1.. SPDX-License-Identifier: GPL-2.0 2 3==================================== 4Virtual Routing and Forwarding (VRF) 5==================================== 6 7The VRF Device 8============== 9 10The VRF device combined with ip rules provides the ability to create virtual 11routing and forwarding domains (aka VRFs, VRF-lite to be specific) in the 12Linux network stack. One use case is the multi-tenancy problem where each 13tenant has their own unique routing tables and in the very least need 14different default gateways. 15 16Processes can be "VRF aware" by binding a socket to the VRF device. Packets 17through the socket then use the routing table associated with the VRF 18device. An important feature of the VRF device implementation is that it 19impacts only Layer 3 and above so L2 tools (e.g., LLDP) are not affected 20(ie., they do not need to be run in each VRF). The design also allows 21the use of higher priority ip rules (Policy Based Routing, PBR) to take 22precedence over the VRF device rules directing specific traffic as desired. 23 24In addition, VRF devices allow VRFs to be nested within namespaces. For 25example network namespaces provide separation of network interfaces at the 26device layer, VLANs on the interfaces within a namespace provide L2 separation 27and then VRF devices provide L3 separation. 28 29Design 30------ 31A VRF device is created with an associated route table. Network interfaces 32are then enslaved to a VRF device:: 33 34 +-----------------------------+ 35 | vrf-blue | ===> route table 10 36 +-----------------------------+ 37 | | | 38 +------+ +------+ +-------------+ 39 | eth1 | | eth2 | ... | bond1 | 40 +------+ +------+ +-------------+ 41 | | 42 +------+ +------+ 43 | eth8 | | eth9 | 44 +------+ +------+ 45 46Packets received on an enslaved device and are switched to the VRF device 47in the IPv4 and IPv6 processing stacks giving the impression that packets 48flow through the VRF device. Similarly on egress routing rules are used to 49send packets to the VRF device driver before getting sent out the actual 50interface. This allows tcpdump on a VRF device to capture all packets into 51and out of the VRF as a whole\ [1]_. Similarly, netfilter\ [2]_ and tc rules 52can be applied using the VRF device to specify rules that apply to the VRF 53domain as a whole. 54 55.. [1] Packets in the forwarded state do not flow through the device, so those 56 packets are not seen by tcpdump. Will revisit this limitation in a 57 future release. 58 59.. [2] Iptables on ingress supports PREROUTING with skb->dev set to the real 60 ingress device and both INPUT and PREROUTING rules with skb->dev set to 61 the VRF device. For egress POSTROUTING and OUTPUT rules can be written 62 using either the VRF device or real egress device. 63 64Setup 65----- 661. VRF device is created with an association to a FIB table. 67 e.g,:: 68 69 ip link add vrf-blue type vrf table 10 70 ip link set dev vrf-blue up 71 722. An l3mdev FIB rule directs lookups to the table associated with the device. 73 A single l3mdev rule is sufficient for all VRFs. The VRF device adds the 74 l3mdev rule for IPv4 and IPv6 when the first device is created with a 75 default preference of 1000. Users may delete the rule if desired and add 76 with a different priority or install per-VRF rules. 