Lustre devel - Jan 2007 - [Bug 9303] repeated lconfs result in EMFILE

Please don''t reply to lustre-devel. Instead, comment in Bugzilla by
using the following link:
https://bugzilla.lustre.org/show_bug.cgi?id=9303



Created an attachment (id=9447)
Please don''t reply to lustre-devel. Instead, comment in Bugzilla by
using the following link:
 --> (https://bugzilla.lustre.org/attachment.cgi?id=9447&action=view)
fix leak anon devices at clustred nfs case.

After some investigations i found sb->s_dev don`t restored to original value
while call put_super and it`s prevent to correctly mark this id is unused.

Please don''t reply to lustre-devel. Instead, comment in Bugzilla by
using the following link:
https://bugzilla.lustre.org/show_bug.cgi?id=9303



(From update of attachment 9447)
>@@ -234,6 +234,7 @@ int lustre_common_fill_super(struct supe
>         if (data->ocd_connect_flags & OBD_CONNECT_JOIN)
>                 sbi->ll_flags |= LL_SBI_JOIN;
> 
>+        sbi->ll_sdev = sb->s_dev;
> #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,6,0))
>         /* We set sb->s_dev equal on all lustre clients in order to
support
>          * NFS export clustering.  NFSD requires that the FSID be the same
Interesting idea.  I had always tried to work harder to ensure that the code
"did the right thing for the VFS", but this is simpler.

Presumably though the #if line disabling the above needs to be uncommented for
2.6 kernels in order for this to be useful?

Please don''t reply to lustre-devel. Instead, comment in Bugzilla by
using the following link:
https://bugzilla.lustre.org/show_bug.cgi?id=9303



i don`t investigate how it work (seems it`s addresed to other bug about nfsd
clustering) but will be test it in near future.

Please don''t reply to lustre-devel. Instead, comment in Bugzilla by
using the following link:
https://bugzilla.lustre.org/show_bug.cgi?id=9303

           What    |Removed                     |Added
----------------------------------------------------------------------------
   Attachment #9447|review?(green@clusterfs.com)|review+
               Flag|                            |


(From update of attachment 9447)
I would much prefer if you put all of that undef #if version < 2.5
since we do not override devices on 2.6 kernels, because kernel actually expect
valid devices all the time in 2.6
But even as it is patch should cause no harm.