse980c9e82009-05-09 22:54:21 -07002571{
2572 struct tun_struct *tun = netdev_priv(to_net_dev(dev));
Eric W. Biederman0625c882012-02-07 16:48:55 -08002573 return uid_valid(tun->owner)?
2574 sprintf(buf, "%u\n",
2575 from_kuid_munged(current_user_ns(), tun->owner)):
2576 sprintf(buf, "-1\n");
David Woodhouse980c9e82009-05-09 22:54:21 -07002577}
2578
YueHaibingbc6d0762021-05-19 10:38:50 +08002579static ssize_t group_show(struct device *dev, struct device_attribute *attr,
2580 char *buf)
David Woodhouse980c9e82009-05-09 22:54:21 -07002581{
2582 struct tun_struct *tun = netdev_priv(to_net_dev(dev));
Eric W. Biederman0625c882012-02-07 16:48:55 -08002583 return gid_valid(tun->group) ?
2584 sprintf(buf, "%u\n",
2585 from_kgid_munged(current_user_ns(), tun->group)):
2586 sprintf(buf, "-1\n");
David Woodhouse980c9e82009-05-09 22:54:21 -07002587}
2588
YueHaibingbc6d0762021-05-19 10:38:50 +08002589static DEVICE_ATTR_RO(tun_flags);
2590static DEVICE_ATTR_RO(owner);
2591static DEVICE_ATTR_RO(group);
David Woodhouse980c9e82009-05-09 22:54:21 -07002592
Takashi Iwaic4d33e22015-02-04 14:37:34 +01002593static struct attribute *tun_dev_attrs[] = {
2594 &dev_attr_tun_flags.attr,
2595 &dev_attr_owner.attr,
2596 &dev_attr_group.attr,
2597 NULL
2598};
2599
2600static const struct attribute_group tun_attr_group = {
2601 .attrs = tun_dev_attrs
2602};
2603
Pavel Emelyanovd647a592008-04-16 00:41:16 -07002604static int tun_set_iff(struct net *net, struct file *file, struct ifreq *ifr)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002605{
2606 struct tun_struct *tun;
Jason Wang54f968d2012-10-31 19:45:57 +00002607 struct tun_file *tfile = file->private_data;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002608 struct net_device *dev;
2609 int err;
2610
Jason Wang7c0c3b12013-01-11 16:59:33 +00002611 if (tfile->detached)
2612 return -EINVAL;
2613
Petar Penkov90e33d42017-09-22 13:49:15 -07002614 if ((ifr->ifr_flags & IFF_NAPI_FRAGS)) {
2615 if (!capable(CAP_NET_ADMIN))
2616 return -EPERM;
2617
2618 if (!(ifr->ifr_flags & IFF_NAPI) ||
2619 (ifr->ifr_flags & TUN_TYPE_MASK) != IFF_TAP)
2620 return -EINVAL;
2621 }
2622
Eric W. Biederman74a3e5a2009-01-20 10:56:20 +00002623 dev = __dev_get_by_name(net, ifr->ifr_name);
2624 if (dev) {
David Woodhousef85ba782009-04-27 03:23:54 -07002625 if (ifr->ifr_flags & IFF_TUN_EXCL)
2626 return -EBUSY;
Eric W. Biederman74a3e5a2009-01-20 10:56:20 +00002627 if ((ifr->ifr_flags & IFF_TUN) && dev->netdev_ops == &tun_netdev_ops)
2628 tun = netdev_priv(dev);
2629 else if ((ifr->ifr_flags & IFF_TAP) && dev->netdev_ops == &tap_netdev_ops)
2630 tun = netdev_priv(dev);
2631 else
2632 return -EINVAL;
2633
Jason Wang8e6d91a2013-05-28 18:32:11 +00002634 if (!!(ifr->ifr_flags & IFF_MULTI_QUEUE) !=
Michael S. Tsirkin40630b82014-11-19 15:17:31 +02002635 !!(tun->flags & IFF_MULTI_QUEUE))
Jason Wang8e6d91a2013-05-28 18:32:11 +00002636 return -EINVAL;
2637
Jason Wangcde8b152012-10-31 19:46:01 +00002638 if (tun_not_capable(tun))
Paul Moore2b980db2009-08-28 18:12:43 -04002639 return -EPERM;
Paul Moore5dbbaf22013-01-14 07:12:19 +00002640 err = security_tun_dev_open(tun->security);
Paul Moore2b980db2009-08-28 18:12:43 -04002641 if (err < 0)
2642 return err;
2643
Petar Penkov94317092017-09-22 13:49:14 -07002644 err = tun_attach(tun, file, ifr->ifr_flags & IFF_NOFILTER,
Eric Dumazetaf3fb242018-09-28 14:51:49 -07002645 ifr->ifr_flags & IFF_NAPI,
Yang Yingliang77f22f92019-09-10 18:56:57 +08002646 ifr->ifr_flags & IFF_NAPI_FRAGS, true);
Eric W. Biedermana7385ba2009-01-20 10:57:48 +00002647 if (err < 0)
2648 return err;
Jason Wang4008e972012-12-13 23:53:30 +00002649
Michael S. Tsirkin40630b82014-11-19 15:17:31 +02002650 if (tun->flags & IFF_MULTI_QUEUE &&
Jason Wange8dbad62013-04-22 20:40:39 +00002651 (tun->numqueues + tun->numdisabled > 1)) {
2652 /* One or more queue has already been attached, no need
2653 * to initialize the device again.
2654 */
Sabrina Dubroca83c1f362018-04-10 16:28:56 +02002655 netdev_state_change(dev);
Jason Wange8dbad62013-04-22 20:40:39 +00002656 return 0;
2657 }
Sabrina Dubroca9fffc5c2018-04-10 16:28:55 +02002658
2659 tun->flags = (tun->flags & ~TUN_FEATURES) |
2660 (ifr->ifr_flags & TUN_FEATURES);
Sabrina Dubroca83c1f362018-04-10 16:28:56 +02002661
2662 netdev_state_change(dev);
2663 } else {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002664 char *name;
2665 unsigned long flags = 0;
Jason Wangedfb6a12013-01-23 03:59:12 +00002666 int queues = ifr->ifr_flags & IFF_MULTI_QUEUE ?
