diff options
author | Daniel Baumann <daniel.baumann@progress-linux.org> | 2023-06-16 11:03:24 +0000 |
---|---|---|
committer | Daniel Baumann <daniel.baumann@progress-linux.org> | 2023-06-16 11:03:24 +0000 |
commit | cd7009797f422a1df7e9fc1edf633a1ec6720b13 (patch) | |
tree | 623d2a0079912a23a15b28179f8fbdb7d6fc9a66 /staslib | |
parent | Releasing debian version 2.2.2-1. (diff) | |
download | nvme-stas-cd7009797f422a1df7e9fc1edf633a1ec6720b13.tar.xz nvme-stas-cd7009797f422a1df7e9fc1edf633a1ec6720b13.zip |
Merging upstream version 2.3~rc1.
Signed-off-by: Daniel Baumann <daniel.baumann@progress-linux.org>
Diffstat (limited to 'staslib')
-rw-r--r-- | staslib/avahi.py | 8 | ||||
-rw-r--r-- | staslib/conf.py | 106 | ||||
-rw-r--r-- | staslib/ctrl.py | 360 | ||||
-rw-r--r-- | staslib/defs.py | 5 | ||||
-rw-r--r-- | staslib/gutil.py | 2 | ||||
-rw-r--r-- | staslib/iputil.py | 305 | ||||
-rw-r--r-- | staslib/meson.build | 1 | ||||
-rw-r--r-- | staslib/nbft.py | 28 | ||||
-rw-r--r-- | staslib/service.py | 12 | ||||
-rw-r--r-- | staslib/singleton.py | 25 | ||||
-rw-r--r-- | staslib/stas.py | 44 | ||||
-rw-r--r-- | staslib/trid.py | 17 | ||||
-rw-r--r-- | staslib/udev.py | 199 |
13 files changed, 806 insertions, 306 deletions
diff --git a/staslib/avahi.py b/staslib/avahi.py index c8a3a0b..26543d9 100644 --- a/staslib/avahi.py +++ b/staslib/avahi.py @@ -29,9 +29,11 @@ def _txt2dict(txt: list): for list_of_chars in txt: try: string = functools.reduce(lambda accumulator, c: accumulator + chr(c), list_of_chars, '') - key, val = string.split("=") - the_dict[key.lower()] = val - except Exception: # pylint: disable=broad-except + if string.isprintable(): + key, val = string.split('=') + if key: # Make sure the key is not an empty string + the_dict[key.lower()] = val + except ValueError: pass return the_dict diff --git a/staslib/conf.py b/staslib/conf.py index a54da98..e5c038b 100644 --- a/staslib/conf.py +++ b/staslib/conf.py @@ -14,7 +14,8 @@ import sys import logging import functools import configparser -from staslib import defs, singleton, timeparse +from urllib.parse import urlparse +from staslib import defs, iputil, nbft, singleton, timeparse __TOKEN_RE = re.compile(r'\s*;\s*') __OPTION_RE = re.compile(r'\s*=\s*') @@ -83,7 +84,8 @@ def _to_ip_family(text): class OrderedMultisetDict(dict): '''This class is used to change the behavior of configparser.ConfigParser and allow multiple configuration parameters with the same key. The - result is a list of values. + result is a list of values, where values are sorted by the order they + appear in the file. ''' def __setitem__(self, key, value): @@ -317,7 +319,7 @@ class SvcConf(metaclass=singleton.Singleton): # pylint: disable=too-many-public option = 'persistent-connections' value = self.get_option(section, option, ignore_default=True) - legacy = self.get_option('Global', 'persistent-connections', ignore_default=True) + legacy = self.get_option('Global', option, ignore_default=True) if value is None and legacy is None: return self._defaults.get((section, option), True) @@ -381,7 +383,7 @@ class SvcConf(metaclass=singleton.Singleton): # pylint: disable=too-many-public controller_list = self.get_option('Controllers', 'exclude') # 2022-09-20: Look for "blacklist". This is for backwards compatibility - # with releases 1.0 to 1.1.6. This is to be phased out (i.e. remove by 2024) + # with releases 1.0 to 1.1.x. This is to be phased out (i.e. remove by 2024) controller_list += self.get_option('Controllers', 'blacklist') excluded = [_parse_controller(controller) for controller in controller_list] @@ -572,7 +574,7 @@ class SysConf(metaclass=singleton.Singleton): try: value = self.__get_value('Host', 'key', defs.NVME_HOSTKEY) except FileNotFoundError as ex: - logging.info('Host key undefined: %s', ex) + logging.debug('Host key undefined: %s', ex) value = None return value @@ -701,3 +703,97 @@ class NvmeOptions(metaclass=singleton.Singleton): def dhchap_ctrlkey_supp(self): '''This option allows specifying the controller DHCHAP key used for authentication.''' return self._supported_options['dhchap_ctrl_secret'] + + +# ****************************************************************************** +class NbftConf(metaclass=singleton.Singleton): + '''Read and cache configuration file.''' + + def __init__(self, root_dir=defs.NBFT_SYSFS_PATH): + self._disc_ctrls = [] + self._subs_ctrls = [] + + nbft_files = nbft.get_nbft_files(root_dir) + if len(nbft_files): + logging.info('NBFT location(s): %s', list(nbft_files.keys())) + + for data in nbft_files.values(): + hfis = data.get('hfi', []) + discovery = data.get('discovery', []) + subsystem = data.get('subsystem', []) + + self._disc_ctrls.extend(NbftConf.__nbft_disc_to_cids(discovery, hfis)) + self._subs_ctrls.extend(NbftConf.__nbft_subs_to_cids(subsystem, hfis)) + + dcs = property(lambda self: self._disc_ctrls) + iocs = property(lambda self: self._subs_ctrls) + + def get_controllers(self): + '''Retrieve the list of controllers. Stafd only cares about + discovery controllers. Stacd only cares about I/O controllers.''' + + # For now, only return DCs. There are still unanswered questions + # regarding I/O controllers, e.g. what if multipathing has been + # configured. + return self.dcs if defs.PROG_NAME == 'stafd' else [] + + @staticmethod + def __nbft_disc_to_cids(discovery, hfis): + cids = [] + + for ctrl in discovery: + cid = NbftConf.__uri2cid(ctrl['uri']) + cid['subsysnqn'] = ctrl['nqn'] + + host_iface = NbftConf.__get_host_iface(ctrl.get('hfi_index'), hfis) + if host_iface: + cid['host-iface'] = host_iface + + cids.append(cid) + + return cids + + @staticmethod + def __nbft_subs_to_cids(subsystem, hfis): + cids = [] + + for ctrl in subsystem: + cid = { + 'transport': ctrl['trtype'], + 'traddr': ctrl['traddr'], + 'trsvcid': ctrl['trsvcid'], + 'subsysnqn': ctrl['subsys_nqn'], + 'hdr-digest': ctrl['pdu_header_digest_required'], + 'data-digest': ctrl['data_digest_required'], + } + + indexes = ctrl.get('hfi_indexes') + if isinstance(indexes, list) and len(indexes) > 0: + host_iface = NbftConf.