xref: /linux/net/openvswitch/vport-internal_dev.c (revision 6fdcba32711044c35c0e1b094cbd8f3f0b4472c9)
1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3  * Copyright (c) 2007-2012 Nicira, Inc.
4  */
5 
6 #include <linux/if_vlan.h>
7 #include <linux/kernel.h>
8 #include <linux/netdevice.h>
9 #include <linux/etherdevice.h>
10 #include <linux/ethtool.h>
11 #include <linux/skbuff.h>
12 
13 #include <net/dst.h>
14 #include <net/xfrm.h>
15 #include <net/rtnetlink.h>
16 
17 #include "datapath.h"
18 #include "vport-internal_dev.h"
19 #include "vport-netdev.h"
20 
21 struct internal_dev {
22 	struct vport *vport;
23 };
24 
25 static struct vport_ops ovs_internal_vport_ops;
26 
27 static struct internal_dev *internal_dev_priv(struct net_device *netdev)
28 {
29 	return netdev_priv(netdev);
30 }
31 
32 /* Called with rcu_read_lock_bh. */
33 static netdev_tx_t
34 internal_dev_xmit(struct sk_buff *skb, struct net_device *netdev)
35 {
36 	int len, err;
37 
38 	len = skb->len;
39 	rcu_read_lock();
40 	err = ovs_vport_receive(internal_dev_priv(netdev)->vport, skb, NULL);
41 	rcu_read_unlock();
42 
43 	if (likely(!err)) {
44 		struct pcpu_sw_netstats *tstats = this_cpu_ptr(netdev->tstats);
45 
46 		u64_stats_update_begin(&tstats->syncp);
47 		tstats->tx_bytes += len;
48 		tstats->tx_packets++;
49 		u64_stats_update_end(&tstats->syncp);
50 	} else {
51 		netdev->stats.tx_errors++;
52 	}
53 	return NETDEV_TX_OK;
54 }
55 
56 static int internal_dev_open(struct net_device *netdev)
57 {
58 	netif_start_queue(netdev);
59 	return 0;
60 }
61 
62 static int internal_dev_stop(struct net_device *netdev)
63 {
64 	netif_stop_queue(netdev);
65 	return 0;
66 }
67 
68 static void internal_dev_getinfo(struct net_device *netdev,
69 				 struct ethtool_drvinfo *info)
70 {
71 	strlcpy(info->driver, "openvswitch", sizeof(info->driver));
72 }
73 
74 static const struct ethtool_ops internal_dev_ethtool_ops = {
75 	.get_drvinfo	= internal_dev_getinfo,
76 	.get_link	= ethtool_op_get_link,
77 };
78 
79 static void internal_dev_destructor(struct net_device *dev)
80 {
81 	struct vport *vport = ovs_internal_dev_get_vport(dev);
82 
83 	ovs_vport_free(vport);
84 }
85 
86 static void
87 internal_get_stats(struct net_device *dev, struct rtnl_link_stats64 *stats)
88 {
89 	int i;
90 
91 	memset(stats, 0, sizeof(*stats));
92 	stats->rx_errors  = dev->stats.rx_errors;
93 	stats->tx_errors  = dev->stats.tx_errors;
94 	stats->tx_dropped = dev->stats.tx_dropped;
95 	stats->rx_dropped = dev->stats.rx_dropped;
96 
97 	for_each_possible_cpu(i) {
98 		const struct pcpu_sw_netstats *percpu_stats;
99 		struct pcpu_sw_netstats local_stats;
100 		unsigned int start;
101 
102 		percpu_stats = per_cpu_ptr(dev->tstats, i);
103 
104 		do {
105 			start = u64_stats_fetch_begin_irq(&percpu_stats->syncp);
106 			local_stats = *percpu_stats;
107 		} while (u64_stats_fetch_retry_irq(&percpu_stats->syncp, start));
108 
109 		stats->rx_bytes         += local_stats.rx_bytes;
110 		stats->rx_packets       += local_stats.rx_packets;
111 		stats->tx_bytes         += local_stats.tx_bytes;
112 		stats->tx_packets       += local_stats.tx_packets;
113 	}
114 }
115 
116 static const struct net_device_ops internal_dev_netdev_ops = {
117 	.ndo_open = internal_dev_open,
118 	.ndo_stop = internal_dev_stop,
119 	.ndo_start_xmit = internal_dev_xmit,
120 	.ndo_set_mac_address = eth_mac_addr,
121 	.ndo_get_stats64 = internal_get_stats,
122 };
123 
124 static struct rtnl_link_ops internal_dev_link_ops __read_mostly = {
125 	.kind = "openvswitch",
126 };
127 
128 static void do_setup(struct net_device *netdev)
129 {
130 	ether_setup(netdev);
131 
132 	netdev->max_mtu = ETH_MAX_MTU;
133 
134 	netdev->netdev_ops = &internal_dev_netdev_ops;
135 
136 	netdev->priv_flags &= ~IFF_TX_SKB_SHARING;
137 	netdev->priv_flags |= IFF_LIVE_ADDR_CHANGE | IFF_OPENVSWITCH |
138 			      IFF_NO_QUEUE;
139 	netdev->needs_free_netdev = true;
140 	netdev->priv_destructor = NULL;
141 	netdev->ethtool_ops = &internal_dev_ethtool_ops;
142 	netdev->rtnl_link_ops = &internal_dev_link_ops;
143 
144 	netdev->features = NETIF_F_LLTX | NETIF_F_SG | NETIF_F_FRAGLIST |
