Address Resolution Protocol - ARP
ARP is a communication protocol that discovers the link layer address, such as a MAC address, associated with a network layer address. The Cumulus Linux ARP implementation differs from standard Debian Linux ARP behavior because Cumulus Linux is an operating system for routers and switches, not servers.
For a definition of ARP, refer to RFC 826.
Standard Debian ARP Behavior and the Tunable ARP Parameters
Debian has these five tunable ARP parameters:
arp_accept
arp_announce
arp_filter
arp_ignore
arp_notify
For a full description of these parameters, refer to the Linux documentation.
The standard Debian installation sets these ARP parameters to 0, leaving the router as wide open and unrestricted as possible. The Linux IP addresses are a property of the device, not an individual interface. Therefore, you can send an ARP request or reply on one interface with an address that resides on a different interface. While this unrestricted behavior makes sense for a server, it is not the normal behavior of a router. Routers expect the MAC and IP address mappings that ARP provides to match the physical topology, so that the IP addresses match the interfaces on which they reside. With these tunable ARP parameters, Cumulus Linux is able to specify the behavior to match the expectations of a router.
ARP Tunable Parameter Settings in Cumulus Linux
Parameter | Default Setting | Type | Description |
---|---|---|---|
arp_accept | 0 | BOOL | Defines the behavior for gratuitous ARP frames when the IP address is not already in the ARP table:
You can set arp_accept on an individual interface which differs from the rest of the switch (see below). |
arp_announce | 2 | INT | Defines different restriction levels for announcing the local source IP address from IP packets in ARP requests that send on an interface:
arp_announce is 0) sends gratuitous ARPs or ARP requests using any local source IP address and does not limit the IP source of the ARP packet to an address residing on the interface that sends the packet.Routers expect a different relationship between the IP address and the physical network. Adjoining routers look for MAC and IP addresses to reach a next hop residing on a connecting interface for transiting traffic. By setting the arp_announce parameter to 2, Cumulus Linux uses the best local address for each ARP request, preferring the primary addresses on the interface that sends the ARP. |
arp_filter | 0 | BOOL |
arp_filter for the interface is on if at least one of conf/{all,interface}/arp_filter is TRUE, it is off otherwise.Cumulus Linux uses the default Debian Linux arp_filter setting of 0.The switch uses arp_filter when multiple interfaces reside in the same subnet and allows certain interfaces to respond to ARP requests. For OSPF with IP unnumbered interfaces, multiple interfaces appear in the same subnet and contain the same address. If you use multiple interfaces between a pair of routers and set arp_filter to 1, forwarding can fail.The arp_filter parameter allows a response on any interface in the subnet, where the arp_ignore setting (below) limits cross-interface ARP behavior. |
arp_ignore | 1 | INT | Defines different modes for sending replies in response to received ARP requests that resolve local target IP addresses:
conf/{all,interface}/arp_ignore when the {interface} receives the ARP request.The default arp_ignore setting of 1 allows the device to reply to an ARP request for any IP address on any interface. While this matches the expectation that an IP address belongs to the device, not an interface, it can cause some unexpected behavior on a router.For example, if arp_ignore is 0 and the switch receives an ARP request on one interface for the IP address residing on a different interface, the switch responds with an ARP reply even if the interface of the target address is down. This can cause traffic loss because the switch does not know if it can reach the next hops and results in troubleshooting challenges for failure conditions.If you set arp_ignore to 2, the switch only replies to ARP requests if the target IP address is a local address and both the sender and target IP addresses are part of the same subnet on the incoming interface. The router does not create stale neighbor entries when a peer device sends an ARP request from a source IP address that is not on the connected subnet. Eventually, the switch sends ARP requests to the host to try to keep the entry fresh. If the host responds, the switch now has reachable neighbor entries for hosts that are not on the connected subnet. |
arp_notify | 1 | BOOL | Defines the mode to notify address and device changes.
arp_notify setting is to remain silent when an interface comes up or the hardware address changes. Because Cumulus Linux often acts as a next hop for several end hosts, it notifies attached devices when an interface comes up or the address changes, which speeds up new information convergence and provides the most rapid support for changes. |
Change Tunable ARP Parameters
You can change the ARP parameter settings in several places, including:
/proc/sys/net/ipv4/conf/all/arp*
(all interfaces)/proc/sys/net/ipv4/conf/default/arp*
(default for future interfaces)/proc/sys/net/ipv4/conf/swp*/arp*
(individual interfaces)
The ARP parameter changes in Cumulus Linux use the default file locations.