__get_host_iface(indexes[0], hfis) + if host_iface: + cid['host-iface'] = host_iface + + cids.append(cid) + + return cids + + @staticmethod + def __get_host_iface(indx, hfis): + if indx is None or indx >= len(hfis): + return None + + mac = hfis[indx].get('mac_addr') + if mac is None: + return None + + return iputil.mac2iface(mac) + + @staticmethod + def __uri2cid(uri: str): + '''Convert a URI of the form "nvme+tcp://100.71.103.50:8009/" to a Controller ID''' + obj = urlparse(uri) + return { + 'transport': obj.scheme.split('+')[1], + 'traddr': obj.hostname, + 'trsvcid': str(obj.port), + } diff --git a/staslib/ctrl.py b/staslib/ctrl.py index 97a1c7b..ad221e0 100644 --- a/staslib/ctrl.py +++ b/staslib/ctrl.py @@ -135,26 +135,7 @@ class Controller(stas.ControllerABC): # pylint: disable=too-many-instance-attri self._root.log_level("debug" if tron else "err") def _on_udev_notification(self, udev_obj): - if self._alive(): - if udev_obj.action == 'change': - nvme_aen = udev_obj.get('NVME_AEN') - nvme_event = udev_obj.get('NVME_EVENT') - if isinstance(nvme_aen, str): - logging.info('%s | %s - Received AEN: %s', self.id, udev_obj.sys_name, nvme_aen) - self._on_aen(int(nvme_aen, 16)) - if isinstance(nvme_event, str): - self._on_nvme_event(nvme_event) - elif udev_obj.action == 'remove': - logging.info('%s | %s - Received "remove" event', self.id, udev_obj.sys_name) - self._on_ctrl_removed(udev_obj) - else: - logging.debug( - 'Controller._on_udev_notification() - %s | %s: Received "%s" event', - self.id, - udev_obj.sys_name, - udev_obj.action, - ) - else: + if not self._alive(): logging.debug( 'Controller._on_udev_notification() - %s | %s: Received event on dead object. udev_obj %s: %s', self.id, @@ -162,6 +143,26 @@ class Controller(stas.ControllerABC): # pylint: disable=too-many-instance-attri udev_obj.action, udev_obj.sys_name, ) + return + + if udev_obj.action == 'change': + nvme_aen = udev_obj.get('NVME_AEN') + nvme_event = udev_obj.get('NVME_EVENT') + if isinstance(nvme_aen, str): + logging.info('%s | %s - Received AEN: %s', self.id, udev_obj.sys_name, nvme_aen) + self._on_aen(int(nvme_aen, 16)) + if isinstance(nvme_event, str): + self._on_nvme_event(nvme_event) + elif udev_obj.action == 'remove': + logging.info('%s | %s - Received "remove" event', self.id, udev_obj.sys_name) + self._on_ctrl_removed(udev_obj) + else: + logging.debug( + 'Controller._on_udev_notification() - %s | %s: Received "%s" event', + self.id, + udev_obj.sys_name, + udev_obj.action, + ) def _on_ctrl_removed(self, udev_obj): # pylint: disable=unused-argument if self._udev: @@ -269,48 +270,51 @@ class Controller(stas.ControllerABC): # pylint: disable=too-many-instance-attri op_obj.kill() self._connect_op = None - if self._alive(): - self._device = self._ctrl.name - logging.info('%s | %s - Connection established!', self.id, self.device) - self._connect_attempts = 0 - self._udev.register_for_device_events(self._device, self._on_udev_notification) - else: + if not self._alive(): logging.debug( 'Controller._on_connect_success() - %s | %s: Received event on dead object. data=%s', self.id, self.device, data, ) + return + + self._device = self._ctrl.name + logging.info('%s | %s - Connection established!', self.id, self.device) + self._connect_attempts = 0 + self._udev.register_for_device_events(self._device, self._on_udev_notification) def _on_connect_fail(self, op_obj: gutil.AsyncTask, err, fail_cnt): # pylint: disable=unused-argument '''@brief Function called when we fail to connect to the Controller.''' op_obj.kill() self._connect_op = None - if self._alive(): - if self._connect_attempts == 1: - # Do a fast re-try on the first failure. - self._retry_connect_tmr.set_timeout(self.FAST_CONNECT_RETRY_PERIOD_SEC) - elif self._connect_attempts == 2: - # If the fast connect re-try fails, then we can print a message to - # indicate the failure, and start a slow re-try period. - self._retry_connect_tmr.set_timeout(self.CONNECT_RETRY_PERIOD_SEC) - logging.error('%s Failed to connect to controller. %s %s', self.id, err.domain, err.message) - - if self._should_try_to_reconnect(): - logging.debug( - 'Controller._on_connect_fail() - %s %s. Retry in %s sec.', - self.id, - err, - self._retry_connect_tmr.get_timeout(), - ) - self._retry_connect_tmr.start() - else: + + if not self._alive(): logging.debug( 'Controller._on_connect_fail() - %s Received event on dead object. %s %s', self.id, err.domain, err.message, ) + return + + if self._connect_attempts == 1: + # Do a fast re-try on the first failure. + self._retry_connect_tmr.set_timeout(self.FAST_CONNECT_RETRY_PERIOD_SEC) + elif self._connect_attempts == 2: + # If the fast connect re-try fails, then we can print a message to + # indicate the failure, and start a slow re-try period. + self._retry_connect_tmr.set_timeout(self.CONNECT_RETRY_PERIOD_SEC) + logging.error('%s Failed to connect to controller. %s %s', self.id, err.domain, err.message) + + if self._should_try_to_reconnect(): + logging.debug( + 'Controller._on_connect_fail() - %s %s. Retry in %s sec.', + self.id, + err, + self._retry_connect_tmr.get_timeout(), + ) + self._retry_connect_tmr.start() def disconnect(self, disconnected_cb, keep_connection): '''@brief Issue an asynchronous disconnect command to a Controller. @@ -507,11 +511,7 @@ class Dc(Controller): return - logging.info( - '%s | %s - Controller not responding. Retrying...', - self.id, - self.device, - ) + logging.info('%s | %s - Controller not responding. Retrying...', self.id, self.device) self._ctrl_unresponsive_time = None self._ctrl_unresponsive_tmr.stop() @@ -555,8 +555,8 @@ class Dc(Controller): def _post_registration_actions(self): # Need to check that supported_log_pages() is available (introduced in libnvme 1.2) - has_supported_log_pages = hasattr(self._ctrl, 'supported_log_pages') - if not has_supported_log_pages: + get_slp = getattr(self._ctrl, 'supported_log_pages', None) + if get_slp is None: logging.warning( '%s | %s - libnvme-%s does not support "Get supported log pages". Please upgrade libnvme.', self.id, @@ -564,9 +564,9 @@ class Dc(Controller): defs.LIBNVME_VERSION, ) - if conf.SvcConf().pleo_enabled and self._is_ddc() and has_supported_log_pages: + if conf.SvcConf().pleo_enabled and self._is_ddc() and get_slp is not None: self._get_supported_op = gutil.AsyncTask( - self._on_get_supported_success, self._on_get_supported_fail, self._ctrl.supported_log_pages + self._on_get_supported_success, self._on_get_supported_fail, get_slp ) self._get_supported_op.run_async() else: @@ -580,21 +580,23 @@ class Dc(Controller): ''' super()._on_connect_success(op_obj, data) - if self._alive(): - self._ctrl_unresponsive_time = None - self._ctrl_unresponsive_tmr.stop() - self._ctrl_unresponsive_tmr.set_timeout(0) - - if self._ctrl.is_registration_supported(): - self._register_op = gutil.AsyncTask( - self._on_registration_success, - self._on_registration_fail, - self._ctrl.registration_ctlr, - nvme.NVMF_DIM_TAS_REGISTER, - ) - self._register_op.run_async() - else: - self._post_registration_actions() + if not self._alive(): + return + + self._ctrl_unresponsive_time = None + self._ctrl_unresponsive_tmr.stop() + self._ctrl_unresponsive_tmr.set_timeout(0) + + if self._ctrl.is_registration_supported(): + self._register_op = gutil.AsyncTask( + self._on_registration_success, + self._on_registration_fail, + self._ctrl.registration_ctlr, + nvme.NVMF_DIM_TAS_REGISTER, + ) + self._register_op.run_async() + else: + self._post_registration_actions() def _on_connect_fail(self, op_obj: gutil.AsyncTask, err, fail_cnt): '''@brief Function called when we fail to connect to the Controller.''' @@ -613,35 +615,25 @@ class Dc(Controller): refers to the fact that a successful exchange was made with the DC. It doesn't mean that the registration itself succeeded. ''' - if self._alive(): - if data is not None: - logging.warning('%s | %s - Registration error. %s.', self.id, self.device, data) - else: - logging.debug('Dc._on_registration_success() - %s | %s', self.id, self.device) - - self._post_registration_actions() - else: + if not self._alive(): logging.debug( 'Dc._on_registration_success() - %s | %s: Received event on dead object.', self.id, self.device ) + return + + if data is not None: + logging.warning('%s | %s - Registration error. %s.', self.id, self.device, data) + else: + logging.debug('Dc._on_registration_success() - %s | %s', self.id, self.device) + + self._post_registration_actions() def _on_registration_fail(self, op_obj: gutil.AsyncTask, err, fail_cnt): '''@brief Function called when we fail to register with the Discovery Controller. See self._register_op object for details. ''' - if self._alive(): - logging.debug( - 'Dc._on_registration_fail() - %s | %s: %s. Retry in %s sec', - self.id, - self.device, - err, - Dc.REGISTRATION_RETRY_RERIOD_SEC, - ) - if fail_cnt == 1: # Throttle the logs. Only print the first time the command fails - logging.error('%s | %s - Failed to register with Discovery Controller. %s', self.id, self.device, err) - op_obj.retry(Dc.REGISTRATION_RETRY_RERIOD_SEC) - else: + if not self._alive(): logging.debug( 'Dc._on_registration_fail() - %s | %s: Received event on dead object. %s', self.id, @@ -649,6 +641,18 @@ class Dc(Controller): err, ) op_obj.kill() + return + + logging.debug( + 'Dc._on_registration_fail() - %s | %s: %s. Retry in %s sec', + self.id, + self.device, + err, + Dc.REGISTRATION_RETRY_RERIOD_SEC, + ) + if fail_cnt == 1: # Throttle the logs. Only print the first time the command fails + logging.error('%s | %s - Failed to register with Discovery Controller. %s', self.id, self.device, err) + op_obj.retry(Dc.REGISTRATION_RETRY_RERIOD_SEC) # -------------------------------------------------------------------------- def _on_get_supported_success(self, op_obj: gutil.AsyncTask, data): # pylint: disable=unused-argument @@ -660,68 +664,70 @@ class Dc(Controller): refers to the fact that a successful exchange was made with the DC. It doesn't mean that the Get Supported Log Page itself succeeded. ''' - if self._alive(): - try: - dlp_supp_opts = data[nvme.NVME_LOG_LID_DISCOVER] >> 16 - except (TypeError, IndexError): - dlp_supp_opts = 0 - + if not self._alive(): logging.debug( - 'Dc._on_get_supported_success() - %s | %s: supported options = 0x%04X = %s', - self.id, - self.device, - dlp_supp_opts, - dlp_supp_opts_as_string(dlp_supp_opts), + 'Dc._on_get_supported_success() - %s | %s: Received event on dead object.', self.id, self.device ) + return - if 'lsp' in inspect.signature(self._ctrl.discover).parameters: - lsp = nvme.NVMF_LOG_DISC_LSP_PLEO if dlp_supp_opts & nvme.NVMF_LOG_DISC_LID_PLEOS else 0 - self._get_log_op = gutil.AsyncTask( - self._on_get_log_success, self._on_get_log_fail, self._ctrl.discover, lsp - ) - else: - logging.warning( - '%s | %s - libnvme-%s does not support setting PLEO bit. Please upgrade.', - self.id, - self.device, - defs.LIBNVME_VERSION, - ) - self._get_log_op = gutil.AsyncTask(self._on_get_log_success, self._on_get_log_fail, self._ctrl.discover) - self._get_log_op.run_async() + try: + dlp_supp_opts = data[nvme.NVME_LOG_LID_DISCOVER] >> 16 + except (TypeError, IndexError): + dlp_supp_opts = 0 + + logging.debug( + 'Dc._on_get_supported_success() - %s | %s: supported options = 0x%04X = %s', + self.id, + self.device, + dlp_supp_opts, + dlp_supp_opts_as_string(dlp_supp_opts), + ) + + if 'lsp' in inspect.signature(self._ctrl.discover).parameters: + lsp = nvme.NVMF_LOG_DISC_LSP_PLEO if dlp_supp_opts & nvme.NVMF_LOG_DISC_LID_PLEOS else 0 + self._get_log_op = gutil.AsyncTask( + self._on_get_log_success, self._on_get_log_fail, self._ctrl.discover, lsp + ) else: - logging.debug( - 'Dc._on_get_supported_success() - %s | %s: Received event on dead object.', self.id, self.device + logging.warning( + '%s | %s - libnvme-%s does not support setting PLEO bit. Please upgrade.', + self.id, + self.device, + defs.LIBNVME_VERSION, ) + self._get_log_op = gutil.AsyncTask(self._on_get_log_success, self._on_get_log_fail, self._ctrl.discover) + self._get_log_op.run_async() def _on_get_supported_fail(self, op_obj: gutil.AsyncTask, err, fail_cnt): '''@brief Function called when we fail to retrieve the supported log page from the Discovery Controller. See self._get_supported_op object for details. ''' - if self._alive(): + if not self._alive(): logging.debug( - 'Dc._on_get_supported_fail() - %s | %s: %s. Retry in %s sec', + 'Dc._on_get_supported_fail() - %s | %s: Received event on dead object. %s', self.id, self.device, err, - Dc.GET_SUPPORTED_RETRY_RERIOD_SEC, ) - if fail_cnt == 1: # Throttle the logs. Only print the first time the command fails - logging.error( - '%s | %s - Failed to Get supported log pages from Discovery Controller. %s', - self.id, - self.device, - err, - ) - op_obj.retry(Dc.GET_SUPPORTED_RETRY_RERIOD_SEC) - else: - logging.debug( - 'Dc._on_get_supported_fail() - %s | %s: Received event on dead object. %s', + op_obj.kill() + return + + logging.debug( + 'Dc._on_get_supported_fail() - %s | %s: %s. Retry in %s sec', + self.id, + self.device, + err, + Dc.GET_SUPPORTED_RETRY_RERIOD_SEC, + ) + if fail_cnt == 1: # Throttle the logs. Only print the first time the command fails + logging.error( + '%s | %s - Failed to Get supported log pages from Discovery Controller. %s', self.id, self.device, err, ) - op_obj.kill() + op_obj.retry(Dc.GET_SUPPORTED_RETRY_RERIOD_SEC) # -------------------------------------------------------------------------- def _on_get_log_success(self, op_obj: gutil.AsyncTask, data): # pylint: disable=unused-argument @@ -729,61 +735,51 @@ class Dc(Controller): from the Discovery Controller. See self._get_log_op object for details. ''' - if self._alive(): - # Note that for historical reasons too long to explain, the CDC may - # return invalid addresses ('0.0.0.0', '::', or ''). Those need to - # be filtered out. - referrals_before = self.referrals() - self._log_pages = ( - [ - {k.strip(): str(v).strip() for k, v in dictionary.items()} - for dictionary in data - if dictionary.get('traddr', '').strip() not in ('0.0.0.0', '::', '') - ] - if