[Gluster-devel] Problem with rm -rf

Harris Landgarten harrisl at lhjonline.com
Mon Jul 2 13:07:41 UTC 2007


That was with 255. I am compiling 256 now.

Harris

----- Original Message -----
From: "Harris Landgarten" <harrisl at lhjonline.com>
To: "Anand Avati" <avati at zresearch.com>
Cc: "gluster-devel" <gluster-devel at nongnu.org>
Sent: Monday, July 2, 2007 9:05:01 AM (GMT-0500) America/New_York
Subject: Re: [Gluster-devel] Problem with rm -rf

avati:

almost but not quite. rm -rf on linux source ran perfectly until 741 items were left and then fell into the lseek behavior.

Harris

----- Original Message -----
From: "Anand Avati" <avati at zresearch.com>
To: "Daniel van Ham Colchete" <daniel.colchete at gmail.com>
Cc: "gluster-devel" <gluster-devel at nongnu.org>
Sent: Monday, July 2, 2007 9:02:20 AM (GMT-0500) America/New_York
Subject: Re: [Gluster-devel] Problem with rm -rf

Daniel,
 Awesome! the bug was in readdir which was not flushing cache on seek
(rewinddir), hence used to return stale (deleted) entries which used to put
it in a loop. So rm -rf should work smooth now..

good catch! and thanks :)