The all and default locations sound similar but they operate in different ways. The all location can potentially change the value for all interfaces running IP, both now and in the future. The all value applies to each parameter using either MAX or OR logic between the all and any port-specific settings, as the following table shows:
ARP Parameter | Condition |
---|---|
arp_accept | OR |
arp_announce | MAX |
arp_filter | OR |
arp_ignore | MAX |
arp_notify | MAX |
For example, if you set the /proc/sys/net/conf/all/arp_ignore
value to 1 and the /proc/sys/net/conf/swp1/arp_ignore
value to 0 to try to disable it on a per-port basis, interface swp1 still uses the value of 1; the port-specific setting does not override the global all setting. Instead, the MAX value between the all value and port-specific value defines the actual behavior.
The default location /proc/sys/net/ipv4/conf/default/arp*
defines the values for all future IP interfaces. Changing the default setting of an ARP parameter does not impact interfaces that already have an IP address. If you make changes to a running system that already has assigned IP addresses, use port-specific settings instead.
Cumulus Linux copies the value of the default parameter to every port-specific location, excluding those that already have an IP address. There is no complicated logic between the default setting and the port-specific setting (unlike the all location).
To determine the current ARP parameter settings for each of the locations, run the following commands:
cumulus@switch:~$ sudo grep . /proc/sys/net/ipv4/conf/all/arp*
/proc/sys/net/ipv4/conf/all/arp_accept:0
/proc/sys/net/ipv4/conf/all/arp_announce:0
/proc/sys/net/ipv4/conf/all/arp_filter:0
/proc/sys/net/ipv4/conf/all/arp_ignore:0
/proc/sys/net/ipv4/conf/all/arp_notify:1
cumulus@switch:~$ sudo grep . /proc/sys/net/ipv4/conf/default/arp*
/proc/sys/net/ipv4/conf/default/arp_accept:0
/proc/sys/net/ipv4/conf/default/arp_announce:2
/proc/sys/net/ipv4/conf/default/arp_filter:0
/proc/sys/net/ipv4/conf/default/arp_ignore:1
/proc/sys/net/ipv4/conf/default/arp_notify:1
cumulus@switch:~$ sudo grep . /proc/sys/net/ipv4/conf/swp1/arp*
/proc/sys/net/ipv4/conf/swp1/arp_accept:0
/proc/sys/net/ipv4/conf/swp1/arp_announce:2
/proc/sys/net/ipv4/conf/swp1/arp_filter:0
/proc/sys/net/ipv4/conf/swp1/arp_ignore:1
/proc/sys/net/ipv4/conf/swp1/arp_notify:1
Cumulus Linux implements this change at boot time using the arp.conf
file in the following location:
cumulus@switch:~$ cat /etc/sysctl.d/arp.conf
net.ipv4.conf.default.arp_announce = 2
net.ipv4.conf.all.arp_notify = 1
net.ipv4.conf.default.arp_notify = 1
net.ipv4.conf.default.arp_ignore=1
Change Port-specific ARP Parameters
To configure port-specific ARP parameters in a running device, run the following command:
cumulus@switch:~$ sudo sh -c "echo 0 > /proc/sys/net/ipv4/conf/swp1/arp_ignore"
cumulus@switch:~$ sudo grep . /proc/sys/net/ipv4/conf/swp1/arp*
/proc/sys/net/ipv4/conf/swp1/arp_accept:0
/proc/sys/net/ipv4/conf/swp1/arp_announce:2
/proc/sys/net/ipv4/conf/swp1/arp_filter:0
/proc/sys/net/ipv4/conf/swp1/arp_ignore:0
/proc/sys/net/ipv4/conf/swp1/arp_notify:1
To make the change persist through reboots, edit the /etc/sysctl.d/arp.conf
file and add your port-specific ARP setting.
Configure Proxy ARP
When you enable proxy ARP, if the switch receives an ARP request for which it has a route to the destination IP address, the switch sends a proxy ARP reply that contains its own MAC address. The host that sent the ARP request then sends its packets to the switch and the switch forwards the packets to the intended host.
Proxy ARP works with IPv4 only; ARP is an IPv4-only protocol.
The following example commands enable proxy ARP on swp1.
Edit the /etc/network/interfaces
file to set /proc/sys/net/ipv4/conf/<interface>/proxy_arp
to 1
in the interface stanza, then run the ifreload -a
command.
cumulus@switch:~$ sudo nano /etc/network/interfaces
...
auto swp1
iface swp1
post-up echo 1 > /proc/sys/net/ipv4/conf/swp1/proxy_arp
...
cumulus@switch:~$ sudo ifreload -a
If you are running two interfaces in the same broadcast domain (typically seen when using VRR, which creates a -v0
interface in the same broadcast domain), set /proc/sys/net/ipv4/conf/<INTERFACE>/medium_id
to 2 on both the base SVI interface and the -v0 interface so that only one of the two interfaces replies when getting an ARP request. This prevents the v0 interface from proxy replying on behalf of the SVI (and the SVI from proxy replying on behalf of the v0 interface). You can only prevent duplicate replies when the ARP request is for the SVI or the v0 interface directly.
