1 // SPDX-License-Identifier: GPL-2.0-only 2 /* 3 * Copyright (c) 2007-2012 Nicira, Inc. 4 */ 5 6 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt 7 8 #include <linux/if_arp.h> 9 #include <linux/if_bridge.h> 10 #include <linux/if_vlan.h> 11 #include <linux/kernel.h> 12 #include <linux/llc.h> 13 #include <linux/rtnetlink.h> 14 #include <linux/skbuff.h> 15 #include <linux/openvswitch.h> 16 #include <linux/export.h> 17 18 #include <net/ip_tunnels.h> 19 #include <net/rtnetlink.h> 20 21 #include "datapath.h" 22 #include "vport.h" 23 #include "vport-internal_dev.h" 24 #include "vport-netdev.h" 25 26 static struct vport_ops ovs_netdev_vport_ops; 27 28 /* Must be called with rcu_read_lock. */ 29 static void netdev_port_receive(struct sk_buff *skb) 30 { 31 struct vport *vport; 32 33 vport = ovs_netdev_get_vport(skb->dev); 34 if (unlikely(!vport)) 35 goto error; 36 37 if (unlikely(skb_warn_if_lro(skb))) 38 goto error; 39 40 /* Make our own copy of the packet. Otherwise we will mangle the 41 * packet for anyone who came before us (e.g. tcpdump via AF_PACKET). 42 */ 43 skb = skb_share_check(skb, GFP_ATOMIC); 44 if (unlikely(!skb)) 45 return; 46 47 if (skb->dev->type == ARPHRD_ETHER) 48 skb_push_rcsum(skb, ETH_HLEN); 49 50 ovs_vport_receive(vport, skb, skb_tunnel_info(skb)); 51 return; 52 error: 53 kfree_skb(skb); 54 } 55 56 /* Called with rcu_read_lock and bottom-halves disabled. */ 57 static rx_handler_result_t netdev_frame_hook(struct sk_buff **pskb) 58 { 59 struct sk_buff *skb = *pskb; 60 61 if (unlikely(skb->pkt_type == PACKET_LOOPBACK)) 62 return RX_HANDLER_PASS; 63 64 netdev_port_receive(skb); 65 return RX_HANDLER_CONSUMED; 66 } 67 68 static struct net_device *get_dpdev(const struct datapath *dp) 69 { 70 struct vport *local; 71 72 local = ovs_vport_ovsl(dp, OVSP_LOCAL); 73 return local->dev; 74 } 75 76 static struct vport *ovs_netdev_link(struct vport *vport) 77 { 78 int err; 79 80 if (WARN_ON_ONCE(!vport->dev)) { 81 err = -ENODEV; 82 goto error_free_vport; 83 } 84 85 rtnl_lock(); 86 /* Do not link devices that are not registered to avoid a potential 87 * race with the NETDEV_UNREGISTER notification in dp_device_event(). 88 */ 89 if (vport->dev->reg_state != NETREG_REGISTERED) { 90 err = -ENODEV; 91 goto error_put_unlock; 92 } 93 94 err = netdev_master_upper_dev_link(vport->dev, 95 get_dpdev(vport->dp), 96 NULL, NULL, NULL); 97 if (err) 98 goto error_put_unlock; 99 100 err = netdev_rx_handler_register(vport->dev, netdev_frame_hook, 101 vport); 102 if (err) 103 goto error_master_upper_dev_unlink; 104 105 dev_disable_lro(vport->dev); 106 dev_set_promiscuity(vport->dev, 1); 107 vport->dev->priv_flags |= IFF_OVS_DATAPATH; 108 rtnl_unlock(); 109 110 return vport; 111 112 error_master_upper_dev_unlink: 113 netdev_upper_dev_unlink(vport->dev, get_dpdev(vport->dp)); 114 error_put_unlock: 115 netdev_put(vport->dev, &vport->dev_tracker); 116 rtnl_unlock(); 117 error_free_vport: 118 ovs_vport_free(vport); 119 return