2667 MAX_TAP_QUEUES : 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002668
Eric W. Biedermanc260b772012-11-18 21:34:11 +00002669 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
David Woodhouseca6bb5d2006-06-22 16:07:52 -07002670 return -EPERM;
Paul Moore2b980db2009-08-28 18:12:43 -04002671 err = security_tun_dev_create();
2672 if (err < 0)
2673 return err;
David Woodhouseca6bb5d2006-06-22 16:07:52 -07002674
Linus Torvalds1da177e2005-04-16 15:20:36 -07002675 /* Set dev type */
2676 if (ifr->ifr_flags & IFF_TUN) {
2677 /* TUN device */
Michael S. Tsirkin40630b82014-11-19 15:17:31 +02002678 flags |= IFF_TUN;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002679 name = "tun%d";
2680 } else if (ifr->ifr_flags & IFF_TAP) {
2681 /* TAP device */
Michael S. Tsirkin40630b82014-11-19 15:17:31 +02002682 flags |= IFF_TAP;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002683 name = "tap%d";
Jeff Garzik6aa20a22006-09-13 13:24:59 -04002684 } else
Kusanagi Kouichi36989b92009-09-16 21:36:13 +00002685 return -EINVAL;
Jeff Garzik6aa20a22006-09-13 13:24:59 -04002686
Linus Torvalds1da177e2005-04-16 15:20:36 -07002687 if (*ifr->ifr_name)
2688 name = ifr->ifr_name;
2689
Jason Wangc8d68e62012-10-31 19:46:00 +00002690 dev = alloc_netdev_mqs(sizeof(struct tun_struct), name,
Tom Gundersenc835a672014-07-14 16:37:24 +02002691 NET_NAME_UNKNOWN, tun_setup, queues,
2692 queues);
Jason Wangedfb6a12013-01-23 03:59:12 +00002693
Linus Torvalds1da177e2005-04-16 15:20:36 -07002694 if (!dev)
2695 return -ENOMEM;
2696
Pavel Emelyanovfc54c652008-04-16 00:41:53 -07002697 dev_net_set(dev, net);
Eric W. Biedermanf019a7a2009-01-21 16:02:16 -08002698 dev->rtnl_link_ops = &tun_link_ops;
Pavel Emelyanovfb7589a2013-08-21 14:31:38 +04002699 dev->ifindex = tfile->ifindex;
Takashi Iwaic4d33e22015-02-04 14:37:34 +01002700 dev->sysfs_groups[0] = &tun_attr_group;
Stephen Hemminger758e43b2008-11-19 22:10:37 -08002701
Linus Torvalds1da177e2005-04-16 15:20:36 -07002702 tun = netdev_priv(dev);
2703 tun->dev = dev;
2704 tun->flags = flags;
Max Krasnyanskyf271b2cc2008-07-14 22:18:19 -07002705 tun->txflt.count = 0;
Michael S. Tsirkind9d52b52010-03-17 17:45:01 +02002706 tun->vnet_hdr_sz = sizeof(struct virtio_net_hdr);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002707
Paolo Abenieaea34b2016-02-26 10:45:40 +01002708 tun->align = NET_SKB_PAD;
Jason Wang54f968d2012-10-31 19:45:57 +00002709 tun->filter_attached = false;
2710 tun->sndbuf = tfile->socket.sk->sk_sndbuf;
Jason Wang5503fce2017-01-18 15:02:03 +08002711 tun->rx_batched = 0;
Jason Wang96f84062017-12-04 17:31:23 +08002712 RCU_INIT_POINTER(tun->steering_prog, NULL);
Herbert Xu33dccbb2009-02-05 21:25:32 -08002713
Heiner Kallweit497a5752020-11-07 21:50:56 +01002714 dev->tstats = netdev_alloc_pcpu_stats(struct pcpu_sw_netstats);
2715 if (!dev->tstats) {
Paolo Abeni608b9972016-04-13 10:52:20 +02002716 err = -ENOMEM;
2717 goto err_free_dev;
2718 }
2719
Jason Wang96442e422012-10-31 19:46:02 +00002720 spin_lock_init(&tun->lock);
2721
Paul Moore5dbbaf22013-01-14 07:12:19 +00002722 err = security_tun_dev_alloc_security(&tun->security);
2723 if (err < 0)
Paolo Abeni608b9972016-04-13 10:52:20 +02002724 goto err_free_stat;
Paul Moore2b980db2009-08-28 18:12:43 -04002725
Linus Torvalds1da177e2005-04-16 15:20:36 -07002726 tun_net_init(dev);
Pavel Emelyanov944a1372013-06-11 17:01:08 +04002727 tun_flow_init(tun);
Jason Wang96442e422012-10-31 19:46:02 +00002728
Michał Mirosław88255372011-04-19 06:13:10 +00002729 dev->hw_features = NETIF_F_SG | NETIF_F_FRAGLIST |
Jason Wang6680ec62013-07-25 13:00:33 +08002730 TUN_USER_FEATURES | NETIF_F_HW_VLAN_CTAG_TX |
2731 NETIF_F_HW_VLAN_STAG_TX;
Paolo Abeni2a2bbf12016-04-14 18:39:39 +02002732 dev->features = dev->hw_features | NETIF_F_LLTX;
Fernando Luis Vazquez Cao6671b222014-02-18 21:20:09 +09002733 dev->vlan_features = dev->features &
2734 ~(NETIF_F_HW_VLAN_CTAG_TX |
2735 NETIF_F_HW_VLAN_STAG_TX);
Michał Mirosław88255372011-04-19 06:13:10 +00002736
Sabrina Dubroca9fffc5c2018-04-10 16:28:55 +02002737 tun->flags = (tun->flags & ~TUN_FEATURES) |
2738 (ifr->ifr_flags & TUN_FEATURES);
2739
Jason Wang4008e972012-12-13 23:53:30 +00002740 INIT_LIST_HEAD(&tun->disabled);
Eric Dumazetaf3fb242018-09-28 14:51:49 -07002741 err = tun_attach(tun, file, false, ifr->ifr_flags & IFF_NAPI,
Yang Yingliang77f22f92019-09-10 18:56:57 +08002742 ifr->ifr_flags & IFF_NAPI_FRAGS, false);
Jason Wangeb0fb362012-12-02 17:19:45 +00002743 if (err < 0)
Jason Wang662ca432013-09-11 18:09:48 +08002744 goto err_free_flow;
Jason Wangeb0fb362012-12-02 17:19:45 +00002745
Linus Torvalds1da177e2005-04-16 15:20:36 -07002746 err = register_netdevice(tun->dev);
2747 if (err < 0)
Jason Wang662ca432013-09-11 18:09:48 +08002748 goto err_detach;
Menglong Dongc2e315b2021-01-18 03:15:39 -08002749 /* free_netdev() won't check refcnt, to avoid race
Yang Yingliang77f22f92019-09-10 18:56:57 +08002750 * with dev_put() we need publish tun after registration.
2751 */
2752 rcu_assign_pointer(tfile->tun, tun);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002753 }
2754
Michael S. Tsirkinaf668b32013-01-28 00:38:02 +00002755 netif_carrier_on(tun->dev);
2756
Max Krasnyanskye35259a2008-07-10 16:59:11 -07002757 /* Make sure persistent devices do not get stuck in
2758 * xoff state.
2759 */
2760 if (netif_running(tun->dev))
Jason Wangc8d68e62012-10-31 19:46:00 +00002761 netif_tx_wake_all_queues(tun->dev);
Max Krasnyanskye35259a2008-07-10 16:59:11 -07002762
Linus Torvalds1da177e2005-04-16 15:20:36 -07002763 strcpy(ifr->ifr_name, tun->dev->name);
2764 return 0;
2765
Jason Wang662ca432013-09-11 18:09:48 +08002766err_detach:
2767 tun_detach_all(dev);
Eric Dumazet11fc7d52019-10-07 12:21:05 -07002768 /* We are here because register_netdevice() has failed.
2769 * If register_netdevice() already called tun_free_netdev()
Heiner Kallweit497a5752020-11-07 21:50:56 +01002770 * while dealing with the error, dev->stats has been cleared.
Eric Dumazet11fc7d52019-10-07 12:21:05 -07002771 */
Heiner Kallweit497a5752020-11-07 21:50:56 +01002772 if (!dev->tstats)
Eric Dumazet11fc7d52019-10-07 12:21:05 -07002773 goto err_free_dev;
Eric Dumazetff244c62017-08-18 13:39:56 -07002774
Jason Wang662ca432013-09-11 18:09:48 +08002775err_free_flow:
2776 tun_flow_uninit(tun);
2777 security_tun_dev_free_security(tun->security);
Paolo Abeni608b9972016-04-13 10:52:20 +02002778err_free_stat:
Heiner Kallweit497a5752020-11-07 21:50:56 +01002779 free_percpu(dev->tstats);
Jason Wang662ca432013-09-11 18:09:48 +08002780err_free_dev:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002781 free_netdev(dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002782 return err;
2783}
2784
Kirill Tkhai12132762019-03-20 12:16:53 +03002785static void tun_get_iff(struct tun_struct *tun, struct ifreq *ifr)
Mark McLoughline3b99552008-08-15 15:09:56 -07002786{
Mark McLoughline3b99552008-08-15 15:09:56 -07002787 strcpy(ifr->ifr_name, tun->dev->name);
2788
David Woodhouse980c9e82009-05-09 22:54:21 -07002789 ifr->ifr_flags = tun_flags(tun);
Mark McLoughline3b99552008-08-15 15:09:56 -07002790
Mark McLoughline3b99552008-08-15 15:09:56 -07002791}
2792
Rusty Russell5228ddc2008-07-03 03:46:16 -07002793/* This is like a cut-down ethtool ops, except done via tun fd so no
2794 * privs required. */
Michał Mirosław88255372011-04-19 06:13:10 +00002795static int set_offload(struct tun_struct *tun, unsigned long arg)
Rusty Russell5228ddc2008-07-03 03:46:16 -07002796{
Michał Mirosławc8f44af2011-11-15 15:29:55 +00002797 netdev_features_t features = 0;
Rusty Russell5228ddc2008-07-03 03:46:16 -07002798
2799 if (arg & TUN_F_CSUM) {
Michał Mirosław88255372011-04-19 06:13:10 +00002800 features |= NETIF_F_HW_CSUM;
Rusty Russell5228ddc2008-07-03 03:46:16 -07002801 arg &= ~TUN_F_CSUM;
