KVMAutotest Networking

KVM autotest networking documentation

Configuration

How to configure to allow all the traffic to be forwarded across the virbr0 bridge:

echo "-I FORWARD -m physdev --physdev-is-bridged -j ACCEPT" > /etc/sysconfig/iptables-forward-bridged
lokkit --custom-rules=ipv4:filter:/etc/sysconfig/iptables-forward-bridged
service libvirtd reload

How to configure Static IP address in KVM autotest

Sometimes, we need to test with guest(s) which have static ip address(es).

• eg. No real/emulated DHCP server in test environment.
• eg. Test with old image we don't want to change the net config.
• eg. Test when DHCP exists problem.

Create a bridge (for example, 'vbr') in host, configure its ip to 192.168.100.1, guest can access host by it. And assign nic(s)' ip in tests.cfg, and execute test as usual.

tests.cfg:

ip_nic1 = 192.168.100.119
nic_mac_nic1 = 11:22:33:44:55:67
bridge = vbr

TestCases

Ntttcp

The Nttcp test suite is a network performance test for windows, developed by Microsoft. It is not a freely redistributable binary, so you must download it from the website, here's the direct link for download (keep in mind it might change):

http://download.microsoft.com/download/f/1/e/f1e1ac7f-e632-48ea-83ac-56b016318735/NT%20Testing%20TCP%20Tool.msi

The knowledge base article associated with it is:

http://msdn.microsoft.com/en-us/windows/hardware/gg463264

You need to add the package to winutils.iso, the iso with utilities used to test windows. First, download the iso. The can help you out with downloading if you like it, but the direct download link is here:

http://people.redhat.com/mrodrigu/kvm/winutils.iso

You need to put all its contents on a folder and create a new iso. Let's say you want to download the iso to /home/kermit/Downloads/winutils.iso. You can create the directory, go to it:

mkdir -p /home/kermit/Downloads
cd /home/kermit/Downloads

Download the iso, create 2 directories, 1 for the mount, another for the contents:

wget http://people.redhat.com/mrodrigu/kvm/winutils.iso
mkdir original
sudo mount -o loop winutils.iso original
mkdir winutils

Copy all contents from the original cd to the new structure:

cp -r original/* winutils/

Create the destination nttcp directory on that new structure:

mkdir -p winutils/NTttcp

Download the installer and copy autoit script to the new structure, unmount the orginal mount:

cd winutils/NTttcp
wget http://download.microsoft.com/download/f/1/e/f1e1ac7f-e632-48ea-83ac-56b016318735/NT%20Testing%20TCP%20Tool.msi -O "winutils/NTttcp/NT Testing TCP Tool.msi"
cp /usr/local/autotest/client/virt/scripts/ntttcp.au3 ./
sudo umount original

Backup the old winutils.iso and create a new winutils.iso using mkisofs:

sudo mv winutils.iso winutils.iso.bak
mkisofs -o winutils.iso -max-iso9660-filenames -relaxed-filenames -D --input-charset iso8859-1 winutils

And that is it. Don't forget to keep winutils in an appropriate location that can be seen by KVM autotest.

Vlan

Bonding

Netperf

Jumbo

Ethtool

Ping

Pxe

VM network infrastructure rewrite

Mid-2012, the virtualization networking code underwent a lot of changes, with the aim to make things cleaner and more flexible for the KVM and libvirt tests. Here is the complete changelog:

• Integrated new network code with kvm module
• Added many logging and debugging statements for new code
• unit tests: New virt_utils_unittests tests
• Updated shared ping test module
• Updated kvm nic_bonding test module
• Updated kvm nic_hotplug test module comment
• virt_utils Env init now always warns when creating an empty env file
• virt_utils Env init prints warning traceback when loading env fails
• Pickling support for all new classes
• Raise exception when adding nic with clashing name
• New virt_utils_unittest deletes temporary testing database
• IP/MAC cache is only updated with new entries
• VM cleanup resets static addresses properly
• Enhanced output of needs_restart() VM comparison function
• Enhanced Env class to output more details on non-fatal load failures
• Old virt_util functions all raise depreciation warnings to aide bug squashing

• remove unnecessary nic_mode from subtests parameter, rely on default

  nettype=bridge instead

• Implimented new nic/net management within libvirt_vm
• Added stubbs for nic activation/deactivation in libvirt_vm
• Removed extranious netdev_id/device_id/tap/ifname support from libvirt_vm
• Enhanced logging for vm.needs_restart()
• Changed libvirt vm to only free mac addresses for undefined VMs
• Improved libvirt vm.is_persistent() to handle non-existing VMs
• Defined mac address prefix convensions for libvirt (kvm/xen)

• virt config: New universal virt. networking configuration

  nettype + netdst parameters determine networking style and destination for virtual NICs. Separate NIC sets can be defined for each VM. Documented examples of other per-nic parameters in config. comments.

