Update Grub on component drives if /boot is on md device

On BIOS systems, previously, if /boot was on md device such as RAID
consisting of multiple partitions on different MBR/GPT partitioned
drives, the part of Grub residing in the 512 Mb after MBR was only
updated for one of the drives. Similar situation occurred on GPT
partitioned drives and the BIOS boot partition. This resulted in
outdated GRUB on the remaining drives which could cause the system to be
unbootable.

Now, Grub is updated on all the component devices of an md array if Grub
was already installed on them before the upgrade.

Jira: OAMG-7835
BZ#2219544
BZ#2140011

repos/system_upgrade/common/actors/checkgrubcore/actor.py

@@ -32,7 +32,7 @@ def process(self):
             grub_info = next(self.consume(GrubInfo), None)
             if not grub_info:
                 raise StopActorExecutionError('Actor did not receive any GrubInfo message.')
-            if grub_info.orig_device_name:
+            if grub_info.orig_devices:
                 create_report([
                     reporting.Title(
                         'GRUB2 core will be automatically updated during the upgrade'
@@ -45,8 +45,9 @@ def process(self):
                 create_report([
                     reporting.Title('Leapp could not identify where GRUB2 core is located'),
                     reporting.Summary(
-                        'We assumed GRUB2 core is located on the same device as /boot, however Leapp could not '
-                        'detect GRUB2 on the device. GRUB2 core needs to be updated maually on legacy (BIOS) systems. '
+                        'We assumed GRUB2 core is located on the same device(s) as /boot, '
+                        'however Leapp could not detect GRUB2 on the device(s). '
+                        'GRUB2 core needs to be updated maually on legacy (BIOS) systems. '
                     ),
                     reporting.Severity(reporting.Severity.HIGH),
                     reporting.Groups([reporting.Groups.BOOT]),

repos/system_upgrade/common/actors/checkgrubcore/tests/test_checkgrubcore.py

@@ -1,18 +1,17 @@
-import pytest
-
-from leapp.exceptions import StopActorExecutionError
 from leapp.libraries.common.config import mock_configs
 from leapp.models import FirmwareFacts, GrubInfo
 from leapp.reporting import Report
 
 NO_GRUB = 'Leapp could not identify where GRUB2 core is located'
+GRUB = 'GRUB2 core will be automatically updated during the upgrade'
 
 
 def test_actor_update_grub(current_actor_context):
     current_actor_context.feed(FirmwareFacts(firmware='bios'))
-    current_actor_context.feed(GrubInfo(orig_device_name='/dev/vda'))
+    current_actor_context.feed(GrubInfo(orig_devices=['/dev/vda', '/dev/vdb']))
     current_actor_context.run(config_model=mock_configs.CONFIG)
     assert current_actor_context.consume(Report)
+    assert current_actor_context.consume(Report)[0].report['title'].startswith(GRUB)
 
 
 def test_actor_no_grub_device(current_actor_context):
@@ -31,6 +30,6 @@ def test_actor_with_efi(current_actor_context):
 
 def test_s390x(current_actor_context):
     current_actor_context.feed(FirmwareFacts(firmware='bios'))
-    current_actor_context.feed(GrubInfo(orig_device_name='/dev/vda'))
+    current_actor_context.feed(GrubInfo(orig_devices=['/dev/vda', '/dev/vdb']))
     current_actor_context.run(config_model=mock_configs.CONFIG_S390X)
     assert not current_actor_context.consume(Report)

repos/system_upgrade/common/actors/scangrubdevice/actor.py

@@ -7,7 +7,7 @@
 
 class ScanGrubDeviceName(Actor):
     """
-    Find the name of the block device where GRUB is located
+    Find the name of the block devices where GRUB is located
     """
 
     name = 'scan_grub_device_name'
@@ -19,8 +19,7 @@ def process(self):
         if architecture.matches_architecture(architecture.ARCH_S390X):
             return
 