2802
2803 if (arg & (TUN_F_TSO4|TUN_F_TSO6)) {
2804 if (arg & TUN_F_TSO_ECN) {
2805 features |= NETIF_F_TSO_ECN;
2806 arg &= ~TUN_F_TSO_ECN;
2807 }
2808 if (arg & TUN_F_TSO4)
2809 features |= NETIF_F_TSO;
2810 if (arg & TUN_F_TSO6)
2811 features |= NETIF_F_TSO6;
2812 arg &= ~(TUN_F_TSO4|TUN_F_TSO6);
2813 }
Willem de Bruijn0c19f8462017-11-21 10:22:25 -05002814
2815 arg &= ~TUN_F_UFO;
Rusty Russell5228ddc2008-07-03 03:46:16 -07002816 }
2817
2818 /* This gives the user a way to test for new features in future by
2819 * trying to set them. */
2820 if (arg)
2821 return -EINVAL;
2822
Michał Mirosław88255372011-04-19 06:13:10 +00002823 tun->set_features = features;
Yaroslav Isakov09050952017-03-16 22:44:10 +03002824 tun->dev->wanted_features &= ~TUN_USER_FEATURES;
2825 tun->dev->wanted_features |= features;
Michał Mirosław88255372011-04-19 06:13:10 +00002826 netdev_update_features(tun->dev);
Rusty Russell5228ddc2008-07-03 03:46:16 -07002827
2828 return 0;
2829}
2830
Jason Wangc8d68e62012-10-31 19:46:00 +00002831static void tun_detach_filter(struct tun_struct *tun, int n)
2832{
2833 int i;
2834 struct tun_file *tfile;
2835
2836 for (i = 0; i < n; i++) {
Jason Wangb8deabd2013-01-11 16:59:32 +00002837 tfile = rtnl_dereference(tun->tfiles[i]);
Hannes Frederic Sowa8ced4252016-04-05 17:10:16 +02002838 lock_sock(tfile->socket.sk);
2839 sk_detach_filter(tfile->socket.sk);
2840 release_sock(tfile->socket.sk);
Jason Wangc8d68e62012-10-31 19:46:00 +00002841 }
2842
2843 tun->filter_attached = false;
2844}
2845
2846static int tun_attach_filter(struct tun_struct *tun)
2847{
2848 int i, ret = 0;
2849 struct tun_file *tfile;
2850
2851 for (i = 0; i < tun->numqueues; i++) {
Jason Wangb8deabd2013-01-11 16:59:32 +00002852 tfile = rtnl_dereference(tun->tfiles[i]);
Hannes Frederic Sowa8ced4252016-04-05 17:10:16 +02002853 lock_sock(tfile->socket.sk);
2854 ret = sk_attach_filter(&tun->fprog, tfile->socket.sk);
2855 release_sock(tfile->socket.sk);
Jason Wangc8d68e62012-10-31 19:46:00 +00002856 if (ret) {
2857 tun_detach_filter(tun, i);
2858 return ret;
2859 }
2860 }
2861
2862 tun->filter_attached = true;
2863 return ret;
2864}
2865
2866static void tun_set_sndbuf(struct tun_struct *tun)
2867{
2868 struct tun_file *tfile;
2869 int i;
2870
2871 for (i = 0; i < tun->numqueues; i++) {
Jason Wangb8deabd2013-01-11 16:59:32 +00002872 tfile = rtnl_dereference(tun->tfiles[i]);
Jason Wangc8d68e62012-10-31 19:46:00 +00002873 tfile->socket.sk->sk_sndbuf = tun->sndbuf;
2874 }
2875}
2876
Jason Wangcde8b152012-10-31 19:46:01 +00002877static int tun_set_queue(struct file *file, struct ifreq *ifr)
2878{
2879 struct tun_file *tfile = file->private_data;
2880 struct tun_struct *tun;
Jason Wangcde8b152012-10-31 19:46:01 +00002881 int ret = 0;
2882
2883 rtnl_lock();
2884
2885 if (ifr->ifr_flags & IFF_ATTACH_QUEUE) {
Jason Wang4008e972012-12-13 23:53:30 +00002886 tun = tfile->detached;
Paul Moore5dbbaf22013-01-14 07:12:19 +00002887 if (!tun) {
Jason Wangcde8b152012-10-31 19:46:01 +00002888 ret = -EINVAL;
Paul Moore5dbbaf22013-01-14 07:12:19 +00002889 goto unlock;
2890 }
2891 ret = security_tun_dev_attach_queue(tun->security);
2892 if (ret < 0)
2893 goto unlock;
Eric Dumazetaf3fb242018-09-28 14:51:49 -07002894 ret = tun_attach(tun, file, false, tun->flags & IFF_NAPI,
Yang Yingliang77f22f92019-09-10 18:56:57 +08002895 tun->flags & IFF_NAPI_FRAGS, true);
Jason Wang4008e972012-12-13 23:53:30 +00002896 } else if (ifr->ifr_flags & IFF_DETACH_QUEUE) {
Jason Wangb8deabd2013-01-11 16:59:32 +00002897 tun = rtnl_dereference(tfile->tun);
Michael S. Tsirkin40630b82014-11-19 15:17:31 +02002898 if (!tun || !(tun->flags & IFF_MULTI_QUEUE) || tfile->detached)
Jason Wang4008e972012-12-13 23:53:30 +00002899 ret = -EINVAL;
2900 else
2901 __tun_detach(tfile, false);
2902 } else
Jason Wangcde8b152012-10-31 19:46:01 +00002903 ret = -EINVAL;
2904
Sabrina Dubroca83c1f362018-04-10 16:28:56 +02002905 if (ret >= 0)
2906 netdev_state_change(tun->dev);
2907
Paul Moore5dbbaf22013-01-14 07:12:19 +00002908unlock:
Jason Wangcde8b152012-10-31 19:46:01 +00002909 rtnl_unlock();
2910 return ret;
2911}
2912
Jason Wang8f3f3302020-07-31 00:17:20 -04002913static int tun_set_ebpf(struct tun_struct *tun, struct tun_prog __rcu **prog_p,
Jason Wangcd5681d2018-01-16 16:31:01 +08002914 void __user *data)
Jason Wang96f84062017-12-04 17:31:23 +08002915{
2916 struct bpf_prog *prog;
2917 int fd;
2918
2919 if (copy_from_user(&fd, data, sizeof(fd)))
2920 return -EFAULT;
2921
2922 if (fd == -1) {
2923 prog = NULL;
2924 } else {
2925 prog = bpf_prog_get_type(fd, BPF_PROG_TYPE_SOCKET_FILTER);
2926 if (IS_ERR(prog))
2927 return PTR_ERR(prog);
2928 }
2929
Jason Wangcd5681d2018-01-16 16:31:01 +08002930 return __tun_set_ebpf(tun, prog_p, prog);
Jason Wang96f84062017-12-04 17:31:23 +08002931}
2932
Phillip Pottercca8ea32021-04-06 18:45:54 +01002933/* Return correct value for tun->dev->addr_len based on tun->dev->type. */
2934static unsigned char tun_get_addr_len(unsigned short type)
2935{
2936 switch (type) {
2937 case ARPHRD_IP6GRE:
2938 case ARPHRD_TUNNEL6:
2939 return sizeof(struct in6_addr);
2940 case ARPHRD_IPGRE:
2941 case ARPHRD_TUNNEL:
2942 case ARPHRD_SIT:
2943 return 4;
2944 case ARPHRD_ETHER:
2945 return ETH_ALEN;
2946 case ARPHRD_IEEE802154:
2947 case ARPHRD_IEEE802154_MONITOR:
2948 return IEEE802154_EXTENDED_ADDR_LEN;
2949 case ARPHRD_PHONET_PIPE:
2950 case ARPHRD_PPP:
2951 case ARPHRD_NONE:
2952 return 0;
2953 case ARPHRD_6LOWPAN:
2954 return EUI64_ADDR_LEN;
2955 case ARPHRD_FDDI:
2956 return FDDI_K_ALEN;
2957 case ARPHRD_HIPPI:
2958 return HIPPI_ALEN;
2959 case ARPHRD_IEEE802:
2960 return FC_ALEN;
2961 case ARPHRD_ROSE:
2962 return ROSE_ADDR_LEN;
2963 case ARPHRD_NETROM:
2964 return AX25_ADDR_LEN;
2965 case ARPHRD_LOCALTLK:
2966 return LTALK_ALEN;
2967 default:
2968 return 0;
2969 }
2970}
2971
Arnd Bergmann50857e22009-11-06 22:52:32 -08002972static long __tun_chr_ioctl(struct file *file, unsigned int cmd,
2973 unsigned long arg, int ifreq_len)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002974{
Eric W. Biederman36b50ba2009-01-20 11:01:48 +00002975 struct tun_file *tfile = file->private_data;
Kirill Tkhaif6637062018-05-08 19:21:34 +03002976 struct net *net = sock_net(&tfile->sk);
Eric W. Biederman631ab462009-01-20 11:00:40 +00002977 struct tun_struct *tun;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002978 void __user* argp = (void __user*)arg;
Nicolas Dichtel26d31922018-11-28 19:12:56 +01002979 unsigned int ifindex, carrier;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002980 struct ifreq ifr;
Eric W. Biederman0625c882012-02-07 16:48:55 -08002981 kuid_t owner;
2982 kgid_t group;
Herbert Xu33dccbb2009-02-05 21:25:32 -08002983 int sndbuf;
Michael S. Tsirkind9d52b52010-03-17 17:45:01 +02002984 int vnet_hdr_sz;
Michael S. Tsirkin1cf8e412014-12-16 15:05:06 +02002985 int le;
Max Krasnyanskyf271b2cc2008-07-14 22:18:19 -07002986 int ret;
Sabrina Dubroca83c1f362018-04-10 16:28:56 +02002987 bool do_notify = false;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002988
Kirill Tkhaif2780d62018-02-14 16:40:14 +03002989 if (cmd == TUNSETIFF || cmd == TUNSETQUEUE ||
2990 (_IOC_TYPE(cmd) == SOCK_IOC_TYPE && cmd != SIOCGSKNS)) {
Arnd Bergmann50857e22009-11-06 22:52:32 -08002991 if (copy_from_user(&ifr, argp, ifreq_len))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002992 return -EFAULT;
David S. Miller8bbb1812012-07-30 14:52:48 -07002993 } else {
Mathias Krausea117dac2012-07-29 19:45:14 +00002994 memset(&ifr, 0, sizeof(ifr));
David S. Miller8bbb1812012-07-30 14:52:48 -07002995 }
Eric W. Biederman631ab462009-01-20 11:00:40 +00002996 if (cmd == TUNGETFEATURES) {
2997 /* Currently this just means: "what IFF flags are valid?".
