Cumulus Linux 3.2.0 Release Notes

Follow

Overview

These release notes support Cumulus Linux 3.2.0 and describe currently available features and known issues. 

Stay up to Date 

  • Please sign in and click Follow above so you can receive a notification when we update these release notes.
  • Subscribe to our product bulletin mailing list to receive important announcements and updates about issues that arise in our products.
  • Subscribe to our security announcement mailing list to receive alerts whenever we update our software for security issues.

{{table_of_contents}}

What's New in Cumulus Linux 3.2.0

Cumulus Linux 3.2.0 includes the following features and a number of improvements:

Early Access Features

The following early access features are included in Cumulus Linux 3.2.0:

Licensing

Cumulus Linux is licensed on a per-instance basis. Each network system is fully operational, enabling any capability to be utilized on the switch with the exception of forwarding on switch panel ports. Only eth0 and console ports are activated on an un-licensed instance of Cumulus Linux. Enabling front panel ports requires a license.

You should have received a license key from Cumulus Networks or an authorized reseller. To install the license, read the Cumulus Linux Quick Start Guide.

Installing Version 3.2.0

If you are upgrading from version 3.0.0 or later, use apt-get to update the software.

  1. Run apt-get update.
  2. Run apt-get upgrade.
  3. Reboot the switch.

New Install or Upgrading from Versions Older than 3.0.0

If you are upgrading from a version older than 3.0.0, or installing Cumulus Linux for the first time, download the Cumulus Linux 3.2.0 installer for Broadcom or Mellanox switches from the Cumulus Networks website, then use ONIE to perform a complete install, following the instructions in the quick start guide.

Note: This method is destructive; any configuration files on the switch will not be saved, so please copy them to a different server before upgrading via ONIE.

Important! After you install, run apt-get update, then apt-get upgrade on your switch to make sure you update Cumulus Linux to include any important or other package updates.

Updating a Deployment that Has MLAG Configured

If you are using MLAG to dual connect two switches in your environment, and those switches are still running Cumulus Linux 2.5 ESR or any other release earlier than 3.0.0, the switches will not be dual-connected after you upgrade the first switch. To ensure a smooth upgrade, follow these steps:

  1. Run cl-img-select -fr to boot the switch in the secondary role into ONIE, then reboot the switch.
  2. Install Cumulus Linux 3.2.0 onto the secondary switch using ONIE. At this time, all traffic is going to the switch in the primary role.
  3. After the install, copy the license file and all the configuration files you backed up, then restart the switchd and networking services.
    cumulus@switch:~$ sudo systemctl restart switchd.service
    cumulus@switch:~$ sudo systemctl restart networking.service
    All traffic is still going to the primary switch.
  4. Run cl-img-select -fr to boot the switch in the primary role into ONIE, then reboot the switch. Now, all traffic is going to the switch in the secondary role that you just upgraded to version 3.2.0.
  5. Install Cumulus Linux 3.2.0 onto the primary switch using ONIE. 
  6. After the install, copy the license file and all the configuration files you backed up, then restart the switchd and networking services.
    cumulus@switch:~$ sudo systemctl restart switchd.service
    cumulus@switch:~$ sudo systemctl restart networking.service
  7. Now the two switches are dual-connected again and traffic flows to both switches.

 SNMP Not Supported in Quagga

There is no SNMP support for Quagga in Cumulus Linux. However, it's possible to get it via SNMP by:

  • Using Nagios
  • Writing a pass persist script in Perl or Python by filling in the OSPF or BGP (rfc) MIBs manually.
  • Creating your own private MIB for the information you need.

Due to this circumstance, you must remove all references to smux in each of the following configuration files. You must also remove these references before upgrading Cumulus Linux using apt-get. If the smux entries are present in the configuration files, the daemons in the 2.5 packaged version of Quagga will not start.

  1. cd /etc/quagga
  2. grep smux *
  3. Delete all lines in the config files containing the smux keyword.

The references to smux that must be removed are:

  • In bgpd.conf, remove this line:
    smux peer 1.3.6.1.4.1.3317.1.2.2 quagga_bgpd
  • In ospf6d.conf, remove this line:
    smux peer 1.3.6.1.4.1.3317.1.2.6 quagga_ospf6d
  • In ospfd.conf, remove this line:
    smux peer 1.3.6.1.4.1.3317.1.2.5 quagga_ospfd
  • In zebra.conf, remove this line:
    smux peer 1.3.6.1.4.1.3317.1.2.1 quagga_zebra

 Perl, Python and BDB Modules

Any Perl scripts that use the DB_File module or Python scripts that use the bsddb module won't run under Cumulus Linux 3.2.0.

Documentation

You can read the technical documentation here.