-        device_name = grub.get_grub_device()
-        if device_name:
-            self.produce(GrubInfo(orig_device_name=device_name))
-        else:
-            self.produce(GrubInfo())
+        devices = grub.get_grub_devices()
+        grub_info = GrubInfo(orig_devices=devices)
+        grub_info.orig_device_name = devices[0] if len(devices) == 1 else None
+        self.produce(grub_info)

repos/system_upgrade/common/actors/scangrubdevice/tests/test_scangrubdevice.py

@@ -0,0 +1,35 @@
+from leapp.libraries.common import grub
+from leapp.libraries.common.config import mock_configs
+from leapp.models import GrubInfo
+
+
+def _get_grub_devices_mocked():
+    return ['/dev/vda', '/dev/vdb']
+
+
+def test_actor_scan_grub_device(current_actor_context, monkeypatch):
+    monkeypatch.setattr(grub, 'get_grub_devices', _get_grub_devices_mocked)
+    current_actor_context.run(config_model=mock_configs.CONFIG)
+    info = current_actor_context.consume(GrubInfo)
+    assert info and info[0].orig_devices == ['/dev/vda', '/dev/vdb']
+    assert len(info) == 1, 'Expected just one GrubInfo message'
+    assert not info[0].orig_device_name
+
+
+def test_actor_scan_grub_device_one(current_actor_context, monkeypatch):
+
+    def _get_grub_devices_mocked():
+        return ['/dev/vda']
+
+    monkeypatch.setattr(grub, 'get_grub_devices', _get_grub_devices_mocked)
+    current_actor_context.run(config_model=mock_configs.CONFIG)
+    info = current_actor_context.consume(GrubInfo)
+    assert info and info[0].orig_devices == ['/dev/vda']
+    assert len(info) == 1, 'Expected just one GrubInfo message'
+    assert info[0].orig_device_name == '/dev/vda'
+
+
+def test_actor_scan_grub_device_s390x(current_actor_context, monkeypatch):
+    monkeypatch.setattr(grub, 'get_grub_devices', _get_grub_devices_mocked)
+    current_actor_context.run(config_model=mock_configs.CONFIG_S390X)
+    assert not current_actor_context.consume(GrubInfo)

repos/system_upgrade/common/actors/updategrubcore/actor.py

@@ -21,8 +21,8 @@ class UpdateGrubCore(Actor):
     def process(self):
         ff = next(self.consume(FirmwareFacts), None)
         if ff and ff.firmware == 'bios':
-            grub_dev = grub.get_grub_device()
-            if grub_dev:
-                update_grub_core(grub_dev)
+            grub_devs = grub.get_grub_devices()
+            if grub_devs:
+                update_grub_core(grub_devs)
             else:
-                api.current_logger().warning('Leapp could not detect GRUB on {}'.format(grub_dev))
+                api.current_logger().warning('Leapp could not detect GRUB devices')

repos/system_upgrade/common/actors/updategrubcore/libraries/updategrubcore.py

@@ -1,35 +1,43 @@
 from leapp import reporting
-from leapp.exceptions import StopActorExecution
 from leapp.libraries.stdlib import api, CalledProcessError, config, run
 
 
-def update_grub_core(grub_dev):
+def update_grub_core(grub_devs):
     """
     Update GRUB core after upgrade from RHEL7 to RHEL8
 
