Commit | Line | Data |
---|---|---|
ccb1352e | 1 | /* |
caf2ee14 | 2 | * Copyright (c) 2007-2012 Nicira, Inc. |
ccb1352e JG |
3 | * |
4 | * This program is free software; you can redistribute it and/or | |
5 | * modify it under the terms of version 2 of the GNU General Public | |
6 | * License as published by the Free Software Foundation. | |
7 | * | |
8 | * This program is distributed in the hope that it will be useful, but | |
9 | * WITHOUT ANY WARRANTY; without even the implied warranty of | |
10 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU | |
11 | * General Public License for more details. | |
12 | * | |
13 | * You should have received a copy of the GNU General Public License | |
14 | * along with this program; if not, write to the Free Software | |
15 | * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA | |
16 | * 02110-1301, USA | |
17 | */ | |
18 | ||
19 | #include <linux/hardirq.h> | |
20 | #include <linux/if_vlan.h> | |
21 | #include <linux/kernel.h> | |
22 | #include <linux/netdevice.h> | |
23 | #include <linux/etherdevice.h> | |
24 | #include <linux/ethtool.h> | |
25 | #include <linux/skbuff.h> | |
ccb1352e | 26 | |
7fe99e2d JG |
27 | #include <net/dst.h> |
28 | #include <net/xfrm.h> | |
5b9e7e16 | 29 | #include <net/rtnetlink.h> |
7fe99e2d | 30 | |
ccb1352e JG |
31 | #include "datapath.h" |
32 | #include "vport-internal_dev.h" | |
33 | #include "vport-netdev.h" | |
34 | ||
35 | struct internal_dev { | |
36 | struct vport *vport; | |
37 | }; | |
38 | ||
62b9c8d0 TG |
39 | static struct vport_ops ovs_internal_vport_ops; |
40 | ||
ccb1352e JG |
41 | static struct internal_dev *internal_dev_priv(struct net_device *netdev) |
42 | { | |
43 | return netdev_priv(netdev); | |
44 | } | |
45 | ||
ccb1352e JG |
46 | /* Called with rcu_read_lock_bh. */ |
47 | static int internal_dev_xmit(struct sk_buff *skb, struct net_device *netdev) | |
48 | { | |
8c876639 PS |
49 | int len, err; |
50 | ||
51 | len = skb->len; | |
ccb1352e | 52 | rcu_read_lock(); |
8c876639 | 53 | err = ovs_vport_receive(internal_dev_priv(netdev)->vport, skb, NULL); |
ccb1352e | 54 | rcu_read_unlock(); |
8c876639 PS |
55 | |
56 | if (likely(!err)) { | |
57 | struct pcpu_sw_netstats *tstats = this_cpu_ptr(netdev->tstats); | |
58 | ||
59 | u64_stats_update_begin(&tstats->syncp); | |
60 | tstats->tx_bytes += len; | |
61 | tstats->tx_packets++; | |
62 | u64_stats_update_end(&tstats->syncp); | |
63 | } else { | |
64 | netdev->stats.tx_errors++; | |
65 | } | |
ccb1352e JG |
66 | return 0; |
67 | } | |
68 | ||
69 | static int internal_dev_open(struct net_device *netdev) | |
70 | { | |
71 | netif_start_queue(netdev); | |
72 | return 0; | |
73 | } | |
74 | ||
75 | static int internal_dev_stop(struct net_device *netdev) | |
76 | { | |
77 | netif_stop_queue(netdev); | |
78 | return 0; | |
79 | } | |
80 | ||
81 | static void internal_dev_getinfo(struct net_device *netdev, | |
82 | struct ethtool_drvinfo *info) | |
83 | { | |
