LSI 1030 and LTO-3 won't work

[Kenneth D. Merry]

On Wed, Oct 08, 2014 at 21:16:41 +0200, Stefan Huerter wrote:
> Guckux
> 
> I have problems with following configuration:
> FreeBSD whisky 10.1-RC1 FreeBSD 10.1-RC1 #0 r272711M: Tue Oct  7,
> LSILogic 1030 Ultra4 Adapter
> Sun StorEdge C2 - LTO-3 library
> 
> I've included the mpt-module to load at boot time in loader.conf.
> 
> part from dmesg:
> mpt0: <LSILogic 1030 Ultra4 Adapter> port 0xc400-0xc4ff mem
> 0xff9a0000-0xff9bffff,0xff980000-0xff99ffff irq 23 at device 2.0 on pci5
> mpt0: MPI Version=1.2.14.0
> mpt1: <LSILogic 1030 Ultra4 Adapter> port 0xc800-0xc8ff mem
> 0xff9c0000-0xff9dffff irq 20 at device 2.1 on pci5
> mpt1: MPI Version=1.2.14.0
> 
> (sa0:mpt0:0:5:0): 32768-byte tape record bigger than supplied buffer
> (sa0:mpt0:0:5:0): 32768-byte tape record bigger than supplied buffer
> 
> I've tried to resize the blocksize to 10240 via mt blocksize 10240.
> 
> Nothing works - dump and tar...
> dump reports shortly, that the End of tape is detected, tar reports
> "write error".
> 
> With the same controller no Problem with Backup Exec under Windows, or
> Solaris with EMC networker.
> 
> Hugh?
> Any hints for me? Or can I give you some other informations, which you need?

The error messages above usually indicate that you're trying to read a
block from the tape (in this case a 32K block) that is bigger than the
blocksize that you specified.  But it can also happen on a write to a tape
when you're in fixed block mode and you try to write less than the
blocksize.

What does 'mt status' show?  If it shows 32KB blocks, then that may be your
problem.

What blocksize are you using with tar and dump?  (I believe they default to
10240 bytes.)

It's generally easier to run in variable blocksize mode.  You can specify
variable blocksize with 'mt blocksize 0'.  You can verify which mode you're
in with 'mt status'.  For instance, this tape drive is in variable block
mode:

Mode      Density              Blocksize      bpi      Compression
Current:  0x5a:LTO-6           variable       384607   0xff
---------available modes---------
0:        0x5a:LTO-6           variable       384607   0xff
1:        0x5a:LTO-6           variable       384607   0xff
2:        0x5a:LTO-6           variable       384607   0xff
3:        0x5a:LTO-6           variable       384607   0xff
---------------------------------
Current Driver State: at rest.
---------------------------------
File Number: 0  Record Number: 0        Residual Count -1

I would try something like this:

mt rewind
mt blocksize 0
mt status
(verify that it says variable block mode)
dump -C 16 -b 64 -0ua -f /dev/nsa0 /

That specifies:
 - 16MB read cache size (for reading the disk)
 - 64K output blocksize
 - level 0 dump
 - update /etc/dumpdates
 - write until the end of tape
 - use the non-rewound tape device
 - dump the root partition

It is generally better to write to the tape drive with the largest
blocksize that your tape drive and controller support.  You'll get better
throughput that way.

The tape driver in FreeBSD 10 will not allow you to read or write a
blocksize that your drive and controller don't support.  To see what they
support, try:

sysctl kern.cam.sa.0

maxio is the what we think you can effectively write given the limits of
the system and the controller.  cpi_maxio is what the controller claims to
support.

Ken
-- 
Kenneth Merry
ken at FreeBSD.ORG