Issues Fixed in Cumulus Linux 3.2.0

The following is a list of issues fixed in Cumulus Linux 3.2.0 from earlier versions of Cumulus Linux.

Release Note ID Summary Description

RN-380 (CM-6110)
ifupdown2: adjust VLAN subinterface MTU based on MTU settings specified under lowerdev by the user

The following kernel error occurs when the MTU is specified under a subinterface rather than under the VLAN interface:

root@dell-s3000-04:~# ifreload -a -X eth0
warning: failed to execute cmd 'ip -force -batch - [addr add 1.1.4.1/24 dev swp52.100
link set dev swp52.100 mtu 9000 
]'(RTNETLINK answers: Numerical result out of range
Command failed -:2)

RN-445 (CM-12370)
An interface cannot have both inet and inet6 DHCP configurations

If you configure an interface so it can to obtain both IPv4 and IPv6 IP addresses via DHCP, ifupdown2 will honor only the first configuration and ignore the second.

In the following example configuration, ifupdown2 will only issue an IPv4 DHCP address for swp1, but not the IPv6 address.

auto swp1
iface swp1 inet dhcp
    link-speed 10000
    link-duplex full
    link-autoneg off

auto swp1
iface swp1 inet6 dhcp

RN-447 (CM-11280)
"portwd: invalid SFF identifier: 0x0c" messages appear continuously in syslog

The following SFF message appears every 5 seconds in syslog:

cumulus@switch:~$ tail -f /var/log/syslog 
2016-08-06T12:18:56.095606-04:00 cumulus portwd: invalid SFF identifier: 0x0c
2016-08-06T12:19:01.113397-04:00 cumulus portwd: invalid SFF identifier: 0x0c
2016-08-06T12:19:01.121068-04:00 cumulus portwd: invalid SFF identifier: 0x0c
2016-08-06T12:19:01.121698-04:00 cumulus portwd: invalid SFF identifier: 0x0c
2016-08-06T12:19:06.139373-04:00 cumulus portwd: invalid SFF identifier: 0x0c
2016-08-06T12:19:06.147045-04:00 cumulus portwd: invalid SFF identifier: 0x0c
2016-08-06T12:19:06.147677-04:00 cumulus portwd: invalid SFF identifier: 0x0c
2016-08-06T12:19:11.165355-04:00 cumulus portwd: invalid SFF identifier: 0x0c
2016-08-06T12:19:11.173134-04:00 cumulus portwd: invalid SFF identifier: 0x0c
2016-08-06T12:19:11.173747-04:00 cumulus portwd: invalid SFF identifier: 0x0c
2016-08-06T12:19:16.191418-04:00 cumulus portwd: invalid SFF identifier: 0x0c
2016-08-06T12:19:16.199154-04:00 cumulus portwd: invalid SFF identifier: 0x0c
2016-08-06T12:19:16.199805-04:00 cumulus portwd: invalid SFF identifier: 0x0c

RN-449 (CM-11584)
In traditional bridge mode, clagd syncs MAC addresses in the wrong VLAN when the peerlink is tagged and the bond is native

When a traditional mode bridge is configured and the peerlink is tagged but the clagd bonds are native VLANs, clagd appears to try and sync the MAC addresses learned using the VLAN tag from the peerlink.

This causes the MAC address not to be synced correctly on the peer.


RN-452 (CM-12363)
Mellanox switch rejects ACL rule in ebtables with both -d and -o options

On a Mellanox switch, cl-acltool rejects an ebtables rule that contains both the destination MAC address and the output interface.


RN-453 (CM-12564)
Default routes learned via DHCP are moved to the management VRF even if they are not in the management VRF

The mgmt-vrf package has a dhclient exit hook that incorrectly assumes that DHCP is used only with the management interface. If management VRF is enabled, it inserts default routes from the DHCP server into the management table.

Until this issue is resolved, do not use DHCP with the front panel (switch) ports. If you need DHCP for a switch port, disable Management VRF and assign the port to a VRF.


RN-454 (CM-12565)
When VRF is enabled, the ICMP "need to fragment" is not respected

Cumulus Linux defaults to using PMTU to discover if a path has a lower MTU than what is set on its local interfaces. If an ICMP - Fragmentation Needed packet is received on an interface associated with a VRF, the MTU change for the remote address is not properly handled. The end result is that packets exceeding the MTU going through that network segment are dropped.

To work around this issue until it is resolved, do one of the following:

  • Disable PMTU. This does not set the "don't fragment" bit in the IP header, allowing nodes to fragment the packet if needed.
    cumulus@switch:~$ sudo sysctl -w net.ipv4.ip_no_pmtu_disc=1
  • Manually reduce the MTU on the interface. Add mtu <value> to the stanza for that interface in /etc/network/interfaces.

