Unknown error 21 when labelling a Tape

[bixente43]

hi,

Since a couple of days, it's impossible to label 9840 Tape on a Networker 6.1.4 server (windows 2000). We have the following error :
"label write, unknown error 32"
or "cannot set record size : block size ioctl failed"

Sometimes a tape can be labelled. When i use it for a backup, it begins correctly but after a few seconds i have this message :
nsrd: media warning: \\.\Tape1 writing: unknown error 21 (0x15), at file 2 record 143
nsrd: media notice: 9840 tape N10860 on \\.\Tape1 is full
nsrd: media notice: 9840 tape N10860 used 36 MB of 20 GB capacity
nsrd: media warning: \\.\Tape1 reading: unknown error 21 (0x15)
...
nsrd: media warning: \\.\Tape1 moving: tape_rewind rewind failed: unknown error 21 (0x15) (268435477)
nsrd: media warning: \\.\Tape1 moving: fsf 2: unknown error 21 (0x15)
nsrd: media emergency: could not position N10860 to file 2, record 139
nsrd: media warning: \\.\Tape1 reading: unknown error 21 (0x15)
nsrd: media notice: Volume "N10860" on device "\\.\Tape1": Cannot decode block. Verify the device configuration. Tape positioning by record is disabled.
nsrd: media warning: verification of volume "N10860", volid 2880035073 failed, can not read record 139 of file 2 on 9840 tape N10860
nsrd: media notice: verification of volume "N10860", volid 2880035073 failed, volume is being marked as full.

The only change we have made was to upgrade the 9840 FW (1.35.109 now, 1.35.105 before).

An idea?
Thanks

 

[605]

If NW labels a tape, it writes ...
- label block (32kB)
- File Mark
- label block (32kB)
- File Mark
- File Mark

Double FMK indicates the logical end-of-tape, aka as end-of-media.

With the first write, it deletes the second of the two consecutive FMKs and writes data blocks (records). Obviously, it has problems here, which are not covered by any internal SCSI error list and are therefore unknown. The drive then can not even run a successful rewind. Later it can not even position to a certain block. Probably due to the rewind problem.

Assume that the media has not been labeled/overwritten by another application, i assume this to be cause dby the firmware, especially if you did not change anything else.

 

[bixente43]

Sorry, i have forgotten something :
I have used the tapes on a Networker 7.2.1 server (Windows 2003). The drives were the same. I wanted to test this "new" version. I have labelled all the tapes (10) and i backed up servers succesfully. Then i stooped the 7.2.1 server and re used the tapes on the 6.1.4 server...

 

[605]

What do you expect ... forward compatibility ?

 

[bixente43]

It was what i thought. But STK has downgraded the FW. But the problem still occurs....

 

[speculos]

It looks like a block size problem!
NT4 default block size for 9840 tapes was 128K,
Windows2K default block size is 64K; (I don't know Windows2003 default block size)
If you want to read 9840 tape labelled with a different bloc size, you can adjust the windows variable NSR_DEV_BLOCK_SIZE_9840 (reboot needed)

You can find the actual block-size of a labelled tape with the scanner command.
Example with a 128K 9840 tape:
D:\>scanner -v -i \\.\Tape0
scanner: Reading the label...
scanner: Reading the label done
scanner: scanning 9840 tape L00221 on \\.\Tape0
scanner: volume id 1984635649 record size 131072

 

[605]

I doubt that. In such case i would expect a clear error message. However, run

scanner -nv device_name

and you should see the block size reported by the label. So stop it soon.

 

[bixente43]

the scanner -nv gives the following result :
scanner: \\.\Tape0: opened for reading
scanner: \\.\Tape0: rewinding
scanner: Cannot set \\.\Tape0 block size: unknown error 21 (0x15)
scanner: error, cannot set record size: block size ioctl failed
scanner: scanning for valid records...
scanner: read: -1 bytes : unknown error 1104 (0x450)
scanner: Cannot continue
scanner: No valid tape records found

Not very good......

 

[speculos]

Try this:
create a system variable called NSR_DEV_BLOCK_SIZE_9840
and set the value to 128, then reboot, and retry scanner.
If the bloc size is too small, you can't read the label

Other test: use tapeexer on a scratch tape to check your hardware.

 

[605]

The block size is stored inside the label. And those are all written in 32kb blocks. They must me readable ... unless you labeled the tape (accidentally) with another utility.

Does the NW CDI command "read_a_block" give you any input?
Or do you have another utility which may read a tape block?

 

[bixente43]

I tried to erase the tape (mt -f \\.\Tape0 erase) and label it. No success...

I don't have the read_a_block command (Windows and NW 6.1.4) and no other utility...

 

[605]

Sorry - that's true. It has been implemented with 7.0.

 

[605]

After re-reading the messages, something comes into my mind: In your second note you mentioned 2 servers.

Could it be that the other server uses drives for the same media but with a higher density? - I do not know the 9840 hardware but with DLT drives/media this is possible.

In this case, you have to first overwrite the data from BOT to be able to switch the density. A label process will NOT do this as it will try to read the existing tape label first. This is done with the default density which an older drive does not have and will fail to read. Is this a possible scenario?

If so, use NW's tapeexercise command and to overwrite the label with the max density before you relabel it with NW.

 

[bixente43]

There are 2 servers but they are both on a SAN using the same drives and the same tapes (FC 9840A). Of course, when a server is on, the other if off (no DDS or something else...).

 

[speculos]

Did you use tapeexer on a scratch tape to check your hardware?

 

[bixente43]

Hi,
I've just tried the tapeexer on a scratch tape. On the both drives, I have the same error :

BasicTest: test begin
rewinding `\\.\Tape1'
ReadTapefile: ReadFile error unknown error 21 (0x15)
write 0 failed, errno 5, Input/output error
EotTest: test begin
\\.\Tape1: open failed, unknown error 32 (0x20)
\\.\Tape1: open failed, unknown error 32 (0x20)
........

Tape1 has been replaced by STK a few days ago.

 

[bixente43]

Information : The jbverify gives the following response :

Summary report of jbverify
======= ====== == ========

Device Handle Blocksize Jukebox Drv No. Status
------------- --------- ------- ------- ------
\\.\Tape0 262144 STK9310E 1 Pass
\\.\Tape1 262144 STK9310E 2 Pass

 

[speculos]

Tapeexer is not successfull, so you have to fix first the hardware problem; it's not a Networker issue.
For your information we use 9840 FW 1.35.105,and it works fine

 

[bixente43]

I have tested different things :
- a new Emulex card (sale model) => NOK
- Downgrade the FW 1.35.105 => NOK
- Upgrading NW 6.1.4 to 7.2.1 => NOK

There are 2 possibilities (because a 7.2.1 NW server connected to the SAN and the 2 tapes is working fine) :
- the fiber link,
- the GBIC on the Brocade switch.

 

[bixente43]

I've just found the reason of the problem. The W2K3 Server on which Networker 7.2 is installed sends messages (??) to the tapes and causes the errors on tapes.

When i turn off the W2k3 server, I can label tapes and backup servers.

When i turn on the server during a backup, errors are present and particularly "Block size is 32768 bytes not 262144".

Do you know how to modify the block size on W2k3 (i have a LP9002L Emulex card connected to a SAN) ?