How to reread disk size?

How to reread disk size?

[Asdo]

Hi all,
I have a system with hot swappable bays but one of the involved 
controllers apparently does not support hot swap:
(I have unfortunately an old kernel: 2.6.24)

00:1f.2 IDE interface: Intel Corporation 631xESB/632xESB/3100 Chipset 
SATA IDE Controller (rev 09) (prog-if 8f [Master SecP SecO PriP PriO])
        Flags: bus master, 66MHz, medium devsel, latency 0, IRQ 19
        I/O ports at 18c8 [size=8]
        I/O ports at 18ac [size=4]
        I/O ports at 18c0 [size=8]
        I/O ports at 18a8 [size=4]
        I/O ports at 18b0 [size=16]
        Memory at da804400 (32-bit, non-prefetchable) [size=1K]
        Capabilities: [70] Power Management version 2


I have swapped the disk with a larger one (750GB -> 1TB) but it blockdev 
--getsize64 still sees the old size. It has not realized I have swapped 
disk.

In the past I was doing:

  blockdev --rereadpt /dev/sdX

and it usually worked on other controllers to reread the size visible 
from "blockdev --getsize" or "blockdev --getsize64". One time I think it 
even worked on exactly *that* controller... but it's not working now, 
it's strange.

Is there a technique, or I am out of luck?
The machine should not be rebooted
I would even enter the size manually if possible: I know how many LBA 
blocks are in that disk.

Thank you

Re: How to reread disk size?

[Robert Hancock]

On 02/19/2010 05:48 PM, Asdo wrote:
> Hi all,
> I have a system with hot swappable bays but one of the involved
> controllers apparently does not support hot swap:
> (I have unfortunately an old kernel: 2.6.24)
>
> 00:1f.2 IDE interface: Intel Corporation 631xESB/632xESB/3100 Chipset
> SATA IDE Controller (rev 09) (prog-if 8f [Master SecP SecO PriP PriO])
> Flags: bus master, 66MHz, medium devsel, latency 0, IRQ 19
> I/O ports at 18c8 [size=8]
> I/O ports at 18ac [size=4]
> I/O ports at 18c0 [size=8]
> I/O ports at 18a8 [size=4]
> I/O ports at 18b0 [size=16]
> Memory at da804400 (32-bit, non-prefetchable) [size=1K]
> Capabilities: [70] Power Management version 2
>
>
> I have swapped the disk with a larger one (750GB -> 1TB) but it blockdev
> --getsize64 still sees the old size. It has not realized I have swapped
> disk.
>
> In the past I was doing:
>
> blockdev --rereadpt /dev/sdX
>
> and it usually worked on other controllers to reread the size visible
> from "blockdev --getsize" or "blockdev --getsize64". One time I think it
> even worked on exactly *that* controller... but it's not working now,
> it's strange.
>
> Is there a technique, or I am out of luck?
> The machine should not be rebooted
> I would even enter the size manually if possible: I know how many LBA
> blocks are in that disk.

What dmesg output do you get when you do this?

Re: How to reread disk size?

[Asdo]

Robert Hancock wrote:
>> [CUT]
>> In the past I was doing:
>>
>> blockdev --rereadpt /dev/sdX
>>
>> and it usually worked on other controllers to reread the size visible
>> from "blockdev --getsize" or "blockdev --getsize64". One time I think it
>> even worked on exactly *that* controller... but it's not working now,
>> it's strange.
>>
>> Is there a technique, or I am out of luck?
>> The machine should not be rebooted
>> I would even enter the size manually if possible: I know how many LBA
>> blocks are in that disk.
>
> What dmesg output do you get when you do this?

with "this" I suppose you mean the blockdev --rereadpt?

This is the dmesg I get upon blockdev --rereadpt:

[508100.472337] sd 3:0:1:0: [sdr] 1465149168 512-byte hardware sectors 
(750156 MB)
[508100.472376] sd 3:0:1:0: [sdr] Write Protect is off
[508100.472381] sd 3:0:1:0: [sdr] Mode Sense: 00 3a 00 00
[508100.472412] sd 3:0:1:0: [sdr] Write cache: enabled, read cache: 
enabled, doesn't support DPO or FUA
[508100.472418]  sdr: unknown partition table


It rereads the wrong size, i.e. the one of the old disk :-(
I am sure sdr is the correct disk, I even identified it by doing dd 
if=/dev/sdr of=/dev/null and then looking at the activity led to confirm 
it's the right drive. It should have read a size of 1TB.

Thanks for your help

Re: How to reread disk size?

[Asdo]

Asdo wrote:
> [...]
> It rereads the wrong size, i.e. the one of the old disk :-(
> I am sure sdr is the correct disk, I even identified it by doing dd 
> if=/dev/sdr of=/dev/null and then looking at the activity led to 
> confirm it's the right drive. It should have read a size of 1TB.

I can also add that smartctl -a /dev/sdr reads the CORRECT size...
(but it does not update the size in the linux's block device obviously)

:-( ...

Re: How to reread disk size?