2998 * This is needed because we never checked for invalid flags on
Michael S. Tsirkin031f5e032014-11-19 14:44:40 +02002999 * TUNSETIFF.
3000 */
3001 return put_user(IFF_TUN | IFF_TAP | TUN_FEATURES,
Eric W. Biederman631ab462009-01-20 11:00:40 +00003002 (unsigned int __user*)argp);
Kirill Tkhaif6637062018-05-08 19:21:34 +03003003 } else if (cmd == TUNSETQUEUE) {
Jason Wangcde8b152012-10-31 19:46:01 +00003004 return tun_set_queue(file, &ifr);
Kirill Tkhaif6637062018-05-08 19:21:34 +03003005 } else if (cmd == SIOCGSKNS) {
3006 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
3007 return -EPERM;
3008 return open_related_ns(&net->ns, get_net_ns);
3009 }
Eric W. Biederman631ab462009-01-20 11:00:40 +00003010
Herbert Xu876bfd42009-08-06 14:22:44 +00003011 rtnl_lock();
3012
yuan linyu9484dc72017-09-23 22:36:52 +08003013 tun = tun_get(tfile);
Gao Feng0f16bc12016-10-25 22:26:09 +08003014 if (cmd == TUNSETIFF) {
3015 ret = -EEXIST;
3016 if (tun)
3017 goto unlock;
3018
Linus Torvalds1da177e2005-04-16 15:20:36 -07003019 ifr.ifr_name[IFNAMSIZ-1] = '\0';
3020
Kirill Tkhaif2780d62018-02-14 16:40:14 +03003021 ret = tun_set_iff(net, file, &ifr);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003022
Herbert Xu876bfd42009-08-06 14:22:44 +00003023 if (ret)
3024 goto unlock;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003025
Arnd Bergmann50857e22009-11-06 22:52:32 -08003026 if (copy_to_user(argp, &ifr, ifreq_len))
Herbert Xu876bfd42009-08-06 14:22:44 +00003027 ret = -EFAULT;
3028 goto unlock;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003029 }
Pavel Emelyanovfb7589a2013-08-21 14:31:38 +04003030 if (cmd == TUNSETIFINDEX) {
3031 ret = -EPERM;
3032 if (tun)
3033 goto unlock;
3034
3035 ret = -EFAULT;
3036 if (copy_from_user(&ifindex, argp, sizeof(ifindex)))
3037 goto unlock;
3038
3039 ret = 0;
3040 tfile->ifindex = ifindex;
3041 goto unlock;
3042 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003043
Herbert Xu876bfd42009-08-06 14:22:44 +00003044 ret = -EBADFD;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003045 if (!tun)
Herbert Xu876bfd42009-08-06 14:22:44 +00003046 goto unlock;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003047
Michal Kubecek34241702020-03-04 17:24:14 +01003048 netif_info(tun, drv, tun->dev, "tun_chr_ioctl cmd %u\n", cmd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003049
Kirill Tkhai0c3e0e32019-03-20 12:16:42 +03003050 net = dev_net(tun->dev);
Eric W. Biederman631ab462009-01-20 11:00:40 +00003051 ret = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003052 switch (cmd) {
Mark McLoughline3b99552008-08-15 15:09:56 -07003053 case TUNGETIFF:
Kirill Tkhai12132762019-03-20 12:16:53 +03003054 tun_get_iff(tun, &ifr);
Mark McLoughline3b99552008-08-15 15:09:56 -07003055
Pavel Emelyanov3d407a82013-08-21 14:32:00 +04003056 if (tfile->detached)
3057 ifr.ifr_flags |= IFF_DETACH_QUEUE;
Pavel Emelyanov849c9b62013-08-21 14:32:21 +04003058 if (!tfile->socket.sk->sk_filter)
3059 ifr.ifr_flags |= IFF_NOFILTER;
Pavel Emelyanov3d407a82013-08-21 14:32:00 +04003060
Arnd Bergmann50857e22009-11-06 22:52:32 -08003061 if (copy_to_user(argp, &ifr, ifreq_len))
Eric W. Biederman631ab462009-01-20 11:00:40 +00003062 ret = -EFAULT;
Mark McLoughline3b99552008-08-15 15:09:56 -07003063 break;
3064
Linus Torvalds1da177e2005-04-16 15:20:36 -07003065 case TUNSETNOCSUM:
3066 /* Disable/Enable checksum */
Linus Torvalds1da177e2005-04-16 15:20:36 -07003067
Michał Mirosław88255372011-04-19 06:13:10 +00003068 /* [unimplemented] */
Michal Kubecek34241702020-03-04 17:24:14 +01003069 netif_info(tun, drv, tun->dev, "ignored: set checksum %s\n",
3070 arg ? "disabled" : "enabled");
Linus Torvalds1da177e2005-04-16 15:20:36 -07003071 break;
3072
3073 case TUNSETPERSIST:
Jason Wang54f968d2012-10-31 19:45:57 +00003074 /* Disable/Enable persist mode. Keep an extra reference to the
3075 * module to prevent the module being unprobed.
