Re: NAA identifiers with scsi_id

Re: NAA identifiers with scsi_id

[Patrick Mansfield]

Hannes -

On Tue, Jun 15, 2004 at 04:39:37PM +0200, Hannes Reinecke wrote:
> Hi all,
> 
> having checked scsi_id, I found out that it will always prefer NAA ids 
> in favour of any other types (EUI-64 et.al.).
> 
> This is not very helpful if scsi_id is used to generate meaningful 
> persistent driver ids, as the user might want to check the links created 
> by udev against the device list presented by the system (e.g. via 
> 'dmesg' or 'cat /proc/scsi/scsi' or even by scanning sysfs).
> 
> Can't we just ignore NAA ids and fall back to page 0x80 + inquiry if 
> these are all we will get via page 0x83? Just having an arbitrary number 
> is not very helpful and totally intransparent to the user.

No.

Some reasons we should not do this:

1) No matter what page or method we use there is a rather arbitrary number,
it is just that for page 0x80 or a page 0x83 vendor specific id, scsi_id
prepends the vendor and model before the id, this is done to ensure the
value is unique across all vendors and models, not so users can more
easily match an id to an actual device.

Some (well not your standard desktop pc) systems will be connected to
multiple LUNs that all have the same vendor and model, and having the
vendor model pre-pended does not make finding them any easier.


2) a device might support only page 0x83 NAA, so falling back to page
0x80 could give nothing.


3) For some disk arrays, page 0x80 gives the same result for all LUNs -
generally it's broken, or it is the serial number of the target not of any
particular LUN. In the past, pre page 0x83 support, vendor and model
specific code had to be added to most disk arrays to create a unique id
based on page 0x80, for example by appending the LUN to the page 0x80
value.

I picked the NAA IEEE extended as the highest priority, as I interpreted
the standard to imply it can be extended to support unique naming of LUNs
on a system, where you can create and remove LUNs. And, that is the only
value I see returned by a fastt 200.

For example, I have a system with one fastt 200 and 32 LUNs, the page 0x80
for each is identical (I don't know why it has T05078426, I also have some
serveraid storage attached that can't handle page 0x80 or 0x83):

0:0:0:0: Invalid page 0x80
./foo.sh failed for device sda
3:0:0:24: SIBM     3542           1T05078426      
3:0:0:25: SIBM     3542           1T05078426      
3:0:0:26: SIBM     3542           1T05078426      
3:0:0:27: SIBM     3542           1T05078426      
3:0:0:28: SIBM     3542           1T05078426      
3:0:0:29: SIBM     3542           1T05078426      
3:0:0:30: SIBM     3542           1T05078426      
3:0:0:31: SIBM     Universal Xport1T05078426      
0:0:1:0: Invalid page 0x80
./foo.sh failed for device sdb
3:0:0:0: SIBM     3542           1T05078426      
3:0:0:1: SIBM     3542           1T05078426      
3:0:0:2: SIBM     3542           1T05078426      
3:0:0:3: SIBM     3542           1T05078426      
3:0:0:4: SIBM     3542           1T05078426      
3:0:0:5: SIBM     3542           1T05078426      
3:0:0:6: SIBM     3542           1T05078426      
3:0:0:7: SIBM     3542           1T05078426      
3:0:0:8: SIBM     3542           1T05078426      
3:0:0:9: SIBM     3542           1T05078426      
3:0:0:10: SIBM     3542           1T05078426      
3:0:0:11: SIBM     3542           1T05078426      
3:0:0:12: SIBM     3542           1T05078426      
3:0:0:13: SIBM     3542           1T05078426      
3:0:0:14: SIBM     3542           1T05078426      
3:0:0:15: SIBM     3542           1T05078426      
3:0:0:16: SIBM     3542           1T05078426      
3:0:0:17: SIBM     3542           1T05078426      
3:0:0:18: SIBM     3542           1T05078426      
3:0:0:19: SIBM     3542           1T05078426      
3:0:0:20: SIBM     3542           1T05078426      
3:0:0:21: SIBM     3542           1T05078426      
3:0:0:22: SIBM     3542           1T05078426      
3:0:0:23: SIBM     3542           1T05078426      


Page 0x83 for all of the above gives (weird that LUNs 1, 3 and 31 have a
different prefix, that is they start with 3600a0b80000b18a5000000 rather
than with 3600a0b80000b1b7f000000):

0:0:0:0: Invalid page 0x83
./foo.sh failed for device sda
3:0:0:24: 3600a0b80000b1b7f0000009b3fd5f394
3:0:0:25: 3600a0b80000b1b7f0000009c3fd5f39c
3:0:0:26: 3600a0b80000b1b7f0000009d3fd5f3a6
3:0:0:27: 3600a0b80000b1b7f0000009e3fd5f3ae
3:0:0:28: 3600a0b80000b1b7f0000009f3fd5f3b8
3:0:0:29: 3600a0b80000b1b7f000000a03fd5f3c0
3:0:0:30: 3600a0b80000b1b7f000000a13fd5f3ca
3:0:0:31: 3600a0b80000b18a50000001d00000000
0:0:1:0: Invalid page 0x83
./foo.sh failed for device sdb
3:0:0:0: 3600a0b80000b1b7f000000853fd5f2bc
3:0:0:1: 3600a0b80000b18a50000001b3fd5f387
3:0:0:2: 3600a0b80000b1b7f000000863fd5f2ce
3:0:0:3: 3600a0b80000b18a50000001c3fd5f399
3:0:0:4: 3600a0b80000b1b7f000000873fd5f2de
3:0:0:5: 3600a0b80000b1b7f000000883fd5f2e8
3:0:0:6: 3600a0b80000b1b7f000000893fd5f2f0
3:0:0:7: 3600a0b80000b1b7f0000008a3fd5f2f8
3:0:0:8: 3600a0b80000b1b7f0000008b3fd5f302
3:0:0:9: 3600a0b80000b1b7f0000008c3fd5f30a
3:0:0:10: 3600a0b80000b1b7f0000008d3fd5f312
3:0:0:11: 3600a0b80000b1b7f0000008e3fd5f31c
3:0:0:12: 3600a0b80000b1b7f0000008f3fd5f324
3:0:0:13: 3600a0b80000b1b7f000000903fd5f32c
3:0:0:14: 3600a0b80000b1b7f000000913fd5f336
3:0:0:15: 3600a0b80000b1b7f000000923fd5f33e
3:0:0:16: 3600a0b80000b1b7f000000933fd5f348
3:0:0:17: 3600a0b80000b1b7f000000943fd5f350
3:0:0:18: 3600a0b80000b1b7f000000953fd5f35a
3:0:0:19: 3600a0b80000b1b7f000000963fd5f364
3:0:0:20: 3600a0b80000b1b7f000000973fd5f36c
3:0:0:21: 3600a0b80000b1b7f000000983fd5f376
3:0:0:22: 3600a0b80000b1b7f000000993fd5f382
3:0:0:23: 3600a0b80000b1b7f0000009a3fd5f38a