[James Bottomley]

On Sat, 2010-02-20 at 22:30 +0100, Asdo wrote:
> Robert Hancock wrote:
> >> [CUT]
> >> In the past I was doing:
> >>
> >> blockdev --rereadpt /dev/sdX
> >>
> >> and it usually worked on other controllers to reread the size visible
> >> from "blockdev --getsize" or "blockdev --getsize64". One time I think it
> >> even worked on exactly *that* controller... but it's not working now,
> >> it's strange.
> >>
> >> Is there a technique, or I am out of luck?
> >> The machine should not be rebooted
> >> I would even enter the size manually if possible: I know how many LBA
> >> blocks are in that disk.
> >
> > What dmesg output do you get when you do this?
> 
> with "this" I suppose you mean the blockdev --rereadpt?
> 
> This is the dmesg I get upon blockdev --rereadpt:
> 
> [508100.472337] sd 3:0:1:0: [sdr] 1465149168 512-byte hardware sectors 
> (750156 MB)
> [508100.472376] sd 3:0:1:0: [sdr] Write Protect is off
> [508100.472381] sd 3:0:1:0: [sdr] Mode Sense: 00 3a 00 00
> [508100.472412] sd 3:0:1:0: [sdr] Write cache: enabled, read cache: 
> enabled, doesn't support DPO or FUA
> [508100.472418]  sdr: unknown partition table
> 
> 
> It rereads the wrong size, i.e. the one of the old disk :-(
> I am sure sdr is the correct disk, I even identified it by doing dd 
> if=/dev/sdr of=/dev/null and then looking at the activity led to confirm 
> it's the right drive. It should have read a size of 1TB.

What's happening is that libata is returning the old cached size to
READ_CAPACITY.  This would likely indicate some type of libata hotplug
failure ... the dmesg  across the unplug/plug would be useful for
diagnosing this.

James

Re: How to reread disk size?

[Asdo]

Robert Hancock wrote:
> ata_piix doesn't have a way to get automatically notified that the
> disk has been hotplugged since the chipset doesn't support it in this
> mode. (The ideal solution would be to switch the controller into AHCI
> mode, but I don't know if that's possible on that chipset and BIOS.)
>
> I think there are some ways to get the code to recheck the disk
> however.. you can try echoing something into the "rescan" sysfs file
> under /sys that matches your device (do a "find /sys/ -name
> scsi_level" to see what directories they're in, the rescan file is in
> the same directory).
>   

Thanks for the hint
It doesn't work unfortunately.
Echoing something into rescan makes it output (in dmesg) basically the 
same message that comes out with blockdev --rereadpt /dev/sdr. The size 
is still detected wrong. Catting "model" also turns out that the old 
brand of disk still being detected. smartctl -a does show the new disk :-)
I tried to write "1" in evt_media_change but it's permission denied :-)

Thank you

Re: How to reread disk size?

[James Bottomley]

On Mon, 2010-02-22 at 01:58 +0100, Asdo wrote:
> Robert Hancock wrote:
> > ata_piix doesn't have a way to get automatically notified that the
> > disk has been hotplugged since the chipset doesn't support it in this
> > mode. (The ideal solution would be to switch the controller into AHCI
> > mode, but I don't know if that's possible on that chipset and BIOS.)
> >
> > I think there are some ways to get the code to recheck the disk
> > however.. you can try echoing something into the "rescan" sysfs file
> > under /sys that matches your device (do a "find /sys/ -name
> > scsi_level" to see what directories they're in, the rescan file is in
> > the same directory).
> >   
> 
> Thanks for the hint
> It doesn't work unfortunately.
> Echoing something into rescan makes it output (in dmesg) basically the 
> same message that comes out with blockdev --rereadpt /dev/sdr. The size 
> is still detected wrong. Catting "model" also turns out that the old 
> brand of disk still being detected. smartctl -a does show the new disk :-)
> I tried to write "1" in evt_media_change but it's permission denied :-)

Right.  libata caches the value from the moment it scans the device.
You might be able to alter it by having the device destroyed and
recreated ... that would be echoing 1 to the delete under the scsi
device sysfs node and then echoing '- - -' to the host rescan parameter.
If that doesn't work, libata will need to be fixed somehow because that
would mean it isn't slaving it's device lifetimes to the SCSI model.

James

Re: How to reread disk size?

[Asdo]

James Bottomley wrote:
> On Mon, 2010-02-22 at 01:58 +0100, Asdo wrote:
>   
>> Robert Hancock wrote:
>>     
>>> ata_piix doesn't have a way to get automatically notified that the
>>> disk has been hotplugged since the chipset doesn't support it in this
>>> mode. (The ideal solution would be to switch the controller into AHCI
>>> mode, but I don't know if that's possible on that chipset and BIOS.)
>>>
>>> I think there are some ways to get the code to recheck the disk
>>> however.. you can try echoing something into the "rescan" sysfs file
>>> under /sys that matches your device (do a "find /sys/ -name
>>> scsi_level" to see what directories they're in, the rescan file is in
>>> the same directory).
>>>   
>>>       
>> Thanks for the hint
>> It doesn't work unfortunately.
>> Echoing something into rescan makes it output (in dmesg) basically the 
>> same message that comes out with blockdev --rereadpt /dev/sdr. The size 
>> is still detected wrong. Catting "model" also turns out that the old 
>> brand of disk still being detected. smartctl -a does show the new disk :-)
>> I tried to write "1" in evt_media_change but it's permission denied :-)
>>     
>
> Right.  libata caches the value from the moment it scans the device.
> You might be able to alter it by having the device destroyed and
> recreated ... that would be echoing 1 to the delete under the scsi
> device sysfs node and then echoing '- - -' to the host rescan parameter.
> If that doesn't work, libata will need to be fixed somehow because that
> would mean it isn't slaving it's device lifetimes to the SCSI model.
>   

Great, it works! Thank you

root@mynode:/sys# echo 1 > 
/sys/devices/pci0000:00/0000:00:1f.2/host3/target3:0:1/3:0:1:0/delete

and then

echo "- - -" > /sys/class/scsi_host/host3/scan

recreated the device with 1TB size!

Acually then I tried to swap disks again and just give the
  echo "- - -" > /sys/class/scsi_host/host3/scan
and it worked already. So deleting the device was not needed.
However the "rescan" in the same directory of scsi_level you suggested 
in previous post did not work. This "scan" in scsi_host is more powerful :-D

Thank you