avati

2007/7/2, Daniel van Ham Colchete <daniel.colchete at gmail.com>:
>
> People,
>
> I tried this here. I'm getting the very same results, but I noticed
> something strace'ing the 'rm -rf *' process here.
>
> All the unlink goes OK untill the rm process calls lseek at the directory
> descriptor:
>
> unlink("/proc/self/fd/5/00-INDEX")      = 0
> unlink("/proc/self/fd/5/Mylex.txt")     = 0
> getdents64(5, /* 0 entries */, 4096)    = 0
> lseek(5, 0, SEEK_SET)                   = 0
> getdents64(5, /* 49 entries */, 4096)   = 1760
> unlink("/proc/self/fd/5/st.txt")        = -1 ENOENT (No such file or
> directory)
> open(".", O_RDONLY|O_LARGEFILE)         = 3
> fchdir(5)                               = 0
> unlink("st.txt")                        = -1 ENOENT (No such file or
> directory)
> fchdir(3)                               = 0
> close(3)                                = 0
> unlink("/proc/self/fd/5/ChangeLog.megaraid") = -1 ENOENT (No such file or
> directory)
> open(".", O_RDONLY|O_LARGEFILE)         = 3
> fchdir(5)                               = 0
> unlink("ChangeLog.megaraid")            = -1 ENOENT (No such file or
> directory)
>
> There is no lseek call before that.
>
> The files being removed after that was already removed before. And now 'rm
> -rf' starts to loop itself trying to lseek and remove the same files over
> and over again.
>
> Best regards,
> Daniel
>
> On 7/1/07, Harris Landgarten <harrisl at lhjonline.com> wrote:
> >
> > This bug is easily reproduced by copying the Linux source tree to
> gluster
> > and then trying to remove it with rm -rf
> >
> > Harris
> >
> > ----- Original Message -----
> > From: "Majied Najjar" <majied.najjar at nationalnet.com>
> > To: "Harris Landgarten" <harrisl at lhjonline.com>
> > Cc: "gluster-devel" <gluster-devel at nongnu.org>
> > Sent: Friday, June 29, 2007 4:01:09 PM (GMT-0500) America/New_York
> > Subject: Re: [Gluster-devel] Problem with rm -rf
> >
> > I have some core file outputs for the same operations:
> >
> > Using host libthread_db library "/lib/tls/i686/cmov/libthread_db.so.1".
> > Core was generated by
> >
> `[glusterfsd]                                                                  '.
> > Program terminated with signal 11, Segmentation fault.
> > #0  0xb7e50639 in ?? ()
> > (gdb) bt
> > #0  0xb7e50639 in ?? ()
> >
> > Using host libthread_db library "/lib/tls/i686/cmov/libthread_db.so.1".
> > Core was generated by
> >
> `[glusterfsd]                                                                  '.
> > Program terminated with signal 11, Segmentation fault.
> > #0  0xb7e92639 in ?? ()
> > (gdb) bt
> > #0  0xb7e92639 in ?? ()
> >
> > Using host libthread_db library "/lib/tls/i686/cmov/libthread_db.so.1".
> > Core was generated by
> > `[glusterfs]                                           '.
> > Program terminated with signal 11, Segmentation fault.
> > #0  0xffffe410 in __kernel_vsyscall ()
> > (gdb) bt
> > #0  0xffffe410 in __kernel_vsyscall ()
> > #1  0xb7f0aa8d in ?? ()
> >
> >
> > On Fri, 29 Jun 2007 13:36:27 -0400 (EDT)
> > Harris Landgarten <harrisl at lhjonline.com> wrote:
> >
> > > Avati,
> > >
> > > More info on rm -rf problem
> > >
> > > rm -rf * and find . -exec rm -rf {} \;
> > >
> > > both begin properly and then fall into a sequence of looking for
> files:
> > >
> > > find . -type f -exec rm {} \;
> > >
> > > works fast and properly
> > >
> > > rm -rf * then works with empty dirs.
> > >
> > > Harris
> > >
> > >
> > >
> > >
> > > ----- Original Message -----
> > > From: "Harris Landgarten" <harrisl at lhjonline.com>
> > > To: "Anand Avati" <avati at zresearch.com>
> > > Cc: "gluster-devel" <gluster-devel at nongnu.org>
> > > Sent: Thursday, June 28, 2007 9:21:46 AM (GMT-0500) America/New_York
> > > Subject: Re: [Gluster-devel] Problem with rm -rf
> > >
> > > the rm -rf hangs. It looks like one or two unlinks are sent to the
> log.
> > I can cntl-C the client and the data is still there. The data was is the
> tmp
> > dir from failed backups. It is gone now. I will investigate more when I
> have
> > more data later today.
> > >
> > > Harris
> > >
> > > ----- Original Message -----
> > > From: "Anand Avati" <avati at zresearch.com>
> > > To: "Harris Landgarten" <harrisl at lhjonline.com>
> > > Cc: "gluster-devel" <gluster-devel at nongnu.org>
> > > Sent: Thursday, June 28, 2007 9:17:41 AM (GMT-0500) America/New_York
> > > Subject: Re: [Gluster-devel] Problem with rm -rf
> > >
> > > Strange,
> > > what is your configuration? At the time of 'hang', is it possible for
> > you to attach gdb to glusterfs and get a backtrace (from every thread,
> by
> > switching as 'thr 1' 'thr 2' etc) ?
> > > rm -rf seems to work fine for me, wondering how find . -exec rm would
> > make a difference.
> > > thanks,
> > > avati
> > >
> > > > I am trying to delete the contents of a tmp dir with 3 trees
> > containing about 1.7G
> > > > as root, from withint the top level tmp dir I issue
> > > >
> > > > rm -rf *
> > > >
> > > > and the command hangs are never returns.
> > > >
> > > > find . -exec rm -rf {} \;
> > > >
> > > > works as expected.
> > > >
> > > >
> > > > Harris
> > >
> > >
> > >
> > >
> > > _______________________________________________
> > > Gluster-devel mailing list
> > > Gluster-devel at nongnu.org
> > > http://lists.nongnu.org/mailman/listinfo/gluster-devel
> > >
> > >
> > > --
> > > Anand V. Avati
> > >
> > >
> > > _______________________________________________
> > > Gluster-devel mailing list
> > > Gluster-devel at nongnu.org
> > > http://lists.nongnu.org/mailman/listinfo/gluster-devel
> > >
> > >
> > >
> > > _______________________________________________
> > > Gluster-devel mailing list
> > > Gluster-devel at nongnu.org
> > > http://lists.nongnu.org/mailman/listinfo/gluster-devel
> >
> >
> >
> > _______________________________________________
> > Gluster-devel mailing list
> > Gluster-devel at nongnu.org
> > http://lists.nongnu.org/mailman/listinfo/gluster-devel
> >
> _______________________________________________
> Gluster-devel mailing list
> Gluster-devel at nongnu.org
> http://lists.nongnu.org/mailman/listinfo/gluster-devel
>



-- 
Anand V. Avati
_______________________________________________
Gluster-devel mailing list
Gluster-devel at nongnu.org
http://lists.nongnu.org/mailman/listinfo/gluster-devel



_______________________________________________
Gluster-devel mailing list
Gluster-devel at nongnu.org
http://lists.nongnu.org/mailman/listinfo/gluster-devel






More information about the Gluster-devel mailing list