77 78 Prior to the v4.8 kernel iif and oif rules are needed for each VRF device:: 79 80 ip ru add oif vrf-blue table 10 81 ip ru add iif vrf-blue table 10 82 833. Set the default route for the table (and hence default route for the VRF):: 84 85 ip route add table 10 unreachable default metric 4278198272 86 87 This high metric value ensures that the default unreachable route can 88 be overridden by a routing protocol suite. FRRouting interprets 89 kernel metrics as a combined admin distance (upper byte) and priority 90 (lower 3 bytes). Thus the above metric translates to [255/8192]. 91 924. Enslave L3 interfaces to a VRF device:: 93 94 ip link set dev eth1 master vrf-blue 95 96 Local and connected routes for enslaved devices are automatically moved to 97 the table associated with VRF device. Any additional routes depending on 98 the enslaved device are dropped and will need to be reinserted to the VRF 99 FIB table following the enslavement. 100 101 The IPv6 sysctl option keep_addr_on_down can be enabled to keep IPv6 global 102 addresses as VRF enslavement changes:: 103 104 sysctl -w net.ipv6.conf.all.keep_addr_on_down=1 105 1065. Additional VRF routes are added to associated table:: 107 108 ip route add table 10 ... 109 110 111Applications 112------------ 113Applications that are to work within a VRF need to bind their socket to the 114VRF device:: 115 116 setsockopt(sd, SOL_SOCKET, SO_BINDTODEVICE, dev, strlen(dev)+1); 117 118or to specify the output device using cmsg and IP_PKTINFO. 119 120By default the scope of the port bindings for unbound sockets is 121limited to the default VRF. That is, it will not be matched by packets 122arriving on interfaces enslaved to an l3mdev and processes may bind to 123the same port if they bind to an l3mdev. 124 125TCP & UDP services running in the default VRF context (ie., not bound 126to any VRF device) can work across all VRF domains by enabling the 127tcp_l3mdev_accept and udp_l3mdev_accept sysctl options:: 128 129 sysctl -w net.ipv4.tcp_l3mdev_accept=1 130 sysctl -w net.ipv4.udp_l3mdev_accept=1 131 132These options are disabled by default so that a socket in a VRF is only 133selected for packets in that VRF. There is a similar option for RAW 134sockets, which is enabled by default for reasons of backwards compatibility. 135This is so as to specify the output device with cmsg and IP_PKTINFO, but 136using a socket not bound to the corresponding VRF. This allows e.g. older ping 137implementations to be run with specifying the device but without executing it 138in the VRF. This option can be disabled so that packets received in a VRF 139context are only handled by a raw socket bound to the VRF, and packets in the 140default VRF are only handled by a socket not bound to any VRF:: 141 142 sysctl -w net.ipv4.raw_l3mdev_accept=0 143 144netfilter rules on the VRF device can be used to limit access to services 145running in the default VRF context as well. 146 147Using VRF-aware applications (applications which simultaneously create sockets 148outside and inside VRFs) in conjunction with ``net.ipv4.tcp_l3mdev_accept=1`` 149is possible but may lead to problems in some situations. With that sysctl 150value, it is unspecified which listening socket will be selected to handle 151connections for VRF traffic; ie. either a socket bound to the VRF or an unbound 152socket may be used to accept new connections from a VRF. This somewhat 153unexpected behavior can lead to problems if sockets are configured with extra 154options (ex. TCP MD5 keys) with the expectation that VRF traffic will 155exclusively be handled by sockets bound to VRFs, as would be the case with 156``net.ipv4.tcp_l3mdev_accept=0``. Finally and as a reminder, regardless of 157which listening socket is selected, established sockets will be created in the 158VRF based on the ingress interface, as documented earlier. 