3076 */
Michael S. Tsirkin40630b82014-11-19 15:17:31 +02003077 if (arg && !(tun->flags & IFF_PERSIST)) {
3078 tun->flags |= IFF_PERSIST;
Jason Wang54f968d2012-10-31 19:45:57 +00003079 __module_get(THIS_MODULE);
Sabrina Dubroca83c1f362018-04-10 16:28:56 +02003080 do_notify = true;
Jason Wangdd38bd82013-01-11 16:59:34 +00003081 }
Michael S. Tsirkin40630b82014-11-19 15:17:31 +02003082 if (!arg && (tun->flags & IFF_PERSIST)) {
3083 tun->flags &= ~IFF_PERSIST;
Jason Wang54f968d2012-10-31 19:45:57 +00003084 module_put(THIS_MODULE);
Sabrina Dubroca83c1f362018-04-10 16:28:56 +02003085 do_notify = true;
Jason Wang54f968d2012-10-31 19:45:57 +00003086 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003087
Michal Kubecek34241702020-03-04 17:24:14 +01003088 netif_info(tun, drv, tun->dev, "persist %s\n",
3089 arg ? "enabled" : "disabled");
Linus Torvalds1da177e2005-04-16 15:20:36 -07003090 break;
3091
3092 case TUNSETOWNER:
3093 /* Set owner of the device */
Eric W. Biederman0625c882012-02-07 16:48:55 -08003094 owner = make_kuid(current_user_ns(), arg);
3095 if (!uid_valid(owner)) {
3096 ret = -EINVAL;
3097 break;
3098 }
3099 tun->owner = owner;
Sabrina Dubroca83c1f362018-04-10 16:28:56 +02003100 do_notify = true;
Michal Kubecek34241702020-03-04 17:24:14 +01003101 netif_info(tun, drv, tun->dev, "owner set to %u\n",
3102 from_kuid(&init_user_ns, tun->owner));
Linus Torvalds1da177e2005-04-16 15:20:36 -07003103 break;
3104
Guido Guenther8c644622007-07-02 22:50:25 -07003105 case TUNSETGROUP:
3106 /* Set group of the device */
Eric W. Biederman0625c882012-02-07 16:48:55 -08003107 group = make_kgid(current_user_ns(), arg);
3108 if (!gid_valid(group)) {
3109 ret = -EINVAL;
3110 break;
3111 }
3112 tun->group = group;
Sabrina Dubroca83c1f362018-04-10 16:28:56 +02003113 do_notify = true;
Michal Kubecek34241702020-03-04 17:24:14 +01003114 netif_info(tun, drv, tun->dev, "group set to %u\n",
3115 from_kgid(&init_user_ns, tun->group));
Guido Guenther8c644622007-07-02 22:50:25 -07003116 break;
3117
Mike Kershawff4cc3a2005-09-01 17:40:05 -07003118 case TUNSETLINK:
3119 /* Only allow setting the type when the interface is down */
3120 if (tun->dev->flags & IFF_UP) {
Michal Kubecek34241702020-03-04 17:24:14 +01003121 netif_info(tun, drv, tun->dev,
3122 "Linktype set failed because interface is up\n");
David S. Miller48abfe02008-04-23 19:37:58 -07003123 ret = -EBUSY;
Mike Kershawff4cc3a2005-09-01 17:40:05 -07003124 } else {
Martin Schiller8e1e33f2020-11-18 07:39:19 +01003125 ret = call_netdevice_notifiers(NETDEV_PRE_TYPE_CHANGE,
3126 tun->dev);
3127 ret = notifier_to_errno(ret);
3128 if (ret) {
3129 netif_info(tun, drv, tun->dev,
3130 "Refused to change device type\n");
3131 break;
3132 }
Mike Kershawff4cc3a2005-09-01 17:40:05 -07003133 tun->dev->type = (int) arg;
Phillip Pottercca8ea32021-04-06 18:45:54 +01003134 tun->dev->addr_len = tun_get_addr_len(tun->dev->type);
Michal Kubecek34241702020-03-04 17:24:14 +01003135 netif_info(tun, drv, tun->dev, "linktype set to %d\n",
3136 tun->dev->type);
Martin Schiller8e1e33f2020-11-18 07:39:19 +01003137 call_netdevice_notifiers(NETDEV_POST_TYPE_CHANGE,
3138 tun->dev);
Mike Kershawff4cc3a2005-09-01 17:40:05 -07003139 }
Eric W. Biederman631ab462009-01-20 11:00:40 +00003140 break;
Mike Kershawff4cc3a2005-09-01 17:40:05 -07003141
Linus Torvalds1da177e2005-04-16 15:20:36 -07003142 case TUNSETDEBUG:
Michal Kubecek34241702020-03-04 17:24:14 +01003143 tun->msg_enable = (u32)arg;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003144 break;
Michal Kubecek34241702020-03-04 17:24:14 +01003145
Rusty Russell5228ddc2008-07-03 03:46:16 -07003146 case TUNSETOFFLOAD:
Michał Mirosław88255372011-04-19 06:13:10 +00003147 ret = set_offload(tun, arg);
Eric W. Biederman631ab462009-01-20 11:00:40 +00003148 break;
Rusty Russell5228ddc2008-07-03 03:46:16 -07003149
Max Krasnyanskyf271b2cc2008-07-14 22:18:19 -07003150 case TUNSETTXFILTER:
3151 /* Can be set only for TAPs */
Eric W. Biederman631ab462009-01-20 11:00:40 +00003152 ret = -EINVAL;
Michael S. Tsirkin40630b82014-11-19 15:17:31 +02003153 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
Eric W. Biederman631ab462009-01-20 11:00:40 +00003154 break;
Harvey Harrisonc0e5a8c2008-07-16 12:45:34 -07003155 ret = update_filter(&tun->txflt, (void __user *)arg);
Eric W. Biederman631ab462009-01-20 11:00:40 +00003156 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003157
3158 case SIOCGIFHWADDR:
Uwe Kleine-Königb5950762010-11-01 15:38:34 -04003159 /* Get hw address */
Cong Wang3b23a322021-02-11 11:34:10 -08003160 dev_get_mac_address(&ifr.ifr_hwaddr, net, tun->dev->name);
Arnd Bergmann50857e22009-11-06 22:52:32 -08003161 if (copy_to_user(argp, &ifr, ifreq_len))
Eric W. Biederman631ab462009-01-20 11:00:40 +00003162 ret = -EFAULT;
3163 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003164
3165 case SIOCSIFHWADDR:
Max Krasnyanskyf271b2cc2008-07-14 22:18:19 -07003166 /* Set hw address */
Cong Wang3b23a322021-02-11 11:34:10 -08003167 ret = dev_set_mac_address_user(tun->dev, &ifr.ifr_hwaddr, NULL);
Eric W. Biederman631ab462009-01-20 11:00:40 +00003168 break;
Herbert Xu33dccbb2009-02-05 21:25:32 -08003169
3170 case TUNGETSNDBUF:
Jason Wang54f968d2012-10-31 19:45:57 +00003171 sndbuf = tfile->socket.sk->sk_sndbuf;
Herbert Xu33dccbb2009-02-05 21:25:32 -08003172 if (copy_to_user(argp, &sndbuf, sizeof(sndbuf)))
3173 ret = -EFAULT;
3174 break;
3175
3176 case TUNSETSNDBUF:
3177 if (copy_from_user(&sndbuf, argp, sizeof(sndbuf))) {
3178 ret = -EFAULT;
3179 break;
3180 }
Craig Gallek931619222017-10-30 18:50:11 -04003181 if (sndbuf <= 0) {
3182 ret = -EINVAL;
3183 break;
3184 }
Herbert Xu33dccbb2009-02-05 21:25:32 -08003185
Jason Wangc8d68e62012-10-31 19:46:00 +00003186 tun->sndbuf = sndbuf;
3187 tun_set_sndbuf(tun);
Herbert Xu33dccbb2009-02-05 21:25:32 -08003188 break;
3189
Michael S. Tsirkind9d52b52010-03-17 17:45:01 +02003190 case TUNGETVNETHDRSZ:
3191 vnet_hdr_sz = tun->vnet_hdr_sz;
3192 if (copy_to_user(argp, &vnet_hdr_sz, sizeof(vnet_hdr_sz)))
3193 ret = -EFAULT;
3194 break;
3195
3196 case TUNSETVNETHDRSZ:
3197 if (copy_from_user(&vnet_hdr_sz, argp, sizeof(vnet_hdr_sz))) {
3198 ret = -EFAULT;
3199 break;
3200 }
3201 if (vnet_hdr_sz < (int)sizeof(struct virtio_net_hdr)) {
3202 ret = -EINVAL;
3203 break;
3204 }
3205
3206 tun->vnet_hdr_sz = vnet_hdr_sz;
3207 break;
3208
Michael S. Tsirkin1cf8e412014-12-16 15:05:06 +02003209 case TUNGETVNETLE:
3210 le = !!(tun->flags & TUN_VNET_LE);
3211 if (put_user(le, (int __user *)argp))
3212 ret = -EFAULT;
3213 break;
3214
3215 case TUNSETVNETLE:
3216 if (get_user(le, (int __user *)argp)) {
3217 ret = -EFAULT;
3218 break;
3219 }
3220 if (le)
3221 tun->flags |= TUN_VNET_LE;
3222 else
3223 tun->flags &= ~TUN_VNET_LE;
3224 break;
3225
