1// SPDX-License-Identifier: GPL-2.0-or-later
2/*
3 * INET		An implementation of the TCP/IP protocol suite for the LINUX
4 *		operating system.  INET is implemented using the  BSD Socket
5 *		interface as the means of communication with the user level.
6 *
7 *		Pseudo-driver for the loopback interface.
8 *
9 * Version:	@(#)loopback.c	1.0.4b	08/16/93
10 *
11 * Authors:	Ross Biro
12 *		Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
13 *		Donald Becker, <becker@scyld.com>
14 *
15 *		Alan Cox	:	Fixed oddments for NET3.014
16 *		Alan Cox	:	Rejig for NET3.029 snap #3
17 *		Alan Cox	:	Fixed NET3.029 bugs and sped up
18 *		Larry McVoy	:	Tiny tweak to double performance
19 *		Alan Cox	:	Backed out LMV's tweak - the linux mm
20 *					can't take it...
21 *              Michael Griffith:       Don't bother computing the checksums
22 *                                      on packets received on the loopback
23 *                                      interface.
24 *		Alexey Kuznetsov:	Potential hang under some extreme
25 *					cases removed.
26 */
27#include <linux/kernel.h>
28#include <linux/jiffies.h>
29#include <linux/module.h>
30#include <linux/interrupt.h>
31#include <linux/fs.h>
32#include <linux/types.h>
33#include <linux/string.h>
34#include <linux/socket.h>
35#include <linux/errno.h>
36#include <linux/fcntl.h>
37#include <linux/in.h>
38
39#include <linux/uaccess.h>
40#include <linux/io.h>
41
42#include <linux/inet.h>
43#include <linux/netdevice.h>
44#include <linux/etherdevice.h>
45#include <linux/skbuff.h>
46#include <linux/ethtool.h>
47#include <net/sch_generic.h>
48#include <net/sock.h>
49#include <net/checksum.h>
50#include <linux/if_ether.h>	/* For the statistics structure. */
51#include <linux/if_arp.h>	/* For ARPHRD_ETHER */
52#include <linux/ip.h>
53#include <linux/tcp.h>
54#include <linux/percpu.h>
55#include <linux/net_tstamp.h>
56#include <net/net_namespace.h>
57#include <linux/u64_stats_sync.h>
58
59/* blackhole_netdev - a device used for dsts that are marked expired!
60 * This is global device (instead of per-net-ns) since it's not needed
61 * to be per-ns and gets initialized at boot time.
62 */
63struct net_device *blackhole_netdev;
64EXPORT_SYMBOL(blackhole_netdev);
65
66/* The higher levels take care of making this non-reentrant (it's
67 * called with bh's disabled).
68 */
69static netdev_tx_t loopback_xmit(struct sk_buff *skb,
70				 struct net_device *dev)
71{
72	int len;
73
74	skb_tx_timestamp(skb);
75
76	/* do not fool net_timestamp_check() with various clock bases */
77	skb_clear_tstamp(skb);
78
79	skb_orphan(skb);
80
81	/* Before queueing this packet to __netif_rx(),
82	 * make sure dst is refcounted.
83	 */
84	skb_dst_force(skb);
85
86	skb->protocol = eth_type_trans(skb, dev);
87
88	len = skb->len;
89	if (likely(__netif_rx(skb) == NET_RX_SUCCESS))
90		dev_lstats_add(dev, len);
91
92	return NETDEV_TX_OK;
93}
94
95void dev_lstats_read(struct net_device *dev, u64 *packets, u64 *bytes)
96{
97	int i;
98
99	*packets = 0;
100	*bytes = 0;
101
102	for_each_possible_cpu(i) {
103		const struct pcpu_lstats *lb_stats;
104		u64 tbytes, tpackets;
105		unsigned int start;
106
107		lb_stats = per_cpu_ptr(dev->lstats, i);
108		do {
109			start = u64_stats_fetch_begin(&lb_stats->syncp);
110			tpackets = u64_stats_read(&lb_stats->packets);
111			tbytes = u64_stats_read(&lb_stats->bytes);
112		} while (u64_stats_fetch_retry(&lb_stats->syncp, start));
113		*bytes   += tbytes;
114		*packets += tpackets;
115	}
116}
117EXPORT_SYMBOL(dev_lstats_read);
118
119static void loopback_get_stats64(struct net_device *dev,
120				 struct rtnl_link_stats64 *stats)
121{
122	u64 packets, bytes;
123
124	dev_lstats_read(dev, &packets, &bytes);
125
126	stats->rx_packets = packets;
127	stats->tx_packets = packets;
128	stats->rx_bytes   = bytes;
129	stats->tx_bytes   = bytes;
130}
131
132static u32 always_on(struct net_device *dev)
133{
134	return 1;
135}
136
137static const struct ethtool_ops loopback_ethtool_ops = {
138	.get_link		= always_on,
139	.get_ts_info		= ethtool_op_get_ts_info,