RN-489 (CM-12643)
IGMP snooping doesn't prevent flooding of unregistered multicast data packets

When broadcast video is sent over multicast with a number of senders running, the flooding approaches caused major issues, as the maximum available bandwidth was saturated on an individual port, and started dropping.


RN-509 (CM-12654)
hsflowd - packet samples missing and default nflog group not set since 3.x

An issue occurrs where only counter samples are generated when hsflowd is configured correctly, but no sFlow sampled packets left the switch.

Manually calling portsamp and specifying an NFLOG group causes the ingress samples to leave the switch. You do this in /proc/bcm/knet/sample_groups:

cumulus@switch:~$ cat /proc/bcm/knet/sample_groups 
Sample NFLOG group (unit 0): 
rx=1
tx=0

RN-524 (CM-13244)
CVE-2016-7042: /proc/keys stack corruption kernel security bug

A Linux kernel vulnerability was found that when the GNU Compiler Collection (gcc) stack protector is enabled, it uses an incorrect buffer size for certain timeout data, which allows local users to cause a denial of service (stack memory corruption and panic) by reading the /proc/keys file.

This issue is fixed in Cumulus Linux 3.2.0.


RN-527 (CM-13588)
Cannot set MTU to 9216 on a Mellanox switch

Setting the MTU to 9216 on a Mellanox switch returns the following error after running ifreload -a to reload the updated configuration:

error: swp1: cmd 'ip -force -batch - [link set dev swp1 mtu 9216
]' failed: returned 1 (RTNETLINK answers: Invalid argument
Command failed -:1
)

RN-531 (CM-13232)
switchd doesn't check for/print more useful info when misprogrammed Mellanox device eeprom is present switchd fails to start when a Mellanox device has a misprogrammed eeprom, as decode-syseeprom relied on output code that contained errors. The error check has been moved out of the decoder, allowing other code to check the validity of the MAC address, fixing the issue.

RN-532 (CM-13277)
dnsmasq port 53 was enabled, causing potential security vulnerability

dnsmasq port 53 has been disabled by default on all platforms except Cumulus RMP, as enabling it creates a security vulnerability.

If your system uses dnsmasq, run the following command to enable it:

cumulus@switch:~$ sudo systemctl enable dnsmasq
 

RN-533 (CM-13348)
TACACS+ client users can't login with eth0 in management VRF An issue occurred when process context was not causing the socket to the server to be bound to the management VRF. This resulted in the TACACS+ client user failing to login with eth0. The issue has now been resolved.

RN-534 (CM-13806)
Memory leak caused by duplicate MAC address entries A memory leak was seen when users manually configured a VLAN-aware bridge along with a controller/vtepd provisioned traditional bridge. This was caused by the same MAC address on different VLANs being seen as duplicates in the bridge MAC entries list. This memory leak has now been corrected.

Known Issues in Cumulus Linux 3.2.0

Issues are categorized for easy review. Some issues are fixed but will be available in a later release.

Release Note ID Summary Description

RN-52 (CM-997,
CM-1013)
Parameters like the router ID and DR priority cannot be changed while OSPFv2/v3 is running Router ID and DR priority can only be changed by shutting down OSPFv2/v3, changing the ID, and restarting the OSPF process.

A change to the DR priority may not properly be reflected in the LSAs that are still aging out.

RN-56 (CM-343)
IPv4/IPv6 forwarding disabled mode not recognized

If either of the following is configured:

net.ipv4.ip_forward == 0 

or:

net.ipv6.conf.all.forwarding == 0 

The hardware still forwards packets if there is a neighbor table entry pointing to the destination.


RN-77 (CM-265)
New routes/ECMPs can evict existing/installed Cumulus Linux syncs routes between the kernel and the switching silicon. If the required resource pools in hardware fill up, new kernel routes can cause existing routes to move from being fully allocated to being partially allocated.

In order to avoid this, routes in the hardware should be monitored and kept below the ASIC limits.

For example, on systems with Trident+ chips, the limits are as follows:
routes: 16384 <<<< if all routes are ipv4 
 long mask routes 256 <<<< i.e., routes with a mask longer 
       than the route mask limit 
 route mask limit 64
 host_routes: 8192 
 ecmp_nhs: 4044 
 ecmp_nhs_per_route: 52 
That translates to about 77 routes with ECMP NHs, if every route has the maximum ECMP NHs.

Monitoring this in Cumulus Linux is performed via the cl-resource-query command:
cumulus@switch:~$ sudo cl-resource-query
 hosts : 3 
 all routes : 29 
 IP4 routes : 17 
 IP6 routes : 12 
 nexthops : 3 
 ecmp_groups : 0
 ecmp_nexthops : 0
 mac entries : 0 / 131072 
 bpdu entries : 500 / 512
The resource to monitor is the ecmp_nexthops. If this count is close to 4044, new ECMPs may evict existing routes.