• virt_utils change etails:

New Class: PropCan

Allow container-like access to fixed set of items (__slots__) This allows NIC/Net handling code to lock-down a fixed set of parameters relevant to a vm_type / driver_type combination. Data is accessible in a dictionary-like manner or as object properties, however empty or None value items will pretend to not exist. The special empty/None handling optimises access and storage by only representing values containing meaningful data.

New Class: Base_VirtIface

Networking information for single guest interface and host connection. This is an abstract base-child-class of PropCan containing the minimal amount of networking keys needed, along with some basic name and mac address handling methods. Some amount of effort is made to ensure stored mac address data is in an expected format - two ASCII character hex-byte representation, separated by :'s, six or less bytes in total. Complete mac addresses are generated by filling in missing trailing bytes from a sequential pool maintained as a class variable. Numbering starts randomly from 0-255, then proceeds incrementally for each byte requested. Sequential byte generation used to ease unit testing.

New Class: Libvirt_VirtIface & KVM_VirtIface

Specializations of Base_VirtIface parent class (and PropCan) with networking keys specific to Libvirt or KVM respectively. If a new type or variation is needed, simply subclass Base_VirtIface and update the VMNet_Style.VMNet_Style_Map as needed. The new or variant class will be used automatically based on resolving vm_type, and/or driver_type from the supplied Params dict (see other classes below)

New Class: VMNet

Abstract base-class collection of networking information, stored within a particular container class (e.g. a subclass of Base_VirtIface). This class behaves like a list of PropCans. It may be initially empty or created from a list-like container of dictionary-like objects holding key/value pairs of networking data. Items within a VMNet instance are indexed by number and the 'nic_name' value of each contained item. Consequently, the 'nic_name' key is required for each contained item, and it's value must be hashable. Updates to the 'mac' value of each item is handled specially, in that storage of incomplete byte lists, and/or values that do not pass the container class's mac_is_valid method, will not be stored.

New Class: VMNet_Style

Make decisions about needed info from vm_type and driver_type params. The VMNet_Style_Map class variable is utilized for initial setup of VMNet derivatives (see below). Essentially it helps provide a container-class type based on the contents of Params dict-like values for vm_type and driver_type. It also establishes a mac address prefix to be used when generating MAC addresses. However, any number of additional VMNet instance variables may also be set with default values. The special 'default' keys are utilized when vm_type and/or driver_type don't match from Params.

New Class: ParamsNet

VMNet sub-class specialized for initialization from a Params dict-like instance. Essentially it filters out and initializes all networking info. for a particular VM. Once initialized, networking information is accessed by NIC index or it's name (e.g. 'nic1', 'nic2', etc.) in a dictionary-like or object property-like manner. For example, instance[0].mac, instance('nic1').mac, instance['nic1']['mac'], etc. Each nic object will be stored in the proper container class as determined by the VMNet_Style class. A utility generator method is provided for iteration over all VM's and NIC's mac address params definitions. This makes validating or checking for mac address conflicts easy.

New Class DbNet:

VMNet sub-class specialized for initialization from a database. Essentially analogous to ParamsNet, with a set of database management and locking methods. Database items are hashed by the key provided at initialization time, so hash-clashes must be managed by implementation code. Changes to any indexed NIC object automatically result in atomic database updates. This also follows for any re-ordering or deletion of contained items. The lock/unlock methods must be called externally prior to any database activity. Also provides a utility generator function for iteration over all MAC addresses contained within the database.

New Class: VirtNet (Re-write former networking code)

Multi-inheritance of the ParamsNet and DbNet classes providing mediated networking info. management behavior combination of both. Provides the former generate_mac_address(), free_mac_address, set_mac_address(), and get_mac_address() functions, which behave exactly as before. The only exception is the 'instance' parameter is no-longer required as it is established at initialization time. Setting and Freeing of arbitrary mac addresses is allowed (as before), however generation of new mac addresses is a sieve over cached values. Deleting the database cache or freeing individual NICs MAC addresses will extend the pool within the confines of any mac prefixes defined by the responsible VMNet_Style definition. A utility generator is provided which wrapps the mac_index generators from ParamsNet and DbNet (in that order). This permits iteration over all known MAC address values possible.

