Re: [smartmontools-support] Smartd/smartctl spin up drive (-n directive versions 5.38 and 5.39)

[Douglas G. <dgi...@in...>]

Christian Franke wrote:
> Hi,
> 
> thanks to Antonio who tested several of my patches, we found the root of 
> the problem:
> 
> hdparm opens the device with O_RDONLY
> smartctl opens the device with O_RDWR
> 
> Interestingly the latter wakes up the drive.
> 
> Does Linux probably send some extra command when an raw device opened 
> for writing is closed ?
> Is the R/W mode necessary for SCSI pass through ?

Christian,
What do you mean by "raw device"
    - /dev/sdb, or
    - /dev/sg3 (for example) ?

In linux SCSI generic devices need to be opened O_RDWR
to send commands unless the command is known to be safe
(and is pretty common), for example, INQUIRY and MODE
SENSE. A minimum of command filtering is done and all
uses of the SCSI ATA PASS-THROUGH commands need O_RDWR.

As for the sd driver, it is not clear to me. There is
definitely code to do a SYNCHRONIZE CACHE command on
close if it thinks the device has its Write Cache Enabled
(WCE). Since that is a property of the disk the block
layer (which receives the open() from the user space)
could clear WCE if the device was opened O_RDONLY.

So try open("/dev/sd<x>", O_RDONLY) and see what happens.
You will soon find out if that is insufficient privilege.

Doug Gilbert