159 160-------------------------------------------------------------------------------- 161 162Using iproute2 for VRFs 163======================= 164iproute2 supports the vrf keyword as of v4.7. For backwards compatibility this 165section lists both commands where appropriate -- with the vrf keyword and the 166older form without it. 167 1681. Create a VRF 169 170 To instantiate a VRF device and associate it with a table:: 171 172 $ ip link add dev NAME type vrf table ID 173 174 As of v4.8 the kernel supports the l3mdev FIB rule where a single rule 175 covers all VRFs. The l3mdev rule is created for IPv4 and IPv6 on first 176 device create. 177 1782. List VRFs 179 180 To list VRFs that have been created:: 181 182 $ ip [-d] link show type vrf 183 NOTE: The -d option is needed to show the table id 184 185 For example:: 186 187 $ ip -d link show type vrf 188 11: mgmt: <NOARP,MASTER,UP,LOWER_UP> mtu 1500 qdisc pfifo_fast state UP mode DEFAULT group default qlen 1000 189 link/ether 72:b3:ba:91:e2:24 brd ff:ff:ff:ff:ff:ff promiscuity 0 190 vrf table 1 addrgenmode eui64 191 12: red: <NOARP,MASTER,UP,LOWER_UP> mtu 1500 qdisc pfifo_fast state UP mode DEFAULT group default qlen 1000 192 link/ether b6:6f:6e:f6:da:73 brd ff:ff:ff:ff:ff:ff promiscuity 0 193 vrf table 10 addrgenmode eui64 194 13: blue: <NOARP,MASTER,UP,LOWER_UP> mtu 1500 qdisc pfifo_fast state UP mode DEFAULT group default qlen 1000 195 link/ether 36:62:e8:7d:bb:8c brd ff:ff:ff:ff:ff:ff promiscuity 0 196 vrf table 66 addrgenmode eui64 197 14: green: <NOARP,MASTER,UP,LOWER_UP> mtu 1500 qdisc pfifo_fast state UP mode DEFAULT group default qlen 1000 198 link/ether e6:28:b8:63:70:bb brd ff:ff:ff:ff:ff:ff promiscuity 0 199 vrf table 81 addrgenmode eui64 200 201 202 Or in brief output:: 203 204 $ ip -br link show type vrf 205 mgmt UP 72:b3:ba:91:e2:24 <NOARP,MASTER,UP,LOWER_UP> 206 red UP b6:6f:6e:f6:da:73 <NOARP,MASTER,UP,LOWER_UP> 207 blue UP 36:62:e8:7d:bb:8c <NOARP,MASTER,UP,LOWER_UP> 208 green UP e6:28:b8:63:70:bb <NOARP,MASTER,UP,LOWER_UP> 209 210 2113. Assign a Network Interface to a VRF 212 213 Network interfaces are assigned to a VRF by enslaving the netdevice to a 214 VRF device:: 215 216 $ ip link set dev NAME master NAME 217 218 On enslavement connected and local routes are automatically moved to the 219 table associated with the VRF device. 220 221 For example:: 222 223 $ ip link set dev eth0 master mgmt 224 225 2264. Show Devices Assigned to a VRF 227 228 To show devices that have been assigned to a specific VRF add the master 229 option to the ip command:: 230 231 $ ip link show vrf NAME 232 $ ip link show master NAME 233 234 For example:: 235 236 $ ip link show vrf red 237 3: eth1: <BROADCAST,MULTICAST,UP,LOWER_UP> mtu 1500 qdisc pfifo_fast master red state UP mode DEFAULT group default qlen 1000 238 link/ether 02:00:00:00:02:02 brd ff:ff:ff:ff:ff:ff 239 4: eth2: <BROADCAST,MULTICAST,UP,LOWER_UP> mtu 1500 qdisc pfifo_fast master red state UP mode DEFAULT group default qlen 1000 240 link/ether 02:00:00:00:02:03 