diff options
Diffstat (limited to 'cloudinit/sources/DataSourceAzure.py')
| -rwxr-xr-x[-rw-r--r--] | cloudinit/sources/DataSourceAzure.py | 474 |
1 files changed, 340 insertions, 134 deletions
diff --git a/cloudinit/sources/DataSourceAzure.py b/cloudinit/sources/DataSourceAzure.py index a06e6e1f..61ec522a 100644..100755 --- a/cloudinit/sources/DataSourceAzure.py +++ b/cloudinit/sources/DataSourceAzure.py @@ -13,7 +13,6 @@ import os import os.path import re from time import time -from subprocess import call from xml.dom import minidom import xml.etree.ElementTree as ET @@ -22,10 +21,20 @@ from cloudinit import net from cloudinit.event import EventType from cloudinit.net.dhcp import EphemeralDHCPv4 from cloudinit import sources -from cloudinit.sources.helpers.azure import get_metadata_from_fabric from cloudinit.sources.helpers import netlink from cloudinit.url_helper import UrlError, readurl, retry_on_url_exc from cloudinit import util +from cloudinit.reporting import events + +from cloudinit.sources.helpers.azure import ( + azure_ds_reporter, + azure_ds_telemetry_reporter, + get_metadata_from_fabric, + get_boot_telemetry, + get_system_info, + report_diagnostic_event, + EphemeralDHCPv4WithReporting, + is_byte_swapped) LOG = logging.getLogger(__name__) @@ -54,8 +63,14 @@ AZURE_CHASSIS_ASSET_TAG = '7783-7084-3265-9085-8269-3286-77' REPROVISION_MARKER_FILE = "/var/lib/cloud/data/poll_imds" REPORTED_READY_MARKER_FILE = "/var/lib/cloud/data/reported_ready" AGENT_SEED_DIR = '/var/lib/waagent' + +# In the event where the IMDS primary server is not +# available, it takes 1s to fallback to the secondary one +IMDS_TIMEOUT_IN_SECONDS = 2 IMDS_URL = "http://169.254.169.254/metadata/" +PLATFORM_ENTROPY_SOURCE = "/sys/firmware/acpi/tables/OEM0" + # List of static scripts and network config artifacts created by # stock ubuntu suported images. UBUNTU_EXTENDED_NETWORK_SCRIPTS = [ @@ -196,6 +211,8 @@ if util.is_FreeBSD(): RESOURCE_DISK_PATH = "/dev/" + res_disk else: LOG.debug("resource disk is None") + # TODO Find where platform entropy data is surfaced + PLATFORM_ENTROPY_SOURCE = None BUILTIN_DS_CONFIG = { 'agent_command': AGENT_START_BUILTIN, @@ -242,6 +259,7 @@ def set_hostname(hostname, hostname_command='hostname'): util.subp([hostname_command, hostname]) +@azure_ds_telemetry_reporter @contextlib.contextmanager def temporary_hostname(temp_hostname, cfg, hostname_command='hostname'): """ @@ -269,11 +287,6 @@ class DataSourceAzure(sources.DataSource): dsname = 'Azure' _negotiated = False _metadata_imds = sources.UNSET - process_name = 'dhclient' - - tmpps = os.popen("ps -Af").read() - if process_name not in tmpps[:]: - call(['/sbin/dhclient', DEFAULT_PRIMARY_NIC]) def __init__(self, sys_cfg, distro, paths): sources.DataSource.__init__(self, sys_cfg, distro, paths) @@ -293,6 +306,7 @@ class DataSourceAzure(sources.DataSource): root = sources.DataSource.__str__(self) return "%s [seed=%s]" % (root, self.seed) + @azure_ds_telemetry_reporter def bounce_network_with_azure_hostname(self): # When using cloud-init to provision, we have to set the hostname from # the metadata and "bounce" the network to force DDNS to update via @@ -318,6 +332,7 @@ class DataSourceAzure(sources.DataSource): util.logexc(LOG, "handling set_hostname failed") return False + @azure_ds_telemetry_reporter def get_metadata_from_agent(self): temp_hostname = self.metadata.get('local-hostname') agent_cmd = self.ds_cfg['agent_command'] @@ -340,22 +355,25 @@ class DataSourceAzure(sources.DataSource): for pk in self.cfg.get('_pubkeys', []): if pk.get('value', None): key_value = pk['value'] - LOG.debug("ssh authentication: using value from fabric") + LOG.debug("SSH authentication: using value from fabric") else: bname = str(pk['fingerprint'] + ".crt") fp_files += [os.path.join(ddir, bname)] - LOG.debug("ssh authentication: " - "using fingerprint from fabirc") - - # wait very long for public SSH keys to arrive - # https://bugs.launchpad.net/cloud-init/+bug/1717611 - missing = util.log_time(logfunc=LOG.debug, - msg="waiting for SSH public key files", - func=util.wait_for_files, - args=(fp_files, 900)) - - if len(missing): - LOG.warning("Did not find files, but going on: %s", missing) + LOG.debug("SSH authentication: " + "using fingerprint from fabric") + + with events.ReportEventStack( + name="waiting-for-ssh-public-key", + description="wait for agents to retrieve SSH keys", + parent=azure_ds_reporter): + # wait very long for public SSH keys to arrive + # https://bugs.launchpad.net/cloud-init/+bug/1717611 + missing = util.log_time(logfunc=LOG.debug, + msg="waiting for SSH public key files", + func=util.wait_for_files, + args=(fp_files, 900)) + if len(missing): + LOG.warning("Did not find files, but going on: %s", missing) metadata = {} metadata['public-keys'] = key_value or pubkeys_from_crt_files(fp_files) @@ -369,6 +387,7 @@ class DataSourceAzure(sources.DataSource): subplatform_type = 'seed-dir' return '%s (%s)' % (subplatform_type, self.seed) + @azure_ds_telemetry_reporter def crawl_metadata(self): """Walk all instance metadata sources returning a dict on success. @@ -399,19 +418,24 @@ class DataSourceAzure(sources.DataSource): elif cdev.startswith("/dev/"): if util.is_FreeBSD(): ret = util.mount_cb(cdev, load_azure_ds_dir, - mtype="udf", sync=False) + mtype="udf") else: ret = util.mount_cb(cdev, load_azure_ds_dir) else: ret = load_azure_ds_dir(cdev) except NonAzureDataSource: + report_diagnostic_event( + "Did not find Azure data source in %s" % cdev) continue except BrokenAzureDataSource as exc: msg = 'BrokenAzureDataSource: %s' % exc + report_diagnostic_event(msg) raise sources.InvalidMetaDataException(msg) except util.MountFailedError: - LOG.warning("%s was not mountable", cdev) + msg = '%s was not mountable' % cdev + report_diagnostic_event(msg) + LOG.warning(msg) continue perform_reprovision = reprovision or self._should_reprovision(ret) @@ -419,10 +443,11 @@ class DataSourceAzure(sources.DataSource): if util.is_FreeBSD(): msg = "Free BSD is not supported for PPS VMs" LOG.error(msg) + report_diagnostic_event(msg) raise sources.InvalidMetaDataException(msg) ret = self._reprovision() imds_md = get_metadata_from_imds( - self.fallback_interface, retries=3) + self.fallback_interface, retries=10) (md, userdata_raw, cfg, files) = ret self.seed = cdev crawled_data.update({ @@ -437,7 +462,9 @@ class DataSourceAzure(sources.DataSource): break if not found: - raise sources.InvalidMetaDataException('No Azure metadata found') + msg = 'No Azure metadata found' + report_diagnostic_event(msg) + raise sources.InvalidMetaDataException(msg) if found == ddir: LOG.debug("using files cached in %s", ddir) @@ -445,8 +472,7 @@ class DataSourceAzure(sources.DataSource): seed = _get_random_seed() if seed: crawled_data['metadata']['random_seed'] = seed - crawled_data['metadata']['instance-id'] = util.read_dmi_data( - 'system-uuid') + crawled_data['metadata']['instance-id'] = self._iid() if perform_reprovision: LOG.info("Reporting ready to Azure after getting ReprovisionData") @@ -456,9 +482,14 @@ class DataSourceAzure(sources.DataSource): self._report_ready(lease=self._ephemeral_dhcp_ctx.lease) self._ephemeral_dhcp_ctx.clean_network() # Teardown ephemeral else: - with EphemeralDHCPv4() as lease: - self._report_ready(lease=lease) - + try: + with EphemeralDHCPv4WithReporting( + azure_ds_reporter) as lease: + self._report_ready(lease=lease) + except Exception as e: + report_diagnostic_event( + "exception while reporting ready: %s" % e) + raise return crawled_data def _is_platform_viable(self): @@ -470,6 +501,7 @@ class DataSourceAzure(sources.DataSource): super(DataSourceAzure, self).clear_cached_attrs(attr_defaults) self._metadata_imds = sources.UNSET + @azure_ds_telemetry_reporter def _get_data(self): """Crawl and process datasource metadata caching metadata as attrs. @@ -479,6 +511,16 @@ class DataSourceAzure(sources.DataSource): if not self._is_platform_viable(): return False try: + get_boot_telemetry() + except Exception as e: + LOG.warning("Failed to get boot telemetry: %s", e) + + try: + get_system_info() + except Exception as e: + LOG.warning("Failed to get system information: %s", e) + + try: crawled_data = util.log_time( logfunc=LOG.debug, msg='Crawl of metadata service', func=self.crawl_metadata) @@ -516,6 +558,17 @@ class DataSourceAzure(sources.DataSource): # quickly (local check only) if self.instance_id is still valid return sources.instance_id_matches_system_uuid(self.get_instance_id()) + def _iid(self, previous=None): + prev_iid_path = os.path.join( + self.paths.get_cpath('data'), 'instance-id') + iid = util.read_dmi_data('system-uuid') + if os.path.exists(prev_iid_path): + previous = util.load_file(prev_iid_path).strip() + if is_byte_swapped(previous, iid): + return previous + return iid + + @azure_ds_telemetry_reporter def setup(self, is_new_instance): if self._negotiated is False: LOG.debug("negotiating for %s (new_instance=%s)", @@ -536,27 +589,55 @@ class DataSourceAzure(sources.DataSource): headers = {"Metadata": "true"} nl_sock = None report_ready = bool(not os.path.isfile(REPORTED_READY_MARKER_FILE)) + self.imds_logging_threshold = 1 + self.imds_poll_counter = 1 + dhcp_attempts = 0 + vnet_switched = False + return_val = None def exc_cb(msg, exception): if isinstance(exception, UrlError) and exception.code == 404: + if self.imds_poll_counter == self.imds_logging_threshold: + # Reducing the logging frequency as we are polling IMDS + self.imds_logging_threshold *= 2 + LOG.debug("Call to IMDS with arguments %s failed " + "with status code %s after %s retries", + msg, exception.code, self.imds_poll_counter) + LOG.debug("Backing off logging threshold for the same " + "exception to %d", self.imds_logging_threshold) + self.imds_poll_counter += 1 return True + # If we get an exception while trying to call IMDS, we # call DHCP and setup the ephemeral network to acquire the new IP. + LOG.debug("Call to IMDS with arguments %s failed with " + "status code %s", msg, exception.code) + report_diagnostic_event("polling IMDS failed with exception %s" + % exception.code) return False LOG.debug("Wait for vnetswitch to happen") while True: try: - # Save our EphemeralDHCPv4 context so we avoid repeated dhcp - self._ephemeral_dhcp_ctx = EphemeralDHCPv4() - lease = self._ephemeral_dhcp_ctx.obtain_lease() + # Save our EphemeralDHCPv4 context to avoid repeated dhcp + with events.ReportEventStack( + name="obtain-dhcp-lease", + description="obtain dhcp lease", + parent=azure_ds_reporter): + self._ephemeral_dhcp_ctx = EphemeralDHCPv4() + lease = self._ephemeral_dhcp_ctx.obtain_lease() + + if vnet_switched: + dhcp_attempts += 1 if report_ready: try: nl_sock = netlink.create_bound_netlink_socket() except netlink.NetlinkCreateSocketError as e: + report_diagnostic_event(e) LOG.warning(e) self._ephemeral_dhcp_ctx.clean_network() - return + break + path = REPORTED_READY_MARKER_FILE LOG.info( "Creating a marker file to report ready: %s", path) @@ -564,17 +645,33 @@ class DataSourceAzure(sources.DataSource): pid=os.getpid(), time=time())) self._report_ready(lease=lease) report_ready = False - try: - netlink.wait_for_media_disconnect_connect( - nl_sock, lease['interface']) - except AssertionError as error: - LOG.error(error) - return + + with events.ReportEventStack( + name="wait-for-media-disconnect-connect", + description="wait