X-Git-Url: https://code.grnet.gr/git/ganeti-local/blobdiff_plain/daa49d6f7761547e5776d9dce65ac33bac8382da..df07c18f5186c3018a45333accd58b5f70a1e581:/lib/hypervisor/hv_kvm.py diff --git a/lib/hypervisor/hv_kvm.py b/lib/hypervisor/hv_kvm.py index a388ec2..1564309 100644 --- a/lib/hypervisor/hv_kvm.py +++ b/lib/hypervisor/hv_kvm.py @@ -1,7 +1,7 @@ # # -# Copyright (C) 2008, 2009, 2010, 2011, 2012 Google Inc. +# Copyright (C) 2008, 2009, 2010, 2011, 2012, 2013 Google Inc. # # This program is free software; you can redistribute it and/or modify # it under the terms of the GNU General Public License as published by @@ -37,10 +37,15 @@ import shutil import socket import stat import StringIO +from bitarray import bitarray try: import affinity # pylint: disable=F0401 except ImportError: affinity = None +try: + import fdsend # pylint: disable=F0401 +except ImportError: + fdsend = None from ganeti import utils from ganeti import constants @@ -79,8 +84,150 @@ _SPICE_ADDITIONAL_PARAMS = frozenset([ constants.HV_KVM_SPICE_USE_TLS, ]) +# Constant bitarray that reflects to a free pci slot +# Use it with bitarray.search() +_AVAILABLE_PCI_SLOT = bitarray("0") + +# below constants show the format of runtime file +# the nics are in second possition, while the disks in 4th (last) +# moreover disk entries are stored in tupples of L{objects.Disk}, dev_path +_KVM_NICS_RUNTIME_INDEX = 1 +_KVM_DISKS_RUNTIME_INDEX = 3 +_DEVICE_RUNTIME_INDEX = { + constants.HOTPLUG_TARGET_DISK: _KVM_DISKS_RUNTIME_INDEX, + constants.HOTPLUG_TARGET_NIC: _KVM_NICS_RUNTIME_INDEX + } +_FIND_RUNTIME_ENTRY = { + constants.HOTPLUG_TARGET_NIC: + lambda nic, kvm_nics: [n for n in kvm_nics if n.uuid == nic.uuid], + constants.HOTPLUG_TARGET_DISK: + lambda disk, kvm_disks: [(d, l) for (d, l) in kvm_disks + if d.uuid == disk.uuid] + } +_RUNTIME_DEVICE = { + constants.HOTPLUG_TARGET_NIC: lambda d: d, + constants.HOTPLUG_TARGET_DISK: lambda (d, e): d + } +_RUNTIME_ENTRY = { + constants.HOTPLUG_TARGET_NIC: lambda d, e: d, + constants.HOTPLUG_TARGET_DISK: lambda d, e: (d, e) + } + + +def _GenerateDeviceKVMId(dev_type, dev): + """Helper function to generate a unique device name used by KVM + + QEMU monitor commands use names to identify devices. Here we use their pci + slot and a part of their UUID to name them. dev.pci might be None for old + devices in the cluster. + + @type dev_type: sting + @param dev_type: device type of param dev + @type dev: L{objects.Disk} or L{objects.NIC} + @param dev: the device object for which we generate a kvm name + @raise errors.HotplugError: in case a device has no pci slot (old devices) + + """ + + if not dev.pci: + raise errors.HotplugError("Hotplug is not supported for %s with UUID %s" % + (dev_type, dev.uuid)) + + return "%s-%s-pci-%d" % (dev_type.lower(), dev.uuid.split("-")[0], dev.pci) + + +def _UpdatePCISlots(dev, pci_reservations): + """Update pci configuration for a stopped instance + + If dev has a pci slot then reserve it, else find first available + in pci_reservations bitarray. It acts on the same objects passed + as params so there is no need to return anything. + + @type dev: L{objects.Disk} or L{objects.NIC} + @param dev: the device object for which we update its pci slot + @type pci_reservations: bitarray + @param pci_reservations: existing pci reservations for an instance + @raise errors.HotplugError: in case an instance has all its slot occupied + + """ + if dev.pci: + free = dev.pci + else: # pylint: disable=E1103 + [free] = pci_reservations.search(_AVAILABLE_PCI_SLOT, 1) + if not free: + raise errors.HypervisorError("All PCI slots occupied") + dev.pci = int(free) + + pci_reservations[free] = True + + +def _GetExistingDeviceInfo(dev_type, device, runtime): + """Helper function to get an existing device inside the runtime file + + Used when an instance is running. Load kvm runtime file and search + for a device based on its type and uuid. + + @type dev_type: sting + @param dev_type: device type of param dev + @type device: L{objects.Disk} or L{objects.NIC} + @param device: the device object for which we generate a kvm name + @type runtime: tuple (cmd, nics, hvparams, disks) + @param runtime: the runtime data to search for the device + @raise errors.HotplugError: in case the requested device does not + exist (e.g. device has been added