RN-120 (CM-477)
ethtool LED blinking does not work with switch ports Linux uses ethtool -p to identify the physical port backing an interface, or to identify the switch itself. Usually this identification is by blinking the port LED until ethtool -p is stopped.

This feature does not apply to switch ports (swpX) in Cumulus Linux.

RN-121 (CM-2123)
PTMD: When a physical interface is in a PTM FAIL state, its subinterface still exchanges information Issue:
When PTMD is incorrectly in a failure state and the Zebra interface is enabled, PIF BGP sessions are not establishing the route, but the subinterface on top of it does establish routes.

If the subinterface is configured on the physical interface and the physical interface is incorrectly marked as being in a PTM FAIL state, routes on the physical interface are not processed in Quagga, but the subinterface is working.

Steps to reproduce:
cumulus@switch:$ sudo vtysh -c 'show int swp8' 
Interface swp8 is up, line protocol is up 
PTM status: fail
index 10 metric 1 mtu 1500 
 flags: <UP,BROADCAST,RUNNING,MULTICAST>
 HWaddr: 44:38:39:00:03:88 
 inet 12.0.0.225/30 broadcast 12.0.0.227 
 inet6 2001:cafe:0:38::1/64 
 inet6 fe80::4638:39ff:fe00:388/64 
cumulus@switch:$ ip addr show | grep swp8 
 10: swp8: <BROADCAST,MULTICAST,UP,LOWER_UP> 
  mtu 1500 qdisc pfifo_fast state UP qlen 500 
  inet 12.0.0.225/30 brd 12.0.0.227 scope global swp8 
 104: swp8.2049@swp8: <BROADCAST,MULTICAST,UP,LOWER_UP> 
  mtu 1500 qdisc noqueue state UP 
  inet 12.0.0.229/30 brd 12.0.0.231 scope global swp8.2049 
 105: swp8.2050@swp8: <BROADCAST,MULTICAST,UP,LOWER_UP> 
  mtu 1500 qdisc noqueue state UP 
  inet 12.0.0.233/30 brd 12.0.0.235 scope global swp8.2050 
 106: swp8.2051@swp8: <BROADCAST,MULTICAST,UP,LOWER_UP> 
  mtu 1500 qdisc noqueue state UP 
  inet 12.0.0.237/30 brd 12.0.0.239 scope global swp8.2051 
 107: swp8.2052@swp8: <BROADCAST,MULTICAST,UP,LOWER_UP> 
  mtu 1500 qdisc noqueue state UP 
  inet 12.0.0.241/30 brd 12.0.0.243 scope global swp8.2052 
 108: swp8.2053@swp8: <BROADCAST,MULTICAST,UP,LOWER_UP>
  mtu 1500 qdisc noqueue state UP 
  inet 12.0.0.245/30 brd 12.0.0.247 scope global swp8.2053 
 109: swp8.2054@swp8: <BROADCAST,MULTICAST,UP,LOWER_UP> 
  mtu 1500 qdisc noqueue state UP 
  inet 12.0.0.249/30 brd 12.0.0.251 scope global swp8.2054
 110: swp8.2055@swp8: <BROADCAST,MULTICAST,UP,LOWER_UP>
  mtu 1500 qdisc noqueue state UP 
  inet 12.0.0.253/30 brd 12.0.0.255 scope global swp8.2055
cumulus@switch:$ bgp sessions: 
 12.0.0.226 ,4 ,64057 , 958 , 1036 , 0 , 0 , 0 ,15:55:42, 0, 10472 
 12.0.0.230 ,4 ,64058 , 958 , 1016 , 0 , 0 , 0 ,15:55:46, 187, 10285
 12.0.0.234 ,4 ,64059 , 958 , 1049 , 0 , 0 , 0 ,15:55:40, 187, 10285 
 12.0.0.238 ,4 ,64060 , 958 , 1039 , 0 , 0 , 0 ,15:55:45, 187, 10285 
 12.0.0.242 ,4 ,64061 , 958 , 1014 , 0 , 0 , 0 ,15:55:46, 187, 10285 
 12.0.0.246 ,4 ,64062 , 958 , 1016 , 0 , 0 , 0 ,15:55:46, 187, 10285 
 12.0.0.250 ,4 ,64063 , 958 , 1029 , 0 , 0 , 0 ,15:55:43, 187, 10285 
 12.0.0.254 ,4 ,64064 , 958 , 1036 , 0 , 0 , 0 ,15:55:44, 187, 10285 

RN-125 (CM-1576)
Network LSA with an old router ID isn't flushed out by the originator
When the router ID is changed, the router should remove the previous network LSA (link-state advertisement) that it generated based on the IP address on the interface in the Network LSA.