data - else list() + if not self._alive(): + logging.debug( + 'Dc._on_get_log_success() - %s | %s: Received event on dead object.', self.id, self.device ) - logging.info( - '%s | %s - Received discovery log pages (num records=%s).', self.id, self.device, len(self._log_pages) + return + + # Note that for historical reasons too long to explain, the CDC may + # return invalid addresses ('0.0.0.0', '::', or ''). Those need to + # be filtered out. + referrals_before = self.referrals() + self._log_pages = ( + [ + {k.strip(): str(v).strip() for k, v in dictionary.items()} + for dictionary in data + if dictionary.get('traddr', '').strip() not in ('0.0.0.0', '::', '') + ] + if data + else list() + ) + logging.info( + '%s | %s - Received discovery log pages (num records=%s).', self.id, self.device, len(self._log_pages) + ) + referrals_after = self.referrals() + self._serv.log_pages_changed(self, self.device) + if referrals_after != referrals_before: + logging.debug( + 'Dc._on_get_log_success() - %s | %s: Referrals before = %s', + self.id, + self.device, + referrals_before, ) - referrals_after = self.referrals() - self._serv.log_pages_changed(self, self.device) - if referrals_after != referrals_before: - logging.debug( - 'Dc._on_get_log_success() - %s | %s: Referrals before = %s', - self.id, - self.device, - referrals_before, - ) - logging.debug( - 'Dc._on_get_log_success() - %s | %s: Referrals after = %s', - self.id, - self.device, - referrals_after, - ) - self._serv.referrals_changed() - else: logging.debug( - 'Dc._on_get_log_success() - %s | %s: Received event on dead object.', self.id, self.device + 'Dc._on_get_log_success() - %s | %s: Referrals after = %s', + self.id, + self.device, + referrals_after, ) + self._serv.referrals_changed() def _on_get_log_fail(self, op_obj: gutil.AsyncTask, err, fail_cnt): '''@brief Function called when we fail to retrieve the log pages from the Discovery Controller. See self._get_log_op object for details. ''' - if self._alive(): - logging.debug( - 'Dc._on_get_log_fail() - %s | %s: %s. Retry in %s sec', - self.id, - self.device, - err, - Dc.GET_LOG_PAGE_RETRY_RERIOD_SEC, - ) - if fail_cnt == 1: # Throttle the logs. Only print the first time the command fails - logging.error('%s | %s - Failed to retrieve log pages. %s', self.id, self.device, err) - op_obj.retry(Dc.GET_LOG_PAGE_RETRY_RERIOD_SEC) - else: + if not self._alive(): logging.debug( 'Dc._on_get_log_fail() - %s | %s: Received event on dead object. %s', self.id, @@ -791,6 +787,18 @@ class Dc(Controller): err, ) op_obj.kill() + return + + logging.debug( + 'Dc._on_get_log_fail() - %s | %s: %s. Retry in %s sec', + self.id, + self.device, + err, + Dc.GET_LOG_PAGE_RETRY_RERIOD_SEC, + ) + if fail_cnt == 1: # Throttle the logs. Only print the first time the command fails + logging.error('%s | %s - Failed to retrieve log pages. %s', self.id, self.device, err) + op_obj.retry(Dc.GET_LOG_PAGE_RETRY_RERIOD_SEC) # ****************************************************************************** diff --git a/staslib/defs.py b/staslib/defs.py index 5a50371..877f6e6 100644 --- a/staslib/defs.py +++ b/staslib/defs.py @@ -12,6 +12,7 @@ import os import sys import shutil import platform +from libnvme import nvme from staslib.version import KernelVersion try: @@ -48,4 +49,8 @@ SYS_CONF_FILE = '/etc/stas/sys.conf' STAFD_CONF_FILE = '/etc/stas/stafd.conf' STACD_CONF_FILE = '/etc/stas/stacd.conf' +HAS_NBFT_SUPPORT = hasattr(nvme, 'nbft_get') +NBFT_SYSFS_PATH = "/sys/firmware/acpi/tables" +NBFT_SYSFS_FILENAME = "NBFT*" + SYSTEMCTL = shutil.which('systemctl') diff --git a/staslib/gutil.py b/staslib/gutil.py index c40b80e..836674a 100644 --- a/staslib/gutil.py +++ b/staslib/gutil.py @@ -309,7 +309,7 @@ class AsyncTask: # pylint: disable=too-many-instance-attributes '''Return object members as a dictionary''' info = { 'fail count': self._fail_cnt, - 'completed': self._task.get_completed(), + 'completed': self._task.get_completed() if self._task else None, 'alive': self._alive(), } diff --git a/staslib/iputil.py b/staslib/iputil.py index 9199a49..96c5d56 100644 --- a/staslib/iputil.py +++ b/staslib/iputil.py @@ -8,45 +8,126 @@ '''A collection of IP address and network interface utilities''' +import struct import socket -import logging import ipaddress -from staslib import conf +RTM_BASE = 16 +RTM_GETLINK = 18 RTM_NEWADDR = 20 RTM_GETADDR = 22 NLM_F_REQUEST = 0x01 NLM_F_ROOT = 0x100 NLMSG_DONE = 3 -IFLA_ADDRESS = 1 -NLMSGHDR_SZ = 16 +NLMSG_HDRLEN = 16 IFADDRMSG_SZ = 8 +IFINFOMSG_SZ = 16 +ARPHRD_ETHER = 1 +ARPHRD_LOOPBACK = 772 +NLMSG_LENGTH = lambda msg_len: msg_len + NLMSG_HDRLEN # pylint: disable=unnecessary-lambda-assignment + RTATTR_SZ = 4 +RTA_ALIGN = lambda length: ((length + 3) & ~3) # pylint: disable=unnecessary-lambda-assignment +IFLA_ADDRESS = 1 +IFLA_IFNAME = 3 + + +def _nlmsghdr(nlmsg_type, nlmsg_flags, nlmsg_seq, nlmsg_pid, msg_len: int): + '''Implement this C struct: + struct nlmsghdr { + __u32 nlmsg_len; /* Length of message including header */ + __u16 nlmsg_type; /* Message content */ + __u16 nlmsg_flags; /* Additional flags */ + __u32 nlmsg_seq; /* Sequence number */ + __u32 nlmsg_pid; /* Sending process port ID */ + }; + ''' + return struct.pack('<LHHLL', NLMSG_LENGTH(msg_len), nlmsg_type, nlmsg_flags, nlmsg_seq, nlmsg_pid) + + +def _ifaddrmsg(family=0, prefixlen=0, flags=0, scope=0, index=0): + '''Implement this C struct: + struct ifaddrmsg { + __u8 ifa_family; + __u8 ifa_prefixlen; /* The prefix length */ + __u8 ifa_flags; /* Flags */ + __u8 ifa_scope; /* Address scope */ + __u32 ifa_index; /* Link index */ + }; + ''' + return struct.pack('<BBBBL', family, prefixlen, flags, scope, index) + + +def _ifinfomsg(family=0, dev_type=0, index=0, flags=0, change=0): + '''Implement this C struct: + struct ifinfomsg { + unsigned char ifi_family; /* AF_UNSPEC */ + unsigned char __ifi_pad; + unsigned short ifi_type; /* Device type: ARPHRD_* */ + int ifi_index; /* Interface index */ + unsigned int ifi_flags; /* Device flags: IFF_* */ + unsigned int ifi_change; /* change mask: IFF_* */ + }; + ''' + return struct.pack('<BBHiII', family, 0, dev_type, index, flags, change) + + +def _nlmsg(nlmsg_type, nlmsg_flags, msg: bytes): + '''Build a Netlink message''' + return _nlmsghdr(nlmsg_type, nlmsg_flags, 0, 0, len(msg)) + msg + + +# Netlink request (Get address command) +GETADDRCMD = _nlmsg(RTM_GETADDR, NLM_F_REQUEST | NLM_F_ROOT, _ifaddrmsg()) # Netlink request (Get address command) -GETADDRCMD = ( - # BEGIN: struct nlmsghdr - b'\0' * 4 # nlmsg_len (placeholder - actual length calculated below) - + (RTM_GETADDR).to_bytes(2, byteorder='little', signed=False) # nlmsg_type - + (NLM_F_REQUEST | NLM_F_ROOT).to_bytes(2, byteorder='little', signed=False) # nlmsg_flags - + b'\0' * 2 # nlmsg_seq - + b'\0' * 2 # nlmsg_pid - # END: struct nlmsghdr - + b'\0' * 8 # struct ifaddrmsg -) -GETADDRCMD = len(GETADDRCMD).to_bytes(4, byteorder='little') + GETADDRCMD[4:] # nlmsg_len +GETLINKCMD = _nlmsg(RTM_GETLINK, NLM_F_REQUEST | NLM_F_ROOT, _ifinfomsg(family=socket.AF_UNSPEC, change=0xFFFFFFFF)) # ****************************************************************************** -def get_ipaddress_obj(ipaddr): - '''@brief Return a IPv4Address or IPv6Address depending on whether @ipaddr - is a valid IPv4 or IPv6 address. Return None otherwise.''' - try: - ip = ipaddress.ip_address(ipaddr) - except ValueError: - return None +def _data_matches_mac(data, mac): + return mac.lower() == ':'.join([f'{x:02x}' for x in data[0:6]]) - return ip + +def mac2iface(mac: str): # pylint: disable=too-many-locals + '''@brief Find the interface that has @mac as its assigned MAC address. + @param mac: The MAC address to match + ''' + with socket.socket(family=socket.AF_NETLINK, type=socket.SOCK_RAW, proto=socket.NETLINK_ROUTE) as sock: + sock.sendall(GETLINKCMD) + nlmsg = sock.recv(8192) + nlmsg_idx = 0 + while True: # pylint: disable=too-many-nested-blocks + if nlmsg_idx >= len(nlmsg): + nlmsg += sock.recv(8192) + + nlmsghdr = nlmsg[nlmsg_idx : nlmsg_idx + NLMSG_HDRLEN] + nlmsg_len, nlmsg_type, _, _, _ = struct.unpack('<LHHLL', nlmsghdr) + + if nlmsg_type == NLMSG_DONE: + break + + if nlmsg_type == RTM_BASE: + msg_indx = nlmsg_idx + NLMSG_HDRLEN + msg = nlmsg[msg_indx : msg_indx + IFINFOMSG_SZ] # ifinfomsg + _, _, ifi_type, ifi_index, _, _ = struct.unpack('<BBHiII', msg) + + if ifi_type in (ARPHRD_LOOPBACK, ARPHRD_ETHER): + rtattr_indx = msg_indx + IFINFOMSG_SZ + while rtattr_indx < (nlmsg_idx + nlmsg_len): + rtattr = nlmsg[rtattr_indx : rtattr_indx + RTATTR_SZ] + rta_len, rta_type = struct.unpack('<HH', rtattr) + if rta_type == IFLA_ADDRESS: + data = nlmsg[rtattr_indx + RTATTR_SZ : rtattr_indx + rta_len] + if _data_matches_mac(data, mac): + return socket.if_indextoname(ifi_index) + + rta_len = RTA_ALIGN(rta_len) # Round up to multiple of 4 + rtattr_indx += rta_len # Move to next rtattr + + nlmsg_idx += nlmsg_len # Move to next Netlink message + + return '' # ****************************************************************************** @@ -71,8 +152,7 @@ def _data_matches_ip(data_family, data, ip): return other_ip == ip -# ****************************************************************************** -def iface_of(src_addr): +def _iface_of(src_addr): # pylint: disable=too-many-locals '''@brief Find the interface that has src_addr as one of its assigned IP addresses. @param src_addr: The IP address to match @type src_addr: Instance of ipaddress.IPv4Address or ipaddress.IPv6Address @@ -85,36 +165,133 @@ def iface_of(src_addr): if nlmsg_idx >= len(nlmsg): nlmsg += sock.recv(8192) - nlmsg_type = int.from_bytes(nlmsg[nlmsg_idx + 4 : nlmsg_idx + 6], byteorder='little', signed=False) + nlmsghdr = nlmsg[nlmsg_idx : nlmsg_idx + NLMSG_HDRLEN] + nlmsg_len, nlmsg_type, _, _, _ = struct.unpack('<LHHLL', nlmsghdr) + if nlmsg_type == NLMSG_DONE: break - if nlmsg_type != RTM_NEWADDR: - break + if nlmsg_type == RTM_NEWADDR: + msg_indx = nlmsg_idx + NLMSG_HDRLEN + msg = nlmsg[msg_indx : msg_indx + IFADDRMSG_SZ] # ifaddrmsg + ifa_family, _, _, _, ifa_index = struct.unpack('<BBBBL', msg) + + rtattr_indx = msg_indx + IFADDRMSG_SZ + while rtattr_indx < (nlmsg_idx + nlmsg_len): + rtattr = nlmsg[rtattr_indx : rtattr_indx + RTATTR_SZ] + rta_len, rta_type = struct.unpack('<HH', rtattr) + if rta_type == IFLA_ADDRESS: + data = nlmsg[rtattr_indx + RTATTR_SZ : rtattr_indx + rta_len] + if _data_matches_ip(ifa_family, data, src_addr): + return socket.if_indextoname(ifa_index) + + rta_len = RTA_ALIGN(rta_len) # Round up to multiple of 4 + rtattr_indx += rta_len # Move to next rtattr + + nlmsg_idx += nlmsg_len # Move to next Netlink message + + return '' + + +# ****************************************************************************** +def get_ipaddress_obj(ipaddr, ipv4_mapped_convert=False): + '''@brief Return a IPv4Address or IPv6Address depending on whether @ipaddr + is a valid IPv4 or IPv6 address. Return None otherwise. + + If ipv4_mapped_resolve is set to True, IPv6 addresses that are IPv4-Mapped, + will be converted to their IPv4 equivalent. + ''' + try: + ip = ipaddress.ip_address(ipaddr) + except ValueError: + return None + + if ipv4_mapped_convert: + ipv4_mapped = getattr(ip, 'ipv4_mapped', None) + if ipv4_mapped is not None: + ip = ipv4_mapped + + return ip + - nlmsg_len = int.from_bytes(nlmsg[nlmsg_idx : nlmsg_idx + 4], byteorder='little', signed=False) - if nlmsg_len % 4: # Is msg length not a multiple of 4? +# ****************************************************************************** +def net_if_addrs(): # pylint: disable=too-many-locals + '''@brief Return a dictionary listing every IP addresses for each interface. + The first IP address of a list is the primary address used as the default + source address. + @example: { + 'wlp0s20f3': { + 4: ['10.0.0.28'], + 6: [ + 'fd5e:9a9e:c5bd:0:5509:890c:1848:3843', + 'fd5e:9a9e:c5bd:0:1fd5:e527:8df7:7912', + '2605:59c8:6128:fb00:c083:1b8:c467:81d2', + '2605:59c8:6128:fb00:e99d:1a02:38e0:ad52', + 'fe80::d71b:e807:d5ee:7614' + ], + }, + 'lo': { + 4: ['127.0.0.1'], + 6: ['::1'], + }, + 'docker0': { + 4: ['172.17.0.1'], + 6: [] + }, + } + ''' + interfaces = {} + with socket.socket(socket.AF_NETLINK, socket.SOCK_RAW) as sock: + sock.sendall(GETADDRCMD) + nlmsg = sock.recv(8192) + nlmsg_idx = 0 + while True: # pylint: disable=too-many-nested-blocks + if nlmsg_idx >= len(nlmsg): + nlmsg += sock.recv(8192) + + nlmsghdr = nlmsg[nlmsg_idx : nlmsg_idx + NLMSG_HDRLEN] + nlmsg_len, nlmsg_type, _, _, _ = struct.unpack('<LHHLL', nlmsghdr) + + if nlmsg_type == NLMSG_DONE: break - ifaddrmsg_indx = nlmsg_idx + NLMSGHDR_SZ - ifa_family = nlmsg[ifaddrmsg_indx] - ifa_index = int.from_bytes(nlmsg[ifaddrmsg_indx + 4 : ifaddrmsg_indx + 8], byteorder='little', signed=False) + if nlmsg_type == RTM_NEWADDR: + msg_indx = nlmsg_idx + NLMSG_HDRLEN + msg = nlmsg[msg_indx : msg_indx + IFADDRMSG_SZ] # ifaddrmsg + ifa_family, _, _, _, ifa_index = struct.unpack('<BBBBL', msg) + + if ifa_family in (socket.AF_INET, socket.AF_INET6): + interfaces.setdefault(ifa_index, {4: [], 6: []}) + + rtattr_indx = msg_indx + IFADDRMSG_SZ + while rtattr_indx < (nlmsg_idx + nlmsg_len): + rtattr = nlmsg[rtattr_indx : rtattr_indx + RTATTR_SZ] + rta_len, rta_type = struct.unpack('<HH', rtattr) - rtattr_indx = ifaddrmsg_indx + IFADDRMSG_SZ - while rtattr_indx < (nlmsg_idx + nlmsg_len): - rta_len = int.from_bytes(nlmsg[rtattr_indx : rtattr_indx + 2], byteorder='little', signed=False) - rta_type = int.from_bytes(nlmsg[rtattr_indx + 2 : rtattr_indx + 4], byteorder='little', signed=False) - if rta_type == IFLA_ADDRESS: - data = nlmsg[rtattr_indx + RTATTR_SZ : rtattr_indx + rta_len] - if _data_matches_ip(ifa_family, data, src_addr): - return socket.if_indextoname(ifa_index) + if rta_type == IFLA_IFNAME: + data = nlmsg[rtattr_indx + RTATTR_SZ : rtattr_indx + rta_len] + ifname = data.rstrip(b'\0').decode() + interfaces[ifa_index]['name'] = ifname - rta_len = (rta_len + 3) & ~3 # Round up to multiple of 4 - rtattr_indx += rta_len # Move to next rtattr + elif rta_type == IFLA_ADDRESS: + data = nlmsg[rtattr_indx + RTATTR_SZ : rtattr_indx + rta_len] + ip = get_ipaddress_obj(data) + if ip: + family = 4 if ifa_family == socket.AF_INET else 6 + interfaces[ifa_index][family].append(ip) + + rta_len = RTA_ALIGN(rta_len) # Round up to multiple of 4 + rtattr_indx += rta_len # Move to next rtattr nlmsg_idx += nlmsg_len # Move to next Netlink message - return '' + if_addrs = {} + for value in interfaces.values(): + name = value.pop('name', None) + if name is not None: + if_addrs[name] = value + + return if_addrs # ****************************************************************************** @@ -128,42 +305,4 @@ def get_interface(src_addr): src_addr = src_addr.split('%')[0] # remove scope-id (if any) src_addr = get_ipaddress_obj(src_addr) - return '' if src_addr is None else iface_of(src_addr) - - -# ****************************************************************************** -def remove_invalid_addresses(controllers: list): - '''@brief Remove controllers with invalid addresses from the list of controllers. - @param controllers: List of TIDs - ''' - service_conf = conf.SvcConf() - valid_controllers = list() - for controller in controllers: - if controller.transport in ('tcp', 'rdma'): - # Let's make sure that traddr is - # syntactically a valid IPv4 or IPv6 address. - ip = get_ipaddress_obj(controller.traddr) - if ip is None: - logging.warning('%s IP address is not valid', controller) - continue - - # Let's make sure the address family is enabled. - if ip.version not in service_conf.ip_family: - logging.debug( - '%s ignored because IPv%s is disabled in %s', - controller, - ip.version, - service_conf.conf_file, - ) - continue - - valid_controllers.append(controller) - - elif controller.transport in ('fc', 'loop'): - # At some point, need to validate FC addresses as well... - valid_controllers.append(controller) - - else: - logging.warning('Invalid transport %s', controller.transport) - - return valid_controllers + return '' if src_addr is None else _iface_of(src_addr) diff --git a/staslib/meson.build b/staslib/meson.build index eb006f0..1658d3c 100644 --- a/staslib/meson.build +++ b/staslib/meson.build @@ -24,6 +24,7 @@ files_to_copy = [ 'gutil.py', 'iputil.py', 'log.py', + 'nbft.py', 'service.py', 'singleton.py', 'stas.py', diff --git a/staslib/nbft.py b/staslib/nbft.py new file mode 100644 index 0000000..c643c9c --- /dev/null +++ b/staslib/nbft.py @@ -0,0 +1,28 @@ +# Copyright (c) 2023, Dell Inc. or its subsidiaries. All rights reserved. +# SPDX-License-Identifier: Apache-2.0 +# See the LICENSE file for details. +# +# This file is part of NVMe STorage Appliance Services (nvme-stas). +# +# Authors: Martin Belanger <Martin.Belanger@dell.com> +# +'''Code used to access the NVMe Boot Firmware Tables''' + +import os +import glob +import logging +from libnvme import nvme +from staslib import defs + + +def get_nbft_files(root_dir=defs.NBFT_SYSFS_PATH): + """Return a dictionary containing the NBFT data for all the NBFT binary files located in @root_dir""" + if not defs.HAS_NBFT_SUPPORT: + logging.warning( + "libnvme-%s does not have NBFT support. Please upgrade libnvme.", + defs.LIBNVME_VERSION, + ) + return {} + + pathname = os.path.join(root_dir, defs.NBFT_SYSFS_FILENAME) + return {fname: nvme.nbft_get(fname) or {} for fname in glob.iglob(pathname=pathname)} # pylint: disable=no-member diff --git a/staslib/service.py b/staslib/service.py index ce4769d..e681f3a 100644 --- a/staslib/service.py +++ b/staslib/service.py @@ -20,7 +20,7 @@ import dasbus.client.proxy from gi.repository import GLib from systemd.daemon import notify as sd_notify -from staslib import avahi, conf, ctrl, defs, gutil, iputil, stas, timeparse, trid, udev +from staslib import avahi, conf, ctrl, defs, gutil, stas, timeparse, trid, udev # ****************************************************************************** @@ -402,7 +402,7 @@ class Stac(Service): logging.debug('Stac._config_ctrls_finish() - discovered_ctrl_list = %s', discovered_ctrl_list) controllers = stas.remove_excluded(configured_ctrl_list + discovered_ctrl_list) - controllers = iputil.remove_invalid_addresses(controllers) + controllers = stas.remove_invalid_addresses(controllers) new_controller_tids = set(controllers) cur_controller_tids = set(self._controllers.keys()) @@ -752,7 +752,7 @@ class Staf(Service): all_ctrls = configured_ctrl_list + discovered_ctrl_list + referral_ctrl_list controllers = stas.remove_excluded(all_ctrls) - controllers = iputil.remove_invalid_addresses(controllers) + controllers = stas.remove_invalid_addresses(controllers) new_controller_tids = set(controllers) cur_controller_tids = set(self._controllers.keys()) @@ -856,7 +856,7 @@ class Staf(Service): return # Did we receive a Change of DLP AEN or an NVME Event indicating 'connect' or 'rediscover'? - if not _is_dlp_changed_aen(udev_obj) and not _event_matches(udev_obj, ('connected', 'rediscover')): + if not _is_dlp_changed_aen(udev_obj) and not _event_matches(udev_obj, ('rediscover',)): return # We need to invoke "nvme connect-all" using nvme-cli's nvmf-connect@.service @@ -873,6 +873,6 @@ class Staf(Service): options = r'\x09'.join( [fr'{option}\x3d{value}' for option, value in cnf if value not in (None, 'none', 'None', '')] ) - logging.info('Invoking: systemctl start nvmf-connect@%s.service', options) - cmd = [defs.SYSTEMCTL, '--quiet', '--no-block', 'start', fr'nvmf-connect@{options}.service'] + logging.debug('Invoking: systemctl restart nvmf-connect@%s.service', options) + cmd = [defs.SYSTEMCTL, '--quiet', '--no-block', 'restart', fr'nvmf-connect@{options}.service'] subprocess.run(cmd, check=False) diff --git a/staslib/singleton.py b/staslib/singleton.py index 2171186..f80fc0b 100644 --- a/staslib/singleton.py +++ b/staslib/singleton.py @@ -21,3 +21,28 @@ class Singleton(type): instance = super(Singleton, cls).