Re: engines/sync.c:26, func=lseek, error=Bad file descriptor and an Assertion

From: Jens Axboe <jens.axboe_at_oracle.com>
Date: Tue, 22 May 2007 11:12:36 +0200

On Tue, May 22 2007, Jens Axboe wrote:
> On Tue, May 22 2007, peter wrote:
> > Hello,
> >
> > I just came across fio, and I want to use it to do some iSCSI tests with
> > the iet target and the iscsi initiator in Centos 5 / RHEL 5.
> >
> > Versions tried: 1.16 and latest git snapshot from today.
> >
> > During some non-iSCSI related testing (that is, using a local ext3
> > filesystem under LVM on top of software RAID5), I repeatedly got:
> >
> > fio: pid=17522, err=9/file:engines/sync.c:26, func=lseek, error=Bad
> > file descriptor
> >
> > just before the job output. fio exits with code 1 in this case. This
> > happens on both Centos 4.2 and Centos 5 based systems.
> >
> > I'm using the following job file:
> >
> > [global]
> > size=128m
> >
> > [job1]
> > rw=rw
> > verify=md5
> > randrepeat=1
> > fadvise_hint=0
> >
> >
> > Sometimes (not easily reproducable) I even get:
> >
> > fio: ioengines.c:188: td_io_queue: Assertion `io_u->file->flags &
> > FIO_FILE_OPEN' failed.
> >
> > So two questions arise:
> >
> > * Is this a bug in fio?
> > * Is this a bug in the linux kernel or filesystem?
> >
> > Personally, I hope it is the former....
>
> It's definitely a fio bug.
>
> > Any other information I can provide?
>
> No I think your report is fine, I can even reproduce it here. It looks
> like the io verification doesn't get the file(s) re-opened properly,
> hence it throws that error. The two different things you see are
> essentially the same problem, it's just a question of timing which one
> you hit.
>
> I'll cook up a fix.

OK, try the current fio snapshot (or do a git pull), it should work now.

-- 
Jens Axboe
Received on Tue May 22 2007 - 11:12:36 CEST

This archive was generated by hypermail 2.2.0 : Tue May 22 2007 - 11:30:01 CEST