Cumulus Networks doesn't remove this LSA, so it will be naturally aged out.

RN-198 (CM-3290)
Port LEDs behave differently on different switch models

It's been observed that port LEDs behave differently depending upon the make and model of the switch. For example:

  • Agema AG-7448CU: the LED is off when the link is up. It blinks on briefly when there is traffic.
  • Edge-Core AS4600-54T: the LED is off when the link is up. It blinks on briefly when there is traffic.
  • QuantaMesh T3048-LY2R: the LED is on when the link is up. It blinks off briefly when there is traffic.

Cumulus Networks is currently working to fix this issue.


RN-199 (CM-2624)
When a Quagga route-map is modified, the switch could use the partial map before edits are completed

Cumulus Linux triggers a route-map update before the user finishes editing the route map, resulting in an incorrect route map being used. The route-map update trigger should only occur when user finishes editing the map.

Cumulus Networks is working to fix this issue.


RN-221 (CM-3926, CM-4501)
BGP graceful restart, including helper mode, not fully supported If you encounter issues with this, please submit a support request and include the output from cl-support with your ticket.

RN-227 (CM-3388)
BGP dynamic capability is not supported BGP peer sessions with dynamic capability are not supported under any version of Cumulus Linux at this time.

RN-322 (CM-7387)
Interfaces disabled using iproute2 become enabled after restarting Quagga By default, all interfaces have a "no shutdown" associated with them in Quagga. Thus, when you restart Quagga, it enables the interfaces. This is expected behavior in Quagga. There is no workaround at this time.

RN-327 (CM-4290)
Changing the route-map parameter of the redistribute command in OSPF and BGP doesn't affect the state of the resulting redistribution in those protocols

To work around this issue, remove any old redistribute command configurations before adding a new one with or without route-map as a parameter.

For example, if OSPF has a redistribute configuration such as redistribute bgp route-map redist-map-name, you would enable redistribution without a route-map by following these steps in OSPF configuration mode:

  1. no redistribute bgp
  2. redistribute bgp

You would perform a similar sequence of commands for redistribution changes in BGP as well.


RN-355 (CM-7994)
OSPFv2 Area ID being implicitly translated from Integer format to dotted decimal format

While OSPF area ID configuration in Quagga allows for the value to be specified in either dotted decimal format, or as an integer, values specified as an integer will be converted into dotted decimal format when displayed, causing potential confusion for the operator.

This issue does not impact OSPF functionality; only the display output. However, it is recommended that the OSPF area ID is specified in dotted decimal format for consistency.

 

RN-382 (CM-6692)
Quagga: Removing bridge via ifupdown2 does not remove it from Quagga Removing a bridge using ifupdown2 does not remove it from the Quagga configuration files. This issue is being investigated; however, restarting Quagga will successfully remove the bridge.

RN-383 (CM-7196)
admin down of link deletes IPv6 nexthop static route entry, but not for IPv4

When a link is admin down and carrier is on, the IPv4 nexthop entry is marked dead, but the IPv6 nexthop entry is deleted, and will not be restored when the link is admin up. However, if carrier is off, the IPv6 nexthop is marked dead and not deleted. This inconsistency in admin down behavior is being investigated.


RN-384 (CM-7684)
Keeping VXLAN single-connected devices up on MLAG secondary node In the current MLAG secondary design, if the VXLAN device is not dual-connected, it is kept in a protodown state. You can keep them up with individual IP addresses rather than anycast IPs when the peerlink is down, so that all single-connected hosts will have connectivity. Further investigation regarding this issue is underway.

RN-387 (CM-8163)
Quagga appears to not honor passive interfaces if VRR is active

In a VRR configuration, any interface-specific routing configuration (e.g., OSPF mode of operation) specified on the subinterface having a virtual IP address does not take effect. This is because when an operator has specified a virtual IP on a bridge, the system creates another internal interface bridge with the virtual IP and MAC. These two interfaces are treated distinctly by Quagga, so any interface-specific routing configuration on the bridge does not get carried over to the second bridge.

In a VRR deployment needing any interface-specific routing configuration on the interface with a virtual IP address, the routing configuration has to be specified against the internally-created virtual interface also.


RN-389 (CM-8410)
switchd supports only port 4789 as the UDP port for VXLAN packets

switchd currently allows only the standard port 4789 as the UDP port for VXLAN packets. There are cases where a hypervisor could be using non-standard UDP port, which would cause VXLAN exchanges with the hardware VTEP to not work. In such a case, packets would not be terminated and encapsulated packets would be sent out on UDP port 4789.