7826d43f | 84 | strlcpy(info->driver, "openvswitch", sizeof(info->driver)); |
ccb1352e JG |
85 | } |
86 | ||
87 | static const struct ethtool_ops internal_dev_ethtool_ops = { | |
88 | .get_drvinfo = internal_dev_getinfo, | |
89 | .get_link = ethtool_op_get_link, | |
90 | }; | |
91 | ||
92 | static int internal_dev_change_mtu(struct net_device *netdev, int new_mtu) | |
93 | { | |
94 | if (new_mtu < 68) | |
95 | return -EINVAL; | |
96 | ||
97 | netdev->mtu = new_mtu; | |
98 | return 0; | |
99 | } | |
100 | ||
101 | static void internal_dev_destructor(struct net_device *dev) | |
102 | { | |
103 | struct vport *vport = ovs_internal_dev_get_vport(dev); | |
104 | ||
105 | ovs_vport_free(vport); | |
106 | free_netdev(dev); | |
107 | } | |
108 | ||
1241365f JM |
109 | static struct rtnl_link_stats64 * |
110 | internal_get_stats(struct net_device *dev, struct rtnl_link_stats64 *stats) | |
111 | { | |
112 | int i; | |
113 | ||
114 | memset(stats, 0, sizeof(*stats)); | |
115 | stats->rx_errors = dev->stats.rx_errors; | |
116 | stats->tx_errors = dev->stats.tx_errors; | |
117 | stats->tx_dropped = dev->stats.tx_dropped; | |
118 | stats->rx_dropped = dev->stats.rx_dropped; | |
119 | ||
120 | for_each_possible_cpu(i) { | |
121 | const struct pcpu_sw_netstats *percpu_stats; | |
122 | struct pcpu_sw_netstats local_stats; | |
123 | unsigned int start; | |
124 | ||
125 | percpu_stats = per_cpu_ptr(dev->tstats, i); | |
126 | ||
127 | do { | |
128 | start = u64_stats_fetch_begin_irq(&percpu_stats->syncp); | |
129 | local_stats = *percpu_stats; | |
130 | } while (u64_stats_fetch_retry_irq(&percpu_stats->syncp, start)); | |
131 | ||
132 | stats->rx_bytes += local_stats.rx_bytes; | |
133 | stats->rx_packets += local_stats.rx_packets; | |
134 | stats->tx_bytes += local_stats.tx_bytes; | |
135 | stats->tx_packets += local_stats.tx_packets; | |
136 | } | |
137 | ||
138 | return stats; | |
139 | } | |
140 | ||
e014e846 | 141 | static void internal_set_rx_headroom(struct net_device *dev, int new_hr) |
3a927bc7 PA |
142 | { |
143 | dev->needed_headroom = new_hr; | |
144 | } | |
145 | ||
ccb1352e JG |
146 | static const struct net_device_ops internal_dev_netdev_ops = { |
147 | .ndo_open = internal_dev_open, | |
148 | .ndo_stop = internal_dev_stop, | |
149 | .ndo_start_xmit = internal_dev_xmit, | |
9151991c | 150 | .ndo_set_mac_address = eth_mac_addr, |
ccb1352e | 151 | .ndo_change_mtu = internal_dev_change_mtu, |
1241365f | 152 | .ndo_get_stats64 = internal_get_stats, |
3a927bc7 | 153 | .ndo_set_rx_headroom = internal_set_rx_headroom, |
ccb1352e JG |
154 | }; |
155 | ||
5b9e7e16 JP |
156 | static struct rtnl_link_ops internal_dev_link_ops __read_mostly = { |
157 | .kind = "openvswitch", | |
158 | }; | |
159 | ||
ccb1352e JG |
160 | static void do_setup(struct net_device *netdev) |
161 | { | |
162 | ether_setup(netdev); | |
163 | ||
164 | netdev->netdev_ops = &internal_dev_netdev_ops; | |
165 | ||