Greg Kurz8b8e6582015-04-24 14:50:36 +02003226 case TUNGETVNETBE:
3227 ret = tun_get_vnet_be(tun, argp);
3228 break;
3229
3230 case TUNSETVNETBE:
3231 ret = tun_set_vnet_be(tun, argp);
3232 break;
3233
Michael S. Tsirkin99405162010-02-14 01:01:10 +00003234 case TUNATTACHFILTER:
3235 /* Can be set only for TAPs */
3236 ret = -EINVAL;
Michael S. Tsirkin40630b82014-11-19 15:17:31 +02003237 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
Michael S. Tsirkin99405162010-02-14 01:01:10 +00003238 break;
3239 ret = -EFAULT;
Jason Wang54f968d2012-10-31 19:45:57 +00003240 if (copy_from_user(&tun->fprog, argp, sizeof(tun->fprog)))
Michael S. Tsirkin99405162010-02-14 01:01:10 +00003241 break;
3242
Jason Wangc8d68e62012-10-31 19:46:00 +00003243 ret = tun_attach_filter(tun);
Michael S. Tsirkin99405162010-02-14 01:01:10 +00003244 break;
3245
3246 case TUNDETACHFILTER:
3247 /* Can be set only for TAPs */
3248 ret = -EINVAL;
Michael S. Tsirkin40630b82014-11-19 15:17:31 +02003249 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
Michael S. Tsirkin99405162010-02-14 01:01:10 +00003250 break;
Jason Wangc8d68e62012-10-31 19:46:00 +00003251 ret = 0;
3252 tun_detach_filter(tun, tun->numqueues);
Michael S. Tsirkin99405162010-02-14 01:01:10 +00003253 break;
3254
Pavel Emelyanov76975e92013-08-21 14:32:39 +04003255 case TUNGETFILTER:
3256 ret = -EINVAL;
Michael S. Tsirkin40630b82014-11-19 15:17:31 +02003257 if ((tun->flags & TUN_TYPE_MASK) != IFF_TAP)
Pavel Emelyanov76975e92013-08-21 14:32:39 +04003258 break;
3259 ret = -EFAULT;
3260 if (copy_to_user(argp, &tun->fprog, sizeof(tun->fprog)))
3261 break;
3262 ret = 0;
3263 break;
3264
Jason Wang96f84062017-12-04 17:31:23 +08003265 case TUNSETSTEERINGEBPF:
Jason Wangcd5681d2018-01-16 16:31:01 +08003266 ret = tun_set_ebpf(tun, &tun->steering_prog, argp);
Jason Wang96f84062017-12-04 17:31:23 +08003267 break;
3268
Jason Wangaff3d702018-01-16 16:31:02 +08003269 case TUNSETFILTEREBPF:
3270 ret = tun_set_ebpf(tun, &tun->filter_prog, argp);
3271 break;
3272
Nicolas Dichtel26d31922018-11-28 19:12:56 +01003273 case TUNSETCARRIER:
3274 ret = -EFAULT;
3275 if (copy_from_user(&carrier, argp, sizeof(carrier)))
3276 goto unlock;
3277
3278 ret = tun_net_change_carrier(tun->dev, (bool)carrier);
3279 break;
3280
Kirill Tkhai0c3e0e32019-03-20 12:16:42 +03003281 case TUNGETDEVNETNS:
3282 ret = -EPERM;
3283 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
3284 goto unlock;
3285 ret = open_related_ns(&net->ns, get_net_ns);
3286 break;
3287
Linus Torvalds1da177e2005-04-16 15:20:36 -07003288 default:
Eric W. Biederman631ab462009-01-20 11:00:40 +00003289 ret = -EINVAL;
3290 break;
Joe Perchesee289b62010-05-17 22:47:34 -07003291 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003292
Sabrina Dubroca83c1f362018-04-10 16:28:56 +02003293 if (do_notify)
3294 netdev_state_change(tun->dev);
3295
Herbert Xu876bfd42009-08-06 14:22:44 +00003296unlock:
3297 rtnl_unlock();
3298 if (tun)
3299 tun_put(tun);
Eric W. Biederman631ab462009-01-20 11:00:40 +00003300 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003301}
3302
Arnd Bergmann50857e22009-11-06 22:52:32 -08003303static long tun_chr_ioctl(struct file *file,
3304 unsigned int cmd, unsigned long arg)
3305{
3306 return __tun_chr_ioctl(file, cmd, arg, sizeof (struct ifreq));
3307}
3308
3309#ifdef CONFIG_COMPAT
3310static long tun_chr_compat_ioctl(struct file *file,
3311 unsigned int cmd, unsigned long arg)
3312{
3313 switch (cmd) {
3314 case TUNSETIFF:
3315 case TUNGETIFF:
3316 case TUNSETTXFILTER:
3317 case TUNGETSNDBUF:
3318 case TUNSETSNDBUF:
3319 case SIOCGIFHWADDR:
3320 case SIOCSIFHWADDR:
3321 arg = (unsigned long)compat_ptr(arg);
3322 break;
3323 default:
3324 arg = (compat_ulong_t)arg;
3325 break;
3326 }
3327
3328 /*
3329 * compat_ifreq is shorter than ifreq, so we must not access beyond
3330 * the end of that structure. All fields that are used in this
3331 * driver are compatible though, we don't need to convert the
3332 * contents.
3333 */
3334 return __tun_chr_ioctl(file, cmd, arg, sizeof(struct compat_ifreq));
3335}
3336#endif /* CONFIG_COMPAT */
3337
Linus Torvalds1da177e2005-04-16 15:20:36 -07003338static int tun_chr_fasync(int fd, struct file *file, int on)
3339{
Jason Wang54f968d2012-10-31 19:45:57 +00003340 struct tun_file *tfile = file->private_data;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003341 int ret;
3342
Jason Wang54f968d2012-10-31 19:45:57 +00003343 if ((ret = fasync_helper(fd, file, on, &tfile->fasync)) < 0)
Jonathan Corbet9d319522008-06-19 15:50:37 -06003344 goto out;
Jeff Garzik6aa20a22006-09-13 13:24:59 -04003345
Linus Torvalds1da177e2005-04-16 15:20:36 -07003346 if (on) {
Eric W. Biederman01919132017-07-16 22:05:57 -05003347 __f_setown(file, task_pid(current), PIDTYPE_TGID, 0);
Jason Wang54f968d2012-10-31 19:45:57 +00003348 tfile->flags |= TUN_FASYNC;
Jeff Garzik6aa20a22006-09-13 13:24:59 -04003349 } else
Jason Wang54f968d2012-10-31 19:45:57 +00003350 tfile->flags &= ~TUN_FASYNC;
Jonathan Corbet9d319522008-06-19 15:50:37 -06003351 ret = 0;
3352out:
Jonathan Corbet9d319522008-06-19 15:50:37 -06003353 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003354}
3355
3356static int tun_chr_open(struct inode *inode, struct file * file)
3357{
Eric W. Biederman140e807da2015-05-08 21:07:08 -05003358 struct net *net = current->nsproxy->net_ns;
Eric W. Biederman631ab462009-01-20 11:00:40 +00003359 struct tun_file *tfile;
Thomas Gleixnerdeed49f2009-10-14 01:19:46 -07003360
Eric W. Biederman140e807da2015-05-08 21:07:08 -05003361 tfile = (struct tun_file *)sk_alloc(net, AF_UNSPEC, GFP_KERNEL,
Eric W. Biederman11aa9c22015-05-08 21:09:13 -05003362 &tun_proto, 0);
Eric W. Biederman631ab462009-01-20 11:00:40 +00003363 if (!tfile)
3364 return -ENOMEM;
Jason Wangb196d882018-05-11 10:49:25 +08003365 if (ptr_ring_init(&tfile->tx_ring, 0, GFP_KERNEL)) {
3366 sk_free(&tfile->sk);
3367 return -ENOMEM;
3368 }
3369
Eric Dumazetc7256f52018-09-28 14:51:48 -07003370 mutex_init(&tfile->napi_mutex);
Monam Agarwalc9566742014-03-24 00:02:32 +05303371 RCU_INIT_POINTER(tfile->tun, NULL);
Jason Wang54f968d2012-10-31 19:45:57 +00003372 tfile->flags = 0;
Pavel Emelyanovfb7589a2013-08-21 14:31:38 +04003373 tfile->ifindex = 0;
Jason Wang54f968d2012-10-31 19:45:57 +00003374
Al Viro333f7902019-07-05 20:14:16 +01003375 init_waitqueue_head(&tfile->socket.wq.wait);
Jason Wang54f968d2012-10-31 19:45:57 +00003376
3377 tfile->socket.file = file;
3378 tfile->socket.ops = &tun_socket_ops;
3379
3380 sock_init_data(&tfile->socket, &tfile->sk);
Jason Wang54f968d2012-10-31 19:45:57 +00003381
3382 tfile->sk.sk_write_space = tun_sock_write_space;
3383 tfile->sk.sk_sndbuf = INT_MAX;
3384
Eric W. Biederman631ab462009-01-20 11:00:40 +00003385 file->private_data = tfile;