for vnet switch", + parent=azure_ds_reporter): + try: + netlink.wait_for_media_disconnect_connect( + nl_sock, lease['interface']) + except AssertionError as error: + report_diagnostic_event(error) + LOG.error(error) + break + + vnet_switched = True self._ephemeral_dhcp_ctx.clean_network() else: - return readurl(url, timeout=1, headers=headers, - exception_cb=exc_cb, infinite=True, - log_req_resp=False).contents + with events.ReportEventStack( + name="get-reprovision-data-from-imds", + description="get reprovision data from imds", + parent=azure_ds_reporter): + return_val = readurl(url, + timeout=IMDS_TIMEOUT_IN_SECONDS, + headers=headers, + exception_cb=exc_cb, + infinite=True, + log_req_resp=False).contents + break except UrlError: # Teardown our EphemeralDHCPv4 context on failure as we retry self._ephemeral_dhcp_ctx.clean_network() @@ -583,6 +680,15 @@ class DataSourceAzure(sources.DataSource): if nl_sock: nl_sock.close() + if vnet_switched: + report_diagnostic_event("attempted dhcp %d times after reuse" % + dhcp_attempts) + report_diagnostic_event("polled imds %d times after reuse" % + self.imds_poll_counter) + + return return_val + + @azure_ds_telemetry_reporter def _report_ready(self, lease): """Tells the fabric provisioning has completed """ try: @@ -620,9 +726,14 @@ class DataSourceAzure(sources.DataSource): def _reprovision(self): """Initiate the reprovisioning workflow.""" contents = self._poll_imds() - md, ud, cfg = read_azure_ovf(contents) - return (md, ud, cfg, {'ovf-env.xml': contents}) - + with events.ReportEventStack( + name="reprovisioning-read-azure-ovf", + description="read azure ovf during reprovisioning", + parent=azure_ds_reporter): + md, ud, cfg = read_azure_ovf(contents) + return (md, ud, cfg, {'ovf-env.xml': contents}) + + @azure_ds_telemetry_reporter def _negotiate(self): """Negotiate with fabric and return data from it. @@ -633,9 +744,11 @@ class DataSourceAzure(sources.DataSource): if self.ds_cfg['agent_command'] == AGENT_START_BUILTIN: self.bounce_network_with_azure_hostname() + pubkey_info = self.cfg.get('_pubkeys', None) metadata_func = partial(get_metadata_from_fabric, fallback_lease_file=self. - dhclient_lease_file) + dhclient_lease_file, + pubkey_info=pubkey_info) else: metadata_func = self.get_metadata_from_agent @@ -643,15 +756,20 @@ class DataSourceAzure(sources.DataSource): self.ds_cfg['agent_command']) try: fabric_data = metadata_func() - except Exception: + except Exception as e: + report_diagnostic_event( + "Error communicating with Azure fabric; You may experience " + "connectivity issues: %s" % e) LOG.warning( - "Error communicating with Azure fabric; You may experience." + "Error communicating with Azure fabric; You may experience " "connectivity issues.", exc_info=True) return False + util.del_file(REPORTED_READY_MARKER_FILE) util.del_file(REPROVISION_MARKER_FILE) return fabric_data + @azure_ds_telemetry_reporter def activate(self, cfg, is_new_instance): address_ephemeral_resize(is_new_instance=is_new_instance, preserve_ntfs=self.ds_cfg.get( @@ -659,6 +777,11 @@ class DataSourceAzure(sources.DataSource): return @property + def availability_zone(self): + return self.metadata.get( + 'imds', {}).get('compute', {}).get('platformFaultDomain') + + @property def network_config(self): """Generate a network config like net.generate_fallback_network() with the following exceptions. @@ -668,7 +791,7 @@ class DataSourceAzure(sources.DataSource): 2. Generate a fallback network config that does not include any of the blacklisted devices. """ - if not self._network_config: + if not self._network_config or self._network_config == sources.UNSET: if self.ds_cfg.get('apply_network_config'): nc_src = self._metadata_imds else: @@ -676,6 +799,10 @@ class DataSourceAzure(sources.DataSource): self._network_config = parse_network_config(nc_src) return self._network_config + @property + def region(self): + return self.metadata.get('imds', {}).get('compute', {}).get('location') + def _partitions_on_device(devpath, maxnum=16): # return a list of tuples (ptnum, path) for each part on devpath @@ -690,12 +817,14 @@ def _partitions_on_device(devpath, maxnum=16): return [] +@azure_ds_telemetry_reporter def _has_ntfs_filesystem(devpath): ntfs_devices = util.find_devs_with("TYPE=ntfs", no_cache=True) LOG.debug('ntfs_devices found = %s', ntfs_devices) return os.path.realpath(devpath) in ntfs_devices +@azure_ds_telemetry_reporter def can_dev_be_reformatted(devpath, preserve_ntfs): """Determine if the ephemeral drive at devpath should be reformatted. @@ -744,43 +873,59 @@ def can_dev_be_reformatted(devpath, preserve_ntfs): (cand_part, cand_path, devpath)) return False, msg + @azure_ds_telemetry_reporter def count_files(mp): ignored = set(['dataloss_warning_readme.txt']) return len([f for f in os.listdir(mp) if f.lower() not in ignored]) bmsg = ('partition %s (%s) on device %s was ntfs formatted' % (cand_part, cand_path, devpath)) - try: - file_count = util.mount_cb(cand_path, count_files, mtype="ntfs", - update_env_for_mount={'LANG': 'C'}) - except util.MountFailedError as e: - if "unknown filesystem type 'ntfs'" in str(e): - return True, (bmsg + ' but this system cannot mount NTFS,' - ' assuming there are no important files.' - ' Formatting allowed.') - return False, bmsg + ' but mount of %s failed: %s' % (cand_part, e) - - if file_count != 0: - LOG.warning("it looks like you're using NTFS on the ephemeral disk, " - 'to ensure that filesystem does not get wiped, set ' - '%s.%s in config', '.'.join(DS_CFG_PATH), - DS_CFG_KEY_PRESERVE_NTFS) - return False, bmsg + ' but had %d files on it.' % file_count + + with events.ReportEventStack( + name="mount-ntfs-and-count", + description="mount-ntfs-and-count", + parent=azure_ds_reporter) as evt: + try: + file_count = util.mount_cb(cand_path, count_files, mtype="ntfs", + update_env_for_mount={'LANG': 'C'}) + except util.MountFailedError as e: + evt.description = "cannot mount ntfs" + if "unknown filesystem type 'ntfs'" in str(e): + return True, (bmsg + ' but this system cannot mount NTFS,' + ' assuming there are no important files.' + ' Formatting allowed.') + return False, bmsg + ' but mount of %s failed: %s' % (cand_part, e) + + if file_count != 0: + evt.description = "mounted and counted %d files" % file_count + LOG.warning("it looks like you're using NTFS on the ephemeral" + " disk, to ensure that filesystem does not get wiped," + " set %s.%s in config", '.'.join(DS_CFG_PATH), + DS_CFG_KEY_PRESERVE_NTFS) + return False, bmsg + ' but had %d files on it.' % file_count return True, bmsg + ' and had no important files. Safe for reformatting.' +@azure_ds_telemetry_reporter def address_ephemeral_resize(devpath=RESOURCE_DISK_PATH, maxwait=120, is_new_instance=False, preserve_ntfs=False): # wait for ephemeral disk to come up naplen = .2 - missing = util.wait_for_files([devpath], maxwait=maxwait, naplen=naplen, - log_pre="Azure ephemeral disk: ") - - if missing: - LOG.warning("ephemeral device '%s' did not appear after %d seconds.", - devpath, maxwait) - return + with events.ReportEventStack( + name="wait-for-ephemeral-disk", + description="wait for ephemeral disk", + parent=azure_ds_reporter): + missing = util.wait_for_files([devpath], + maxwait=maxwait, + naplen=naplen, + log_pre="Azure ephemeral disk: ") + + if missing: + LOG.warning("ephemeral device '%s' did" + " not appear after %d seconds.", + devpath, maxwait) + return result = False msg = None @@ -808,6 +953,7 @@ def address_ephemeral_resize(devpath=RESOURCE_DISK_PATH, maxwait=120, return +@azure_ds_telemetry_reporter def perform_hostname_bounce(hostname, cfg, prev_hostname): # set the hostname to 'hostname' if it is not already set to that. # then, if policy is not off, bounce the interface using command @@ -843,6 +989,7 @@ def perform_hostname_bounce(hostname, cfg, prev_hostname): return True +@azure_ds_telemetry_reporter def crtfile_to_pubkey(fname, data=None): pipeline = ('openssl x509 -noout -pubkey < "$0" |' 'ssh-keygen -i -m PKCS8 -f /dev/stdin') @@ -851,6 +998,7 @@ def crtfile_to_pubkey(fname, data=None): return out.rstrip() +@azure_ds_telemetry_reporter def pubkeys_from_crt_files(flist): pubkeys = [] errors = [] @@ -866,6 +1014,7 @@ def pubkeys_from_crt_files(flist): return pubkeys +@azure_ds_telemetry_reporter def write_files(datadir, files, dirmode=None): def _redact_password(cnt, fname): @@ -893,6 +1042,7 @@ def write_files(datadir, files, dirmode=None): util.write_file(filename=fname, content=content, mode=0o600) +@azure_ds_telemetry_reporter def invoke_agent(cmd): # this is a function itself to simplify patching it for test if cmd: @@ -912,16 +1062,19 @@ def find_child(node, filter_func): return ret +@azure_ds_telemetry_reporter def load_azure_ovf_pubkeys(sshnode): # This parses a 'SSH' node formatted like below, and returns # an array of dicts. - # [{'fp': '6BE7A7C3C8A8F4B123CCA5D0C2F1BE4CA7B63ED7', - # 'path': 'where/to/go'}] + # [{'fingerprint': '6BE7A7C3C8A8F4B123CCA5D0C2F1BE4CA7B63ED7', + # 'path': '/where/to/go'}] # # <SSH><PublicKeys> - # <PublicKey><Fingerprint>ABC</FingerPrint><Path>/ABC</Path> + # <PublicKey><Fingerprint>ABC</FingerPrint><Path>/x/y/z</Path> # ... # </PublicKeys></SSH> + # Under some circumstances, there may be a <Value> element along with the + # Fingerprint and Path. Pass those along if they appear. results = find_child(sshnode, lambda n: n.localName == "PublicKeys") if len(results) == 0: return [] @@ -962,11 +1115,14 @@ def load_azure_ovf_pubkeys(sshnode): return found +@azure_ds_telemetry_reporter def read_azure_ovf(contents): try: dom = minidom.parseString(contents) except Exception as e: - raise BrokenAzureDataSource("Invalid ovf-env.xml: %s" % e) + error_str = "Invalid ovf-env.xml: %s" % e + report_diagnostic_event(error_str) + raise BrokenAzureDataSource(error_str) results = find_child(dom.documentElement, lambda n: n.localName == "ProvisioningSection") @@ -986,8 +1142,8 @@ def read_azure_ovf(contents): raise NonAzureDataSource("No LinuxProvisioningConfigurationSet") if len(lpcs_nodes) > 1: raise BrokenAzureDataSource("found '%d' %ss" % - ("LinuxProvisioningConfigurationSet", - len(lpcs_nodes))) + (len(lpcs_nodes), + "LinuxProvisioningConfigurationSet")) lpcs = lpcs_nodes[0] if not lpcs.hasChildNodes(): @@ -1047,9 +1203,10 @@ def read_azure_ovf(contents): defuser = {} if username: defuser['name'] = username - if password and DEF_PASSWD_REDACTION != password: - defuser['passwd'] = encrypt_pass(password) + if password: defuser['lock_passwd'] = False + if DEF_PASSWD_REDACTION != password: + defuser['passwd'] = encrypt_pass(password) if defuser: cfg['system_info'] = {'default_user': defuser} @@ -1062,6 +1219,7 @@ def read_azure_ovf(contents): return (md, ud, cfg) +@azure_ds_telemetry_reporter def _extract_preprovisioned_vm_setting(dom): """Read the preprovision flag from the ovf. It should not exist unless true.""" @@ -1090,6 +1248,7 @@ def encrypt_pass(password, salt_id="$6$"): return crypt.crypt(password, salt_id + util.rand_str(strlen=16)) +@azure_ds_telemetry_reporter def _check_freebsd_cdrom(cdrom_dev): """Return boolean indicating path to cdrom device has content.""" try: @@ -1101,18 +1260,31 @@ def _check_freebsd_cdrom(cdrom_dev): return False -def _get_random_seed(): +@azure_ds_telemetry_reporter +def _get_random_seed(source=PLATFORM_ENTROPY_SOURCE): """Return content random seed file if available, otherwise, return None.""" # azure / hyper-v provides random data here - # TODO. find the seed on FreeBSD platform # now update ds_cfg to reflect contents pass in config - if util.is_FreeBSD(): + if source is None: return None - return util.load_file("/sys/firmware/acpi/tables/OEM0", - quiet=True, decode=False) + seed = util.load_file(source, quiet=True, decode=False) + + # The seed generally contains non-Unicode characters. load_file puts + # them into a str (in python 2) or bytes (in python 3). In python 2, + # bad octets in a str cause util.json_dumps() to throw an exception. In + # python 3, bytes is a non-serializable type, and the handler load_file + # uses applies b64 encoding *again* to handle it. The simplest solution + # is to just b64encode the data and then decode it to a serializable + # string. Same number of bits of entropy, just with 25% more zeroes. + # There's no need to undo this base64-encoding when the random seed is + # actually used in cc_seed_random.py. + seed = base64.b64encode(seed).decode() + return seed + +@azure_ds_telemetry_reporter def list_possible_azure_ds_devs(): devlist = [] if util.is_FreeBSD(): @@ -1127,6 +1299,7 @@ def list_possible_azure_ds_devs(): return devlist +@azure_ds_telemetry_reporter def load_azure_ds_dir(source_dir): ovf_file = os.path.join(source_dir, "ovf-env.xml") @@ -1149,47 +1322,62 @@ def parse_network_config(imds_metadata): @param: imds_metadata: Dict of content read from IMDS network service. @return: Dictionary containing network version 2 standard configuration. """ - if imds_metadata != sources.UNSET and imds_metadata: - netconfig = {'version': 2, 'ethernets': {}} - LOG.debug('Azure: generating network configuration from IMDS') - network_metadata = imds_metadata['network'] - for idx, intf in enumerate(network_metadata['interface']): - nicname = 'eth{idx}'.format(idx=idx) - dev_config = {} - for addr4 in intf['ipv4']['ipAddress']: - privateIpv4 = addr4['privateIpAddress'] - if privateIpv4: - if dev_config.get('dhcp4', False): - # Append static address config for nic > 1 - netPrefix = intf['ipv4']['subnet'][0].get( - 'prefix', '24') + with events.ReportEventStack( + name="parse_network_config", + description="", + parent=azure_ds_reporter) as evt: + if imds_metadata != sources.UNSET and imds_metadata: + netconfig = {'version': 2, 'ethernets': {}} + LOG.debug('Azure: generating network configuration from IMDS') + network_metadata = imds_metadata['network'] + for idx, intf in enumerate(network_metadata['interface']): + # First IPv4 and/or IPv6 address will be obtained via DHCP. + # Any additional IPs of each type will be set as static + # addresses. + nicname = 'eth{idx}'.format(idx=idx) + dhcp_override = {'route-metric': (idx + 1) * 100} + dev_config = {'dhcp4': True, 'dhcp4-overrides': dhcp_override, + 'dhcp6': False} + for addr_type in ('ipv4', 'ipv6'): + addresses = intf.get(addr_type, {}).get('ipAddress', []) + if addr_type == 'ipv4': + default_prefix = '24' + else: + default_prefix = '128' + if addresses: + dev_config['dhcp6'] = True + # non-primary interfaces should have a higher + # route-metric (cost) so default routes prefer + # primary nic due to lower route-metric value + dev_config['dhcp6-overrides'] = dhcp_override + for addr in addresses[1:]: + # Append static address config for ip > 1 + netPrefix = intf[addr_type]['subnet'][0].get( + 'prefix', default_prefix) + privateIp = addr['privateIpAddress'] if not dev_config.get('addresses'): dev_config['addresses'] = [] dev_config['addresses'].append( '{ip}/{prefix}'.format( - ip=privateIpv4, prefix=netPrefix)) - else: - dev_config['dhcp4'] = True - for addr6 in intf['ipv6']['ipAddress']: - privateIpv6 = addr6['privateIpAddress'] - if privateIpv6: - dev_config['dhcp6'] = True - break - if dev_config: - mac = ':'.join(re.findall(r'..', intf['macAddress'])) - dev_config.update( - {'match': {'macaddress': mac.lower()}, - 'set-name': nicname}) - netconfig['ethernets'][nicname] = dev_config - else: - blacklist = ['mlx4_core'] - LOG.debug('Azure: generating fallback configuration') - # generate a network config, blacklist picking mlx4_core devs - netconfig = net.generate_fallback_config( - blacklist_drivers=blacklist, config_driver=True) - return netconfig + ip=privateIp, prefix=netPrefix)) + if dev_config: + mac = ':'.join(re.findall(r'..', intf['macAddress'])) + dev_config.update( + {'match': {'macaddress': mac.lower()}, + 'set-name': nicname}) + netconfig['ethernets'][nicname] = dev_config + evt.description = "network config from imds" + else: + blacklist = ['mlx4_core'] + LOG.debug('Azure: generating fallback configuration') + # generate a network config, blacklist picking mlx4_core devs + netconfig = net.generate_fallback_config( + blacklist_drivers=blacklist, config_driver=True) + evt.description = "network config from fallback" + return netconfig +@azure_ds_telemetry_reporter def get_metadata_from_imds(fallback_nic, retries): """Query Azure's network metadata service, returning a dictionary. @@ -1210,29 +1398,39 @@ def get_metadata_from_imds(fallback_nic, retries): if net.is_up(fallback_nic): return util.log_time(**kwargs) else: - with EphemeralDHCPv4(fallback_nic): - return util.log_time(**kwargs) + try: + with EphemeralDHCPv4WithReporting( + azure_ds_reporter, fallback_nic): + return util.log_time(**kwargs) + except Exception as e: + report_diagnostic_event("exception while getting metadata: %s" % e) + raise +@azure_ds_telemetry_reporter def _get_metadata_from_imds(retries): url = IMDS_URL + "instance?api-version=2017-12-01" headers = {"Metadata": "true"} try: response = readurl( - url, timeout=1, headers=headers, retries=retries, - exception_cb=retry_on_url_exc) + url, timeout=IMDS_TIMEOUT_IN_SECONDS, headers=headers, + retries=retries, exception_cb=retry_on_url_exc) except Exception as e: - LOG.debug('Ignoring IMDS instance metadata: %s', e) + msg = 'Ignoring IMDS instance metadata: %s' % e + report_diagnostic_event(msg) + LOG.debug(msg) return {} try: return util.load_json(str(response)) - except json.decoder.JSONDecodeError: + except json.decoder.JSONDecodeError as e: + report_diagnostic_event('non-json imds response' % e) LOG.warning( 'Ignoring non-json IMDS instance metadata: %s', str(response)) return {} +@azure_ds_telemetry_reporter def maybe_remove_ubuntu_network_config_scripts(paths=None): """Remove Azure-specific ubuntu network config for non-primary nics. @@ -1270,14 +1468,22 @@ def maybe_remove_ubuntu_network_config_scripts(paths=None): def _is_platform_viable(seed_dir): - """Check platform environment to report if this datasource may run.""" - asset_tag = util.read_dmi_data('chassis-asset-tag') - if asset_tag == AZURE_CHASSIS_ASSET_TAG: - return True - LOG.debug("Non-Azure DMI asset tag '%s' discovered.", asset_tag) - if os.path.exists(os.path.join(seed_dir, 'ovf-env.xml')): - return True - return False + with events.ReportEventStack( + name="check-platform-viability", + description="found azure asset tag", + parent=azure_ds_reporter) as evt: + + """Check platform environment to report if this datasource may run.""" + asset_tag = util.read_dmi_data('chassis-asset-tag') + if asset_tag == AZURE_CHASSIS_ASSET_TAG: + return True + msg = "Non-Azure DMI asset tag '%s' discovered." % asset_tag + LOG.debug(msg) + evt.description = msg + report_diagnostic_event(msg) + if os.path.exists(os.path.join(seed_dir, 'ovf-env.xml')): + return True + return False class BrokenAzureDataSource(Exception): |