• Libvirt change details:

  Similar to KVM (below), however activate/deactivate_nic() are not

  yet implimented - no change from previous, unsupported hotplugging

  KVM/Xen guests make use of specialized MAC address prefixes

  • vm_type/driver_type combination chooses networking parameters
  • vm_type/driver_type combination dictates mac address prefix

• KVM change details:

  Intended CodePath:

  Base Class __init__:

  Load/Parse Cartesian params + address_pool cache

  Base Class nic addition:

  BaseVM.add_nic BaseVM.activate_nic <testing> BaseVM.deactivate_nic BaseVM.del_nic

  KVM VM class __init__:

  Call BaseVM.__init__

  KVM VM create:

  vm.create:

  vm.add_nic -> vm.add_netdev specialized nic & netdev activation

  <testing>

  KVM VM nic addition:

  vm.add_netdev -or- vm.add_nic -> vm.add_netdev

  Uses BaseVM.add_nic

  vm.activate_netdev vm.activate_nic <testing> vm.deactivate_netdev vm.del_netdev vm.deactivate_nic vm.del_nic

  Modules Changes summary:

  BaseVM module

  Before:

  Different behavior per sub-class

  After:

  Defined new VMUnknownNetTypeError, raised when validating 'nettype' parameter doesn't match supported type. Supported types left up to sub-classes to define.

  BaseVM.init

  Before:

  Subclasses and methods deal with params and cache DB ad hoc

  After:

  Common setup/validate networking from parameters and cache DB input

  BaseVM._generate_unique_id

  Before:

  vm.instance unique across thousands of VM instances created per second

  After:

  vm.instance unqie across millions of VM instances created per second

  BaseVM.get_mac_address

  Before:

  Obtain mac defined in params or cache DB input. Raise VMMACAddressMissingError if no mac defined (note: doesn't actually check vm object in any way).

  After:

  Loading mac from params and/or cache db moved to __init__. Check instance variables for mac and return or raise VMMACAddressMissingError

  BaseVM.get_address

  Before:

  Require sub-classes to impliment.

  After:

  Rely on __init__ to parse options and cache DB. If nic has mac defined, and corresponding IP in IP/MAC cache, verify in host arp cache then return IP. Return localhost if no mac address defined for nic.

  BaseVM.free_mac_address

  Before:

  Different behavior per sub-class

  After:

  Unified behavior to drop mac reference only if they're not staticly defined. When static, the original mac from params is re-established for the nic.

  BaseVM.add_nic

  Before:

  Left up to subclasses

  After:

  Adds new or updates existing NIC definition on vm instance with

  new or changed options. Generates random nic_name if none specified. If IP and MAC are defined, updates IP/MAC cache. Otherwise, generates random mac address. Returns standardized instance of dict-like object containing NIC new/existing (possibly changed) parameters.

  BaseVM.del_nic

  Before:

  Left up to subclasses

  After:

  Drop references to mac in cache DB. Drop references to IP

  from IP/MAC cache. Ensure NIC parameters removed from vm instance.

  BaseVM.activate_nic

  Before:

  Host resource allocation bundled with sub-class add_nic method.

  After:

  Require sub-classes to implement separate from add_nic.

  BaseVM.deactivate_nic

  Before:

  Host resource deallocation bundled with sub-class add_nic method.

  After:

  Require sub-classes to implement separate from del_nic

  kvm_vm.init

  Before: Track nic-indexed netdev_id, device_id, and tapfd separately. After: Moved into BaseVM.__init__, single virt_utils.virtnet class.

  kvm_vm.clone

  unchanged- rely on __init__ to produce new virtnet instance

  kvm_vm.__make_qemu_command

  add_nic: unchanged add_net: unchanged Body: Referenced names, and assignment-order changes / updates

  so may be called independently of vm.create().

  kvm_vm.create

  Before:

  For each nic- Parse params, generating random netdev_id, device_id. For tap a.k.a. bridge devices, obtain ifname, bridge name, tapfd,

  add to bridge, then raise interface

  For each nic- obtain mac or generate random mac

  if params define ip address,update address cache

  __make_qemu_command background-run qemu command for each open tapfd- close tapfd

  After:

  (Rely on __init__ to parse params and cache db content) For each nic, use common add_nic(), but specialize activation:

  if nettype is bridge/tap:

  open tap device node register tapfd and tapfd_id to qemu add interface to bridge raise interface

  if nettype is user:

  log info

  __make_qemu_command background-run qemu command for each already parsed nic:

  if tapfd open, close tapfd

  kvm_vm.destroy

  Same behavior, updated names

  kvm_vm.get_port

  Same behavior, updated names

  kvm_vm.get_peer

  Unchanged

  kvm_vm.get_address

  Before:

  Parses params for IP, if bridge/TAP mode defined for nic, parses params for mac address. Consults IP/MAC cache DB for IP. Returns localhost if no IP defined or nic_mode is not 'tap'.

  After:

  Functionality made common by implimentation in BaseVM class.

  kvm_vm.get_ifname

  Before:

  Return the letter 't' followed by the vm's nic index, followed by a '-' and the last 11 characters of vm.instance.

  After:

  Retrieve ifname (generated in vm.add_netdev) for NIC or raise KeyError.

  kvm_vm.get_mac_address

  Before:

  Duplicated code from old BaseVM method with same name

  After:

  Functionality made common by implimentation in BaseVM class.

  kvm_vm.free_mac_address

  Before:

  Use virt_utils function to drop mac reference from cache DB.

  After:

  Functionality made common by implementation in BaseVM class.

  kvm_vm.add_netdev

  Before:

  Parse params for bridge name, generate random netdev_id, set vlan to number of other tap NICs on VM. Generate ifname, open tap device node, set tapfd, bring up interface, generate random tapfd_id, run 'netdev_add' monitor command. Verify netdev_id shows up in network info. Return netdev_id.

  After:

  Rely on BaseVM.__init__ to parse options and cache DB for bridge name. Validate supported nettype specified in parameters ('bridge' or 'user') Generate random netdev_id, use defined vlan or set based on # total of nics on VM., and generate ifname. Open tap device node moved to vm.activate_netdev(). Generate random tapfd_id. Running 'netdev_add' moved to vm.activate_netdev(). Return netdev_id. Moved running of 'netdev_add' functionality into vm.activate_netdev.

  kvm_vm.del_netdev

  Before:

  Run 'netdev_del' monitor command. Validate netdev_id no longer shows in network info.

  After:

  Moved 'netdev_del' monitor command and validation to vm.deactivate_nic(). Remove properties from nic: netdev_id, vlan, ifname, tapfd, tapfd_id.

  kvm_vm.add_nic

  Before:

  Take fixed set of nic parameters, generate a name for nic (nic_id) if not specified. Call add_netdev() if netdev_id parameter not specified. Generate random mac address if mac parameter not specified. Setup/Run device_add monitor command with id=nic_id. Return dict containing nic parameters. Verify nic_id shows up in qtree.

  After:

  Take arbitrary set of key/value nic prameters. Call BaseVM.add_nic method with nic parameters. Generate random device_id. Call add_netdev with parameters if no netdev_id specified. Return standardized dict-like object containing nic parameters. Moved call to device_add command into vm.activate_nic method.

  kvm_vm.del_nic

  Before:

  Runs device_del monitor command. Checks for wait timeout for nic to not show in qtree.

  After:

  Call BaseVM's del_nic method. Functionality to call device_del moved to vm.deactivate_nic

  kvm_vm.activate_netdev

  Before:

  Functionality combined with vm.add_netdev

  After:

  Re-validate nettype is supported, setup netdev_add monitor command with parameters from nic based on nettype value. Run netdev_add monitor command. Verify netdev_id shows up in network info.

  kvm_vm.deactivate_netdev

  Before:

  Functionality combined with vm.del_netdev

  After:

  Call netdev_del monitor command, verify netdev_id no longer shows in network info.

  kvm_vm.activate_nic

  Before:

  Functionality combined into vm.add_nic

  After:

  Setup/run device_add monitor command with id=nic_name. Verify nic_name shows up in qtree.

  kvm_vm.deactivate_nic

  Before:

  Functionality combined with vm.del_nic

  After:

  Run 'device_del' monitor command. Validate netdev_id unknown to monitor. Raise virt_vm.VMDelNicError if qtree shows nic_name after wait time.

  kvm_vm.migrate

  Unchanged