Jason Wang4008e972012-12-13 23:53:30 +00003386 INIT_LIST_HEAD(&tfile->next);
Jason Wang54f968d2012-10-31 19:45:57 +00003387
Jason Wang19a6afb2013-06-08 14:17:41 +08003388 sock_set_flag(&tfile->sk, SOCK_ZEROCOPY);
3389
Linus Torvalds1da177e2005-04-16 15:20:36 -07003390 return 0;
3391}
3392
3393static int tun_chr_close(struct inode *inode, struct file *file)
3394{
Eric W. Biederman631ab462009-01-20 11:00:40 +00003395 struct tun_file *tfile = file->private_data;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003396
Jason Wangc8d68e62012-10-31 19:46:00 +00003397 tun_detach(tfile, true);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003398
3399 return 0;
3400}
3401
Masatake YAMATO93e14b62014-01-29 16:43:31 +09003402#ifdef CONFIG_PROC_FS
yuan linyu9484dc72017-09-23 22:36:52 +08003403static void tun_chr_show_fdinfo(struct seq_file *m, struct file *file)
Masatake YAMATO93e14b62014-01-29 16:43:31 +09003404{
yuan linyu9484dc72017-09-23 22:36:52 +08003405 struct tun_file *tfile = file->private_data;
Masatake YAMATO93e14b62014-01-29 16:43:31 +09003406 struct tun_struct *tun;
3407 struct ifreq ifr;
3408
3409 memset(&ifr, 0, sizeof(ifr));
3410
3411 rtnl_lock();
yuan linyu9484dc72017-09-23 22:36:52 +08003412 tun = tun_get(tfile);
Masatake YAMATO93e14b62014-01-29 16:43:31 +09003413 if (tun)
Kirill Tkhai12132762019-03-20 12:16:53 +03003414 tun_get_iff(tun, &ifr);
Masatake YAMATO93e14b62014-01-29 16:43:31 +09003415 rtnl_unlock();
3416
3417 if (tun)
3418 tun_put(tun);
3419
Joe Perchesa3816ab2014-09-29 16:08:25 -07003420 seq_printf(m, "iff:\t%s\n", ifr.ifr_name);
Masatake YAMATO93e14b62014-01-29 16:43:31 +09003421}
3422#endif
3423
Arjan van de Vend54b1fd2007-02-12 00:55:34 -08003424static const struct file_operations tun_fops = {
Jeff Garzik6aa20a22006-09-13 13:24:59 -04003425 .owner = THIS_MODULE,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003426 .llseek = no_llseek,
Al Viro9b067032014-11-07 13:52:07 -05003427 .read_iter = tun_chr_read_iter,
Al Virof5ff53b2014-06-19 15:36:49 -04003428 .write_iter = tun_chr_write_iter,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003429 .poll = tun_chr_poll,
Arnd Bergmann50857e22009-11-06 22:52:32 -08003430 .unlocked_ioctl = tun_chr_ioctl,
3431#ifdef CONFIG_COMPAT
3432 .compat_ioctl = tun_chr_compat_ioctl,
3433#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07003434 .open = tun_chr_open,
3435 .release = tun_chr_close,
Masatake YAMATO93e14b62014-01-29 16:43:31 +09003436 .fasync = tun_chr_fasync,
3437#ifdef CONFIG_PROC_FS
3438 .show_fdinfo = tun_chr_show_fdinfo,
3439#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07003440};
3441
3442static struct miscdevice tun_miscdev = {
3443 .minor = TUN_MINOR,
3444 .name = "tun",
Kay Sieverse454cea2009-09-18 23:01:12 +02003445 .nodename = "net/tun",
Linus Torvalds1da177e2005-04-16 15:20:36 -07003446 .fops = &tun_fops,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003447};
3448
3449/* ethtool interface */
3450
Chas Williams4e24f2d2018-06-02 17:49:53 -04003451static void tun_default_link_ksettings(struct net_device *dev,
3452 struct ethtool_link_ksettings *cmd)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003453{
Philippe Reynes29ccc492017-03-11 22:03:50 +01003454 ethtool_link_ksettings_zero_link_mode(cmd, supported);
3455 ethtool_link_ksettings_zero_link_mode(cmd, advertising);
3456 cmd->base.speed = SPEED_10;
3457 cmd->base.duplex = DUPLEX_FULL;
3458 cmd->base.port = PORT_TP;
3459 cmd->base.phy_address = 0;
3460 cmd->base.autoneg = AUTONEG_DISABLE;
Chas Williams4e24f2d2018-06-02 17:49:53 -04003461}
3462
3463static int tun_get_link_ksettings(struct net_device *dev,
3464 struct ethtool_link_ksettings *cmd)
3465{
3466 struct tun_struct *tun = netdev_priv(dev);
3467
3468 memcpy(cmd, &tun->link_ksettings, sizeof(*cmd));
3469 return 0;
3470}
3471
3472static int tun_set_link_ksettings(struct net_device *dev,
3473 const struct ethtool_link_ksettings *cmd)
3474{
3475 struct tun_struct *tun = netdev_priv(dev);
3476
3477 memcpy(&tun->link_ksettings, cmd, sizeof(*cmd));
Linus Torvalds1da177e2005-04-16 15:20:36 -07003478 return 0;
3479}
3480
3481static void tun_get_drvinfo(struct net_device *dev, struct ethtool_drvinfo *info)
3482{
3483 struct tun_struct *tun = netdev_priv(dev);
3484
Rick Jones33a5ba12011-11-15 14:59:53 +00003485 strlcpy(info->driver, DRV_NAME, sizeof(info->driver));
3486 strlcpy(info->version, DRV_VERSION, sizeof(info->version));
Linus Torvalds1da177e2005-04-16 15:20:36 -07003487
3488 switch (tun->flags & TUN_TYPE_MASK) {
Michael S. Tsirkin40630b82014-11-19 15:17:31 +02003489 case IFF_TUN:
Rick Jones33a5ba12011-11-15 14:59:53 +00003490 strlcpy(info->bus_info, "tun", sizeof(info->bus_info));
Linus Torvalds1da177e2005-04-16 15:20:36 -07003491 break;
Michael S. Tsirkin40630b82014-11-19 15:17:31 +02003492 case IFF_TAP:
Rick Jones33a5ba12011-11-15 14:59:53 +00003493 strlcpy(info->bus_info, "tap", sizeof(info->bus_info));
Linus Torvalds1da177e2005-04-16 15:20:36 -07003494 break;
3495 }
3496}
3497
3498static u32 tun_get_msglevel(struct net_device *dev)
3499{
Linus Torvalds1da177e2005-04-16 15:20:36 -07003500 struct tun_struct *tun = netdev_priv(dev);
Michal Kubecek34241702020-03-04 17:24:14 +01003501
3502 return tun->msg_enable;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003503}
3504
3505static void tun_set_msglevel(struct net_device *dev, u32 value)
3506{
Linus Torvalds1da177e2005-04-16 15:20:36 -07003507 struct tun_struct *tun = netdev_priv(dev);
Michal Kubecek34241702020-03-04 17:24:14 +01003508
3509 tun->msg_enable = value;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003510}
3511
Jason Wang5503fce2017-01-18 15:02:03 +08003512static int tun_get_coalesce(struct net_device *dev,
3513 struct ethtool_coalesce *ec)
3514{
3515 struct tun_struct *tun = netdev_priv(dev);
3516
3517 ec->rx_max_coalesced_frames = tun->rx_batched;
3518
3519 return 0;
3520}
3521
3522static int tun_set_coalesce(struct net_device *dev,
3523 struct ethtool_coalesce *ec)
3524{
3525 struct tun_struct *tun = netdev_priv(dev);
3526
3527 if (ec->rx_max_coalesced_frames > NAPI_POLL_WEIGHT)
3528 tun->rx_batched = NAPI_POLL_WEIGHT;
3529 else
3530 tun->rx_batched = ec->rx_max_coalesced_frames;
3531
3532 return 0;
3533}
3534
Jeff Garzik7282d492006-09-13 14:30:00 -04003535static const struct ethtool_ops tun_ethtool_ops = {
Jakub Kicinskie5ad00b2020-03-05 17:05:58 -08003536 .supported_coalesce_params = ETHTOOL_COALESCE_RX_MAX_FRAMES,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003537 .get_drvinfo = tun_get_drvinfo,
3538 .get_msglevel = tun_get_msglevel,
3539 .set_msglevel = tun_set_msglevel,
Nolan Leakebee31362010-07-27 13:53:43 +00003540 .get_link = ethtool_op_get_link,
Richard Cochraneda29772013-07-19 19:40:10 +02003541 .get_ts_info = ethtool_op_get_ts_info,