145 			   NETIF_F_HIGHDMA | NETIF_F_HW_CSUM |
146 			   NETIF_F_GSO_SOFTWARE | NETIF_F_GSO_ENCAP_ALL;
147 
148 	netdev->vlan_features = netdev->features;
149 	netdev->hw_enc_features = netdev->features;
150 	netdev->features |= NETIF_F_HW_VLAN_CTAG_TX | NETIF_F_HW_VLAN_STAG_TX;
151 	netdev->hw_features = netdev->features & ~NETIF_F_LLTX;
152 
153 	eth_hw_addr_random(netdev);
154 }
155 
156 static struct vport *internal_dev_create(const struct vport_parms *parms)
157 {
158 	struct vport *vport;
159 	struct internal_dev *internal_dev;
160 	struct net_device *dev;
161 	int err;
162 
163 	vport = ovs_vport_alloc(0, &ovs_internal_vport_ops, parms);
164 	if (IS_ERR(vport)) {
165 		err = PTR_ERR(vport);
166 		goto error;
167 	}
168 
169 	dev = alloc_netdev(sizeof(struct internal_dev),
170 			   parms->name, NET_NAME_USER, do_setup);
171 	vport->dev = dev;
172 	if (!vport->dev) {
173 		err = -ENOMEM;
174 		goto error_free_vport;
175 	}
176 	vport->dev->tstats = netdev_alloc_pcpu_stats(struct pcpu_sw_netstats);
177 	if (!vport->dev->tstats) {
178 		err = -ENOMEM;
179 		goto error_free_netdev;
180 	}
181 
182 	dev_net_set(vport->dev, ovs_dp_get_net(vport->dp));
183 	internal_dev = internal_dev_priv(vport->dev);
184 	internal_dev->vport = vport;
185 
186 	/* Restrict bridge port to current netns. */
187 	if (vport->port_no == OVSP_LOCAL)
188 		vport->dev->features |= NETIF_F_NETNS_LOCAL;
189 
190 	rtnl_lock();
191 	err = register_netdevice(vport->dev);
192 	if (err)
193 		goto error_unlock;
194 	vport->dev->priv_destructor = internal_dev_destructor;
195 
196 	dev_set_promiscuity(vport->dev, 1);
197 	rtnl_unlock();
198 	netif_start_queue(vport->dev);
199 
200 	return vport;
201 
202 error_unlock:
203 	rtnl_unlock();
204 	free_percpu(dev->tstats);
205 error_free_netdev:
206 	free_netdev(dev);
207 error_free_vport:
208 	ovs_vport_free(vport);
209 error:
210 	return ERR_PTR(err);
211 }
212 
213 static void internal_dev_destroy(struct vport *vport)
214 {
215 	netif_stop_queue(vport->dev);
216 	rtnl_lock();
217 	dev_set_promiscuity(vport->dev, -1);
218 
219 	/* unregister_netdevice() waits for an RCU grace period. */
220 	unregister_netdevice(vport->dev);
221 	free_percpu(vport->dev->tstats);
222 	rtnl_unlock();
223 }
224 
225 static netdev_tx_t internal_dev_recv(struct sk_buff *skb)
226 {
227 	struct net_device *netdev = skb->dev;
228 	struct pcpu_sw_netstats *stats;
229 
230 	if (unlikely(!(netdev->flags & IFF_UP))) {
231 		kfree_skb(skb);
232 		netdev->stats.rx_dropped++;
233 		return NETDEV_TX_OK;
234 	}
235 
236 	skb_dst_drop(skb);
237 	nf_reset_ct(skb);
238 	secpath_reset(skb);
239 
240 	skb->pkt_type = PACKET_HOST;
241 	skb->protocol = eth_type_trans(skb, netdev);
242 	skb_postpull_rcsum(skb, eth_hdr(skb), ETH_HLEN);
243 
244 	stats = this_cpu_ptr(netdev->tstats);
245 	u64_stats_update_begin(&stats->syncp);
246 	stats->rx_packets++;
247 	stats->rx_bytes += skb->len;
248 	u64_stats_update_end(&stats->syncp);
249 
250 	netif_rx(skb);
251 	return NETDEV_TX_OK;
252 }
253 
254 static struct vport_ops ovs_internal_vport_ops = {
255 	.type		= OVS_VPORT_TYPE_INTERNAL,
256 	.create		= internal_dev_create,
257 	.destroy	= internal_dev_destroy,
258 	.send		= internal_dev_recv,
259 };
260 
261 int ovs_is_internal_dev(const struct net_device *netdev)
262 {
263 	return netdev->netdev_ops == &internal_dev_netdev_ops;
264 }
265 
266 struct vport *ovs_internal_dev_get_vport(struct net_device *netdev)
267 {
268 	if (!ovs_is_internal_dev(netdev))
269 		return NULL;
270 
271 	return internal_dev_priv(netdev)->vport;
272 }
273 
274 int ovs_internal_dev_rtnl_link_register(void)
275 {
276 	int err;
277 
278 	err = rtnl_link_register(&internal_dev_link_ops);
279 	if (err < 0)
280 		return err;
281 
282 	err = ovs_vport_ops_register(&ovs_internal_vport_ops);
283 	if (err < 0)
284 		rtnl_link_unregister(&internal_dev_link_ops);
285 
286 	return err;
287 }
288 
289 void ovs_internal_dev_rtnl_link_unregister(void)
290 {
291 	ovs_vport_ops_unregister(&ovs_internal_vport_ops);
292 	rtnl_link_unregister(&internal_dev_link_ops);
293 }
294