without --hotplug option) or + device info has not pci slot (e.g. old devices in the cluster) -def _ProbeTapVnetHdr(fd): + """ + index = _DEVICE_RUNTIME_INDEX[dev_type] + found = _FIND_RUNTIME_ENTRY[dev_type](device, runtime[index]) + if not found: + raise errors.HotplugError("Cannot find runtime info for %s with UUID %s" % + (dev_type, device.uuid)) + + return found[0] + + +def _AnalyzeSerializedRuntime(serialized_runtime): + """Return runtime entries for a serialized runtime file + + @type serialized_runtime: string + @param serialized_runtime: raw text data read from actual runtime file + @return: (cmd, nics, hvparams, bdevs) + @rtype: list + + """ + loaded_runtime = serializer.Load(serialized_runtime) + if len(loaded_runtime) == 3: + serialized_blockdevs = [] + kvm_cmd, serialized_nics, hvparams = loaded_runtime + else: + kvm_cmd, serialized_nics, hvparams, serialized_blockdevs = loaded_runtime + + kvm_nics = [objects.NIC.FromDict(snic) for snic in serialized_nics] + block_devices = [(objects.Disk.FromDict(sdisk), link) + for sdisk, link in serialized_blockdevs] + + return (kvm_cmd, kvm_nics, hvparams, block_devices) + + +def _GetTunFeatures(fd, _ioctl=fcntl.ioctl): + """Retrieves supported TUN features from file descriptor. + + @see: L{_ProbeTapVnetHdr} + + """ + req = struct.pack("I", 0) + try: + buf = _ioctl(fd, TUNGETFEATURES, req) + except EnvironmentError, err: + logging.warning("ioctl(TUNGETFEATURES) failed: %s", err) + return None + else: + (flags, ) = struct.unpack("I", buf) + return flags + + +def _ProbeTapVnetHdr(fd, _features_fn=_GetTunFeatures): """Check whether to enable the IFF_VNET_HDR flag. To do this, _all_ of the following conditions must be met: @@ -97,20 +244,19 @@ def _ProbeTapVnetHdr(fd): @param fd: the file descriptor of /dev/net/tun """ - req = struct.pack("I", 0) - try: - res = fcntl.ioctl(fd, TUNGETFEATURES, req) - except EnvironmentError: - logging.warning("TUNGETFEATURES ioctl() not implemented") - return False + flags = _features_fn(fd) - tunflags = struct.unpack("I", res)[0] - if tunflags & IFF_VNET_HDR: - return True - else: - logging.warning("Host does not support IFF_VNET_HDR, not enabling") + if flags is None: + # Not supported return False + result = bool(flags & IFF_VNET_HDR) + + if not result: + logging.warning("Kernel does not support IFF_VNET_HDR, not enabling") + + return result + def _OpenTap(vnet_hdr=True): """Open a new tap device and return its file descriptor. @@ -139,39 +285,15 @@ def _OpenTap(vnet_hdr=True): try: res = fcntl.ioctl(tapfd, TUNSETIFF, ifr) - except EnvironmentError: - raise errors.HypervisorError("Failed to allocate a new TAP device") + except EnvironmentError, err: + raise errors.HypervisorError("Failed to allocate a new TAP device: %s" % + err) # Get the interface name from the ioctl ifname = struct.unpack("16sh", res)[0].strip("\x00") return (ifname, tapfd) -def _BuildNetworkEnv(name, network, gateway, network6, gateway6, - network_type, mac_prefix, tags, env): - """Build environment variables concerning a Network. - - """ - if name: - env["NETWORK_NAME"] = name - if network: - env["NETWORK_SUBNET"] = network - if gateway: - env["NETWORK_GATEWAY"] = gateway - if network6: - env["NETWORK_SUBNET6"] = network6 - if gateway6: - env["NETWORK_GATEWAY6"] = gateway6 - if mac_prefix: - env["NETWORK_MAC_PREFIX"] = mac_prefix - if network_type: - env["NETWORK_TYPE"] = network_type - if tags: - env["NETWORK_TAGS"] = " ".join(tags) - - return env - - class QmpMessage: """QEMU Messaging Protocol (QMP) message. @@ -226,30 +348,15 @@ class QmpMessage: return self.data == other.data -class QmpConnection: - """Connection to the QEMU Monitor using the QEMU Monitor Protocol (QMP). - - """ - _FIRST_MESSAGE_KEY = "QMP" - _EVENT_KEY = "event" - _ERROR_KEY = "error" - _RETURN_KEY = RETURN_KEY = "return" - _ACTUAL_KEY = ACTUAL_KEY = "actual" - _ERROR_CLASS_KEY = "class" - _ERROR_DATA_KEY = "data" - _ERROR_DESC_KEY = "desc" - _EXECUTE_KEY = "execute" - _ARGUMENTS_KEY = "arguments" - _CAPABILITIES_COMMAND = "qmp_capabilities" - _MESSAGE_END_TOKEN = "\r\n" +class MonitorSocket(object): _SOCKET_TIMEOUT = 5 def __init__(self, monitor_filename): - """Instantiates the QmpConnection object. + """Instantiates the MonitorSocket object. @type monitor_filename: string @param monitor_filename: the filename of the UNIX raw socket on which the - QMP monitor is listening + monitor (QMP or simple one) is listening """ self.monitor_filename = monitor_filename @@ -258,7 +365,6 @@ class QmpConnection: # in a reasonable amount of time self.sock.settimeout(self._SOCKET_TIMEOUT) self._connected = False - self._buf = "" def _check_socket(self): sock_stat = None @@ -266,29 +372,27 @@ class QmpConnection: sock_stat = os.stat(self.monitor_filename) except EnvironmentError, err: if err.errno == errno.ENOENT: - raise errors.HypervisorError("No qmp socket found") + raise errors.HypervisorError("No monitor socket found") else: - raise errors.HypervisorError("Error checking qmp socket: %s", + raise errors.HypervisorError("Error checking monitor socket: %s", utils.ErrnoOrStr(err)) if not stat.S_ISSOCK(sock_stat.st_mode): - raise errors.HypervisorError("Qmp socket is not a socket") + raise errors.HypervisorError("Monitor socket is not a socket") def _check_connection(self): """Make sure that the connection is established. """ if not self._connected: - raise errors.ProgrammerError("To use a QmpConnection you need to first" + raise errors.ProgrammerError("To use a MonitorSocket you need to first" " invoke connect() on it") def connect(self): - """Connects to the QMP monitor. + """Connects to the monitor. - Connects to the UNIX socket and makes sure that we can actually send and - receive data to the kvm instance via QMP. + Connects to the UNIX socket @raise errors.HypervisorError: when there are communication errors - @raise errors.ProgrammerError: when there are data serialization errors """ if self._connected: @@ -303,12 +407,53 @@ class QmpConnection: raise errors.HypervisorError("Can't connect to qmp socket") self._connected = True + def close(self): + """Closes the socket + + It cannot be used after this call. + + """ + self.sock.close() + + +class QmpConnection(MonitorSocket): + """Connection to the QEMU Monitor using the QEMU Monitor Protocol (QMP). + + """ + _FIRST_MESSAGE_KEY = "QMP" + _EVENT_KEY = "event" + _ERROR_KEY = "error" + _RETURN_KEY = RETURN_KEY = "return" + _ACTUAL_KEY = ACTUAL_KEY = "actual" + _ERROR_CLASS_KEY = "class" + _ERROR_DATA_KEY = "data" + _ERROR_DESC_KEY = "desc" + _EXECUTE_KEY = "execute" + _ARGUMENTS_KEY = "arguments" + _CAPABILITIES_COMMAND = "qmp_capabilities" + _MESSAGE_END_TOKEN = "\r\n" + + def __init__(self, monitor_filename): + super(QmpConnection, self).__init__(monitor_filename) + self._buf = "" + + def connect(self): + """Connects to the QMP monitor. + + Connects to the UNIX socket and makes sure that we can actually send and + receive data to the kvm instance via QMP. + + @raise errors.HypervisorError: when there are communication errors + @raise errors.ProgrammerError: when there are data serialization errors + + """ + super(QmpConnection, self).connect() # Check if we receive a correct greeting message from the server # (As per the QEMU Protocol Specification 0.1 - section 2.2) greeting = self._Recv() if not greeting[self._FIRST_MESSAGE_KEY]: self._connected = False - raise errors.HypervisorError("kvm: qmp communication error (wrong" + raise errors.HypervisorError("kvm: QMP communication error (wrong" " server greeting") # Let's put the monitor in command mode using the qmp_capabilities @@ -460,16 +605,18 @@ class KVMHypervisor(hv_base.BaseHypervisor): _CHROOT_DIR, _CHROOT_QUARANTINE_DIR, _KEYMAP_DIR] PARAMETERS = { + constants.HV_KVM_PATH: hv_base.REQ_FILE_CHECK, constants.HV_KERNEL_PATH: hv_base.OPT_FILE_CHECK, constants.HV_INITRD_PATH: hv_base.OPT_FILE_CHECK, constants.HV_ROOT_PATH: hv_base.NO_CHECK, constants.HV_KERNEL_ARGS: hv_base.NO_CHECK, constants.HV_ACPI: hv_base.NO_CHECK, constants.HV_SERIAL_CONSOLE: hv_base.NO_CHECK, + constants.HV_SERIAL_SPEED: hv_base.NO_CHECK, constants.HV_VNC_BIND_ADDRESS: (False, lambda x: (netutils.IP4Address.IsValid(x) or utils.IsNormAbsPath(x)), - "the VNC bind address must be either a valid IP address or an absolute" + "The VNC bind address must be either a valid IP address or an absolute" " pathname", None, None), constants.HV_VNC_TLS: hv_base.NO_CHECK, constants.HV_VNC_X509: hv_base.OPT_DIR_CHECK, @@ -479,7 +626,7 @@ class KVMHypervisor(hv_base.BaseHypervisor): constants.HV_KVM_SPICE_IP_VERSION: (False, lambda x: (x == constants.IFACE_NO_IP_VERSION_SPECIFIED