ERR_PTR(err); 120 } 121 122 static struct vport *netdev_create(const struct vport_parms *parms) 123 { 124 struct vport *vport; 125 int err; 126 127 vport = ovs_vport_alloc(0, &ovs_netdev_vport_ops, parms); 128 if (IS_ERR(vport)) 129 return vport; 130 131 vport->dev = dev_get_by_name(ovs_dp_get_net(vport->dp), parms->name); 132 if (!vport->dev) { 133 err = -ENODEV; 134 goto error_free_vport; 135 } 136 netdev_tracker_alloc(vport->dev, &vport->dev_tracker, GFP_KERNEL); 137 138 /* Ensure that the provided name is not an alias. */ 139 if (strcmp(parms->name, ovs_vport_name(vport))) { 140 err = -ENODEV; 141 goto error_put; 142 } 143 144 if (vport->dev->flags & IFF_LOOPBACK || 145 (vport->dev->type != ARPHRD_ETHER && 146 vport->dev->type != ARPHRD_NONE) || 147 ovs_is_internal_dev(vport->dev)) { 148 err = -EINVAL; 149 goto error_put; 150 } 151 152 return ovs_netdev_link(vport); 153 error_put: 154 netdev_put(vport->dev, &vport->dev_tracker); 155 error_free_vport: 156 ovs_vport_free(vport); 157 return ERR_PTR(err); 158 } 159 160 static void vport_netdev_free(struct rcu_head *rcu) 161 { 162 struct vport *vport = container_of(rcu, struct vport, rcu); 163 164 netdev_put(vport->dev, &vport->dev_tracker); 165 ovs_vport_free(vport); 166 } 167 168 void ovs_netdev_detach_dev(struct vport *vport) 169 { 170 ASSERT_RTNL(); 171 netdev_rx_handler_unregister(vport->dev); 172 netdev_upper_dev_unlink(vport->dev, 173 netdev_master_upper_dev_get(vport->dev)); 174 dev_set_promiscuity(vport->dev, -1); 175 176 /* paired with smp_mb() in netdev_destroy() */ 177 smp_wmb(); 178 179 vport->dev->priv_flags &= ~IFF_OVS_DATAPATH; 180 } 181 182 static void netdev_destroy(struct vport *vport) 183 { 184 /* When called from ovs_db_notify_wq() after a dp_device_event(), the 185 * port has already been detached, so we can avoid taking the RTNL by 186 * checking this first. 187 */ 188 if (netif_is_ovs_port(vport->dev)) { 189 rtnl_lock(); 190 /* Check again while holding the lock to ensure we don't race 191 * with the netdev notifier and detach twice. 192 */ 193 if (netif_is_ovs_port(vport->dev)) 194 ovs_netdev_detach_dev(vport); 195 rtnl_unlock(); 196 } 197 198 /* paired with smp_wmb() in ovs_netdev_detach_dev() */ 199 smp_mb(); 200 201 call_rcu(&vport->rcu, vport_netdev_free); 202 } 203 204 /* Returns null if this device is not attached to a datapath. */ 205 struct vport *ovs_netdev_get_vport(struct net_device *dev) 206 { 207 if (likely(netif_is_ovs_port(dev))) 208 return (struct vport *) 209 rcu_dereference_rtnl(dev->rx_handler_data); 210 else 211 return NULL; 212 } 213 214 static struct vport_ops ovs_netdev_vport_ops = { 215 .type = OVS_VPORT_TYPE_NETDEV, 216 .create = netdev_create, 217 .destroy = netdev_destroy, 218 .send = dev_queue_xmit, 219 }; 220 221 int __init ovs_netdev_init(void) 222 { 223 return ovs_vport_ops_register(&ovs_netdev_vport_ops); 224 } 225 226 void ovs_netdev_exit(void) 227 { 228 ovs_vport_ops_unregister(&ovs_netdev_vport_ops); 229 } 230