lkml.org 
[lkml]   [2005]   [Mar]   [2]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
/
Date
From
SubjectRe: Problems with SCSI tape rewind / verify on 2.4.29
On Wed, 2 Mar 2005, Marcelo Tosatti wrote:

> On Wed, Mar 02, 2005 at 11:15:42AM -0000, Mark Yeatman wrote:
> > Hi
> >
> > Never had to log a bug before, hope this is correctly done.
> >
> > Thanks
> >
> > Mark
> >
> > Detail....
> >
> > [1.] One line summary of the problem:
> > SCSI tape drive is refusing to rewind after backup to allow verify and
> > causing illegal seek error
> >
> > [2.] Full description of the problem/report:
> > On backup the tape drive is reporting the following error and failing
> > it's backups.
> >
> > tar: /dev/st0: Warning: Cannot seek: Illegal seek
> >
> > I have traced this back to failing at an upgrade of the kernel to 2.4.29
> > on Feb 8th. The backups have not worked since. Replacement Drives have
> > been tried and cables to no avail. I noticed in the the changelog that a
> > patch by Solar Designer to the Scsi tape return code had been made.

BTW, this "fix" by Solar Designer introduces a bug to 2.4.29: a tape
driver is supposed to return ENOMEM in the case that was changed to return
EIO ;-(

>
> v2.6 also contains the same problem BTW.
>
> Try this:
>
> --- a/drivers/scsi/st.c.orig 2005-03-02 09:02:13.637158144 -0300
> +++ b/drivers/scsi/st.c 2005-03-02 09:02:20.208159200 -0300
> @@ -3778,7 +3778,6 @@
> read: st_read,
> write: st_write,
> ioctl: st_ioctl,
> - llseek: no_llseek,
> open: st_open,
> flush: st_flush,
> release: st_release,

This change covers up the problem. The real bug is in tar. The following
code is from tar is supposed to reposition the tape to the beginning of
the file jus written:

#ifdef MTIOCTOP
{
struct mtop operation;
int status;

operation.mt_op = MTBSF;
operation.mt_count = 1;
if (status = rmtioctl (archive, MTIOCTOP, (char *) &operation), status
< 0)
{
if (errno != EIO
|| (status = rmtioctl (archive, MTIOCTOP, (char *)
&operation),
status < 0))
{
#endif
if (rmtlseek (archive, (off_t) 0, SEEK_SET) != 0)
{
/* Lseek failed. Try a different method. */
seek_warn (archive_name_array[0]);
return;
}
#ifdef MTIOCTOP
}
}
}
#endif


Here is output from strace showing what happens with 'tar -c -W' applied
at the beginning of the tape (this is using kernel 2.6.11-rc4 but the same
probably happens with 2.4.29):
...
ioctl(3, MGSL_IOCGPARAMS or MTIOCTOP or SNDCTL_MIDI_MPUMODE,
0x7fffffffecd0) = -1 EIO (Input/output error)
ioctl(3, MGSL_IOCGPARAMS or MTIOCTOP or SNDCTL_MIDI_MPUMODE,
0x7fffffffecd0) = -1 EIO (Input/output error)
lseek(3, 0, SEEK_SET) = -1 ESPIPE (Illegal seek)

So, both tape positioning commands fail and the code falls back to lseek.
Earlier it has returned success even though it has not done anything (this
was on purpose because it is the way some other Unices behave and with
reason). In that case this tar succeeded but it was pure luck. The first
BSF did position the tape correctly although it did fail.

The 2.6 st driver does contain this near the beginning of st_open():

nonseekable_open(inode, filp);

This probably makes lseek fail. This code has been in st.c since 2.6.8.




--
Kai

-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/

\
 
 \ /
  Last update: 2005-03-22 14:10    [W:0.102 / U:0.404 seconds]
©2003-2020 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site