brd ff:ff:ff:ff:ff:ff 241 7: eth5: <BROADCAST,MULTICAST> mtu 1500 qdisc noop master red state DOWN mode DEFAULT group default qlen 1000 242 link/ether 02:00:00:00:02:06 brd ff:ff:ff:ff:ff:ff 243 244 245 Or using the brief output:: 246 247 $ ip -br link show vrf red 248 eth1 UP 02:00:00:00:02:02 <BROADCAST,MULTICAST,UP,LOWER_UP> 249 eth2 UP 02:00:00:00:02:03 <BROADCAST,MULTICAST,UP,LOWER_UP> 250 eth5 DOWN 02:00:00:00:02:06 <BROADCAST,MULTICAST> 251 252 2535. Show Neighbor Entries for a VRF 254 255 To list neighbor entries associated with devices enslaved to a VRF device 256 add the master option to the ip command:: 257 258 $ ip [-6] neigh show vrf NAME 259 $ ip [-6] neigh show master NAME 260 261 For example:: 262 263 $ ip neigh show vrf red 264 10.2.1.254 dev eth1 lladdr a6:d9:c7:4f:06:23 REACHABLE 265 10.2.2.254 dev eth2 lladdr 5e:54:01:6a:ee:80 REACHABLE 266 267 $ ip -6 neigh show vrf red 268 2002:1::64 dev eth1 lladdr a6:d9:c7:4f:06:23 REACHABLE 269 270 2716. Show Addresses for a VRF 272 273 To show addresses for interfaces associated with a VRF add the master 274 option to the ip command:: 275 276 $ ip addr show vrf NAME 277 $ ip addr show master NAME 278 279 For example:: 280 281 $ ip addr show vrf red 282 3: eth1: <BROADCAST,MULTICAST,UP,LOWER_UP> mtu 1500 qdisc pfifo_fast master red state UP group default qlen 1000 283 link/ether 02:00:00:00:02:02 brd ff:ff:ff:ff:ff:ff 284 inet 10.2.1.2/24 brd 10.2.1.255 scope global eth1 285 valid_lft forever preferred_lft forever 286 inet6 2002:1::2/120 scope global 287 valid_lft forever preferred_lft forever 288 inet6 fe80::ff:fe00:202/64 scope link 289 valid_lft forever preferred_lft forever 290 4: eth2: <BROADCAST,MULTICAST,UP,LOWER_UP> mtu 1500 qdisc pfifo_fast master red state UP group default qlen 1000 291 link/ether 02:00:00:00:02:03 brd ff:ff:ff:ff:ff:ff 292 inet 10.2.2.2/24 brd 10.2.2.255 scope global eth2 293 valid_lft forever preferred_lft forever 294 inet6 2002:2::2/120 scope global 295 valid_lft forever preferred_lft forever 296 inet6 fe80::ff:fe00:203/64 scope link 297 valid_lft forever preferred_lft forever 298 7: eth5: <BROADCAST,MULTICAST> mtu 1500 qdisc noop master red state DOWN group default qlen 1000 299 link/ether 02:00:00:00:02:06 brd ff:ff:ff:ff:ff:ff 300 301 Or in brief format:: 302 303 $ ip -br addr show vrf red 304 eth1 UP 10.2.1.2/24 2002:1::2/120 fe80::ff:fe00:202/64 305 eth2 UP 10.2.2.2/24 2002:2::2/120 fe80::ff:fe00:203/64 306 eth5 DOWN 307 308 3097. Show Routes for a VRF 310 311 To show routes for a VRF use the ip command to display the table associated 312 with the VRF device:: 313 314 $ ip [-6] route show vrf NAME 315 $ ip [-6] route show table ID 316 317 For example:: 318 319 $ ip route show vrf red 320 unreachable default metric 4278198272 321 broadcast 10.2.1.0 dev eth1 proto kernel scope link src 10.2.1.2 322 10.2.1.0/24 dev eth1 proto kernel scope link src 10.2.1.2 323 local 10.2.1.2 dev eth1 proto kernel scope host src 10.2.1.2 324 broadcast 10.2.1.255 dev eth1 proto kernel scope link src 10.2.1.2 325 broadcast 10.2.2.0 dev eth2 proto kernel scope link src 10.2.2.2 326 10.2.2.0/24 dev eth2 proto kernel scope link src 10.2.2.2 327 local 10.2.2.2 dev eth2 proto kernel scope host src 10.2.2.2 328 broadcast 10.2.2.255 dev eth2 proto kernel scope link src 10.2.2.2 329 330 $ ip -6 route show vrf red 331 local 2002:1:: dev lo proto none metric 0 pref medium 332 