RN-390 (CM-9055)
If you’re logged into the serial console and type reboot, the system may hang indefinitely

In Cumulus Linux, if the serial console terminal settings are changed and clocal (modem carrier) is turned off,  systemd may do all of the following:

  • Block and stop handling systemctl changes
  • Fail to start or restart services
  • Prevent reboot and shutdown

For example, running the /usr/bin/reset command may cause this problem when run on the serial console. Once in this state, a new login session will not be started on the serial console (/dev/ttyS0 or ttyS1) after logout, and systemd will not respond to systemctl commands, including reboot.

To work around this issue, whenever you are logged in to the serial console and run reset, run stty clocal afterwards.


RN-391 (CM-9631)
Dell S4048 unresponsive after TX Unit Hang detected

After booting a Dell S4048 switch, the switch becomes unresponsive and errors like the following appear in the console log:

[ 1206.440277] igb 0000:00:14.0: Detected Tx Unit Hang
[ 1206.440277]   Tx Queue             <0>
[ 1206.440277]   TDH                  <2d>
[ 1206.440277]   TDT                  <2e>
[ 1206.440277]   next_to_use          <2e>
[ 1206.440277]   next_to_clean        <2d>
[ 1206.440277] buffer_info[next_to_clean]
[ 1206.440277]   time_stamp           <1000dcd20>
[ 1206.440277]   next_to_watch        <ffff88007d81b2d0>
[ 1206.440277]   jiffies              <1000dd5d4>
[ 1206.440277]   desc.status          <300000>
[ 1208.439856] igb 0000:00:14.0: Detected Tx Unit Hang
[ 1208.439856]   Tx Queue             <0>
[ 1208.439856]   TDH                  <2d>
[ 1208.439856]   TDT                  <2e>
[ 1208.439856]   next_to_use          <2e>
[ 1208.439856]   next_to_clean        <2d>
[ 1208.439856] buffer_info[next_to_clean]
[ 1208.439856]   time_stamp           <1000dcd20>
[ 1208.439856]   next_to_watch        <ffff88007d81b2d0>
[ 1208.439856]   jiffies              <1000ddda4>
[ 1208.439856]   desc.status          <300000>
[ 1210.439414] igb 0000:00:14.0: Detected Tx Unit Hang
[ 1210.439414]   Tx Queue             <0>
[ 1210.439414]   TDH                  <2d>
[ 1210.439414]   TDT                  <2e>
[ 1210.439414]   next_to_use          <2e>
[ 1210.439414]   next_to_clean        <2d>
[ 1210.439414] buffer_info[next_to_clean]
[ 1210.439414]   time_stamp           <1000dcd20>
[ 1210.439414]   next_to_watch        <ffff88007d81b2d0>
[ 1210.439414]   jiffies              <1000de574>
[ 1210.439414]   desc.status          <300000>
[ 1212.438966] igb 0000:00:14.0: Detected Tx Unit Hang
[ 1212.438966]   Tx Queue             <0>
[ 1212.438966]   TDH                  <2d>
[ 1212.438966]   TDT                  <2e>
[ 1212.438966]   next_to_use          <2e>
[ 1212.438966]   next_to_clean        <2d>
[ 1212.438966] buffer_info[next_to_clean]
[ 1212.438966]   time_stamp           <1000dcd20>
[ 1212.438966]   next_to_watch        <ffff88007d81b2d0>
[ 1212.438966]   jiffies              <1000ded44>
[ 1212.438966]   desc.status          <300000>
[ 1212.490329] igb 0000:00:14.0 eth0: Reset adapter

Rebooting the switch again stops the behavior.


RN-404 (CM-4407)
Aggregating routes in BGP with as-set can result in high CPU usage

When BGP is configured with aggregate addresses with as-set configuration and there are many routes to be aggregated, the BGP process gets into high CPU usage.

To work around this issue, do not specify the as-set parameter for the aggregate-address configuration.


RN-409 (CM-10054)
BGP may show an inaccessible path as the best path

Existing BGP issues caused peering between a VRF device and a loopback BGP session to stay up if the loopback session doesn’t advertise its local address.

This issue will be fixed in a future release.


RN-446 (CM-10513)
Redistribute neighbor does not work with more than 1024 interfaces

The rdnbrd service crashes because it cannot work with more than 1024 interfaces.

This issue should be fixed in a future release of Cumulus Linux.


RN-448 (CM-11302)
Using the json option in the "show ip bgp" command causes peer session flaps

This issue causes peer session flaps on Penguin Arctica 4806XP and Supermicro SSE-X3648S switches. It occurs with 16K IPv4 prefixes and only when you run show ip bgp json.

However, on switches with Tomahawk ASICs, with 61K IPv4 prefixes and default timers, the same show ip bgp json command causes all peer sessions to go down.

This is a known issue that should be fixed in a future release of Cumulus Linux.


