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

Previously, if /boot was on md device such as RAID consisting of
multiple partitions on different drives, the part of Grub residing in
the 512 Mb after MBR was only updated for one of the drives. This
resulted in broken Grub.

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

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/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 +19,5 @@ 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()
+        self.produce(GrubInfo(orig_devices=devices))

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

@@ -3,33 +3,39 @@
 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 = []
+    for dev in grub_devs:
+        cmd = ['grub2-install', 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(dev))
+            raise StopActorExecution()
+
+        successful.append(dev)
+
     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/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 = 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 +1,41 @@
+from leapp.exceptions import StopActorExecution
+from leapp.libraries.stdlib import api, CalledProcessError, run
+
+
+def is_mdraid_dev(dev):
+    """
+    Check if a given device is an md (Multiple Device) device
+
+    :raises: StopActorExecution in case of error
+    """
+    try:
+        result = run(['mdadm', '--query', dev])
+    except CalledProcessError as err:
+        api.current_logger().warning(
+                'Could not check if device is an md device: {}'.format(err)
+        )
+        raise StopActorExecution()
+    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
+    """
+    try:
+        # using both --verbose and --brief for medium verbosity
+        result = run(['mdadm', '--detail', '--verbose', '--brief', raid_dev])
+    except 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 result['stdout'].rsplit('=', 2)[-1].strip().split(',')

repos/system_upgrade/common/libraries/tests/test_grub.py

@@ -3,14 +3,17 @@
 import pytest
 
 from leapp.exceptions import StopActorExecution
-from leapp.libraries.common import grub
+from leapp.libraries.common import grub, mdraid
 from leapp.libraries.common.testutils import logger_mocked
 from leapp.libraries.stdlib import api, CalledProcessError
 from leapp.models import DefaultGrub, DefaultGrubInfo
 
 BOOT_PARTITION = '/dev/vda1'
 BOOT_DEVICE = '/dev/vda'
 
+MD_BOOT_DEVICE = '/dev/md0'
+MD_BOOT_REAL_DEVICES = ['/dev/sda', '/dev/sdb']
+
 VALID_DD = b'GRUB GeomHard DiskRead Error'
 INVALID_DD = b'Nothing to see here!'
 
@@ -27,10 +30,11 @@ def raise_call_error(args=None):
 
 class RunMocked(object):
 
-    def __init__(self, raise_err=False):
+    def __init__(self, raise_err=False, boot_on_raid=False):
         self.called = 0
         self.args = None
         self.raise_err = raise_err
+        self.boot_on_raid = boot_on_raid
 
     def __call__(self, args, encoding=None):
         self.called += 1
@@ -39,18 +43,24 @@ def __call__(self, args, encoding=None):
             raise_call_error(args)
 
         if self.args == ['grub2-probe', '--target=device', '/boot']:
-            stdout = BOOT_PARTITION
+            stdout = MD_BOOT_DEVICE if self.boot_on_raid else BOOT_PARTITION
 
         elif self.args == ['lsblk', '-spnlo', 'name', BOOT_PARTITION]:
             stdout = BOOT_DEVICE
+        elif self.args == ['lsblk', '-spnlo', 'name', '/dev/sda1']:
+            stdout = '/dev/sda'
+        elif self.args == ['lsblk', '-spnlo', 'name', '/dev/sdb1']:
+            stdout = '/dev/sdb'
 
         return {'stdout': stdout}
 
 
 def open_mocked(fn, flags):
-    return open(
-        os.path.join(CUR_DIR, 'grub_valid') if fn == BOOT_DEVICE else os.path.join(CUR_DIR, 'grub_invalid'), 'r'
-    )
+    if fn == BOOT_DEVICE or fn in MD_BOOT_REAL_DEVICES:
+        path = os.path.join(CUR_DIR, 'grub_valid')
+    else:
+        path = os.path.join(CUR_DIR, 'grub_invalid')
+    return open(path, 'r')
 