166 | netdev->priv_flags &= ~IFF_TX_SKB_SHARING; | |
3a927bc7 | 167 | netdev->priv_flags |= IFF_LIVE_ADDR_CHANGE | IFF_OPENVSWITCH | |
4272cc51 | 168 | IFF_PHONY_HEADROOM | IFF_NO_QUEUE; |
ccb1352e | 169 | netdev->destructor = internal_dev_destructor; |
7ad24ea4 | 170 | netdev->ethtool_ops = &internal_dev_ethtool_ops; |
5b9e7e16 | 171 | netdev->rtnl_link_ops = &internal_dev_link_ops; |
ccb1352e JG |
172 | |
173 | netdev->features = NETIF_F_LLTX | NETIF_F_SG | NETIF_F_FRAGLIST | | |
f6eec614 PS |
174 | NETIF_F_HIGHDMA | NETIF_F_HW_CSUM | |
175 | NETIF_F_GSO_SOFTWARE | NETIF_F_GSO_ENCAP_ALL; | |
ccb1352e JG |
176 | |
177 | netdev->vlan_features = netdev->features; | |
f6eec614 | 178 | netdev->hw_enc_features = netdev->features; |
f646968f | 179 | netdev->features |= NETIF_F_HW_VLAN_CTAG_TX; |
ccb1352e | 180 | netdev->hw_features = netdev->features & ~NETIF_F_LLTX; |
f6eec614 | 181 | |
7ce5d222 | 182 | eth_hw_addr_random(netdev); |
ccb1352e JG |
183 | } |
184 | ||
185 | static struct vport *internal_dev_create(const struct vport_parms *parms) | |
186 | { | |
187 | struct vport *vport; | |
ccb1352e JG |
188 | struct internal_dev *internal_dev; |
189 | int err; | |
190 | ||
be4ace6e | 191 | vport = ovs_vport_alloc(0, &ovs_internal_vport_ops, parms); |
ccb1352e JG |
192 | if (IS_ERR(vport)) { |
193 | err = PTR_ERR(vport); | |
194 | goto error; | |
195 | } | |
196 | ||
be4ace6e TG |
197 | vport->dev = alloc_netdev(sizeof(struct internal_dev), |
198 | parms->name, NET_NAME_UNKNOWN, do_setup); | |
199 | if (!vport->dev) { | |
ccb1352e JG |
200 | err = -ENOMEM; |
201 | goto error_free_vport; | |
202 | } | |
1241365f JM |
203 | vport->dev->tstats = netdev_alloc_pcpu_stats(struct pcpu_sw_netstats); |
204 | if (!vport->dev->tstats) { | |
205 | err = -ENOMEM; | |
206 | goto error_free_netdev; | |
207 | } | |
3a927bc7 | 208 | vport->dev->needed_headroom = vport->dp->max_headroom; |
ccb1352e | 209 | |
be4ace6e TG |
210 | dev_net_set(vport->dev, ovs_dp_get_net(vport->dp)); |
211 | internal_dev = internal_dev_priv(vport->dev); | |
ccb1352e JG |
212 | internal_dev->vport = vport; |
213 | ||
46df7b81 PS |
214 | /* Restrict bridge port to current netns. */ |
215 | if (vport->port_no == OVSP_LOCAL) | |
be4ace6e | 216 | vport->dev->features |= NETIF_F_NETNS_LOCAL; |
46df7b81 | 217 | |
8e4e1713 | 218 | rtnl_lock(); |
be4ace6e | 219 | err = register_netdevice(vport->dev); |
ccb1352e | 220 | if (err) |
1241365f | 221 | goto error_unlock; |
ccb1352e | 222 | |
be4ace6e | 223 | dev_set_promiscuity(vport->dev, 1); |
8e4e1713 | 224 | rtnl_unlock(); |
be4ace6e | 225 | netif_start_queue(vport->dev); |
ccb1352e JG |
226 | |
227 | return vport; | |
228 | ||
1241365f | 229 | error_unlock: |
8e4e1713 | 230 | rtnl_unlock(); |
1241365f JM |
231 | free_percpu(vport->dev->tstats); |
232 | error_free_netdev: | |
be4ace6e | 233 | free_netdev(vport->dev); |
ccb1352e JG |
234 | error_free_vport: |
235 | ovs_vport_free(vport); | |
236 | error: | |
