Re: star quitting when changing tapes - FIXED!!!!

"Ruud Baart" <[email protected]> Sat, 26 Aug 2006 14:54:13 +0200
Newsgroups gmane.comp.archivers.star.user
Organization IT's Prompt or never ...
Message-ID <[email protected]>
I use Debian servers. With the Debian kernel 2.6.16 and 2.6.17 I get 
always a device busy message. I use an Adaptec 29160 SCSI controller 
with a Sony AIT SX900V backup-unit. I can rewind the tape unit (mt -f 
/dev/st0 rewind), but I can not extract or list files from tape (star -t 
f=/dev/st0 bs=...).

I used the latest star version star-1.5a75 and recompiled it with 2.6.17 
headers. No luck.

After downgrading to 2.6.15 the self-compiled star-1.5a75 worked, as far 
as ik can see now, without a problem. 

So I can confirm there is a kernel problem but as far as my debian servers 
concerned 2.6.16 and 2.6.17 kernel won't work.

> The bug in the Linux kernel that kills star when backing up to
> multiple tapes is fixed in the latest series of kernels!  Here's the
> list:
> 
> Linux kernel:
> 2.6.15.x - WORKS FINE
> 2.6.16.x - BROKEN
> 2.6.17.x - WORKS FINE
> 
> The README in the star distribution should probably be updated to let
> users know about this kernel bug.
> 
> Thanks for your help, Joerg.
> 
> -Dan
> 
> Joerg Schilling <[email protected]> wrote: Dan  wrote:
> 
> > open("/dev/nst0", O_RDWR|O_CREAT|O_LARGEFILE, 0666) = 4
> > fcntl64(4, F_GETFL)                     = 0x8002 (flags
> > O_RDWR|O_LARGEFILE) fstat64(4, {st_mode=S_IFCHR|0660,
> > st_rdev=makedev(9, 128), ...}) = 0 ioctl(4, SNDCTL_TMR_TIMEBASE or
> > TCGETS, 0xbfa0b804) = -1 EINVAL (Invalid argument)
> >
> > Joerg, can you point me to where in the code this ioctl is sent?  A
> > kernel guy I emailed said this might be sending something that's
> > screwing up the new st.c driver.  If this is causing an error to
> > occur (REQUEST SENSE, ASC/ASCQ kind of error), the driver won't know
> > about it until the next I/O is done.  At that point, the driver will
> > block waiting for someone to read the additional error codes, thus
> > causing the EBUSY.  I know EBUSY should never happen on a write(2),
> > and he agrees.  I need to do more troubleshooting, and disabling
> > this ioctl command is the first step.
> 
> I would guess this is a call to isatty() from within stdio.
> 
> BTW: the correct error code for a non-tty fd would be ENOTTY.
> 
> 
> 
> 
> ---------------------------------
> Do you Yahoo!?
>  Everyone is raving about the  all-new Yahoo! Mail Beta.


Met vriendelijke groet/Regards,
Ruud Baart

Prompt, Kerkstraat 173
5261 CW Vught, Netherlands
Tel: +31 73 6567041
http://www.prompt.nl