or x in constants.VALID_IP_VERSIONS), - "the SPICE IP version should be 4 or 6", + "The SPICE IP version should be 4 or 6", None, None), constants.HV_KVM_SPICE_PASSWORD_FILE: hv_base.OPT_FILE_CHECK, constants.HV_KVM_SPICE_LOSSLESS_IMG_COMPR: @@ -513,8 +660,8 @@ class KVMHypervisor(hv_base.BaseHypervisor): hv_base.ParamInSet(False, constants.HT_KVM_VALID_MOUSE_TYPES), constants.HV_KEYMAP: hv_base.NO_CHECK, constants.HV_MIGRATION_PORT: hv_base.REQ_NET_PORT_CHECK, - constants.HV_MIGRATION_BANDWIDTH: hv_base.NO_CHECK, - constants.HV_MIGRATION_DOWNTIME: hv_base.NO_CHECK, + constants.HV_MIGRATION_BANDWIDTH: hv_base.REQ_NONNEGATIVE_INT_CHECK, + constants.HV_MIGRATION_DOWNTIME: hv_base.REQ_NONNEGATIVE_INT_CHECK, constants.HV_MIGRATION_MODE: hv_base.MIGRATION_MODE_CHECK, constants.HV_USE_LOCALTIME: hv_base.NO_CHECK, constants.HV_DISK_CACHE: @@ -531,8 +678,21 @@ class KVMHypervisor(hv_base.BaseHypervisor): hv_base.ParamInSet(True, constants.REBOOT_BEHAVIORS), constants.HV_CPU_MASK: hv_base.OPT_MULTI_CPU_MASK_CHECK, constants.HV_CPU_TYPE: hv_base.NO_CHECK, + constants.HV_CPU_CORES: hv_base.OPT_NONNEGATIVE_INT_CHECK, + constants.HV_CPU_THREADS: hv_base.OPT_NONNEGATIVE_INT_CHECK, + constants.HV_CPU_SOCKETS: hv_base.OPT_NONNEGATIVE_INT_CHECK, + constants.HV_SOUNDHW: hv_base.NO_CHECK, + constants.HV_USB_DEVICES: hv_base.NO_CHECK, + constants.HV_VGA: hv_base.NO_CHECK, + constants.HV_KVM_EXTRA: hv_base.NO_CHECK, + constants.HV_KVM_MACHINE_VERSION: hv_base.NO_CHECK, + constants.HV_VNET_HDR: hv_base.NO_CHECK, } + _VIRTIO = "virtio" + _VIRTIO_NET_PCI = "virtio-net-pci" + _VIRTIO_BLK_PCI = "virtio-blk-pci" + _MIGRATION_STATUS_RE = re.compile("Migration\s+status:\s+(\w+)", re.M | re.I) _MIGRATION_PROGRESS_RE = \ @@ -549,6 +709,33 @@ class KVMHypervisor(hv_base.BaseHypervisor): _CPU_INFO_CMD = "info cpus" _CONT_CMD = "cont" + _DEFAULT_MACHINE_VERSION_RE = re.compile(r"^(\S+).*\(default\)", re.M) + _CHECK_MACHINE_VERSION_RE = \ + staticmethod(lambda x: re.compile(r"^(%s)[ ]+.*PC" % x, re.M)) + + _QMP_RE = re.compile(r"^-qmp\s", re.M) + _SPICE_RE = re.compile(r"^-spice\s", re.M) + _VHOST_RE = re.compile(r"^-net\s.*,vhost=on|off", re.M) + _ENABLE_KVM_RE = re.compile(r"^-enable-kvm\s", re.M) + _DISABLE_KVM_RE = re.compile(r"^-disable-kvm\s", re.M) + _NETDEV_RE = re.compile(r"^-netdev\s", re.M) + _DISPLAY_RE = re.compile(r"^-display\s", re.M) + _MACHINE_RE = re.compile(r"^-machine\s", re.M) + _VIRTIO_NET_RE = re.compile(r"^name \"%s\"" % _VIRTIO_NET_PCI, re.M) + _VIRTIO_BLK_RE = re.compile(r"^name \"%s\"" % _VIRTIO_BLK_PCI, re.M) + # match -drive.*boot=on|off on different lines, but in between accept only + # dashes not preceeded by a new line (which would mean another option + # different than -drive is starting) + _BOOT_RE = re.compile(r"^-drive\s([^-]|(?= (0, 12): - nic_model = "virtio-net-pci" - vnet_hdr = True - else: - nic_model = "virtio" + nic_model = self._VIRTIO + try: + if self._VIRTIO_NET_RE.search(devlist): + nic_model = self._VIRTIO_NET_PCI + vnet_hdr = up_hvp[constants.HV_VNET_HDR] + except errors.HypervisorError, _: + # Older versions of kvm don't support DEVICE_LIST, but they don't + # have new virtio syntax either. + pass if up_hvp[constants.HV_VHOST_NET]: - # vhost_net is only available from version 0.13.0 or newer - if (v_major, v_min) >= (0, 13): + # check for vhost_net support + if self._VHOST_RE.search(kvmhelp): tap_extra = ",vhost=on" else: raise errors.HypervisorError("vhost_net is configured" @@ -1443,13 +1765,25 @@ class KVMHypervisor(hv_base.BaseHypervisor): else: nic_model = nic_type + kvm_supports_netdev = self._NETDEV_RE.search(kvmhelp) + for nic_seq, nic in enumerate(kvm_nics): - tapname, tapfd = _OpenTap(vnet_hdr) + tapname, tapfd = _OpenTap(vnet_hdr=vnet_hdr) tapfds.append(tapfd) taps.append(tapname) - if (v_major, v_min) >= (0, 12): - nic_val = "%s,mac=%s,netdev=netdev%s" % (nic_model, nic.mac, nic_seq) - tap_val = "type=tap,id=netdev%s,fd=%d%s" % (nic_seq, tapfd, tap_extra) + if kvm_supports_netdev: + nic_val = "%s,mac=%s" % (nic_model, nic.mac) + try: + # kvm_nics already exist in old runtime files and thus there might + # be some entries without pci slot (therefore try: except:) + kvm_devid = _GenerateDeviceKVMId(constants.HOTPLUG_TARGET_NIC, nic) + netdev = kvm_devid + nic_val += (",id=%s,bus=pci.0,addr=%s" % (kvm_devid, hex(nic.pci))) + except errors.HotplugError: + netdev = "netdev%d" % nic_seq + nic_val += (",netdev=%s" % netdev) + tap_val = ("type=tap,id=%s,fd=%d%s" % + (netdev, tapfd, tap_extra)) kvm_cmd.extend(["-netdev", tap_val, "-device", nic_val]) else: nic_val = "nic,vlan=%s,macaddr=%s,model=%s" % (nic_seq, @@ -1478,7 +1812,7 @@ class KVMHypervisor(hv_base.BaseHypervisor): constants.SECURE_DIR_MODE)]) # Automatically enable QMP if version is >= 0.14 - if (v_major, v_min) >= (0, 14): + if self._QMP_RE.search(kvmhelp): logging.debug("Enabling QMP") kvm_cmd.extend(["-qmp", "unix:%s,server,nowait" % self._InstanceQmpMonitor(instance.name)]) @@ -1491,6 +1825,11 @@ class KVMHypervisor(hv_base.BaseHypervisor): continue self._ConfigureNIC(instance, nic_seq, nic, taps[nic_seq]) + bdev_opts = self._GenerateKVMBlockDevicesOptions(instance, + block_devices, + kvmhelp, + devlist) + kvm_cmd.extend(bdev_opts) # CPU affinity requires kvm to start paused, so we set this flag if the # instance is not already paused and if we are not going to accept a # migrating instance. In the latter case, pausing is not needed. @@ -1577,29 +1916,195 @@ class KVMHypervisor(hv_base.BaseHypervisor): """ self._CheckDown(instance.name) + kvmpath = instance.hvparams[constants.HV_KVM_PATH] + kvmhelp = self._GetKVMOutput(kvmpath, self._KVMOPT_HELP) kvm_runtime = self._GenerateKVMRuntime(instance, block_devices, - startup_paused) + startup_paused, kvmhelp) self._SaveKVMRuntime(instance, kvm_runtime) - self._ExecuteKVMRuntime(instance, kvm_runtime) + self._ExecuteKVMRuntime(instance, kvm_runtime, kvmhelp) - def _CallMonitorCommand(self, instance_name, command): + def _CallMonitorCommand(self, instance_name, command, timeout=None): """Invoke a command on the instance monitor. """ - socat = ("echo %s | %s STDIO UNIX-CONNECT:%s" % + if timeout is not None: + timeout_cmd = "timeout %s" % (timeout, ) + else: + timeout_cmd = "" + + # TODO: Replace monitor calls with QMP once KVM >= 0.14 is the minimum + # version. The monitor protocol is designed for human consumption, whereas + # QMP is made for programmatic usage. In the worst case QMP can also + # execute monitor commands. As it is, all calls to socat take at least + # 500ms and likely more: socat can't detect the end of the reply and waits + # for 500ms of no data received before exiting (500 ms is the default for + # the "-t" parameter). + socat = ("echo %s | %s %s STDIO UNIX-CONNECT:%s" % (utils.ShellQuote(command), + timeout_cmd, constants.SOCAT_PATH, utils.ShellQuote(self._InstanceMonitor(instance_name)))) + result = utils.RunCmd(socat) if result.failed: - msg = ("Failed to send command '%s' to instance %s." - " output: %s, error: %s, fail_reason: %s" % - (command, instance_name, - result.stdout, result.stderr, result.fail_reason)) + msg = ("Failed to send command '%s' to instance '%s', reason '%s'," + " output: %s" % + (command, instance_name, result.fail_reason, result.output)) raise errors.HypervisorError(msg) return result + def _GetFreePCISlot(self, instance, dev): + """Get the first available pci slot of a runnung instance. + + """ + slots = bitarray(32) + slots.setall(False) # pylint: disable=E1101 + output = self._CallMonitorCommand(instance.name, self._INFO_PCI_CMD) + for line in output.stdout.splitlines(): + match = self._INFO_PCI_RE.search(line) + if match: + slot = int(match.group(1)) + slots[slot] = True + + [free] = slots.search(_AVAILABLE_PCI_SLOT, 1) # pylint: disable=E1101 + if not free: + raise errors.HypervisorError("All PCI slots occupied") + + dev.pci = int(free) + + def HotplugSupported(self, instance, action, dev_type): + """Check if hotplug is supported. + + Hotplug is *not* supported in case of: + - qemu versions < 1.0 + - security models and chroot (disk hotplug) + - fdsend module is missing (nic hot-add) + + @raise errors.HypervisorError: in previous cases + + """ + output = self._CallMonitorCommand(instance.name, self._INFO_VERSION_CMD) + # TODO: search for netdev_add, drive_add, device_add..... + match = self._INFO_VERSION_RE.search(output.stdout) + if not match: + raise errors.HotplugError("Try hotplug only in running instances.") + v_major, v_min, _, _ = match.groups() + if (int(v_major), int(v_min)) < (1, 0): + raise errors.HotplugError("Hotplug not supported for qemu