 
 def open_invalid(fn, flags):
@@ -122,3 +132,45 @@ def test_is_blscfg_library(monkeypatch, enabled):
         assert result
     else:
         assert not result
+
+
+def is_mdraid_dev_mocked(dev):
+    return dev == '/dev/md0'
+
+
+def test_get_grub_devices_one_device(monkeypatch):
+    run_mocked = RunMocked()
+    monkeypatch.setattr(grub, 'run', run_mocked)
+    monkeypatch.setattr(os, 'open', open_mocked)
+    monkeypatch.setattr(os, 'read', read_mocked)
+    monkeypatch.setattr(os, 'close', close_mocked)
+    monkeypatch.setattr(api, 'current_logger', logger_mocked())
+    monkeypatch.setattr(mdraid, 'is_mdraid_dev', is_mdraid_dev_mocked)
+
+    result = grub.get_grub_devices()
+    assert grub.run.called == 2
+    assert [BOOT_DEVICE] == result
+    assert not api.current_logger.warnmsg
+    assert 'GRUB is installed on {}'.format(",".join(result)) in api.current_logger.infomsg
+
+
+def test_get_grub_devices_raid_device(monkeypatch):
+    run_mocked = RunMocked(boot_on_raid=True)
+    monkeypatch.setattr(grub, 'run', run_mocked)
+    monkeypatch.setattr(os, 'open', open_mocked)
+    monkeypatch.setattr(os, 'read', read_mocked)
+    monkeypatch.setattr(os, 'close', close_mocked)
+    monkeypatch.setattr(api, 'current_logger', logger_mocked())
+    monkeypatch.setattr(mdraid, 'is_mdraid_dev', is_mdraid_dev_mocked)
+
+    def get_component_devices_mocked(raid_dev):
+        assert raid_dev == MD_BOOT_DEVICE
+        return ['/dev/sda1', '/dev/sdb1']
+
+    monkeypatch.setattr(mdraid, 'get_component_devices', get_component_devices_mocked)
+
+    result = grub.get_grub_devices()
+    assert grub.run.called == 3  # grub2-probe + 2x lsblk
+    assert ['/dev/sda', '/dev/sdb'] == result
+    assert not api.current_logger.warnmsg
+    assert 'GRUB is installed on {}'.format(",".join(result)) in api.current_logger.infomsg

repos/system_upgrade/common/libraries/tests/test_mdraid.py

@@ -0,0 +1,78 @@
+import os
+
+import pytest
+
+from leapp.libraries.common import mdraid
+from leapp.libraries.common.testutils import logger_mocked
+from leapp.libraries.stdlib import api, CalledProcessError
+
+MD_DEVICE = '/dev/md0'
+NOT_MD_DEVICE = '/dev/sda'
+
+CUR_DIR = os.path.dirname(os.path.abspath(__file__))
+
+
+def raise_call_error(args=None):
+    raise CalledProcessError(
+        message='A Leapp Command Error occurred.',
+        command=args,
+        result={'signal': None, 'exit_code': 1, 'pid': 0, 'stdout': 'fake', 'stderr': 'fake'}
+    )
+
+
+class RunMocked(object):
+
+    def __init__(self, raise_err=False):
+        self.called = 0
+        self.args = None
+        self.raise_err = raise_err
+
+    def __call__(self, args, encoding=None):
+        self.called += 1
+        self.args = args
+        if self.raise_err:
+            raise_call_error(args)
+
+        if self.args == ['mdadm', '--query', MD_DEVICE]:
+            stdout = '/dev/md0: 1022.00MiB raid1 2 devices, 0 spares. Use mdadm --detail for more detail.'
+        elif self.args == ['mdadm', '--query', NOT_MD_DEVICE]:
+            stdout = '/dev/sda: is not an md array'
+
+        elif self.args == ['mdadm', '--detail', '--verbose', '--brief', MD_DEVICE]:
+            stdout = 'ARRAY /dev/md0 level=raid1 num-devices=2 metadata=1.2 name=localhost.localdomain:0 UUID=c4acea6e:d56e1598:91822e3f:fb26832c\n    devices=/dev/sda1,/dev/sdb1'  # noqa: E501; pylint: disable=line-too-long
+        elif self.args == ['mdadm', '--detail', '--verbose', '--brief', NOT_MD_DEVICE]:
+            stdout = 'mdadm: /dev/sda does not appear to be an md device'
+
+        return {'stdout': stdout}
+
+
+@pytest.mark.parametrize('dev,expected', [(MD_DEVICE, True), (NOT_MD_DEVICE, False)])
+def test_is_mdraid_dev(monkeypatch, dev, expected):
+    run_mocked = RunMocked()
+    monkeypatch.setattr(mdraid, 'run', run_mocked)
+    monkeypatch.setattr(api, 'current_logger', logger_mocked())
+
+    result = mdraid.is_mdraid_dev(dev)
+    assert mdraid.run.called == 1
+    assert expected == result
+    assert not api.current_logger.warnmsg
+
+
+def test_get_component_devices_ok(monkeypatch):
+    run_mocked = RunMocked()
+    monkeypatch.setattr(mdraid, 'run', run_mocked)
+    monkeypatch.setattr(api, 'current_logger', logger_mocked())
+
+    result = mdraid.get_component_devices(MD_DEVICE)
+    assert mdraid.run.called == 1
+    assert ['/dev/sda1', '/dev/sdb1'] == result
+    assert not api.current_logger.warnmsg
+
+
+def test_get_component_devices_not_md_device(monkeypatch):
+    run_mocked = RunMocked()
+    monkeypatch.setattr(mdraid, 'run', run_mocked)
+
+    with pytest.raises(ValueError):
+        mdraid.get_component_devices(NOT_MD_DEVICE)
+    assert mdraid.run.called == 1