local 2002:1::2 dev lo proto none metric 0 pref medium 333 2002:1::/120 dev eth1 proto kernel metric 256 pref medium 334 local 2002:2:: dev lo proto none metric 0 pref medium 335 local 2002:2::2 dev lo proto none metric 0 pref medium 336 2002:2::/120 dev eth2 proto kernel metric 256 pref medium 337 local fe80:: dev lo proto none metric 0 pref medium 338 local fe80:: dev lo proto none metric 0 pref medium 339 local fe80::ff:fe00:202 dev lo proto none metric 0 pref medium 340 local fe80::ff:fe00:203 dev lo proto none metric 0 pref medium 341 fe80::/64 dev eth1 proto kernel metric 256 pref medium 342 fe80::/64 dev eth2 proto kernel metric 256 pref medium 343 ff00::/8 dev red metric 256 pref medium 344 ff00::/8 dev eth1 metric 256 pref medium 345 ff00::/8 dev eth2 metric 256 pref medium 346 unreachable default dev lo metric 4278198272 error -101 pref medium 347 3488. Route Lookup for a VRF 349 350 A test route lookup can be done for a VRF:: 351 352 $ ip [-6] route get vrf NAME ADDRESS 353 $ ip [-6] route get oif NAME ADDRESS 354 355 For example:: 356 357 $ ip route get 10.2.1.40 vrf red 358 10.2.1.40 dev eth1 table red src 10.2.1.2 359 cache 360 361 $ ip -6 route get 2002:1::32 vrf red 362 2002:1::32 from :: dev eth1 table red proto kernel src 2002:1::2 metric 256 pref medium 363 364 3659. Removing Network Interface from a VRF 366 367 Network interfaces are removed from a VRF by breaking the enslavement to 368 the VRF device:: 369 370 $ ip link set dev NAME nomaster 371 372 Connected routes are moved back to the default table and local entries are 373 moved to the local table. 374 375 For example:: 376 377 $ ip link set dev eth0 nomaster 378 379-------------------------------------------------------------------------------- 380 381Commands used in this example:: 382 383 cat >> /etc/iproute2/rt_tables.d/vrf.conf <<EOF 384 1 mgmt 385 10 red 386 66 blue 387 81 green 388 EOF 389 390 function vrf_create 391 { 392 VRF=$1 393 TBID=$2 394 395 # create VRF device 396 ip link add ${VRF} type vrf table ${TBID} 397 398 if [ "${VRF}" != "mgmt" ]; then 399 ip route add table ${TBID} unreachable default metric 4278198272 400 fi 401 ip link set dev ${VRF} up 402 } 403 404 vrf_create mgmt 1 405 ip link set dev eth0 master mgmt 406 407 vrf_create red 10 408 ip link set dev eth1 master red 409 ip link set dev eth2 master red 410 ip link set dev eth5 master red 411 412 vrf_create blue 66 413 ip link set dev eth3 master blue 414 415 vrf_create green 81 416 ip link set dev eth4 master green 417 418 419 Interface addresses from /etc/network/interfaces: 420 auto eth0 421 iface eth0 inet static 422 address 10.0.0.2 423 netmask 255.255.255.0 424 gateway 10.0.0.254 425 426 iface eth0 inet6 static 427 address 2000:1::2 428 netmask 120 429 430 auto eth1 431 iface eth1 inet static 432 address 10.2.1.2 433 netmask 255.255.255.0 434 435 iface eth1 inet6 static 436 address 2002:1::2 437 netmask 120 438 439 auto eth2 440 iface eth2 inet static 441 address 10.2.2.2 442 netmask 255.255.255.0 443 444 iface eth2 inet6 static 445 address 2002:2::2 446 netmask 120 447 448 auto eth3 449 iface eth3 inet static 450 address 10.2.3.2 451 netmask 255.255.255.0 452 453 iface eth3 inet6 static 454 address 2002:3::2 455 netmask 120 456 457 auto eth4 458 iface eth4 inet static 459 address 10.2.4.2 460 netmask 255.255.255.0 461 462 iface eth4 inet6 static 463 address 2002:4::2 464 netmask 120 465