versions < 1.0") + + if dev_type == constants.HOTPLUG_TARGET_DISK: + hvp = instance.hvparams + security_model = hvp[constants.HV_SECURITY_MODEL] + use_chroot = hvp[constants.HV_KVM_USE_CHROOT] + if use_chroot: + raise errors.HotplugError("Disk hotplug is not supported" + " in case of chroot.") + if security_model != constants.HT_SM_NONE: + raise errors.HotplugError("Disk Hotplug is not supported in case" + " security models are used.") + + if (dev_type == constants.HOTPLUG_TARGET_NIC and + action == constants.HOTPLUG_ACTION_ADD and not fdsend): + raise errors.HotplugError("Cannot hot-add NIC." + " fdsend python module is missing.") + return True + + def _CallHotplugCommand(self, name, cmd): + output = self._CallMonitorCommand(name, cmd) + # TODO: parse output and check if succeeded + for line in output.stdout.splitlines(): + logging.info("%s", line) + + def HotAddDevice(self, instance, dev_type, device, extra, seq): + """ Helper method to hot-add a new device + + It gets free pci slot generates the device name and invokes the + device specific method. + + """ + # in case of hot-mod this is given + if device.pci is None: + self._GetFreePCISlot(instance, device) + kvm_devid = _GenerateDeviceKVMId(dev_type, device) + runtime = self._LoadKVMRuntime(instance) + if dev_type == constants.HOTPLUG_TARGET_DISK: + command = "drive_add dummy file=%s,if=none,id=%s,format=raw\n" % \ + (extra, kvm_devid) + command += ("device_add virtio-blk-pci,bus=pci.0,addr=%s,drive=%s,id=%s" % + (hex(device.pci), kvm_devid, kvm_devid)) + elif dev_type == constants.HOTPLUG_TARGET_NIC: + (tap, fd) = _OpenTap() + self._ConfigureNIC(instance, seq, device, tap) + self._PassTapFd(instance, fd, device) + command = "netdev_add tap,id=%s,fd=%s\n" % (kvm_devid, kvm_devid) + args = "virtio-net-pci,bus=pci.0,addr=%s,mac=%s,netdev=%s,id=%s" % \ + (hex(device.pci), device.mac, kvm_devid, kvm_devid) + command += "device_add %s" % args + utils.WriteFile(self._InstanceNICFile(instance.name, seq), data=tap) + + self._CallHotplugCommand(instance.name, command) + # update relevant entries in runtime file + index = _DEVICE_RUNTIME_INDEX[dev_type] + entry = _RUNTIME_ENTRY[dev_type](device, extra) + runtime[index].append(entry) + self._SaveKVMRuntime(instance, runtime) + + def HotDelDevice(self, instance, dev_type, device, _, seq): + """ Helper method for hot-del device + + It gets device info from runtime file, generates the device name and + invokes the device specific method. + + """ + runtime = self._LoadKVMRuntime(instance) + entry = _GetExistingDeviceInfo(dev_type, device, runtime) + kvm_device = _RUNTIME_DEVICE[dev_type](entry) + kvm_devid = _GenerateDeviceKVMId(dev_type, kvm_device) + if dev_type == constants.HOTPLUG_TARGET_DISK: + command = "device_del %s" % kvm_devid + elif dev_type == constants.HOTPLUG_TARGET_NIC: + command = "device_del %s\n" % kvm_devid + command += "netdev_del %s" % kvm_devid + utils.RemoveFile(self._InstanceNICFile(instance.name, seq)) + self._CallHotplugCommand(instance.name, command) + index = _DEVICE_RUNTIME_INDEX[dev_type] + runtime[index].remove(entry) + self._SaveKVMRuntime(instance, runtime) + + return kvm_device.pci + + def HotModDevice(self, instance, dev_type, device, _, seq): + """ Helper method for hot-mod device + + It gets device info from runtime file, generates the device name and + invokes the device specific method. Currently only NICs support hot-mod + + """ + if dev_type == constants.HOTPLUG_TARGET_NIC: + # putting it back in the same pci slot + device.pci = self.HotDelDevice(instance, dev_type, device, _, seq) + # TODO: remove sleep when socat gets removed + time.sleep(2) + self.HotAddDevice(instance, dev_type, device, _, seq) + + def _PassTapFd(self, instance, fd, nic): + """Pass file descriptor to kvm process via monitor socket using SCM_RIGHTS + + """ + # TODO: factor out code related to unix sockets. + # squash common parts between monitor and qmp + kvm_devid = _GenerateDeviceKVMId(constants.HOTPLUG_TARGET_NIC, nic) + command = "getfd %s\n" % kvm_devid + fds = [fd] + logging.info("%s", fds) + try: + monsock = MonitorSocket(self._InstanceMonitor(instance.name)) + monsock.connect() + fdsend.sendfds(monsock.sock, command, fds=fds) + finally: + monsock.close() + @classmethod def _ParseKVMVersion(cls, text): """Parse the KVM