     On legacy systems, GRUB core does not get automatically updated when GRUB packages
     are updated.
     """
-    cmd = ['grub2-install', grub_dev]
-    if config.is_debug():
-        cmd += ['-v']
-    try:
-        run(cmd)
-    except CalledProcessError as err:
-        reporting.create_report([
-            reporting.Title('GRUB core update failed'),
-            reporting.Summary(str(err)),
-            reporting.Groups([reporting.Groups.BOOT]),
-            reporting.Severity(reporting.Severity.HIGH),
-            reporting.Remediation(
-                hint='Please run "grub2-install <GRUB_DEVICE>" manually after upgrade'
-            )
-        ])
-        api.current_logger().warning('GRUB core update on {} failed'.format(grub_dev))
-        raise StopActorExecution()
+
+    successful = []
+    failed = []
+    for dev in grub_devs:
+        cmd = ['grub2-install', dev]
+        if config.is_debug():
+            cmd += ['-v']
+        try:
+            run(cmd)
+        except CalledProcessError as err:
+            api.current_logger().warning('GRUB core update on {} failed: {}'.format(dev, err))
+            failed.append(dev)
+            continue
+
+        successful.append(dev)
+
+    reporting.create_report([
+        reporting.Title('GRUB core update failed'),
+        reporting.Summary('Leapp failed to update GRUB on {}'.format(', '.join(failed))),
+        reporting.Groups([reporting.Groups.BOOT]),
+        reporting.Severity(reporting.Severity.HIGH),
+        reporting.Remediation(
+            hint='Please run "grub2-install <GRUB_DEVICE>" manually after upgrade'
+        )
+    ])
+
     reporting.create_report([
         reporting.Title('GRUB core successfully updated'),
-        reporting.Summary('GRUB core on {} was successfully updated'.format(grub_dev)),
+        reporting.Summary('GRUB core on {} was successfully updated'.format(', '.join(successful))),
         reporting.Groups([reporting.Groups.BOOT]),
         reporting.Severity(reporting.Severity.INFO)
     ])

repos/system_upgrade/common/actors/updategrubcore/tests/test_updategrubcore.py

@@ -1,7 +1,6 @@
 import pytest
 
 from leapp import reporting
-from leapp.exceptions import StopActorExecution
 from leapp.libraries.actor import updategrubcore
 from leapp.libraries.common import testutils
 from leapp.libraries.stdlib import api, CalledProcessError
@@ -32,21 +31,45 @@ def __call__(self, *args):
             raise_call_error(args)
 
 
-def test_update_grub(monkeypatch):
+@pytest.mark.parametrize('devices', [['/dev/vda'], ['/dev/vda', '/dev/vdb']])
+def test_update_grub(monkeypatch, devices):
     monkeypatch.setattr(reporting, "create_report", testutils.create_report_mocked())
     monkeypatch.setattr(updategrubcore, 'run', run_mocked())
-    updategrubcore.update_grub_core('/dev/vda')
+    updategrubcore.update_grub_core(devices)
     assert reporting.create_report.called
-    assert UPDATE_OK_TITLE == reporting.create_report.report_fields['title']
+    assert UPDATE_OK_TITLE == reporting.create_report.reports[1]['title']
+    assert all(dev in reporting.create_report.reports[1]['summary'] for dev in devices)
 
 
-def test_update_grub_failed(monkeypatch):
+@pytest.mark.parametrize('devices', [['/dev/vda'], ['/dev/vda', '/dev/vdb']])
+def test_update_grub_failed(monkeypatch, devices):
     monkeypatch.setattr(reporting, "create_report", testutils.create_report_mocked())
     monkeypatch.setattr(updategrubcore, 'run', run_mocked(raise_err=True))
-    with pytest.raises(StopActorExecution):
-        updategrubcore.update_grub_core('/dev/vda')
+    updategrubcore.update_grub_core(devices)
     assert reporting.create_report.called
-    assert UPDATE_FAILED_TITLE == reporting.create_report.report_fields['title']
+    assert UPDATE_FAILED_TITLE == reporting.create_report.reports[0]['title']
+    assert all(dev in reporting.create_report.reports[0]['summary'] for dev in devices)
+
+
+def test_update_grub_success_and_fail(monkeypatch):
+    monkeypatch.setattr(reporting, "create_report", testutils.create_report_mocked())
+
+    def run_mocked(args):
+        if args == ['grub2-install', '/dev/vdb']:
+            raise_call_error(args)
+        else:
+            assert args == ['grub2-install', '/dev/vda']
+
+    monkeypatch.setattr(updategrubcore, 'run', run_mocked)
+
+    devices = ['/dev/vda', '/dev/vdb']
+    updategrubcore.update_grub_core(devices)
+
+    assert reporting.create_report.called
+    assert UPDATE_FAILED_TITLE == reporting.create_report.reports[0]['title']
+    assert '/dev/vdb' in reporting.create_report.reports[0]['summary']
+    assert UPDATE_OK_TITLE == reporting.create_report.reports[1]['title']
+    assert '/dev/vda' in reporting.create_report.reports[1]['summary']
 