Here's the script I used to generate the above, I ran it with:

sh ./foo.sh -g -i -p 0x80

And:

sh ./foo.sh -g -i -p 0x83

-------------- cut here ---------------
#! /bin/sh

scsi_id=/sbin/scsi_id
opts=$*

sysfs_dir=$(mount | awk '$5 == "sysfs" {print $3}')
if [ "$sysfs_dir" = "" ]
then
        sysfs_dir="/sys"
        echo "Using sysfs mount point \"$sysfs_dir\"" >&2
fi

c=$(ls /${sysfs_dir}/block/sd* 2>/dev/null | wc -l)
if [ $c = 0 ]
then
        echo $0 no block devices present >&2
        exit 1
fi

cd ${sysfs_dir}/block
for name in sd*
do
        id=$($scsi_id ${opts} -s /block/$name)
        if [ $? != 0 ]
        then
                echo $0 failed for device $name >&2
        else
                echo ${id}
        fi
done

-- Patrick Mansfield

Re: NAA identifiers with scsi_id

[Douglas Gilbert]

Patrick Mansfield wrote:
> Hannes -
> 
> On Tue, Jun 15, 2004 at 04:39:37PM +0200, Hannes Reinecke wrote:
> 
>>Hi all,
>>
>>having checked scsi_id, I found out that it will always prefer NAA ids 
>>in favour of any other types (EUI-64 et.al.).
>>
>>This is not very helpful if scsi_id is used to generate meaningful 
>>persistent driver ids, as the user might want to check the links created 
>>by udev against the device list presented by the system (e.g. via 
>>'dmesg' or 'cat /proc/scsi/scsi' or even by scanning sysfs).
>>
>>Can't we just ignore NAA ids and fall back to page 0x80 + inquiry if 
>>these are all we will get via page 0x83? Just having an arbitrary number 
>>is not very helpful and totally intransparent to the user.
> 
> 
> No.
> 
> Some reasons we should not do this:
> 
> 1) No matter what page or method we use there is a rather arbitrary number,
> it is just that for page 0x80 or a page 0x83 vendor specific id, scsi_id
> prepends the vendor and model before the id, this is done to ensure the
> value is unique across all vendors and models, not so users can more
> easily match an id to an actual device.
> 
> Some (well not your standard desktop pc) systems will be connected to
> multiple LUNs that all have the same vendor and model, and having the
> vendor model pre-pended does not make finding them any easier.
> 
> 
> 2) a device might support only page 0x83 NAA, so falling back to page
> 0x80 could give nothing.

Recent SPC-3 drafts have marked support for the Device
Identification VPD page (i.e. 0x83) in INQUIRY as mandatory
while support for the serial number VPD page (0x80) is still
optional.

The allocation length (of the response) of an INQUIRY has
been expanded from 1 to 2 bytes. This, no doubt, is to
cope with large responses associated with VPD page 0x83.

I am in the process of adding 0x83 VPD page
support to sg_inq (in sg3_utils) to decode its
multiple descriptors.

<snip/>

Doug Gilbert

Re: NAA identifiers with scsi_id

[Bryan Henderson]

> Can't we just ignore NAA ids and fall back to page 0x80 + inquiry if 
> these are all we will get via page 0x83?  Just having an arbitrary 
number 
> is not very helpful and totally intransparent to the user.

It's extremely helpful if you're trying to figure out where to send a CDB 
so it reaches the same logical unit as a CDB sent to some address from 
another system or at another time from the same system.  In that case, an 
arbitrary but absolutely unique number is all you need.  And that's the 
most important case.

The matter of identifying a logical unit in human terms (maybe, "the old 
Hitachi disk," or "the disk drive I bought last week that is on the left 
end of the rack") is quite a bit more complex; I'd expect to see that in a 
Device Description page rather than the Device Identification (0x83) or 
Unit Serial Number (0x80) page.  That could give you a make and model, the 
serial number stamped on the box, technology, or whatever other 
information a person uses to identify a logical unit.  But I wouldn't use 
it in anything automated.

--
Bryan Henderson                          IBM Almaden Research Center
San Jose CA                              Filesystems


-------------------------------------------------------
This SF.Net email is sponsored by The 2004 JavaOne(SM) Conference
Learn from the experts at JavaOne(SM), Sun's Worldwide Java Developer
Conference, June 28 - July 1 at the Moscone Center in San Francisco, CA
REGISTER AND SAVE! http://java.sun.com/javaone/sf Priority Code NWMGYKND