Edit the /etc/network/interfaces
file, then run the ifreload -a
command. For example:
cumulus@switch:~$ sudo nano /etc/network/interfaces
...
auto swp1
iface swp1
post-up echo 1 > /proc/sys/net/ipv4/conf/swp1/proxy_arp
post-up echo 2 > /proc/sys/net/ipv4/conf/swp1/medium_id
auto swp1-v0
iface swp1-v0
post-up echo 1 > /proc/sys/net/ipv4/conf/swp1-v0/proxy_arp
post-up echo 2 > /proc/sys/net/ipv4/conf/swp1-v0/medium_id
...
cumulus@switch:~$ sudo ifreload -a
If you are running proxy ARP on a VRR interface, add a post-up line to the VRR interface stanza similar to the following. For example, if vlan100 is the VRR interface for the configuration above:
Edit the /etc/network/interfaces
file, then run the ifreload -a
command. For example:
cumulus@switch:~$ sudo nano /etc/networks/interfaces
...
auto vlan100
iface vlan100
post-up echo 1 > /proc/sys/net/ipv4/conf/swp1-v0/proxy_arp
post-up echo 1 > /proc/sys/net/ipv4/conf/swp1/proxy_arp
post-up echo 2 > /proc/sys/net/ipv4/conf/swp1-v0/medium_id
post-up echo 2 > /proc/sys/net/ipv4/conf/swp1/medium_id
vlan-id 100
...
cumulus@switch:~$ sudo ifreload -a
Duplicate Address Detection (Windows Hosts)
In centralized VXLAN environments with ARP and ND suppression, if the SVIs on the leafs but do not have an IP address within the subnet, problems with the Duplicate Address Detection process on Microsoft Windows hosts occur. For example, in a pure layer 2 scenario or with SVIs that have the ip-forward
option off, the SVI does not have an IP address. The neighmgrd
service selects a source IP address for an ARP probe based on the subnet match on the neighbor IP address. Because the SVI that learns this neighbor does not have an IP address, the subnet match fails and neighmgrd
uses UNSPEC (0.0.0.0 for IPv4) as the source IP address in the ARP probe.
To work around this issue, run the neighmgrctl setsrcipv4 <ipaddress>
command to specify a non-0.0.0.0 address for the source; for example:
cumulus@switch:~$ neighmgrctl setsrcipv4 10.1.0.2
The configuration above does not persist if you reboot the switch. To make the changes apply persistently:
Create a new file called
/etc/cumulus/neighmgr.conf
and add thesetsrcipv4 <ipaddress>
option; for example:cumulus@switch:~$ sudo nano /etc/cumulus/neighmgr.conf [main] setsrcipv4: 10.1.0.2
Restart the
neighmgrd
service:cumulus@switch:~$ sudo systemctl restart neighmgrd
Neighbor Base Reachable Timer
You can set how long a neighbor cache entry is valid with the NVUE nv set system global arp base-reachable-time
command. The entry is valid for at least the value between the base reachable time divided by two and three times the base reachable time divided by two. You can specify a value between 30 and 2147483 seconds. The default value is auto
; NVUE derives the value for auto
from the /etc/sysctl.d/neigh.conf
file.
The following example configures the neighbor base reachable timer to 50 seconds.
cumulus@leaf01:~$ nv set system global arp base-reachable-time 50
cumulus@leaf01:~$ nv config apply
To reset the neighbor base reachable timer to the default setting, run the nv unset system global arp base-reachable-time
command.
NVIDIA recommends that you run the NVUE command to change the neighbor base reachable timer instead of modifying the /etc/sysctl.d/neigh.conf
file manually.
To show the neighbor base reachable timer setting, run the nv show system global arp
command:
cumulus@leaf02:mgmt:~$ nv show system global arp
operational applied
---------------------------- ----------- -------
base-reachable-time 50 50
garbage-collection-threshold
effective 35840
maximum 40960
minimum 128
ARP Refresh
Cumulus Linux does not interact directly with end systems as much as end systems interact with each another. Therefore, after ARP places a neighbor into a reachable state, if Cumulus Linux does not interact with the client again for a long enough period of time, the neighbor can move into a stale state. To keep neighbors in the reachable state, Cumulus Linux includes a background process (/usr/bin/neighmgrd
). The background process can track neighbors that move into a stale, delay, or probe state, and attempt to refresh their state before removing them from the Linux kernel and from hardware forwarding. If you want the neighmgrd
process to add a neighbor if the sender IP address in the ARP packet is in one of the SVI’s subnets, create the /etc/cumulus/neighmgr.conf
file and add the subnet_checks=1
parameter under the [snooper]
header. By default, the subnet_checks
option is set to 0 (disabled) so that neighmgrd
allows out-of-network neighbors to be processed from SVIs.
The ARP refresh timer defaults to 1080 seconds (18 minutes).
cumulus@leaf02:mgmt:~$ sudo nano /etc/cumulus/neighmgr.conf
[snooper]
subnet_checks=1