__call__(*args, **kwargs) cls._instances[cls] = instance return cls._instances[cls] + + def destroy(cls): + '''Delete a singleton instance. + + This is to be invoked using the derived class Name. + + For example: + + class Child(Singleton): + pass + + child1 = Child() # Instantiate singleton + child2 = Child() # Get a reference to the singleton + + print(f'{child1 is child2}') # True + + Child.destroy() # Delete the singleton + + print(f'{child1 is child2}') # Still True because child1 and child2 still hold reference to the singleton + + child1 = Child() # Instantiate a new singleton and assign to child1 + + print(f'{child1 is child2}') # False + ''' + cls._instances.pop(cls, None) diff --git a/staslib/stas.py b/staslib/stas.py index 95afb94..e333b90 100644 --- a/staslib/stas.py +++ b/staslib/stas.py @@ -18,7 +18,7 @@ import logging import dasbus.connection from gi.repository import Gio, GLib from systemd.daemon import notify as sd_notify -from staslib import conf, defs, gutil, log, trid +from staslib import conf, defs, gutil, iputil, log, trid try: # Python 3.9 or later @@ -87,6 +87,44 @@ def check_if_allowed_to_continue(): # ****************************************************************************** +def remove_invalid_addresses(controllers: list): + '''@brief Remove controllers with invalid addresses from the list of controllers. + @param controllers: List of TIDs + ''' + service_conf = conf.SvcConf() + valid_controllers = list() + for controller in controllers: + if controller.transport in ('tcp', 'rdma'): + # Let's make sure that traddr is + # syntactically a valid IPv4 or IPv6 address. + ip = iputil.get_ipaddress_obj(controller.traddr) + if ip is None: + logging.warning('%s IP address is not valid', controller) + continue + + # Let's make sure the address family is enabled. + if ip.version not in service_conf.ip_family: + logging.debug( + '%s ignored because IPv%s is disabled in %s', + controller, + ip.version, + service_conf.conf_file, + ) + continue + + valid_controllers.append(controller) + + elif controller.transport in ('fc', 'loop'): + # At some point, need to validate FC addresses as well... + valid_controllers.append(controller) + + else: + logging.warning('Invalid transport %s', controller.transport) + + return valid_controllers + + +# ****************************************************************************** def tid_from_dlpe(dlpe, host_traddr, host_iface): '''@brief Take a Discovery Log Page Entry and return a Controller ID as a dict.''' cid = { @@ -492,7 +530,9 @@ class ServiceABC(abc.ABC): # pylint: disable=too-many-instance-attributes # elements after name resolution is complete (i.e. in the calback # function _config_ctrls_finish) logging.debug('ServiceABC._config_ctrls()') - configured_controllers = [trid.TID(cid) for cid in conf.SvcConf().get_controllers()] + configured_controllers = [ + trid.TID(cid) for cid in conf.SvcConf().get_controllers() + conf.NbftConf().get_controllers() + ] configured_controllers = remove_excluded(configured_controllers) self._resolver.resolve_ctrl_async(self._cancellable, configured_controllers, self._config_ctrls_finish) diff --git a/staslib/trid.py b/staslib/trid.py index d9a05ba..11065ea 100644 --- a/staslib/trid.py +++ b/staslib/trid.py @@ -61,7 +61,6 @@ class TID: # pylint: disable=too-many-instance-attributes self._host_traddr = cid.get('host-traddr', '') self._host_iface = '' if conf.SvcConf().ignore_iface else cid.get('host-iface', '') self._subsysnqn = cid.get('subsysnqn', '') - self._shortkey = (self._transport, self._traddr, self._trsvcid, self._subsysnqn, self._host_traddr) self._key = (self._transport, self._traddr, self._trsvcid, self._subsysnqn, self._host_traddr, self._host_iface) self._hash = int.from_bytes( hashlib.md5(''.join(self._key).encode('utf-8')).digest(), 'big' @@ -121,22 +120,10 @@ class TID: # pylint: disable=too-many-instance-attributes return self._id def __eq__(self, other): - if not isinstance(other, self.__class__): - return False - - if self._host_iface and other._host_iface: - return self._key == other._key - - return self._shortkey == other._shortkey + return isinstance(other, self.__class__) and self._key == other._key def __ne__(self, other): - if not isinstance(other, self.__class__): - return True - - if self._host_iface and other._host_iface: - return self._key != other._key - - return self._shortkey != other._shortkey + return not isinstance(other, self.__class__) or self._key != other._key def __hash__(self): return self._hash diff --git a/staslib/udev.py b/staslib/udev.py index 12ef61b..80555dd 100644 --- a/staslib/udev.py +++ b/staslib/udev.py @@ -153,6 +153,169 @@ class Udev: return False + @staticmethod + def _cid_matches_tcp_tid_legacy(tid, cid): # pylint: disable=too-many-return-statements,too-many-branches + '''On kernels older than 6.1, the src_addr parameter is not available + from the sysfs. Therefore, we need to infer a match based on other + parameters. And there are a few cases where we're simply not sure + whether an existing connection (cid) matches the candidate + connection (tid). + ''' + cid_host_iface = cid['host-iface'] + cid_host_traddr = iputil.get_ipaddress_obj(cid['host-traddr'], ipv4_mapped_convert=True) + + if not cid_host_iface: # cid.host_iface is undefined + if not cid_host_traddr: # cid.host_traddr is undefined + # When the existing cid.src_addr, cid.host_traddr, and cid.host_iface + # are all undefined (which can only happen on kernels prior to 6.1), + # we can't know for sure on which interface an existing connection + # was made. In this case, we can only declare a match if both + # tid.host_iface and tid.host_traddr are undefined as well. + logging.debug( + 'Udev._cid_matches_tcp_tid_legacy() - cid=%s, tid=%s - Not enough info. Assume "match" but this could be wrong.', + cid, + tid, + ) + return True + + # cid.host_traddr is defined. If tid.host_traddr is also + # defined, then it must match the existing cid.host_traddr. + if tid.host_traddr: + tid_host_traddr = iputil.get_ipaddress_obj(tid.host_traddr, ipv4_mapped_convert=True) + if tid_host_traddr != cid_host_traddr: + return False + + # If tid.host_iface is defined, then the interface where + # the connection is located must match. If tid.host_iface + # is not defined, then we don't really care on which + # interface the connection was made and we can skip this test. + if tid.host_iface: + # With the existing cid.host_traddr, we can find the + # interface of the exisiting connection. + connection_iface = iputil.get_interface(str(cid_host_traddr)) + if tid.host_iface != connection_iface: + return False + + return True + + # cid.host_iface is defined + if not cid_host_traddr: # cid.host_traddr is undefined + if tid.host_iface and tid.host_iface != cid_host_iface: + return False + + if tid.host_traddr: + # It's impossible to tell the existing connection source + # address. So, we can't tell if it matches tid.host_traddr. + # However, if the existing