140};
141
142static int loopback_dev_init(struct net_device *dev)
143{
144	dev->lstats = netdev_alloc_pcpu_stats(struct pcpu_lstats);
145	if (!dev->lstats)
146		return -ENOMEM;
147	netdev_lockdep_set_classes(dev);
148	return 0;
149}
150
151static void loopback_dev_free(struct net_device *dev)
152{
153	dev_net(dev)->loopback_dev = NULL;
154	free_percpu(dev->lstats);
155}
156
157static const struct net_device_ops loopback_ops = {
158	.ndo_init        = loopback_dev_init,
159	.ndo_start_xmit  = loopback_xmit,
160	.ndo_get_stats64 = loopback_get_stats64,
161	.ndo_set_mac_address = eth_mac_addr,
162};
163
164static void gen_lo_setup(struct net_device *dev,
165			 unsigned int mtu,
166			 const struct ethtool_ops *eth_ops,
167			 const struct header_ops *hdr_ops,
168			 const struct net_device_ops *dev_ops,
169			 void (*dev_destructor)(struct net_device *dev))
170{
171	dev->mtu		= mtu;
172	dev->hard_header_len	= ETH_HLEN;	/* 14	*/
173	dev->min_header_len	= ETH_HLEN;	/* 14	*/
174	dev->addr_len		= ETH_ALEN;	/* 6	*/
175	dev->type		= ARPHRD_LOOPBACK;	/* 0x0001*/
176	dev->flags		= IFF_LOOPBACK;
177	dev->priv_flags		|= IFF_LIVE_ADDR_CHANGE | IFF_NO_QUEUE;
178	netif_keep_dst(dev);
179	dev->hw_features	= NETIF_F_GSO_SOFTWARE;
180	dev->features		= NETIF_F_SG | NETIF_F_FRAGLIST
181		| NETIF_F_GSO_SOFTWARE
182		| NETIF_F_HW_CSUM
183		| NETIF_F_RXCSUM
184		| NETIF_F_SCTP_CRC
185		| NETIF_F_HIGHDMA
186		| NETIF_F_LLTX
187		| NETIF_F_NETNS_LOCAL
188		| NETIF_F_VLAN_CHALLENGED
189		| NETIF_F_LOOPBACK;
190	dev->ethtool_ops	= eth_ops;
191	dev->header_ops		= hdr_ops;
192	dev->netdev_ops		= dev_ops;
193	dev->needs_free_netdev	= true;
194	dev->priv_destructor	= dev_destructor;
195
196	netif_set_tso_max_size(dev, GSO_MAX_SIZE);
197}
198
199/* The loopback device is special. There is only one instance
200 * per network namespace.
201 */
202static void loopback_setup(struct net_device *dev)
203{
204	gen_lo_setup(dev, (64 * 1024), &loopback_ethtool_ops, &eth_header_ops,
205		     &loopback_ops, loopback_dev_free);
206}
207
208/* Setup and register the loopback device. */
209static __net_init int loopback_net_init(struct net *net)
210{
211	struct net_device *dev;
212	int err;
213
214	err = -ENOMEM;
215	dev = alloc_netdev(0, "lo", NET_NAME_PREDICTABLE, loopback_setup);
216	if (!dev)
217		goto out;
218
219	dev_net_set(dev, net);
220	err = register_netdev(dev);
221	if (err)
222		goto out_free_netdev;
223
224	BUG_ON(dev->ifindex != LOOPBACK_IFINDEX);
225	net->loopback_dev = dev;
226	return 0;
227
228out_free_netdev:
229	free_netdev(dev);
230out:
231	if (net_eq(net, &init_net))
232		panic("loopback: Failed to register netdevice: %d\n", err);
233	return err;
234}
235
236/* Registered in net/core/dev.c */
237struct pernet_operations __net_initdata loopback_net_ops = {
238	.init = loopback_net_init,
239};
240
241/* blackhole netdevice */
242static netdev_tx_t blackhole_netdev_xmit(struct sk_buff *skb,
243					 struct net_device *dev)
244{
245	kfree_skb(skb);
246	net_warn_ratelimited("%s(): Dropping skb.\n", __func__);
247	return NETDEV_TX_OK;
248}
249
250static const struct net_device_ops blackhole_netdev_ops = {
251	.ndo_start_xmit = blackhole_netdev_xmit,
252};
253
254/* This is a dst-dummy device used specifically for invalidated
255 * DSTs and unlike loopback, this is not per-ns.
256 */
257static void blackhole_netdev_setup(struct net_device *dev)
258{
259	gen_lo_setup(dev, ETH_MIN_MTU, NULL, NULL, &blackhole_netdev_ops, NULL);
260}
261
262/* Setup and register the blackhole_netdev. */
263static int __init blackhole_netdev_init(void)
264{
265	blackhole_netdev = alloc_netdev(0, "blackhole_dev", NET_NAME_UNKNOWN,
266					blackhole_netdev_setup);
267	if (!blackhole_netdev)
268		return -ENOMEM;
269
270	rtnl_lock();
271	dev_init_scheduler(blackhole_netdev);
272	dev_activate(blackhole_netdev);
273	rtnl_unlock();
274
275	blackhole_netdev->flags |= IFF_UP | IFF_RUNNING;
276	dev_net_set(blackhole_netdev, &init_net);
277
278	return 0;
279}
280
281device_initcall(blackhole_netdev_init);
282