NAS-130745 / 25.04 / Update smartctl usage to use JSON (#14479)

src/middlewared/middlewared/api/v25_04_0/__init__.py

@@ -8,3 +8,4 @@
 from .privilege import *  # noqa
 from .user import *  # noqa
 from .vendor import *  # noqa
+from .smartctl import * # noqa

src/middlewared/middlewared/api/v25_04_0/smartctl.py

@@ -0,0 +1,38 @@
+from typing import Any
+
+from middlewared.api.base import BaseModel
+
+__all__ = ["AtaSelfTest", "NvmeSelfTest", "ScsiSelfTest"]
+
+
+class AtaSelfTest(BaseModel):
+    num: int
+    description: str
+    status: str
+    status_verbose: str
+    remaining: float
+    lifetime: int
+    lba_of_first_error: int | None = None
+
+
+class NvmeSelfTest(BaseModel):
+    num: int
+    description: str
+    status: str
+    status_verbose: str
+    power_on_hours: int
+    failing_lba: int | None = None
+    nsid: int | None = None
+    seg: int | None = None
+    sct: int | None = 0x0
+    code: int | None = 0x0
+
+
+class ScsiSelfTest(BaseModel):
+    num: int
+    description: str
+    status: str
+    status_verbose: str
+    segment_number: int | None = None
+    lifetime: int | None = None
+    lba_of_first_error: int | None = None

src/middlewared/middlewared/etc_files/smartd.py

@@ -2,6 +2,7 @@
 import re
 import shlex
 import subprocess
+import json
 
 from middlewared.common.smart.smartctl import get_smartctl_args, smartctl, SMARTCTX
 from middlewared.plugins.smart_.schedule import SMARTD_SCHEDULE_PIECES, smartd_schedule_piece
@@ -23,15 +24,16 @@ async def ensure_smart_enabled(args):
     if any(arg.startswith("/dev/nvme") for arg in args):
         return True
 
-    p = await smartctl(args + ["-i"], stderr=subprocess.STDOUT, check=False, encoding="utf8", errors="ignore")
-    if not re.search("SMART.*abled", p.stdout):
+    p = await smartctl(args + ["-i", "--json=c"], check=False, stderr=subprocess.STDOUT, encoding="utf8", errors="ignore")
+    pjson = json.loads(p.stdout)
+    if not pjson["smart_support"]["available"]:
         logger.debug("SMART is not supported on %r", args)
         return False
 
-    if re.search("SMART.*Enabled", p.stdout):
+    if pjson["smart_support"]["enabled"]:
         return True
 
-    p = await smartctl(args + ["-s", "on"], stderr=subprocess.STDOUT, check=False)
+    p = await smartctl(args + ["-s", "on"], check=False, stderr=subprocess.STDOUT)
     if p.returncode == 0:
         return True
     else:

src/middlewared/middlewared/plugins/disk.py

@@ -12,7 +12,6 @@
 
 RE_SED_RDLOCK_EN = re.compile(r'(RLKEna = Y|ReadLockEnabled:\s*1)', re.M)
 RE_SED_WRLOCK_EN = re.compile(r'(WLKEna = Y|WriteLockEnabled:\s*1)', re.M)
-RE_SMART_AVAILABLE = re.compile(r'SMART support is:\s+Available')
 
 
 class DiskModel(sa.Model):
@@ -140,12 +139,11 @@ async def disk_extend(self, disk, context):
 
         disk['supports_smart'] = None
         if context['supports_smart']:
-            if await self.middleware.call('truenas.is_ix_hardware'):
+            if await self.middleware.call('truenas.is_ix_hardware') or disk['name'].startswith('nvme'):
                 disk['supports_smart'] = True
             else:
-                disk['supports_smart'] = disk['name'].startswith('nvme') or bool(RE_SMART_AVAILABLE.search(
-                    await self.middleware.call('disk.smartctl', disk['name'], ['-a'], {'silent': True}) or ''
-                ))
+                disk_query = await self.middleware.call('disk.smartctl', disk['name'], ['-a', '--json=c'], {'silent': True})
+                disk['supports_smart'] = disk_query.get('smart_support', {}).get('available', False)
 
         if disk['name'] in context['boot_pool_disks']:
             disk['pool'] = context['boot_pool_name']

src/middlewared/middlewared/plugins/disk_/smart_attributes.py

@@ -38,12 +38,16 @@ async def smart_attributes(self, name):
         """
         Returns S.M.A.R.T. attributes values for specified disk name.
         """
-        output = json.loads(await self.middleware.call('disk.smartctl', name, ['-A', '-j']))
+        output = json.loads(await self.middleware.call('disk.smartctl', name, ['-a', '--json=c']))
 
         if 'ata_smart_attributes' in output:
             return output['ata_smart_attributes']['table']
+        if 'nvme_smart_health_information_log' in output:
+            return output['nvme_smart_health_information_log']
+        if 'scsi_error_counter_log' in output and 'scsi_grown_defect_list' in output:
+            return {'scsi_error_counter_log': output['scsi_error_counter_log'], 'scsi_grown_defect_list': output['scsi_grown_defect_list']}
 
-        raise CallError('Only ATA device support S.M.A.R.T. attributes')
+        raise CallError('Only ATA/SCSI/NVMe devices support S.M.A.R.T. attributes')
 
     @private
     async def sata_dom_lifetime_left(self, name):

src/middlewared/middlewared/plugins/disk_/temperature.py

@@ -1,6 +1,7 @@
 import asyncio
 import datetime
 import time
+import json
 
 import async_timeout
 
@@ -61,9 +62,8 @@ async def temperature(self, name, options):
 
     @private
     async def temperature_uncached(self, name, powermode):
-        output = await self.middleware.call('disk.smartctl', name, ['-a', '-n', powermode.lower()], {'silent': True})
-        if output is not None:
-            return parse_smartctl_for_temperature_output(output)
+        if output := await self.middleware.call('disk.smartctl', name, ['-a', '-n', powermode.lower(), '--json=c'], {'silent': True}):
+            return parse_smartctl_for_temperature_output(json.loads(output))
 
     @private
     async def reset_temperature_cache(self):