host_iface has only one source + # address assigned to it, we can assume that the source + # address used for the existing connection is that address. + if_addrs = iputil.net_if_addrs().get(cid_host_iface, {4: [], 6: []}) + tid_host_traddr = iputil.get_ipaddress_obj(tid.host_traddr, ipv4_mapped_convert=True) + source_addrs = if_addrs[tid_host_traddr.version] + if len(source_addrs) != 1: + return False + + src_addr0 = iputil.get_ipaddress_obj(source_addrs[0], ipv4_mapped_convert=True) + if src_addr0 != tid_host_traddr: + return False + + return True + + # cid.host_traddr is defined + if tid.host_iface and tid.host_iface != cid_host_iface: + return False + + if tid.host_traddr: + tid_host_traddr = iputil.get_ipaddress_obj(tid.host_traddr, ipv4_mapped_convert=True) + if tid_host_traddr != cid_host_traddr: + return False + + return True + + @staticmethod + def _cid_matches_tid(tid, cid): # pylint: disable=too-many-return-statements,too-many-branches + '''Check if existing controller's cid matches candidate controller's tid. + @param cid: The Connection ID of an existing controller (from the sysfs). + @param tid: The Transport ID of a candidate controller. + + We're trying to find if an existing connection (specified by cid) can + be re-used for the candidate controller (specified by tid). + + We do not have a match if the candidate's tid.transport, tid.traddr, + tid.trsvcid, and tid.subsysnqn are not identical to those of the cid. + These 4 parameters are mandatory for a match. + + The tid.host_traddr and tid.host_iface depend on the transport type. + These parameters may not apply or have a different syntax/meaning + depending on the transport type. + + For TCP only: + With regards to the candidate's tid.host_traddr and tid.host_iface, + if those are defined but do not match the existing cid.host_traddr + and cid.host_iface, we may still be able to find a match by taking + the existing cid.src_addr into consideration since that parameter + identifies the actual source address of the connection and therefore + can be used to infer the interface of the connection. However, the + cid.src_addr can only be read from the sysfs starting with kernel + 6.1. + ''' + # 'transport', 'traddr', 'trsvcid', and 'subsysnqn' must exactly match. + if cid['transport'] != tid.transport or cid['trsvcid'] != tid.trsvcid or cid['subsysnqn'] != tid.subsysnqn: + return False + + if tid.transport in ('tcp', 'rdma'): + # Need to convert to ipaddress objects to properly + # handle all variations of IPv6 addresses. + tid_traddr = iputil.get_ipaddress_obj(tid.traddr, ipv4_mapped_convert=True) + cid_traddr = iputil.get_ipaddress_obj(cid['traddr'], ipv4_mapped_convert=True) + else: + cid_traddr = cid['traddr'] + tid_traddr = tid.traddr + + if cid_traddr != tid_traddr: + return False + + # We need to know the type of transport to compare 'host-traddr' and + # 'host-iface'. These parameters don't apply to all transport types + # and may have a different meaning/syntax. + if tid.transport == 'tcp': + if tid.host_traddr or tid.host_iface: + src_addr = iputil.get_ipaddress_obj(cid['src-addr'], ipv4_mapped_convert=True) + if not src_addr: + # For legacy kernels (i.e. older than 6.1), the existing cid.src_addr + # is always undefined. We need to use advanced logic to determine + # whether cid and tid match. + return Udev._cid_matches_tcp_tid_legacy(tid, cid) + + # The existing controller's cid.src_addr is always defined for kernel + # 6.1 and later. We can use the existing controller's cid.src_addr to + # find the interface on which the connection was made and therefore + # match it to the candidate's tid.host_iface. And the cid.src_addr + # can also be used to match the candidate's tid.host_traddr. + if tid.host_traddr: + tid_host_traddr = iputil.get_ipaddress_obj(tid.host_traddr, ipv4_mapped_convert=True) + if tid_host_traddr != src_addr: + return False + + # host-iface is an optional tcp-only parameter. + if tid.host_iface and tid.host_iface != iputil.get_interface(str(src_addr)): + return False + + elif tid.transport == 'fc': + # host-traddr is mandatory for FC. + if tid.host_traddr != cid['host-traddr']: + return False + + elif tid.transport == 'rdma': + # host-traddr is optional for RDMA and is expressed as an IP address. + if tid.host_traddr: + tid_host_traddr = iputil.get_ipaddress_obj(tid.host_traddr, ipv4_mapped_convert=True) + cid_host_traddr = iputil.get_ipaddress_obj(cid['host-traddr'], ipv4_mapped_convert=True) + if tid_host_traddr != cid_host_traddr: + return False + + return True + def find_nvme_dc_device(self, tid): '''@brief Find the nvme device associated with the specified Discovery Controller. @@ -164,7 +327,8 @@ class Udev: if not self.is_dc_device(device): continue - if self.get_tid(device) != tid: + cid = self.get_cid(device) + if not self._cid_matches_tid(tid, cid): continue return device @@ -182,7 +346,8 @@ class Udev: if not self.is_ioc_device(device): continue - if self.get_tid(device) != tid: + cid = self.get_cid(device) + if not self._cid_matches_tid(tid, cid): continue return device @@ -300,28 +465,32 @@ class Udev: return attr_str[start:end] @staticmethod - def _get_host_iface(device): - host_iface = Udev._get_property(device, 'NVME_HOST_IFACE') - if not host_iface: - # We'll try to find the interface from the source address on - # the connection. Only available if kernel exposes the source - # address (src_addr) in the "address" attribute. - src_addr = Udev.get_key_from_attr(device, 'address', 'src_addr=') - host_iface = iputil.get_interface(src_addr) - return host_iface - - @staticmethod def get_tid(device): '''@brief return the Transport ID associated with a udev device''' + cid = Udev.get_cid(device) + if cid['transport'] == 'tcp': + src_addr = cid['src-addr'] + if not cid['host-iface'] and src_addr: + # We'll try to find the interface from the source address on + # the connection. Only available if kernel exposes the source + # address (src_addr) in the "address" attribute. + cid['host-iface'] = iputil.get_interface(src_addr) + + return trid.TID(cid) + + @staticmethod + def get_cid(device): + '''@brief return the Connection ID associated with a udev device''' cid = { 'transport': Udev._get_property(device, 'NVME_TRTYPE'), 'traddr': Udev._get_property(device, 'NVME_TRADDR'), 'trsvcid': Udev._get_property(device, 'NVME_TRSVCID'), 'host-traddr': Udev._get_property(device, 'NVME_HOST_TRADDR'), - 'host-iface': Udev._get_host_iface(device), + 'host-iface': Udev._get_property(device, 'NVME_HOST_IFACE'), 'subsysnqn': Udev._get_attribute(device, 'subsysnqn'), + 'src-addr': Udev.get_key_from_attr(device, 'address', 'src_addr='), } - return trid.TID(cid) + return cid UDEV = Udev() # Singleton |