Jason Wang5503fce2017-01-18 15:02:03 +08003542 .get_coalesce = tun_get_coalesce,
3543 .set_coalesce = tun_set_coalesce,
Philippe Reynes29ccc492017-03-11 22:03:50 +01003544 .get_link_ksettings = tun_get_link_ksettings,
Chas Williams4e24f2d2018-06-02 17:49:53 -04003545 .set_link_ksettings = tun_set_link_ksettings,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003546};
3547
Jason Wang1576d982016-06-30 14:45:36 +08003548static int tun_queue_resize(struct tun_struct *tun)
3549{
3550 struct net_device *dev = tun->dev;
3551 struct tun_file *tfile;
Jason Wang5990a302018-01-04 11:14:27 +08003552 struct ptr_ring **rings;
Jason Wang1576d982016-06-30 14:45:36 +08003553 int n = tun->numqueues + tun->numdisabled;
3554 int ret, i;
3555
Jason Wang5990a302018-01-04 11:14:27 +08003556 rings = kmalloc_array(n, sizeof(*rings), GFP_KERNEL);
3557 if (!rings)
Jason Wang1576d982016-06-30 14:45:36 +08003558 return -ENOMEM;
3559
3560 for (i = 0; i < tun->numqueues; i++) {
3561 tfile = rtnl_dereference(tun->tfiles[i]);
Jason Wang5990a302018-01-04 11:14:27 +08003562 rings[i] = &tfile->tx_ring;
Jason Wang1576d982016-06-30 14:45:36 +08003563 }
3564 list_for_each_entry(tfile, &tun->disabled, next)
Jason Wang5990a302018-01-04 11:14:27 +08003565 rings[i++] = &tfile->tx_ring;
Jason Wang1576d982016-06-30 14:45:36 +08003566
Jason Wang5990a302018-01-04 11:14:27 +08003567 ret = ptr_ring_resize_multiple(rings, n,
3568 dev->tx_queue_len, GFP_KERNEL,
Jason Wangfc72d1d2018-01-04 11:14:28 +08003569 tun_ptr_free);
Jason Wang1576d982016-06-30 14:45:36 +08003570
Jason Wang5990a302018-01-04 11:14:27 +08003571 kfree(rings);
Jason Wang1576d982016-06-30 14:45:36 +08003572 return ret;
3573}
3574
3575static int tun_device_event(struct notifier_block *unused,
3576 unsigned long event, void *ptr)
3577{
3578 struct net_device *dev = netdev_notifier_info_to_dev(ptr);
3579 struct tun_struct *tun = netdev_priv(dev);
Fei Li72b319d2019-06-17 21:26:36 +08003580 int i;
Jason Wang1576d982016-06-30 14:45:36 +08003581
Craig Gallek86dfb4ac2016-07-06 18:44:20 -04003582 if (dev->rtnl_link_ops != &tun_link_ops)
3583 return NOTIFY_DONE;
3584
Jason Wang1576d982016-06-30 14:45:36 +08003585 switch (event) {
3586 case NETDEV_CHANGE_TX_QUEUE_LEN:
3587 if (tun_queue_resize(tun))
3588 return NOTIFY_BAD;
3589 break;
Fei Li72b319d2019-06-17 21:26:36 +08003590 case NETDEV_UP:
3591 for (i = 0; i < tun->numqueues; i++) {
3592 struct tun_file *tfile;
3593
3594 tfile = rtnl_dereference(tun->tfiles[i]);
3595 tfile->socket.sk->sk_write_space(tfile->socket.sk);
3596 }
3597 break;
Jason Wang1576d982016-06-30 14:45:36 +08003598 default:
3599 break;
3600 }
3601
3602 return NOTIFY_DONE;
3603}
3604
3605static struct notifier_block tun_notifier_block __read_mostly = {
3606 .notifier_call = tun_device_event,
3607};
Pavel Emelyanov79d17602008-04-16 00:40:46 -07003608
Linus Torvalds1da177e2005-04-16 15:20:36 -07003609static int __init tun_init(void)
3610{
3611 int ret = 0;
3612
Joe Perches6b8a66e2011-03-02 07:18:10 +00003613 pr_info("%s, %s\n", DRV_DESCRIPTION, DRV_VERSION);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003614
Eric W. Biedermanf019a7a2009-01-21 16:02:16 -08003615 ret = rtnl_link_register(&tun_link_ops);
Pavel Emelyanov79d17602008-04-16 00:40:46 -07003616 if (ret) {
Joe Perches6b8a66e2011-03-02 07:18:10 +00003617 pr_err("Can't register link_ops\n");
Eric W. Biedermanf019a7a2009-01-21 16:02:16 -08003618 goto err_linkops;
Pavel Emelyanov79d17602008-04-16 00:40:46 -07003619 }
3620
Linus Torvalds1da177e2005-04-16 15:20:36 -07003621 ret = misc_register(&tun_miscdev);
Pavel Emelyanov79d17602008-04-16 00:40:46 -07003622 if (ret) {
Joe Perches6b8a66e2011-03-02 07:18:10 +00003623 pr_err("Can't register misc device %d\n", TUN_MINOR);
Pavel Emelyanov79d17602008-04-16 00:40:46 -07003624 goto err_misc;
3625 }
Jason Wang1576d982016-06-30 14:45:36 +08003626
Tonghao Zhang5edfbd32017-07-20 02:41:34 -07003627 ret = register_netdevice_notifier(&tun_notifier_block);
3628 if (ret) {
3629 pr_err("Can't register netdevice notifier\n");
3630 goto err_notifier;
3631 }
3632
Eric W. Biedermanf019a7a2009-01-21 16:02:16 -08003633 return 0;
Tonghao Zhang5edfbd32017-07-20 02:41:34 -07003634
3635err_notifier:
3636 misc_deregister(&tun_miscdev);
Pavel Emelyanov79d17602008-04-16 00:40:46 -07003637err_misc:
Eric W. Biedermanf019a7a2009-01-21 16:02:16 -08003638 rtnl_link_unregister(&tun_link_ops);
3639err_linkops:
Linus Torvalds1da177e2005-04-16 15:20:36 -07003640 return ret;
3641}
3642
3643static void tun_cleanup(void)
3644{
Jeff Garzik6aa20a22006-09-13 13:24:59 -04003645 misc_deregister(&tun_miscdev);
Eric W. Biedermanf019a7a2009-01-21 16:02:16 -08003646 rtnl_link_unregister(&tun_link_ops);
Jason Wang1576d982016-06-30 14:45:36 +08003647 unregister_netdevice_notifier(&tun_notifier_block);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003648}
3649
Michael S. Tsirkin05c28282010-01-14 06:17:09 +00003650/* Get an underlying socket object from tun file. Returns error unless file is
3651 * attached to a device. The returned object works like a packet socket, it
3652 * can be used for sock_sendmsg/sock_recvmsg. The caller is responsible for
3653 * holding a reference to the file for as long as the socket is in use. */
3654struct socket *tun_get_socket(struct file *file)
3655{
Jason Wang6e914fc2012-10-31 19:45:58 +00003656 struct tun_file *tfile;
Michael S. Tsirkin05c28282010-01-14 06:17:09 +00003657 if (file->f_op != &tun_fops)
3658 return ERR_PTR(-EINVAL);
Jason Wang6e914fc2012-10-31 19:45:58 +00003659 tfile = file->private_data;
3660 if (!tfile)
Michael S. Tsirkin05c28282010-01-14 06:17:09 +00003661 return ERR_PTR(-EBADFD);
Jason Wang54f968d2012-10-31 19:45:57 +00003662 return &tfile->socket;
Michael S. Tsirkin05c28282010-01-14 06:17:09 +00003663}
3664EXPORT_SYMBOL_GPL(tun_get_socket);
3665
Jason Wang5990a302018-01-04 11:14:27 +08003666struct ptr_ring *tun_get_tx_ring(struct file *file)
Jason Wang83339c62017-05-17 12:14:41 +08003667{
3668 struct tun_file *tfile;
3669
3670 if (file->f_op != &tun_fops)
3671 return ERR_PTR(-EINVAL);
3672 tfile = file->private_data;
3673 if (!tfile)
3674 return ERR_PTR(-EBADFD);
Jason Wang5990a302018-01-04 11:14:27 +08003675 return &tfile->tx_ring;
Jason Wang83339c62017-05-17 12:14:41 +08003676}
Jason Wang5990a302018-01-04 11:14:27 +08003677EXPORT_SYMBOL_GPL(tun_get_tx_ring);
Jason Wang83339c62017-05-17 12:14:41 +08003678
Linus Torvalds1da177e2005-04-16 15:20:36 -07003679module_init(tun_init);
3680module_exit(tun_cleanup);
3681MODULE_DESCRIPTION(DRV_DESCRIPTION);
3682MODULE_AUTHOR(DRV_COPYRIGHT);
3683MODULE_LICENSE("GPL");
3684MODULE_ALIAS_MISCDEV(TUN_MINOR);
Kay Sievers578454f2010-05-20 18:07:20 +02003685MODULE_ALIAS("devname:net/tun");