version from the --help output. @@ -1624,22 +2129,56 @@ class KVMHypervisor(hv_base.BaseHypervisor): return (v_all, v_maj, v_min, v_rev) @classmethod - def _GetKVMVersion(cls): + def _GetKVMOutput(cls, kvm_path, option): + """Return the output of a kvm invocation + + @type kvm_path: string + @param kvm_path: path to the kvm executable + @type option: a key of _KVMOPTS_CMDS + @param option: kvm option to fetch the output from + @return: output a supported kvm invocation + @raise errors.HypervisorError: when the KVM help output cannot be retrieved + + """ + assert option in cls._KVMOPTS_CMDS, "Invalid output option" + + optlist, can_fail = cls._KVMOPTS_CMDS[option] + + result = utils.RunCmd([kvm_path] + optlist) + if result.failed and not can_fail: + raise errors.HypervisorError("Unable to get KVM %s output" % + " ".join(optlist)) + return result.output + + @classmethod + def _GetKVMVersion(cls, kvm_path): """Return the installed KVM version. @return: (version, v_maj, v_min, v_rev) @raise errors.HypervisorError: when the KVM version cannot be retrieved """ - result = utils.RunCmd([constants.KVM_PATH, "--help"]) - if result.failed: - raise errors.HypervisorError("Unable to get KVM version") - return cls._ParseKVMVersion(result.output) + return cls._ParseKVMVersion(cls._GetKVMOutput(kvm_path, cls._KVMOPT_HELP)) + + @classmethod + def _GetDefaultMachineVersion(cls, kvm_path): + """Return the default hardware revision (e.g. pc-1.1) + + """ + output = cls._GetKVMOutput(kvm_path, cls._KVMOPT_MLIST) + match = cls._DEFAULT_MACHINE_VERSION_RE.search(output) + if match: + return match.group(1) + else: + return "pc" - def StopInstance(self, instance, force=False, retry=False, name=None): + def StopInstance(self, instance, force=False, retry=False, name=None, + timeout=None): """Stop an instance. """ + assert(timeout is None or force is not None) + if name is not None and not force: raise errors.HypervisorError("Cannot shutdown cleanly by name only") if name is None: @@ -1652,7 +2191,7 @@ class KVMHypervisor(hv_base.BaseHypervisor): if force or not acpi: utils.KillProcess(pid) else: - self._CallMonitorCommand(name, "system_powerdown") + self._CallMonitorCommand(name, "system_powerdown", timeout) def CleanupInstance(self, instance_name): """Cleanup after a stopped instance @@ -1682,7 +2221,9 @@ class KVMHypervisor(hv_base.BaseHypervisor): self.StopInstance(instance, force=True) # ...and finally we can save it again, and execute it... self._SaveKVMRuntime(instance, kvm_runtime) - self._ExecuteKVMRuntime(instance, kvm_runtime) + kvmpath = instance.hvparams[constants.HV_KVM_PATH] + kvmhelp = self._GetKVMOutput(kvmpath, self._KVMOPT_HELP) + self._ExecuteKVMRuntime(instance, kvm_runtime, kvmhelp) def MigrationInfo(self, instance): """Get instance information to perform a migration. @@ -1708,7 +2249,10 @@ class KVMHypervisor(hv_base.BaseHypervisor): """ kvm_runtime = self._LoadKVMRuntime(instance, serialized_runtime=info) incoming_address = (target, instance.hvparams[constants.HV_MIGRATION_PORT]) - self._ExecuteKVMRuntime(instance, kvm_runtime, incoming=incoming_address) + kvmpath = instance.hvparams[constants.HV_KVM_PATH] + kvmhelp = self._GetKVMOutput(kvmpath, self._KVMOPT_HELP) + self._ExecuteKVMRuntime(instance, kvm_runtime, kvmhelp, + incoming=incoming_address) def FinalizeMigrationDst(self, instance, info, success): """Finalize the instance migration on the target node. @@ -1855,7 +2399,10 @@ class KVMHypervisor(hv_base.BaseHypervisor): """ result = self.GetLinuxNodeInfo() - _, v_major, v_min, v_rev = self._GetKVMVersion() + # FIXME: this is the global kvm version, but the actual version can be + # customized as an hv parameter. we should use the nodegroup's default kvm + # path parameter here. + _, v_major, v_min, v_rev = self._GetKVMVersion(constants.KVM_PATH) result[constants.HV_NODEINFO_KEY_VERSION] = (v_major, v_min, v_rev) return result @@ -1900,13 +2447,22 @@ class KVMHypervisor(hv_base.BaseHypervisor): def Verify(self): """Verify the hypervisor. - Check that the binary exists. + Check that the required binaries exist. + + @return: Problem description if something is wrong, C{None} otherwise """ + msgs = [] + # FIXME: this is the global kvm binary, but the actual path can be + # customized as an hv parameter; we should use the