237 | return ERR_PTR(err); | |
238 | } | |
239 | ||
240 | static void internal_dev_destroy(struct vport *vport) | |
241 | { | |
be4ace6e | 242 | netif_stop_queue(vport->dev); |
8e4e1713 | 243 | rtnl_lock(); |
be4ace6e | 244 | dev_set_promiscuity(vport->dev, -1); |
ccb1352e JG |
245 | |
246 | /* unregister_netdevice() waits for an RCU grace period. */ | |
be4ace6e | 247 | unregister_netdevice(vport->dev); |
1241365f | 248 | free_percpu(vport->dev->tstats); |
8e4e1713 | 249 | rtnl_unlock(); |
ccb1352e JG |
250 | } |
251 | ||
aec15924 | 252 | static netdev_tx_t internal_dev_recv(struct sk_buff *skb) |
ccb1352e | 253 | { |
aec15924 | 254 | struct net_device *netdev = skb->dev; |
8c876639 | 255 | struct pcpu_sw_netstats *stats; |
ccb1352e | 256 | |
e1f9c356 CL |
257 | if (unlikely(!(netdev->flags & IFF_UP))) { |
258 | kfree_skb(skb); | |
8c876639 | 259 | netdev->stats.rx_dropped++; |
aec15924 | 260 | return NETDEV_TX_OK; |
e1f9c356 CL |
261 | } |
262 | ||
7fe99e2d JG |
263 | skb_dst_drop(skb); |
264 | nf_reset(skb); | |
265 | secpath_reset(skb); | |
266 | ||
ccb1352e JG |
267 | skb->pkt_type = PACKET_HOST; |
268 | skb->protocol = eth_type_trans(skb, netdev); | |
b34df5e8 | 269 | skb_postpull_rcsum(skb, eth_hdr(skb), ETH_HLEN); |
ccb1352e | 270 | |
8c876639 PS |
271 | stats = this_cpu_ptr(netdev->tstats); |
272 | u64_stats_update_begin(&stats->syncp); | |
273 | stats->rx_packets++; | |
274 | stats->rx_bytes += skb->len; | |
275 | u64_stats_update_end(&stats->syncp); | |
ccb1352e | 276 | |
8c876639 | 277 | netif_rx(skb); |
aec15924 | 278 | return NETDEV_TX_OK; |
ccb1352e JG |
279 | } |
280 | ||
62b9c8d0 | 281 | static struct vport_ops ovs_internal_vport_ops = { |
ccb1352e JG |
282 | .type = OVS_VPORT_TYPE_INTERNAL, |
283 | .create = internal_dev_create, | |
284 | .destroy = internal_dev_destroy, | |
ccb1352e JG |
285 | .send = internal_dev_recv, |
286 | }; | |
287 | ||
288 | int ovs_is_internal_dev(const struct net_device *netdev) | |
289 | { | |
290 | return netdev->netdev_ops == &internal_dev_netdev_ops; | |
291 | } | |
292 | ||
293 | struct vport *ovs_internal_dev_get_vport(struct net_device *netdev) | |
294 | { | |
295 | if (!ovs_is_internal_dev(netdev)) | |
296 | return NULL; | |
297 | ||
298 | return internal_dev_priv(netdev)->vport; | |
299 | } | |
5b9e7e16 JP |
300 | |
301 | int ovs_internal_dev_rtnl_link_register(void) | |
302 | { | |
62b9c8d0 TG |
303 | int err; |
304 | ||
305 | err = rtnl_link_register(&internal_dev_link_ops); | |
306 | if (err < 0) | |
307 | return err; | |
308 | ||
309 | err = ovs_vport_ops_register(&ovs_internal_vport_ops); | |
310 | if (err < 0) | |
311 | rtnl_link_unregister(&internal_dev_link_ops); | |
312 | ||
313 | return err; | |
5b9e7e16 JP |
314 | } |
315 | ||
316 | void ovs_internal_dev_rtnl_link_unregister(void) | |
317 | { | |
62b9c8d0 | 318 | ovs_vport_ops_unregister(&ovs_internal_vport_ops); |
5b9e7e16 JP |
319 | rtnl_link_unregister(&internal_dev_link_ops); |
320 | } |