Re: [smartmontools-support] smartd vs. smartctl: Fusion MPT SAS/SATA RAID controller and physical disks (/dev/sg*)

[Douglas G. <do...@to...>]

Wolfram Schlich wrote:
> * Douglas Gilbert <do...@to...> [2008-02-16 17:13]:
>> Bruce Allen wrote:
>>> Christian, Doug,
>>> Could one of you have a look at this?  The report is that 'smartctl -d 
>>> sat' works correctly but smartd with '-d sat' in /etc/smartd.conf does NOT 
>>> work correctly.  I would like to sort this out before doing a new release.
>>> Cheers,
>>>      Bruce
>>> On Tue, 5 Feb 2008, Wolfram Schlich wrote:
>>>> * Bruce Allen <ba...@gr...> [2008-02-05 18:21]:
>>>>> And perhaps check against current CVS HEAD?
>>>> Done -- current CVS HEAD has the same behaviour.
>> I connected a SATA disk (Seagate 7200.7) to a MPT SAS
>> Fusion HBA and smartd worked for me with:
>>    /dev/sda -d sat -a
>> in /etc/smartd.conf (and DEVICESCAN commented out).
>> That is with lk 2.6.24-rc6 and a version of smartmontools
>> near the CVS head.
>>
>> I know that Dell's Perc 5i is a Megaraid card (rather than
>> MPT SAS fusion, both from LSI). Since a RAID is involved
>> the Perc 6i may be blocking the SCSI ATA PASS_THROUGH
>> command used by '-d sat'. If you could send a SCSI WRITE
>> command, or the ATA equivalent via the ATA PASS_THROUGH
>> command then the RAID integrity could be compromised.
>>
>> This looks like a specialized bug, associated with the
>> Perc 6i to me, rather than a general failure in
>> smartd. [smartd make need tweaking, or perhaps LSI need
>> to be more selective in what they allow through /dev/sg1).
> 
> The controller I tried with is NOT a PERC one, it is a
> MPT SAS fusion one, called SAS6/iR (NOT PERC6!) at Dell.
> 
> As I said, it worked fine with smartctl, but not with smartd.

Wolfram,
Whenever I dive into the smartd code I come away with a
sore head. This was no exception. Since a release is pending
I went for a relatively low level (i.e. os_linux.cpp) patch
limited to the strange case in point.

See my comment in the deviceopen() function in os_linux.cpp
for more details (also found in the CHANGELOG).

Could you retry the CVS HEAD now and see if it makes a
difference?

Doug Gilbert