 
 def test_update_grub_negative(current_actor_context):

repos/system_upgrade/common/libraries/grub.py

@@ -1,7 +1,9 @@
 import os
 
 from leapp.exceptions import StopActorExecution
+from leapp.libraries.common import mdraid
 from leapp.libraries.stdlib import api, CalledProcessError, run
+from leapp.utils.deprecation import deprecated
 
 
 def has_grub(blk_dev):
@@ -59,6 +61,32 @@ def get_boot_partition():
     return boot_partition
 
 
+def get_grub_devices():
+    """
+    Get block devices where GRUB is located. We assume GRUB is on the same device
+    as /boot partition is. In case that device is an md (Multiple Device) device, all
+    of the component devices of such a device are considered.
+
+    :return: Devices where GRUB is located
+    :rtype: list
+    """
+    boot_device = get_boot_partition()
+    devices = []
+    if mdraid.is_mdraid_dev(boot_device):
+        component_devs = mdraid.get_component_devices(boot_device)
+        blk_devs = [blk_dev_from_partition(dev) for dev in component_devs]
+        # remove duplicates as there might be raid on partitions on the same drive
+        # even if that's very unusual
+        devices = sorted(list(set(blk_devs)))
+    else:
+        devices.append(blk_dev_from_partition(boot_device))
+
+    have_grub = [dev for dev in devices if has_grub(dev)]
+    api.current_logger().info('GRUB is installed on {}'.format(",".join(have_grub)))
+    return have_grub
+
+
+@deprecated(since='2023-06-23', message='This function has been replaced by get_grub_devices')
 def get_grub_device():
     """
     Get block device where GRUB is located. We assume GRUB is on the same device

repos/system_upgrade/common/libraries/mdraid.py

@@ -0,0 +1,48 @@
+from leapp.libraries.stdlib import api, CalledProcessError, run
+
+
+def is_mdraid_dev(dev):
+    """
+    Check if a given device is an md (Multiple Device) device
+
+    It is expected that the "mdadm" command is available,
+    if it's not it is assumed the device is not an md device.
+
+    :return: True if the device is an md device, False otherwise
+    :raises CalledProcessError: If an error occurred
+    """
+    fail_msg = 'Could not check if device "{}" is an md device: {}'
+    try:
+        result = run(['mdadm', '--query', dev])
+    except OSError as err:
+        api.current_logger().warning(fail_msg.format(dev, err))
+        return False
+    except CalledProcessError as err:
+        err.message = fail_msg.format(dev, err)
+        raise  # let the calling actor handle the exception
+
+    return '--detail' in result['stdout']
+
+
+def get_component_devices(raid_dev):
+    """
+    Get list of component devices in an md (Multiple Device) array
+
+    :return: The list of component devices or None in case of error
+    :raises ValueError: If the device is not an mdraid device
+    """
+    try:
+        # using both --verbose and --brief for medium verbosity
+        result = run(['mdadm', '--detail', '--verbose', '--brief', raid_dev])
+    except (OSError, CalledProcessError) as err:
+        api.current_logger().warning(
+            'Could not get md array component devices: {}'.format(err)
+        )
+        return None
+    # example output:
+    # ARRAY /dev/md0 level=raid1 num-devices=2 metadata=1.2 name=localhost.localdomain:0 UUID=c4acea6e:d56e1598:91822e3f:fb26832c # noqa: E501; pylint: disable=line-too-long
+    #     devices=/dev/vda1,/dev/vdb1
+    if 'does not appear to be an md device' in result['stdout']:
+        raise ValueError("Expected md device, but got: {}".format(raid_dev))
+
+    return sorted(result['stdout'].rsplit('=', 2)[-1].strip().split(','))