RN-450 (CM-12252)
802.1p remark in traffic.conf behaves differently on Mellanox vs. Broadcom switches

The 802.1p remark defined in traffic.conf acts differently on a Mellanox switch when compared to a Broadcom switch.

On the Mellanox platform, the remark defined in the traffic.conf file takes precedence even if there is an ACL rule that is matched.

On the Broadcom platform, the ACL rule takes precedence over the remark defined in the traffic.conf file.


RN-451 (CM-12344)
Mellanox switch rejects SPAN ACL rule for an output interface that is a subinterface

This is a known issue at this time.


RN-455 (CM-12578)
The cumulus-poe package is not installed on ARM switches after upgrading to version 3.1

After upgrading Cumulus Linux from a binary image installation of version 3.1 or earlier, the cumulus-poe package does not get installed on ARM-based switches. THe following message appears after you reboot the switch:

[ OK ] Stopped Cumulus Linux POE Daemon.
Starting Cumulus Linux POE Daemon...
[FAILED] Failed to start Cumulus Linux POE Daemon.

In order to use Power over Ethernet (PoE) on an ARM switch, you need to install the cumulus-poe package:

cumulus@switch:~$ sudo apt-get update
cumulus@switch:~$ sudo apt-get install cumulus-poe
cumulus@switch:~$ sudo apt-get upgrade

RN-525 (CM-12715)
On a Quanta IX1 switch, 100G OSI LR4 module (QSFP28-LR4-OSI) doesn't advertise 40G in transceiver codes, so cannot be used at 40G speed

Cumulus Linux can only support speeds on modules that are advertised in each module's transceiver codes. You can determine this by running ethtool -m:

cumulus@switch:~$ sudo ethtool -m
...

    Transceiver codes                       : 0x80 0x00 0x00 0x00 0x00 0x00 0x00 0x00
    Transceiver type                        : 100G Ethernet: 100G Base-LR4

...

The example above shows that this module supports only 100G; it cannot support 40G speeds. Similarly, if a module advertises 40G support only, it cannot support 100G speeds.


RN-526 (CM-13037)
Upgrading the clag package fails when logrotate contains an invalid date

While upgrading from Cumulus Linux 3.0.1 to 3.1.z, the clag package does not update if the logrotate file contains an invalid date. This can occur due to bad batteries in the switch or RTC clock chip issues. The error may look like the following:

error: bad year 1929 for file /var/log/boot.log in state file /var/lib/logrotate/status
dpkg: error processing package clag (--configure):
 subprocess installed post-installation script returned error exit status 1
Errors were encountered while processing:
 clag
E: Sub-process /usr/bin/dpkg returned an error code (1)

You can work around this issue by removing the /var/lib/logrotate/status file, and forcing a logrotate. After you do this, the upgrade should be successful.


RN-536 (CM-12628)
Port LEDs on ARM 1G-T platforms do not show link status (green), but blink on activity

On the DNI-3048UP, Edge-Core 4610-54P and HP 4610-54T, there is no link LED for 1G-T ports. The SFP+ ports correctly show the link LEDs.

The Dell S3048-ON, which has a Helix4 ASIC, correctly shows the link LED.

This issue is currently being investigated.


RN-537 (CM-12967)
Pause frames sent by a Tomahawk switch are not honored by the upstream switch

An issue exists when link pause or priority flow control (PFC) is enabled on a Broadcom Tomahawk-based switch, and there is over-subscription on a link, where the ASIC sends pause frames aggressively, causing the upstream switch not to throttle enough.

If you need link pause or PFC functionality, then to work around this issue, you must use a switch that does not use the Tomahawk ASIC.


RN-538 (CM-13106)
Link issue between Edge-Core AS4610-54P and Supermicro server with embedded NIC x552 2x10G SFP+

The Supermicro server reports via dmesg that when the link starts to come up, it begins flapping, and eventually comes up after 15-45 minutes. The link remains stable if no link down event occurs.

If the link is brought down (due to cable disconnect, or link set down/up), then the flap occurs again the next time the link tries to come up.

This issue is currently being investigated.


RN-539 (CM-13385)
On Penguin Arctica 4804IP, the default ebtables rules to send LACP/LLDP packets to queue 7/6 does not work as expected

An issue exists where INPUT chain rules with input interface match (including swp+ wildcard interface match) that match multicast packets will not be hit. This means that default rules that match and set class on these packets (LLDP/LACP for example) will not take effect. Their class will be zero.

The workaround for this issue is to match these packets in the FORWARD chain. The side effect of this is that even forwarded copies will have a higher class set.


RN-540 (CM-13428)
Quagga reload fills FIB after renaming the table map

Trying to rename a table map — essentially deleting the current table map and adding a new one — causes Quagga to try to install all the routes again, which fills the FIB. The table map begins functioning again on its own after some time and without any intervention, and the FIB usage returns to a normal level.