nodegroup's + # default kvm path parameter here. if not os.path.exists(constants.KVM_PATH): - return "The kvm binary ('%s') does not exist." % constants.KVM_PATH + msgs.append("The KVM binary ('%s') does not exist" % constants.KVM_PATH) if not os.path.exists(constants.SOCAT_PATH): - return "The socat binary ('%s') does not exist." % constants.SOCAT_PATH + msgs.append("The socat binary ('%s') does not exist" % + constants.SOCAT_PATH) + + return self._FormatVerifyResults(msgs) @classmethod def CheckParameterSyntax(cls, hvparams): @@ -1931,6 +2487,14 @@ class KVMHypervisor(hv_base.BaseHypervisor): (constants.HV_VNC_X509, constants.HV_VNC_X509_VERIFY)) + if hvparams[constants.HV_SERIAL_CONSOLE]: + serial_speed = hvparams[constants.HV_SERIAL_SPEED] + valid_speeds = constants.VALID_SERIAL_SPEEDS + if not serial_speed or serial_speed not in valid_speeds: + raise errors.HypervisorError("Invalid serial console speed, must be" + " one of: %s" % + utils.CommaJoin(valid_speeds)) + boot_order = hvparams[constants.HV_BOOT_ORDER] if (boot_order == constants.HT_BO_CDROM and not hvparams[constants.HV_CDROM_IMAGE_PATH]): @@ -1956,13 +2520,13 @@ class KVMHypervisor(hv_base.BaseHypervisor): # IP of that family if (netutils.IP4Address.IsValid(spice_bind) and spice_ip_version != constants.IP4_VERSION): - raise errors.HypervisorError("spice: got an IPv4 address (%s), but" + raise errors.HypervisorError("SPICE: Got an IPv4 address (%s), but" " the specified IP version is %s" % (spice_bind, spice_ip_version)) if (netutils.IP6Address.IsValid(spice_bind) and spice_ip_version != constants.IP6_VERSION): - raise errors.HypervisorError("spice: got an IPv6 address (%s), but" + raise errors.HypervisorError("SPICE: Got an IPv6 address (%s), but" " the specified IP version is %s" % (spice_bind, spice_ip_version)) else: @@ -1970,7 +2534,7 @@ class KVMHypervisor(hv_base.BaseHypervisor): # error if any of them is set without it. for param in _SPICE_ADDITIONAL_PARAMS: if hvparams[param]: - raise errors.HypervisorError("spice: %s requires %s to be set" % + raise errors.HypervisorError("SPICE: %s requires %s to be set" % (param, constants.HV_KVM_SPICE_BIND)) @classmethod @@ -1984,6 +2548,8 @@ class KVMHypervisor(hv_base.BaseHypervisor): """ super(KVMHypervisor, cls).ValidateParameters(hvparams) + kvm_path = hvparams[constants.HV_KVM_PATH] + security_model = hvparams[constants.HV_SECURITY_MODEL] if security_model == constants.HT_SM_USER: username = hvparams[constants.HV_SECURITY_DOMAIN] @@ -1997,24 +2563,31 @@ class KVMHypervisor(hv_base.BaseHypervisor): if spice_bind: # only one of VNC and SPICE can be used currently. if hvparams[constants.HV_VNC_BIND_ADDRESS]: - raise errors.HypervisorError("both SPICE and VNC are configured, but" + raise errors.HypervisorError("Both SPICE and VNC are configured, but" " only one of them can be used at a" - " given time.") + " given time") - # KVM version should be >= 0.14.0 - _, v_major, v_min, _ = cls._GetKVMVersion() - if (v_major, v_min) < (0, 14): - raise errors.HypervisorError("spice is configured, but it is not" - " available in versions of KVM < 0.14") + # check that KVM supports SPICE + kvmhelp = cls._GetKVMOutput(kvm_path, cls._KVMOPT_HELP) + if not cls._SPICE_RE.search(kvmhelp): + raise errors.HypervisorError("SPICE is configured, but it is not" + " supported according to 'kvm --help'") # if spice_bind is not an IP address, it must be a valid interface - bound_to_addr = (netutils.IP4Address.IsValid(spice_bind) - or netutils.IP6Address.IsValid(spice_bind)) + bound_to_addr = (netutils.IP4Address.IsValid(spice_bind) or + netutils.IP6Address.IsValid(spice_bind)) if not bound_to_addr and not netutils.IsValidInterface(spice_bind): - raise errors.HypervisorError("spice: the %s parameter must be either" + raise errors.HypervisorError("SPICE: The %s parameter must be either" " a valid IP address or interface name" % constants.HV_KVM_SPICE_BIND) + machine_version = hvparams[constants.HV_KVM_MACHINE_VERSION] + if machine_version: + output = cls._GetKVMOutput(kvm_path, cls._KVMOPT_MLIST) + if not cls._CHECK_MACHINE_VERSION_RE(machine_version).search(output): + raise errors.HypervisorError("Unsupported machine version: %s" % + machine_version) + @classmethod def PowercycleNode(cls): """KVM powercycle, just a wrapper over Linux powercycle.