This issue is currently being investigated.


RN-541 (CM-13460)
ebtables counters get cleared after an iptables/ip6tables incremental rule update

On a Penguin Arctica 4804IP switch, when any iptables or ip6tables ACL rule gets updated, the ebtables counters get reset.

This issue is currently being investigated. There is no workaround at this time.


RN-542 (CM-13461)
Polling the BGP RIB with "show ip bgp" causes the peer to flap if the RIB has more than 600K entries

This is a known issue that's currently being investigated. The Quagga log shows these commands taking a very long to execute.

To work around this issue, Cumulus Networks recommends you use larger keepalive/hold timers — 60 and 180 seconds, respectively.


RN-543 (CM-13649)
Adding a large number of ACL rules with NCLU takes an excessively long time

Using the network command line utility (NCLU) to add a large number of ACL rules may take a very long time. For example, it was reported that adding 100 rules took about 19 seconds. In addition, CPU usage is high during this time. The system doesn't hang, it's just launching an instance of netd every time a rule is added.

This issue is currently being investigated.


RN-544 (CM-13739)
Speed 1000 breaks porttab on Edge-Core AS5812-54X, changing the interface name from xe to ge When the port speed is changed from 10G/40G to 1G, the SDK dport interface name changes from xe to ge. This causes switchd APIs that use the BCP dport name to fail when the logical port is not generated. There is currently no fix for a persistent configuration.

RN-545 (CM-13800)
OSPFv3 redistribute connected with route-map broken at reboot (or ospf6d start) This issue only affects OSPFv3 (IPv6) and is being investigated at this time.

RN-546 (CM-14051)
netd crashes at "snapper list" after running "net show commit history"

This issue has been seen on switches that upgraded from a version of Cumulus Linux earlier than 3.2.0. To work around the issue, install the cumulus-snapshot package on the switch. This activates the NCLU rollback capability.

cumulus@switch:~$ sudo apt-get install cumulus-snapshot

RN-547 (CM-14060)
First Quagga reload causes routing changes for an older configuration (earlier than version 3.0)

For older style Quagga configurations (from Cumulus Linux versions earlier than 3.0), the first time Quagga is reloaded after upgrading to Cumulus Linux 3.y.z, the configuration gets overwritten with the version 3.y.z defaults. This affects these two configuration settings specifically:

  • The "bgp bestpath as-path multipath-relax no-as-set" setting, which was replaced with no-as-set as the default, so it no longer needs to be specified. This causes all BGP sessions to be reset.
  • The IP import-table, which is used by redistribute neighbor. This causes Cumulus Linux to unimport the routes and then add them back again, causing routing prefix changes throughout the network, potentially blackholing traffic.

This issue is currently being investigated.


RN-548 (CM-14061)
ethtool --show-fec does not reflect correct state The ethtool --show-fec command does not accurately report the current link state. The cause of this issue is being investigated.

RN-549 (CM-14106)
For a 100G capable port and cable, enabling auto-negotiation fails when the port speed is set to 40G in ports.conf An issue is present on 100G capable ports and cables, where auto-negotiation fails when set to 40G in the /etc/cumulus/ports.conf file. This issue is currently being investigated.

RN-550 (CM-13674)
The ZTP daemon shuts itself down after 5 minutes of inactivity

The zero touch provisioning (ZTP) daemon ztpd shuts itself down after 5 minutes of inactivity but the service remains enabled for the next reboot.

This can affect deployments where a switch might be powered up in a remote data center for weeks without ever being configured. In such a case, there is no way to automatically initiate the ZTP process.

This is a known issue that will be fixed in a future release of Cumulus Linux.


RN-571 (CM-14791)
Management VRF loses default gateway with gateway command

When a management VRF was configured using NCLU, the route to the default gateway failed and was considered invalid.

To work around this issue, manually add post-up ip route add vrf mgmt default via IP_ADDRESS to the /etc/network/interfaces file, then run ifreload -a.


RN-576 (CM-14908)
TACACS sends authentication requests out of the default VRF, not the management VRF

If a management VRF if configured, TACACS won't send authentication requests out of the management VRF. Instead, it sends these requests out of the default VRF.

To work around this issue, run the following commands, which restrict inbound SSH to only the management VRF interface and disable inbound SSH via the switch ports. Note that using SSH via the front panel ports is not a workaround.

cumulus@switch:~$ sudo systemctl disable ssh.service
cumulus@switch:~$ sudo systemctl stop ssh.service
cumulus@switch:~$ sudo systemctl enable ssh@mgmt.service
cumulus@switch:~$ sudo systemctl start ssh@mgmt.service
Have more questions? Submit a request

Comments

Powered by Zendesk